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 <asm/div64.h>
  29 #include "compat.h"
  30 #include "ctree.h"
  31 #include "extent_map.h"
  32 #include "disk-io.h"
  33 #include "transaction.h"
  34 #include "print-tree.h"
  35 #include "volumes.h"
  36 #include "async-thread.h"
  37 #include "check-integrity.h"
  38 #include "rcu-string.h"
  39
  40 static int init_first_rw_device(struct btrfs_trans_handle *trans,
  41                                 struct btrfs_root *root,
  42                                 struct btrfs_device *device);
  43 static int btrfs_relocate_sys_chunks(struct btrfs_root *root);
  44 static void __btrfs_reset_dev_stats(struct btrfs_device *dev);
  45 static void btrfs_dev_stat_print_on_load(struct btrfs_device *device);
  46
  47 static DEFINE_MUTEX(uuid_mutex);
  48 static LIST_HEAD(fs_uuids);
  49
  50 static void lock_chunks(struct btrfs_root *root)
  51 {
  52         mutex_lock(&root->fs_info->chunk_mutex);
  53 }
  54
  55 static void unlock_chunks(struct btrfs_root *root)
  56 {
  57         mutex_unlock(&root->fs_info->chunk_mutex);
  58 }
  59
  60 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
  61 {
  62         struct btrfs_device *device;
  63         WARN_ON(fs_devices->opened);
  64         while (!list_empty(&fs_devices->devices)) {
  65                 device = list_entry(fs_devices->devices.next,
  66                                     struct btrfs_device, dev_list);
  67                 list_del(&device->dev_list);
  68                 rcu_string_free(device->name);
  69                 kfree(device);
  70         }
  71         kfree(fs_devices);
  72 }
  73
  74 void btrfs_cleanup_fs_uuids(void)
  75 {
  76         struct btrfs_fs_devices *fs_devices;
  77
  78         while (!list_empty(&fs_uuids)) {
  79                 fs_devices = list_entry(fs_uuids.next,
  80                                         struct btrfs_fs_devices, list);
  81                 list_del(&fs_devices->list);
  82                 free_fs_devices(fs_devices);
  83         }
  84 }
  85
  86 static noinline struct btrfs_device *__find_device(struct list_head *head,
  87                                                    u64 devid, u8 *uuid)
  88 {
  89         struct btrfs_device *dev;
  90
  91         list_for_each_entry(dev, head, dev_list) {
  92                 if (dev->devid == devid &&
  93                     (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
  94                         return dev;
  95                 }
  96         }
  97         return NULL;
  98 }
  99
 100 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
 101 {
 102         struct btrfs_fs_devices *fs_devices;
 103
 104         list_for_each_entry(fs_devices, &fs_uuids, list) {
 105                 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
 106                         return fs_devices;
 107         }
 108         return NULL;
 109 }
 110
 111 static void requeue_list(struct btrfs_pending_bios *pending_bios,
 112                         struct bio *head, struct bio *tail)
 113 {
 114
 115         struct bio *old_head;
 116
 117         old_head = pending_bios->head;
 118         pending_bios->head = head;
 119         if (pending_bios->tail)
 120                 tail->bi_next = old_head;
 121         else
 122                 pending_bios->tail = tail;
 123 }
 124
 125 /*
 126  * we try to collect pending bios for a device so we don't get a large
 127  * number of procs sending bios down to the same device.  This greatly
 128  * improves the schedulers ability to collect and merge the bios.
 129  *
 130  * But, it also turns into a long list of bios to process and that is sure
 131  * to eventually make the worker thread block.  The solution here is to
 132  * make some progress and then put this work struct back at the end of
 133  * the list if the block device is congested.  This way, multiple devices
 134  * can make progress from a single worker thread.
 135  */
 136 static noinline void run_scheduled_bios(struct btrfs_device *device)
 137 {
 138         struct bio *pending;
 139         struct backing_dev_info *bdi;
 140         struct btrfs_fs_info *fs_info;
 141         struct btrfs_pending_bios *pending_bios;
 142         struct bio *tail;
 143         struct bio *cur;
 144         int again = 0;
 145         unsigned long num_run;
 146         unsigned long batch_run = 0;
 147         unsigned long limit;
 148         unsigned long last_waited = 0;
 149         int force_reg = 0;
 150         int sync_pending = 0;
 151         struct blk_plug plug;
 152
 153         /*
 154          * this function runs all the bios we've collected for
 155          * a particular device.  We don't want to wander off to
 156          * another device without first sending all of these down.
 157          * So, setup a plug here and finish it off before we return
 158          */
 159         blk_start_plug(&plug);
 160
 161         bdi = blk_get_backing_dev_info(device->bdev);
 162         fs_info = device->dev_root->fs_info;
 163         limit = btrfs_async_submit_limit(fs_info);
 164         limit = limit * 2 / 3;
 165
 166 loop:
 167         spin_lock(&device->io_lock);
 168
 169 loop_lock:
 170         num_run = 0;
 171
 172         /* take all the bios off the list at once and process them
 173          * later on (without the lock held).  But, remember the
 174          * tail and other pointers so the bios can be properly reinserted
 175          * into the list if we hit congestion
 176          */
 177         if (!force_reg && device->pending_sync_bios.head) {
 178                 pending_bios = &device->pending_sync_bios;
 179                 force_reg = 1;
 180         } else {
 181                 pending_bios = &device->pending_bios;
 182                 force_reg = 0;
 183         }
 184
 185         pending = pending_bios->head;
 186         tail = pending_bios->tail;
 187         WARN_ON(pending && !tail);
 188
 189         /*
 190          * if pending was null this time around, no bios need processing
 191          * at all and we can stop.  Otherwise it'll loop back up again
 192          * and do an additional check so no bios are missed.
 193          *
 194          * device->running_pending is used to synchronize with the
 195          * schedule_bio code.
 196          */
 197         if (device->pending_sync_bios.head == NULL &&
 198             device->pending_bios.head == NULL) {
 199                 again = 0;
 200                 device->running_pending = 0;
 201         } else {
 202                 again = 1;
 203                 device->running_pending = 1;
 204         }
 205
 206         pending_bios->head = NULL;
 207         pending_bios->tail = NULL;
 208
 209         spin_unlock(&device->io_lock);
 210
 211         while (pending) {
 212
 213                 rmb();
 214                 /* we want to work on both lists, but do more bios on the
 215                  * sync list than the regular list
 216                  */
 217                 if ((num_run > 32 &&
 218                     pending_bios != &device->pending_sync_bios &&
 219                     device->pending_sync_bios.head) ||
 220                    (num_run > 64 && pending_bios == &device->pending_sync_bios &&
 221                     device->pending_bios.head)) {
 222                         spin_lock(&device->io_lock);
 223                         requeue_list(pending_bios, pending, tail);
 224                         goto loop_lock;
 225                 }
 226
 227                 cur = pending;
 228                 pending = pending->bi_next;
 229                 cur->bi_next = NULL;
 230                 atomic_dec(&fs_info->nr_async_bios);
 231
 232                 if (atomic_read(&fs_info->nr_async_bios) < limit &&
 233                     waitqueue_active(&fs_info->async_submit_wait))
 234                         wake_up(&fs_info->async_submit_wait);
 235
 236                 BUG_ON(atomic_read(&cur->bi_cnt) == 0);
 237
 238                 /*
 239                  * if we're doing the sync list, record that our
 240                  * plug has some sync requests on it
 241                  *
 242                  * If we're doing the regular list and there are
 243                  * sync requests sitting around, unplug before
 244                  * we add more
 245                  */
 246                 if (pending_bios == &device->pending_sync_bios) {
 247                         sync_pending = 1;
 248                 } else if (sync_pending) {
 249                         blk_finish_plug(&plug);
 250                         blk_start_plug(&plug);
 251                         sync_pending = 0;
 252                 }
 253
 254                 btrfsic_submit_bio(cur->bi_rw, cur);
 255                 num_run++;
 256                 batch_run++;
 257                 if (need_resched())
 258                         cond_resched();
 259
 260                 /*
 261                  * we made progress, there is more work to do and the bdi
 262                  * is now congested.  Back off and let other work structs
 263                  * run instead
 264                  */
 265                 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
 266                     fs_info->fs_devices->open_devices > 1) {
 267                         struct io_context *ioc;
 268
 269                         ioc = current->io_context;
 270
 271                         /*
 272                          * the main goal here is that we don't want to
 273                          * block if we're going to be able to submit
 274                          * more requests without blocking.
 275                          *
 276                          * This code does two great things, it pokes into
 277                          * the elevator code from a filesystem _and_
 278                          * it makes assumptions about how batching works.
 279                          */
 280                         if (ioc && ioc->nr_batch_requests > 0 &&
 281                             time_before(jiffies, ioc->last_waited + HZ/50UL) &&
 282                             (last_waited == 0 ||
 283                              ioc->last_waited == last_waited)) {
 284                                 /*
 285                                  * we want to go through our batch of
 286                                  * requests and stop.  So, we copy out
 287                                  * the ioc->last_waited time and test
 288                                  * against it before looping
 289                                  */
 290                                 last_waited = ioc->last_waited;
 291                                 if (need_resched())
 292                                         cond_resched();
 293                                 continue;
 294                         }
 295                         spin_lock(&device->io_lock);
 296                         requeue_list(pending_bios, pending, tail);
 297                         device->running_pending = 1;
 298
 299                         spin_unlock(&device->io_lock);
 300                         btrfs_requeue_work(&device->work);
 301                         goto done;
 302                 }
 303                 /* unplug every 64 requests just for good measure */
 304                 if (batch_run % 64 == 0) {
 305                         blk_finish_plug(&plug);
 306                         blk_start_plug(&plug);
 307                         sync_pending = 0;
 308                 }
 309         }
 310
 311         cond_resched();
 312         if (again)
 313                 goto loop;
 314
 315         spin_lock(&device->io_lock);
 316         if (device->pending_bios.head || device->pending_sync_bios.head)
 317                 goto loop_lock;
 318         spin_unlock(&device->io_lock);
 319
 320 done:
 321         blk_finish_plug(&plug);
 322 }
 323
 324 static void pending_bios_fn(struct btrfs_work *work)
 325 {
 326         struct btrfs_device *device;
 327
 328         device = container_of(work, struct btrfs_device, work);
 329         run_scheduled_bios(device);
 330 }
 331
 332 static noinline int device_list_add(const char *path,
 333                            struct btrfs_super_block *disk_super,
 334                            u64 devid, struct btrfs_fs_devices **fs_devices_ret)
 335 {
 336         struct btrfs_device *device;
 337         struct btrfs_fs_devices *fs_devices;
 338         struct rcu_string *name;
 339         u64 found_transid = btrfs_super_generation(disk_super);
 340
 341         fs_devices = find_fsid(disk_super->fsid);
 342         if (!fs_devices) {
 343                 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
 344                 if (!fs_devices)
 345                         return -ENOMEM;
 346                 INIT_LIST_HEAD(&fs_devices->devices);
 347                 INIT_LIST_HEAD(&fs_devices->alloc_list);
 348                 list_add(&fs_devices->list, &fs_uuids);
 349                 memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
 350                 fs_devices->latest_devid = devid;
 351                 fs_devices->latest_trans = found_transid;
 352                 mutex_init(&fs_devices->device_list_mutex);
 353                 device = NULL;
 354         } else {
 355                 device = __find_device(&fs_devices->devices, devid,
 356                                        disk_super->dev_item.uuid);
 357         }
 358         if (!device) {
 359                 if (fs_devices->opened)
 360                         return -EBUSY;
 361
 362                 device = kzalloc(sizeof(*device), GFP_NOFS);
 363                 if (!device) {
 364                         /* we can safely leave the fs_devices entry around */
 365                         return -ENOMEM;
 366                 }
 367                 device->devid = devid;
 368                 device->dev_stats_valid = 0;
 369                 device->work.func = pending_bios_fn;
 370                 memcpy(device->uuid, disk_super->dev_item.uuid,
 371                        BTRFS_UUID_SIZE);
 372                 spin_lock_init(&device->io_lock);
 373
 374                 name = rcu_string_strdup(path, GFP_NOFS);
 375                 if (!name) {
 376                         kfree(device);
 377                         return -ENOMEM;
 378                 }
 379                 rcu_assign_pointer(device->name, name);
 380                 INIT_LIST_HEAD(&device->dev_alloc_list);
 381
 382                 /* init readahead state */
 383                 spin_lock_init(&device->reada_lock);
 384                 device->reada_curr_zone = NULL;
 385                 atomic_set(&device->reada_in_flight, 0);
 386                 device->reada_next = 0;
 387                 INIT_RADIX_TREE(&device->reada_zones, GFP_NOFS & ~__GFP_WAIT);
 388                 INIT_RADIX_TREE(&device->reada_extents, GFP_NOFS & ~__GFP_WAIT);
 389
 390                 mutex_lock(&fs_devices->device_list_mutex);
 391                 list_add_rcu(&device->dev_list, &fs_devices->devices);
 392                 mutex_unlock(&fs_devices->device_list_mutex);
 393
 394                 device->fs_devices = fs_devices;
 395                 fs_devices->num_devices++;
 396         } else if (!device->name || strcmp(device->name->str, path)) {
 397                 name = rcu_string_strdup(path, GFP_NOFS);
 398                 if (!name)
 399                         return -ENOMEM;
 400                 rcu_string_free(device->name);
 401                 rcu_assign_pointer(device->name, name);
 402                 if (device->missing) {
 403                         fs_devices->missing_devices--;
 404                         device->missing = 0;
 405                 }
 406         }
 407
 408         if (found_transid > fs_devices->latest_trans) {
 409                 fs_devices->latest_devid = devid;
 410                 fs_devices->latest_trans = found_transid;
 411         }
 412         *fs_devices_ret = fs_devices;
 413         return 0;
 414 }
 415
 416 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
 417 {
 418         struct btrfs_fs_devices *fs_devices;
 419         struct btrfs_device *device;
 420         struct btrfs_device *orig_dev;
 421
 422         fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
 423         if (!fs_devices)
 424                 return ERR_PTR(-ENOMEM);
 425
 426         INIT_LIST_HEAD(&fs_devices->devices);
 427         INIT_LIST_HEAD(&fs_devices->alloc_list);
 428         INIT_LIST_HEAD(&fs_devices->list);
 429         mutex_init(&fs_devices->device_list_mutex);
 430         fs_devices->latest_devid = orig->latest_devid;
 431         fs_devices->latest_trans = orig->latest_trans;
 432         memcpy(fs_devices->fsid, orig->fsid, sizeof(fs_devices->fsid));
 433
 434         /* We have held the volume lock, it is safe to get the devices. */
 435         list_for_each_entry(orig_dev, &orig->devices, dev_list) {
 436                 struct rcu_string *name;
 437
 438                 device = kzalloc(sizeof(*device), GFP_NOFS);
 439                 if (!device)
 440                         goto error;
 441
 442                 /*
 443                  * This is ok to do without rcu read locked because we hold the
 444                  * uuid mutex so nothing we touch in here is going to disappear.
