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