fs/btrfs/volumes.c

   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 <linux/capability.h>
  26 #include <linux/ratelimit.h>
  27 #include <linux/kthread.h>
  28 #include <linux/raid/pq.h>
  29 #include <linux/semaphore.h>
  30 #include <asm/div64.h>
  31 #include "ctree.h"
  32 #include "extent_map.h"
  33 #include "disk-io.h"
  34 #include "transaction.h"
  35 #include "print-tree.h"
  36 #include "volumes.h"
  37 #include "raid56.h"
  38 #include "async-thread.h"
  39 #include "check-integrity.h"
  40 #include "rcu-string.h"
  41 #include "math.h"
  42 #include "dev-replace.h"
  43
  44 static int init_first_rw_device(struct btrfs_trans_handle *trans,
  45                                 struct btrfs_root *root,
  46                                 struct btrfs_device *device);
  47 static int btrfs_relocate_sys_chunks(struct btrfs_root *root);
  48 static void __btrfs_reset_dev_stats(struct btrfs_device *dev);
  49 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev);
  50 static void btrfs_dev_stat_print_on_load(struct btrfs_device *device);
  51
  52 static DEFINE_MUTEX(uuid_mutex);
  53 static LIST_HEAD(fs_uuids);
  54
  55 static void lock_chunks(struct btrfs_root *root)
  56 {
  57         mutex_lock(&root->fs_info->chunk_mutex);
  58 }
  59
  60 static void unlock_chunks(struct btrfs_root *root)
  61 {
  62         mutex_unlock(&root->fs_info->chunk_mutex);
  63 }
  64
  65 static struct btrfs_fs_devices *__alloc_fs_devices(void)
  66 {
  67         struct btrfs_fs_devices *fs_devs;
  68
  69         fs_devs = kzalloc(sizeof(*fs_devs), GFP_NOFS);
  70         if (!fs_devs)
  71                 return ERR_PTR(-ENOMEM);
  72
  73         mutex_init(&fs_devs->device_list_mutex);
  74
  75         INIT_LIST_HEAD(&fs_devs->devices);
  76         INIT_LIST_HEAD(&fs_devs->alloc_list);
  77         INIT_LIST_HEAD(&fs_devs->list);
  78
  79         return fs_devs;
  80 }
  81
  82 /**
  83  * alloc_fs_devices - allocate struct btrfs_fs_devices
  84  * @fsid:       a pointer to UUID for this FS.  If NULL a new UUID is
  85  *              generated.
  86  *
  87  * Return: a pointer to a new &struct btrfs_fs_devices on success;
  88  * ERR_PTR() on error.  Returned struct is not linked onto any lists and
  89  * can be destroyed with kfree() right away.
  90  */
  91 static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid)
  92 {
  93         struct btrfs_fs_devices *fs_devs;
  94
  95         fs_devs = __alloc_fs_devices();
  96         if (IS_ERR(fs_devs))
  97                 return fs_devs;
  98
  99         if (fsid)
 100                 memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE);
 101         else
 102                 generate_random_uuid(fs_devs->fsid);
 103
 104         return fs_devs;
 105 }
 106
 107 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
 108 {
 109         struct btrfs_device *device;
 110         WARN_ON(fs_devices->opened);
 111         while (!list_empty(&fs_devices->devices)) {
 112                 device = list_entry(fs_devices->devices.next,
 113                                     struct btrfs_device, dev_list);
 114                 list_del(&device->dev_list);
 115                 rcu_string_free(device->name);
 116                 kfree(device);
 117         }
 118         kfree(fs_devices);
 119 }
 120
 121 static void btrfs_kobject_uevent(struct block_device *bdev,
 122                                  enum kobject_action action)
 123 {
 124         int ret;
 125
 126         ret = kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, action);
 127         if (ret)
 128                 pr_warn("BTRFS: Sending event '%d' to kobject: '%s' (%p): failed\n",
 129                         action,
 130                         kobject_name(&disk_to_dev(bdev->bd_disk)->kobj),
 131                         &disk_to_dev(bdev->bd_disk)->kobj);
 132 }
 133
 134 void btrfs_cleanup_fs_uuids(void)
 135 {
 136         struct btrfs_fs_devices *fs_devices;
 137
 138         while (!list_empty(&fs_uuids)) {
 139                 fs_devices = list_entry(fs_uuids.next,
 140                                         struct btrfs_fs_devices, list);
 141                 list_del(&fs_devices->list);
 142                 free_fs_devices(fs_devices);
 143         }
 144 }
 145
 146 static struct btrfs_device *__alloc_device(void)
 147 {
 148         struct btrfs_device *dev;
 149
 150         dev = kzalloc(sizeof(*dev), GFP_NOFS);
 151         if (!dev)
 152                 return ERR_PTR(-ENOMEM);
 153
 154         INIT_LIST_HEAD(&dev->dev_list);
 155         INIT_LIST_HEAD(&dev->dev_alloc_list);
 156
 157         spin_lock_init(&dev->io_lock);
 158
 159         spin_lock_init(&dev->reada_lock);
 160         atomic_set(&dev->reada_in_flight, 0);
 161         INIT_RADIX_TREE(&dev->reada_zones, GFP_NOFS & ~__GFP_WAIT);
 162         INIT_RADIX_TREE(&dev->reada_extents, GFP_NOFS & ~__GFP_WAIT);
 163
 164         return dev;
 165 }
 166
 167 static noinline struct btrfs_device *__find_device(struct list_head *head,
 168                                                    u64 devid, u8 *uuid)
 169 {
 170         struct btrfs_device *dev;
 171
 172         list_for_each_entry(dev, head, dev_list) {
 173                 if (dev->devid == devid &&
 174                     (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
 175                         return dev;
 176                 }
 177         }
 178         return NULL;
 179 }
 180
 181 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
 182 {
 183         struct btrfs_fs_devices *fs_devices;
 184
 185         list_for_each_entry(fs_devices, &fs_uuids, list) {
 186                 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
 187                         return fs_devices;
 188         }
 189         return NULL;
 190 }
 191
 192 static int
 193 btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder,
 194                       int flush, struct block_device **bdev,
 195                       struct buffer_head **bh)
 196 {
 197         int ret;
 198
 199         *bdev = blkdev_get_by_path(device_path, flags, holder);
 200
 201         if (IS_ERR(*bdev)) {
 202                 ret = PTR_ERR(*bdev);
 203                 printk(KERN_INFO "BTRFS: open %s failed\n", device_path);
 204                 goto error;
 205         }
 206
 207         if (flush)
 208                 filemap_write_and_wait((*bdev)->bd_inode->i_mapping);
 209         ret = set_blocksize(*bdev, 4096);
 210         if (ret) {
 211                 blkdev_put(*bdev, flags);
 212                 goto error;
 213         }
 214         invalidate_bdev(*bdev);
 215         *bh = btrfs_read_dev_super(*bdev);
 216         if (!*bh) {
 217                 ret = -EINVAL;
 218                 blkdev_put(*bdev, flags);
 219                 goto error;
 220         }
 221
 222         return 0;
 223
 224 error:
 225         *bdev = NULL;
 226         *bh = NULL;
 227         return ret;
 228 }
 229
 230 static void requeue_list(struct btrfs_pending_bios *pending_bios,
 231                         struct bio *head, struct bio *tail)
 232 {
 233
 234         struct bio *old_head;
 235
 236         old_head = pending_bios->head;
 237         pending_bios->head = head;
 238         if (pending_bios->tail)
 239                 tail->bi_next = old_head;
 240         else
 241                 pending_bios->tail = tail;
 242 }
 243
 244 /*
 245  * we try to collect pending bios for a device so we don't get a large
 246  * number of procs sending bios down to the same device.  This greatly
 247  * improves the schedulers ability to collect and merge the bios.
 248  *
 249  * But, it also turns into a long list of bios to process and that is sure
 250  * to eventually make the worker thread block.  The solution here is to
 251  * make some progress and then put this work struct back at the end of
 252  * the list if the block device is congested.  This way, multiple devices
 253  * can make progress from a single worker thread.
 254  */
 255 static noinline void run_scheduled_bios(struct btrfs_device *device)
 256 {
 257         struct bio *pending;
 258         struct backing_dev_info *bdi;
 259         struct btrfs_fs_info *fs_info;
 260         struct btrfs_pending_bios *pending_bios;
 261         struct bio *tail;
 262         struct bio *cur;
 263         int again = 0;
 264         unsigned long num_run;
 265         unsigned long batch_run = 0;
 266         unsigned long limit;
 267         unsigned long last_waited = 0;
 268         int force_reg = 0;
 269         int sync_pending = 0;
 270         struct blk_plug plug;
 271
 272         /*
 273          * this function runs all the bios we've collected for
 274          * a particular device.  We don't want to wander off to
 275          * another device without first sending all of these down.
 276          * So, setup a plug here and finish it off before we return
 277          */
 278         blk_start_plug(&plug);
 279
 280         bdi = blk_get_backing_dev_info(device->bdev);
 281         fs_info = device->dev_root->fs_info;
 282         limit = btrfs_async_submit_limit(fs_info);
 283         limit = limit * 2 / 3;
 284
 285 loop:
 286         spin_lock(&device->io_lock);
 287
 288 loop_lock:
 289         num_run = 0;
 290
 291         /* take all the bios off the list at once and process them
 292          * later on (without the lock held).  But, remember the
 293          * tail and other pointers so the bios can be properly reinserted
 294          * into the list if we hit congestion
 295          */
 296         if (!force_reg && device->pending_sync_bios.head) {
 297                 pending_bios = &device->pending_sync_bios;
 298                 force_reg = 1;
 299         } else {
 300                 pending_bios = &device->pending_bios;
 301                 force_reg = 0;
 302         }
 303
 304         pending = pending_bios->head;
 305         tail = pending_bios->tail;
 306         WARN_ON(pending && !tail);
 307
 308         /*
 309          * if pending was null this time around, no bios need processing
 310          * at all and we can stop.  Otherwise it'll loop back up again
 311          * and do an additional check so no bios are missed.
 312          *
 313          * device->running_pending is used to synchronize with the
 314          * schedule_bio code.
 315          */
 316         if (device->pending_sync_bios.head == NULL &&
 317             device->pending_bios.head == NULL) {
 318                 again = 0;
 319                 device->running_pending = 0;
 320         } else {
 321                 again = 1;
 322                 device->running_pending = 1;
 323         }
 324
 325         pending_bios->head = NULL;
 326         pending_bios->tail = NULL;
 327
 328         spin_unlock(&device->io_lock);
 329
 330         while (pending) {
 331
 332                 rmb();
 333                 /* we want to work on both lists, but do more bios on the
 334                  * sync list than the regular list
 335                  */
 336                 if ((num_run > 32 &&
 337                     pending_bios != &device->pending_sync_bios &&
 338                     device->pending_sync_bios.head) ||
 339                    (num_run > 64 && pending_bios == &device->pending_sync_bios &&
 340                     device->pending_bios.head)) {
 341                         spin_lock(&device->io_lock);
 342                         requeue_list(pending_bios, pending, tail);
 343                         goto loop_lock;
 344                 }
 345
 346                 cur = pending;
 347                 pending = pending->bi_next;
 348                 cur->bi_next = NULL;
 349
 350                 if (atomic_dec_return(&fs_info->nr_async_bios) < limit &&
 351                     waitqueue_active(&fs_info->async_submit_wait))
 352                         wake_up(&fs_info->async_submit_wait);
 353
 354                 BUG_ON(atomic_read(&cur->bi_cnt) == 0);
 355
 356                 /*
 357                  * if we're doing the sync list, record that our
 358                  * plug has some sync requests on it
 359                  *
 360                  * If we're doing the regular list and there are
 361                  * sync requests sitting around, unplug before
 362                  * we add more
 363                  */
 364                 if (pending_bios == &device->pending_sync_bios) {
 365                         sync_pending = 1;
 366                 } else if (sync_pending) {
 367                         blk_finish_plug(&plug);
 368                         blk_start_plug(&plug);
 369                         sync_pending = 0;
 370                 }
 371
 372                 btrfsic_submit_bio(cur->bi_rw, cur);
 373                 num_run++;
 374                 batch_run++;
 375                 if (need_resched())
 376                         cond_resched();
 377
 378                 /*
 379                  * we made progress, there is more work to do and the bdi
 380                  * is now congested.  Back off and let other work structs
 381                  * run instead
 382                  */
 383                 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
 384                     fs_info->fs_devices->open_devices > 1) {
 385                         struct io_context *ioc;
 386
 387                         ioc = current->io_context;
 388
 389                         /*
 390                          * the main goal here is that we don't want to
 391                          * block if we're going to be able to submit
 392                          * more requests without blocking.
 393                          *
 394                          * This code does two great things, it pokes into
 395                          * the elevator code from a filesystem _and_
 396                          * it makes assumptions about how batching works.
 397                          */
 398                         if (ioc && ioc->nr_batch_requests > 0 &&
 399                             time_before(jiffies, ioc->last_waited + HZ/50UL) &&
 400                             (last_waited == 0 ||
 401                              ioc->last_waited == last_waited)) {
 402                                 /*
 403                                  * we want to go through our batch of
 404                                  * requests and stop.  So, we copy out
 405                                  * the ioc->last_waited time and test
 406                                  * against it before looping
 407                                  */
 408                                 last_waited = ioc->last_waited;
 409                                 if (need_resched())
 410                                         cond_resched();
 411                                 continue;
 412                         }
 413                         spin_lock(&device->io_lock);
 414                         requeue_list(pending_bios, pending, tail);
 415                         device->running_pending = 1;
 416
 417                         spin_unlock(&device->io_lock);
 418                         btrfs_queue_work(fs_info->submit_workers,
 419                                          &device->work);
 420                         goto done;
 421                 }
 422                 /* unplug every 64 requests just for good measure */
 423                 if (batch_run % 64 == 0) {
 424                         blk_finish_plug(&plug);
 425                         blk_start_plug(&plug);
 426                         sync_pending = 0;
 427                 }
 428         }
 429
 430         cond_resched();
 431         if (again)
 432                 goto loop;
 433
 434         spin_lock(&device->io_lock);
 435         if (device->pending_bios.head || device->pending_sync_bios.head)
 436                 goto loop_lock;
 437         spin_unlock(&device->io_lock);
 438
 439 done:
 440         blk_finish_plug(&plug);
 441 }
 442
 443 static void pending_bios_fn(struct btrfs_work *work)
 444 {
 445         struct btrfs_device *device;
 446
 447         device = container_of(work, struct btrfs_device, work);
 448         run_scheduled_bios(device);
 449 }
 450
 451 static noinline int device_list_add(const char *path,
 452                            struct btrfs_super_block *disk_super,
 453                            u64 devid, struct btrfs_fs_devices **fs_devices_ret)
 454 {
 455         struct btrfs_device *device;
 456         struct btrfs_fs_devices *fs_devices;
 457         struct rcu_string *name;
 458         u64 found_transid = btrfs_super_generation(disk_super);
 459
 460         fs_devices = find_fsid(disk_super->fsid);
 461         if (!fs_devices) {
 462                 fs_devices = alloc_fs_devices(disk_super->fsid);
 463                 if (IS_ERR(fs_devices))
 464                         return PTR_ERR(fs_devices);
 465
 466                 list_add(&fs_devices->list, &fs_uuids);
 467                 fs_devices->latest_devid = devid;
 468                 fs_devices->latest_trans = found_transid;
 469
 470                 device = NULL;
 471         } else {
 472                 device = __find_device(&fs_devices->devices, devid,
 473                                        disk_super->dev_item.uuid);
 474         }
 475         if (!device) {
 476                 if (fs_devices->opened)
 477                         return -EBUSY;
 478
 479                 device = btrfs_alloc_device(NULL, &devid,
 480                                             disk_super->dev_item.uuid);
 481                 if (IS_ERR(device)) {
 482                         /* we can safely leave the fs_devices entry around */
 483                         return PTR_ERR(device);
 484                 }
 485
 486                 name = rcu_string_strdup(path, GFP_NOFS);
 487                 if (!name) {
 488                         kfree(device);
 489                         return -ENOMEM;
 490                 }
 491                 rcu_assign_pointer(device->name, name);
 492
 493                 mutex_lock(&fs_devices->device_list_mutex);
 494                 list_add_rcu(&device->dev_list, &fs_devices->devices);
 495                 fs_devices->num_devices++;
 496                 mutex_unlock(&fs_devices->device_list_mutex);
 497
 498                 device->fs_devices = fs_devices;
 499         } else if (!device->name || strcmp(device->name->str, path)) {
 500                 name = rcu_string_strdup(path, GFP_NOFS);
 501                 if (!name)
 502                         return -ENOMEM;
 503                 rcu_string_free(device->name);
 504                 rcu_assign_pointer(device->name, name);
 505                 if (device->missing) {
 506                         fs_devices->missing_devices--;
 507                         device->missing = 0;
 508                 }
 509         }
 510
 511         if (found_transid > fs_devices->latest_trans) {
 512                 fs_devices->latest_devid = devid;
 513                 fs_devices->latest_trans = found_transid;
 514         }
 515         *fs_devices_ret = fs_devices;
 516         return 0;
 517 }
 518
 519 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
 520 {
 521         struct btrfs_fs_devices *fs_devices;
 522         struct btrfs_device *device;
 523         struct btrfs_device *orig_dev;
 524
 525         fs_devices = alloc_fs_devices(orig->fsid);
 526         if (IS_ERR(fs_devices))
 527                 return fs_devices;
 528
 529         fs_devices->latest_devid = orig->latest_devid;
 530         fs_devices->latest_trans = orig->latest_trans;
 531         fs_devices->total_devices = orig->total_devices;
 532
 533         /* We have held the volume lock, it is safe to get the devices. */
 534         list_for_each_entry(orig_dev, &orig->devices, dev_list) {
 535                 struct rcu_string *name;
 536
 537                 device = btrfs_alloc_device(NULL, &orig_dev->devid,
 538                                             orig_dev->uuid);
 539                 if (IS_ERR(device))
 540                         goto error;
 541
 542                 /*
 543                  * This is ok to do without rcu read locked because we hold the
 544                  * uuid mutex so nothing we touch in here is going to disappear.
 545                  */
 546                 name = rcu_string_strdup(orig_dev->name->str, GFP_NOFS);
 547                 if (!name) {
 548                         kfree(device);
 549                         goto error;
 550                 }
 551                 rcu_assign_pointer(device->name, name);
 552
 553                 list_add(&device->dev_list, &fs_devices->devices);
 554                 device->fs_devices = fs_devices;
 555                 fs_devices->num_devices++;
 556         }
 557         return fs_devices;
 558 error:
 559         free_fs_devices(fs_devices);
 560         return ERR_PTR(-ENOMEM);
 561 }
 562
 563 void btrfs_close_extra_devices(struct btrfs_fs_info *fs_info,
 564                                struct btrfs_fs_devices *fs_devices, int step)
 565 {
 566         struct btrfs_device *device, *next;
 567
 568         struct block_device *latest_bdev = NULL;
 569         u64 latest_devid = 0;
 570         u64 latest_transid = 0;
 571
 572         mutex_lock(&uuid_mutex);
 573 again:
 574         /* This is the initialized path, it is safe to release the devices. */
 575         list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
 576                 if (device->in_fs_metadata) {
 577                         if (!device->is_tgtdev_for_dev_replace &&
 578                             (!latest_transid ||
 579                              device->generation > latest_transid)) {
 580                                 latest_devid = device->devid;
 581                                 latest_transid = device->generation;
 582                                 latest_bdev = device->bdev;
 583                         }
 584                         continue;
 585                 }
 586
 587                 if (device->devid == BTRFS_DEV_REPLACE_DEVID) {
 588                         /*
 589                          * In the first step, keep the device which has
 590                          * the correct fsid and the devid that is used
 591                          * for the dev_replace procedure.
 592                          * In the second step, the dev_replace state is
 593                          * read from the device tree and it is known
 594                          * whether the procedure is really active or
 595                          * not, which means whether this device is
 596                          * used or whether it should be removed.
 597                          */
 598                         if (step == 0 || device->is_tgtdev_for_dev_replace) {
 599                                 continue;
 600                         }
 601                 }
 602                 if (device->bdev) {
 603                         blkdev_put(device->bdev, device->mode);
 604                         device->bdev = NULL;
 605                         fs_devices->open_devices--;
 606                 }
 607                 if (device->writeable) {
 608                         list_del_init(&device->dev_alloc_list);
 609                         device->writeable = 0;
 610                         if (!device->is_tgtdev_for_dev_replace)
 611                                 fs_devices->rw_devices--;
 612                 }
 613                 list_del_init(&device->dev_list);
 614                 fs_devices->num_devices--;
 615                 rcu_string_free(device->name);
 616                 kfree(device);
 617         }
 618
 619         if (fs_devices->seed) {
 620                 fs_devices = fs_devices->seed;
 621                 goto again;
 622         }
 623
 624         fs_devices->latest_bdev = latest_bdev;
 625         fs_devices->latest_devid = latest_devid;
 626         fs_devices->latest_trans = latest_transid;
 627
 628         mutex_unlock(&uuid_mutex);
 629 }
 630
 631 static void __free_device(struct work_struct *work)
 632 {
 633         struct btrfs_device *device;
 634
 635         device = container_of(work, struct btrfs_device, rcu_work);
 636
 637         if (device->bdev)
 638                 blkdev_put(device->bdev, device->mode);
 639
 640         rcu_string_free(device->name);
 641         kfree(device);
 642 }
 643
 644 static void free_device(struct rcu_head *head)
 645 {
 646         struct btrfs_device *device;
 647
 648         device = container_of(head, struct btrfs_device, rcu);
 649
 650         INIT_WORK(&device->rcu_work, __free_device);
 651         schedule_work(&device->rcu_work);
 652 }
 653
 654 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
 655 {
 656         struct btrfs_device *device;
 657
 658         if (--fs_devices->opened > 0)
 659                 return 0;
 660
 661         mutex_lock(&fs_devices->device_list_mutex);
 662         list_for_each_entry(device, &fs_devices->devices, dev_list) {
 663                 struct btrfs_device *new_device;
 664                 struct rcu_string *name;
 665
 666                 if (device->bdev)
 667                         fs_devices->open_devices--;
 668
 669                 if (device->writeable &&
 670                     device->devid != BTRFS_DEV_REPLACE_DEVID) {
 671                         list_del_init(&device->dev_alloc_list);
 672                         fs_devices->rw_devices--;
 673                 }
 674
 675                 if (device->can_discard)
 676                         fs_devices->num_can_discard--;
 677                 if (device->missing)
 678                         fs_devices->missing_devices--;
 679
 680                 new_device = btrfs_alloc_device(NULL, &device->devid,
 681                                                 device->uuid);
 682                 BUG_ON(IS_ERR(new_device)); /* -ENOMEM */
 683
 684                 /* Safe because we are under uuid_mutex */
 685                 if (device->name) {
 686                         name = rcu_string_strdup(device->name->str, GFP_NOFS);
 687                         BUG_ON(!name); /* -ENOMEM */
 688                         rcu_assign_pointer(new_device->name, name);
 689                 }
 690
 691                 list_replace_rcu(&device->dev_list, &new_device->dev_list);
 692                 new_device->fs_devices = device->fs_devices;
 693
 694                 call_rcu(&device->rcu, free_device);
 695         }
 696         mutex_unlock(&fs_devices->device_list_mutex);
 697
 698         WARN_ON(fs_devices->open_devices);
 699         WARN_ON(fs_devices->rw_devices);
 700         fs_devices->opened = 0;
 701         fs_devices->seeding = 0;
 702
 703         return 0;
 704 }
 705
 706 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
 707 {
 708         struct btrfs_fs_devices *seed_devices = NULL;
 709         int ret;
 710
 711         mutex_lock(&uuid_mutex);
 712         ret = __btrfs_close_devices(fs_devices);
 713         if (!fs_devices->opened) {
 714                 seed_devices = fs_devices->seed;
 715                 fs_devices->seed = NULL;
 716         }
 717         mutex_unlock(&uuid_mutex);
 718
 719         while (seed_devices) {
 720                 fs_devices = seed_devices;
 721                 seed_devices = fs_devices->seed;
 722                 __btrfs_close_devices(fs_devices);
 723                 free_fs_devices(fs_devices);
 724         }
 725         /*
 726          * Wait for rcu kworkers under __btrfs_close_devices
 727          * to finish all blkdev_puts so device is really
 728          * free when umount is done.
 729          */
 730         rcu_barrier();
 731         return ret;
 732 }
 733
 734 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
 735                                 fmode_t flags, void *holder)
 736 {
 737         struct request_queue *q;
 738         struct block_device *bdev;
 739         struct list_head *head = &fs_devices->devices;
 740         struct btrfs_device *device;
 741         struct block_device *latest_bdev = NULL;
 742         struct buffer_head *bh;
 743         struct btrfs_super_block *disk_super;
 744         u64 latest_devid = 0;
 745         u64 latest_transid = 0;
 746         u64 devid;
 747         int seeding = 1;
 748         int ret = 0;
 749
 750         flags |= FMODE_EXCL;
 751
 752         list_for_each_entry(device, head, dev_list) {
 753                 if (device->bdev)
 754                         continue;
 755                 if (!device->name)
 756                         continue;
 757
 758                 /* Just open everything we can; ignore failures here */
 759                 if (btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1,
 760                                             &bdev, &bh))
 761                         continue;
 762
 763                 disk_super = (struct btrfs_super_block *)bh->b_data;
 764                 devid = btrfs_stack_device_id(&disk_super->dev_item);
 765                 if (devid != device->devid)
 766                         goto error_brelse;
 767
 768                 if (memcmp(device->uuid, disk_super->dev_item.uuid,
 769                            BTRFS_UUID_SIZE))
 770                         goto error_brelse;
 771
 772                 device->generation = btrfs_super_generation(disk_super);
 773                 if (!latest_transid || device->generation > latest_transid) {
 774                         latest_devid = devid;
 775                         latest_transid = device->generation;
 776                         latest_bdev = bdev;
 777                 }
 778
 779                 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
 780                         device->writeable = 0;
 781                 } else {
 782                         device->writeable = !bdev_read_only(bdev);
 783                         seeding = 0;
 784                 }
 785
 786                 q = bdev_get_queue(bdev);
 787                 if (blk_queue_discard(q)) {
 788                         device->can_discard = 1;
 789                         fs_devices->num_can_discard++;
 790                 }
 791
 792                 device->bdev = bdev;
 793                 device->in_fs_metadata = 0;
 794                 device->mode = flags;
 795
 796                 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
 797                         fs_devices->rotating = 1;
 798
 799                 fs_devices->open_devices++;
 800                 if (device->writeable &&
 801                     device->devid != BTRFS_DEV_REPLACE_DEVID) {
 802                         fs_devices->rw_devices++;
 803                         list_add(&device->dev_alloc_list,
 804                                  &fs_devices->alloc_list);
 805                 }
 806                 brelse(bh);
 807                 continue;
 808
 809 error_brelse:
 810                 brelse(bh);
 811                 blkdev_put(bdev, flags);
 812                 continue;
 813         }
 814         if (fs_devices->open_devices == 0) {
 815                 ret = -EINVAL;
 816                 goto out;
 817         }
 818         fs_devices->seeding = seeding;
 819         fs_devices->opened = 1;
 820         fs_devices->latest_bdev = latest_bdev;
 821         fs_devices->latest_devid = latest_devid;
 822         fs_devices->latest_trans = latest_transid;
 823         fs_devices->total_rw_bytes = 0;
 824 out:
 825         return ret;
 826 }
 827
 828 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
 829                        fmode_t flags, void *holder)
 830 {
 831         int ret;
 832
 833         mutex_lock(&uuid_mutex);
 834         if (fs_devices->opened) {
 835                 fs_devices->opened++;
 836                 ret = 0;
 837         } else {
 838                 ret = __btrfs_open_devices(fs_devices, flags, holder);
 839         }
 840         mutex_unlock(&uuid_mutex);
 841         return ret;
 842 }
 843
 844 /*
 845  * Look for a btrfs signature on a device. This may be called out of the mount path
 846  * and we are not allowed to call set_blocksize during the scan. The superblock
 847  * is read via pagecache
 848  */
 849 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
 850                           struct btrfs_fs_devices **fs_devices_ret)
 851 {
 852         struct btrfs_super_block *disk_super;
 853         struct block_device *bdev;
 854         struct page *page;
 855         void *p;
 856         int ret = -EINVAL;
 857         u64 devid;
 858         u64 transid;
 859         u64 total_devices;
 860         u64 bytenr;
 861         pgoff_t index;
 862
 863         /*
 864          * we would like to check all the supers, but that would make
 865          * a btrfs mount succeed after a mkfs from a different FS.