 445                  */
 446                 name = rcu_string_strdup(orig_dev->name->str, GFP_NOFS);
 447                 if (!name) {
 448                         kfree(device);
 449                         goto error;
 450                 }
 451                 rcu_assign_pointer(device->name, name);
 452
 453                 device->devid = orig_dev->devid;
 454                 device->work.func = pending_bios_fn;
 455                 memcpy(device->uuid, orig_dev->uuid, sizeof(device->uuid));
 456                 spin_lock_init(&device->io_lock);
 457                 INIT_LIST_HEAD(&device->dev_list);
 458                 INIT_LIST_HEAD(&device->dev_alloc_list);
 459
 460                 list_add(&device->dev_list, &fs_devices->devices);
 461                 device->fs_devices = fs_devices;
 462                 fs_devices->num_devices++;
 463         }
 464         return fs_devices;
 465 error:
 466         free_fs_devices(fs_devices);
 467         return ERR_PTR(-ENOMEM);
 468 }
 469
 470 void btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices)
 471 {
 472         struct btrfs_device *device, *next;
 473
 474         struct block_device *latest_bdev = NULL;
 475         u64 latest_devid = 0;
 476         u64 latest_transid = 0;
 477
 478         mutex_lock(&uuid_mutex);
 479 again:
 480         /* This is the initialized path, it is safe to release the devices. */
 481         list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
 482                 if (device->in_fs_metadata) {
 483                         if (!latest_transid ||
 484                             device->generation > latest_transid) {
 485                                 latest_devid = device->devid;
 486                                 latest_transid = device->generation;
 487                                 latest_bdev = device->bdev;
 488                         }
 489                         continue;
 490                 }
 491
 492                 if (device->bdev) {
 493                         blkdev_put(device->bdev, device->mode);
 494                         device->bdev = NULL;
 495                         fs_devices->open_devices--;
 496                 }
 497                 if (device->writeable) {
 498                         list_del_init(&device->dev_alloc_list);
 499                         device->writeable = 0;
 500                         fs_devices->rw_devices--;
 501                 }
 502                 list_del_init(&device->dev_list);
 503                 fs_devices->num_devices--;
 504                 rcu_string_free(device->name);
 505                 kfree(device);
 506         }
 507
 508         if (fs_devices->seed) {
 509                 fs_devices = fs_devices->seed;
 510                 goto again;
 511         }
 512
 513         fs_devices->latest_bdev = latest_bdev;
 514         fs_devices->latest_devid = latest_devid;
 515         fs_devices->latest_trans = latest_transid;
 516
 517         mutex_unlock(&uuid_mutex);
 518 }
 519
 520 static void __free_device(struct work_struct *work)
 521 {
 522         struct btrfs_device *device;
 523
 524         device = container_of(work, struct btrfs_device, rcu_work);
 525
 526         if (device->bdev)
 527                 blkdev_put(device->bdev, device->mode);
 528
 529         rcu_string_free(device->name);
 530         kfree(device);
 531 }
 532
 533 static void free_device(struct rcu_head *head)
 534 {
 535         struct btrfs_device *device;
 536
 537         device = container_of(head, struct btrfs_device, rcu);
 538
 539         INIT_WORK(&device->rcu_work, __free_device);
 540         schedule_work(&device->rcu_work);
 541 }
 542
 543 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
 544 {
 545         struct btrfs_device *device;
 546
 547         if (--fs_devices->opened > 0)
 548                 return 0;
 549
 550         mutex_lock(&fs_devices->device_list_mutex);
 551         list_for_each_entry(device, &fs_devices->devices, dev_list) {
 552                 struct btrfs_device *new_device;
 553                 struct rcu_string *name;
 554
 555                 if (device->bdev)
 556                         fs_devices->open_devices--;
 557
 558                 if (device->writeable) {
 559                         list_del_init(&device->dev_alloc_list);
 560                         fs_devices->rw_devices--;
 561                 }
 562
 563                 if (device->can_discard)
 564                         fs_devices->num_can_discard--;
 565
 566                 new_device = kmalloc(sizeof(*new_device), GFP_NOFS);
 567                 BUG_ON(!new_device); /* -ENOMEM */
 568                 memcpy(new_device, device, sizeof(*new_device));
 569
 570                 /* Safe because we are under uuid_mutex */
 571                 name = rcu_string_strdup(device->name->str, GFP_NOFS);
 572                 BUG_ON(device->name && !name); /* -ENOMEM */
 573                 rcu_assign_pointer(new_device->name, name);
 574                 new_device->bdev = NULL;
 575                 new_device->writeable = 0;
 576                 new_device->in_fs_metadata = 0;
 577                 new_device->can_discard = 0;
 578                 list_replace_rcu(&device->dev_list, &new_device->dev_list);
 579
 580                 call_rcu(&device->rcu, free_device);
 581         }
 582         mutex_unlock(&fs_devices->device_list_mutex);
 583
 584         WARN_ON(fs_devices->open_devices);
 585         WARN_ON(fs_devices->rw_devices);
 586         fs_devices->opened = 0;
 587         fs_devices->seeding = 0;
 588
 589         return 0;
 590 }
 591
 592 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
 593 {
 594         struct btrfs_fs_devices *seed_devices = NULL;
 595         int ret;
 596
 597         mutex_lock(&uuid_mutex);
 598         ret = __btrfs_close_devices(fs_devices);
 599         if (!fs_devices->opened) {
 600                 seed_devices = fs_devices->seed;
 601                 fs_devices->seed = NULL;
 602         }
 603         mutex_unlock(&uuid_mutex);
 604
 605         while (seed_devices) {
 606                 fs_devices = seed_devices;
 607                 seed_devices = fs_devices->seed;
 608                 __btrfs_close_devices(fs_devices);
 609                 free_fs_devices(fs_devices);
 610         }
 611         return ret;
 612 }
 613
 614 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
 615                                 fmode_t flags, void *holder)
 616 {
 617         struct request_queue *q;
 618         struct block_device *bdev;
 619         struct list_head *head = &fs_devices->devices;
 620         struct btrfs_device *device;
 621         struct block_device *latest_bdev = NULL;
 622         struct buffer_head *bh;
 623         struct btrfs_super_block *disk_super;
 624         u64 latest_devid = 0;
 625         u64 latest_transid = 0;
 626         u64 devid;
 627         int seeding = 1;
 628         int ret = 0;
 629
 630         flags |= FMODE_EXCL;
 631
 632         list_for_each_entry(device, head, dev_list) {
 633                 if (device->bdev)
 634                         continue;
 635                 if (!device->name)
 636                         continue;
 637
 638                 bdev = blkdev_get_by_path(device->name->str, flags, holder);
 639                 if (IS_ERR(bdev)) {
 640                         printk(KERN_INFO "open %s failed\n", device->name->str);
 641                         goto error;
 642                 }
 643                 filemap_write_and_wait(bdev->bd_inode->i_mapping);
 644                 invalidate_bdev(bdev);
 645                 set_blocksize(bdev, 4096);
 646
 647                 bh = btrfs_read_dev_super(bdev);
 648                 if (!bh)
 649                         goto error_close;
 650
 651                 disk_super = (struct btrfs_super_block *)bh->b_data;
 652                 devid = btrfs_stack_device_id(&disk_super->dev_item);
 653                 if (devid != device->devid)
 654                         goto error_brelse;
 655
 656                 if (memcmp(device->uuid, disk_super->dev_item.uuid,
 657                            BTRFS_UUID_SIZE))
 658                         goto error_brelse;
 659
 660                 device->generation = btrfs_super_generation(disk_super);
 661                 if (!latest_transid || device->generation > latest_transid) {
 662                         latest_devid = devid;
 663                         latest_transid = device->generation;
 664                         latest_bdev = bdev;
 665                 }
 666
 667                 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
 668                         device->writeable = 0;
 669                 } else {
 670                         device->writeable = !bdev_read_only(bdev);
 671                         seeding = 0;
 672                 }
 673
 674                 q = bdev_get_queue(bdev);
 675                 if (blk_queue_discard(q)) {
 676                         device->can_discard = 1;
 677                         fs_devices->num_can_discard++;
 678                 }
 679
 680                 device->bdev = bdev;
 681                 device->in_fs_metadata = 0;
 682                 device->mode = flags;
 683
 684                 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
 685                         fs_devices->rotating = 1;
 686
 687                 fs_devices->open_devices++;
 688                 if (device->writeable) {
 689                         fs_devices->rw_devices++;
 690                         list_add(&device->dev_alloc_list,
 691                                  &fs_devices->alloc_list);
 692                 }
 693                 brelse(bh);
 694                 continue;
 695
 696 error_brelse:
 697                 brelse(bh);
 698 error_close:
 699                 blkdev_put(bdev, flags);
 700 error:
 701                 continue;
 702         }
 703         if (fs_devices->open_devices == 0) {
 704                 ret = -EINVAL;
 705                 goto out;
 706         }
 707         fs_devices->seeding = seeding;
 708         fs_devices->opened = 1;
 709         fs_devices->latest_bdev = latest_bdev;
 710         fs_devices->latest_devid = latest_devid;
 711         fs_devices->latest_trans = latest_transid;
 712         fs_devices->total_rw_bytes = 0;
 713 out:
 714         return ret;
 715 }
 716
 717 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
 718                        fmode_t flags, void *holder)
 719 {
 720         int ret;
 721
 722         mutex_lock(&uuid_mutex);
 723         if (fs_devices->opened) {
 724                 fs_devices->opened++;
 725                 ret = 0;
 726         } else {
 727                 ret = __btrfs_open_devices(fs_devices, flags, holder);
 728         }
 729         mutex_unlock(&uuid_mutex);
 730         return ret;
 731 }
 732
 733 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
 734                           struct btrfs_fs_devices **fs_devices_ret)
 735 {
 736         struct btrfs_super_block *disk_super;
 737         struct block_device *bdev;
 738         struct buffer_head *bh;
 739         int ret;
 740         u64 devid;
 741         u64 transid;
 742
 743         flags |= FMODE_EXCL;
 744         bdev = blkdev_get_by_path(path, flags, holder);
 745
 746         if (IS_ERR(bdev)) {
 747                 ret = PTR_ERR(bdev);
 748                 goto error;
 749         }
 750
 751         mutex_lock(&uuid_mutex);
 752         ret = set_blocksize(bdev, 4096);
 753         if (ret)
 754                 goto error_close;
 755         bh = btrfs_read_dev_super(bdev);
 756         if (!bh) {
 757                 ret = -EINVAL;
 758                 goto error_close;
 759         }
 760         disk_super = (struct btrfs_super_block *)bh->b_data;
 761         devid = btrfs_stack_device_id(&disk_super->dev_item);
 762         transid = btrfs_super_generation(disk_super);
 763         if (disk_super->label[0])
 764                 printk(KERN_INFO "device label %s ", disk_super->label);
 765         else
 766                 printk(KERN_INFO "device fsid %pU ", disk_super->fsid);
 767         printk(KERN_CONT "devid %llu transid %llu %s\n",
 768                (unsigned long long)devid, (unsigned long long)transid, path);
 769         ret = device_list_add(path, disk_super, devid, fs_devices_ret);
 770
 771         brelse(bh);
 772 error_close:
 773         mutex_unlock(&uuid_mutex);
 774         blkdev_put(bdev, flags);
 775 error:
 776         return ret;
 777 }
 778
 779 /* helper to account the used device space in the range */
 780 int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
 781                                    u64 end, u64 *length)
 782 {
 783         struct btrfs_key key;
 784         struct btrfs_root *root = device->dev_root;
 785         struct btrfs_dev_extent *dev_extent;
 786         struct btrfs_path *path;
 787         u64 extent_end;
 788         int ret;
 789         int slot;
 790         struct extent_buffer *l;
 791
 792         *length = 0;
 793
 794         if (start >= device->total_bytes)
 795                 return 0;
 796
 797         path = btrfs_alloc_path();
 798         if (!path)
 799                 return -ENOMEM;
 800         path->reada = 2;
 801
 802         key.objectid = device->devid;
 803         key.offset = start;
 804         key.type = BTRFS_DEV_EXTENT_KEY;
 805
 806         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 807         if (ret < 0)
 808                 goto out;
 809         if (ret > 0) {
 810                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
 811                 if (ret < 0)
 812                         goto out;
 813         }
 814
 815         while (1) {
 816                 l = path->nodes[0];
 817                 slot = path->slots[0];
 818                 if (slot >= btrfs_header_nritems(l)) {
 819                         ret = btrfs_next_leaf(root, path);
 820                         if (ret == 0)
 821                                 continue;
 822                         if (ret < 0)
 823                                 goto out;
 824
 825                         break;
 826                 }
 827                 btrfs_item_key_to_cpu(l, &key, slot);
 828
 829                 if (key.objectid < device->devid)
 830                         goto next;
 831
 832                 if (key.objectid > device->devid)
 833                         break;
 834
 835                 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
 836                         goto next;
 837
 838                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
 839                 extent_end = key.offset + btrfs_dev_extent_length(l,
 840                                                                   dev_extent);
 841                 if (key.offset <= start && extent_end > end) {
 842                         *length = end - start + 1;
 843                         break;
 844                 } else if (key.offset <= start && extent_end > start)
 845                         *length += extent_end - start;
 846                 else if (key.offset > start && extent_end <= end)
 847                         *length += extent_end - key.offset;
 848                 else if (key.offset > start && key.offset <= end) {
 849                         *length += end - key.offset + 1;
 850                         break;
 851                 } else if (key.offset > end)
 852                         break;
 853
 854 next:
 855                 path->slots[0]++;
 856         }
 857         ret = 0;
 858 out:
 859         btrfs_free_path(path);
 860         return ret;
 861 }
 862
 863 /*
 864  * find_free_dev_extent - find free space in the specified device
 865  * @device:     the device which we search the free space in
 866  * @num_bytes:  the size of the free space that we need
 867  * @start:      store the start of the free space.
 868  * @len:        the size of the free space. that we find, or the size of the max
 869  *              free space if we don't find suitable free space
 870  *
 871  * this uses a pretty simple search, the expectation is that it is
 872  * called very infrequently and that a given device has a small number
 873  * of extents
 874  *
 875  * @start is used to store the start of the free space if we find. But if we
 876  * don't find suitable free space, it will be used to store the start position
 877  * of the max free space.
 878  *
 879  * @len is used to store the size of the free space that we find.
 880  * But if we don't find suitable free space, it is used to store the size of
 881  * the max free space.
 882  */
 883 int find_free_dev_extent(struct btrfs_device *device, u64 num_bytes,
 884                          u64 *start, u64 *len)
 885 {
 886         struct btrfs_key key;
 887         struct btrfs_root *root = device->dev_root;
 888         struct btrfs_dev_extent *dev_extent;
 889         struct btrfs_path *path;
 890         u64 hole_size;
 891         u64 max_hole_start;
 892         u64 max_hole_size;
 893         u64 extent_end;
 894         u64 search_start;
 895         u64 search_end = device->total_bytes;
 896         int ret;
 897         int slot;
 898         struct extent_buffer *l;
 899
 900         /* FIXME use last free of some kind */
 901
 902         /* we don't want to overwrite the superblock on the drive,
 903          * so we make sure to start at an offset of at least 1MB
 904          */
 905         search_start = max(root->fs_info->alloc_start, 1024ull * 1024);
 906
 907         max_hole_start = search_start;
 908         max_hole_size = 0;
 909         hole_size = 0;
 910
 911         if (search_start >= search_end) {
 912                 ret = -ENOSPC;
 913                 goto error;
 914         }
 915
 916         path = btrfs_alloc_path();
 917         if (!path) {
 918                 ret = -ENOMEM;
 919                 goto error;
 920         }
 921         path->reada = 2;
 922
 923         key.objectid = device->devid;
 924         key.offset = search_start;
 925         key.type = BTRFS_DEV_EXTENT_KEY;
 926
 927         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 928         if (ret < 0)
 929                 goto out;
 930         if (ret > 0) {
 931                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
 932                 if (ret < 0)
 933                         goto out;
 934         }
 935
 936         while (1) {
 937                 l = path->nodes[0];
 938                 slot = path->slots[0];
 939                 if (slot >= btrfs_header_nritems(l)) {
 940                         ret = btrfs_next_leaf(root, path);
 941                         if (ret == 0)
 942                                 continue;
 943                         if (ret < 0)
 944                                 goto out;
 945
 946                         break;
 947                 }
 948                 btrfs_item_key_to_cpu(l, &key, slot);
 949
 950                 if (key.objectid < device->devid)
 951                         goto next;
 952
 953                 if (key.objectid > device->devid)
 954                         break;
 955
 956                 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
 957                         goto next;
 958
 959                 if (key.offset > search_start) {
 960                         hole_size = key.offset - search_start;
 961
 962                         if (hole_size > max_hole_size) {
 963                                 max_hole_start = search_start;
 964                                 max_hole_size = hole_size;
 965                         }
 966
 967                         /*
 968                          * If this free space is greater than which we need,
 969                          * it must be the max free space that we have found
 970                          * until now, so max_hole_start must point to the start
 971                          * of this free space and the length of this free space
 972                          * is stored in max_hole_size. Thus, we return
 973                          * max_hole_start and max_hole_size and go back to the
 974                          * caller.
 975                          */
 976                         if (hole_size >= num_bytes) {
 977                                 ret = 0;
 978                                 goto out;
 979                         }
 980                 }
 981
 982                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
 983                 extent_end = key.offset + btrfs_dev_extent_length(l,
 984                                                                   dev_extent);
 985                 if (extent_end > search_start)
 986                         search_start = extent_end;
 987 next:
 988                 path->slots[0]++;
 989                 cond_resched();
 990         }
 991
 992         /*
 993          * At this point, search_start should be the end of
 994          * allocated dev extents, and when shrinking the device,
 995          * search_end may be smaller than search_start.
 996          */
 997         if (search_end > search_start)
 998                 hole_size = search_end - search_start;
 999
1000         if (hole_size > max_hole_size) {
1001                 max_hole_start = search_start;
1002                 max_hole_size = hole_size;
1003         }
1004
1005         /* See above. */
1006         if (hole_size < num_bytes)
1007                 ret = -ENOSPC;
1008         else
1009                 ret = 0;
1010
1011 out:
1012         btrfs_free_path(path);
1013 error:
1014         *start = max_hole_start;
1015         if (len)
1016                 *len = max_hole_size;
1017         return ret;
1018 }
1019
1020 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1021                           struct btrfs_device *device,
1022                           u64 start)
1023 {
1024         int ret;
1025         struct btrfs_path *path;
1026         struct btrfs_root *root = device->dev_root;
1027         struct btrfs_key key;
1028         struct btrfs_key found_key;
1029         struct extent_buffer *leaf = NULL;
1030         struct btrfs_dev_extent *extent = NULL;
1031
1032         path = btrfs_alloc_path();
1033         if (!path)
1034                 return -ENOMEM;
1035
1036         key.objectid = device->devid;
1037         key.offset = start;
1038         key.type = BTRFS_DEV_EXTENT_KEY;
1039 again:
1040         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1041         if (ret > 0) {
1042                 ret = btrfs_previous_item(root, path, key.objectid,
1043                                           BTRFS_DEV_EXTENT_KEY);
1044                 if (ret)
1045                         goto out;
1046                 leaf = path->nodes[0];
1047                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1048                 extent = btrfs_item_ptr(leaf, path->slots[0],
1049                                         struct btrfs_dev_extent);
1050                 BUG_ON(found_key.offset > start || found_key.offset +
1051                        btrfs_dev_extent_length(leaf, extent) < start);
1052                 key = found_key;
1053                 btrfs_release_path(path);
1054                 goto again;
1055         } else if (ret == 0) {
1056                 leaf = path->nodes[0];
1057                 extent = btrfs_item_ptr(leaf, path->slots[0],
1058                                         struct btrfs_dev_extent);
1059         } else {
1060                 btrfs_error(root->fs_info, ret, "Slot search failed");
1061                 goto out;
1062         }
1063
1064         if (device->bytes_used > 0) {
1065                 u64 len = btrfs_dev_extent_length(leaf, extent);
1066                 device->bytes_used -= len;
1067                 spin_lock(&root->fs_info->free_chunk_lock);
1068                 root->fs_info->free_chunk_space += len;
1069                 spin_unlock(&root->fs_info->free_chunk_lock);
1070         }
1071         ret = btrfs_del_item(trans, root, path);
1072         if (ret) {
1073                 btrfs_error(root->fs_info, ret,
1074                             "Failed to remove dev extent item");
1075         }
1076 out:
1077         btrfs_free_path(path);
1078         return ret;
1079 }
1080
1081 int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1082                            struct btrfs_device *device,
1083                            u64 chunk_tree, u64 chunk_objectid,
1084                            u64 chunk_offset, u64 start, u64 num_bytes)
1085 {
1086         int ret;
1087         struct btrfs_path *path;
1088         struct btrfs_root *root = device->dev_root;
1089         struct btrfs_dev_extent *extent;
1090         struct extent_buffer *leaf;
1091         struct btrfs_key key;
1092
1093         WARN_ON(!device->in_fs_metadata);
1094         path = btrfs_alloc_path();
1095         if (!path)
1096                 return -ENOMEM;
1097
1098         key.objectid = device->devid;
1099         key.offset = start;
1100         key.type = BTRFS_DEV_EXTENT_KEY;
1101         ret = btrfs_insert_empty_item(trans, root, path, &key,
1102                                       sizeof(*extent));
1103         if (ret)
1104                 goto out;
1105
1106         leaf = path->nodes[0];
1107         extent = btrfs_item_ptr(leaf, path->slots[0],
1108                                 struct btrfs_dev_extent);
1109         btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
1110         btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
1111         btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1112
1113         write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
1114                     (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
1115                     BTRFS_UUID_SIZE);
1116
1117         btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1118         btrfs_mark_buffer_dirty(leaf);
1119 out:
1120         btrfs_free_path(path);
1121         return ret;
1122 }
1123
1124 static noinline int find_next_chunk(struct btrfs_root *root,
1125                                     u64 objectid, u64 *offset)
1126 {
1127         struct btrfs_path *path;
1128         int ret;
1129         struct btrfs_key key;
1130         struct btrfs_chunk *chunk;
1131         struct btrfs_key found_key;
1132
1133         path = btrfs_alloc_path();
1134         if (!path)
1135                 return -ENOMEM;
1136
1137         key.objectid = objectid;
1138         key.offset = (u64)-1;
1139         key.type = BTRFS_CHUNK_ITEM_KEY;
1140
1141         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1142         if (ret < 0)
1143                 goto error;
1144
1145         BUG_ON(ret == 0); /* Corruption */
1146
1147         ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
1148         if (ret) {
1149                 *offset = 0;
1150         } else {
1151                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1152                                       path->slots[0]);
1153                 if (found_key.objectid != objectid)
1154                         *offset = 0;
1155                 else {
1156                         chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
1157                                                struct btrfs_chunk);
1158                         *offset = found_key.offset +
1159                                 btrfs_chunk_length(path->nodes[0], chunk);
1160                 }
1161         }
1162         ret = 0;
1163 error:
1164         btrfs_free_path(path);
1165         return ret;
1166 }
1167
1168 static noinline int find_next_devid(struct btrfs_root *root, u64 *objectid)
1169 {
1170         int ret;
1171         struct btrfs_key key;
1172         struct btrfs_key found_key;
1173         struct btrfs_path *path;
1174
1175         root = root->fs_info->chunk_root;
1176
1177         path = btrfs_alloc_path();
1178         if (!path)
1179                 return -ENOMEM;
1180
1181         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1182         key.type = BTRFS_DEV_ITEM_KEY;
1183         key.offset = (u64)-1;
1184
1185         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1186         if (ret < 0)
1187                 goto error;
1188
1189         BUG_ON(ret == 0); /* Corruption */
1190
1191         ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
1192                                   BTRFS_DEV_ITEM_KEY);
1193         if (ret) {
1194                 *objectid = 1;
1195         } else {
1196                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1197                                       path->slots[0]);
1198                 *objectid = found_key.offset + 1;
1199         }
1200         ret = 0;
1201 error:
1202         btrfs_free_path(path);
1203         return ret;
1204 }
1205
1206 /*
1207  * the device information is stored in the chunk root
1208  * the btrfs_device struct should be fully filled in
1209  */
1210 int btrfs_add_device(struct btrfs_trans_handle *trans,
1211                      struct btrfs_root *root,
1212                      struct btrfs_device *device)
1213 {
1214         int ret;
1215         struct btrfs_path *path;
1216         struct btrfs_dev_item *dev_item;
1217         struct extent_buffer *leaf;
1218         struct btrfs_key key;
1219         unsigned long ptr;
1220
1221         root = root->fs_info->chunk_root;
1222
1223         path = btrfs_alloc_path();
1224         if (!path)
1225                 return -ENOMEM;
1226
1227         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1228         key.type = BTRFS_DEV_ITEM_KEY;
1229         key.offset = device->devid;
1230
1231         ret = btrfs_insert_empty_item(trans, root, path, &key,
1232                                       sizeof(*dev_item));
1233         if (ret)
1234                 goto out;
1235
1236         leaf = path->nodes[0];
1237         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1238
1239         btrfs_set_device_id(leaf, dev_item, device->devid);
1240         btrfs_set_device_generation(leaf, dev_item, 0);
1241         btrfs_set_device_type(leaf, dev_item, device->type);
1242         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1243         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1244         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1245         btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
1246         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1247         btrfs_set_device_group(leaf, dev_item, 0);
1248         btrfs_set_device_seek_speed(leaf, dev_item, 0);
1249         btrfs_set_device_bandwidth(leaf, dev_item, 0);
1250         btrfs_set_device_start_offset(leaf, dev_item, 0);
1251
1252         ptr = (unsigned long)btrfs_device_uuid(dev_item);
1253         write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1254         ptr = (unsigned long)btrfs_device_fsid(dev_item);
1255         write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1256         btrfs_mark_buffer_dirty(leaf);
1257
1258         ret = 0;
1259 out:
1260         btrfs_free_path(path);
1261         return ret;
1262 }
1263
1264 static int btrfs_rm_dev_item(struct btrfs_root *root,
1265                              struct btrfs_device *device)
1266 {
1267         int ret;
1268         struct btrfs_path *path;
1269         struct btrfs_key key;
1270         struct btrfs_trans_handle *trans;
1271
1272         root = root->fs_info->chunk_root;
1273
1274         path = btrfs_alloc_path();
1275         if (!path)
1276                 return -ENOMEM;
1277
1278         trans = btrfs_start_transaction(root, 0);
1279         if (IS_ERR(trans)) {
1280                 btrfs_free_path(path);
1281                 return PTR_ERR(trans);
1282         }
1283         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1284         key.type = BTRFS_DEV_ITEM_KEY;
1285         key.offset = device->devid;
1286         lock_chunks(root);
1287
1288         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1289         if (ret < 0)
1290                 goto out;
1291
1292         if (ret > 0) {
1293                 ret = -ENOENT;
1294                 goto out;
1295         }
1296
1297         ret = btrfs_del_item(trans, root, path);
1298         if (ret)
1299                 goto out;
1300 out:
1301         btrfs_free_path(path);
1302         unlock_chunks(root);
1303         btrfs_commit_transaction(trans, root);
1304         return ret;
1305 }
1306
1307 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
1308 {
1309         struct btrfs_device *device;
1310         struct btrfs_device *next_device;
1311         struct block_device *bdev;
1312         struct buffer_head *bh = NULL;
1313         struct btrfs_super_block *disk_super;
1314         struct btrfs_fs_devices *cur_devices;
1315         u64 all_avail;
1316         u64 devid;
1317         u64 num_devices;
1318         u8 *dev_uuid;
1319         int ret = 0;
1320         bool clear_super = false;
1321
1322         mutex_lock(&uuid_mutex);
1323
1324         all_avail = root->fs_info->avail_data_alloc_bits |
1325                 root->fs_info->avail_system_alloc_bits |
1326                 root->fs_info->avail_metadata_alloc_bits;
1327
1328         if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) &&
1329             root->fs_info->fs_devices->num_devices <= 4) {
1330                 printk(KERN_ERR "btrfs: unable to go below four devices "
1331                        "on raid10\n");
1332                 ret = -EINVAL;
1333                 goto out;
1334         }
1335
1336         if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) &&
1337             root->fs_info->fs_devices->num_devices <= 2) {
1338                 printk(KERN_ERR "btrfs: unable to go below two "
1339                        "devices on raid1\n");
1340                 ret = -EINVAL;
1341                 goto out;
1342         }
1343
1344         if (strcmp(device_path, "missing") == 0) {
1345                 struct list_head *devices;
1346                 struct btrfs_device *tmp;
1347
1348                 device = NULL;
1349                 devices = &root->fs_info->fs_devices->devices;
1350                 /*
1351                  * It is safe to read the devices since the volume_mutex
1352                  * is held.
1353                  */
1354                 list_for_each_entry(tmp, devices, dev_list) {
1355                         if (tmp->in_fs_metadata && !tmp->bdev) {
1356                                 device = tmp;
1357                                 break;
1358                         }
1359                 }
1360                 bdev = NULL;
1361                 bh = NULL;
1362                 disk_super = NULL;
1363                 if (!device) {
1364                         printk(KERN_ERR "btrfs: no missing devices found to "
1365                                "remove\n");
1366                         goto out;
1367                 }
1368         } else {
1369                 bdev = blkdev_get_by_path(device_path, FMODE_READ | FMODE_EXCL,
1370                                           root->fs_info->bdev_holder);
1371                 if (IS_ERR(bdev)) {
1372                         ret = PTR_ERR(bdev);
1373                         goto out;
1374                 }
1375
1376                 set_blocksize(bdev, 4096);
1377                 invalidate_bdev(bdev);
1378                 bh = btrfs_read_dev_super(bdev);
1379                 if (!bh) {
1380                         ret = -EINVAL;
1381                         goto error_close;
1382                 }
1383                 disk_super = (struct btrfs_super_block *)bh->b_data;
1384                 devid = btrfs_stack_device_id(&disk_super->dev_item);
1385                 dev_uuid = disk_super->dev_item.uuid;
1386                 device = btrfs_find_device(root, devid, dev_uuid,
1387                                            disk_super->fsid);
1388                 if (!device) {
1389                         ret = -ENOENT;
1390                         goto error_brelse;
1391                 }
1392         }
1393
1394         if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1395                 printk(KERN_ERR "btrfs: unable to remove the only writeable "
1396                        "device\n");
1397                 ret = -EINVAL;
1398                 goto error_brelse;
1399         }
1400
1401         if (device->writeable) {
1402                 lock_chunks(root);
1403                 list_del_init(&device->dev_alloc_list);
1404                 unlock_chunks(root);
1405                 root->fs_info->fs_devices->rw_devices--;
1406                 clear_super = true;
1407         }
1408
1409         ret = btrfs_shrink_device(device, 0);
1410         if (ret)
1411                 goto error_undo;
1412
1413         ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1414         if (ret)
1415                 goto error_undo;
1416
1417         spin_lock(&root->fs_info->free_chunk_lock);
1418         root->fs_info->free_chunk_space = device->total_bytes -
1419                 device->bytes_used;
1420         spin_unlock(&root->fs_info->free_chunk_lock);
1421
1422         device->in_fs_metadata = 0;
1423         btrfs_scrub_cancel_dev(root, device);
1424
1425         /*
1426          * the device list mutex makes sure that we don't change
1427          * the device list while someone else is writing out all
1428          * the device supers.
1429          */
1430
1431         cur_devices = device->fs_devices;
1432         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1433         list_del_rcu(&device->dev_list);
1434
1435         device->fs_devices->num_devices--;
1436
1437         if (device->missing)
1438                 root->fs_info->fs_devices->missing_devices--;
1439
1440         next_device = list_entry(root->fs_info->fs_devices->devices.next,
1441                                  struct btrfs_device, dev_list);
1442         if (device->bdev == root->fs_info->sb->s_bdev)
1443                 root->fs_info->sb->s_bdev = next_device->bdev;
1444         if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1445                 root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1446
1447         if (device->bdev)
1448                 device->fs_devices->open_devices--;
1449
1450         call_rcu(&device->rcu, free_device);
1451         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1452
1453         num_devices = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
1454         btrfs_set_super_num_devices(root->fs_info->super_copy, num_devices);
1455
1456         if (cur_devices->open_devices == 0) {
1457                 struct btrfs_fs_devices *fs_devices;
1458                 fs_devices = root->fs_info->fs_devices;
1459                 while (fs_devices) {
1460                         if (fs_devices->seed == cur_devices)
1461                                 break;
1462                         fs_devices = fs_devices->seed;
1463                 }
1464                 fs_devices->seed = cur_devices->seed;
1465                 cur_devices->seed = NULL;
1466                 lock_chunks(root);
1467                 __btrfs_close_devices(cur_devices);
1468                 unlock_chunks(root);
1469                 free_fs_devices(cur_devices);
1470         }
1471
1472         /*
1473          * at this point, the device is zero sized.  We want to
1474          * remove it from the devices list and zero out the old super
1475          */
1476         if (clear_super) {
1477                 /* make sure this device isn't detected as part of
1478                  * the FS anymore
1479                  */
1480                 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1481                 set_buffer_dirty(bh);
1482                 sync_dirty_buffer(bh);
1483         }
1484
1485         ret = 0;
1486
1487 error_brelse:
1488         brelse(bh);
1489 error_close:
1490         if (bdev)
1491                 blkdev_put(bdev, FMODE_READ | FMODE_EXCL);
1492 out:
1493         mutex_unlock(&uuid_mutex);
1494         return ret;
1495 error_undo:
1496         if (device->writeable) {
1497                 lock_chunks(root);
1498                 list_add(&device->dev_alloc_list,
1499                          &root->fs_info->fs_devices->alloc_list);
1500                 unlock_chunks(root);
1501                 root->fs_info->fs_devices->rw_devices++;
1502         }
1503         goto error_brelse;
1504 }
1505
1506 /*
1507  * does all the dirty work required for changing file system's UUID.
1508  */
1509 static int btrfs_prepare_sprout(struct btrfs_root *root)
1510 {
1511         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
1512         struct btrfs_fs_devices *old_devices;
1513         struct btrfs_fs_devices *seed_devices;
1514         struct btrfs_super_block *disk_super = root->fs_info->super_copy;
1515         struct btrfs_device *device;
1516         u64 super_flags;
1517
1518         BUG_ON(!mutex_is_locked(&uuid_mutex));
1519         if (!fs_devices->seeding)
1520                 return -EINVAL;
1521
1522         seed_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
1523         if (!seed_devices)
1524                 return -ENOMEM;
1525
1526         old_devices = clone_fs_devices(fs_devices);
1527         if (IS_ERR(old_devices)) {
1528                 kfree(seed_devices);
1529                 return PTR_ERR(old_devices);
1530         }
1531
1532         list_add(&old_devices->list, &fs_uuids);
1533
1534         memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
1535         seed_devices->opened = 1;
1536         INIT_LIST_HEAD(&seed_devices->devices);
1537         INIT_LIST_HEAD(&seed_devices->alloc_list);
1538         mutex_init(&seed_devices->device_list_mutex);
1539
1540         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1541         list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
1542                               synchronize_rcu);
1543         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1544
1545         list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
1546         list_for_each_entry(device, &seed_devices->devices, dev_list) {
1547                 device->fs_devices = seed_devices;
1548         }
1549
1550         fs_devices->seeding = 0;
1551         fs_devices->num_devices = 0;
1552         fs_devices->open_devices = 0;
1553         fs_devices->seed = seed_devices;
1554
1555         generate_random_uuid(fs_devices->fsid);
1556         memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1557         memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1558         super_flags = btrfs_super_flags(disk_super) &
1559                       ~BTRFS_SUPER_FLAG_SEEDING;
1560         btrfs_set_super_flags(disk_super, super_flags);
1561
1562         return 0;
1563 }
1564
1565 /*
1566  * strore the expected generation for seed devices in device items.
1567  */
1568 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
1569                                struct btrfs_root *root)
1570 {
1571         struct btrfs_path *path;
1572         struct extent_buffer *leaf;
1573         struct btrfs_dev_item *dev_item;
1574         struct btrfs_device *device;
1575         struct btrfs_key key;
1576         u8 fs_uuid[BTRFS_UUID_SIZE];
1577         u8 dev_uuid[BTRFS_UUID_SIZE];
1578         u64 devid;
1579         int ret;
1580
1581         path = btrfs_alloc_path();
1582         if (!path)
1583                 return -ENOMEM;
1584
1585         root = root->fs_info->chunk_root;
1586         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1587         key.offset = 0;
1588         key.type = BTRFS_DEV_ITEM_KEY;
1589
1590         while (1) {
1591                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1592                 if (ret < 0)
1593                         goto error;
1594
1595                 leaf = path->nodes[0];
1596 next_slot:
1597                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1598                         ret = btrfs_next_leaf(root, path);
1599                         if (ret > 0)
1600                                 break;
1601                         if (ret < 0)
1602                                 goto error;
1603                         leaf = path->nodes[0];
1604                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1605                         btrfs_release_path(path);
1606                         continue;
1607                 }
1608
1609                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1610                 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
1611                     key.type != BTRFS_DEV_ITEM_KEY)
1612                         break;
1613
1614                 dev_item = btrfs_item_ptr(leaf, path->slots[0],
1615                                           struct btrfs_dev_item);
1616                 devid = btrfs_device_id(leaf, dev_item);
1617                 read_extent_buffer(leaf, dev_uuid,
1618                                    (unsigned long)btrfs_device_uuid(dev_item),
1619                                    BTRFS_UUID_SIZE);
1620                 read_extent_buffer(leaf, fs_uuid,
1621                                    (unsigned long)btrfs_device_fsid(dev_item),
1622                                    BTRFS_UUID_SIZE);
1623                 device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
1624                 BUG_ON(!device); /* Logic error */
1625
1626                 if (device->fs_devices->seeding) {
1627                         btrfs_set_device_generation(leaf, dev_item,
1628                                                     device->generation);
1629                         btrfs_mark_buffer_dirty(leaf);
1630                 }
1631
1632                 path->slots[0]++;
1633                 goto next_slot;
1634         }
1635         ret = 0;
1636 error:
1637         btrfs_free_path(path);
1638         return ret;
1639 }
1640
1641 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
1642 {
1643         struct request_queue *q;
1644         struct btrfs_trans_handle *trans;
1645         struct btrfs_device *device;
1646         struct block_device *bdev;
1647         struct list_head *devices;
1648         struct super_block *sb = root->fs_info->sb;
1649         struct rcu_string *name;
1650         u64 total_bytes;
1651         int seeding_dev = 0;
1652         int ret = 0;
1653
1654         if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
1655                 return -EROFS;
1656
1657         bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
1658                                   root->fs_info->bdev_holder);
1659         if (IS_ERR(bdev))
1660                 return PTR_ERR(bdev);
1661
1662         if (root->fs_info->fs_devices->seeding) {
1663                 seeding_dev = 1;
1664                 down_write(&sb->s_umount);
1665                 mutex_lock(&uuid_mutex);
1666         }
1667
1668         filemap_write_and_wait(bdev->bd_inode->i_mapping);
1669
1670         devices = &root->fs_info->fs_devices->devices;
1671         /*
1672          * we have the volume lock, so we don't need the extra
1673          * device list mutex while reading the list here.
1674          */
1675         list_for_each_entry(device, devices, dev_list) {
1676                 if (device->bdev == bdev) {
1677                         ret = -EEXIST;
1678                         goto error;
1679                 }
1680         }
1681
1682         device = kzalloc(sizeof(*device), GFP_NOFS);
1683         if (!device) {
1684                 /* we can safely leave the fs_devices entry around */
1685                 ret = -ENOMEM;
1686                 goto error;
1687         }
1688
1689         name = rcu_string_strdup(device_path, GFP_NOFS);
1690         if (!name) {
1691                 kfree(device);
1692                 ret = -ENOMEM;
1693                 goto error;
1694         }
1695         rcu_assign_pointer(device->name, name);
1696
1697         ret = find_next_devid(root, &device->devid);
1698         if (ret) {
1699                 rcu_string_free(device->name);
1700                 kfree(device);
1701                 goto error;
1702         }
1703
1704         trans = btrfs_start_transaction(root, 0);
1705         if (IS_ERR(trans)) {
1706                 rcu_string_free(device->name);
1707                 kfree(device);
1708                 ret = PTR_ERR(trans);
1709                 goto error;
1710         }
1711
1712         lock_chunks(root);
1713
1714         q = bdev_get_queue(bdev);
1715         if (blk_queue_discard(q))
1716                 device->can_discard = 1;
1717         device->writeable = 1;
1718         device->work.func = pending_bios_fn;
1719         generate_random_uuid(device->uuid);
1720         spin_lock_init(&device->io_lock);
1721         device->generation = trans->transid;
1722         device->io_width = root->sectorsize;
1723         device->io_align = root->sectorsize;
1724         device->sector_size = root->sectorsize;
1725         device->total_bytes = i_size_read(bdev->bd_inode);
1726         device->disk_total_bytes = device->total_bytes;
1727         device->dev_root = root->fs_info->dev_root;
1728         device->bdev = bdev;
1729         device->in_fs_metadata = 1;
1730         device->mode = FMODE_EXCL;
1731         set_blocksize(device->bdev, 4096);
1732
1733         if (seeding_dev) {
1734                 sb->s_flags &= ~MS_RDONLY;
1735                 ret = btrfs_prepare_sprout(root);
1736                 BUG_ON(ret); /* -ENOMEM */
1737         }
1738
1739         device->fs_devices = root->fs_info->fs_devices;
1740
1741         /*
1742          * we don't want write_supers to jump in here with our device
1743          * half setup
1744          */
1745         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1746         list_add_rcu(&device->dev_list, &root->fs_info->fs_devices->devices);
1747         list_add(&device->dev_alloc_list,
1748                  &root->fs_info->fs_devices->alloc_list);
1749         root->fs_info->fs_devices->num_devices++;
1750         root->fs_info->fs_devices->open_devices++;
1751         root->fs_info->fs_devices->rw_devices++;
1752         if (device->can_discard)
1753                 root->fs_info->fs_devices->num_can_discard++;
1754         root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
1755
1756         spin_lock(&root->fs_info->free_chunk_lock);
1757         root->fs_info->free_chunk_space += device->total_bytes;
1758         spin_unlock(&root->fs_info->free_chunk_lock);
1759
1760         if (!blk_queue_nonrot(bdev_get_queue(bdev)))
1761                 root->fs_info->fs_devices->rotating = 1;
1762
1763         total_bytes = btrfs_super_total_bytes(root->fs_info->super_copy);
1764         btrfs_set_super_total_bytes(root->fs_info->super_copy,
1765                                     total_bytes + device->total_bytes);
1766
1767         total_bytes = btrfs_super_num_devices(root->fs_info->super_copy);
1768         btrfs_set_super_num_devices(root->fs_info->super_copy,
1769                                     total_bytes + 1);
1770         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1771
1772         if (seeding_dev) {
1773                 ret = init_first_rw_device(trans, root, device);
1774                 if (ret)
1775                         goto error_trans;
1776                 ret = btrfs_finish_sprout(trans, root);
1777                 if (ret)
1778                         goto error_trans;
1779         } else {
1780                 ret = btrfs_add_device(trans, root, device);
1781                 if (ret)
1782                         goto error_trans;
1783         }
1784
1785         /*
1786          * we've got more storage, clear any full flags on the space
1787          * infos
1788          */
1789         btrfs_clear_space_info_full(root->fs_info);
1790
1791         unlock_chunks(root);
1792         ret = btrfs_commit_transaction(trans, root);
1793
1794         if (seeding_dev) {
1795                 mutex_unlock(&uuid_mutex);
1796                 up_write(&sb->s_umount);
1797
1798                 if (ret) /* transaction commit */
1799                         return ret;
1800
1801                 ret = btrfs_relocate_sys_chunks(root);
1802                 if (ret < 0)
1803                         btrfs_error(root->fs_info, ret,
1804                                     "Failed to relocate sys chunks after "
1805                                     "device initialization. This can be fixed "
1806                                     "using the \"btrfs balance\" command.");
1807         }
1808
1809         return ret;
1810
1811 error_trans:
1812         unlock_chunks(root);
1813         btrfs_abort_transaction(trans, root, ret);
1814         btrfs_end_transaction(trans, root);
1815         rcu_string_free(device->name);
1816         kfree(device);
1817 error:
1818         blkdev_put(bdev, FMODE_EXCL);
1819         if (seeding_dev) {
1820                 mutex_unlock(&uuid_mutex);
1821                 up_write(&sb->s_umount);
1822         }
1823         return ret;
1824 }
1825
1826 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
1827                                         struct btrfs_device *device)
1828 {
1829         int ret;
1830         struct btrfs_path *path;
1831         struct btrfs_root *root;
1832         struct btrfs_dev_item *dev_item;
1833         struct extent_buffer *leaf;
1834         struct btrfs_key key;
1835
1836         root = device->dev_root->fs_info->chunk_root;
1837
1838         path = btrfs_alloc_path();
1839         if (!path)
1840                 return -ENOMEM;
1841
1842         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1843         key.type = BTRFS_DEV_ITEM_KEY;
1844         key.offset = device->devid;
1845
1846         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1847         if (ret < 0)
1848                 goto out;
1849
1850         if (ret > 0) {
1851                 ret = -ENOENT;
1852                 goto out;
1853         }
1854
1855         leaf = path->nodes[0];
1856         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1857
1858         btrfs_set_device_id(leaf, dev_item, device->devid);
1859         btrfs_set_device_type(leaf, dev_item, device->type);
1860         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1861         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1862         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1863         btrfs_set_device_total_bytes(leaf, dev_item, device->disk_total_bytes);
1864         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1865         btrfs_mark_buffer_dirty(leaf);
1866
1867 out:
1868         btrfs_free_path(path);
1869         return ret;
1870 }
1871
1872 static int __btrfs_grow_device(struct btrfs_trans_handle *trans,
1873                       struct btrfs_device *device, u64 new_size)
1874 {
1875         struct btrfs_super_block *super_copy =
1876                 device->dev_root->fs_info->super_copy;
1877         u64 old_total = btrfs_super_total_bytes(super_copy);
1878         u64 diff = new_size - device->total_bytes;
1879
1880         if (!device->writeable)
1881                 return -EACCES;
1882         if (new_size <= device->total_bytes)
1883                 return -EINVAL;
1884
1885         btrfs_set_super_total_bytes(super_copy, old_total + diff);
1886         device->fs_devices->total_rw_bytes += diff;
1887
1888         device->total_bytes = new_size;
1889         device->disk_total_bytes = new_size;
1890         btrfs_clear_space_info_full(device->dev_root->fs_info);
1891
1892         return btrfs_update_device(trans, device);
1893 }
1894
1895 int btrfs_grow_device(struct btrfs_trans_handle *trans,
1896                       struct btrfs_device *device, u64 new_size)
1897 {
1898         int ret;
1899         lock_chunks(device->dev_root);
1900         ret = __btrfs_grow_device(trans, device, new_size);
1901         unlock_chunks(device->dev_root);
1902         return ret;
1903 }
1904
1905 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
1906                             struct btrfs_root *root,
1907                             u64 chunk_tree, u64 chunk_objectid,
1908                             u64 chunk_offset)
1909 {
1910         int ret;
1911         struct btrfs_path *path;
1912         struct btrfs_key key;
1913
1914         root = root->fs_info->chunk_root;
1915         path = btrfs_alloc_path();
1916         if (!path)
1917                 return -ENOMEM;
1918
1919         key.objectid = chunk_objectid;
1920         key.offset = chunk_offset;
1921         key.type = BTRFS_CHUNK_ITEM_KEY;
1922
1923         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1924         if (ret < 0)
1925                 goto out;
1926         else if (ret > 0) { /* Logic error or corruption */
1927                 btrfs_error(root->fs_info, -ENOENT,
1928                             "Failed lookup while freeing chunk.");
1929                 ret = -ENOENT;
1930                 goto out;
1931         }
1932
1933         ret = btrfs_del_item(trans, root, path);
1934         if (ret < 0)
1935                 btrfs_error(root->fs_info, ret,
1936                             "Failed to delete chunk item.");
1937 out:
1938         btrfs_free_path(path);
1939         return ret;
1940 }
1941
1942 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
1943                         chunk_offset)
1944 {
1945         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
1946         struct btrfs_disk_key *disk_key;
1947         struct btrfs_chunk *chunk;
1948         u8 *ptr;
1949         int ret = 0;
1950         u32 num_stripes;
1951         u32 array_size;
1952         u32 len = 0;
1953         u32 cur;
1954         struct btrfs_key key;
1955
1956         array_size = btrfs_super_sys_array_size(super_copy);
1957
1958         ptr = super_copy->sys_chunk_array;
1959         cur = 0;
1960
1961         while (cur < array_size) {
1962                 disk_key = (struct btrfs_disk_key *)ptr;
1963                 btrfs_disk_key_to_cpu(&key, disk_key);
1964
1965                 len = sizeof(*disk_key);
1966
1967                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
1968                         chunk = (struct btrfs_chunk *)(ptr + len);
1969                         num_stripes = btrfs_stack_chunk_num_stripes(chunk);
1970                         len += btrfs_chunk_item_size(num_stripes);
1971                 } else {
1972                         ret = -EIO;
1973                         break;
1974                 }
1975                 if (key.objectid == chunk_objectid &&
1976                     key.offset == chunk_offset) {
1977                         memmove(ptr, ptr + len, array_size - (cur + len));
1978                         array_size -= len;
1979                         btrfs_set_super_sys_array_size(super_copy, array_size);
1980                 } else {
1981                         ptr += len;
1982                         cur += len;
1983                 }
1984         }
1985         return ret;
1986 }
1987
1988 static int btrfs_relocate_chunk(struct btrfs_root *root,
1989                          u64 chunk_tree, u64 chunk_objectid,
1990                          u64 chunk_offset)
1991 {
1992         struct extent_map_tree *em_tree;
1993         struct btrfs_root *extent_root;
1994         struct btrfs_trans_handle *trans;
1995         struct extent_map *em;
1996         struct map_lookup *map;
1997         int ret;
1998         int i;
1999
2000         root = root->fs_info->chunk_root;
2001         extent_root = root->fs_info->extent_root;
2002         em_tree = &root->fs_info->mapping_tree.map_tree;
2003
2004         ret = btrfs_can_relocate(extent_root, chunk_offset);
2005         if (ret)
2006                 return -ENOSPC;
2007
2008         /* step one, relocate all the extents inside this chunk */
2009         ret = btrfs_relocate_block_group(extent_root, chunk_offset);
2010         if (ret)
2011                 return ret;
2012
2013         trans = btrfs_start_transaction(root, 0);
2014         BUG_ON(IS_ERR(trans));
2015
2016         lock_chunks(root);
2017
2018         /*
2019          * step two, delete the device extents and the
2020          * chunk tree entries
2021          */
2022         read_lock(&em_tree->lock);