 866          * So, we need to add a special mount option to scan for
 867          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
 868          */
 869         bytenr = btrfs_sb_offset(0);
 870         flags |= FMODE_EXCL;
 871         mutex_lock(&uuid_mutex);
 872
 873         bdev = blkdev_get_by_path(path, flags, holder);
 874
 875         if (IS_ERR(bdev)) {
 876                 ret = PTR_ERR(bdev);
 877                 goto error;
 878         }
 879
 880         /* make sure our super fits in the device */
 881         if (bytenr + PAGE_CACHE_SIZE >= i_size_read(bdev->bd_inode))
 882                 goto error_bdev_put;
 883
 884         /* make sure our super fits in the page */
 885         if (sizeof(*disk_super) > PAGE_CACHE_SIZE)
 886                 goto error_bdev_put;
 887
 888         /* make sure our super doesn't straddle pages on disk */
 889         index = bytenr >> PAGE_CACHE_SHIFT;
 890         if ((bytenr + sizeof(*disk_super) - 1) >> PAGE_CACHE_SHIFT != index)
 891                 goto error_bdev_put;
 892
 893         /* pull in the page with our super */
 894         page = read_cache_page_gfp(bdev->bd_inode->i_mapping,
 895                                    index, GFP_NOFS);
 896
 897         if (IS_ERR_OR_NULL(page))
 898                 goto error_bdev_put;
 899
 900         p = kmap(page);
 901
 902         /* align our pointer to the offset of the super block */
 903         disk_super = p + (bytenr & ~PAGE_CACHE_MASK);
 904
 905         if (btrfs_super_bytenr(disk_super) != bytenr ||
 906             btrfs_super_magic(disk_super) != BTRFS_MAGIC)
 907                 goto error_unmap;
 908
 909         devid = btrfs_stack_device_id(&disk_super->dev_item);
 910         transid = btrfs_super_generation(disk_super);
 911         total_devices = btrfs_super_num_devices(disk_super);
 912
 913         if (disk_super->label[0]) {
 914                 if (disk_super->label[BTRFS_LABEL_SIZE - 1])
 915                         disk_super->label[BTRFS_LABEL_SIZE - 1] = '\0';
 916                 printk(KERN_INFO "BTRFS: device label %s ", disk_super->label);
 917         } else {
 918                 printk(KERN_INFO "BTRFS: device fsid %pU ", disk_super->fsid);
 919         }
 920
 921         printk(KERN_CONT "devid %llu transid %llu %s\n", devid, transid, path);
 922
 923         ret = device_list_add(path, disk_super, devid, fs_devices_ret);
 924         if (!ret && fs_devices_ret)
 925                 (*fs_devices_ret)->total_devices = total_devices;
 926
 927 error_unmap:
 928         kunmap(page);
 929         page_cache_release(page);
 930
 931 error_bdev_put:
 932         blkdev_put(bdev, flags);
 933 error:
 934         mutex_unlock(&uuid_mutex);
 935         return ret;
 936 }
 937
 938 /* helper to account the used device space in the range */
 939 int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
 940                                    u64 end, u64 *length)
 941 {
 942         struct btrfs_key key;
 943         struct btrfs_root *root = device->dev_root;
 944         struct btrfs_dev_extent *dev_extent;
 945         struct btrfs_path *path;
 946         u64 extent_end;
 947         int ret;
 948         int slot;
 949         struct extent_buffer *l;
 950
 951         *length = 0;
 952
 953         if (start >= device->total_bytes || device->is_tgtdev_for_dev_replace)
 954                 return 0;
 955
 956         path = btrfs_alloc_path();
 957         if (!path)
 958                 return -ENOMEM;
 959         path->reada = 2;
 960
 961         key.objectid = device->devid;
 962         key.offset = start;
 963         key.type = BTRFS_DEV_EXTENT_KEY;
 964
 965         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 966         if (ret < 0)
 967                 goto out;
 968         if (ret > 0) {
 969                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
 970                 if (ret < 0)
 971                         goto out;
 972         }
 973
 974         while (1) {
 975                 l = path->nodes[0];
 976                 slot = path->slots[0];
 977                 if (slot >= btrfs_header_nritems(l)) {
 978                         ret = btrfs_next_leaf(root, path);
 979                         if (ret == 0)
 980                                 continue;
 981                         if (ret < 0)
 982                                 goto out;
 983
 984                         break;
 985                 }
 986                 btrfs_item_key_to_cpu(l, &key, slot);
 987
 988                 if (key.objectid < device->devid)
 989                         goto next;
 990
 991                 if (key.objectid > device->devid)
 992                         break;
 993
 994                 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
 995                         goto next;
 996
 997                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
 998                 extent_end = key.offset + btrfs_dev_extent_length(l,
 999                                                                   dev_extent);
1000                 if (key.offset <= start && extent_end > end) {
1001                         *length = end - start + 1;
1002                         break;
1003                 } else if (key.offset <= start && extent_end > start)
1004                         *length += extent_end - start;
1005                 else if (key.offset > start && extent_end <= end)
1006                         *length += extent_end - key.offset;
1007                 else if (key.offset > start && key.offset <= end) {
1008                         *length += end - key.offset + 1;
1009                         break;
1010                 } else if (key.offset > end)
1011                         break;
1012
1013 next:
1014                 path->slots[0]++;
1015         }
1016         ret = 0;
1017 out:
1018         btrfs_free_path(path);
1019         return ret;
1020 }
1021
1022 static int contains_pending_extent(struct btrfs_trans_handle *trans,
1023                                    struct btrfs_device *device,
1024                                    u64 *start, u64 len)
1025 {
1026         struct extent_map *em;
1027         int ret = 0;
1028
1029         list_for_each_entry(em, &trans->transaction->pending_chunks, list) {
1030                 struct map_lookup *map;
1031                 int i;
1032
1033                 map = (struct map_lookup *)em->bdev;
1034                 for (i = 0; i < map->num_stripes; i++) {
1035                         if (map->stripes[i].dev != device)
1036                                 continue;
1037                         if (map->stripes[i].physical >= *start + len ||
1038                             map->stripes[i].physical + em->orig_block_len <=
1039                             *start)
1040                                 continue;
1041                         *start = map->stripes[i].physical +
1042                                 em->orig_block_len;
1043                         ret = 1;
1044                 }
1045         }
1046
1047         return ret;
1048 }
1049
1050
1051 /*
1052  * find_free_dev_extent - find free space in the specified device
1053  * @device:     the device which we search the free space in
1054  * @num_bytes:  the size of the free space that we need
1055  * @start:      store the start of the free space.
1056  * @len:        the size of the free space. that we find, or the size of the max
1057  *              free space if we don't find suitable free space
1058  *
1059  * this uses a pretty simple search, the expectation is that it is
1060  * called very infrequently and that a given device has a small number
1061  * of extents
1062  *
1063  * @start is used to store the start of the free space if we find. But if we
1064  * don't find suitable free space, it will be used to store the start position
1065  * of the max free space.
1066  *
1067  * @len is used to store the size of the free space that we find.
1068  * But if we don't find suitable free space, it is used to store the size of
1069  * the max free space.
1070  */
1071 int find_free_dev_extent(struct btrfs_trans_handle *trans,
1072                          struct btrfs_device *device, u64 num_bytes,
1073                          u64 *start, u64 *len)
1074 {
1075         struct btrfs_key key;
1076         struct btrfs_root *root = device->dev_root;
1077         struct btrfs_dev_extent *dev_extent;
1078         struct btrfs_path *path;
1079         u64 hole_size;
1080         u64 max_hole_start;
1081         u64 max_hole_size;
1082         u64 extent_end;
1083         u64 search_start;
1084         u64 search_end = device->total_bytes;
1085         int ret;
1086         int slot;
1087         struct extent_buffer *l;
1088
1089         /* FIXME use last free of some kind */
1090
1091         /* we don't want to overwrite the superblock on the drive,
1092          * so we make sure to start at an offset of at least 1MB
1093          */
1094         search_start = max(root->fs_info->alloc_start, 1024ull * 1024);
1095
1096         path = btrfs_alloc_path();
1097         if (!path)
1098                 return -ENOMEM;
1099 again:
1100         max_hole_start = search_start;
1101         max_hole_size = 0;
1102         hole_size = 0;
1103
1104         if (search_start >= search_end || device->is_tgtdev_for_dev_replace) {
1105                 ret = -ENOSPC;
1106                 goto out;
1107         }
1108
1109         path->reada = 2;
1110         path->search_commit_root = 1;
1111         path->skip_locking = 1;
1112
1113         key.objectid = device->devid;
1114         key.offset = search_start;
1115         key.type = BTRFS_DEV_EXTENT_KEY;
1116
1117         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1118         if (ret < 0)
1119                 goto out;
1120         if (ret > 0) {
1121                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1122                 if (ret < 0)
1123                         goto out;
1124         }
1125
1126         while (1) {
1127                 l = path->nodes[0];
1128                 slot = path->slots[0];
1129                 if (slot >= btrfs_header_nritems(l)) {
1130                         ret = btrfs_next_leaf(root, path);
1131                         if (ret == 0)
1132                                 continue;
1133                         if (ret < 0)
1134                                 goto out;
1135
1136                         break;
1137                 }
1138                 btrfs_item_key_to_cpu(l, &key, slot);
1139
1140                 if (key.objectid < device->devid)
1141                         goto next;
1142
1143                 if (key.objectid > device->devid)
1144                         break;
1145
1146                 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
1147                         goto next;
1148
1149                 if (key.offset > search_start) {
1150                         hole_size = key.offset - search_start;
1151
1152                         /*
1153                          * Have to check before we set max_hole_start, otherwise
1154                          * we could end up sending back this offset anyway.
1155                          */
1156                         if (contains_pending_extent(trans, device,
1157                                                     &search_start,
1158                                                     hole_size))
1159                                 hole_size = 0;
1160
1161                         if (hole_size > max_hole_size) {
1162                                 max_hole_start = search_start;
1163                                 max_hole_size = hole_size;
1164                         }
1165
1166                         /*
1167                          * If this free space is greater than which we need,
1168                          * it must be the max free space that we have found
1169                          * until now, so max_hole_start must point to the start
1170                          * of this free space and the length of this free space
1171                          * is stored in max_hole_size. Thus, we return
1172                          * max_hole_start and max_hole_size and go back to the
1173                          * caller.
1174                          */
1175                         if (hole_size >= num_bytes) {
1176                                 ret = 0;
1177                                 goto out;
1178                         }
1179                 }
1180
1181                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1182                 extent_end = key.offset + btrfs_dev_extent_length(l,
1183                                                                   dev_extent);
1184                 if (extent_end > search_start)
1185                         search_start = extent_end;
1186 next:
1187                 path->slots[0]++;
1188                 cond_resched();
1189         }
1190
1191         /*
1192          * At this point, search_start should be the end of
1193          * allocated dev extents, and when shrinking the device,
1194          * search_end may be smaller than search_start.
1195          */
1196         if (search_end > search_start)
1197                 hole_size = search_end - search_start;
1198
1199         if (hole_size > max_hole_size) {
1200                 max_hole_start = search_start;
1201                 max_hole_size = hole_size;
1202         }
1203
1204         if (contains_pending_extent(trans, device, &search_start, hole_size)) {
1205                 btrfs_release_path(path);
1206                 goto again;
1207         }
1208
1209         /* See above. */
1210         if (hole_size < num_bytes)
1211                 ret = -ENOSPC;
1212         else
1213                 ret = 0;
1214
1215 out:
1216         btrfs_free_path(path);
1217         *start = max_hole_start;
1218         if (len)
1219                 *len = max_hole_size;
1220         return ret;
1221 }
1222
1223 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1224                           struct btrfs_device *device,
1225                           u64 start)
1226 {
1227         int ret;
1228         struct btrfs_path *path;
1229         struct btrfs_root *root = device->dev_root;
1230         struct btrfs_key key;
1231         struct btrfs_key found_key;
1232         struct extent_buffer *leaf = NULL;
1233         struct btrfs_dev_extent *extent = NULL;
1234
1235         path = btrfs_alloc_path();
1236         if (!path)
1237                 return -ENOMEM;
1238
1239         key.objectid = device->devid;
1240         key.offset = start;
1241         key.type = BTRFS_DEV_EXTENT_KEY;
1242 again:
1243         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1244         if (ret > 0) {
1245                 ret = btrfs_previous_item(root, path, key.objectid,
1246                                           BTRFS_DEV_EXTENT_KEY);
1247                 if (ret)
1248                         goto out;
1249                 leaf = path->nodes[0];
1250                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1251                 extent = btrfs_item_ptr(leaf, path->slots[0],
1252                                         struct btrfs_dev_extent);
1253                 BUG_ON(found_key.offset > start || found_key.offset +
1254                        btrfs_dev_extent_length(leaf, extent) < start);
1255                 key = found_key;
1256                 btrfs_release_path(path);
1257                 goto again;
1258         } else if (ret == 0) {
1259                 leaf = path->nodes[0];
1260                 extent = btrfs_item_ptr(leaf, path->slots[0],
1261                                         struct btrfs_dev_extent);
1262         } else {
1263                 btrfs_error(root->fs_info, ret, "Slot search failed");
1264                 goto out;
1265         }
1266
1267         if (device->bytes_used > 0) {
1268                 u64 len = btrfs_dev_extent_length(leaf, extent);
1269                 device->bytes_used -= len;
1270                 spin_lock(&root->fs_info->free_chunk_lock);
1271                 root->fs_info->free_chunk_space += len;
1272                 spin_unlock(&root->fs_info->free_chunk_lock);
1273         }
1274         ret = btrfs_del_item(trans, root, path);
1275         if (ret) {
1276                 btrfs_error(root->fs_info, ret,
1277                             "Failed to remove dev extent item");
1278         }
1279 out:
1280         btrfs_free_path(path);
1281         return ret;
1282 }
1283
1284 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1285                                   struct btrfs_device *device,
1286                                   u64 chunk_tree, u64 chunk_objectid,
1287                                   u64 chunk_offset, u64 start, u64 num_bytes)
1288 {
1289         int ret;
1290         struct btrfs_path *path;
1291         struct btrfs_root *root = device->dev_root;
1292         struct btrfs_dev_extent *extent;
1293         struct extent_buffer *leaf;
1294         struct btrfs_key key;
1295
1296         WARN_ON(!device->in_fs_metadata);
1297         WARN_ON(device->is_tgtdev_for_dev_replace);
1298         path = btrfs_alloc_path();
1299         if (!path)
1300                 return -ENOMEM;
1301
1302         key.objectid = device->devid;
1303         key.offset = start;
1304         key.type = BTRFS_DEV_EXTENT_KEY;
1305         ret = btrfs_insert_empty_item(trans, root, path, &key,
1306                                       sizeof(*extent));
1307         if (ret)
1308                 goto out;
1309
1310         leaf = path->nodes[0];
1311         extent = btrfs_item_ptr(leaf, path->slots[0],
1312                                 struct btrfs_dev_extent);
1313         btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
1314         btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
1315         btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1316
1317         write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
1318                     btrfs_dev_extent_chunk_tree_uuid(extent), BTRFS_UUID_SIZE);
1319
1320         btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1321         btrfs_mark_buffer_dirty(leaf);
1322 out:
1323         btrfs_free_path(path);
1324         return ret;
1325 }
1326
1327 static u64 find_next_chunk(struct btrfs_fs_info *fs_info)
1328 {
1329         struct extent_map_tree *em_tree;
1330         struct extent_map *em;
1331         struct rb_node *n;
1332         u64 ret = 0;
1333
1334         em_tree = &fs_info->mapping_tree.map_tree;
1335         read_lock(&em_tree->lock);
1336         n = rb_last(&em_tree->map);
1337         if (n) {
1338                 em = rb_entry(n, struct extent_map, rb_node);
1339                 ret = em->start + em->len;
1340         }
1341         read_unlock(&em_tree->lock);
1342
1343         return ret;
1344 }
1345
1346 static noinline int find_next_devid(struct btrfs_fs_info *fs_info,
1347                                     u64 *devid_ret)
1348 {
1349         int ret;
1350         struct btrfs_key key;
1351         struct btrfs_key found_key;
1352         struct btrfs_path *path;
1353
1354         path = btrfs_alloc_path();
1355         if (!path)
1356                 return -ENOMEM;
1357
1358         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1359         key.type = BTRFS_DEV_ITEM_KEY;
1360         key.offset = (u64)-1;
1361
1362         ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
1363         if (ret < 0)
1364                 goto error;
1365
1366         BUG_ON(ret == 0); /* Corruption */
1367
1368         ret = btrfs_previous_item(fs_info->chunk_root, path,
1369                                   BTRFS_DEV_ITEMS_OBJECTID,
1370                                   BTRFS_DEV_ITEM_KEY);
1371         if (ret) {
1372                 *devid_ret = 1;
1373         } else {
1374                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1375                                       path->slots[0]);
1376                 *devid_ret = found_key.offset + 1;
1377         }
1378         ret = 0;
1379 error:
1380         btrfs_free_path(path);
1381         return ret;
1382 }
1383
1384 /*
1385  * the device information is stored in the chunk root
1386  * the btrfs_device struct should be fully filled in
1387  */
1388 static int btrfs_add_device(struct btrfs_trans_handle *trans,
1389                             struct btrfs_root *root,
1390                             struct btrfs_device *device)
1391 {
1392         int ret;
1393         struct btrfs_path *path;
1394         struct btrfs_dev_item *dev_item;
1395         struct extent_buffer *leaf;
1396         struct btrfs_key key;
1397         unsigned long ptr;
1398
1399         root = root->fs_info->chunk_root;
1400
1401         path = btrfs_alloc_path();
1402         if (!path)
1403                 return -ENOMEM;
1404
1405         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1406         key.type = BTRFS_DEV_ITEM_KEY;
1407         key.offset = device->devid;
1408
1409         ret = btrfs_insert_empty_item(trans, root, path, &key,
1410                                       sizeof(*dev_item));
1411         if (ret)
1412                 goto out;
1413
1414         leaf = path->nodes[0];
1415         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1416
1417         btrfs_set_device_id(leaf, dev_item, device->devid);
1418         btrfs_set_device_generation(leaf, dev_item, 0);
1419         btrfs_set_device_type(leaf, dev_item, device->type);
1420         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1421         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1422         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1423         btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
1424         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1425         btrfs_set_device_group(leaf, dev_item, 0);
1426         btrfs_set_device_seek_speed(leaf, dev_item, 0);
1427         btrfs_set_device_bandwidth(leaf, dev_item, 0);
1428         btrfs_set_device_start_offset(leaf, dev_item, 0);
1429
1430         ptr = btrfs_device_uuid(dev_item);
1431         write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1432         ptr = btrfs_device_fsid(dev_item);
1433         write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1434         btrfs_mark_buffer_dirty(leaf);
1435
1436         ret = 0;
1437 out:
1438         btrfs_free_path(path);
1439         return ret;
1440 }
1441
1442 static int btrfs_rm_dev_item(struct btrfs_root *root,
1443                              struct btrfs_device *device)
1444 {
1445         int ret;
1446         struct btrfs_path *path;
1447         struct btrfs_key key;
1448         struct btrfs_trans_handle *trans;
1449
1450         root = root->fs_info->chunk_root;
1451
1452         path = btrfs_alloc_path();
1453         if (!path)
1454                 return -ENOMEM;
1455
1456         trans = btrfs_start_transaction(root, 0);
1457         if (IS_ERR(trans)) {
1458                 btrfs_free_path(path);
1459                 return PTR_ERR(trans);
1460         }
1461         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1462         key.type = BTRFS_DEV_ITEM_KEY;
1463         key.offset = device->devid;
1464         lock_chunks(root);
1465
1466         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1467         if (ret < 0)
1468                 goto out;
1469
1470         if (ret > 0) {
1471                 ret = -ENOENT;
1472                 goto out;
1473         }
1474
1475         ret = btrfs_del_item(trans, root, path);
1476         if (ret)
1477                 goto out;
1478 out:
1479         btrfs_free_path(path);
1480         unlock_chunks(root);
1481         btrfs_commit_transaction(trans, root);
1482         return ret;
1483 }
1484
1485 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
1486 {
1487         struct btrfs_device *device;
1488         struct btrfs_device *next_device;
1489         struct block_device *bdev;
1490         struct buffer_head *bh = NULL;
1491         struct btrfs_super_block *disk_super;
1492         struct btrfs_fs_devices *cur_devices;
1493         u64 all_avail;
1494         u64 devid;
1495         u64 num_devices;
1496         u8 *dev_uuid;
1497         unsigned seq;
1498         int ret = 0;
1499         bool clear_super = false;
1500
1501         mutex_lock(&uuid_mutex);
1502
1503         do {
1504                 seq = read_seqbegin(&root->fs_info->profiles_lock);
1505
1506                 all_avail = root->fs_info->avail_data_alloc_bits |
1507                             root->fs_info->avail_system_alloc_bits |
1508                             root->fs_info->avail_metadata_alloc_bits;
1509         } while (read_seqretry(&root->fs_info->profiles_lock, seq));
1510
1511         num_devices = root->fs_info->fs_devices->num_devices;
1512         btrfs_dev_replace_lock(&root->fs_info->dev_replace);
1513         if (btrfs_dev_replace_is_ongoing(&root->fs_info->dev_replace)) {
1514                 WARN_ON(num_devices < 1);
1515                 num_devices--;
1516         }
1517         btrfs_dev_replace_unlock(&root->fs_info->dev_replace);
1518
1519         if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) && num_devices <= 4) {
1520                 ret = BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET;
1521                 goto out;
1522         }
1523
1524         if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) && num_devices <= 2) {
1525                 ret = BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET;
1526                 goto out;
1527         }
1528
1529         if ((all_avail & BTRFS_BLOCK_GROUP_RAID5) &&
1530             root->fs_info->fs_devices->rw_devices <= 2) {
1531                 ret = BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET;
1532                 goto out;
1533         }
1534         if ((all_avail & BTRFS_BLOCK_GROUP_RAID6) &&
1535             root->fs_info->fs_devices->rw_devices <= 3) {
1536                 ret = BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET;
1537                 goto out;
1538         }
1539
1540         if (strcmp(device_path, "missing") == 0) {
1541                 struct list_head *devices;
1542                 struct btrfs_device *tmp;
1543
1544                 device = NULL;
1545                 devices = &root->fs_info->fs_devices->devices;
1546                 /*
1547                  * It is safe to read the devices since the volume_mutex
1548                  * is held.
1549                  */
1550                 list_for_each_entry(tmp, devices, dev_list) {
1551                         if (tmp->in_fs_metadata &&
1552                             !tmp->is_tgtdev_for_dev_replace &&
1553                             !tmp->bdev) {
1554                                 device = tmp;
1555                                 break;
1556                         }
1557                 }
1558                 bdev = NULL;
1559                 bh = NULL;
1560                 disk_super = NULL;
1561                 if (!device) {
1562                         ret = BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
1563                         goto out;
1564                 }
1565         } else {
1566                 ret = btrfs_get_bdev_and_sb(device_path,
1567                                             FMODE_WRITE | FMODE_EXCL,
1568                                             root->fs_info->bdev_holder, 0,
1569                                             &bdev, &bh);
1570                 if (ret)
1571                         goto out;
1572                 disk_super = (struct btrfs_super_block *)bh->b_data;
1573                 devid = btrfs_stack_device_id(&disk_super->dev_item);
1574                 dev_uuid = disk_super->dev_item.uuid;
1575                 device = btrfs_find_device(root->fs_info, devid, dev_uuid,
1576                                            disk_super->fsid);
1577                 if (!device) {
1578                         ret = -ENOENT;
1579                         goto error_brelse;
1580                 }
1581         }
1582
1583         if (device->is_tgtdev_for_dev_replace) {
1584                 ret = BTRFS_ERROR_DEV_TGT_REPLACE;
1585                 goto error_brelse;
1586         }
1587
1588         if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1589                 ret = BTRFS_ERROR_DEV_ONLY_WRITABLE;
1590                 goto error_brelse;
1591         }
1592
1593         if (device->writeable) {
1594                 lock_chunks(root);
1595                 list_del_init(&device->dev_alloc_list);
1596                 unlock_chunks(root);
1597                 root->fs_info->fs_devices->rw_devices--;
1598                 clear_super = true;
1599         }
1600
1601         mutex_unlock(&uuid_mutex);
1602         ret = btrfs_shrink_device(device, 0);
1603         mutex_lock(&uuid_mutex);
1604         if (ret)
1605                 goto error_undo;
1606
1607         /*
1608          * TODO: the superblock still includes this device in its num_devices
1609          * counter although write_all_supers() is not locked out. This
1610          * could give a filesystem state which requires a degraded mount.