2023         em = lookup_extent_mapping(em_tree, chunk_offset, 1);
2024         read_unlock(&em_tree->lock);
2025
2026         BUG_ON(!em || em->start > chunk_offset ||
2027                em->start + em->len < chunk_offset);
2028         map = (struct map_lookup *)em->bdev;
2029
2030         for (i = 0; i < map->num_stripes; i++) {
2031                 ret = btrfs_free_dev_extent(trans, map->stripes[i].dev,
2032                                             map->stripes[i].physical);
2033                 BUG_ON(ret);
2034
2035                 if (map->stripes[i].dev) {
2036                         ret = btrfs_update_device(trans, map->stripes[i].dev);
2037                         BUG_ON(ret);
2038                 }
2039         }
2040         ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
2041                                chunk_offset);
2042
2043         BUG_ON(ret);
2044
2045         trace_btrfs_chunk_free(root, map, chunk_offset, em->len);
2046
2047         if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2048                 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
2049                 BUG_ON(ret);
2050         }
2051
2052         ret = btrfs_remove_block_group(trans, extent_root, chunk_offset);
2053         BUG_ON(ret);
2054
2055         write_lock(&em_tree->lock);
2056         remove_extent_mapping(em_tree, em);
2057         write_unlock(&em_tree->lock);
2058
2059         kfree(map);
2060         em->bdev = NULL;
2061
2062         /* once for the tree */
2063         free_extent_map(em);
2064         /* once for us */
2065         free_extent_map(em);
2066
2067         unlock_chunks(root);
2068         btrfs_end_transaction(trans, root);
2069         return 0;
2070 }
2071
2072 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
2073 {
2074         struct btrfs_root *chunk_root = root->fs_info->chunk_root;
2075         struct btrfs_path *path;
2076         struct extent_buffer *leaf;
2077         struct btrfs_chunk *chunk;
2078         struct btrfs_key key;
2079         struct btrfs_key found_key;
2080         u64 chunk_tree = chunk_root->root_key.objectid;
2081         u64 chunk_type;
2082         bool retried = false;
2083         int failed = 0;
2084         int ret;
2085
2086         path = btrfs_alloc_path();
2087         if (!path)
2088                 return -ENOMEM;
2089
2090 again:
2091         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2092         key.offset = (u64)-1;
2093         key.type = BTRFS_CHUNK_ITEM_KEY;
2094
2095         while (1) {
2096                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2097                 if (ret < 0)
2098                         goto error;
2099                 BUG_ON(ret == 0); /* Corruption */
2100
2101                 ret = btrfs_previous_item(chunk_root, path, key.objectid,
2102                                           key.type);
2103                 if (ret < 0)
2104                         goto error;
2105                 if (ret > 0)
2106                         break;
2107
2108                 leaf = path->nodes[0];
2109                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2110
2111                 chunk = btrfs_item_ptr(leaf, path->slots[0],
2112                                        struct btrfs_chunk);
2113                 chunk_type = btrfs_chunk_type(leaf, chunk);
2114                 btrfs_release_path(path);
2115
2116                 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
2117                         ret = btrfs_relocate_chunk(chunk_root, chunk_tree,
2118                                                    found_key.objectid,
2119                                                    found_key.offset);
2120                         if (ret == -ENOSPC)
2121                                 failed++;
2122                         else if (ret)
2123                                 BUG();
2124                 }
2125
2126                 if (found_key.offset == 0)
2127                         break;
2128                 key.offset = found_key.offset - 1;
2129         }
2130         ret = 0;
2131         if (failed && !retried) {
2132                 failed = 0;
2133                 retried = true;
2134                 goto again;
2135         } else if (failed && retried) {
2136                 WARN_ON(1);
2137                 ret = -ENOSPC;
2138         }
2139 error:
2140         btrfs_free_path(path);
2141         return ret;
2142 }
2143
2144 static int insert_balance_item(struct btrfs_root *root,
2145                                struct btrfs_balance_control *bctl)
2146 {
2147         struct btrfs_trans_handle *trans;
2148         struct btrfs_balance_item *item;
2149         struct btrfs_disk_balance_args disk_bargs;
2150         struct btrfs_path *path;
2151         struct extent_buffer *leaf;
2152         struct btrfs_key key;
2153         int ret, err;
2154
2155         path = btrfs_alloc_path();
2156         if (!path)
2157                 return -ENOMEM;
2158
2159         trans = btrfs_start_transaction(root, 0);
2160         if (IS_ERR(trans)) {
2161                 btrfs_free_path(path);
2162                 return PTR_ERR(trans);
2163         }
2164
2165         key.objectid = BTRFS_BALANCE_OBJECTID;
2166         key.type = BTRFS_BALANCE_ITEM_KEY;
2167         key.offset = 0;
2168
2169         ret = btrfs_insert_empty_item(trans, root, path, &key,
2170                                       sizeof(*item));
2171         if (ret)
2172                 goto out;
2173
2174         leaf = path->nodes[0];
2175         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
2176
2177         memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
2178
2179         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
2180         btrfs_set_balance_data(leaf, item, &disk_bargs);
2181         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
2182         btrfs_set_balance_meta(leaf, item, &disk_bargs);
2183         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
2184         btrfs_set_balance_sys(leaf, item, &disk_bargs);
2185
2186         btrfs_set_balance_flags(leaf, item, bctl->flags);
2187
2188         btrfs_mark_buffer_dirty(leaf);
2189 out:
2190         btrfs_free_path(path);
2191         err = btrfs_commit_transaction(trans, root);
2192         if (err && !ret)
2193                 ret = err;
2194         return ret;
2195 }
2196
2197 static int del_balance_item(struct btrfs_root *root)
2198 {
2199         struct btrfs_trans_handle *trans;
2200         struct btrfs_path *path;
2201         struct btrfs_key key;
2202         int ret, err;
2203
2204         path = btrfs_alloc_path();
2205         if (!path)
2206                 return -ENOMEM;
2207
2208         trans = btrfs_start_transaction(root, 0);
2209         if (IS_ERR(trans)) {
2210                 btrfs_free_path(path);
2211                 return PTR_ERR(trans);
2212         }
2213
2214         key.objectid = BTRFS_BALANCE_OBJECTID;
2215         key.type = BTRFS_BALANCE_ITEM_KEY;
2216         key.offset = 0;
2217
2218         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2219         if (ret < 0)
2220                 goto out;
2221         if (ret > 0) {
2222                 ret = -ENOENT;
2223                 goto out;
2224         }
2225
2226         ret = btrfs_del_item(trans, root, path);
2227 out:
2228         btrfs_free_path(path);
2229         err = btrfs_commit_transaction(trans, root);
2230         if (err && !ret)
2231                 ret = err;
2232         return ret;
2233 }
2234
2235 /*
2236  * This is a heuristic used to reduce the number of chunks balanced on
2237  * resume after balance was interrupted.
2238  */
2239 static void update_balance_args(struct btrfs_balance_control *bctl)
2240 {
2241         /*
2242          * Turn on soft mode for chunk types that were being converted.
2243          */
2244         if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
2245                 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
2246         if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
2247                 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
2248         if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
2249                 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
2250
2251         /*
2252          * Turn on usage filter if is not already used.  The idea is
2253          * that chunks that we have already balanced should be
2254          * reasonably full.  Don't do it for chunks that are being
2255          * converted - that will keep us from relocating unconverted
2256          * (albeit full) chunks.
2257          */
2258         if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2259             !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2260                 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
2261                 bctl->data.usage = 90;
2262         }
2263         if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2264             !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2265                 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
2266                 bctl->sys.usage = 90;
2267         }
2268         if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2269             !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2270                 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
2271                 bctl->meta.usage = 90;
2272         }
2273 }
2274
2275 /*
2276  * Should be called with both balance and volume mutexes held to
2277  * serialize other volume operations (add_dev/rm_dev/resize) with
2278  * restriper.  Same goes for unset_balance_control.
2279  */
2280 static void set_balance_control(struct btrfs_balance_control *bctl)
2281 {
2282         struct btrfs_fs_info *fs_info = bctl->fs_info;
2283
2284         BUG_ON(fs_info->balance_ctl);
2285
2286         spin_lock(&fs_info->balance_lock);
2287         fs_info->balance_ctl = bctl;
2288         spin_unlock(&fs_info->balance_lock);
2289 }
2290
2291 static void unset_balance_control(struct btrfs_fs_info *fs_info)
2292 {
2293         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
2294
2295         BUG_ON(!fs_info->balance_ctl);
2296
2297         spin_lock(&fs_info->balance_lock);
2298         fs_info->balance_ctl = NULL;
2299         spin_unlock(&fs_info->balance_lock);
2300
2301         kfree(bctl);
2302 }
2303
2304 /*
2305  * Balance filters.  Return 1 if chunk should be filtered out
2306  * (should not be balanced).
2307  */
2308 static int chunk_profiles_filter(u64 chunk_type,
2309                                  struct btrfs_balance_args *bargs)
2310 {
2311         chunk_type = chunk_to_extended(chunk_type) &
2312                                 BTRFS_EXTENDED_PROFILE_MASK;
2313
2314         if (bargs->profiles & chunk_type)
2315                 return 0;
2316
2317         return 1;
2318 }
2319
2320 static u64 div_factor_fine(u64 num, int factor)
2321 {
2322         if (factor <= 0)
2323                 return 0;
2324         if (factor >= 100)
2325                 return num;
2326
2327         num *= factor;
2328         do_div(num, 100);
2329         return num;
2330 }
2331
2332 static int chunk_usage_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
2333                               struct btrfs_balance_args *bargs)
2334 {
2335         struct btrfs_block_group_cache *cache;
2336         u64 chunk_used, user_thresh;
2337         int ret = 1;
2338
2339         cache = btrfs_lookup_block_group(fs_info, chunk_offset);
2340         chunk_used = btrfs_block_group_used(&cache->item);
2341
2342         user_thresh = div_factor_fine(cache->key.offset, bargs->usage);
2343         if (chunk_used < user_thresh)
2344                 ret = 0;
2345
2346         btrfs_put_block_group(cache);
2347         return ret;
2348 }
2349
2350 static int chunk_devid_filter(struct extent_buffer *leaf,
2351                               struct btrfs_chunk *chunk,
2352                               struct btrfs_balance_args *bargs)
2353 {
2354         struct btrfs_stripe *stripe;
2355         int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2356         int i;
2357
2358         for (i = 0; i < num_stripes; i++) {
2359                 stripe = btrfs_stripe_nr(chunk, i);
2360                 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
2361                         return 0;
2362         }
2363
2364         return 1;
2365 }
2366
2367 /* [pstart, pend) */
2368 static int chunk_drange_filter(struct extent_buffer *leaf,
2369                                struct btrfs_chunk *chunk,
2370                                u64 chunk_offset,
2371                                struct btrfs_balance_args *bargs)
2372 {
2373         struct btrfs_stripe *stripe;
2374         int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2375         u64 stripe_offset;
2376         u64 stripe_length;
2377         int factor;
2378         int i;
2379
2380         if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
2381                 return 0;
2382
2383         if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP |
2384              BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10))
2385                 factor = 2;
2386         else
2387                 factor = 1;
2388         factor = num_stripes / factor;
2389
2390         for (i = 0; i < num_stripes; i++) {
2391                 stripe = btrfs_stripe_nr(chunk, i);
2392                 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
2393                         continue;
2394
2395                 stripe_offset = btrfs_stripe_offset(leaf, stripe);
2396                 stripe_length = btrfs_chunk_length(leaf, chunk);
2397                 do_div(stripe_length, factor);
2398
2399                 if (stripe_offset < bargs->pend &&
2400                     stripe_offset + stripe_length > bargs->pstart)
2401                         return 0;
2402         }
2403
2404         return 1;
2405 }
2406
2407 /* [vstart, vend) */
2408 static int chunk_vrange_filter(struct extent_buffer *leaf,
2409                                struct btrfs_chunk *chunk,
2410                                u64 chunk_offset,
2411                                struct btrfs_balance_args *bargs)
2412 {
2413         if (chunk_offset < bargs->vend &&
2414             chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
2415                 /* at least part of the chunk is inside this vrange */
2416                 return 0;
2417
2418         return 1;
2419 }
2420
2421 static int chunk_soft_convert_filter(u64 chunk_type,
2422                                      struct btrfs_balance_args *bargs)
2423 {
2424         if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
2425                 return 0;
2426
2427         chunk_type = chunk_to_extended(chunk_type) &
2428                                 BTRFS_EXTENDED_PROFILE_MASK;
2429
2430         if (bargs->target == chunk_type)
2431                 return 1;
2432
2433         return 0;
2434 }
2435
2436 static int should_balance_chunk(struct btrfs_root *root,
2437                                 struct extent_buffer *leaf,
2438                                 struct btrfs_chunk *chunk, u64 chunk_offset)
2439 {
2440         struct btrfs_balance_control *bctl = root->fs_info->balance_ctl;
2441         struct btrfs_balance_args *bargs = NULL;
2442         u64 chunk_type = btrfs_chunk_type(leaf, chunk);
2443
2444         /* type filter */
2445         if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
2446               (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
2447                 return 0;
2448         }
2449
2450         if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
2451                 bargs = &bctl->data;
2452         else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
2453                 bargs = &bctl->sys;
2454         else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
2455                 bargs = &bctl->meta;
2456
2457         /* profiles filter */
2458         if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
2459             chunk_profiles_filter(chunk_type, bargs)) {
2460                 return 0;
2461         }
2462
2463         /* usage filter */
2464         if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
2465             chunk_usage_filter(bctl->fs_info, chunk_offset, bargs)) {
2466                 return 0;
2467         }
2468
2469         /* devid filter */
2470         if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
2471             chunk_devid_filter(leaf, chunk, bargs)) {
2472                 return 0;
2473         }
2474
2475         /* drange filter, makes sense only with devid filter */
2476         if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
2477             chunk_drange_filter(leaf, chunk, chunk_offset, bargs)) {
2478                 return 0;
2479         }
2480
2481         /* vrange filter */
2482         if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
2483             chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
2484                 return 0;
2485         }
2486
2487         /* soft profile changing mode */
2488         if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
2489             chunk_soft_convert_filter(chunk_type, bargs)) {
2490                 return 0;
2491         }
2492
2493         return 1;
2494 }
2495
2496 static u64 div_factor(u64 num, int factor)
2497 {
2498         if (factor == 10)
2499                 return num;
2500         num *= factor;
2501         do_div(num, 10);
2502         return num;
2503 }
2504
2505 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
2506 {
2507         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
2508         struct btrfs_root *chunk_root = fs_info->chunk_root;
2509         struct btrfs_root *dev_root = fs_info->dev_root;
2510         struct list_head *devices;
2511         struct btrfs_device *device;
2512         u64 old_size;
2513         u64 size_to_free;
2514         struct btrfs_chunk *chunk;
2515         struct btrfs_path *path;
2516         struct btrfs_key key;
2517         struct btrfs_key found_key;
2518         struct btrfs_trans_handle *trans;
2519         struct extent_buffer *leaf;
2520         int slot;
2521         int ret;
2522         int enospc_errors = 0;
2523         bool counting = true;
2524
2525         /* step one make some room on all the devices */
2526         devices = &fs_info->fs_devices->devices;
2527         list_for_each_entry(device, devices, dev_list) {
2528                 old_size = device->total_bytes;
2529                 size_to_free = div_factor(old_size, 1);
2530                 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
2531                 if (!device->writeable ||
2532                     device->total_bytes - device->bytes_used > size_to_free)
2533                         continue;
2534
2535                 ret = btrfs_shrink_device(device, old_size - size_to_free);
2536                 if (ret == -ENOSPC)
2537                         break;
2538                 BUG_ON(ret);
2539
2540                 trans = btrfs_start_transaction(dev_root, 0);
2541                 BUG_ON(IS_ERR(trans));
2542
2543                 ret = btrfs_grow_device(trans, device, old_size);
2544                 BUG_ON(ret);
2545
2546                 btrfs_end_transaction(trans, dev_root);
2547         }
2548
2549         /* step two, relocate all the chunks */
2550         path = btrfs_alloc_path();
2551         if (!path) {
2552                 ret = -ENOMEM;
2553                 goto error;
2554         }
2555
2556         /* zero out stat counters */
2557         spin_lock(&fs_info->balance_lock);
2558         memset(&bctl->stat, 0, sizeof(bctl->stat));
2559         spin_unlock(&fs_info->balance_lock);
2560 again:
2561         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2562         key.offset = (u64)-1;
2563         key.type = BTRFS_CHUNK_ITEM_KEY;
2564
2565         while (1) {
2566                 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
2567                     atomic_read(&fs_info->balance_cancel_req)) {
2568                         ret = -ECANCELED;
2569                         goto error;
2570                 }
2571
2572                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2573                 if (ret < 0)
2574                         goto error;
2575
2576                 /*
2577                  * this shouldn't happen, it means the last relocate
2578                  * failed
2579                  */
2580                 if (ret == 0)
2581                         BUG(); /* FIXME break ? */
2582
2583                 ret = btrfs_previous_item(chunk_root, path, 0,
2584                                           BTRFS_CHUNK_ITEM_KEY);
2585                 if (ret) {
2586                         ret = 0;
2587                         break;
2588                 }
2589
2590                 leaf = path->nodes[0];
2591                 slot = path->slots[0];
2592                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2593
2594                 if (found_key.objectid != key.objectid)
2595                         break;
2596
2597                 /* chunk zero is special */
2598                 if (found_key.offset == 0)
2599                         break;
2600
2601                 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
2602
2603                 if (!counting) {
2604                         spin_lock(&fs_info->balance_lock);
2605                         bctl->stat.considered++;
2606                         spin_unlock(&fs_info->balance_lock);
2607                 }
2608
2609                 ret = should_balance_chunk(chunk_root, leaf, chunk,
2610                                            found_key.offset);
2611                 btrfs_release_path(path);
2612                 if (!ret)
2613                         goto loop;
2614
2615                 if (counting) {
2616                         spin_lock(&fs_info->balance_lock);
2617                         bctl->stat.expected++;
2618                         spin_unlock(&fs_info->balance_lock);
2619                         goto loop;
2620                 }
2621
2622                 ret = btrfs_relocate_chunk(chunk_root,
2623                                            chunk_root->root_key.objectid,
2624                                            found_key.objectid,
2625                                            found_key.offset);
2626                 if (ret && ret != -ENOSPC)
2627                         goto error;
2628                 if (ret == -ENOSPC) {
2629                         enospc_errors++;
2630                 } else {
2631                         spin_lock(&fs_info->balance_lock);
2632                         bctl->stat.completed++;
2633                         spin_unlock(&fs_info->balance_lock);
2634                 }
2635 loop:
2636                 key.offset = found_key.offset - 1;
2637         }
2638
2639         if (counting) {
2640                 btrfs_release_path(path);
2641                 counting = false;
2642                 goto again;
2643         }
2644 error:
2645         btrfs_free_path(path);
2646         if (enospc_errors) {
2647                 printk(KERN_INFO "btrfs: %d enospc errors during balance\n",
2648                        enospc_errors);
2649                 if (!ret)
2650                         ret = -ENOSPC;
2651         }
2652
2653         return ret;
2654 }
2655
2656 /**
2657  * alloc_profile_is_valid - see if a given profile is valid and reduced
2658  * @flags: profile to validate
2659  * @extended: if true @flags is treated as an extended profile
2660  */
2661 static int alloc_profile_is_valid(u64 flags, int extended)
2662 {
2663         u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
2664                                BTRFS_BLOCK_GROUP_PROFILE_MASK);
2665
2666         flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
2667
2668         /* 1) check that all other bits are zeroed */
2669         if (flags & ~mask)
2670                 return 0;
2671
2672         /* 2) see if profile is reduced */
2673         if (flags == 0)
2674                 return !extended; /* "0" is valid for usual profiles */
2675
2676         /* true if exactly one bit set */
2677         return (flags & (flags - 1)) == 0;
2678 }
2679
2680 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
2681 {
2682         /* cancel requested || normal exit path */
2683         return atomic_read(&fs_info->balance_cancel_req) ||
2684                 (atomic_read(&fs_info->balance_pause_req) == 0 &&
2685                  atomic_read(&fs_info->balance_cancel_req) == 0);
2686 }
2687
2688 static void __cancel_balance(struct btrfs_fs_info *fs_info)
2689 {
2690         int ret;
2691
2692         unset_balance_control(fs_info);
2693         ret = del_balance_item(fs_info->tree_root);
2694         BUG_ON(ret);
2695 }
2696
2697 void update_ioctl_balance_args(struct btrfs_fs_info *fs_info, int lock,
2698                                struct btrfs_ioctl_balance_args *bargs);
2699
2700 /*
2701  * Should be called with both balance and volume mutexes held
2702  */
2703 int btrfs_balance(struct btrfs_balance_control *bctl,
2704                   struct btrfs_ioctl_balance_args *bargs)
2705 {
2706         struct btrfs_fs_info *fs_info = bctl->fs_info;
2707         u64 allowed;
2708         int mixed = 0;
2709         int ret;
2710
2711         if (btrfs_fs_closing(fs_info) ||
2712             atomic_read(&fs_info->balance_pause_req) ||
2713             atomic_read(&fs_info->balance_cancel_req)) {
2714                 ret = -EINVAL;
2715                 goto out;
2716         }
2717
2718         allowed = btrfs_super_incompat_flags(fs_info->super_copy);
2719         if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
2720                 mixed = 1;
2721
2722         /*
2723          * In case of mixed groups both data and meta should be picked,
2724          * and identical options should be given for both of them.
2725          */
2726         allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
2727         if (mixed && (bctl->flags & allowed)) {
2728                 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
2729                     !(bctl->flags & BTRFS_BALANCE_METADATA) ||
2730                     memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
2731                         printk(KERN_ERR "btrfs: with mixed groups data and "
2732                                "metadata balance options must be the same\n");
2733                         ret = -EINVAL;
2734                         goto out;
2735                 }
2736         }
2737
2738         allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE;
2739         if (fs_info->fs_devices->num_devices == 1)
2740                 allowed |= BTRFS_BLOCK_GROUP_DUP;
2741         else if (fs_info->fs_devices->num_devices < 4)
2742                 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1);
2743         else
2744                 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
2745                                 BTRFS_BLOCK_GROUP_RAID10);
2746
2747         if ((bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
2748             (!alloc_profile_is_valid(bctl->data.target, 1) ||
2749              (bctl->data.target & ~allowed))) {
2750                 printk(KERN_ERR "btrfs: unable to start balance with target "
2751                        "data profile %llu\n",
2752                        (unsigned long long)bctl->data.target);
2753                 ret = -EINVAL;
2754                 goto out;
2755         }
2756         if ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
2757             (!alloc_profile_is_valid(bctl->meta.target, 1) ||
2758              (bctl->meta.target & ~allowed))) {
2759                 printk(KERN_ERR "btrfs: unable to start balance with target "
2760                        "metadata profile %llu\n",
2761                        (unsigned long long)bctl->meta.target);
2762                 ret = -EINVAL;
2763                 goto out;
2764         }
2765         if ((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
2766             (!alloc_profile_is_valid(bctl->sys.target, 1) ||
2767              (bctl->sys.target & ~allowed))) {
2768                 printk(KERN_ERR "btrfs: unable to start balance with target "
2769                        "system profile %llu\n",
2770                        (unsigned long long)bctl->sys.target);
2771                 ret = -EINVAL;
2772                 goto out;
2773         }
2774
2775         /* allow dup'ed data chunks only in mixed mode */
2776         if (!mixed && (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
2777             (bctl->data.target & BTRFS_BLOCK_GROUP_DUP)) {
2778                 printk(KERN_ERR "btrfs: dup for data is not allowed\n");
2779                 ret = -EINVAL;
2780                 goto out;
2781         }
2782
2783         /* allow to reduce meta or sys integrity only if force set */
2784         allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
2785                         BTRFS_BLOCK_GROUP_RAID10;
2786         if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
2787              (fs_info->avail_system_alloc_bits & allowed) &&
2788              !(bctl->sys.target & allowed)) ||
2789             ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
2790              (fs_info->avail_metadata_alloc_bits & allowed) &&
2791              !(bctl->meta.target & allowed))) {
2792                 if (bctl->flags & BTRFS_BALANCE_FORCE) {
2793                         printk(KERN_INFO "btrfs: force reducing metadata "
2794                                "integrity\n");
2795                 } else {
2796                         printk(KERN_ERR "btrfs: balance will reduce metadata "
2797                                "integrity, use force if you want this\n");
2798                         ret = -EINVAL;
2799                         goto out;
2800                 }
2801         }
2802
2803         ret = insert_balance_item(fs_info->tree_root, bctl);
2804         if (ret && ret != -EEXIST)
2805                 goto out;
2806
2807         if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
2808                 BUG_ON(ret == -EEXIST);
2809                 set_balance_control(bctl);
2810         } else {
2811                 BUG_ON(ret != -EEXIST);
2812                 spin_lock(&fs_info->balance_lock);
2813                 update_balance_args(bctl);
2814                 spin_unlock(&fs_info->balance_lock);
2815         }
2816
2817         atomic_inc(&fs_info->balance_running);
2818         mutex_unlock(&fs_info->balance_mutex);
2819
2820         ret = __btrfs_balance(fs_info);
2821
2822         mutex_lock(&fs_info->balance_mutex);
2823         atomic_dec(&fs_info->balance_running);
2824
2825         if (bargs) {
2826                 memset(bargs, 0, sizeof(*bargs));
2827                 update_ioctl_balance_args(fs_info, 0, bargs);
2828         }
2829
2830         if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
2831             balance_need_close(fs_info)) {
2832                 __cancel_balance(fs_info);
2833         }
2834
2835         wake_up(&fs_info->balance_wait_q);
2836
2837         return ret;
2838 out:
2839         if (bctl->flags & BTRFS_BALANCE_RESUME)
2840                 __cancel_balance(fs_info);
2841         else
2842                 kfree(bctl);
2843         return ret;
2844 }
2845
2846 static int balance_kthread(void *data)
2847 {
2848         struct btrfs_fs_info *fs_info = data;
2849         int ret = 0;
2850
2851         mutex_lock(&fs_info->volume_mutex);
2852         mutex_lock(&fs_info->balance_mutex);
2853
2854         if (fs_info->balance_ctl) {
2855                 printk(KERN_INFO "btrfs: continuing balance\n");
2856                 ret = btrfs_balance(fs_info->balance_ctl, NULL);
2857         }
2858
2859         mutex_unlock(&fs_info->balance_mutex);
2860         mutex_unlock(&fs_info->volume_mutex);
2861
2862         return ret;
2863 }
2864
2865 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
2866 {
2867         struct task_struct *tsk;
2868
2869         spin_lock(&fs_info->balance_lock);
2870         if (!fs_info->balance_ctl) {
2871                 spin_unlock(&fs_info->balance_lock);
2872                 return 0;
2873         }
2874         spin_unlock(&fs_info->balance_lock);
2875
2876         if (btrfs_test_opt(fs_info->tree_root, SKIP_BALANCE)) {
2877                 printk(KERN_INFO "btrfs: force skipping balance\n");
2878                 return 0;
2879         }
2880
2881         tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
2882         if (IS_ERR(tsk))
2883                 return PTR_ERR(tsk);
2884
2885         return 0;
2886 }
2887
2888 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
2889 {
2890         struct btrfs_balance_control *bctl;
2891         struct btrfs_balance_item *item;
2892         struct btrfs_disk_balance_args disk_bargs;
2893         struct btrfs_path *path;
2894         struct extent_buffer *leaf;
2895         struct btrfs_key key;
2896         int ret;
2897
2898         path = btrfs_alloc_path();
2899         if (!path)
2900                 return -ENOMEM;
2901
2902         key.objectid = BTRFS_BALANCE_OBJECTID;
2903         key.type = BTRFS_BALANCE_ITEM_KEY;
2904         key.offset = 0;
2905
2906         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
2907         if (ret < 0)
2908                 goto out;
2909         if (ret > 0) { /* ret = -ENOENT; */
2910                 ret = 0;
2911                 goto out;
2912         }
2913
2914         bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
2915         if (!bctl) {
2916                 ret = -ENOMEM;
2917                 goto out;
2918         }
2919
2920         leaf = path->nodes[0];
2921         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
2922
2923         bctl->fs_info = fs_info;
2924         bctl->flags = btrfs_balance_flags(leaf, item);
2925         bctl->flags |= BTRFS_BALANCE_RESUME;
2926
2927         btrfs_balance_data(leaf, item, &disk_bargs);
2928         btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
2929         btrfs_balance_meta(leaf, item, &disk_bargs);
2930         btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
2931         btrfs_balance_sys(leaf, item, &disk_bargs);
2932         btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
2933
2934         mutex_lock(&fs_info->volume_mutex);
2935         mutex_lock(&fs_info->balance_mutex);
2936
2937         set_balance_control(bctl);
2938
2939         mutex_unlock(&fs_info->balance_mutex);
2940         mutex_unlock(&fs_info->volume_mutex);
2941 out:
2942         btrfs_free_path(path);
2943         return ret;
2944 }
2945
2946 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
2947 {
2948         int ret = 0;
2949
2950         mutex_lock(&fs_info->balance_mutex);
2951         if (!fs_info->balance_ctl) {
2952                 mutex_unlock(&fs_info->balance_mutex);
2953                 return -ENOTCONN;
2954         }
2955
2956         if (atomic_read(&fs_info->balance_running)) {
2957                 atomic_inc(&fs_info->balance_pause_req);
2958                 mutex_unlock(&fs_info->balance_mutex);
2959
2960                 wait_event(fs_info->balance_wait_q,
2961                            atomic_read(&fs_info->balance_running) == 0);
2962
2963                 mutex_lock(&fs_info->balance_mutex);
2964                 /* we are good with balance_ctl ripped off from under us */
2965                 BUG_ON(atomic_read(&fs_info->balance_running));
2966                 atomic_dec(&fs_info->balance_pause_req);
2967         } else {
2968                 ret = -ENOTCONN;
2969         }
2970
2971         mutex_unlock(&fs_info->balance_mutex);
2972         return ret;
2973 }
2974
2975 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
2976 {
2977         mutex_lock(&fs_info->balance_mutex);
2978         if (!fs_info->balance_ctl) {
2979                 mutex_unlock(&fs_info->balance_mutex);
2980                 return -ENOTCONN;
2981         }
2982
2983         atomic_inc(&fs_info->balance_cancel_req);
2984         /*
2985          * if we are running just wait and return, balance item is
2986          * deleted in btrfs_balance in this case
2987          */
2988         if (atomic_read(&fs_info->balance_running)) {
2989                 mutex_unlock(&fs_info->balance_mutex);
2990                 wait_event(fs_info->balance_wait_q,
2991                            atomic_read(&fs_info->balance_running) == 0);
2992                 mutex_lock(&fs_info->balance_mutex);
2993         } else {
2994                 /* __cancel_balance needs volume_mutex */
2995                 mutex_unlock(&fs_info->balance_mutex);
2996                 mutex_lock(&fs_info->volume_mutex);
2997                 mutex_lock(&fs_info->balance_mutex);
2998
2999                 if (fs_info->balance_ctl)
3000                         __cancel_balance(fs_info);
3001
3002                 mutex_unlock(&fs_info->volume_mutex);
3003         }
3004
3005         BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running));
3006         atomic_dec(&fs_info->balance_cancel_req);
3007         mutex_unlock(&fs_info->balance_mutex);
3008         return 0;
3009 }
3010
3011 /*
3012  * shrinking a device means finding all of the device extents past
3013  * the new size, and then following the back refs to the chunks.
3014  * The chunk relocation code actually frees the device extent
3015  */
3016 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
3017 {
3018         struct btrfs_trans_handle *trans;
3019         struct btrfs_root *root = device->dev_root;
3020         struct btrfs_dev_extent *dev_extent = NULL;
3021         struct btrfs_path *path;
3022         u64 length;
3023         u64 chunk_tree;
3024         u64 chunk_objectid;
3025         u64 chunk_offset;
3026         int ret;
3027         int slot;
3028         int failed = 0;
3029         bool retried = false;
3030         struct extent_buffer *l;
3031         struct btrfs_key key;
3032         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
3033         u64 old_total = btrfs_super_total_bytes(super_copy);
3034         u64 old_size = device->total_bytes;
3035         u64 diff = device->total_bytes - new_size;
3036
3037         if (new_size >= device->total_bytes)
3038                 return -EINVAL;
3039
3040         path = btrfs_alloc_path();
3041         if (!path)
3042                 return -ENOMEM;
3043
3044         path->reada = 2;
3045
3046         lock_chunks(root);
3047
3048         device->total_bytes = new_size;
3049         if (device->writeable) {
3050                 device->fs_devices->total_rw_bytes -= diff;
3051                 spin_lock(&root->fs_info->free_chunk_lock);
3052                 root->fs_info->free_chunk_space -= diff;
3053                 spin_unlock(&root->fs_info->free_chunk_lock);
3054         }
3055         unlock_chunks(root);
3056
3057 again:
3058         key.objectid = device->devid;
3059         key.offset = (u64)-1;
3060         key.type = BTRFS_DEV_EXTENT_KEY;
3061
3062         do {
3063                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3064                 if (ret < 0)
3065                         goto done;
3066
3067                 ret = btrfs_previous_item(root, path, 0, key.type);
3068                 if (ret < 0)
3069                         goto done;
3070                 if (ret) {
3071                         ret = 0;
3072                         btrfs_release_path(path);
3073                         break;
3074                 }
3075
3076                 l = path->nodes[0];
3077                 slot = path->slots[0];
3078                 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
3079
3080                 if (key.objectid != device->devid) {
3081                         btrfs_release_path(path);
3082                         break;
3083                 }
3084
3085                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
3086                 length = btrfs_dev_extent_length(l, dev_extent);
3087
3088                 if (key.offset + length <= new_size) {
3089                         btrfs_release_path(path);
3090                         break;
3091                 }
3092
3093                 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
3094                 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
3095                 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
3096                 btrfs_release_path(path);
3097
3098                 ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
3099                                            chunk_offset);
3100                 if (ret && ret != -ENOSPC)
3101                         goto done;
3102                 if (ret == -ENOSPC)
3103                         failed++;
3104         } while (key.offset-- > 0);
3105
3106         if (failed && !retried) {
3107                 failed = 0;
3108                 retried = true;
3109                 goto again;
3110         } else if (failed && retried) {
3111                 ret = -ENOSPC;
3112                 lock_chunks(root);
3113
3114                 device->total_bytes = old_size;
3115                 if (device->writeable)
3116                         device->fs_devices->total_rw_bytes += diff;
3117                 spin_lock(&root->fs_info->free_chunk_lock);
3118                 root->fs_info->free_chunk_space += diff;
3119                 spin_unlock(&root->fs_info->free_chunk_lock);
3120                 unlock_chunks(root);
3121                 goto done;
3122         }
3123
3124         /* Shrinking succeeded, else we would be at "done". */
3125         trans = btrfs_start_transaction(root, 0);
3126         if (IS_ERR(trans)) {
3127                 ret = PTR_ERR(trans);
3128                 goto done;
3129         }
3130
3131         lock_chunks(root);
3132
3133         device->disk_total_bytes = new_size;
3134         /* Now btrfs_update_device() will change the on-disk size. */
3135         ret = btrfs_update_device(trans, device);
3136         if (ret) {
3137                 unlock_chunks(root);
3138                 btrfs_end_transaction(trans, root);
3139                 goto done;
3140         }
3141         WARN_ON(diff > old_total);
3142         btrfs_set_super_total_bytes(super_copy, old_total - diff);
3143         unlock_chunks(root);
3144         btrfs_end_transaction(trans, root);
3145 done:
3146         btrfs_free_path(path);
3147         return ret;
3148 }
3149
3150 static int btrfs_add_system_chunk(struct btrfs_root *root,
3151                            struct btrfs_key *key,
3152                            struct btrfs_chunk *chunk, int item_size)
3153 {
3154         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
3155         struct btrfs_disk_key disk_key;
3156         u32 array_size;
3157         u8 *ptr;
3158
3159         array_size = btrfs_super_sys_array_size(super_copy);
3160         if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
3161                 return -EFBIG;
3162
3163         ptr = super_copy->sys_chunk_array + array_size;
3164         btrfs_cpu_key_to_disk(&disk_key, key);
3165         memcpy(ptr, &disk_key, sizeof(disk_key));
3166         ptr += sizeof(disk_key);
3167         memcpy(ptr, chunk, item_size);
3168         item_size += sizeof(disk_key);
3169         btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
3170         return 0;
3171 }
3172
3173 /*
3174  * sort the devices in descending order by max_avail, total_avail
3175  */
3176 static int btrfs_cmp_device_info(const void *a, const void *b)
3177 {
3178         const struct btrfs_device_info *di_a = a;
3179         const struct btrfs_device_info *di_b = b;
3180
3181         if (di_a->max_avail > di_b->max_avail)
3182                 return -1;
3183         if (di_a->max_avail < di_b->max_avail)
3184                 return 1;
3185         if (di_a->total_avail > di_b->total_avail)
3186                 return -1;
3187         if (di_a->total_avail < di_b->total_avail)
3188                 return 1;
3189         return 0;
3190 }
3191
3192 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
3193                                struct btrfs_root *extent_root,
3194                                struct map_lookup **map_ret,
3195                                u64 *num_bytes_out, u64 *stripe_size_out,
3196                                u64 start, u64 type)
3197 {
3198         struct btrfs_fs_info *info = extent_root->fs_info;
3199         struct btrfs_fs_devices *fs_devices = info->fs_devices;
3200         struct list_head *cur;
3201         struct map_lookup *map = NULL;
3202         struct extent_map_tree *em_tree;
3203         struct extent_map *em;
3204         struct btrfs_device_info *devices_info = NULL;
3205         u64 total_avail;
3206         int num_stripes;        /* total number of stripes to allocate */
3207         int sub_stripes;        /* sub_stripes info for map */
3208         int dev_stripes;        /* stripes per dev */
3209         int devs_max;           /* max devs to use */
3210         int devs_min;           /* min devs needed */
3211         int devs_increment;     /* ndevs has to be a multiple of this */
3212         int ncopies;            /* how many copies to data has */
3213         int ret;
3214         u64 max_stripe_size;
3215         u64 max_chunk_size;
3216         u64 stripe_size;
3217         u64 num_bytes;
3218         int ndevs;
3219         int i;
3220         int j;
3221
3222         BUG_ON(!alloc_profile_is_valid(type, 0));
3223
3224         if (list_empty(&fs_devices->alloc_list))
3225                 return -ENOSPC;
3226
3227         sub_stripes = 1;
3228         dev_stripes = 1;
3229         devs_increment = 1;
3230         ncopies = 1;
3231         devs_max = 0;   /* 0 == as many as possible */
3232         devs_min = 1;
3233
3234         /*
3235          * define the properties of each RAID type.