1611          */
1612         ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1613         if (ret)
1614                 goto error_undo;
1615
1616         spin_lock(&root->fs_info->free_chunk_lock);
1617         root->fs_info->free_chunk_space = device->total_bytes -
1618                 device->bytes_used;
1619         spin_unlock(&root->fs_info->free_chunk_lock);
1620
1621         device->in_fs_metadata = 0;
1622         btrfs_scrub_cancel_dev(root->fs_info, device);
1623
1624         /*
1625          * the device list mutex makes sure that we don't change
1626          * the device list while someone else is writing out all
1627          * the device supers. Whoever is writing all supers, should
1628          * lock the device list mutex before getting the number of
1629          * devices in the super block (super_copy). Conversely,
1630          * whoever updates the number of devices in the super block
1631          * (super_copy) should hold the device list mutex.
1632          */
1633
1634         cur_devices = device->fs_devices;
1635         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1636         list_del_rcu(&device->dev_list);
1637
1638         device->fs_devices->num_devices--;
1639         device->fs_devices->total_devices--;
1640
1641         if (device->missing)
1642                 root->fs_info->fs_devices->missing_devices--;
1643
1644         next_device = list_entry(root->fs_info->fs_devices->devices.next,
1645                                  struct btrfs_device, dev_list);
1646         if (device->bdev == root->fs_info->sb->s_bdev)
1647                 root->fs_info->sb->s_bdev = next_device->bdev;
1648         if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1649                 root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1650
1651         if (device->bdev)
1652                 device->fs_devices->open_devices--;
1653
1654         call_rcu(&device->rcu, free_device);
1655
1656         num_devices = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
1657         btrfs_set_super_num_devices(root->fs_info->super_copy, num_devices);
1658         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1659
1660         if (cur_devices->open_devices == 0) {
1661                 struct btrfs_fs_devices *fs_devices;
1662                 fs_devices = root->fs_info->fs_devices;
1663                 while (fs_devices) {
1664                         if (fs_devices->seed == cur_devices)
1665                                 break;
1666                         fs_devices = fs_devices->seed;
1667                 }
1668                 fs_devices->seed = cur_devices->seed;
1669                 cur_devices->seed = NULL;
1670                 lock_chunks(root);
1671                 __btrfs_close_devices(cur_devices);
1672                 unlock_chunks(root);
1673                 free_fs_devices(cur_devices);
1674         }
1675
1676         root->fs_info->num_tolerated_disk_barrier_failures =
1677                 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
1678
1679         /*
1680          * at this point, the device is zero sized.  We want to
1681          * remove it from the devices list and zero out the old super
1682          */
1683         if (clear_super && disk_super) {
1684                 /* make sure this device isn't detected as part of
1685                  * the FS anymore
1686                  */
1687                 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1688                 set_buffer_dirty(bh);
1689                 sync_dirty_buffer(bh);
1690         }
1691
1692         ret = 0;
1693
1694         /* Notify udev that device has changed */
1695         if (bdev)
1696                 btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
1697
1698 error_brelse:
1699         brelse(bh);
1700         if (bdev)
1701                 blkdev_put(bdev, FMODE_READ | FMODE_EXCL);
1702 out:
1703         mutex_unlock(&uuid_mutex);
1704         return ret;
1705 error_undo:
1706         if (device->writeable) {
1707                 lock_chunks(root);
1708                 list_add(&device->dev_alloc_list,
1709                          &root->fs_info->fs_devices->alloc_list);
1710                 unlock_chunks(root);
1711                 root->fs_info->fs_devices->rw_devices++;
1712         }
1713         goto error_brelse;
1714 }
1715
1716 void btrfs_rm_dev_replace_srcdev(struct btrfs_fs_info *fs_info,
1717                                  struct btrfs_device *srcdev)
1718 {
1719         WARN_ON(!mutex_is_locked(&fs_info->fs_devices->device_list_mutex));
1720
1721         list_del_rcu(&srcdev->dev_list);
1722         list_del_rcu(&srcdev->dev_alloc_list);
1723         fs_info->fs_devices->num_devices--;
1724         if (srcdev->missing) {
1725                 fs_info->fs_devices->missing_devices--;
1726                 fs_info->fs_devices->rw_devices++;
1727         }
1728         if (srcdev->can_discard)
1729                 fs_info->fs_devices->num_can_discard--;
1730         if (srcdev->bdev) {
1731                 fs_info->fs_devices->open_devices--;
1732
1733                 /* zero out the old super */
1734                 btrfs_scratch_superblock(srcdev);
1735         }
1736
1737         call_rcu(&srcdev->rcu, free_device);
1738 }
1739
1740 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
1741                                       struct btrfs_device *tgtdev)
1742 {
1743         struct btrfs_device *next_device;
1744
1745         WARN_ON(!tgtdev);
1746         mutex_lock(&fs_info->fs_devices->device_list_mutex);
1747         if (tgtdev->bdev) {
1748                 btrfs_scratch_superblock(tgtdev);
1749                 fs_info->fs_devices->open_devices--;
1750         }
1751         fs_info->fs_devices->num_devices--;
1752         if (tgtdev->can_discard)
1753                 fs_info->fs_devices->num_can_discard++;
1754
1755         next_device = list_entry(fs_info->fs_devices->devices.next,
1756                                  struct btrfs_device, dev_list);
1757         if (tgtdev->bdev == fs_info->sb->s_bdev)
1758                 fs_info->sb->s_bdev = next_device->bdev;
1759         if (tgtdev->bdev == fs_info->fs_devices->latest_bdev)
1760                 fs_info->fs_devices->latest_bdev = next_device->bdev;
1761         list_del_rcu(&tgtdev->dev_list);
1762
1763         call_rcu(&tgtdev->rcu, free_device);
1764
1765         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
1766 }
1767
1768 static int btrfs_find_device_by_path(struct btrfs_root *root, char *device_path,
1769                                      struct btrfs_device **device)
1770 {
1771         int ret = 0;
1772         struct btrfs_super_block *disk_super;
1773         u64 devid;
1774         u8 *dev_uuid;
1775         struct block_device *bdev;
1776         struct buffer_head *bh;
1777
1778         *device = NULL;
1779         ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
1780                                     root->fs_info->bdev_holder, 0, &bdev, &bh);
1781         if (ret)
1782                 return ret;
1783         disk_super = (struct btrfs_super_block *)bh->b_data;
1784         devid = btrfs_stack_device_id(&disk_super->dev_item);
1785         dev_uuid = disk_super->dev_item.uuid;
1786         *device = btrfs_find_device(root->fs_info, devid, dev_uuid,
1787                                     disk_super->fsid);
1788         brelse(bh);
1789         if (!*device)
1790                 ret = -ENOENT;
1791         blkdev_put(bdev, FMODE_READ);
1792         return ret;
1793 }
1794
1795 int btrfs_find_device_missing_or_by_path(struct btrfs_root *root,
1796                                          char *device_path,
1797                                          struct btrfs_device **device)
1798 {
1799         *device = NULL;
1800         if (strcmp(device_path, "missing") == 0) {
1801                 struct list_head *devices;
1802                 struct btrfs_device *tmp;
1803
1804                 devices = &root->fs_info->fs_devices->devices;
1805                 /*
1806                  * It is safe to read the devices since the volume_mutex
1807                  * is held by the caller.
1808                  */
1809                 list_for_each_entry(tmp, devices, dev_list) {
1810                         if (tmp->in_fs_metadata && !tmp->bdev) {
1811                                 *device = tmp;
1812                                 break;
1813                         }
1814                 }
1815
1816                 if (!*device) {
1817                         btrfs_err(root->fs_info, "no missing device found");
1818                         return -ENOENT;
1819                 }
1820
1821                 return 0;
1822         } else {
1823                 return btrfs_find_device_by_path(root, device_path, device);
1824         }
1825 }
1826
1827 /*
1828  * does all the dirty work required for changing file system's UUID.
1829  */
1830 static int btrfs_prepare_sprout(struct btrfs_root *root)
1831 {
1832         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
1833         struct btrfs_fs_devices *old_devices;
1834         struct btrfs_fs_devices *seed_devices;
1835         struct btrfs_super_block *disk_super = root->fs_info->super_copy;
1836         struct btrfs_device *device;
1837         u64 super_flags;
1838
1839         BUG_ON(!mutex_is_locked(&uuid_mutex));
1840         if (!fs_devices->seeding)
1841                 return -EINVAL;
1842
1843         seed_devices = __alloc_fs_devices();
1844         if (IS_ERR(seed_devices))
1845                 return PTR_ERR(seed_devices);
1846
1847         old_devices = clone_fs_devices(fs_devices);
1848         if (IS_ERR(old_devices)) {
1849                 kfree(seed_devices);
1850                 return PTR_ERR(old_devices);
1851         }
1852
1853         list_add(&old_devices->list, &fs_uuids);
1854
1855         memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
1856         seed_devices->opened = 1;
1857         INIT_LIST_HEAD(&seed_devices->devices);
1858         INIT_LIST_HEAD(&seed_devices->alloc_list);
1859         mutex_init(&seed_devices->device_list_mutex);
1860
1861         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1862         list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
1863                               synchronize_rcu);
1864
1865         list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
1866         list_for_each_entry(device, &seed_devices->devices, dev_list) {
1867                 device->fs_devices = seed_devices;
1868         }
1869
1870         fs_devices->seeding = 0;
1871         fs_devices->num_devices = 0;
1872         fs_devices->open_devices = 0;
1873         fs_devices->total_devices = 0;
1874         fs_devices->seed = seed_devices;
1875
1876         generate_random_uuid(fs_devices->fsid);
1877         memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1878         memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1879         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1880
1881         super_flags = btrfs_super_flags(disk_super) &
1882                       ~BTRFS_SUPER_FLAG_SEEDING;
1883         btrfs_set_super_flags(disk_super, super_flags);
1884
1885         return 0;
1886 }
1887
1888 /*
1889  * strore the expected generation for seed devices in device items.
1890  */
1891 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
1892                                struct btrfs_root *root)
1893 {
1894         struct btrfs_path *path;
1895         struct extent_buffer *leaf;
1896         struct btrfs_dev_item *dev_item;
1897         struct btrfs_device *device;
1898         struct btrfs_key key;
1899         u8 fs_uuid[BTRFS_UUID_SIZE];
1900         u8 dev_uuid[BTRFS_UUID_SIZE];
1901         u64 devid;
1902         int ret;
1903
1904         path = btrfs_alloc_path();
1905         if (!path)
1906                 return -ENOMEM;
1907
1908         root = root->fs_info->chunk_root;
1909         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1910         key.offset = 0;
1911         key.type = BTRFS_DEV_ITEM_KEY;
1912
1913         while (1) {
1914                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1915                 if (ret < 0)
1916                         goto error;
1917
1918                 leaf = path->nodes[0];
1919 next_slot:
1920                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1921                         ret = btrfs_next_leaf(root, path);
1922                         if (ret > 0)
1923                                 break;
1924                         if (ret < 0)
1925                                 goto error;
1926                         leaf = path->nodes[0];
1927                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1928                         btrfs_release_path(path);
1929                         continue;
1930                 }
1931
1932                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1933                 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
1934                     key.type != BTRFS_DEV_ITEM_KEY)
1935                         break;
1936
1937                 dev_item = btrfs_item_ptr(leaf, path->slots[0],
1938                                           struct btrfs_dev_item);
1939                 devid = btrfs_device_id(leaf, dev_item);
1940                 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
1941                                    BTRFS_UUID_SIZE);
1942                 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
1943                                    BTRFS_UUID_SIZE);
1944                 device = btrfs_find_device(root->fs_info, devid, dev_uuid,
1945                                            fs_uuid);
1946                 BUG_ON(!device); /* Logic error */
1947
1948                 if (device->fs_devices->seeding) {
1949                         btrfs_set_device_generation(leaf, dev_item,
1950                                                     device->generation);
1951                         btrfs_mark_buffer_dirty(leaf);
1952                 }
1953
1954                 path->slots[0]++;
1955                 goto next_slot;
1956         }
1957         ret = 0;
1958 error:
1959         btrfs_free_path(path);
1960         return ret;
1961 }
1962
1963 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
1964 {
1965         struct request_queue *q;
1966         struct btrfs_trans_handle *trans;
1967         struct btrfs_device *device;
1968         struct block_device *bdev;
1969         struct list_head *devices;
1970         struct super_block *sb = root->fs_info->sb;
1971         struct rcu_string *name;
1972         u64 total_bytes;
1973         int seeding_dev = 0;
1974         int ret = 0;
1975
1976         if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
1977                 return -EROFS;
1978
1979         bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
1980                                   root->fs_info->bdev_holder);
1981         if (IS_ERR(bdev))
1982                 return PTR_ERR(bdev);
1983
1984         if (root->fs_info->fs_devices->seeding) {
1985                 seeding_dev = 1;
1986                 down_write(&sb->s_umount);
1987                 mutex_lock(&uuid_mutex);
1988         }
1989
1990         filemap_write_and_wait(bdev->bd_inode->i_mapping);
1991
1992         devices = &root->fs_info->fs_devices->devices;
1993
1994         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1995         list_for_each_entry(device, devices, dev_list) {
1996                 if (device->bdev == bdev) {
1997                         ret = -EEXIST;
1998                         mutex_unlock(
1999                                 &root->fs_info->fs_devices->device_list_mutex);
2000                         goto error;
2001                 }
2002         }
2003         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2004
2005         device = btrfs_alloc_device(root->fs_info, NULL, NULL);
2006         if (IS_ERR(device)) {
2007                 /* we can safely leave the fs_devices entry around */
2008                 ret = PTR_ERR(device);
2009                 goto error;
2010         }
2011
2012         name = rcu_string_strdup(device_path, GFP_NOFS);
2013         if (!name) {
2014                 kfree(device);
2015                 ret = -ENOMEM;
2016                 goto error;
2017         }
2018         rcu_assign_pointer(device->name, name);
2019
2020         trans = btrfs_start_transaction(root, 0);
2021         if (IS_ERR(trans)) {
2022                 rcu_string_free(device->name);
2023                 kfree(device);
2024                 ret = PTR_ERR(trans);
2025                 goto error;
2026         }
2027
2028         lock_chunks(root);
2029
2030         q = bdev_get_queue(bdev);
2031         if (blk_queue_discard(q))
2032                 device->can_discard = 1;
2033         device->writeable = 1;
2034         device->generation = trans->transid;
2035         device->io_width = root->sectorsize;
2036         device->io_align = root->sectorsize;
2037         device->sector_size = root->sectorsize;
2038         device->total_bytes = i_size_read(bdev->bd_inode);
2039         device->disk_total_bytes = device->total_bytes;
2040         device->dev_root = root->fs_info->dev_root;
2041         device->bdev = bdev;
2042         device->in_fs_metadata = 1;
2043         device->is_tgtdev_for_dev_replace = 0;
2044         device->mode = FMODE_EXCL;
2045         device->dev_stats_valid = 1;
2046         set_blocksize(device->bdev, 4096);
2047
2048         if (seeding_dev) {
2049                 sb->s_flags &= ~MS_RDONLY;
2050                 ret = btrfs_prepare_sprout(root);
2051                 BUG_ON(ret); /* -ENOMEM */
2052         }
2053
2054         device->fs_devices = root->fs_info->fs_devices;
2055
2056         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2057         list_add_rcu(&device->dev_list, &root->fs_info->fs_devices->devices);
2058         list_add(&device->dev_alloc_list,
2059                  &root->fs_info->fs_devices->alloc_list);
2060         root->fs_info->fs_devices->num_devices++;
2061         root->fs_info->fs_devices->open_devices++;
2062         root->fs_info->fs_devices->rw_devices++;
2063         root->fs_info->fs_devices->total_devices++;
2064         if (device->can_discard)
2065                 root->fs_info->fs_devices->num_can_discard++;
2066         root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
2067
2068         spin_lock(&root->fs_info->free_chunk_lock);
2069         root->fs_info->free_chunk_space += device->total_bytes;
2070         spin_unlock(&root->fs_info->free_chunk_lock);
2071
2072         if (!blk_queue_nonrot(bdev_get_queue(bdev)))
2073                 root->fs_info->fs_devices->rotating = 1;
2074
2075         total_bytes = btrfs_super_total_bytes(root->fs_info->super_copy);
2076         btrfs_set_super_total_bytes(root->fs_info->super_copy,
2077                                     total_bytes + device->total_bytes);
2078
2079         total_bytes = btrfs_super_num_devices(root->fs_info->super_copy);
2080         btrfs_set_super_num_devices(root->fs_info->super_copy,
2081                                     total_bytes + 1);
2082         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2083
2084         if (seeding_dev) {
2085                 ret = init_first_rw_device(trans, root, device);
2086                 if (ret) {
2087                         btrfs_abort_transaction(trans, root, ret);
2088                         goto error_trans;
2089                 }
2090                 ret = btrfs_finish_sprout(trans, root);
2091                 if (ret) {
2092                         btrfs_abort_transaction(trans, root, ret);
2093                         goto error_trans;
2094                 }
2095         } else {
2096                 ret = btrfs_add_device(trans, root, device);
2097                 if (ret) {
2098                         btrfs_abort_transaction(trans, root, ret);
2099                         goto error_trans;
2100                 }
2101         }
2102
2103         /*
2104          * we've got more storage, clear any full flags on the space
2105          * infos
2106          */
2107         btrfs_clear_space_info_full(root->fs_info);
2108
2109         unlock_chunks(root);
2110         root->fs_info->num_tolerated_disk_barrier_failures =
2111                 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
2112         ret = btrfs_commit_transaction(trans, root);
2113
2114         if (seeding_dev) {
2115                 mutex_unlock(&uuid_mutex);
2116                 up_write(&sb->s_umount);
2117
2118                 if (ret) /* transaction commit */
2119                         return ret;
2120
2121                 ret = btrfs_relocate_sys_chunks(root);
2122                 if (ret < 0)
2123                         btrfs_error(root->fs_info, ret,
2124                                     "Failed to relocate sys chunks after "
2125                                     "device initialization. This can be fixed "
2126                                     "using the \"btrfs balance\" command.");
2127                 trans = btrfs_attach_transaction(root);
2128                 if (IS_ERR(trans)) {
2129                         if (PTR_ERR(trans) == -ENOENT)
2130                                 return 0;
2131                         return PTR_ERR(trans);
2132                 }
2133                 ret = btrfs_commit_transaction(trans, root);
2134         }
2135
2136         return ret;
2137
2138 error_trans:
2139         unlock_chunks(root);
2140         btrfs_end_transaction(trans, root);
2141         rcu_string_free(device->name);
2142         kfree(device);
2143 error:
2144         blkdev_put(bdev, FMODE_EXCL);
2145         if (seeding_dev) {
2146                 mutex_unlock(&uuid_mutex);
2147                 up_write(&sb->s_umount);
2148         }
2149         return ret;
2150 }
2151
2152 int btrfs_init_dev_replace_tgtdev(struct btrfs_root *root, char *device_path,
2153                                   struct btrfs_device **device_out)
2154 {
2155         struct request_queue *q;
2156         struct btrfs_device *device;
2157         struct block_device *bdev;
2158         struct btrfs_fs_info *fs_info = root->fs_info;
2159         struct list_head *devices;
2160         struct rcu_string *name;
2161         u64 devid = BTRFS_DEV_REPLACE_DEVID;
2162         int ret = 0;
2163
2164         *device_out = NULL;
2165         if (fs_info->fs_devices->seeding)
2166                 return -EINVAL;
2167
2168         bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2169                                   fs_info->bdev_holder);
2170         if (IS_ERR(bdev))
2171                 return PTR_ERR(bdev);
2172
2173         filemap_write_and_wait(bdev->bd_inode->i_mapping);
2174
2175         devices = &fs_info->fs_devices->devices;
2176         list_for_each_entry(device, devices, dev_list) {
2177                 if (device->bdev == bdev) {
2178                         ret = -EEXIST;
2179                         goto error;
2180                 }
2181         }
2182
2183         device = btrfs_alloc_device(NULL, &devid, NULL);
2184         if (IS_ERR(device)) {
2185                 ret = PTR_ERR(device);
2186                 goto error;
2187         }
2188
2189         name = rcu_string_strdup(device_path, GFP_NOFS);
2190         if (!name) {
2191                 kfree(device);
2192                 ret = -ENOMEM;
2193                 goto error;
2194         }
2195         rcu_assign_pointer(device->name, name);
2196
2197         q = bdev_get_queue(bdev);
2198         if (blk_queue_discard(q))
2199                 device->can_discard = 1;
2200         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2201         device->writeable = 1;
2202         device->generation = 0;
2203         device->io_width = root->sectorsize;
2204         device->io_align = root->sectorsize;
2205         device->sector_size = root->sectorsize;
2206         device->total_bytes = i_size_read(bdev->bd_inode);
2207         device->disk_total_bytes = device->total_bytes;
2208         device->dev_root = fs_info->dev_root;
2209         device->bdev = bdev;
2210         device->in_fs_metadata = 1;
2211         device->is_tgtdev_for_dev_replace = 1;
2212         device->mode = FMODE_EXCL;
2213         device->dev_stats_valid = 1;
2214         set_blocksize(device->bdev, 4096);
2215         device->fs_devices = fs_info->fs_devices;
2216         list_add(&device->dev_list, &fs_info->fs_devices->devices);
2217         fs_info->fs_devices->num_devices++;
2218         fs_info->fs_devices->open_devices++;
2219         if (device->can_discard)
2220                 fs_info->fs_devices->num_can_discard++;
2221         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2222
2223         *device_out = device;
2224         return ret;
2225
2226 error:
2227         blkdev_put(bdev, FMODE_EXCL);
2228         return ret;
2229 }
2230
2231 void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info *fs_info,
2232                                               struct btrfs_device *tgtdev)
2233 {
2234         WARN_ON(fs_info->fs_devices->rw_devices == 0);
2235         tgtdev->io_width = fs_info->dev_root->sectorsize;
2236         tgtdev->io_align = fs_info->dev_root->sectorsize;
2237         tgtdev->sector_size = fs_info->dev_root->sectorsize;
2238         tgtdev->dev_root = fs_info->dev_root;
2239         tgtdev->in_fs_metadata = 1;
2240 }
2241
2242 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2243                                         struct btrfs_device *device)
2244 {
2245         int ret;
2246         struct btrfs_path *path;
2247         struct btrfs_root *root;
2248         struct btrfs_dev_item *dev_item;
2249         struct extent_buffer *leaf;
2250         struct btrfs_key key;
2251
2252         root = device->dev_root->fs_info->chunk_root;
2253
2254         path = btrfs_alloc_path();
2255         if (!path)
2256                 return -ENOMEM;
2257
2258         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2259         key.type = BTRFS_DEV_ITEM_KEY;
2260         key.offset = device->devid;
2261
2262         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2263         if (ret < 0)
2264                 goto out;
2265
2266         if (ret > 0) {
2267                 ret = -ENOENT;
2268                 goto out;
2269         }
2270
2271         leaf = path->nodes[0];
2272         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2273
2274         btrfs_set_device_id(leaf, dev_item, device->devid);
2275         btrfs_set_device_type(leaf, dev_item, device->type);
2276         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2277         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2278         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2279         btrfs_set_device_total_bytes(leaf, dev_item, device->disk_total_bytes);
2280         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
2281         btrfs_mark_buffer_dirty(leaf);
2282
2283 out:
2284         btrfs_free_path(path);
2285         return ret;
2286 }
2287
2288 static int __btrfs_grow_device(struct btrfs_trans_handle *trans,
2289                       struct btrfs_device *device, u64 new_size)
2290 {
2291         struct btrfs_super_block *super_copy =
2292                 device->dev_root->fs_info->super_copy;
2293         u64 old_total = btrfs_super_total_bytes(super_copy);
2294         u64 diff = new_size - device->total_bytes;
2295
2296         if (!device->writeable)
2297                 return -EACCES;
2298         if (new_size <= device->total_bytes ||
2299             device->is_tgtdev_for_dev_replace)
2300                 return -EINVAL;
2301
2302         btrfs_set_super_total_bytes(super_copy, old_total + diff);
2303         device->fs_devices->total_rw_bytes += diff;
2304
2305         device->total_bytes = new_size;
2306         device->disk_total_bytes = new_size;
2307         btrfs_clear_space_info_full(device->dev_root->fs_info);
2308
2309         return btrfs_update_device(trans, device);
2310 }
2311
2312 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2313                       struct btrfs_device *device, u64 new_size)
2314 {
2315         int ret;
2316         lock_chunks(device->dev_root);
2317         ret = __btrfs_grow_device(trans, device, new_size);
2318         unlock_chunks(device->dev_root);
2319         return ret;
2320 }
2321
2322 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
2323                             struct btrfs_root *root,
2324                             u64 chunk_tree, u64 chunk_objectid,
2325                             u64 chunk_offset)
2326 {
2327         int ret;
2328         struct btrfs_path *path;
2329         struct btrfs_key key;
2330
2331         root = root->fs_info->chunk_root;
2332         path = btrfs_alloc_path();
2333         if (!path)
2334                 return -ENOMEM;
2335
2336         key.objectid = chunk_objectid;
2337         key.offset = chunk_offset;
2338         key.type = BTRFS_CHUNK_ITEM_KEY;
2339
2340         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2341         if (ret < 0)
2342                 goto out;
2343         else if (ret > 0) { /* Logic error or corruption */
2344                 btrfs_error(root->fs_info, -ENOENT,
2345                             "Failed lookup while freeing chunk.");
2346                 ret = -ENOENT;