3236          * FIXME: move this to a global table and use it in all RAID
3237          * calculation code
3238          */
3239         if (type & (BTRFS_BLOCK_GROUP_DUP)) {
3240                 dev_stripes = 2;
3241                 ncopies = 2;
3242                 devs_max = 1;
3243         } else if (type & (BTRFS_BLOCK_GROUP_RAID0)) {
3244                 devs_min = 2;
3245         } else if (type & (BTRFS_BLOCK_GROUP_RAID1)) {
3246                 devs_increment = 2;
3247                 ncopies = 2;
3248                 devs_max = 2;
3249                 devs_min = 2;
3250         } else if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
3251                 sub_stripes = 2;
3252                 devs_increment = 2;
3253                 ncopies = 2;
3254                 devs_min = 4;
3255         } else {
3256                 devs_max = 1;
3257         }
3258
3259         if (type & BTRFS_BLOCK_GROUP_DATA) {
3260                 max_stripe_size = 1024 * 1024 * 1024;
3261                 max_chunk_size = 10 * max_stripe_size;
3262         } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
3263                 /* for larger filesystems, use larger metadata chunks */
3264                 if (fs_devices->total_rw_bytes > 50ULL * 1024 * 1024 * 1024)
3265                         max_stripe_size = 1024 * 1024 * 1024;
3266                 else
3267                         max_stripe_size = 256 * 1024 * 1024;
3268                 max_chunk_size = max_stripe_size;
3269         } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
3270                 max_stripe_size = 32 * 1024 * 1024;
3271                 max_chunk_size = 2 * max_stripe_size;
3272         } else {
3273                 printk(KERN_ERR "btrfs: invalid chunk type 0x%llx requested\n",
3274                        type);
3275                 BUG_ON(1);
3276         }
3277
3278         /* we don't want a chunk larger than 10% of writeable space */
3279         max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
3280                              max_chunk_size);
3281
3282         devices_info = kzalloc(sizeof(*devices_info) * fs_devices->rw_devices,
3283                                GFP_NOFS);
3284         if (!devices_info)
3285                 return -ENOMEM;
3286
3287         cur = fs_devices->alloc_list.next;
3288
3289         /*
3290          * in the first pass through the devices list, we gather information
3291          * about the available holes on each device.
3292          */
3293         ndevs = 0;
3294         while (cur != &fs_devices->alloc_list) {
3295                 struct btrfs_device *device;
3296                 u64 max_avail;
3297                 u64 dev_offset;
3298
3299                 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
3300
3301                 cur = cur->next;
3302
3303                 if (!device->writeable) {
3304                         printk(KERN_ERR
3305                                "btrfs: read-only device in alloc_list\n");
3306                         WARN_ON(1);
3307                         continue;
3308                 }
3309
3310                 if (!device->in_fs_metadata)
3311                         continue;
3312
3313                 if (device->total_bytes > device->bytes_used)
3314                         total_avail = device->total_bytes - device->bytes_used;
3315                 else
3316                         total_avail = 0;
3317
3318                 /* If there is no space on this device, skip it. */
3319                 if (total_avail == 0)
3320                         continue;
3321
3322                 ret = find_free_dev_extent(device,
3323                                            max_stripe_size * dev_stripes,
3324                                            &dev_offset, &max_avail);
3325                 if (ret && ret != -ENOSPC)
3326                         goto error;
3327
3328                 if (ret == 0)
3329                         max_avail = max_stripe_size * dev_stripes;
3330
3331                 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes)
3332                         continue;
3333
3334                 devices_info[ndevs].dev_offset = dev_offset;
3335                 devices_info[ndevs].max_avail = max_avail;
3336                 devices_info[ndevs].total_avail = total_avail;
3337                 devices_info[ndevs].dev = device;
3338                 ++ndevs;
3339         }
3340
3341         /*
3342          * now sort the devices by hole size / available space
3343          */
3344         sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
3345              btrfs_cmp_device_info, NULL);
3346
3347         /* round down to number of usable stripes */
3348         ndevs -= ndevs % devs_increment;
3349
3350         if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) {
3351                 ret = -ENOSPC;
3352                 goto error;
3353         }
3354
3355         if (devs_max && ndevs > devs_max)
3356                 ndevs = devs_max;
3357         /*
3358          * the primary goal is to maximize the number of stripes, so use as many
3359          * devices as possible, even if the stripes are not maximum sized.
3360          */
3361         stripe_size = devices_info[ndevs-1].max_avail;
3362         num_stripes = ndevs * dev_stripes;
3363
3364         if (stripe_size * ndevs > max_chunk_size * ncopies) {
3365                 stripe_size = max_chunk_size * ncopies;
3366                 do_div(stripe_size, ndevs);
3367         }
3368
3369         do_div(stripe_size, dev_stripes);
3370
3371         /* align to BTRFS_STRIPE_LEN */
3372         do_div(stripe_size, BTRFS_STRIPE_LEN);
3373         stripe_size *= BTRFS_STRIPE_LEN;
3374
3375         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
3376         if (!map) {
3377                 ret = -ENOMEM;
3378                 goto error;
3379         }
3380         map->num_stripes = num_stripes;
3381
3382         for (i = 0; i < ndevs; ++i) {
3383                 for (j = 0; j < dev_stripes; ++j) {
3384                         int s = i * dev_stripes + j;
3385                         map->stripes[s].dev = devices_info[i].dev;
3386                         map->stripes[s].physical = devices_info[i].dev_offset +
3387                                                    j * stripe_size;
3388                 }
3389         }
3390         map->sector_size = extent_root->sectorsize;
3391         map->stripe_len = BTRFS_STRIPE_LEN;
3392         map->io_align = BTRFS_STRIPE_LEN;
3393         map->io_width = BTRFS_STRIPE_LEN;
3394         map->type = type;
3395         map->sub_stripes = sub_stripes;
3396
3397         *map_ret = map;
3398         num_bytes = stripe_size * (num_stripes / ncopies);
3399
3400         *stripe_size_out = stripe_size;
3401         *num_bytes_out = num_bytes;
3402
3403         trace_btrfs_chunk_alloc(info->chunk_root, map, start, num_bytes);
3404
3405         em = alloc_extent_map();
3406         if (!em) {
3407                 ret = -ENOMEM;
3408                 goto error;
3409         }
3410         em->bdev = (struct block_device *)map;
3411         em->start = start;
3412         em->len = num_bytes;
3413         em->block_start = 0;
3414         em->block_len = em->len;
3415
3416         em_tree = &extent_root->fs_info->mapping_tree.map_tree;
3417         write_lock(&em_tree->lock);
3418         ret = add_extent_mapping(em_tree, em);
3419         write_unlock(&em_tree->lock);
3420         free_extent_map(em);
3421         if (ret)
3422                 goto error;
3423
3424         ret = btrfs_make_block_group(trans, extent_root, 0, type,
3425                                      BTRFS_FIRST_CHUNK_TREE_OBJECTID,
3426                                      start, num_bytes);
3427         if (ret)
3428                 goto error;
3429
3430         for (i = 0; i < map->num_stripes; ++i) {
3431                 struct btrfs_device *device;
3432                 u64 dev_offset;
3433
3434                 device = map->stripes[i].dev;
3435                 dev_offset = map->stripes[i].physical;
3436
3437                 ret = btrfs_alloc_dev_extent(trans, device,
3438                                 info->chunk_root->root_key.objectid,
3439                                 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
3440                                 start, dev_offset, stripe_size);
3441                 if (ret) {
3442                         btrfs_abort_transaction(trans, extent_root, ret);
3443                         goto error;
3444                 }
3445         }
3446
3447         kfree(devices_info);
3448         return 0;
3449
3450 error:
3451         kfree(map);
3452         kfree(devices_info);
3453         return ret;
3454 }
3455
3456 static int __finish_chunk_alloc(struct btrfs_trans_handle *trans,
3457                                 struct btrfs_root *extent_root,
3458                                 struct map_lookup *map, u64 chunk_offset,
3459                                 u64 chunk_size, u64 stripe_size)
3460 {
3461         u64 dev_offset;
3462         struct btrfs_key key;
3463         struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
3464         struct btrfs_device *device;
3465         struct btrfs_chunk *chunk;
3466         struct btrfs_stripe *stripe;
3467         size_t item_size = btrfs_chunk_item_size(map->num_stripes);
3468         int index = 0;
3469         int ret;
3470
3471         chunk = kzalloc(item_size, GFP_NOFS);
3472         if (!chunk)
3473                 return -ENOMEM;
3474
3475         index = 0;
3476         while (index < map->num_stripes) {
3477                 device = map->stripes[index].dev;
3478                 device->bytes_used += stripe_size;
3479                 ret = btrfs_update_device(trans, device);
3480                 if (ret)
3481                         goto out_free;
3482                 index++;
3483         }
3484
3485         spin_lock(&extent_root->fs_info->free_chunk_lock);
3486         extent_root->fs_info->free_chunk_space -= (stripe_size *
3487                                                    map->num_stripes);
3488         spin_unlock(&extent_root->fs_info->free_chunk_lock);
3489
3490         index = 0;
3491         stripe = &chunk->stripe;
3492         while (index < map->num_stripes) {
3493                 device = map->stripes[index].dev;
3494                 dev_offset = map->stripes[index].physical;
3495
3496                 btrfs_set_stack_stripe_devid(stripe, device->devid);
3497                 btrfs_set_stack_stripe_offset(stripe, dev_offset);
3498                 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
3499                 stripe++;
3500                 index++;
3501         }
3502
3503         btrfs_set_stack_chunk_length(chunk, chunk_size);
3504         btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
3505         btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
3506         btrfs_set_stack_chunk_type(chunk, map->type);
3507         btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
3508         btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
3509         btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
3510         btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
3511         btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
3512
3513         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3514         key.type = BTRFS_CHUNK_ITEM_KEY;
3515         key.offset = chunk_offset;
3516
3517         ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
3518
3519         if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
3520                 /*
3521                  * TODO: Cleanup of inserted chunk root in case of
3522                  * failure.
3523                  */
3524                 ret = btrfs_add_system_chunk(chunk_root, &key, chunk,
3525                                              item_size);
3526         }
3527
3528 out_free:
3529         kfree(chunk);
3530         return ret;
3531 }
3532
3533 /*
3534  * Chunk allocation falls into two parts. The first part does works
3535  * that make the new allocated chunk useable, but not do any operation
3536  * that modifies the chunk tree. The second part does the works that
3537  * require modifying the chunk tree. This division is important for the
3538  * bootstrap process of adding storage to a seed btrfs.
3539  */
3540 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
3541                       struct btrfs_root *extent_root, u64 type)
3542 {
3543         u64 chunk_offset;
3544         u64 chunk_size;
3545         u64 stripe_size;
3546         struct map_lookup *map;
3547         struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
3548         int ret;
3549
3550         ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
3551                               &chunk_offset);
3552         if (ret)
3553                 return ret;
3554
3555         ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
3556                                   &stripe_size, chunk_offset, type);
3557         if (ret)
3558                 return ret;
3559
3560         ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
3561                                    chunk_size, stripe_size);
3562         if (ret)
3563                 return ret;
3564         return 0;
3565 }
3566
3567 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
3568                                          struct btrfs_root *root,
3569                                          struct btrfs_device *device)
3570 {
3571         u64 chunk_offset;
3572         u64 sys_chunk_offset;
3573         u64 chunk_size;
3574         u64 sys_chunk_size;
3575         u64 stripe_size;
3576         u64 sys_stripe_size;
3577         u64 alloc_profile;
3578         struct map_lookup *map;
3579         struct map_lookup *sys_map;
3580         struct btrfs_fs_info *fs_info = root->fs_info;
3581         struct btrfs_root *extent_root = fs_info->extent_root;
3582         int ret;
3583
3584         ret = find_next_chunk(fs_info->chunk_root,
3585                               BTRFS_FIRST_CHUNK_TREE_OBJECTID, &chunk_offset);
3586         if (ret)
3587                 return ret;
3588
3589         alloc_profile = BTRFS_BLOCK_GROUP_METADATA |
3590                                 fs_info->avail_metadata_alloc_bits;
3591         alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
3592
3593         ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
3594                                   &stripe_size, chunk_offset, alloc_profile);
3595         if (ret)
3596                 return ret;
3597
3598         sys_chunk_offset = chunk_offset + chunk_size;
3599
3600         alloc_profile = BTRFS_BLOCK_GROUP_SYSTEM |
3601                                 fs_info->avail_system_alloc_bits;
3602         alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
3603
3604         ret = __btrfs_alloc_chunk(trans, extent_root, &sys_map,
3605                                   &sys_chunk_size, &sys_stripe_size,
3606                                   sys_chunk_offset, alloc_profile);
3607         if (ret)
3608                 goto abort;
3609
3610         ret = btrfs_add_device(trans, fs_info->chunk_root, device);
3611         if (ret)
3612                 goto abort;
3613
3614         /*
3615          * Modifying chunk tree needs allocating new blocks from both
3616          * system block group and metadata block group. So we only can
3617          * do operations require modifying the chunk tree after both
3618          * block groups were created.