2347                 goto out;
2348         }
2349
2350         ret = btrfs_del_item(trans, root, path);
2351         if (ret < 0)
2352                 btrfs_error(root->fs_info, ret,
2353                             "Failed to delete chunk item.");
2354 out:
2355         btrfs_free_path(path);
2356         return ret;
2357 }
2358
2359 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
2360                         chunk_offset)
2361 {
2362         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
2363         struct btrfs_disk_key *disk_key;
2364         struct btrfs_chunk *chunk;
2365         u8 *ptr;
2366         int ret = 0;
2367         u32 num_stripes;
2368         u32 array_size;
2369         u32 len = 0;
2370         u32 cur;
2371         struct btrfs_key key;
2372
2373         array_size = btrfs_super_sys_array_size(super_copy);
2374
2375         ptr = super_copy->sys_chunk_array;
2376         cur = 0;
2377
2378         while (cur < array_size) {
2379                 disk_key = (struct btrfs_disk_key *)ptr;
2380                 btrfs_disk_key_to_cpu(&key, disk_key);
2381
2382                 len = sizeof(*disk_key);
2383
2384                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2385                         chunk = (struct btrfs_chunk *)(ptr + len);
2386                         num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2387                         len += btrfs_chunk_item_size(num_stripes);
2388                 } else {
2389                         ret = -EIO;
2390                         break;
2391                 }
2392                 if (key.objectid == chunk_objectid &&
2393                     key.offset == chunk_offset) {
2394                         memmove(ptr, ptr + len, array_size - (cur + len));
2395                         array_size -= len;
2396                         btrfs_set_super_sys_array_size(super_copy, array_size);
2397                 } else {
2398                         ptr += len;
2399                         cur += len;
2400                 }
2401         }
2402         return ret;
2403 }
2404
2405 static int btrfs_relocate_chunk(struct btrfs_root *root,
2406                          u64 chunk_tree, u64 chunk_objectid,
2407                          u64 chunk_offset)
2408 {
2409         struct extent_map_tree *em_tree;
2410         struct btrfs_root *extent_root;
2411         struct btrfs_trans_handle *trans;
2412         struct extent_map *em;
2413         struct map_lookup *map;
2414         int ret;
2415         int i;
2416
2417         root = root->fs_info->chunk_root;
2418         extent_root = root->fs_info->extent_root;
2419         em_tree = &root->fs_info->mapping_tree.map_tree;
2420
2421         ret = btrfs_can_relocate(extent_root, chunk_offset);
2422         if (ret)
2423                 return -ENOSPC;
2424
2425         /* step one, relocate all the extents inside this chunk */
2426         ret = btrfs_relocate_block_group(extent_root, chunk_offset);
2427         if (ret)
2428                 return ret;
2429
2430         trans = btrfs_start_transaction(root, 0);
2431         if (IS_ERR(trans)) {
2432                 ret = PTR_ERR(trans);
2433                 btrfs_std_error(root->fs_info, ret);
2434                 return ret;
2435         }
2436
2437         lock_chunks(root);
2438
2439         /*
2440          * step two, delete the device extents and the
2441          * chunk tree entries
2442          */
2443         read_lock(&em_tree->lock);
2444         em = lookup_extent_mapping(em_tree, chunk_offset, 1);
2445         read_unlock(&em_tree->lock);
2446
2447         BUG_ON(!em || em->start > chunk_offset ||
2448                em->start + em->len < chunk_offset);
2449         map = (struct map_lookup *)em->bdev;
2450
2451         for (i = 0; i < map->num_stripes; i++) {
2452                 ret = btrfs_free_dev_extent(trans, map->stripes[i].dev,
2453                                             map->stripes[i].physical);
2454                 BUG_ON(ret);
2455
2456                 if (map->stripes[i].dev) {
2457                         ret = btrfs_update_device(trans, map->stripes[i].dev);
2458                         BUG_ON(ret);
2459                 }
2460         }
2461         ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
2462                                chunk_offset);
2463
2464         BUG_ON(ret);
2465
2466         trace_btrfs_chunk_free(root, map, chunk_offset, em->len);
2467
2468         if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2469                 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
2470                 BUG_ON(ret);
2471         }
2472
2473         ret = btrfs_remove_block_group(trans, extent_root, chunk_offset);
2474         BUG_ON(ret);
2475
2476         write_lock(&em_tree->lock);
2477         remove_extent_mapping(em_tree, em);
2478         write_unlock(&em_tree->lock);
2479
2480         kfree(map);
2481         em->bdev = NULL;
2482
2483         /* once for the tree */
2484         free_extent_map(em);
2485         /* once for us */
2486         free_extent_map(em);
2487
2488         unlock_chunks(root);
2489         btrfs_end_transaction(trans, root);
2490         return 0;
2491 }
2492
2493 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
2494 {
2495         struct btrfs_root *chunk_root = root->fs_info->chunk_root;
2496         struct btrfs_path *path;
2497         struct extent_buffer *leaf;
2498         struct btrfs_chunk *chunk;
2499         struct btrfs_key key;
2500         struct btrfs_key found_key;
2501         u64 chunk_tree = chunk_root->root_key.objectid;
2502         u64 chunk_type;
2503         bool retried = false;
2504         int failed = 0;
2505         int ret;
2506
2507         path = btrfs_alloc_path();
2508         if (!path)
2509                 return -ENOMEM;
2510
2511 again:
2512         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2513         key.offset = (u64)-1;
2514         key.type = BTRFS_CHUNK_ITEM_KEY;
2515
2516         while (1) {
2517                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2518                 if (ret < 0)
2519                         goto error;
2520                 BUG_ON(ret == 0); /* Corruption */
2521
2522                 ret = btrfs_previous_item(chunk_root, path, key.objectid,
2523                                           key.type);
2524                 if (ret < 0)
2525                         goto error;
2526                 if (ret > 0)
2527                         break;
2528
2529                 leaf = path->nodes[0];
2530                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2531
2532                 chunk = btrfs_item_ptr(leaf, path->slots[0],
2533                                        struct btrfs_chunk);
2534                 chunk_type = btrfs_chunk_type(leaf, chunk);
2535                 btrfs_release_path(path);
2536
2537                 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
2538                         ret = btrfs_relocate_chunk(chunk_root, chunk_tree,
2539                                                    found_key.objectid,
2540                                                    found_key.offset);
2541                         if (ret == -ENOSPC)
2542                                 failed++;
2543                         else if (ret)
2544                                 BUG();
2545                 }
2546
2547                 if (found_key.offset == 0)
2548                         break;
2549                 key.offset = found_key.offset - 1;
2550         }
2551         ret = 0;
2552         if (failed && !retried) {
2553                 failed = 0;
2554                 retried = true;
2555                 goto again;
2556         } else if (WARN_ON(failed && retried)) {
2557                 ret = -ENOSPC;
2558         }
2559 error:
2560         btrfs_free_path(path);
2561         return ret;
2562 }
2563
2564 static int insert_balance_item(struct btrfs_root *root,
2565                                struct btrfs_balance_control *bctl)
2566 {
2567         struct btrfs_trans_handle *trans;
2568         struct btrfs_balance_item *item;
2569         struct btrfs_disk_balance_args disk_bargs;
2570         struct btrfs_path *path;
2571         struct extent_buffer *leaf;
2572         struct btrfs_key key;
2573         int ret, err;
2574
2575         path = btrfs_alloc_path();
2576         if (!path)
2577                 return -ENOMEM;
2578
2579         trans = btrfs_start_transaction(root, 0);
2580         if (IS_ERR(trans)) {
2581                 btrfs_free_path(path);
2582                 return PTR_ERR(trans);
2583         }
2584
2585         key.objectid = BTRFS_BALANCE_OBJECTID;
2586         key.type = BTRFS_BALANCE_ITEM_KEY;
2587         key.offset = 0;
2588
2589         ret = btrfs_insert_empty_item(trans, root, path, &key,
2590                                       sizeof(*item));
2591         if (ret)
2592                 goto out;
2593
2594         leaf = path->nodes[0];
2595         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
2596
2597         memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
2598
2599         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
2600         btrfs_set_balance_data(leaf, item, &disk_bargs);
2601         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
2602         btrfs_set_balance_meta(leaf, item, &disk_bargs);
2603         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
2604         btrfs_set_balance_sys(leaf, item, &disk_bargs);
2605
2606         btrfs_set_balance_flags(leaf, item, bctl->flags);
2607
2608         btrfs_mark_buffer_dirty(leaf);
2609 out:
2610         btrfs_free_path(path);
2611         err = btrfs_commit_transaction(trans, root);
2612         if (err && !ret)
2613                 ret = err;
2614         return ret;
2615 }
2616
2617 static int del_balance_item(struct btrfs_root *root)
2618 {
2619         struct btrfs_trans_handle *trans;
2620         struct btrfs_path *path;
2621         struct btrfs_key key;
2622         int ret, err;
2623
2624         path = btrfs_alloc_path();
2625         if (!path)
2626                 return -ENOMEM;
2627
2628         trans = btrfs_start_transaction(root, 0);
2629         if (IS_ERR(trans)) {
2630                 btrfs_free_path(path);
2631                 return PTR_ERR(trans);
2632         }
2633
2634         key.objectid = BTRFS_BALANCE_OBJECTID;
2635         key.type = BTRFS_BALANCE_ITEM_KEY;
2636         key.offset = 0;
2637
2638         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2639         if (ret < 0)
2640                 goto out;
2641         if (ret > 0) {
2642                 ret = -ENOENT;
2643                 goto out;
2644         }
2645
2646         ret = btrfs_del_item(trans, root, path);
2647 out:
2648         btrfs_free_path(path);
2649         err = btrfs_commit_transaction(trans, root);
2650         if (err && !ret)
2651                 ret = err;
2652         return ret;
2653 }
2654
2655 /*
2656  * This is a heuristic used to reduce the number of chunks balanced on
2657  * resume after balance was interrupted.
2658  */
2659 static void update_balance_args(struct btrfs_balance_control *bctl)
2660 {
2661         /*
2662          * Turn on soft mode for chunk types that were being converted.
2663          */
2664         if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
2665                 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
2666         if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
2667                 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
2668         if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
2669                 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
2670
2671         /*
2672          * Turn on usage filter if is not already used.  The idea is
2673          * that chunks that we have already balanced should be
2674          * reasonably full.  Don't do it for chunks that are being
2675          * converted - that will keep us from relocating unconverted
2676          * (albeit full) chunks.
2677          */
2678         if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2679             !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2680                 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
2681                 bctl->data.usage = 90;
2682         }
2683         if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2684             !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2685                 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
2686                 bctl->sys.usage = 90;
2687         }
2688         if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2689             !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2690                 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
2691                 bctl->meta.usage = 90;
2692         }
2693 }
2694
2695 /*
2696  * Should be called with both balance and volume mutexes held to
2697  * serialize other volume operations (add_dev/rm_dev/resize) with
2698  * restriper.  Same goes for unset_balance_control.
2699  */
2700 static void set_balance_control(struct btrfs_balance_control *bctl)
2701 {
2702         struct btrfs_fs_info *fs_info = bctl->fs_info;
2703
2704         BUG_ON(fs_info->balance_ctl);
2705
2706         spin_lock(&fs_info->balance_lock);
2707         fs_info->balance_ctl = bctl;
2708         spin_unlock(&fs_info->balance_lock);
2709 }
2710
2711 static void unset_balance_control(struct btrfs_fs_info *fs_info)
2712 {
2713         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
2714
2715         BUG_ON(!fs_info->balance_ctl);
2716
2717         spin_lock(&fs_info->balance_lock);
2718         fs_info->balance_ctl = NULL;
2719         spin_unlock(&fs_info->balance_lock);
2720
2721         kfree(bctl);
2722 }
2723
2724 /*
2725  * Balance filters.  Return 1 if chunk should be filtered out
2726  * (should not be balanced).
2727  */
2728 static int chunk_profiles_filter(u64 chunk_type,
2729                                  struct btrfs_balance_args *bargs)
2730 {
2731         chunk_type = chunk_to_extended(chunk_type) &
2732                                 BTRFS_EXTENDED_PROFILE_MASK;
2733
2734         if (bargs->profiles & chunk_type)
2735                 return 0;
2736
2737         return 1;
2738 }
2739
2740 static int chunk_usage_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
2741                               struct btrfs_balance_args *bargs)
2742 {
2743         struct btrfs_block_group_cache *cache;
2744         u64 chunk_used, user_thresh;
2745         int ret = 1;
2746
2747         cache = btrfs_lookup_block_group(fs_info, chunk_offset);
2748         chunk_used = btrfs_block_group_used(&cache->item);
2749
2750         if (bargs->usage == 0)
2751                 user_thresh = 1;
2752         else if (bargs->usage > 100)
2753                 user_thresh = cache->key.offset;
2754         else
2755                 user_thresh = div_factor_fine(cache->key.offset,
2756                                               bargs->usage);
2757
2758         if (chunk_used < user_thresh)
2759                 ret = 0;
2760
2761         btrfs_put_block_group(cache);
2762         return ret;
2763 }
2764
2765 static int chunk_devid_filter(struct extent_buffer *leaf,
2766                               struct btrfs_chunk *chunk,
2767                               struct btrfs_balance_args *bargs)
2768 {
2769         struct btrfs_stripe *stripe;
2770         int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2771         int i;
2772
2773         for (i = 0; i < num_stripes; i++) {
2774                 stripe = btrfs_stripe_nr(chunk, i);
2775                 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
2776                         return 0;
2777         }
2778
2779         return 1;
2780 }
2781
2782 /* [pstart, pend) */
2783 static int chunk_drange_filter(struct extent_buffer *leaf,
2784                                struct btrfs_chunk *chunk,
2785                                u64 chunk_offset,
2786                                struct btrfs_balance_args *bargs)
2787 {
2788         struct btrfs_stripe *stripe;
2789         int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2790         u64 stripe_offset;
2791         u64 stripe_length;
2792         int factor;
2793         int i;
2794
2795         if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
2796                 return 0;
2797
2798         if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP |
2799              BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) {
2800                 factor = num_stripes / 2;
2801         } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) {
2802                 factor = num_stripes - 1;
2803         } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) {
2804                 factor = num_stripes - 2;
2805         } else {
2806                 factor = num_stripes;
2807         }
2808
2809         for (i = 0; i < num_stripes; i++) {
2810                 stripe = btrfs_stripe_nr(chunk, i);
2811                 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
2812                         continue;
2813
2814                 stripe_offset = btrfs_stripe_offset(leaf, stripe);
2815                 stripe_length = btrfs_chunk_length(leaf, chunk);
2816                 do_div(stripe_length, factor);
2817
2818                 if (stripe_offset < bargs->pend &&
2819                     stripe_offset + stripe_length > bargs->pstart)
2820                         return 0;
2821         }
2822
2823         return 1;
2824 }
2825
2826 /* [vstart, vend) */
2827 static int chunk_vrange_filter(struct extent_buffer *leaf,
2828                                struct btrfs_chunk *chunk,
2829                                u64 chunk_offset,
2830                                struct btrfs_balance_args *bargs)
2831 {
2832         if (chunk_offset < bargs->vend &&
2833             chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
2834                 /* at least part of the chunk is inside this vrange */
2835                 return 0;
2836
2837         return 1;
2838 }
2839
2840 static int chunk_soft_convert_filter(u64 chunk_type,
2841                                      struct btrfs_balance_args *bargs)
2842 {
2843         if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
2844                 return 0;
2845
2846         chunk_type = chunk_to_extended(chunk_type) &
2847                                 BTRFS_EXTENDED_PROFILE_MASK;
2848
2849         if (bargs->target == chunk_type)
2850                 return 1;
2851
2852         return 0;
2853 }
2854
2855 static int should_balance_chunk(struct btrfs_root *root,
2856                                 struct extent_buffer *leaf,
2857                                 struct btrfs_chunk *chunk, u64 chunk_offset)
2858 {
2859         struct btrfs_balance_control *bctl = root->fs_info->balance_ctl;
2860         struct btrfs_balance_args *bargs = NULL;
2861         u64 chunk_type = btrfs_chunk_type(leaf, chunk);
2862
2863         /* type filter */
2864         if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
2865               (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
2866                 return 0;
2867         }
2868
2869         if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
2870                 bargs = &bctl->data;
2871         else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
2872                 bargs = &bctl->sys;
2873         else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
2874                 bargs = &bctl->meta;
2875
2876         /* profiles filter */
2877         if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
2878             chunk_profiles_filter(chunk_type, bargs)) {
2879                 return 0;
2880         }
2881
2882         /* usage filter */
2883         if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
2884             chunk_usage_filter(bctl->fs_info, chunk_offset, bargs)) {
2885                 return 0;
2886         }
2887
2888         /* devid filter */
2889         if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
2890             chunk_devid_filter(leaf, chunk, bargs)) {
2891                 return 0;
2892         }
2893
2894         /* drange filter, makes sense only with devid filter */
2895         if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
2896             chunk_drange_filter(leaf, chunk, chunk_offset, bargs)) {
2897                 return 0;
2898         }
2899
2900         /* vrange filter */
2901         if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
2902             chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
2903                 return 0;
2904         }
2905
2906         /* soft profile changing mode */
2907         if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
2908             chunk_soft_convert_filter(chunk_type, bargs)) {
2909                 return 0;
2910         }
2911
2912         return 1;
2913 }
2914
2915 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
2916 {
2917         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
2918         struct btrfs_root *chunk_root = fs_info->chunk_root;
2919         struct btrfs_root *dev_root = fs_info->dev_root;
2920         struct list_head *devices;
2921         struct btrfs_device *device;
2922         u64 old_size;
2923         u64 size_to_free;
2924         struct btrfs_chunk *chunk;
2925         struct btrfs_path *path;
2926         struct btrfs_key key;
2927         struct btrfs_key found_key;
2928         struct btrfs_trans_handle *trans;
2929         struct extent_buffer *leaf;
2930         int slot;
2931         int ret;
2932         int enospc_errors = 0;
2933         bool counting = true;
2934
2935         /* step one make some room on all the devices */
2936         devices = &fs_info->fs_devices->devices;
2937         list_for_each_entry(device, devices, dev_list) {
2938                 old_size = device->total_bytes;
2939                 size_to_free = div_factor(old_size, 1);
2940                 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
2941                 if (!device->writeable ||
2942                     device->total_bytes - device->bytes_used > size_to_free ||
2943                     device->is_tgtdev_for_dev_replace)
2944                         continue;
2945
2946                 ret = btrfs_shrink_device(device, old_size - size_to_free);
2947                 if (ret == -ENOSPC)
2948                         break;
2949                 BUG_ON(ret);
2950
2951                 trans = btrfs_start_transaction(dev_root, 0);
2952                 BUG_ON(IS_ERR(trans));
2953
2954                 ret = btrfs_grow_device(trans, device, old_size);
2955                 BUG_ON(ret);
2956
2957                 btrfs_end_transaction(trans, dev_root);
2958         }
2959
2960         /* step two, relocate all the chunks */
2961         path = btrfs_alloc_path();
2962         if (!path) {
2963                 ret = -ENOMEM;
2964                 goto error;
2965         }
2966
2967         /* zero out stat counters */
2968         spin_lock(&fs_info->balance_lock);
2969         memset(&bctl->stat, 0, sizeof(bctl->stat));
2970         spin_unlock(&fs_info->balance_lock);
2971 again:
2972         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2973         key.offset = (u64)-1;
2974         key.type = BTRFS_CHUNK_ITEM_KEY;
2975
2976         while (1) {
2977                 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
2978                     atomic_read(&fs_info->balance_cancel_req)) {
2979                         ret = -ECANCELED;
2980                         goto error;
2981                 }
2982
2983                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2984                 if (ret < 0)
2985                         goto error;
2986
2987                 /*
2988                  * this shouldn't happen, it means the last relocate
2989                  * failed
2990                  */
2991                 if (ret == 0)
2992                         BUG(); /* FIXME break ? */
2993
2994                 ret = btrfs_previous_item(chunk_root, path, 0,
2995                                           BTRFS_CHUNK_ITEM_KEY);
2996                 if (ret) {
2997                         ret = 0;
2998                         break;
2999                 }
3000
3001                 leaf = path->nodes[0];
3002                 slot = path->slots[0];
3003                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3004
3005                 if (found_key.objectid != key.objectid)
3006                         break;
3007
3008                 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3009
3010                 if (!counting) {
3011                         spin_lock(&fs_info->balance_lock);
3012                         bctl->stat.considered++;
3013                         spin_unlock(&fs_info->balance_lock);
3014                 }
3015
3016                 ret = should_balance_chunk(chunk_root, leaf, chunk,
3017                                            found_key.offset);
3018                 btrfs_release_path(path);
3019                 if (!ret)
3020                         goto loop;
3021
3022                 if (counting) {
3023                         spin_lock(&fs_info->balance_lock);
3024                         bctl->stat.expected++;
3025                         spin_unlock(&fs_info->balance_lock);
3026                         goto loop;
3027                 }
3028
3029                 ret = btrfs_relocate_chunk(chunk_root,
3030                                            chunk_root->root_key.objectid,
3031                                            found_key.objectid,
3032                                            found_key.offset);
3033                 if (ret && ret != -ENOSPC)
3034                         goto error;
3035                 if (ret == -ENOSPC) {
3036                         enospc_errors++;
3037                 } else {
3038                         spin_lock(&fs_info->balance_lock);
3039                         bctl->stat.completed++;
3040                         spin_unlock(&fs_info->balance_lock);
3041                 }
3042 loop:
3043                 if (found_key.offset == 0)
3044                         break;
3045                 key.offset = found_key.offset - 1;
3046         }
3047
3048         if (counting) {
3049                 btrfs_release_path(path);
3050                 counting = false;
3051                 goto again;
3052         }
3053 error:
3054         btrfs_free_path(path);
3055         if (enospc_errors) {
3056                 btrfs_info(fs_info, "%d enospc errors during balance",
3057                        enospc_errors);
3058                 if (!ret)
3059                         ret = -ENOSPC;
3060         }
3061
3062         return ret;
3063 }
3064
3065 /**
3066  * alloc_profile_is_valid - see if a given profile is valid and reduced
3067  * @flags: profile to validate
3068  * @extended: if true @flags is treated as an extended profile
3069  */
3070 static int alloc_profile_is_valid(u64 flags, int extended)
3071 {
3072         u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
3073                                BTRFS_BLOCK_GROUP_PROFILE_MASK);
3074
3075         flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
3076
3077         /* 1) check that all other bits are zeroed */
3078         if (flags & ~mask)
3079                 return 0;
3080
3081         /* 2) see if profile is reduced */
3082         if (flags == 0)
3083                 return !extended; /* "0" is valid for usual profiles */
3084
3085         /* true if exactly one bit set */
3086         return (flags & (flags - 1)) == 0;
3087 }
3088
3089 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
3090 {
3091         /* cancel requested || normal exit path */
3092         return atomic_read(&fs_info->balance_cancel_req) ||
3093                 (atomic_read(&fs_info->balance_pause_req) == 0 &&
3094                  atomic_read(&fs_info->balance_cancel_req) == 0);
3095 }
3096
3097 static void __cancel_balance(struct btrfs_fs_info *fs_info)
3098 {
3099         int ret;
3100
3101         unset_balance_control(fs_info);
3102         ret = del_balance_item(fs_info->tree_root);
3103         if (ret)
3104                 btrfs_std_error(fs_info, ret);
3105
3106         atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3107 }
3108
3109 /*
3110  * Should be called with both balance and volume mutexes held
3111  */
3112 int btrfs_balance(struct btrfs_balance_control *bctl,
3113                   struct btrfs_ioctl_balance_args *bargs)
3114 {
3115         struct btrfs_fs_info *fs_info = bctl->fs_info;
3116         u64 allowed;
3117         int mixed = 0;
3118         int ret;
3119         u64 num_devices;
3120         unsigned seq;
3121
3122         if (btrfs_fs_closing(fs_info) ||
3123             atomic_read(&fs_info->balance_pause_req) ||
3124             atomic_read(&fs_info->balance_cancel_req)) {
3125                 ret = -EINVAL;
3126                 goto out;
3127         }
3128
3129         allowed = btrfs_super_incompat_flags(fs_info->super_copy);
3130         if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
3131                 mixed = 1;
3132
3133         /*
3134          * In case of mixed groups both data and meta should be picked,
3135          * and identical options should be given for both of them.
3136          */
3137         allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
3138         if (mixed && (bctl->flags & allowed)) {
3139                 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
3140                     !(bctl->flags & BTRFS_BALANCE_METADATA) ||
3141                     memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
3142                         btrfs_err(fs_info, "with mixed groups data and "
3143                                    "metadata balance options must be the same");
3144                         ret = -EINVAL;
3145                         goto out;
3146                 }
3147         }
3148
3149         num_devices = fs_info->fs_devices->num_devices;
3150         btrfs_dev_replace_lock(&fs_info->dev_replace);
3151         if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
3152                 BUG_ON(num_devices < 1);
3153                 num_devices--;
3154         }
3155         btrfs_dev_replace_unlock(&fs_info->dev_replace);
3156         allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE;
3157         if (num_devices == 1)
3158                 allowed |= BTRFS_BLOCK_GROUP_DUP;
3159         else if (num_devices > 1)
3160                 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1);
3161         if (num_devices > 2)
3162                 allowed |= BTRFS_BLOCK_GROUP_RAID5;
3163         if (num_devices > 3)
3164                 allowed |= (BTRFS_BLOCK_GROUP_RAID10 |
3165                             BTRFS_BLOCK_GROUP_RAID6);
3166         if ((bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3167             (!alloc_profile_is_valid(bctl->data.target, 1) ||
3168              (bctl->data.target & ~allowed))) {
3169                 btrfs_err(fs_info, "unable to start balance with target "
3170                            "data profile %llu",
3171                        bctl->data.target);
3172                 ret = -EINVAL;
3173                 goto out;
3174         }
3175         if ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3176             (!alloc_profile_is_valid(bctl->meta.target, 1) ||
3177              (bctl->meta.target & ~allowed))) {
3178                 btrfs_err(fs_info,
3179                            "unable to start balance with target metadata profile %llu",
3180                        bctl->meta.target);
3181                 ret = -EINVAL;
3182                 goto out;
3183         }
3184         if ((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3185             (!alloc_profile_is_valid(bctl->sys.target, 1) ||
3186              (bctl->sys.target & ~allowed))) {
3187                 btrfs_err(fs_info,
3188                            "unable to start balance with target system profile %llu",
3189                        bctl->sys.target);
3190                 ret = -EINVAL;
3191                 goto out;
3192         }
3193
3194         /* allow dup'ed data chunks only in mixed mode */
3195         if (!mixed && (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3196             (bctl->data.target & BTRFS_BLOCK_GROUP_DUP)) {
3197                 btrfs_err(fs_info, "dup for data is not allowed");
3198                 ret = -EINVAL;
3199                 goto out;
3200         }
3201
3202         /* allow to reduce meta or sys integrity only if force set */
3203         allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3204                         BTRFS_BLOCK_GROUP_RAID10 |
3205                         BTRFS_BLOCK_GROUP_RAID5 |
3206                         BTRFS_BLOCK_GROUP_RAID6;
3207         do {
3208                 seq = read_seqbegin(&fs_info->profiles_lock);
3209
3210                 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3211                      (fs_info->avail_system_alloc_bits & allowed) &&
3212                      !(bctl->sys.target & allowed)) ||
3213                     ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3214                      (fs_info->avail_metadata_alloc_bits & allowed) &&
3215                      !(bctl->meta.target & allowed))) {
3216                         if (bctl->flags & BTRFS_BALANCE_FORCE) {
3217                                 btrfs_info(fs_info, "force reducing metadata integrity");
3218                         } else {
3219                                 btrfs_err(fs_info, "balance will reduce metadata "
3220                                            "integrity, use force if you want this");
3221                                 ret = -EINVAL;
3222                                 goto out;
3223                         }
3224                 }
3225         } while (read_seqretry(&fs_info->profiles_lock, seq));
3226
3227         if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3228                 int num_tolerated_disk_barrier_failures;
3229                 u64 target = bctl->sys.target;
3230
3231                 num_tolerated_disk_barrier_failures =
3232                         btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3233                 if (num_tolerated_disk_barrier_failures > 0 &&
3234                     (target &
3235                      (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID0 |
3236                       BTRFS_AVAIL_ALLOC_BIT_SINGLE)))
3237                         num_tolerated_disk_barrier_failures = 0;
3238                 else if (num_tolerated_disk_barrier_failures > 1 &&
3239                          (target &
3240                           (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)))
3241                         num_tolerated_disk_barrier_failures = 1;
3242
3243                 fs_info->num_tolerated_disk_barrier_failures =
3244                         num_tolerated_disk_barrier_failures;
3245         }
3246
3247         ret = insert_balance_item(fs_info->tree_root, bctl);
3248         if (ret && ret != -EEXIST)
3249                 goto out;
3250
3251         if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
3252                 BUG_ON(ret == -EEXIST);
3253                 set_balance_control(bctl);
3254         } else {
3255                 BUG_ON(ret != -EEXIST);
3256                 spin_lock(&fs_info->balance_lock);
3257                 update_balance_args(bctl);
3258                 spin_unlock(&fs_info->balance_lock);
3259         }
3260
3261         atomic_inc(&fs_info->balance_running);
3262         mutex_unlock(&fs_info->balance_mutex);
3263
3264         ret = __btrfs_balance(fs_info);
3265
3266         mutex_lock(&fs_info->balance_mutex);
3267         atomic_dec(&fs_info->balance_running);
3268
3269         if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3270                 fs_info->num_tolerated_disk_barrier_failures =
3271                         btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3272         }
3273
3274         if (bargs) {
3275                 memset(bargs, 0, sizeof(*bargs));
3276                 update_ioctl_balance_args(fs_info, 0, bargs);
3277         }
3278
3279         if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
3280             balance_need_close(fs_info)) {
3281                 __cancel_balance(fs_info);
3282         }
3283
3284         wake_up(&fs_info->balance_wait_q);
3285
3286         return ret;
3287 out:
3288         if (bctl->flags & BTRFS_BALANCE_RESUME)
3289                 __cancel_balance(fs_info);
3290         else {
3291                 kfree(bctl);
3292                 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3293         }
3294         return ret;
3295 }
3296
3297 static int balance_kthread(void *data)
3298 {
3299         struct btrfs_fs_info *fs_info = data;
3300         int ret = 0;
3301
3302         mutex_lock(&fs_info->volume_mutex);
3303         mutex_lock(&fs_info->balance_mutex);
3304
3305         if (fs_info->balance_ctl) {
3306                 btrfs_info(fs_info, "continuing balance");
3307                 ret = btrfs_balance(fs_info->balance_ctl, NULL);
3308         }
3309
3310         mutex_unlock(&fs_info->balance_mutex);
3311         mutex_unlock(&fs_info->volume_mutex);
3312
3313         return ret;
3314 }
3315
3316 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
3317 {
3318         struct task_struct *tsk;
3319
3320         spin_lock(&fs_info->balance_lock);
3321         if (!fs_info->balance_ctl) {
3322                 spin_unlock(&fs_info->balance_lock);
3323                 return 0;
3324         }
3325         spin_unlock(&fs_info->balance_lock);
3326
3327         if (btrfs_test_opt(fs_info->tree_root, SKIP_BALANCE)) {
3328                 btrfs_info(fs_info, "force skipping balance");
3329                 return 0;
3330         }
3331
3332         tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
3333         return PTR_ERR_OR_ZERO(tsk);
3334 }
3335
3336 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
3337 {
3338         struct btrfs_balance_control *bctl;
3339         struct btrfs_balance_item *item;
3340         struct btrfs_disk_balance_args disk_bargs;
3341         struct btrfs_path *path;
3342         struct extent_buffer *leaf;
3343         struct btrfs_key key;
3344         int ret;
3345
3346         path = btrfs_alloc_path();
3347         if (!path)
3348                 return -ENOMEM;
3349
3350         key.objectid = BTRFS_BALANCE_OBJECTID;
3351         key.type = BTRFS_BALANCE_ITEM_KEY;
3352         key.offset = 0;
3353
3354         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
3355         if (ret < 0)
3356                 goto out;
3357         if (ret > 0) { /* ret = -ENOENT; */
3358                 ret = 0;
3359                 goto out;
3360         }
3361
3362         bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
3363         if (!bctl) {
3364                 ret = -ENOMEM;
3365                 goto out;
3366         }
3367
3368         leaf = path->nodes[0];
3369         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3370
3371         bctl->fs_info = fs_info;
3372         bctl->flags = btrfs_balance_flags(leaf, item);
3373         bctl->flags |= BTRFS_BALANCE_RESUME;
3374
3375         btrfs_balance_data(leaf, item, &disk_bargs);
3376         btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
3377         btrfs_balance_meta(leaf, item, &disk_bargs);
3378         btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
3379         btrfs_balance_sys(leaf, item, &disk_bargs);
3380         btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
3381
3382         WARN_ON(atomic_xchg(&fs_info->mutually_exclusive_operation_running, 1));
3383
3384         mutex_lock(&fs_info->volume_mutex);
3385         mutex_lock(&fs_info->balance_mutex);
3386
3387         set_balance_control(bctl);
3388
3389         mutex_unlock(&fs_info->balance_mutex);
3390         mutex_unlock(&fs_info->volume_mutex);
3391 out:
3392         btrfs_free_path(path);
3393         return ret;
3394 }
3395
3396 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
3397 {
3398         int ret = 0;
3399
3400         mutex_lock(&fs_info->balance_mutex);
3401         if (!fs_info->balance_ctl) {
3402                 mutex_unlock(&fs_info->balance_mutex);
3403                 return -ENOTCONN;
3404         }
3405
3406         if (atomic_read(&fs_info->balance_running)) {
3407                 atomic_inc(&fs_info->balance_pause_req);
3408                 mutex_unlock(&fs_info->balance_mutex);
3409
3410                 wait_event(fs_info->balance_wait_q,
3411                            atomic_read(&fs_info->balance_running) == 0);
3412
3413                 mutex_lock(&fs_info->balance_mutex);
3414                 /* we are good with balance_ctl ripped off from under us */
3415                 BUG_ON(atomic_read(&fs_info->balance_running));
3416                 atomic_dec(&fs_info->balance_pause_req);
3417         } else {
3418                 ret = -ENOTCONN;
3419         }
3420
3421         mutex_unlock(&fs_info->balance_mutex);
3422         return ret;
3423 }
3424
3425 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
3426 {
3427         if (fs_info->sb->s_flags & MS_RDONLY)
3428                 return -EROFS;
3429
3430         mutex_lock(&fs_info->balance_mutex);
3431         if (!fs_info->balance_ctl) {
3432                 mutex_unlock(&fs_info->balance_mutex);
3433                 return -ENOTCONN;
3434         }
3435
3436         atomic_inc(&fs_info->balance_cancel_req);
3437         /*
3438          * if we are running just wait and return, balance item is
3439          * deleted in btrfs_balance in this case
3440          */
3441         if (atomic_read(&fs_info->balance_running)) {
3442                 mutex_unlock(&fs_info->balance_mutex);
3443                 wait_event(fs_info->balance_wait_q,
3444                            atomic_read(&fs_info->balance_running) == 0);
3445                 mutex_lock(&fs_info->balance_mutex);
3446         } else {
3447                 /* __cancel_balance needs volume_mutex */
3448                 mutex_unlock(&fs_info->balance_mutex);
3449                 mutex_lock(&fs_info->volume_mutex);
3450                 mutex_lock(&fs_info->balance_mutex);
3451
3452                 if (fs_info->balance_ctl)
3453                         __cancel_balance(fs_info);
3454
3455                 mutex_unlock(&fs_info->volume_mutex);
3456         }
3457
3458         BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running));
3459         atomic_dec(&fs_info->balance_cancel_req);
3460         mutex_unlock(&fs_info->balance_mutex);
3461         return 0;
3462 }
3463
3464 static int btrfs_uuid_scan_kthread(void *data)
3465 {
3466         struct btrfs_fs_info *fs_info = data;
3467         struct btrfs_root *root = fs_info->tree_root;
3468         struct btrfs_key key;
3469         struct btrfs_key max_key;
3470         struct btrfs_path *path = NULL;
3471         int ret = 0;
3472         struct extent_buffer *eb;
3473         int slot;
3474         struct btrfs_root_item root_item;
3475         u32 item_size;
3476         struct btrfs_trans_handle *trans = NULL;
3477
3478         path = btrfs_alloc_path();
3479         if (!path) {
3480                 ret = -ENOMEM;
3481                 goto out;
3482         }
3483
3484         key.objectid = 0;
3485         key.type = BTRFS_ROOT_ITEM_KEY;
3486         key.offset = 0;
3487
3488         max_key.objectid = (u64)-1;
3489         max_key.type = BTRFS_ROOT_ITEM_KEY;
3490         max_key.offset = (u64)-1;
3491
3492         path->keep_locks = 1;
3493
3494         while (1) {
3495                 ret = btrfs_search_forward(root, &key, path, 0);
3496                 if (ret) {
3497                         if (ret > 0)
3498                                 ret = 0;
3499                         break;
3500                 }
3501
3502                 if (key.type != BTRFS_ROOT_ITEM_KEY ||
3503                     (key.objectid < BTRFS_FIRST_FREE_OBJECTID &&
3504                      key.objectid != BTRFS_FS_TREE_OBJECTID) ||
3505                     key.objectid > BTRFS_LAST_FREE_OBJECTID)
3506                         goto skip;
3507
3508                 eb = path->nodes[0];
3509                 slot = path->slots[0];
3510                 item_size = btrfs_item_size_nr(eb, slot);
3511                 if (item_size < sizeof(root_item))
3512                         goto skip;
3513
3514                 read_extent_buffer(eb, &root_item,
3515                                    btrfs_item_ptr_offset(eb, slot),
3516                                    (int)sizeof(root_item));
3517                 if (btrfs_root_refs(&root_item) == 0)
3518                         goto skip;
3519
3520                 if (!btrfs_is_empty_uuid(root_item.uuid) ||
3521                     !btrfs_is_empty_uuid(root_item.received_uuid)) {
3522                         if (trans)
3523                                 goto update_tree;
3524
3525                         btrfs_release_path(path);
3526                         /*
3527                          * 1 - subvol uuid item
3528                          * 1 - received_subvol uuid item
3529                          */
3530                         trans = btrfs_start_transaction(fs_info->uuid_root, 2);
3531                         if (IS_ERR(trans)) {
3532                                 ret = PTR_ERR(trans);
3533                                 break;
3534                         }
3535                         continue;
3536                 } else {
3537                         goto skip;
3538                 }
3539 update_tree:
3540                 if (!btrfs_is_empty_uuid(root_item.uuid)) {
3541                         ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
3542                                                   root_item.uuid,
3543                                                   BTRFS_UUID_KEY_SUBVOL,
3544                                                   key.objectid);
3545                         if (ret < 0) {
3546                                 btrfs_warn(fs_info, "uuid_tree_add failed %d",
3547                                         ret);
3548                                 break;
3549                         }
3550                 }
3551
3552                 if (!btrfs_is_empty_uuid(root_item.received_uuid)) {
3553                         ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
3554                                                   root_item.received_uuid,
3555                                                  BTRFS_UUID_KEY_RECEIVED_SUBVOL,
3556                                                   key.objectid);
3557                         if (ret < 0) {
3558                                 btrfs_warn(fs_info, "uuid_tree_add failed %d",
3559                                         ret);
3560                                 break;
3561                         }
3562                 }
3563
3564 skip:
3565                 if (trans) {
3566                         ret = btrfs_end_transaction(trans, fs_info->uuid_root);
3567                         trans = NULL;
3568                         if (ret)
3569                                 break;
3570                 }
3571
3572                 btrfs_release_path(path);
3573                 if (key.offset < (u64)-1) {
3574                         key.offset++;
3575                 } else if (key.type < BTRFS_ROOT_ITEM_KEY) {
3576                         key.offset = 0;
3577                         key.type = BTRFS_ROOT_ITEM_KEY;
3578                 } else if (key.objectid < (u64)-1) {
3579                         key.offset = 0;
3580                         key.type = BTRFS_ROOT_ITEM_KEY;
3581                         key.objectid++;
3582                 } else {
3583                         break;
3584                 }
3585                 cond_resched();
3586         }
3587
3588 out:
3589         btrfs_free_path(path);
3590         if (trans && !IS_ERR(trans))
3591                 btrfs_end_transaction(trans, fs_info->uuid_root);
3592         if (ret)
3593                 btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret);
3594         else
3595                 fs_info->update_uuid_tree_gen = 1;
3596         up(&fs_info->uuid_tree_rescan_sem);
3597         return 0;
3598 }
3599
3600 /*
3601  * Callback for btrfs_uuid_tree_iterate().
3602  * returns:
3603  * 0    check succeeded, the entry is not outdated.
3604  * < 0  if an error occured.
3605  * > 0  if the check failed, which means the caller shall remove the entry.
3606  */
3607 static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info *fs_info,
3608                                        u8 *uuid, u8 type, u64 subid)
3609 {
3610         struct btrfs_key key;
3611         int ret = 0;
3612         struct btrfs_root *subvol_root;
3613
3614         if (type != BTRFS_UUID_KEY_SUBVOL &&
3615             type != BTRFS_UUID_KEY_RECEIVED_SUBVOL)
3616                 goto out;
3617
3618         key.objectid = subid;
3619         key.type = BTRFS_ROOT_ITEM_KEY;
3620         key.offset = (u64)-1;
3621         subvol_root = btrfs_read_fs_root_no_name(fs_info, &key);
3622         if (IS_ERR(subvol_root)) {
3623                 ret = PTR_ERR(subvol_root);
3624                 if (ret == -ENOENT)
3625                         ret = 1;
3626                 goto out;
3627         }
3628
3629         switch (type) {
3630         case BTRFS_UUID_KEY_SUBVOL:
3631                 if (memcmp(uuid, subvol_root->root_item.uuid, BTRFS_UUID_SIZE))
3632                         ret = 1;
3633                 break;
3634         case BTRFS_UUID_KEY_RECEIVED_SUBVOL:
3635                 if (memcmp(uuid, subvol_root->root_item.received_uuid,
3636                            BTRFS_UUID_SIZE))
3637                         ret = 1;
3638                 break;
3639         }
3640
3641 out:
3642         return ret;
3643 }
3644
3645 static int btrfs_uuid_rescan_kthread(void *data)
3646 {
3647         struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data;
3648         int ret;
3649
3650         /*
3651          * 1st step is to iterate through the existing UUID tree and
3652          * to delete all entries that contain outdated data.
3653          * 2nd step is to add all missing entries to the UUID tree.
3654          */
3655         ret = btrfs_uuid_tree_iterate(fs_info, btrfs_check_uuid_tree_entry);
3656         if (ret < 0) {
3657                 btrfs_warn(fs_info, "iterating uuid_tree failed %d", ret);
3658                 up(&fs_info->uuid_tree_rescan_sem);
3659                 return ret;
3660         }
3661         return btrfs_uuid_scan_kthread(data);
3662 }
3663
3664 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info)
3665 {
3666         struct btrfs_trans_handle *trans;
3667         struct btrfs_root *tree_root = fs_info->tree_root;
3668         struct btrfs_root *uuid_root;
3669         struct task_struct *task;
3670         int ret;
3671
3672         /*
3673          * 1 - root node
3674          * 1 - root item
3675          */
3676         trans = btrfs_start_transaction(tree_root, 2);
3677         if (IS_ERR(trans))
3678                 return PTR_ERR(trans);
3679
3680         uuid_root = btrfs_create_tree(trans, fs_info,
3681                                       BTRFS_UUID_TREE_OBJECTID);
3682         if (IS_ERR(uuid_root)) {
3683                 btrfs_abort_transaction(trans, tree_root,
3684                                         PTR_ERR(uuid_root));
3685                 return PTR_ERR(uuid_root);
3686         }
3687
3688         fs_info->uuid_root = uuid_root;
3689
3690         ret = btrfs_commit_transaction(trans, tree_root);
3691         if (ret)
3692                 return ret;
3693
3694         down(&fs_info->uuid_tree_rescan_sem);
3695         task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid");
3696         if (IS_ERR(task)) {
3697                 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
3698                 btrfs_warn(fs_info, "failed to start uuid_scan task");
3699                 up(&fs_info->uuid_tree_rescan_sem);
3700                 return PTR_ERR(task);
3701         }
3702
3703         return 0;
3704 }
3705
3706 int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
3707 {
3708         struct task_struct *task;
3709
3710         down(&fs_info->uuid_tree_rescan_sem);
3711         task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
3712         if (IS_ERR(task)) {
3713                 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
3714                 btrfs_warn(fs_info, "failed to start uuid_rescan task");
3715                 up(&fs_info->uuid_tree_rescan_sem);
3716                 return PTR_ERR(task);
3717         }
3718
3719         return 0;
3720 }
3721
3722 /*
3723  * shrinking a device means finding all of the device extents past
3724  * the new size, and then following the back refs to the chunks.
3725  * The chunk relocation code actually frees the device extent
3726  */
3727 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
3728 {
3729         struct btrfs_trans_handle *trans;
3730         struct btrfs_root *root = device->dev_root;
3731         struct btrfs_dev_extent *dev_extent = NULL;
3732         struct btrfs_path *path;
3733         u64 length;
3734         u64 chunk_tree;
3735         u64 chunk_objectid;
3736         u64 chunk_offset;
3737         int ret;
3738         int slot;
3739         int failed = 0;
3740         bool retried = false;
3741         struct extent_buffer *l;
3742         struct btrfs_key key;
3743         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
3744         u64 old_total = btrfs_super_total_bytes(super_copy);
3745         u64 old_size = device->total_bytes;
3746         u64 diff = device->total_bytes - new_size;
3747
3748         if (device->is_tgtdev_for_dev_replace)
3749                 return -EINVAL;
3750
3751         path = btrfs_alloc_path();
3752         if (!path)
3753                 return -ENOMEM;
3754
3755         path->reada = 2;
3756
3757         lock_chunks(root);
3758
3759         device->total_bytes = new_size;
3760         if (device->writeable) {
3761                 device->fs_devices->total_rw_bytes -= diff;
3762                 spin_lock(&root->fs_info->free_chunk_lock);
3763                 root->fs_info->free_chunk_space -= diff;
3764                 spin_unlock(&root->fs_info->free_chunk_lock);
3765         }
3766         unlock_chunks(root);
3767
3768 again:
3769         key.objectid = device->devid;
3770         key.offset = (u64)-1;
3771         key.type = BTRFS_DEV_EXTENT_KEY;
3772
3773         do {
3774                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3775                 if (ret < 0)
3776                         goto done;
3777
3778                 ret = btrfs_previous_item(root, path, 0, key.type);
3779                 if (ret < 0)
3780                         goto done;
3781                 if (ret) {
3782                         ret = 0;
3783                         btrfs_release_path(path);
3784                         break;
3785                 }
3786
3787                 l = path->nodes[0];
3788                 slot = path->slots[0];
3789                 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
3790
3791                 if (key.objectid != device->devid) {
3792                         btrfs_release_path(path);
3793                         break;
3794                 }
3795
3796                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
3797                 length = btrfs_dev_extent_length(l, dev_extent);
3798
3799                 if (key.offset + length <= new_size) {
3800                         btrfs_release_path(path);
3801                         break;
3802                 }
3803
3804                 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
3805                 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
3806                 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
3807                 btrfs_release_path(path);
3808
3809                 ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
3810                                            chunk_offset);
3811                 if (ret && ret != -ENOSPC)
3812                         goto done;
3813                 if (ret == -ENOSPC)
3814                         failed++;
3815         } while (key.offset-- > 0);
3816
3817         if (failed && !retried) {
3818                 failed = 0;
3819                 retried = true;
3820                 goto again;
3821         } else if (failed && retried) {
3822                 ret = -ENOSPC;
3823                 lock_chunks(root);
3824
3825                 device->total_bytes = old_size;
3826                 if (device->writeable)
3827                         device->fs_devices->total_rw_bytes += diff;
3828                 spin_lock(&root->fs_info->free_chunk_lock);
3829                 root->fs_info->free_chunk_space += diff;
3830                 spin_unlock(&root->fs_info->free_chunk_lock);
3831                 unlock_chunks(root);
3832                 goto done;
3833         }
3834
3835         /* Shrinking succeeded, else we would be at "done". */
3836         trans = btrfs_start_transaction(root, 0);
3837         if (IS_ERR(trans)) {
3838                 ret = PTR_ERR(trans);
3839                 goto done;
3840         }
3841
3842         lock_chunks(root);
3843
3844         device->disk_total_bytes = new_size;
3845         /* Now btrfs_update_device() will change the on-disk size. */
3846         ret = btrfs_update_device(trans, device);
3847         if (ret) {
3848                 unlock_chunks(root);
3849                 btrfs_end_transaction(trans, root);
3850                 goto done;
3851         }
3852         WARN_ON(diff > old_total);
3853         btrfs_set_super_total_bytes(super_copy, old_total - diff);
3854         unlock_chunks(root);
3855         btrfs_end_transaction(trans, root);
3856 done:
3857         btrfs_free_path(path);
3858         return ret;
3859 }
3860
3861 static int btrfs_add_system_chunk(struct btrfs_root *root,
3862                            struct btrfs_key *key,
3863                            struct btrfs_chunk *chunk, int item_size)
3864 {
3865         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
3866         struct btrfs_disk_key disk_key;
3867         u32 array_size;
3868         u8 *ptr;
3869
3870         array_size = btrfs_super_sys_array_size(super_copy);
3871         if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
3872                 return -EFBIG;
3873
3874         ptr = super_copy->sys_chunk_array + array_size;
3875         btrfs_cpu_key_to_disk(&disk_key, key);
3876         memcpy(ptr, &disk_key, sizeof(disk_key));
3877         ptr += sizeof(disk_key);
3878         memcpy(ptr, chunk, item_size);
3879         item_size += sizeof(disk_key);
3880         btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
3881         return 0;
3882 }
3883
3884 /*
3885  * sort the devices in descending order by max_avail, total_avail
3886  */
3887 static int btrfs_cmp_device_info(const void *a, const void *b)
3888 {
3889         const struct btrfs_device_info *di_a = a;
3890         const struct btrfs_device_info *di_b = b;
3891
3892         if (di_a->max_avail > di_b->max_avail)
3893                 return -1;
3894         if (di_a->max_avail < di_b->max_avail)
3895                 return 1;
3896         if (di_a->total_avail > di_b->total_avail)
3897                 return -1;
3898         if (di_a->total_avail < di_b->total_avail)
3899                 return 1;
3900         return 0;
3901 }
3902
3903 static struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = {
3904         [BTRFS_RAID_RAID10] = {
3905                 .sub_stripes    = 2,
3906                 .dev_stripes    = 1,
3907                 .devs_max       = 0,    /* 0 == as many as possible */
3908                 .devs_min       = 4,
3909                 .devs_increment = 2,
3910                 .ncopies        = 2,
3911         },
3912         [BTRFS_RAID_RAID1] = {
3913                 .sub_stripes    = 1,
3914                 .dev_stripes    = 1,
3915                 .devs_max       = 2,
3916                 .devs_min       = 2,
3917                 .devs_increment = 2,
3918                 .ncopies        = 2,
3919         },
3920         [BTRFS_RAID_DUP] = {
3921                 .sub_stripes    = 1,
3922                 .dev_stripes    = 2,
3923                 .devs_max       = 1,
3924                 .devs_min       = 1,
3925                 .devs_increment = 1,
3926                 .ncopies        = 2,
3927         },
3928         [BTRFS_RAID_RAID0] = {
3929                 .sub_stripes    = 1,
3930                 .dev_stripes    = 1,
3931                 .devs_max       = 0,
3932                 .devs_min       = 2,
3933                 .devs_increment = 1,
3934                 .ncopies        = 1,
3935         },
3936         [BTRFS_RAID_SINGLE] = {
3937                 .sub_stripes    = 1,
3938                 .dev_stripes    = 1,
3939                 .devs_max       = 1,
3940                 .devs_min       = 1,
3941                 .devs_increment = 1,
3942                 .ncopies        = 1,
3943         },
3944         [BTRFS_RAID_RAID5] = {
3945                 .sub_stripes    = 1,
3946                 .dev_stripes    = 1,
3947                 .devs_max       = 0,
3948                 .devs_min       = 2,
3949                 .devs_increment = 1,
3950                 .ncopies        = 2,
3951         },
3952         [BTRFS_RAID_RAID6] = {
3953                 .sub_stripes    = 1,
3954                 .dev_stripes    = 1,
3955                 .devs_max       = 0,
3956                 .devs_min       = 3,
3957                 .devs_increment = 1,
3958                 .ncopies        = 3,
3959         },
3960 };
3961
3962 static u32 find_raid56_stripe_len(u32 data_devices, u32 dev_stripe_target)
3963 {
3964         /* TODO allow them to set a preferred stripe size */
3965         return 64 * 1024;
3966 }
3967
3968 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
3969 {
3970         if (!(type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)))
3971                 return;
3972
3973         btrfs_set_fs_incompat(info, RAID56);
3974 }
3975
3976 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
3977                                struct btrfs_root *extent_root, u64 start,
3978                                u64 type)
3979 {
3980         struct btrfs_fs_info *info = extent_root->fs_info;
3981         struct btrfs_fs_devices *fs_devices = info->fs_devices;
3982         struct list_head *cur;
3983         struct map_lookup *map = NULL;
3984         struct extent_map_tree *em_tree;
3985         struct extent_map *em;
3986         struct btrfs_device_info *devices_info = NULL;
3987         u64 total_avail;
3988         int num_stripes;        /* total number of stripes to allocate */
3989         int data_stripes;       /* number of stripes that count for
3990                                    block group size */
3991         int sub_stripes;        /* sub_stripes info for map */
3992         int dev_stripes;        /* stripes per dev */
3993         int devs_max;           /* max devs to use */
3994         int devs_min;           /* min devs needed */
3995         int devs_increment;     /* ndevs has to be a multiple of this */
3996         int ncopies;            /* how many copies to data has */
3997         int ret;
3998         u64 max_stripe_size;
3999         u64 max_chunk_size;
4000         u64 stripe_size;
4001         u64 num_bytes;
4002         u64 raid_stripe_len = BTRFS_STRIPE_LEN;
4003         int ndevs;
4004         int i;
4005         int j;
4006         int index;
4007
4008         BUG_ON(!alloc_profile_is_valid(type, 0));
4009
4010         if (list_empty(&fs_devices->alloc_list))
4011                 return -ENOSPC;
4012
4013         index = __get_raid_index(type);
4014
4015         sub_stripes = btrfs_raid_array[index].sub_stripes;
4016         dev_stripes = btrfs_raid_array[index].dev_stripes;
4017         devs_max = btrfs_raid_array[index].devs_max;
4018         devs_min = btrfs_raid_array[index].devs_min;
4019         devs_increment = btrfs_raid_array[index].devs_increment;
4020         ncopies = btrfs_raid_array[index].ncopies;
4021
4022         if (type & BTRFS_BLOCK_GROUP_DATA) {
4023                 max_stripe_size = 1024 * 1024 * 1024;
4024                 max_chunk_size = 10 * max_stripe_size;
4025         } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
4026                 /* for larger filesystems, use larger metadata chunks */
4027                 if (fs_devices->total_rw_bytes > 50ULL * 1024 * 1024 * 1024)
4028                         max_stripe_size = 1024 * 1024 * 1024;
4029                 else
4030                         max_stripe_size = 256 * 1024 * 1024;
4031                 max_chunk_size = max_stripe_size;
4032         } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
4033                 max_stripe_size = 32 * 1024 * 1024;
4034                 max_chunk_size = 2 * max_stripe_size;
4035         } else {
4036                 btrfs_err(info, "invalid chunk type 0x%llx requested\n",
4037                        type);
4038                 BUG_ON(1);
4039         }
4040
4041         /* we don't want a chunk larger than 10% of writeable space */
4042         max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
4043                              max_chunk_size);
4044
4045         devices_info = kzalloc(sizeof(*devices_info) * fs_devices->rw_devices,
4046                                GFP_NOFS);
4047         if (!devices_info)
4048                 return -ENOMEM;
4049
4050         cur = fs_devices->alloc_list.next;
4051
4052         /*
4053          * in the first pass through the devices list, we gather information
4054          * about the available holes on each device.