3619          */
3620         ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
3621                                    chunk_size, stripe_size);
3622         if (ret)
3623                 goto abort;
3624
3625         ret = __finish_chunk_alloc(trans, extent_root, sys_map,
3626                                    sys_chunk_offset, sys_chunk_size,
3627                                    sys_stripe_size);
3628         if (ret)
3629                 goto abort;
3630
3631         return 0;
3632
3633 abort:
3634         btrfs_abort_transaction(trans, root, ret);
3635         return ret;
3636 }
3637
3638 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
3639 {
3640         struct extent_map *em;
3641         struct map_lookup *map;
3642         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
3643         int readonly = 0;
3644         int i;
3645
3646         read_lock(&map_tree->map_tree.lock);
3647         em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
3648         read_unlock(&map_tree->map_tree.lock);
3649         if (!em)
3650                 return 1;
3651
3652         if (btrfs_test_opt(root, DEGRADED)) {
3653                 free_extent_map(em);
3654                 return 0;
3655         }
3656
3657         map = (struct map_lookup *)em->bdev;
3658         for (i = 0; i < map->num_stripes; i++) {
3659                 if (!map->stripes[i].dev->writeable) {
3660                         readonly = 1;
3661                         break;
3662                 }
3663         }
3664         free_extent_map(em);
3665         return readonly;
3666 }
3667
3668 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
3669 {
3670         extent_map_tree_init(&tree->map_tree);
3671 }
3672
3673 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
3674 {
3675         struct extent_map *em;
3676
3677         while (1) {
3678                 write_lock(&tree->map_tree.lock);
3679                 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
3680                 if (em)
3681                         remove_extent_mapping(&tree->map_tree, em);
3682                 write_unlock(&tree->map_tree.lock);
3683                 if (!em)
3684                         break;
3685                 kfree(em->bdev);
3686                 /* once for us */
3687                 free_extent_map(em);
3688                 /* once for the tree */
3689                 free_extent_map(em);
3690         }
3691 }
3692
3693 int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len)
3694 {
3695         struct extent_map *em;
3696         struct map_lookup *map;
3697         struct extent_map_tree *em_tree = &map_tree->map_tree;
3698         int ret;
3699
3700         read_lock(&em_tree->lock);
3701         em = lookup_extent_mapping(em_tree, logical, len);
3702         read_unlock(&em_tree->lock);
3703         BUG_ON(!em);
3704
3705         BUG_ON(em->start > logical || em->start + em->len < logical);
3706         map = (struct map_lookup *)em->bdev;
3707         if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
3708                 ret = map->num_stripes;
3709         else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
3710                 ret = map->sub_stripes;
3711         else
3712                 ret = 1;
3713         free_extent_map(em);
3714         return ret;
3715 }
3716
3717 static int find_live_mirror(struct map_lookup *map, int first, int num,
3718                             int optimal)
3719 {
3720         int i;
3721         if (map->stripes[optimal].dev->bdev)
3722                 return optimal;
3723         for (i = first; i < first + num; i++) {
3724                 if (map->stripes[i].dev->bdev)
3725                         return i;
3726         }
3727         /* we couldn't find one that doesn't fail.  Just return something
3728          * and the io error handling code will clean up eventually
3729          */
3730         return optimal;
3731 }
3732
3733 static int __btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
3734                              u64 logical, u64 *length,
3735                              struct btrfs_bio **bbio_ret,
3736                              int mirror_num)
3737 {
3738         struct extent_map *em;
3739         struct map_lookup *map;
3740         struct extent_map_tree *em_tree = &map_tree->map_tree;
3741         u64 offset;
3742         u64 stripe_offset;
3743         u64 stripe_end_offset;
3744         u64 stripe_nr;
3745         u64 stripe_nr_orig;
3746         u64 stripe_nr_end;
3747         int stripe_index;
3748         int i;
3749         int ret = 0;
3750         int num_stripes;
3751         int max_errors = 0;
3752         struct btrfs_bio *bbio = NULL;
3753
3754         read_lock(&em_tree->lock);
3755         em = lookup_extent_mapping(em_tree, logical, *length);
3756         read_unlock(&em_tree->lock);
3757
3758         if (!em) {
3759                 printk(KERN_CRIT "unable to find logical %llu len %llu\n",
3760                        (unsigned long long)logical,
3761                        (unsigned long long)*length);
3762                 BUG();
3763         }
3764
3765         BUG_ON(em->start > logical || em->start + em->len < logical);
3766         map = (struct map_lookup *)em->bdev;
3767         offset = logical - em->start;
3768
3769         if (mirror_num > map->num_stripes)
3770                 mirror_num = 0;
3771
3772         stripe_nr = offset;
3773         /*
3774          * stripe_nr counts the total number of stripes we have to stride
3775          * to get to this block
3776          */
3777         do_div(stripe_nr, map->stripe_len);
3778
3779         stripe_offset = stripe_nr * map->stripe_len;
3780         BUG_ON(offset < stripe_offset);
3781
3782         /* stripe_offset is the offset of this block in its stripe*/
3783         stripe_offset = offset - stripe_offset;
3784
3785         if (rw & REQ_DISCARD)
3786                 *length = min_t(u64, em->len - offset, *length);
3787         else if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
3788                 /* we limit the length of each bio to what fits in a stripe */
3789                 *length = min_t(u64, em->len - offset,
3790                                 map->stripe_len - stripe_offset);
3791         } else {
3792                 *length = em->len - offset;
3793         }
3794
3795         if (!bbio_ret)
3796                 goto out;
3797
3798         num_stripes = 1;
3799         stripe_index = 0;
3800         stripe_nr_orig = stripe_nr;
3801         stripe_nr_end = (offset + *length + map->stripe_len - 1) &
3802                         (~(map->stripe_len - 1));
3803         do_div(stripe_nr_end, map->stripe_len);
3804         stripe_end_offset = stripe_nr_end * map->stripe_len -
3805                             (offset + *length);
3806         if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
3807                 if (rw & REQ_DISCARD)
3808                         num_stripes = min_t(u64, map->num_stripes,
3809                                             stripe_nr_end - stripe_nr_orig);
3810                 stripe_index = do_div(stripe_nr, map->num_stripes);
3811         } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
3812                 if (rw & (REQ_WRITE | REQ_DISCARD))
3813                         num_stripes = map->num_stripes;
3814                 else if (mirror_num)
3815                         stripe_index = mirror_num - 1;
3816                 else {
3817                         stripe_index = find_live_mirror(map, 0,
3818                                             map->num_stripes,
3819                                             current->pid % map->num_stripes);
3820                         mirror_num = stripe_index + 1;
3821                 }
3822
3823         } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
3824                 if (rw & (REQ_WRITE | REQ_DISCARD)) {
3825                         num_stripes = map->num_stripes;
3826                 } else if (mirror_num) {
3827                         stripe_index = mirror_num - 1;
3828                 } else {
3829                         mirror_num = 1;
3830                 }
3831
3832         } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
3833                 int factor = map->num_stripes / map->sub_stripes;
3834
3835                 stripe_index = do_div(stripe_nr, factor);
3836                 stripe_index *= map->sub_stripes;
3837
3838                 if (rw & REQ_WRITE)
3839                         num_stripes = map->sub_stripes;
3840                 else if (rw & REQ_DISCARD)
3841                         num_stripes = min_t(u64, map->sub_stripes *
3842                                             (stripe_nr_end - stripe_nr_orig),
3843                                             map->num_stripes);
3844                 else if (mirror_num)
3845                         stripe_index += mirror_num - 1;
3846                 else {
3847                         int old_stripe_index = stripe_index;
3848                         stripe_index = find_live_mirror(map, stripe_index,
3849                                               map->sub_stripes, stripe_index +
3850                                               current->pid % map->sub_stripes);
3851                         mirror_num = stripe_index - old_stripe_index + 1;
3852                 }
3853         } else {
3854                 /*
3855                  * after this do_div call, stripe_nr is the number of stripes
3856                  * on this device we have to walk to find the data, and
3857                  * stripe_index is the number of our device in the stripe array
3858                  */
3859                 stripe_index = do_div(stripe_nr, map->num_stripes);
3860                 mirror_num = stripe_index + 1;
3861         }
3862         BUG_ON(stripe_index >= map->num_stripes);
3863
3864         bbio = kzalloc(btrfs_bio_size(num_stripes), GFP_NOFS);
3865         if (!bbio) {
3866                 ret = -ENOMEM;
3867                 goto out;
3868         }
3869         atomic_set(&bbio->error, 0);
3870
3871         if (rw & REQ_DISCARD) {
3872                 int factor = 0;
3873                 int sub_stripes = 0;
3874                 u64 stripes_per_dev = 0;
3875                 u32 remaining_stripes = 0;
3876                 u32 last_stripe = 0;
3877
3878                 if (map->type &
3879                     (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID10)) {
3880                         if (map->type & BTRFS_BLOCK_GROUP_RAID0)
3881                                 sub_stripes = 1;
3882                         else
3883                                 sub_stripes = map->sub_stripes;
3884
3885                         factor = map->num_stripes / sub_stripes;
3886                         stripes_per_dev = div_u64_rem(stripe_nr_end -
3887                                                       stripe_nr_orig,
3888                                                       factor,
3889                                                       &remaining_stripes);
3890                         div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
3891                         last_stripe *= sub_stripes;
3892                 }
3893
3894                 for (i = 0; i < num_stripes; i++) {
3895                         bbio->stripes[i].physical =
3896                                 map->stripes[stripe_index].physical +
3897                                 stripe_offset + stripe_nr * map->stripe_len;
3898                         bbio->stripes[i].dev = map->stripes[stripe_index].dev;
3899
3900                         if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
3901                                          BTRFS_BLOCK_GROUP_RAID10)) {
3902                                 bbio->stripes[i].length = stripes_per_dev *
3903                                                           map->stripe_len;
3904
3905                                 if (i / sub_stripes < remaining_stripes)
3906                                         bbio->stripes[i].length +=
3907                                                 map->stripe_len;
3908
3909                                 /*
3910                                  * Special for the first stripe and
3911                                  * the last stripe:
3912                                  *
3913                                  * |-------|...|-------|
3914                                  *     |----------|
3915                                  *    off     end_off
3916                                  */
3917                                 if (i < sub_stripes)
3918                                         bbio->stripes[i].length -=
3919                                                 stripe_offset;
3920
3921                                 if (stripe_index >= last_stripe &&
3922                                     stripe_index <= (last_stripe +
3923                                                      sub_stripes - 1))
3924                                         bbio->stripes[i].length -=
3925                                                 stripe_end_offset;
3926
3927                                 if (i == sub_stripes - 1)
3928                                         stripe_offset = 0;
3929                         } else
3930                                 bbio->stripes[i].length = *length;
3931
3932                         stripe_index++;
3933                         if (stripe_index == map->num_stripes) {
3934                                 /* This could only happen for RAID0/10 */
3935                                 stripe_index = 0;
3936                                 stripe_nr++;
3937                         }
3938                 }
3939         } else {
3940                 for (i = 0; i < num_stripes; i++) {
3941                         bbio->stripes[i].physical =
3942                                 map->stripes[stripe_index].physical +
3943                                 stripe_offset +
3944                                 stripe_nr * map->stripe_len;
3945                         bbio->stripes[i].dev =
3946                                 map->stripes[stripe_index].dev;
3947                         stripe_index++;
3948                 }
3949         }
3950
3951         if (rw & REQ_WRITE) {
3952                 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
3953                                  BTRFS_BLOCK_GROUP_RAID10 |
3954                                  BTRFS_BLOCK_GROUP_DUP)) {
3955                         max_errors = 1;
3956                 }
3957         }
3958
3959         *bbio_ret = bbio;
3960         bbio->num_stripes = num_stripes;
3961         bbio->max_errors = max_errors;
3962         bbio->mirror_num = mirror_num;
3963 out:
3964         free_extent_map(em);
3965         return ret;
3966 }
3967
3968 int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
3969                       u64 logical, u64 *length,
3970                       struct btrfs_bio **bbio_ret, int mirror_num)
3971 {
3972         return __btrfs_map_block(map_tree, rw, logical, length, bbio_ret,
3973                                  mirror_num);
3974 }
3975
3976 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
3977                      u64 chunk_start, u64 physical, u64 devid,
3978                      u64 **logical, int *naddrs, int *stripe_len)
3979 {
3980         struct extent_map_tree *em_tree = &map_tree->map_tree;
3981         struct extent_map *em;
3982         struct map_lookup *map;
3983         u64 *buf;
3984         u64 bytenr;
3985         u64 length;
3986         u64 stripe_nr;
3987         int i, j, nr = 0;
3988
3989         read_lock(&em_tree->lock);
3990         em = lookup_extent_mapping(em_tree, chunk_start, 1);
3991         read_unlock(&em_tree->lock);
3992
3993         BUG_ON(!em || em->start != chunk_start);
3994         map = (struct map_lookup *)em->bdev;
3995
3996         length = em->len;
3997         if (map->type & BTRFS_BLOCK_GROUP_RAID10)
3998                 do_div(length, map->num_stripes / map->sub_stripes);
3999         else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
4000                 do_div(length, map->num_stripes);
4001
4002         buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
4003         BUG_ON(!buf); /* -ENOMEM */
4004
4005         for (i = 0; i < map->num_stripes; i++) {
4006                 if (devid && map->stripes[i].dev->devid != devid)
4007                         continue;
4008                 if (map->stripes[i].physical > physical ||
4009                     map->stripes[i].physical + length <= physical)
4010                         continue;
4011
4012                 stripe_nr = physical - map->stripes[i].physical;
4013                 do_div(stripe_nr, map->stripe_len);
4014
4015                 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
4016                         stripe_nr = stripe_nr * map->num_stripes + i;
4017                         do_div(stripe_nr, map->sub_stripes);
4018                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
4019                         stripe_nr = stripe_nr * map->num_stripes + i;
4020                 }
4021                 bytenr = chunk_start + stripe_nr * map->stripe_len;
4022                 WARN_ON(nr >= map->num_stripes);
4023                 for (j = 0; j < nr; j++) {
4024                         if (buf[j] == bytenr)
4025                                 break;
4026                 }
4027                 if (j == nr) {
4028                         WARN_ON(nr >= map->num_stripes);
4029                         buf[nr++] = bytenr;
4030                 }
4031         }
4032
4033         *logical = buf;
4034         *naddrs = nr;
4035         *stripe_len = map->stripe_len;
4036
4037         free_extent_map(em);
4038         return 0;
4039 }
4040
4041 static void *merge_stripe_index_into_bio_private(void *bi_private,
4042                                                  unsigned int stripe_index)
4043 {
4044         /*
4045          * with single, dup, RAID0, RAID1 and RAID10, stripe_index is
4046          * at most 1.
4047          * The alternative solution (instead of stealing bits from the
4048          * pointer) would be to allocate an intermediate structure
4049          * that contains the old private pointer plus the stripe_index.
4050          */
4051         BUG_ON((((uintptr_t)bi_private) & 3) != 0);
4052         BUG_ON(stripe_index > 3);
4053         return (void *)(((uintptr_t)bi_private) | stripe_index);
4054 }
4055
4056 static struct btrfs_bio *extract_bbio_from_bio_private(void *bi_private)
4057 {
4058         return (struct btrfs_bio *)(((uintptr_t)bi_private) & ~((uintptr_t)3));
4059 }
4060
4061 static unsigned int extract_stripe_index_from_bio_private(void *bi_private)
4062 {
4063         return (unsigned int)((uintptr_t)bi_private) & 3;
4064 }
4065
4066 static void btrfs_end_bio(struct bio *bio, int err)
4067 {
4068         struct btrfs_bio *bbio = extract_bbio_from_bio_private(bio->bi_private);
4069         int is_orig_bio = 0;
4070
4071         if (err) {
4072                 atomic_inc(&bbio->error);
4073                 if (err == -EIO || err == -EREMOTEIO) {
4074                         unsigned int stripe_index =
4075                                 extract_stripe_index_from_bio_private(
4076                                         bio->bi_private);
4077                         struct btrfs_device *dev;
4078
4079                         BUG_ON(stripe_index >= bbio->num_stripes);
4080                         dev = bbio->stripes[stripe_index].dev;
4081                         if (dev->bdev) {
4082                                 if (bio->bi_rw & WRITE)
4083                                         btrfs_dev_stat_inc(dev,
4084                                                 BTRFS_DEV_STAT_WRITE_ERRS);
4085                                 else
4086                                         btrfs_dev_stat_inc(dev,
4087                                                 BTRFS_DEV_STAT_READ_ERRS);
4088                                 if ((bio->bi_rw & WRITE_FLUSH) == WRITE_FLUSH)
4089                                         btrfs_dev_stat_inc(dev,
4090                                                 BTRFS_DEV_STAT_FLUSH_ERRS);
4091                                 btrfs_dev_stat_print_on_error(dev);
4092                         }
4093                 }
4094         }
4095
4096         if (bio == bbio->orig_bio)
4097                 is_orig_bio = 1;
4098
4099         if (atomic_dec_and_test(&bbio->stripes_pending)) {
4100                 if (!is_orig_bio) {
4101                         bio_put(bio);
4102                         bio = bbio->orig_bio;
4103                 }
4104                 bio->bi_private = bbio->private;
4105                 bio->bi_end_io = bbio->end_io;
4106                 bio->bi_bdev = (struct block_device *)
4107                                         (unsigned long)bbio->mirror_num;
4108                 /* only send an error to the higher layers if it is
4109                  * beyond the tolerance of the multi-bio
4110                  */
4111                 if (atomic_read(&bbio->error) > bbio->max_errors) {
4112                         err = -EIO;
4113                 } else {
4114                         /*
4115                          * this bio is actually up to date, we didn't
4116                          * go over the max number of errors
4117                          */
4118                         set_bit(BIO_UPTODATE, &bio->bi_flags);
4119                         err = 0;
4120                 }
4121                 kfree(bbio);
4122
4123                 bio_endio(bio, err);
4124         } else if (!is_orig_bio) {
4125                 bio_put(bio);
4126         }
4127 }
4128
4129 struct async_sched {
4130         struct bio *bio;
4131         int rw;
4132         struct btrfs_fs_info *info;
4133         struct btrfs_work work;
4134 };
4135
4136 /*
4137  * see run_scheduled_bios for a description of why bios are collected for
4138  * async submit.
4139  *
4140  * This will add one bio to the pending list for a device and make sure
4141  * the work struct is scheduled.
4142  */
4143 static noinline void schedule_bio(struct btrfs_root *root,
4144                                  struct btrfs_device *device,
4145                                  int rw, struct bio *bio)
4146 {
4147         int should_queue = 1;
4148         struct btrfs_pending_bios *pending_bios;
4149
4150         /* don't bother with additional async steps for reads, right now */
4151         if (!(rw & REQ_WRITE)) {
4152                 bio_get(bio);
4153                 btrfsic_submit_bio(rw, bio);
4154                 bio_put(bio);
4155                 return;
4156         }
4157
4158         /*
4159          * nr_async_bios allows us to reliably return congestion to the
4160          * higher layers.  Otherwise, the async bio makes it appear we have
4161          * made progress against dirty pages when we've really just put it
4162          * on a queue for later
4163          */
4164         atomic_inc(&root->fs_info->nr_async_bios);
4165         WARN_ON(bio->bi_next);
4166         bio->bi_next = NULL;
4167         bio->bi_rw |= rw;
4168
4169         spin_lock(&device->io_lock);
4170         if (bio->bi_rw & REQ_SYNC)
4171                 pending_bios = &device->pending_sync_bios;
4172         else
4173                 pending_bios = &device->pending_bios;
4174
4175         if (pending_bios->tail)
4176                 pending_bios->tail->bi_next = bio;
4177
4178         pending_bios->tail = bio;
4179         if (!pending_bios->head)
4180                 pending_bios->head = bio;
4181         if (device->running_pending)
4182                 should_queue = 0;
4183
4184         spin_unlock(&device->io_lock);
4185
4186         if (should_queue)
4187                 btrfs_queue_worker(&root->fs_info->submit_workers,
4188                                    &device->work);
4189 }
4190
4191 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
4192                   int mirror_num, int async_submit)
4193 {
4194         struct btrfs_mapping_tree *map_tree;
4195         struct btrfs_device *dev;
4196         struct bio *first_bio = bio;
4197         u64 logical = (u64)bio->bi_sector << 9;
4198         u64 length = 0;
4199         u64 map_length;
4200         int ret;
4201         int dev_nr = 0;
4202         int total_devs = 1;
4203         struct btrfs_bio *bbio = NULL;
4204
4205         length = bio->bi_size;
4206         map_tree = &root->fs_info->mapping_tree;
4207         map_length = length;
4208
4209         ret = btrfs_map_block(map_tree, rw, logical, &map_length, &bbio,
4210                               mirror_num);
4211         if (ret) /* -ENOMEM */
4212                 return ret;
4213
4214         total_devs = bbio->num_stripes;
4215         if (map_length < length) {
4216                 printk(KERN_CRIT "mapping failed logical %llu bio len %llu "
4217                        "len %llu\n", (unsigned long long)logical,
4218                        (unsigned long long)length,
4219                        (unsigned long long)map_length);
4220                 BUG();
4221         }
4222
4223         bbio->orig_bio = first_bio;
4224         bbio->private = first_bio->bi_private;
4225         bbio->end_io = first_bio->bi_end_io;
4226         atomic_set(&bbio->stripes_pending, bbio->num_stripes);
4227
4228         while (dev_nr < total_devs) {
4229                 if (dev_nr < total_devs - 1) {
4230                         bio = bio_clone(first_bio, GFP_NOFS);
4231                         BUG_ON(!bio); /* -ENOMEM */
4232                 } else {
4233                         bio = first_bio;
4234                 }
4235                 bio->bi_private = bbio;
4236                 bio->bi_private = merge_stripe_index_into_bio_private(
4237                                 bio->bi_private, (unsigned int)dev_nr);
4238                 bio->bi_end_io = btrfs_end_bio;
4239                 bio->bi_sector = bbio->stripes[dev_nr].physical >> 9;
4240                 dev = bbio->stripes[dev_nr].dev;
4241                 if (dev && dev->bdev && (rw != WRITE || dev->writeable)) {
4242 #ifdef DEBUG
4243                         struct rcu_string *name;
4244
4245                         rcu_read_lock();
4246                         name = rcu_dereference(dev->name);
4247                         pr_debug("btrfs_map_bio: rw %d, secor=%llu, dev=%lu "
4248                                  "(%s id %llu), size=%u\n", rw,
4249                                  (u64)bio->bi_sector, (u_long)dev->bdev->bd_dev,
4250                                  name->str, dev->devid, bio->bi_size);
4251                         rcu_read_unlock();
4252 #endif
4253                         bio->bi_bdev = dev->bdev;
4254                         if (async_submit)
4255                                 schedule_bio(root, dev, rw, bio);
4256                         else
4257                                 btrfsic_submit_bio(rw, bio);
4258                 } else {
4259                         bio->bi_bdev = root->fs_info->fs_devices->latest_bdev;
4260                         bio->bi_sector = logical >> 9;
4261                         bio_endio(bio, -EIO);
4262                 }
4263                 dev_nr++;
4264         }
4265         return 0;
4266 }
4267
4268 struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid,
4269                                        u8 *uuid, u8 *fsid)
4270 {
4271         struct btrfs_device *device;
4272         struct btrfs_fs_devices *cur_devices;