4055          */
4056         ndevs = 0;
4057         while (cur != &fs_devices->alloc_list) {
4058                 struct btrfs_device *device;
4059                 u64 max_avail;
4060                 u64 dev_offset;
4061
4062                 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
4063
4064                 cur = cur->next;
4065
4066                 if (!device->writeable) {
4067                         WARN(1, KERN_ERR
4068                                "BTRFS: read-only device in alloc_list\n");
4069                         continue;
4070                 }
4071
4072                 if (!device->in_fs_metadata ||
4073                     device->is_tgtdev_for_dev_replace)
4074                         continue;
4075
4076                 if (device->total_bytes > device->bytes_used)
4077                         total_avail = device->total_bytes - device->bytes_used;
4078                 else
4079                         total_avail = 0;
4080
4081                 /* If there is no space on this device, skip it. */
4082                 if (total_avail == 0)
4083                         continue;
4084
4085                 ret = find_free_dev_extent(trans, device,
4086                                            max_stripe_size * dev_stripes,
4087                                            &dev_offset, &max_avail);
4088                 if (ret && ret != -ENOSPC)
4089                         goto error;
4090
4091                 if (ret == 0)
4092                         max_avail = max_stripe_size * dev_stripes;
4093
4094                 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes)
4095                         continue;
4096
4097                 if (ndevs == fs_devices->rw_devices) {
4098                         WARN(1, "%s: found more than %llu devices\n",
4099                              __func__, fs_devices->rw_devices);
4100                         break;
4101                 }
4102                 devices_info[ndevs].dev_offset = dev_offset;
4103                 devices_info[ndevs].max_avail = max_avail;
4104                 devices_info[ndevs].total_avail = total_avail;
4105                 devices_info[ndevs].dev = device;
4106                 ++ndevs;
4107         }
4108
4109         /*
4110          * now sort the devices by hole size / available space
4111          */
4112         sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
4113              btrfs_cmp_device_info, NULL);
4114
4115         /* round down to number of usable stripes */
4116         ndevs -= ndevs % devs_increment;
4117
4118         if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) {
4119                 ret = -ENOSPC;
4120                 goto error;
4121         }
4122
4123         if (devs_max && ndevs > devs_max)
4124                 ndevs = devs_max;
4125         /*
4126          * the primary goal is to maximize the number of stripes, so use as many
4127          * devices as possible, even if the stripes are not maximum sized.
4128          */
4129         stripe_size = devices_info[ndevs-1].max_avail;
4130         num_stripes = ndevs * dev_stripes;
4131
4132         /*
4133          * this will have to be fixed for RAID1 and RAID10 over
4134          * more drives
4135          */
4136         data_stripes = num_stripes / ncopies;
4137
4138         if (type & BTRFS_BLOCK_GROUP_RAID5) {
4139                 raid_stripe_len = find_raid56_stripe_len(ndevs - 1,
4140                                  btrfs_super_stripesize(info->super_copy));
4141                 data_stripes = num_stripes - 1;
4142         }
4143         if (type & BTRFS_BLOCK_GROUP_RAID6) {
4144                 raid_stripe_len = find_raid56_stripe_len(ndevs - 2,
4145                                  btrfs_super_stripesize(info->super_copy));
4146                 data_stripes = num_stripes - 2;
4147         }
4148
4149         /*
4150          * Use the number of data stripes to figure out how big this chunk
4151          * is really going to be in terms of logical address space,
4152          * and compare that answer with the max chunk size
4153          */
4154         if (stripe_size * data_stripes > max_chunk_size) {
4155                 u64 mask = (1ULL << 24) - 1;
4156                 stripe_size = max_chunk_size;
4157                 do_div(stripe_size, data_stripes);
4158
4159                 /* bump the answer up to a 16MB boundary */
4160                 stripe_size = (stripe_size + mask) & ~mask;
4161
4162                 /* but don't go higher than the limits we found
4163                  * while searching for free extents
4164                  */
4165                 if (stripe_size > devices_info[ndevs-1].max_avail)
4166                         stripe_size = devices_info[ndevs-1].max_avail;
4167         }
4168
4169         do_div(stripe_size, dev_stripes);
4170
4171         /* align to BTRFS_STRIPE_LEN */
4172         do_div(stripe_size, raid_stripe_len);
4173         stripe_size *= raid_stripe_len;
4174
4175         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
4176         if (!map) {
4177                 ret = -ENOMEM;
4178                 goto error;
4179         }
4180         map->num_stripes = num_stripes;
4181
4182         for (i = 0; i < ndevs; ++i) {
4183                 for (j = 0; j < dev_stripes; ++j) {
4184                         int s = i * dev_stripes + j;
4185                         map->stripes[s].dev = devices_info[i].dev;
4186                         map->stripes[s].physical = devices_info[i].dev_offset +
4187                                                    j * stripe_size;
4188                 }
4189         }
4190         map->sector_size = extent_root->sectorsize;
4191         map->stripe_len = raid_stripe_len;
4192         map->io_align = raid_stripe_len;
4193         map->io_width = raid_stripe_len;
4194         map->type = type;
4195         map->sub_stripes = sub_stripes;
4196
4197         num_bytes = stripe_size * data_stripes;
4198
4199         trace_btrfs_chunk_alloc(info->chunk_root, map, start, num_bytes);
4200
4201         em = alloc_extent_map();
4202         if (!em) {
4203                 ret = -ENOMEM;
4204                 goto error;
4205         }
4206         em->bdev = (struct block_device *)map;
4207         em->start = start;
4208         em->len = num_bytes;
4209         em->block_start = 0;
4210         em->block_len = em->len;
4211         em->orig_block_len = stripe_size;
4212
4213         em_tree = &extent_root->fs_info->mapping_tree.map_tree;
4214         write_lock(&em_tree->lock);
4215         ret = add_extent_mapping(em_tree, em, 0);
4216         if (!ret) {
4217                 list_add_tail(&em->list, &trans->transaction->pending_chunks);
4218                 atomic_inc(&em->refs);
4219         }
4220         write_unlock(&em_tree->lock);
4221         if (ret) {
4222                 free_extent_map(em);
4223                 goto error;
4224         }
4225
4226         ret = btrfs_make_block_group(trans, extent_root, 0, type,
4227                                      BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4228                                      start, num_bytes);
4229         if (ret)
4230                 goto error_del_extent;
4231
4232         free_extent_map(em);
4233         check_raid56_incompat_flag(extent_root->fs_info, type);
4234
4235         kfree(devices_info);
4236         return 0;
4237
4238 error_del_extent:
4239         write_lock(&em_tree->lock);
4240         remove_extent_mapping(em_tree, em);
4241         write_unlock(&em_tree->lock);
4242
4243         /* One for our allocation */
4244         free_extent_map(em);
4245         /* One for the tree reference */
4246         free_extent_map(em);
4247 error:
4248         kfree(map);
4249         kfree(devices_info);
4250         return ret;
4251 }
4252
4253 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans,
4254                                 struct btrfs_root *extent_root,
4255                                 u64 chunk_offset, u64 chunk_size)
4256 {
4257         struct btrfs_key key;
4258         struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
4259         struct btrfs_device *device;
4260         struct btrfs_chunk *chunk;
4261         struct btrfs_stripe *stripe;
4262         struct extent_map_tree *em_tree;
4263         struct extent_map *em;
4264         struct map_lookup *map;
4265         size_t item_size;
4266         u64 dev_offset;
4267         u64 stripe_size;
4268         int i = 0;
4269         int ret;
4270
4271         em_tree = &extent_root->fs_info->mapping_tree.map_tree;
4272         read_lock(&em_tree->lock);
4273         em = lookup_extent_mapping(em_tree, chunk_offset, chunk_size);
4274         read_unlock(&em_tree->lock);
4275
4276         if (!em) {
4277                 btrfs_crit(extent_root->fs_info, "unable to find logical "
4278                            "%Lu len %Lu", chunk_offset, chunk_size);
4279                 return -EINVAL;
4280         }
4281
4282         if (em->start != chunk_offset || em->len != chunk_size) {
4283                 btrfs_crit(extent_root->fs_info, "found a bad mapping, wanted"
4284                           " %Lu-%Lu, found %Lu-%Lu\n", chunk_offset,
4285                           chunk_size, em->start, em->len);
4286                 free_extent_map(em);
4287                 return -EINVAL;
4288         }
4289
4290         map = (struct map_lookup *)em->bdev;
4291         item_size = btrfs_chunk_item_size(map->num_stripes);
4292         stripe_size = em->orig_block_len;
4293
4294         chunk = kzalloc(item_size, GFP_NOFS);
4295         if (!chunk) {
4296                 ret = -ENOMEM;
4297                 goto out;
4298         }
4299
4300         for (i = 0; i < map->num_stripes; i++) {
4301                 device = map->stripes[i].dev;
4302                 dev_offset = map->stripes[i].physical;
4303
4304                 device->bytes_used += stripe_size;
4305                 ret = btrfs_update_device(trans, device);
4306                 if (ret)
4307                         goto out;
4308                 ret = btrfs_alloc_dev_extent(trans, device,
4309                                              chunk_root->root_key.objectid,
4310                                              BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4311                                              chunk_offset, dev_offset,
4312                                              stripe_size);
4313                 if (ret)
4314                         goto out;
4315         }
4316
4317         spin_lock(&extent_root->fs_info->free_chunk_lock);
4318         extent_root->fs_info->free_chunk_space -= (stripe_size *
4319                                                    map->num_stripes);
4320         spin_unlock(&extent_root->fs_info->free_chunk_lock);
4321
4322         stripe = &chunk->stripe;
4323         for (i = 0; i < map->num_stripes; i++) {
4324                 device = map->stripes[i].dev;
4325                 dev_offset = map->stripes[i].physical;
4326
4327                 btrfs_set_stack_stripe_devid(stripe, device->devid);
4328                 btrfs_set_stack_stripe_offset(stripe, dev_offset);
4329                 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
4330                 stripe++;
4331         }
4332
4333         btrfs_set_stack_chunk_length(chunk, chunk_size);
4334         btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
4335         btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
4336         btrfs_set_stack_chunk_type(chunk, map->type);
4337         btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
4338         btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
4339         btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
4340         btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
4341         btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
4342
4343         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
4344         key.type = BTRFS_CHUNK_ITEM_KEY;
4345         key.offset = chunk_offset;
4346
4347         ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
4348         if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
4349                 /*
4350                  * TODO: Cleanup of inserted chunk root in case of
4351                  * failure.
4352                  */
4353                 ret = btrfs_add_system_chunk(chunk_root, &key, chunk,
4354                                              item_size);
4355         }
4356
4357 out:
4358         kfree(chunk);
4359         free_extent_map(em);
4360         return ret;
4361 }
4362
4363 /*
4364  * Chunk allocation falls into two parts. The first part does works
4365  * that make the new allocated chunk useable, but not do any operation
4366  * that modifies the chunk tree. The second part does the works that
4367  * require modifying the chunk tree. This division is important for the
4368  * bootstrap process of adding storage to a seed btrfs.
4369  */
4370 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4371                       struct btrfs_root *extent_root, u64 type)
4372 {
4373         u64 chunk_offset;
4374
4375         chunk_offset = find_next_chunk(extent_root->fs_info);
4376         return __btrfs_alloc_chunk(trans, extent_root, chunk_offset, type);
4377 }
4378
4379 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
4380                                          struct btrfs_root *root,
4381                                          struct btrfs_device *device)
4382 {
4383         u64 chunk_offset;
4384         u64 sys_chunk_offset;
4385         u64 alloc_profile;
4386         struct btrfs_fs_info *fs_info = root->fs_info;
4387         struct btrfs_root *extent_root = fs_info->extent_root;
4388         int ret;
4389
4390         chunk_offset = find_next_chunk(fs_info);
4391         alloc_profile = btrfs_get_alloc_profile(extent_root, 0);
4392         ret = __btrfs_alloc_chunk(trans, extent_root, chunk_offset,
4393                                   alloc_profile);
4394         if (ret)
4395                 return ret;
4396
4397         sys_chunk_offset = find_next_chunk(root->fs_info);
4398         alloc_profile = btrfs_get_alloc_profile(fs_info->chunk_root, 0);
4399         ret = __btrfs_alloc_chunk(trans, extent_root, sys_chunk_offset,
4400                                   alloc_profile);
4401         if (ret) {
4402                 btrfs_abort_transaction(trans, root, ret);
4403                 goto out;
4404         }
4405
4406         ret = btrfs_add_device(trans, fs_info->chunk_root, device);
4407         if (ret)
4408                 btrfs_abort_transaction(trans, root, ret);
4409 out:
4410         return ret;
4411 }
4412
4413 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
4414 {
4415         struct extent_map *em;
4416         struct map_lookup *map;
4417         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
4418         int readonly = 0;
4419         int i;
4420
4421         read_lock(&map_tree->map_tree.lock);
4422         em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
4423         read_unlock(&map_tree->map_tree.lock);
4424         if (!em)
4425                 return 1;
4426
4427         if (btrfs_test_opt(root, DEGRADED)) {
4428                 free_extent_map(em);
4429                 return 0;
4430         }
4431
4432         map = (struct map_lookup *)em->bdev;
4433         for (i = 0; i < map->num_stripes; i++) {
4434                 if (!map->stripes[i].dev->writeable) {
4435                         readonly = 1;
4436                         break;
4437                 }
4438         }
4439         free_extent_map(em);
4440         return readonly;
4441 }
4442
4443 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
4444 {
4445         extent_map_tree_init(&tree->map_tree);
4446 }
4447
4448 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
4449 {
4450         struct extent_map *em;
4451
4452         while (1) {
4453                 write_lock(&tree->map_tree.lock);
4454                 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
4455                 if (em)
4456                         remove_extent_mapping(&tree->map_tree, em);
4457                 write_unlock(&tree->map_tree.lock);
4458                 if (!em)
4459                         break;
4460                 kfree(em->bdev);
4461                 /* once for us */
4462                 free_extent_map(em);
4463                 /* once for the tree */
4464                 free_extent_map(em);
4465         }
4466 }
4467
4468 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
4469 {
4470         struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
4471         struct extent_map *em;
4472         struct map_lookup *map;
4473         struct extent_map_tree *em_tree = &map_tree->map_tree;
4474         int ret;
4475
4476         read_lock(&em_tree->lock);
4477         em = lookup_extent_mapping(em_tree, logical, len);
4478         read_unlock(&em_tree->lock);
4479
4480         /*
4481          * We could return errors for these cases, but that could get ugly and
4482          * we'd probably do the same thing which is just not do anything else
4483          * and exit, so return 1 so the callers don't try to use other copies.
4484          */
4485         if (!em) {
4486                 btrfs_crit(fs_info, "No mapping for %Lu-%Lu\n", logical,
4487                             logical+len);
4488                 return 1;
4489         }
4490
4491         if (em->start > logical || em->start + em->len < logical) {
4492                 btrfs_crit(fs_info, "Invalid mapping for %Lu-%Lu, got "
4493                             "%Lu-%Lu\n", logical, logical+len, em->start,
4494                             em->start + em->len);
4495                 free_extent_map(em);
4496                 return 1;
4497         }
4498
4499         map = (struct map_lookup *)em->bdev;
4500         if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
4501                 ret = map->num_stripes;
4502         else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
4503                 ret = map->sub_stripes;
4504         else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
4505                 ret = 2;
4506         else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
4507                 ret = 3;
4508         else
4509                 ret = 1;
4510         free_extent_map(em);
4511
4512         btrfs_dev_replace_lock(&fs_info->dev_replace);
4513         if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace))
4514                 ret++;
4515         btrfs_dev_replace_unlock(&fs_info->dev_replace);
4516
4517         return ret;
4518 }
4519
4520 unsigned long btrfs_full_stripe_len(struct btrfs_root *root,
4521                                     struct btrfs_mapping_tree *map_tree,
4522                                     u64 logical)
4523 {
4524         struct extent_map *em;
4525         struct map_lookup *map;
4526         struct extent_map_tree *em_tree = &map_tree->map_tree;
4527         unsigned long len = root->sectorsize;
4528
4529         read_lock(&em_tree->lock);
4530         em = lookup_extent_mapping(em_tree, logical, len);
4531         read_unlock(&em_tree->lock);
4532         BUG_ON(!em);
4533
4534         BUG_ON(em->start > logical || em->start + em->len < logical);
4535         map = (struct map_lookup *)em->bdev;
4536         if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
4537                          BTRFS_BLOCK_GROUP_RAID6)) {
4538                 len = map->stripe_len * nr_data_stripes(map);
4539         }
4540         free_extent_map(em);
4541         return len;
4542 }
4543
4544 int btrfs_is_parity_mirror(struct btrfs_mapping_tree *map_tree,
4545                            u64 logical, u64 len, int mirror_num)
4546 {
4547         struct extent_map *em;
4548         struct map_lookup *map;
4549         struct extent_map_tree *em_tree = &map_tree->map_tree;
4550         int ret = 0;
4551
4552         read_lock(&em_tree->lock);
4553         em = lookup_extent_mapping(em_tree, logical, len);
4554         read_unlock(&em_tree->lock);
4555         BUG_ON(!em);
4556
4557         BUG_ON(em->start > logical || em->start + em->len < logical);
4558         map = (struct map_lookup *)em->bdev;
4559         if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
4560                          BTRFS_BLOCK_GROUP_RAID6))
4561                 ret = 1;
4562         free_extent_map(em);
4563         return ret;
4564 }
4565
4566 static int find_live_mirror(struct btrfs_fs_info *fs_info,
4567                             struct map_lookup *map, int first, int num,
4568                             int optimal, int dev_replace_is_ongoing)
4569 {
4570         int i;
4571         int tolerance;
4572         struct btrfs_device *srcdev;
4573
4574         if (dev_replace_is_ongoing &&
4575             fs_info->dev_replace.cont_reading_from_srcdev_mode ==
4576              BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
4577                 srcdev = fs_info->dev_replace.srcdev;
4578         else
4579                 srcdev = NULL;
4580
4581         /*
4582          * try to avoid the drive that is the source drive for a
4583          * dev-replace procedure, only choose it if no other non-missing
4584          * mirror is available
4585          */
4586         for (tolerance = 0; tolerance < 2; tolerance++) {
4587                 if (map->stripes[optimal].dev->bdev &&
4588                     (tolerance || map->stripes[optimal].dev != srcdev))
4589                         return optimal;
4590                 for (i = first; i < first + num; i++) {
4591                         if (map->stripes[i].dev->bdev &&
4592                             (tolerance || map->stripes[i].dev != srcdev))
4593                                 return i;
4594                 }
4595         }
4596
4597         /* we couldn't find one that doesn't fail.  Just return something
4598          * and the io error handling code will clean up eventually
4599          */
4600         return optimal;
4601 }
4602
4603 static inline int parity_smaller(u64 a, u64 b)
4604 {
4605         return a > b;
4606 }
4607
4608 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
4609 static void sort_parity_stripes(struct btrfs_bio *bbio, u64 *raid_map)
4610 {
4611         struct btrfs_bio_stripe s;
4612         int i;
4613         u64 l;
4614         int again = 1;
4615
4616         while (again) {
4617                 again = 0;
4618                 for (i = 0; i < bbio->num_stripes - 1; i++) {
4619                         if (parity_smaller(raid_map[i], raid_map[i+1])) {
4620                                 s = bbio->stripes[i];
4621                                 l = raid_map[i];
4622                                 bbio->stripes[i] = bbio->stripes[i+1];
4623                                 raid_map[i] = raid_map[i+1];
4624                                 bbio->stripes[i+1] = s;
4625                                 raid_map[i+1] = l;
4626                                 again = 1;
4627                         }
4628                 }
4629         }
4630 }
4631
4632 static int __btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
4633                              u64 logical, u64 *length,
4634                              struct btrfs_bio **bbio_ret,
4635                              int mirror_num, u64 **raid_map_ret)
4636 {
4637         struct extent_map *em;
4638         struct map_lookup *map;
4639         struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
4640         struct extent_map_tree *em_tree = &map_tree->map_tree;
4641         u64 offset;
4642         u64 stripe_offset;
4643         u64 stripe_end_offset;
4644         u64 stripe_nr;
4645         u64 stripe_nr_orig;
4646         u64 stripe_nr_end;
4647         u64 stripe_len;
4648         u64 *raid_map = NULL;
4649         int stripe_index;
4650         int i;
4651         int ret = 0;
4652         int num_stripes;
4653         int max_errors = 0;
4654         struct btrfs_bio *bbio = NULL;
4655         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
4656         int dev_replace_is_ongoing = 0;
4657         int num_alloc_stripes;
4658         int patch_the_first_stripe_for_dev_replace = 0;
4659         u64 physical_to_patch_in_first_stripe = 0;
4660         u64 raid56_full_stripe_start = (u64)-1;
4661
4662         read_lock(&em_tree->lock);
4663         em = lookup_extent_mapping(em_tree, logical, *length);
4664         read_unlock(&em_tree->lock);
4665
4666         if (!em) {
4667                 btrfs_crit(fs_info, "unable to find logical %llu len %llu",
4668                         logical, *length);
4669                 return -EINVAL;
4670         }
4671
4672         if (em->start > logical || em->start + em->len < logical) {
4673                 btrfs_crit(fs_info, "found a bad mapping, wanted %Lu, "
4674                            "found %Lu-%Lu\n", logical, em->start,
4675                            em->start + em->len);
4676                 free_extent_map(em);
4677                 return -EINVAL;
4678         }
4679
4680         map = (struct map_lookup *)em->bdev;
4681         offset = logical - em->start;
4682
4683         stripe_len = map->stripe_len;
4684         stripe_nr = offset;
4685         /*
4686          * stripe_nr counts the total number of stripes we have to stride
4687          * to get to this block
4688          */
4689         do_div(stripe_nr, stripe_len);
4690
4691         stripe_offset = stripe_nr * stripe_len;
4692         BUG_ON(offset < stripe_offset);
4693
4694         /* stripe_offset is the offset of this block in its stripe*/
4695         stripe_offset = offset - stripe_offset;
4696
4697         /* if we're here for raid56, we need to know the stripe aligned start */
4698         if (map->type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)) {
4699                 unsigned long full_stripe_len = stripe_len * nr_data_stripes(map);
4700                 raid56_full_stripe_start = offset;
4701
4702                 /* allow a write of a full stripe, but make sure we don't
4703                  * allow straddling of stripes
4704                  */
4705                 do_div(raid56_full_stripe_start, full_stripe_len);
4706                 raid56_full_stripe_start *= full_stripe_len;
4707         }
4708
4709         if (rw & REQ_DISCARD) {
4710                 /* we don't discard raid56 yet */
4711                 if (map->type &
4712                     (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)) {
4713                         ret = -EOPNOTSUPP;
4714                         goto out;
4715                 }
4716                 *length = min_t(u64, em->len - offset, *length);
4717         } else if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
4718                 u64 max_len;
4719                 /* For writes to RAID[56], allow a full stripeset across all disks.
4720                    For other RAID types and for RAID[56] reads, just allow a single
4721                    stripe (on a single disk). */
4722                 if (map->type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6) &&
4723                     (rw & REQ_WRITE)) {
4724                         max_len = stripe_len * nr_data_stripes(map) -
4725                                 (offset - raid56_full_stripe_start);
4726                 } else {
4727                         /* we limit the length of each bio to what fits in a stripe */
4728                         max_len = stripe_len - stripe_offset;
4729                 }
4730                 *length = min_t(u64, em->len - offset, max_len);
4731         } else {
4732                 *length = em->len - offset;
4733         }
4734
4735         /* This is for when we're called from btrfs_merge_bio_hook() and all
4736            it cares about is the length */
4737         if (!bbio_ret)
4738                 goto out;
4739
4740         btrfs_dev_replace_lock(dev_replace);
4741         dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
4742         if (!dev_replace_is_ongoing)
4743                 btrfs_dev_replace_unlock(dev_replace);
4744
4745         if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
4746             !(rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) &&
4747             dev_replace->tgtdev != NULL) {
4748                 /*
4749                  * in dev-replace case, for repair case (that's the only
4750                  * case where the mirror is selected explicitly when
4751                  * calling btrfs_map_block), blocks left of the left cursor
4752                  * can also be read from the target drive.
4753                  * For REQ_GET_READ_MIRRORS, the target drive is added as
4754                  * the last one to the array of stripes. For READ, it also
4755                  * needs to be supported using the same mirror number.
4756                  * If the requested block is not left of the left cursor,
4757                  * EIO is returned. This can happen because btrfs_num_copies()
4758                  * returns one more in the dev-replace case.
4759                  */
4760                 u64 tmp_length = *length;
4761                 struct btrfs_bio *tmp_bbio = NULL;
4762                 int tmp_num_stripes;
4763                 u64 srcdev_devid = dev_replace->srcdev->devid;
4764                 int index_srcdev = 0;
4765                 int found = 0;
4766                 u64 physical_of_found = 0;
4767
4768                 ret = __btrfs_map_block(fs_info, REQ_GET_READ_MIRRORS,
4769                              logical, &tmp_length, &tmp_bbio, 0, NULL);
4770                 if (ret) {
4771                         WARN_ON(tmp_bbio != NULL);
4772                         goto out;
4773                 }
4774
4775                 tmp_num_stripes = tmp_bbio->num_stripes;
4776                 if (mirror_num > tmp_num_stripes) {
4777                         /*
4778                          * REQ_GET_READ_MIRRORS does not contain this
4779                          * mirror, that means that the requested area
4780                          * is not left of the left cursor
4781                          */
4782                         ret = -EIO;
4783                         kfree(tmp_bbio);
4784                         goto out;
4785                 }
4786
4787                 /*
4788                  * process the rest of the function using the mirror_num
4789                  * of the source drive. Therefore look it up first.
4790                  * At the end, patch the device pointer to the one of the
4791                  * target drive.
4792                  */
4793                 for (i = 0; i < tmp_num_stripes; i++) {
4794                         if (tmp_bbio->stripes[i].dev->devid == srcdev_devid) {
4795                                 /*
4796                                  * In case of DUP, in order to keep it
4797                                  * simple, only add the mirror with the
4798                                  * lowest physical address
4799                                  */
4800                                 if (found &&
4801                                     physical_of_found <=
4802                                      tmp_bbio->stripes[i].physical)
4803                                         continue;
4804                                 index_srcdev = i;
4805                                 found = 1;
4806                                 physical_of_found =
4807                                         tmp_bbio->stripes[i].physical;
4808                         }
4809                 }
4810
4811                 if (found) {
4812                         mirror_num = index_srcdev + 1;
4813                         patch_the_first_stripe_for_dev_replace = 1;
4814                         physical_to_patch_in_first_stripe = physical_of_found;
4815                 } else {
4816                         WARN_ON(1);
4817                         ret = -EIO;
4818                         kfree(tmp_bbio);
4819                         goto out;
4820                 }
4821
4822                 kfree(tmp_bbio);
4823         } else if (mirror_num > map->num_stripes) {
4824                 mirror_num = 0;
4825         }
4826
4827         num_stripes = 1;
4828         stripe_index = 0;
4829         stripe_nr_orig = stripe_nr;
4830         stripe_nr_end = ALIGN(offset + *length, map->stripe_len);
4831         do_div(stripe_nr_end, map->stripe_len);
4832         stripe_end_offset = stripe_nr_end * map->stripe_len -
4833                             (offset + *length);
4834
4835         if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
4836                 if (rw & REQ_DISCARD)
4837                         num_stripes = min_t(u64, map->num_stripes,
4838                                             stripe_nr_end - stripe_nr_orig);
4839                 stripe_index = do_div(stripe_nr, map->num_stripes);
4840         } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
4841                 if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS))
4842                         num_stripes = map->num_stripes;
4843                 else if (mirror_num)
4844                         stripe_index = mirror_num - 1;
4845                 else {
4846                         stripe_index = find_live_mirror(fs_info, map, 0,
4847                                             map->num_stripes,
4848                                             current->pid % map->num_stripes,
4849                                             dev_replace_is_ongoing);
4850                         mirror_num = stripe_index + 1;
4851                 }
4852
4853         } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
4854                 if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) {
4855                         num_stripes = map->num_stripes;
4856                 } else if (mirror_num) {
4857                         stripe_index = mirror_num - 1;
4858                 } else {
4859                         mirror_num = 1;
4860                 }
4861
4862         } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
4863                 int factor = map->num_stripes / map->sub_stripes;
4864
4865                 stripe_index = do_div(stripe_nr, factor);
4866                 stripe_index *= map->sub_stripes;
4867
4868                 if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS))
4869                         num_stripes = map->sub_stripes;
4870                 else if (rw & REQ_DISCARD)
4871                         num_stripes = min_t(u64, map->sub_stripes *
4872                                             (stripe_nr_end - stripe_nr_orig),
4873                                             map->num_stripes);
4874                 else if (mirror_num)
4875                         stripe_index += mirror_num - 1;
4876                 else {
4877                         int old_stripe_index = stripe_index;
4878                         stripe_index = find_live_mirror(fs_info, map,
4879                                               stripe_index,
4880                                               map->sub_stripes, stripe_index +
4881                                               current->pid % map->sub_stripes,
4882                                               dev_replace_is_ongoing);
4883                         mirror_num = stripe_index - old_stripe_index + 1;
4884                 }
4885
4886         } else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
4887                                 BTRFS_BLOCK_GROUP_RAID6)) {
4888                 u64 tmp;
4889
4890                 if (bbio_ret && ((rw & REQ_WRITE) || mirror_num > 1)
4891                     && raid_map_ret) {
4892                         int i, rot;
4893
4894                         /* push stripe_nr back to the start of the full stripe */
4895                         stripe_nr = raid56_full_stripe_start;
4896                         do_div(stripe_nr, stripe_len);
4897
4898                         stripe_index = do_div(stripe_nr, nr_data_stripes(map));
4899
4900                         /* RAID[56] write or recovery. Return all stripes */
4901                         num_stripes = map->num_stripes;
4902                         max_errors = nr_parity_stripes(map);
4903
4904                         raid_map = kmalloc_array(num_stripes, sizeof(u64),
4905                                            GFP_NOFS);
4906                         if (!raid_map) {
4907                                 ret = -ENOMEM;
4908                                 goto out;
4909                         }
4910
4911                         /* Work out the disk rotation on this stripe-set */
4912                         tmp = stripe_nr;
4913                         rot = do_div(tmp, num_stripes);
4914
4915                         /* Fill in the logical address of each stripe */
4916                         tmp = stripe_nr * nr_data_stripes(map);
4917                         for (i = 0; i < nr_data_stripes(map); i++)
4918                                 raid_map[(i+rot) % num_stripes] =
4919                                         em->start + (tmp + i) * map->stripe_len;
4920
4921                         raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE;
4922                         if (map->type & BTRFS_BLOCK_GROUP_RAID6)
4923                                 raid_map[(i+rot+1) % num_stripes] =
4924                                         RAID6_Q_STRIPE;
4925
4926                         *length = map->stripe_len;
4927                         stripe_index = 0;
4928                         stripe_offset = 0;
4929                 } else {
4930                         /*
4931                          * Mirror #0 or #1 means the original data block.
4932                          * Mirror #2 is RAID5 parity block.
4933                          * Mirror #3 is RAID6 Q block.
4934                          */
4935                         stripe_index = do_div(stripe_nr, nr_data_stripes(map));
4936                         if (mirror_num > 1)
4937                                 stripe_index = nr_data_stripes(map) +
4938                                                 mirror_num - 2;
4939
4940                         /* We distribute the parity blocks across stripes */
4941                         tmp = stripe_nr + stripe_index;
4942                         stripe_index = do_div(tmp, map->num_stripes);
4943                 }
4944         } else {
4945                 /*
4946                  * after this do_div call, stripe_nr is the number of stripes
4947                  * on this device we have to walk to find the data, and
4948                  * stripe_index is the number of our device in the stripe array
4949                  */
4950                 stripe_index = do_div(stripe_nr, map->num_stripes);
4951                 mirror_num = stripe_index + 1;
4952         }
4953         BUG_ON(stripe_index >= map->num_stripes);
4954
4955         num_alloc_stripes = num_stripes;
4956         if (dev_replace_is_ongoing) {
4957                 if (rw & (REQ_WRITE | REQ_DISCARD))
4958                         num_alloc_stripes <<= 1;
4959                 if (rw & REQ_GET_READ_MIRRORS)
4960                         num_alloc_stripes++;
4961         }
4962         bbio = kzalloc(btrfs_bio_size(num_alloc_stripes), GFP_NOFS);
4963         if (!bbio) {
4964                 kfree(raid_map);
4965                 ret = -ENOMEM;
4966                 goto out;
4967         }
4968         atomic_set(&bbio->error, 0);
4969
4970         if (rw & REQ_DISCARD) {
4971                 int factor = 0;
4972                 int sub_stripes = 0;
4973                 u64 stripes_per_dev = 0;
4974                 u32 remaining_stripes = 0;
4975                 u32 last_stripe = 0;
4976
4977                 if (map->type &
4978                     (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID10)) {
4979                         if (map->type & BTRFS_BLOCK_GROUP_RAID0)
4980                                 sub_stripes = 1;
4981                         else
4982                                 sub_stripes = map->sub_stripes;
4983
4984                         factor = map->num_stripes / sub_stripes;
4985                         stripes_per_dev = div_u64_rem(stripe_nr_end -
4986                                                       stripe_nr_orig,
4987                                                       factor,
4988                                                       &remaining_stripes);
4989                         div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
4990                         last_stripe *= sub_stripes;
4991                 }
4992
4993                 for (i = 0; i < num_stripes; i++) {
4994                         bbio->stripes[i].physical =
4995                                 map->stripes[stripe_index].physical +
4996                                 stripe_offset + stripe_nr * map->stripe_len;
4997                         bbio->stripes[i].dev = map->stripes[stripe_index].dev;
4998
4999                         if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5000                                          BTRFS_BLOCK_GROUP_RAID10)) {
5001                                 bbio->stripes[i].length = stripes_per_dev *
5002                                                           map->stripe_len;
5003
5004                                 if (i / sub_stripes < remaining_stripes)
5005                                         bbio->stripes[i].length +=
5006                                                 map->stripe_len;
5007
5008                                 /*
5009                                  * Special for the first stripe and
5010                                  * the last stripe:
5011                                  *
5012                                  * |-------|...|-------|
5013                                  *     |----------|
5014                                  *    off     end_off
5015                                  */
5016                                 if (i < sub_stripes)
5017                                         bbio->stripes[i].length -=
5018                                                 stripe_offset;
5019
5020                                 if (stripe_index >= last_stripe &&
5021                                     stripe_index <= (last_stripe +
5022                                                      sub_stripes - 1))
5023                                         bbio->stripes[i].length -=
5024                                                 stripe_end_offset;
5025
5026                                 if (i == sub_stripes - 1)
5027                                         stripe_offset = 0;
5028                         } else
5029                                 bbio->stripes[i].length = *length;
5030
5031                         stripe_index++;
5032                         if (stripe_index == map->num_stripes) {
5033                                 /* This could only happen for RAID0/10 */
5034                                 stripe_index = 0;
5035                                 stripe_nr++;
5036                         }
5037                 }
5038         } else {
5039                 for (i = 0; i < num_stripes; i++) {
5040                         bbio->stripes[i].physical =
5041                                 map->stripes[stripe_index].physical +
5042                                 stripe_offset +
5043                                 stripe_nr * map->stripe_len;
5044                         bbio->stripes[i].dev =
5045                                 map->stripes[stripe_index].dev;
5046                         stripe_index++;
5047                 }
5048         }
5049
5050         if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS)) {
5051                 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
5052                                  BTRFS_BLOCK_GROUP_RAID10 |
5053                                  BTRFS_BLOCK_GROUP_RAID5 |
5054                                  BTRFS_BLOCK_GROUP_DUP)) {
5055                         max_errors = 1;
5056                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID6) {
5057                         max_errors = 2;
5058                 }
5059         }
5060
5061         if (dev_replace_is_ongoing && (rw & (REQ_WRITE | REQ_DISCARD)) &&
5062             dev_replace->tgtdev != NULL) {
5063                 int index_where_to_add;
5064                 u64 srcdev_devid = dev_replace->srcdev->devid;
5065
5066                 /*
5067                  * duplicate the write operations while the dev replace
5068                  * procedure is running. Since the copying of the old disk
5069                  * to the new disk takes place at run time while the
5070                  * filesystem is mounted writable, the regular write
5071                  * operations to the old disk have to be duplicated to go
5072                  * to the new disk as well.
5073                  * Note that device->missing is handled by the caller, and
5074                  * that the write to the old disk is already set up in the
5075                  * stripes array.
5076                  */
5077                 index_where_to_add = num_stripes;
5078                 for (i = 0; i < num_stripes; i++) {
5079                         if (bbio->stripes[i].dev->devid == srcdev_devid) {
5080                                 /* write to new disk, too */
5081                                 struct btrfs_bio_stripe *new =
5082                                         bbio->stripes + index_where_to_add;
5083                                 struct btrfs_bio_stripe *old =
5084                                         bbio->stripes + i;
5085
5086                                 new->physical = old->physical;
5087                                 new->length = old->length;
5088                                 new->dev = dev_replace->tgtdev;
5089                                 index_where_to_add++;
5090                                 max_errors++;
5091                         }
5092                 }
5093                 num_stripes = index_where_to_add;
5094         } else if (dev_replace_is_ongoing && (rw & REQ_GET_READ_MIRRORS) &&
5095                    dev_replace->tgtdev != NULL) {
5096                 u64 srcdev_devid = dev_replace->srcdev->devid;
5097                 int index_srcdev = 0;
5098                 int found = 0;
5099                 u64 physical_of_found = 0;
5100
5101                 /*
5102                  * During the dev-replace procedure, the target drive can
5103                  * also be used to read data in case it is needed to repair
5104                  * a corrupt block elsewhere. This is possible if the
5105                  * requested area is left of the left cursor. In this area,
5106                  * the target drive is a full copy of the source drive.
5107                  */
5108                 for (i = 0; i < num_stripes; i++) {
5109                         if (bbio->stripes[i].dev->devid == srcdev_devid) {
5110                                 /*
5111                                  * In case of DUP, in order to keep it
5112                                  * simple, only add the mirror with the
5113                                  * lowest physical address
5114                                  */
5115                                 if (found &&
5116                                     physical_of_found <=
5117                                      bbio->stripes[i].physical)
5118                                         continue;
5119                                 index_srcdev = i;
5120                                 found = 1;
5121                                 physical_of_found = bbio->stripes[i].physical;
5122                         }
5123                 }
5124                 if (found) {
5125                         u64 length = map->stripe_len;
5126
5127                         if (physical_of_found + length <=
5128                             dev_replace->cursor_left) {
5129                                 struct btrfs_bio_stripe *tgtdev_stripe =
5130                                         bbio->stripes + num_stripes;
5131
5132                                 tgtdev_stripe->physical = physical_of_found;
5133                                 tgtdev_stripe->length =
5134                                         bbio->stripes[index_srcdev].length;
5135                                 tgtdev_stripe->dev = dev_replace->tgtdev;
5136
5137                                 num_stripes++;
5138                         }
5139                 }
5140         }
5141
5142         *bbio_ret = bbio;
5143         bbio->num_stripes = num_stripes;
5144         bbio->max_errors = max_errors;
5145         bbio->mirror_num = mirror_num;
5146
5147         /*
5148          * this is the case that REQ_READ && dev_replace_is_ongoing &&
5149          * mirror_num == num_stripes + 1 && dev_replace target drive is
5150          * available as a mirror
5151          */
5152         if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) {
5153                 WARN_ON(num_stripes > 1);
5154                 bbio->stripes[0].dev = dev_replace->tgtdev;
5155                 bbio->stripes[0].physical = physical_to_patch_in_first_stripe;
5156                 bbio->mirror_num = map->num_stripes + 1;
5157         }
5158         if (raid_map) {
5159                 sort_parity_stripes(bbio, raid_map);
5160                 *raid_map_ret = raid_map;
5161         }
5162 out:
5163         if (dev_replace_is_ongoing)
5164                 btrfs_dev_replace_unlock(dev_replace);
5165         free_extent_map(em);
5166         return ret;
5167 }
5168
5169 int btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
5170                       u64 logical, u64 *length,
5171                       struct btrfs_bio **bbio_ret, int mirror_num)
5172 {
5173         return __btrfs_map_block(fs_info, rw, logical, length, bbio_ret,
5174                                  mirror_num, NULL);
5175 }
5176
5177 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
5178                      u64 chunk_start, u64 physical, u64 devid,
5179                      u64 **logical, int *naddrs, int *stripe_len)
5180 {
5181         struct extent_map_tree *em_tree = &map_tree->map_tree;
5182         struct extent_map *em;
5183         struct map_lookup *map;
5184         u64 *buf;
5185         u64 bytenr;
5186         u64 length;
5187         u64 stripe_nr;
5188         u64 rmap_len;
5189         int i, j, nr = 0;
5190
5191         read_lock(&em_tree->lock);
5192         em = lookup_extent_mapping(em_tree, chunk_start, 1);
5193         read_unlock(&em_tree->lock);
5194
5195         if (!em) {
5196                 printk(KERN_ERR "BTRFS: couldn't find em for chunk %Lu\n",
5197                        chunk_start);
5198                 return -EIO;
5199         }
5200
5201         if (em->start != chunk_start) {
5202                 printk(KERN_ERR "BTRFS: bad chunk start, em=%Lu, wanted=%Lu\n",
5203                        em->start, chunk_start);
5204                 free_extent_map(em);
5205                 return -EIO;
5206         }
5207         map = (struct map_lookup *)em->bdev;
5208
5209         length = em->len;
5210         rmap_len = map->stripe_len;
5211
5212         if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5213                 do_div(length, map->num_stripes / map->sub_stripes);
5214         else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5215                 do_div(length, map->num_stripes);
5216         else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
5217                               BTRFS_BLOCK_GROUP_RAID6)) {
5218                 do_div(length, nr_data_stripes(map));
5219                 rmap_len = map->stripe_len * nr_data_stripes(map);
5220         }
5221
5222         buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
5223         BUG_ON(!buf); /* -ENOMEM */
5224
5225         for (i = 0; i < map->num_stripes; i++) {
5226                 if (devid && map->stripes[i].dev->devid != devid)
5227                         continue;
5228                 if (map->stripes[i].physical > physical ||
5229                     map->stripes[i].physical + length <= physical)
5230                         continue;
5231
5232                 stripe_nr = physical - map->stripes[i].physical;
5233                 do_div(stripe_nr, map->stripe_len);
5234
5235                 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5236                         stripe_nr = stripe_nr * map->num_stripes + i;
5237                         do_div(stripe_nr, map->sub_stripes);
5238                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5239                         stripe_nr = stripe_nr * map->num_stripes + i;
5240                 } /* else if RAID[56], multiply by nr_data_stripes().
5241                    * Alternatively, just use rmap_len below instead of
5242                    * map->stripe_len */
5243
5244                 bytenr = chunk_start + stripe_nr * rmap_len;
5245                 WARN_ON(nr >= map->num_stripes);
5246                 for (j = 0; j < nr; j++) {
5247                         if (buf[j] == bytenr)
5248                                 break;
5249                 }
5250                 if (j == nr) {
5251                         WARN_ON(nr >= map->num_stripes);
5252                         buf[nr++] = bytenr;
5253                 }
5254         }
5255
5256         *logical = buf;
5257         *naddrs = nr;
5258         *stripe_len = rmap_len;
5259
5260         free_extent_map(em);
5261         return 0;
5262 }
5263
5264 static void btrfs_end_bio(struct bio *bio, int err)
5265 {
5266         struct btrfs_bio *bbio = bio->bi_private;
5267         struct btrfs_device *dev = bbio->stripes[0].dev;
5268         int is_orig_bio = 0;
5269
5270         if (err) {
5271                 atomic_inc(&bbio->error);
5272                 if (err == -EIO || err == -EREMOTEIO) {
5273                         unsigned int stripe_index =
5274                                 btrfs_io_bio(bio)->stripe_index;
5275
5276                         BUG_ON(stripe_index >= bbio->num_stripes);
5277                         dev = bbio->stripes[stripe_index].dev;
5278                         if (dev->bdev) {
5279                                 if (bio->bi_rw & WRITE)
5280                                         btrfs_dev_stat_inc(dev,
5281                                                 BTRFS_DEV_STAT_WRITE_ERRS);
5282                                 else
5283                                         btrfs_dev_stat_inc(dev,
5284                                                 BTRFS_DEV_STAT_READ_ERRS);
5285                                 if ((bio->bi_rw & WRITE_FLUSH) == WRITE_FLUSH)
5286                                         btrfs_dev_stat_inc(dev,
5287                                                 BTRFS_DEV_STAT_FLUSH_ERRS);
5288                                 btrfs_dev_stat_print_on_error(dev);
5289                         }
5290                 }
5291         }
5292
5293         if (bio == bbio->orig_bio)
5294                 is_orig_bio = 1;
5295
5296         btrfs_bio_counter_dec(bbio->fs_info);
5297
5298         if (atomic_dec_and_test(&bbio->stripes_pending)) {
5299                 if (!is_orig_bio) {
5300                         bio_put(bio);
5301                         bio = bbio->orig_bio;
5302                 }
5303                 bio->bi_private = bbio->private;
5304                 bio->bi_end_io = bbio->end_io;
5305                 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
5306                 /* only send an error to the higher layers if it is
5307                  * beyond the tolerance of the btrfs bio
5308                  */
5309                 if (atomic_read(&bbio->error) > bbio->max_errors) {
5310                         err = -EIO;
5311                 } else {
5312                         /*
5313                          * this bio is actually up to date, we didn't
5314                          * go over the max number of errors
5315                          */
5316                         set_bit(BIO_UPTODATE, &bio->bi_flags);
5317                         err = 0;
5318                 }
5319                 kfree(bbio);
5320
5321                 bio_endio(bio, err);
5322         } else if (!is_orig_bio) {
5323                 bio_put(bio);
5324         }
5325 }
5326
5327 /*
5328  * see run_scheduled_bios for a description of why bios are collected for
5329  * async submit.
5330  *
5331  * This will add one bio to the pending list for a device and make sure
5332  * the work struct is scheduled.
5333  */
5334 static noinline void btrfs_schedule_bio(struct btrfs_root *root,
5335                                         struct btrfs_device *device,
5336                                         int rw, struct bio *bio)
5337 {
5338         int should_queue = 1;
5339         struct btrfs_pending_bios *pending_bios;
5340
5341         if (device->missing || !device->bdev) {
5342                 bio_endio(bio, -EIO);
5343                 return;
5344         }
5345
5346         /* don't bother with additional async steps for reads, right now */
5347         if (!(rw & REQ_WRITE)) {
5348                 bio_get(bio);
5349                 btrfsic_submit_bio(rw, bio);
5350                 bio_put(bio);
5351                 return;
5352         }
5353
5354         /*
5355          * nr_async_bios allows us to reliably return congestion to the
5356          * higher layers.  Otherwise, the async bio makes it appear we have
5357          * made progress against dirty pages when we've really just put it
5358          * on a queue for later
5359          */
5360         atomic_inc(&root->fs_info->nr_async_bios);
5361         WARN_ON(bio->bi_next);
5362         bio->bi_next = NULL;
5363         bio->bi_rw |= rw;
5364
5365         spin_lock(&device->io_lock);
5366         if (bio->bi_rw & REQ_SYNC)
5367                 pending_bios = &device->pending_sync_bios;
5368         else
5369                 pending_bios = &device->pending_bios;
5370
5371         if (pending_bios->tail)
5372                 pending_bios->tail->bi_next = bio;
5373
5374         pending_bios->tail = bio;
5375         if (!pending_bios->head)
5376                 pending_bios->head = bio;
5377         if (device->running_pending)
5378                 should_queue = 0;
5379
5380         spin_unlock(&device->io_lock);
5381
5382         if (should_queue)
5383                 btrfs_queue_work(root->fs_info->submit_workers,
5384                                  &device->work);
5385 }
5386
5387 static int bio_size_ok(struct block_device *bdev, struct bio *bio,
5388                        sector_t sector)
5389 {
5390         struct bio_vec *prev;
5391         struct request_queue *q = bdev_get_queue(bdev);
5392         unsigned int max_sectors = queue_max_sectors(q);
5393         struct bvec_merge_data bvm = {
5394                 .bi_bdev = bdev,
5395                 .bi_sector = sector,
5396                 .bi_rw = bio->bi_rw,
5397         };
5398
5399         if (WARN_ON(bio->bi_vcnt == 0))
5400                 return 1;
5401
5402         prev = &bio->bi_io_vec[bio->bi_vcnt - 1];
5403         if (bio_sectors(bio) > max_sectors)
5404                 return 0;
5405
5406         if (!q->merge_bvec_fn)
5407                 return 1;
5408
5409         bvm.bi_size = bio->bi_size - prev->bv_len;
5410         if (q->merge_bvec_fn(q, &bvm, prev) < prev->bv_len)
5411                 return 0;
5412         return 1;
5413 }
5414
5415 static void submit_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio,
5416                               struct bio *bio, u64 physical, int dev_nr,
5417                               int rw, int async)
5418 {
5419         struct btrfs_device *dev = bbio->stripes[dev_nr].dev;
5420
5421         bio->bi_private = bbio;
5422         btrfs_io_bio(bio)->stripe_index = dev_nr;
5423         bio->bi_end_io = btrfs_end_bio;
5424         bio->bi_sector = physical >> 9;
5425 #ifdef DEBUG
5426         {
5427                 struct rcu_string *name;
5428
5429                 rcu_read_lock();
5430                 name = rcu_dereference(dev->name);
5431                 pr_debug("btrfs_map_bio: rw %d, sector=%llu, dev=%lu "
5432                          "(%s id %llu), size=%u\n", rw,
5433                          (u64)bio->bi_sector, (u_long)dev->bdev->bd_dev,
5434                          name->str, dev->devid, bio->bi_size);
5435                 rcu_read_unlock();
5436         }
5437 #endif
5438         bio->bi_bdev = dev->bdev;
5439
5440         btrfs_bio_counter_inc_noblocked(root->fs_info);
5441
5442         if (async)
5443                 btrfs_schedule_bio(root, dev, rw, bio);
5444         else
5445                 btrfsic_submit_bio(rw, bio);
5446 }
5447
5448 static int breakup_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio,
5449                               struct bio *first_bio, struct btrfs_device *dev,
5450                               int dev_nr, int rw, int async)
5451 {
5452         struct bio_vec *bvec = first_bio->bi_io_vec;
5453         struct bio *bio;
5454         int nr_vecs = bio_get_nr_vecs(dev->bdev);
5455         u64 physical = bbio->stripes[dev_nr].physical;
5456
5457 again:
5458         bio = btrfs_bio_alloc(dev->bdev, physical >> 9, nr_vecs, GFP_NOFS);
5459         if (!bio)
5460                 return -ENOMEM;
5461
5462         while (bvec <= (first_bio->bi_io_vec + first_bio->bi_vcnt - 1)) {
5463                 if (bio_add_page(bio, bvec->bv_page, bvec->bv_len,
5464                                  bvec->bv_offset) < bvec->bv_len) {
5465                         u64 len = bio->bi_size;
5466
5467                         atomic_inc(&bbio->stripes_pending);
5468                         submit_stripe_bio(root, bbio, bio, physical, dev_nr,
5469                                           rw, async);
5470                         physical += len;
5471                         goto again;
5472                 }
5473                 bvec++;
5474         }
5475
5476         submit_stripe_bio(root, bbio, bio, physical, dev_nr, rw, async);
5477         return 0;
5478 }
5479
5480 static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical)
5481 {
5482         atomic_inc(&bbio->error);
5483         if (atomic_dec_and_test(&bbio->stripes_pending)) {
5484                 bio->bi_private = bbio->private;
5485                 bio->bi_end_io = bbio->end_io;
5486                 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
5487                 bio->bi_sector = logical >> 9;
5488                 kfree(bbio);
5489                 bio_endio(bio, -EIO);
5490         }
5491 }
5492
5493 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
5494                   int mirror_num, int async_submit)
5495 {
5496         struct btrfs_device *dev;
5497         struct bio *first_bio = bio;
5498         u64 logical = (u64)bio->bi_sector << 9;
5499         u64 length = 0;
5500         u64 map_length;
5501         u64 *raid_map = NULL;
5502         int ret;
5503         int dev_nr = 0;
5504         int total_devs = 1;
5505         struct btrfs_bio *bbio = NULL;
5506
5507         length = bio->bi_size;
5508         map_length = length;
5509
5510         btrfs_bio_counter_inc_blocked(root->fs_info);
5511         ret = __btrfs_map_block(root->fs_info, rw, logical, &map_length, &bbio,
5512                               mirror_num, &raid_map);
5513         if (ret) {
5514                 btrfs_bio_counter_dec(root->fs_info);
5515                 return ret;
5516         }
5517
5518         total_devs = bbio->num_stripes;
5519         bbio->orig_bio = first_bio;
5520         bbio->private = first_bio->bi_private;
5521         bbio->end_io = first_bio->bi_end_io;
5522         bbio->fs_info = root->fs_info;
5523         atomic_set(&bbio->stripes_pending, bbio->num_stripes);
5524
5525         if (raid_map) {
5526                 /* In this case, map_length has been set to the length of
5527                    a single stripe; not the whole write */
5528                 if (rw & WRITE) {
5529                         ret = raid56_parity_write(root, bio, bbio,
5530                                                   raid_map, map_length);
5531                 } else {
5532                         ret = raid56_parity_recover(root, bio, bbio,
5533                                                     raid_map, map_length,
5534                                                     mirror_num);
5535                 }
5536                 /*
5537                  * FIXME, replace dosen't support raid56 yet, please fix
5538                  * it in the future.