4273
4274         cur_devices = root->fs_info->fs_devices;
4275         while (cur_devices) {
4276                 if (!fsid ||
4277                     !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
4278                         device = __find_device(&cur_devices->devices,
4279                                                devid, uuid);
4280                         if (device)
4281                                 return device;
4282                 }
4283                 cur_devices = cur_devices->seed;
4284         }
4285         return NULL;
4286 }
4287
4288 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
4289                                             u64 devid, u8 *dev_uuid)
4290 {
4291         struct btrfs_device *device;
4292         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
4293
4294         device = kzalloc(sizeof(*device), GFP_NOFS);
4295         if (!device)
4296                 return NULL;
4297         list_add(&device->dev_list,
4298                  &fs_devices->devices);
4299         device->dev_root = root->fs_info->dev_root;
4300         device->devid = devid;
4301         device->work.func = pending_bios_fn;
4302         device->fs_devices = fs_devices;
4303         device->missing = 1;
4304         fs_devices->num_devices++;
4305         fs_devices->missing_devices++;
4306         spin_lock_init(&device->io_lock);
4307         INIT_LIST_HEAD(&device->dev_alloc_list);
4308         memcpy(device->uuid, dev_uuid, BTRFS_UUID_SIZE);
4309         return device;
4310 }
4311
4312 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
4313                           struct extent_buffer *leaf,
4314                           struct btrfs_chunk *chunk)
4315 {
4316         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
4317         struct map_lookup *map;
4318         struct extent_map *em;
4319         u64 logical;
4320         u64 length;
4321         u64 devid;
4322         u8 uuid[BTRFS_UUID_SIZE];
4323         int num_stripes;
4324         int ret;
4325         int i;
4326
4327         logical = key->offset;
4328         length = btrfs_chunk_length(leaf, chunk);
4329
4330         read_lock(&map_tree->map_tree.lock);
4331         em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
4332         read_unlock(&map_tree->map_tree.lock);
4333
4334         /* already mapped? */
4335         if (em && em->start <= logical && em->start + em->len > logical) {
4336                 free_extent_map(em);
4337                 return 0;
4338         } else if (em) {
4339                 free_extent_map(em);
4340         }
4341
4342         em = alloc_extent_map();
4343         if (!em)
4344                 return -ENOMEM;
4345         num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
4346         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
4347         if (!map) {
4348                 free_extent_map(em);
4349                 return -ENOMEM;
4350         }
4351
4352         em->bdev = (struct block_device *)map;
4353         em->start = logical;
4354         em->len = length;
4355         em->block_start = 0;
4356         em->block_len = em->len;
4357
4358         map->num_stripes = num_stripes;
4359         map->io_width = btrfs_chunk_io_width(leaf, chunk);
4360         map->io_align = btrfs_chunk_io_align(leaf, chunk);
4361         map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
4362         map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
4363         map->type = btrfs_chunk_type(leaf, chunk);
4364         map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
4365         for (i = 0; i < num_stripes; i++) {
4366                 map->stripes[i].physical =
4367                         btrfs_stripe_offset_nr(leaf, chunk, i);
4368                 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
4369                 read_extent_buffer(leaf, uuid, (unsigned long)
4370                                    btrfs_stripe_dev_uuid_nr(chunk, i),
4371                                    BTRFS_UUID_SIZE);
4372                 map->stripes[i].dev = btrfs_find_device(root, devid, uuid,
4373                                                         NULL);
4374                 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
4375                         kfree(map);
4376                         free_extent_map(em);
4377                         return -EIO;
4378                 }
4379                 if (!map->stripes[i].dev) {
4380                         map->stripes[i].dev =
4381                                 add_missing_dev(root, devid, uuid);
4382                         if (!map->stripes[i].dev) {
4383                                 kfree(map);
4384                                 free_extent_map(em);
4385                                 return -EIO;
4386                         }
4387                 }
4388                 map->stripes[i].dev->in_fs_metadata = 1;
4389         }
4390
4391         write_lock(&map_tree->map_tree.lock);
4392         ret = add_extent_mapping(&map_tree->map_tree, em);
4393         write_unlock(&map_tree->map_tree.lock);
4394         BUG_ON(ret); /* Tree corruption */
4395         free_extent_map(em);
4396
4397         return 0;
4398 }
4399
4400 static void fill_device_from_item(struct extent_buffer *leaf,
4401                                  struct btrfs_dev_item *dev_item,
4402                                  struct btrfs_device *device)
4403 {
4404         unsigned long ptr;
4405
4406         device->devid = btrfs_device_id(leaf, dev_item);
4407         device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
4408         device->total_bytes = device->disk_total_bytes;
4409         device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
4410         device->type = btrfs_device_type(leaf, dev_item);
4411         device->io_align = btrfs_device_io_align(leaf, dev_item);
4412         device->io_width = btrfs_device_io_width(leaf, dev_item);
4413         device->sector_size = btrfs_device_sector_size(leaf, dev_item);
4414
4415         ptr = (unsigned long)btrfs_device_uuid(dev_item);
4416         read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
4417 }
4418
4419 static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
4420 {
4421         struct btrfs_fs_devices *fs_devices;
4422         int ret;
4423
4424         BUG_ON(!mutex_is_locked(&uuid_mutex));
4425
4426         fs_devices = root->fs_info->fs_devices->seed;
4427         while (fs_devices) {
4428                 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
4429                         ret = 0;
4430                         goto out;
4431                 }
4432                 fs_devices = fs_devices->seed;
4433         }
4434
4435         fs_devices = find_fsid(fsid);
4436         if (!fs_devices) {
4437                 ret = -ENOENT;
4438                 goto out;
4439         }
4440
4441         fs_devices = clone_fs_devices(fs_devices);
4442         if (IS_ERR(fs_devices)) {
4443                 ret = PTR_ERR(fs_devices);
4444                 goto out;
4445         }
4446
4447         ret = __btrfs_open_devices(fs_devices, FMODE_READ,
4448                                    root->fs_info->bdev_holder);
4449         if (ret) {
4450                 free_fs_devices(fs_devices);
4451                 goto out;
4452         }
4453
4454         if (!fs_devices->seeding) {
4455                 __btrfs_close_devices(fs_devices);
4456                 free_fs_devices(fs_devices);
4457                 ret = -EINVAL;
4458                 goto out;
4459         }
4460
4461         fs_devices->seed = root->fs_info->fs_devices->seed;
4462         root->fs_info->fs_devices->seed = fs_devices;
4463 out:
4464         return ret;
4465 }
4466
4467 static int read_one_dev(struct btrfs_root *root,
4468                         struct extent_buffer *leaf,
4469                         struct btrfs_dev_item *dev_item)
4470 {
4471         struct btrfs_device *device;
4472         u64 devid;
4473         int ret;
4474         u8 fs_uuid[BTRFS_UUID_SIZE];
4475         u8 dev_uuid[BTRFS_UUID_SIZE];
4476
4477         devid = btrfs_device_id(leaf, dev_item);
4478         read_extent_buffer(leaf, dev_uuid,
4479                            (unsigned long)btrfs_device_uuid(dev_item),
4480                            BTRFS_UUID_SIZE);
4481         read_extent_buffer(leaf, fs_uuid,
4482                            (unsigned long)btrfs_device_fsid(dev_item),
4483                            BTRFS_UUID_SIZE);
4484
4485         if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
4486                 ret = open_seed_devices(root, fs_uuid);
4487                 if (ret && !btrfs_test_opt(root, DEGRADED))
4488                         return ret;
4489         }
4490
4491         device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
4492         if (!device || !device->bdev) {
4493                 if (!btrfs_test_opt(root, DEGRADED))
4494                         return -EIO;
4495
4496                 if (!device) {
4497                         printk(KERN_WARNING "warning devid %llu missing\n",
4498                                (unsigned long long)devid);
4499                         device = add_missing_dev(root, devid, dev_uuid);
4500                         if (!device)
4501                                 return -ENOMEM;
4502                 } else if (!device->missing) {
4503                         /*
4504                          * this happens when a device that was properly setup
4505                          * in the device info lists suddenly goes bad.
4506                          * device->bdev is NULL, and so we have to set
4507                          * device->missing to one here
4508                          */
4509                         root->fs_info->fs_devices->missing_devices++;
4510                         device->missing = 1;
4511                 }
4512         }
4513
4514         if (device->fs_devices != root->fs_info->fs_devices) {
4515                 BUG_ON(device->writeable);
4516                 if (device->generation !=
4517                     btrfs_device_generation(leaf, dev_item))
4518                         return -EINVAL;
4519         }
4520
4521         fill_device_from_item(leaf, dev_item, device);
4522         device->dev_root = root->fs_info->dev_root;
4523         device->in_fs_metadata = 1;
4524         if (device->writeable) {
4525                 device->fs_devices->total_rw_bytes += device->total_bytes;
4526                 spin_lock(&root->fs_info->free_chunk_lock);
4527                 root->fs_info->free_chunk_space += device->total_bytes -
4528                         device->bytes_used;
4529                 spin_unlock(&root->fs_info->free_chunk_lock);
4530         }
4531         ret = 0;
4532         return ret;
4533 }
4534
4535 int btrfs_read_sys_array(struct btrfs_root *root)
4536 {
4537         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
4538         struct extent_buffer *sb;
4539         struct btrfs_disk_key *disk_key;
4540         struct btrfs_chunk *chunk;
4541         u8 *ptr;
4542         unsigned long sb_ptr;
4543         int ret = 0;
4544         u32 num_stripes;
4545         u32 array_size;
4546         u32 len = 0;
4547         u32 cur;
4548         struct btrfs_key key;
4549
4550         sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
4551                                           BTRFS_SUPER_INFO_SIZE);
4552         if (!sb)
4553                 return -ENOMEM;
4554         btrfs_set_buffer_uptodate(sb);
4555         btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
4556         /*
4557          * The sb extent buffer is artifical and just used to read the system array.
4558          * btrfs_set_buffer_uptodate() call does not properly mark all it's
4559          * pages up-to-date when the page is larger: extent does not cover the
4560          * whole page and consequently check_page_uptodate does not find all
4561          * the page's extents up-to-date (the hole beyond sb),
4562          * write_extent_buffer then triggers a WARN_ON.
4563          *
4564          * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
4565          * but sb spans only this function. Add an explicit SetPageUptodate call
4566          * to silence the warning eg. on PowerPC 64.
4567          */
4568         if (PAGE_CACHE_SIZE > BTRFS_SUPER_INFO_SIZE)
4569                 SetPageUptodate(sb->pages[0]);
4570
4571         write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
4572         array_size = btrfs_super_sys_array_size(super_copy);
4573
4574         ptr = super_copy->sys_chunk_array;
4575         sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
4576         cur = 0;
4577
4578         while (cur < array_size) {
4579                 disk_key = (struct btrfs_disk_key *)ptr;
4580                 btrfs_disk_key_to_cpu(&key, disk_key);
4581
4582                 len = sizeof(*disk_key); ptr += len;
4583                 sb_ptr += len;
4584                 cur += len;
4585
4586                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
4587                         chunk = (struct btrfs_chunk *)sb_ptr;
4588                         ret = read_one_chunk(root, &key, sb, chunk);
4589                         if (ret)
4590                                 break;
4591                         num_stripes = btrfs_chunk_num_stripes(sb, chunk);
4592                         len = btrfs_chunk_item_size(num_stripes);
4593                 } else {
4594                         ret = -EIO;
4595                         break;
4596                 }
4597                 ptr += len;
4598                 sb_ptr += len;
4599                 cur += len;
4600         }
4601         free_extent_buffer(sb);
4602         return ret;
4603 }
4604
4605 struct btrfs_device *btrfs_find_device_for_logical(struct btrfs_root *root,
4606                                                    u64 logical, int mirror_num)
4607 {
4608         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
4609         int ret;
4610         u64 map_length = 0;
4611         struct btrfs_bio *bbio = NULL;
4612         struct btrfs_device *device;
4613
4614         BUG_ON(mirror_num == 0);
4615         ret = btrfs_map_block(map_tree, WRITE, logical, &map_length, &bbio,
4616                               mirror_num);
4617         if (ret) {
4618                 BUG_ON(bbio != NULL);
4619                 return NULL;
4620         }
4621         BUG_ON(mirror_num != bbio->mirror_num);
4622         device = bbio->stripes[mirror_num - 1].dev;
4623         kfree(bbio);
4624         return device;
4625 }
4626
4627 int btrfs_read_chunk_tree(struct btrfs_root *root)
4628 {
4629         struct btrfs_path *path;
4630         struct extent_buffer *leaf;
4631         struct btrfs_key key;
4632         struct btrfs_key found_key;
4633         int ret;
4634         int slot;
4635
4636         root = root->fs_info->chunk_root;
4637
4638         path = btrfs_alloc_path();
4639         if (!path)
4640                 return -ENOMEM;
4641
4642         mutex_lock(&uuid_mutex);
4643         lock_chunks(root);
4644
4645         /* first we search for all of the device items, and then we
4646          * read in all of the chunk items.  This way we can create chunk
4647          * mappings that reference all of the devices that are afound
4648          */
4649         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
4650         key.offset = 0;
4651         key.type = 0;
4652 again:
4653         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4654         if (ret < 0)
4655                 goto error;
4656         while (1) {
4657                 leaf = path->nodes[0];
4658                 slot = path->slots[0];
4659                 if (slot >= btrfs_header_nritems(leaf)) {
4660                         ret = btrfs_next_leaf(root, path);
4661                         if (ret == 0)
4662                                 continue;
4663                         if (ret < 0)
4664                                 goto error;
4665                         break;
4666                 }
4667                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4668                 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
4669                         if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID)
4670                                 break;
4671                         if (found_key.type == BTRFS_DEV_ITEM_KEY) {
4672                                 struct btrfs_dev_item *dev_item;
4673                                 dev_item = btrfs_item_ptr(leaf, slot,
4674                                                   struct btrfs_dev_item);
4675                                 ret = read_one_dev(root, leaf, dev_item);
4676                                 if (ret)
4677                                         goto error;
4678                         }
4679                 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
4680                         struct btrfs_chunk *chunk;
4681                         chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
4682                         ret = read_one_chunk(root, &found_key, leaf, chunk);
4683                         if (ret)
4684                                 goto error;
4685                 }
4686                 path->slots[0]++;
4687         }
4688         if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
4689                 key.objectid = 0;
4690                 btrfs_release_path(path);
4691                 goto again;
4692         }
4693         ret = 0;
4694 error:
4695         unlock_chunks(root);
4696         mutex_unlock(&uuid_mutex);
4697
4698         btrfs_free_path(path);
4699         return ret;
4700 }
4701
4702 static void __btrfs_reset_dev_stats(struct btrfs_device *dev)
4703 {
4704         int i;
4705
4706         for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
4707                 btrfs_dev_stat_reset(dev, i);
4708 }
4709
4710 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
4711 {
4712         struct btrfs_key key;
4713         struct btrfs_key found_key;
4714         struct btrfs_root *dev_root = fs_info->dev_root;
4715         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
4716         struct extent_buffer *eb;
4717         int slot;
4718         int ret = 0;
4719         struct btrfs_device *device;
4720         struct btrfs_path *path = NULL;
4721         int i;
4722
4723         path = btrfs_alloc_path();
4724         if (!path) {
4725                 ret = -ENOMEM;
4726                 goto out;
4727         }
4728
4729         mutex_lock(&fs_devices->device_list_mutex);
4730         list_for_each_entry(device, &fs_devices->devices, dev_list) {
4731                 int item_size;
4732                 struct btrfs_dev_stats_item *ptr;
4733
4734                 key.objectid = 0;
4735                 key.type = BTRFS_DEV_STATS_KEY;
4736                 key.offset = device->devid;
4737                 ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
4738                 if (ret) {
4739                         printk_in_rcu(KERN_WARNING "btrfs: no dev_stats entry found for device %s (devid %llu) (OK on first mount after mkfs)\n",
4740                                       rcu_str_deref(device->name),
4741                                       (unsigned long long)device->devid);
4742                         __btrfs_reset_dev_stats(device);
4743                         device->dev_stats_valid = 1;
4744                         btrfs_release_path(path);
4745                         continue;
4746                 }
4747                 slot = path->slots[0];
4748                 eb = path->nodes[0];
4749                 btrfs_item_key_to_cpu(eb, &found_key, slot);
4750                 item_size = btrfs_item_size_nr(eb, slot);
4751
4752                 ptr = btrfs_item_ptr(eb, slot,
4753                                      struct btrfs_dev_stats_item);
4754
4755                 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
4756                         if (item_size >= (1 + i) * sizeof(__le64))
4757                                 btrfs_dev_stat_set(device, i,
4758                                         btrfs_dev_stats_value(eb, ptr, i));
4759                         else
4760                                 btrfs_dev_stat_reset(device, i);
4761                 }
4762
4763                 device->dev_stats_valid = 1;
4764                 btrfs_dev_stat_print_on_load(device);
4765                 btrfs_release_path(path);
4766         }
4767         mutex_unlock(&fs_devices->device_list_mutex);
4768
4769 out:
4770         btrfs_free_path(path);
4771         return ret < 0 ? ret : 0;
4772 }
4773
4774 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
4775                                 struct btrfs_root *dev_root,
4776                                 struct btrfs_device *device)
4777 {
4778         struct btrfs_path *path;
4779         struct btrfs_key key;
4780         struct extent_buffer *eb;
4781         struct btrfs_dev_stats_item *ptr;
4782         int ret;
4783         int i;
4784
4785         key.objectid = 0;
4786         key.type = BTRFS_DEV_STATS_KEY;
4787         key.offset = device->devid;
4788
4789         path = btrfs_alloc_path();
4790         BUG_ON(!path);
4791         ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
4792         if (ret < 0) {
4793                 printk_in_rcu(KERN_WARNING "btrfs: error %d while searching for dev_stats item for device %s!\n",
4794                               ret, rcu_str_deref(device->name));
4795                 goto out;
4796         }
4797
4798         if (ret == 0 &&
4799             btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
4800                 /* need to delete old one and insert a new one */
4801                 ret = btrfs_del_item(trans, dev_root, path);
4802                 if (ret != 0) {
4803                         printk_in_rcu(KERN_WARNING "btrfs: delete too small dev_stats item for device %s failed %d!\n",
4804                                       rcu_str_deref(device->name), ret);
4805                         goto out;
4806                 }
4807                 ret = 1;
4808         }
4809
4810         if (ret == 1) {
4811                 /* need to insert a new item */
4812                 btrfs_release_path(path);
4813                 ret = btrfs_insert_empty_item(trans, dev_root, path,
4814                                               &key, sizeof(*ptr));
4815                 if (ret < 0) {
4816                         printk_in_rcu(KERN_WARNING "btrfs: insert dev_stats item for device %s failed %d!\n",
4817                                       rcu_str_deref(device->name), ret);
4818                         goto out;
4819                 }
4820         }
4821
4822         eb = path->nodes[0];
4823         ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
4824         for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
4825                 btrfs_set_dev_stats_value(eb, ptr, i,
4826                                           btrfs_dev_stat_read(device, i));
4827         btrfs_mark_buffer_dirty(eb);
4828
4829 out:
4830         btrfs_free_path(path);
4831         return ret;
4832 }
4833
4834 /*
4835  * called from commit_transaction. Writes all changed device stats to disk.
4836  */
4837 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans,
4838                         struct btrfs_fs_info *fs_info)
4839 {
4840         struct btrfs_root *dev_root = fs_info->dev_root;
4841         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
4842         struct btrfs_device *device;
4843         int ret = 0;
4844
4845         mutex_lock(&fs_devices->device_list_mutex);
4846         list_for_each_entry(device, &fs_devices->devices, dev_list) {
4847                 if (!device->dev_stats_valid || !device->dev_stats_dirty)
4848                         continue;
4849
4850                 ret = update_dev_stat_item(trans, dev_root, device);
4851                 if (!ret)
4852                         device->dev_stats_dirty = 0;
4853         }
4854         mutex_unlock(&fs_devices->device_list_mutex);
4855
4856         return ret;
4857 }
4858
4859 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
4860 {
4861         btrfs_dev_stat_inc(dev, index);
4862         btrfs_dev_stat_print_on_error(dev);
4863 }
4864
4865 void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
4866 {
4867         if (!dev->dev_stats_valid)
4868                 return;
4869         printk_ratelimited_in_rcu(KERN_ERR
4870                            "btrfs: bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
4871                            rcu_str_deref(dev->name),
4872                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
4873                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
4874                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
4875                            btrfs_dev_stat_read(dev,
4876                                                BTRFS_DEV_STAT_CORRUPTION_ERRS),
4877                            btrfs_dev_stat_read(dev,
4878                                                BTRFS_DEV_STAT_GENERATION_ERRS));
4879 }
4880
4881 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
4882 {
4883         printk_in_rcu(KERN_INFO "btrfs: bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
4884                rcu_str_deref(dev->name),
4885                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
4886                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
4887                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
4888                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
4889                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
4890 }
4891
4892 int btrfs_get_dev_stats(struct btrfs_root *root,
4893                         struct btrfs_ioctl_get_dev_stats *stats,
4894                         int reset_after_read)
4895 {
4896         struct btrfs_device *dev;
4897         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
4898         int i;
4899
4900         mutex_lock(&fs_devices->device_list_mutex);
4901         dev = btrfs_find_device(root, stats->devid, NULL, NULL);
4902         mutex_unlock(&fs_devices->device_list_mutex);
4903
4904         if (!dev) {
4905                 printk(KERN_WARNING
4906                        "btrfs: get dev_stats failed, device not found\n");
4907                 return -ENODEV;
4908         } else if (!dev->dev_stats_valid) {
4909                 printk(KERN_WARNING
4910                        "btrfs: get dev_stats failed, not yet valid\n");
4911                 return -ENODEV;
4912         } else if (reset_after_read) {
4913                 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
4914                         if (stats->nr_items > i)
4915                                 stats->values[i] =
4916                                         btrfs_dev_stat_read_and_reset(dev, i);
4917                         else
4918                                 btrfs_dev_stat_reset(dev, i);
4919                 }
4920         } else {
4921                 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
4922                         if (stats->nr_items > i)
4923                                 stats->values[i] = btrfs_dev_stat_read(dev, i);
4924         }
4925         if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
4926                 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
4927         return 0;
4928 }