5539                  */
5540                 btrfs_bio_counter_dec(root->fs_info);
5541                 return ret;
5542         }
5543
5544         if (map_length < length) {
5545                 btrfs_crit(root->fs_info, "mapping failed logical %llu bio len %llu len %llu",
5546                         logical, length, map_length);
5547                 BUG();
5548         }
5549
5550         while (dev_nr < total_devs) {
5551                 dev = bbio->stripes[dev_nr].dev;
5552                 if (!dev || !dev->bdev || (rw & WRITE && !dev->writeable)) {
5553                         bbio_error(bbio, first_bio, logical);
5554                         dev_nr++;
5555                         continue;
5556                 }
5557
5558                 /*
5559                  * Check and see if we're ok with this bio based on it's size
5560                  * and offset with the given device.
5561                  */
5562                 if (!bio_size_ok(dev->bdev, first_bio,
5563                                  bbio->stripes[dev_nr].physical >> 9)) {
5564                         ret = breakup_stripe_bio(root, bbio, first_bio, dev,
5565                                                  dev_nr, rw, async_submit);
5566                         BUG_ON(ret);
5567                         dev_nr++;
5568                         continue;
5569                 }
5570
5571                 if (dev_nr < total_devs - 1) {
5572                         bio = btrfs_bio_clone(first_bio, GFP_NOFS);
5573                         BUG_ON(!bio); /* -ENOMEM */
5574                 } else {
5575                         bio = first_bio;
5576                 }
5577
5578                 submit_stripe_bio(root, bbio, bio,
5579                                   bbio->stripes[dev_nr].physical, dev_nr, rw,
5580                                   async_submit);
5581                 dev_nr++;
5582         }
5583         btrfs_bio_counter_dec(root->fs_info);
5584         return 0;
5585 }
5586
5587 struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid,
5588                                        u8 *uuid, u8 *fsid)
5589 {
5590         struct btrfs_device *device;
5591         struct btrfs_fs_devices *cur_devices;
5592
5593         cur_devices = fs_info->fs_devices;
5594         while (cur_devices) {
5595                 if (!fsid ||
5596                     !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
5597                         device = __find_device(&cur_devices->devices,
5598                                                devid, uuid);
5599                         if (device)
5600                                 return device;
5601                 }
5602                 cur_devices = cur_devices->seed;
5603         }
5604         return NULL;
5605 }
5606
5607 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
5608                                             u64 devid, u8 *dev_uuid)
5609 {
5610         struct btrfs_device *device;
5611         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
5612
5613         device = btrfs_alloc_device(NULL, &devid, dev_uuid);
5614         if (IS_ERR(device))
5615                 return NULL;
5616
5617         list_add(&device->dev_list, &fs_devices->devices);
5618         device->fs_devices = fs_devices;
5619         fs_devices->num_devices++;
5620
5621         device->missing = 1;
5622         fs_devices->missing_devices++;
5623
5624         return device;
5625 }
5626
5627 /**
5628  * btrfs_alloc_device - allocate struct btrfs_device
5629  * @fs_info:    used only for generating a new devid, can be NULL if
5630  *              devid is provided (i.e. @devid != NULL).
5631  * @devid:      a pointer to devid for this device.  If NULL a new devid
5632  *              is generated.
5633  * @uuid:       a pointer to UUID for this device.  If NULL a new UUID
5634  *              is generated.
5635  *
5636  * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
5637  * on error.  Returned struct is not linked onto any lists and can be
5638  * destroyed with kfree() right away.
5639  */
5640 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
5641                                         const u64 *devid,
5642                                         const u8 *uuid)
5643 {
5644         struct btrfs_device *dev;
5645         u64 tmp;
5646
5647         if (WARN_ON(!devid && !fs_info))
5648                 return ERR_PTR(-EINVAL);
5649
5650         dev = __alloc_device();
5651         if (IS_ERR(dev))
5652                 return dev;
5653
5654         if (devid)
5655                 tmp = *devid;
5656         else {
5657                 int ret;
5658
5659                 ret = find_next_devid(fs_info, &tmp);
5660                 if (ret) {
5661                         kfree(dev);
5662                         return ERR_PTR(ret);
5663                 }
5664         }
5665         dev->devid = tmp;
5666
5667         if (uuid)
5668                 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE);
5669         else
5670                 generate_random_uuid(dev->uuid);
5671
5672         btrfs_init_work(&dev->work, pending_bios_fn, NULL, NULL);
5673
5674         return dev;
5675 }
5676
5677 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
5678                           struct extent_buffer *leaf,
5679                           struct btrfs_chunk *chunk)
5680 {
5681         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
5682         struct map_lookup *map;
5683         struct extent_map *em;
5684         u64 logical;
5685         u64 length;
5686         u64 devid;
5687         u8 uuid[BTRFS_UUID_SIZE];
5688         int num_stripes;
5689         int ret;
5690         int i;
5691
5692         logical = key->offset;
5693         length = btrfs_chunk_length(leaf, chunk);
5694
5695         read_lock(&map_tree->map_tree.lock);
5696         em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
5697         read_unlock(&map_tree->map_tree.lock);
5698
5699         /* already mapped? */
5700         if (em && em->start <= logical && em->start + em->len > logical) {
5701                 free_extent_map(em);
5702                 return 0;
5703         } else if (em) {
5704                 free_extent_map(em);
5705         }
5706
5707         em = alloc_extent_map();
5708         if (!em)
5709                 return -ENOMEM;
5710         num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
5711         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
5712         if (!map) {
5713                 free_extent_map(em);
5714                 return -ENOMEM;
5715         }
5716
5717         em->bdev = (struct block_device *)map;
5718         em->start = logical;
5719         em->len = length;
5720         em->orig_start = 0;
5721         em->block_start = 0;
5722         em->block_len = em->len;
5723
5724         map->num_stripes = num_stripes;
5725         map->io_width = btrfs_chunk_io_width(leaf, chunk);
5726         map->io_align = btrfs_chunk_io_align(leaf, chunk);
5727         map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
5728         map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
5729         map->type = btrfs_chunk_type(leaf, chunk);
5730         map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
5731         for (i = 0; i < num_stripes; i++) {
5732                 map->stripes[i].physical =
5733                         btrfs_stripe_offset_nr(leaf, chunk, i);
5734                 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
5735                 read_extent_buffer(leaf, uuid, (unsigned long)
5736                                    btrfs_stripe_dev_uuid_nr(chunk, i),
5737                                    BTRFS_UUID_SIZE);
5738                 map->stripes[i].dev = btrfs_find_device(root->fs_info, devid,
5739                                                         uuid, NULL);
5740                 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
5741                         kfree(map);
5742                         free_extent_map(em);
5743                         return -EIO;
5744                 }
5745                 if (!map->stripes[i].dev) {
5746                         map->stripes[i].dev =
5747                                 add_missing_dev(root, devid, uuid);
5748                         if (!map->stripes[i].dev) {
5749                                 kfree(map);
5750                                 free_extent_map(em);
5751                                 return -EIO;
5752                         }
5753                 }
5754                 map->stripes[i].dev->in_fs_metadata = 1;
5755         }
5756
5757         write_lock(&map_tree->map_tree.lock);
5758         ret = add_extent_mapping(&map_tree->map_tree, em, 0);
5759         write_unlock(&map_tree->map_tree.lock);
5760         BUG_ON(ret); /* Tree corruption */
5761         free_extent_map(em);
5762
5763         return 0;
5764 }
5765
5766 static void fill_device_from_item(struct extent_buffer *leaf,
5767                                  struct btrfs_dev_item *dev_item,
5768                                  struct btrfs_device *device)
5769 {
5770         unsigned long ptr;
5771
5772         device->devid = btrfs_device_id(leaf, dev_item);
5773         device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
5774         device->total_bytes = device->disk_total_bytes;
5775         device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
5776         device->type = btrfs_device_type(leaf, dev_item);
5777         device->io_align = btrfs_device_io_align(leaf, dev_item);
5778         device->io_width = btrfs_device_io_width(leaf, dev_item);
5779         device->sector_size = btrfs_device_sector_size(leaf, dev_item);
5780         WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
5781         device->is_tgtdev_for_dev_replace = 0;
5782
5783         ptr = btrfs_device_uuid(dev_item);
5784         read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
5785 }
5786
5787 static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
5788 {
5789         struct btrfs_fs_devices *fs_devices;
5790         int ret;
5791
5792         BUG_ON(!mutex_is_locked(&uuid_mutex));
5793
5794         fs_devices = root->fs_info->fs_devices->seed;
5795         while (fs_devices) {
5796                 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
5797                         ret = 0;
5798                         goto out;
5799                 }
5800                 fs_devices = fs_devices->seed;
5801         }
5802
5803         fs_devices = find_fsid(fsid);
5804         if (!fs_devices) {
5805                 ret = -ENOENT;
5806                 goto out;
5807         }
5808
5809         fs_devices = clone_fs_devices(fs_devices);
5810         if (IS_ERR(fs_devices)) {
5811                 ret = PTR_ERR(fs_devices);
5812                 goto out;
5813         }
5814
5815         ret = __btrfs_open_devices(fs_devices, FMODE_READ,
5816                                    root->fs_info->bdev_holder);
5817         if (ret) {
5818                 free_fs_devices(fs_devices);
5819                 goto out;
5820         }
5821
5822         if (!fs_devices->seeding) {
5823                 __btrfs_close_devices(fs_devices);
5824                 free_fs_devices(fs_devices);
5825                 ret = -EINVAL;
5826                 goto out;
5827         }
5828
5829         fs_devices->seed = root->fs_info->fs_devices->seed;
5830         root->fs_info->fs_devices->seed = fs_devices;
5831 out:
5832         return ret;
5833 }
5834
5835 static int read_one_dev(struct btrfs_root *root,
5836                         struct extent_buffer *leaf,
5837                         struct btrfs_dev_item *dev_item)
5838 {
5839         struct btrfs_device *device;
5840         u64 devid;
5841         int ret;
5842         u8 fs_uuid[BTRFS_UUID_SIZE];
5843         u8 dev_uuid[BTRFS_UUID_SIZE];
5844
5845         devid = btrfs_device_id(leaf, dev_item);
5846         read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
5847                            BTRFS_UUID_SIZE);
5848         read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
5849                            BTRFS_UUID_SIZE);
5850
5851         if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
5852                 ret = open_seed_devices(root, fs_uuid);
5853                 if (ret && !btrfs_test_opt(root, DEGRADED))
5854                         return ret;
5855         }
5856
5857         device = btrfs_find_device(root->fs_info, devid, dev_uuid, fs_uuid);
5858         if (!device || !device->bdev) {
5859                 if (!btrfs_test_opt(root, DEGRADED))
5860                         return -EIO;
5861
5862                 if (!device) {
5863                         btrfs_warn(root->fs_info, "devid %llu missing", devid);
5864                         device = add_missing_dev(root, devid, dev_uuid);
5865                         if (!device)
5866                                 return -ENOMEM;
5867                 } else if (!device->missing) {
5868                         /*
5869                          * this happens when a device that was properly setup
5870                          * in the device info lists suddenly goes bad.
5871                          * device->bdev is NULL, and so we have to set
5872                          * device->missing to one here
5873                          */
5874                         root->fs_info->fs_devices->missing_devices++;
5875                         device->missing = 1;
5876                 }
5877         }
5878
5879         if (device->fs_devices != root->fs_info->fs_devices) {
5880                 BUG_ON(device->writeable);
5881                 if (device->generation !=
5882                     btrfs_device_generation(leaf, dev_item))
5883                         return -EINVAL;
5884         }
5885
5886         fill_device_from_item(leaf, dev_item, device);
5887         device->in_fs_metadata = 1;
5888         if (device->writeable && !device->is_tgtdev_for_dev_replace) {
5889                 device->fs_devices->total_rw_bytes += device->total_bytes;
5890                 spin_lock(&root->fs_info->free_chunk_lock);
5891                 root->fs_info->free_chunk_space += device->total_bytes -
5892                         device->bytes_used;
5893                 spin_unlock(&root->fs_info->free_chunk_lock);
5894         }
5895         ret = 0;
5896         return ret;
5897 }
5898
5899 int btrfs_read_sys_array(struct btrfs_root *root)
5900 {
5901         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
5902         struct extent_buffer *sb;
5903         struct btrfs_disk_key *disk_key;
5904         struct btrfs_chunk *chunk;
5905         u8 *ptr;
5906         unsigned long sb_ptr;
5907         int ret = 0;
5908         u32 num_stripes;
5909         u32 array_size;
5910         u32 len = 0;
5911         u32 cur;
5912         struct btrfs_key key;
5913
5914         sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
5915                                           BTRFS_SUPER_INFO_SIZE);
5916         if (!sb)
5917                 return -ENOMEM;
5918         btrfs_set_buffer_uptodate(sb);
5919         btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
5920         /*
5921          * The sb extent buffer is artifical and just used to read the system array.
5922          * btrfs_set_buffer_uptodate() call does not properly mark all it's
5923          * pages up-to-date when the page is larger: extent does not cover the
5924          * whole page and consequently check_page_uptodate does not find all
5925          * the page's extents up-to-date (the hole beyond sb),
5926          * write_extent_buffer then triggers a WARN_ON.
5927          *
5928          * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
5929          * but sb spans only this function. Add an explicit SetPageUptodate call
5930          * to silence the warning eg. on PowerPC 64.
5931          */
5932         if (PAGE_CACHE_SIZE > BTRFS_SUPER_INFO_SIZE)
5933                 SetPageUptodate(sb->pages[0]);
5934
5935         write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
5936         array_size = btrfs_super_sys_array_size(super_copy);
5937
5938         ptr = super_copy->sys_chunk_array;
5939         sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
5940         cur = 0;
5941
5942         while (cur < array_size) {
5943                 disk_key = (struct btrfs_disk_key *)ptr;
5944                 btrfs_disk_key_to_cpu(&key, disk_key);
5945
5946                 len = sizeof(*disk_key); ptr += len;
5947                 sb_ptr += len;
5948                 cur += len;
5949
5950                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
5951                         chunk = (struct btrfs_chunk *)sb_ptr;
5952                         ret = read_one_chunk(root, &key, sb, chunk);
5953                         if (ret)
5954                                 break;
5955                         num_stripes = btrfs_chunk_num_stripes(sb, chunk);
5956                         len = btrfs_chunk_item_size(num_stripes);
5957                 } else {
5958                         ret = -EIO;
5959                         break;
5960                 }
5961                 ptr += len;
5962                 sb_ptr += len;
5963                 cur += len;
5964         }
5965         free_extent_buffer(sb);
5966         return ret;
5967 }
5968
5969 int btrfs_read_chunk_tree(struct btrfs_root *root)
5970 {
5971         struct btrfs_path *path;
5972         struct extent_buffer *leaf;
5973         struct btrfs_key key;
5974         struct btrfs_key found_key;
5975         int ret;
5976         int slot;
5977
5978         root = root->fs_info->chunk_root;
5979
5980         path = btrfs_alloc_path();
5981         if (!path)
5982                 return -ENOMEM;
5983
5984         mutex_lock(&uuid_mutex);
5985         lock_chunks(root);
5986
5987         /*
5988          * Read all device items, and then all the chunk items. All
5989          * device items are found before any chunk item (their object id
5990          * is smaller than the lowest possible object id for a chunk
5991          * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
5992          */
5993         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
5994         key.offset = 0;
5995         key.type = 0;
5996         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5997         if (ret < 0)
5998                 goto error;
5999         while (1) {
6000                 leaf = path->nodes[0];
6001                 slot = path->slots[0];
6002                 if (slot >= btrfs_header_nritems(leaf)) {
6003                         ret = btrfs_next_leaf(root, path);
6004                         if (ret == 0)
6005                                 continue;
6006                         if (ret < 0)
6007                                 goto error;
6008                         break;
6009                 }
6010                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6011                 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
6012                         struct btrfs_dev_item *dev_item;
6013                         dev_item = btrfs_item_ptr(leaf, slot,
6014                                                   struct btrfs_dev_item);
6015                         ret = read_one_dev(root, leaf, dev_item);
6016                         if (ret)
6017                                 goto error;
6018                 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
6019                         struct btrfs_chunk *chunk;
6020                         chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
6021                         ret = read_one_chunk(root, &found_key, leaf, chunk);
6022                         if (ret)
6023                                 goto error;
6024                 }
6025                 path->slots[0]++;
6026         }
6027         ret = 0;
6028 error:
6029         unlock_chunks(root);
6030         mutex_unlock(&uuid_mutex);
6031
6032         btrfs_free_path(path);
6033         return ret;
6034 }
6035
6036 void btrfs_init_devices_late(struct btrfs_fs_info *fs_info)
6037 {
6038         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6039         struct btrfs_device *device;
6040
6041         mutex_lock(&fs_devices->device_list_mutex);
6042         list_for_each_entry(device, &fs_devices->devices, dev_list)
6043                 device->dev_root = fs_info->dev_root;
6044         mutex_unlock(&fs_devices->device_list_mutex);
6045 }
6046
6047 static void __btrfs_reset_dev_stats(struct btrfs_device *dev)
6048 {
6049         int i;
6050
6051         for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6052                 btrfs_dev_stat_reset(dev, i);
6053 }
6054
6055 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
6056 {
6057         struct btrfs_key key;
6058         struct btrfs_key found_key;
6059         struct btrfs_root *dev_root = fs_info->dev_root;
6060         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6061         struct extent_buffer *eb;
6062         int slot;
6063         int ret = 0;
6064         struct btrfs_device *device;
6065         struct btrfs_path *path = NULL;
6066         int i;
6067
6068         path = btrfs_alloc_path();
6069         if (!path) {
6070                 ret = -ENOMEM;
6071                 goto out;
6072         }
6073
6074         mutex_lock(&fs_devices->device_list_mutex);
6075         list_for_each_entry(device, &fs_devices->devices, dev_list) {
6076                 int item_size;
6077                 struct btrfs_dev_stats_item *ptr;
6078
6079                 key.objectid = 0;
6080                 key.type = BTRFS_DEV_STATS_KEY;
6081                 key.offset = device->devid;
6082                 ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
6083                 if (ret) {
6084                         __btrfs_reset_dev_stats(device);
6085                         device->dev_stats_valid = 1;
6086                         btrfs_release_path(path);
6087                         continue;
6088                 }
6089                 slot = path->slots[0];
6090                 eb = path->nodes[0];
6091                 btrfs_item_key_to_cpu(eb, &found_key, slot);
6092                 item_size = btrfs_item_size_nr(eb, slot);
6093
6094                 ptr = btrfs_item_ptr(eb, slot,
6095                                      struct btrfs_dev_stats_item);
6096
6097                 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
6098                         if (item_size >= (1 + i) * sizeof(__le64))
6099                                 btrfs_dev_stat_set(device, i,
6100                                         btrfs_dev_stats_value(eb, ptr, i));
6101                         else
6102                                 btrfs_dev_stat_reset(device, i);
6103                 }
6104
6105                 device->dev_stats_valid = 1;
6106                 btrfs_dev_stat_print_on_load(device);
6107                 btrfs_release_path(path);
6108         }
6109         mutex_unlock(&fs_devices->device_list_mutex);
6110
6111 out:
6112         btrfs_free_path(path);
6113         return ret < 0 ? ret : 0;
6114 }
6115
6116 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
6117                                 struct btrfs_root *dev_root,
6118                                 struct btrfs_device *device)
6119 {
6120         struct btrfs_path *path;
6121         struct btrfs_key key;
6122         struct extent_buffer *eb;
6123         struct btrfs_dev_stats_item *ptr;
6124         int ret;
6125         int i;
6126
6127         key.objectid = 0;
6128         key.type = BTRFS_DEV_STATS_KEY;
6129         key.offset = device->devid;
6130
6131         path = btrfs_alloc_path();
6132         BUG_ON(!path);
6133         ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
6134         if (ret < 0) {
6135                 printk_in_rcu(KERN_WARNING "BTRFS: "
6136                         "error %d while searching for dev_stats item for device %s!\n",
6137                               ret, rcu_str_deref(device->name));
6138                 goto out;
6139         }
6140
6141         if (ret == 0 &&
6142             btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
6143                 /* need to delete old one and insert a new one */
6144                 ret = btrfs_del_item(trans, dev_root, path);
6145                 if (ret != 0) {
6146                         printk_in_rcu(KERN_WARNING "BTRFS: "
6147                                 "delete too small dev_stats item for device %s failed %d!\n",
6148                                       rcu_str_deref(device->name), ret);
6149                         goto out;
6150                 }
6151                 ret = 1;
6152         }
6153
6154         if (ret == 1) {
6155                 /* need to insert a new item */
6156                 btrfs_release_path(path);
6157                 ret = btrfs_insert_empty_item(trans, dev_root, path,
6158                                               &key, sizeof(*ptr));
6159                 if (ret < 0) {
6160                         printk_in_rcu(KERN_WARNING "BTRFS: "
6161                                           "insert dev_stats item for device %s failed %d!\n",
6162                                       rcu_str_deref(device->name), ret);
6163                         goto out;
6164                 }
6165         }
6166
6167         eb = path->nodes[0];
6168         ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
6169         for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6170                 btrfs_set_dev_stats_value(eb, ptr, i,
6171                                           btrfs_dev_stat_read(device, i));
6172         btrfs_mark_buffer_dirty(eb);
6173
6174 out:
6175         btrfs_free_path(path);
6176         return ret;
6177 }
6178
6179 /*
6180  * called from commit_transaction. Writes all changed device stats to disk.
6181  */
6182 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans,
6183                         struct btrfs_fs_info *fs_info)
6184 {
6185         struct btrfs_root *dev_root = fs_info->dev_root;
6186         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6187         struct btrfs_device *device;
6188         int ret = 0;
6189
6190         mutex_lock(&fs_devices->device_list_mutex);
6191         list_for_each_entry(device, &fs_devices->devices, dev_list) {
6192                 if (!device->dev_stats_valid || !device->dev_stats_dirty)
6193                         continue;
6194
6195                 ret = update_dev_stat_item(trans, dev_root, device);
6196                 if (!ret)
6197                         device->dev_stats_dirty = 0;
6198         }
6199         mutex_unlock(&fs_devices->device_list_mutex);
6200
6201         return ret;
6202 }
6203
6204 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
6205 {
6206         btrfs_dev_stat_inc(dev, index);
6207         btrfs_dev_stat_print_on_error(dev);
6208 }
6209
6210 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
6211 {
6212         if (!dev->dev_stats_valid)
6213                 return;
6214         printk_ratelimited_in_rcu(KERN_ERR "BTRFS: "
6215                            "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
6216                            rcu_str_deref(dev->name),
6217                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
6218                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
6219                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
6220                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
6221                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
6222 }
6223
6224 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
6225 {
6226         int i;
6227
6228         for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6229                 if (btrfs_dev_stat_read(dev, i) != 0)
6230                         break;
6231         if (i == BTRFS_DEV_STAT_VALUES_MAX)
6232                 return; /* all values == 0, suppress message */
6233
6234         printk_in_rcu(KERN_INFO "BTRFS: "
6235                    "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
6236                rcu_str_deref(dev->name),
6237                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
6238                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
6239                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
6240                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
6241                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
6242 }
6243
6244 int btrfs_get_dev_stats(struct btrfs_root *root,
6245                         struct btrfs_ioctl_get_dev_stats *stats)
6246 {
6247         struct btrfs_device *dev;
6248         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
6249         int i;
6250
6251         mutex_lock(&fs_devices->device_list_mutex);
6252         dev = btrfs_find_device(root->fs_info, stats->devid, NULL, NULL);
6253         mutex_unlock(&fs_devices->device_list_mutex);
6254
6255         if (!dev) {
6256                 btrfs_warn(root->fs_info, "get dev_stats failed, device not found");
6257                 return -ENODEV;
6258         } else if (!dev->dev_stats_valid) {
6259                 btrfs_warn(root->fs_info, "get dev_stats failed, not yet valid");
6260                 return -ENODEV;
6261         } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
6262                 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
6263                         if (stats->nr_items > i)
6264                                 stats->values[i] =
6265                                         btrfs_dev_stat_read_and_reset(dev, i);
6266                         else
6267                                 btrfs_dev_stat_reset(dev, i);
6268                 }
6269         } else {
6270                 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6271                         if (stats->nr_items > i)
6272                                 stats->values[i] = btrfs_dev_stat_read(dev, i);
6273         }
6274         if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
6275                 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
6276         return 0;
6277 }
6278
6279 int btrfs_scratch_superblock(struct btrfs_device *device)
6280 {
6281         struct buffer_head *bh;
6282         struct btrfs_super_block *disk_super;
6283
6284         bh = btrfs_read_dev_super(device->bdev);
6285         if (!bh)
6286                 return -EINVAL;
6287         disk_super = (struct btrfs_super_block *)bh->b_data;
6288
6289         memset(&disk_super->magic, 0, sizeof(disk_super->magic));
6290         set_buffer_dirty(bh);
6291         sync_dirty_buffer(bh);
6292         brelse(bh);
6293
6294         return 0;
6295 }