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 "compat.h"
  29 #include "ctree.h"
  30 #include "extent_map.h"
  31 #include "disk-io.h"
  32 #include "transaction.h"
  33 #include "print-tree.h"
  34 #include "volumes.h"
  35 #include "async-thread.h"
  36 #include "check-integrity.h"
  37 #include "rcu-string.h"
  38 #include "math.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 int
 112 btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder,
 113                       int flush, struct block_device **bdev,
 114                       struct buffer_head **bh)
 115 {
 116         int ret;
 117
 118         *bdev = blkdev_get_by_path(device_path, flags, holder);
 119
 120         if (IS_ERR(*bdev)) {
 121                 ret = PTR_ERR(*bdev);
 122                 printk(KERN_INFO "btrfs: open %s failed\n", device_path);
 123                 goto error;
 124         }
 125
 126         if (flush)
 127                 filemap_write_and_wait((*bdev)->bd_inode->i_mapping);
 128         ret = set_blocksize(*bdev, 4096);
 129         if (ret) {
 130                 blkdev_put(*bdev, flags);
 131                 goto error;
 132         }
 133         invalidate_bdev(*bdev);
 134         *bh = btrfs_read_dev_super(*bdev);
 135         if (!*bh) {
 136                 ret = -EINVAL;
 137                 blkdev_put(*bdev, flags);
 138                 goto error;
 139         }
 140
 141         return 0;
 142
 143 error:
 144         *bdev = NULL;
 145         *bh = NULL;
 146         return ret;
 147 }
 148
 149 static void requeue_list(struct btrfs_pending_bios *pending_bios,
 150                         struct bio *head, struct bio *tail)
 151 {
 152
 153         struct bio *old_head;
 154
 155         old_head = pending_bios->head;
 156         pending_bios->head = head;
 157         if (pending_bios->tail)
 158                 tail->bi_next = old_head;
 159         else
 160                 pending_bios->tail = tail;
 161 }
 162
 163 /*
 164  * we try to collect pending bios for a device so we don't get a large
 165  * number of procs sending bios down to the same device.  This greatly
 166  * improves the schedulers ability to collect and merge the bios.
 167  *
 168  * But, it also turns into a long list of bios to process and that is sure
 169  * to eventually make the worker thread block.  The solution here is to
 170  * make some progress and then put this work struct back at the end of
 171  * the list if the block device is congested.  This way, multiple devices
 172  * can make progress from a single worker thread.
 173  */
 174 static noinline void run_scheduled_bios(struct btrfs_device *device)
 175 {
 176         struct bio *pending;
 177         struct backing_dev_info *bdi;
 178         struct btrfs_fs_info *fs_info;
 179         struct btrfs_pending_bios *pending_bios;
 180         struct bio *tail;
 181         struct bio *cur;
 182         int again = 0;
 183         unsigned long num_run;
 184         unsigned long batch_run = 0;
 185         unsigned long limit;
 186         unsigned long last_waited = 0;
 187         int force_reg = 0;
 188         int sync_pending = 0;
 189         struct blk_plug plug;
 190
 191         /*
 192          * this function runs all the bios we've collected for
 193          * a particular device.  We don't want to wander off to
 194          * another device without first sending all of these down.
 195          * So, setup a plug here and finish it off before we return
 196          */
 197         blk_start_plug(&plug);
 198
 199         bdi = blk_get_backing_dev_info(device->bdev);
 200         fs_info = device->dev_root->fs_info;
 201         limit = btrfs_async_submit_limit(fs_info);
 202         limit = limit * 2 / 3;
 203
 204 loop:
 205         spin_lock(&device->io_lock);
 206
 207 loop_lock:
 208         num_run = 0;
 209
 210         /* take all the bios off the list at once and process them
 211          * later on (without the lock held).  But, remember the
 212          * tail and other pointers so the bios can be properly reinserted
 213          * into the list if we hit congestion
 214          */
 215         if (!force_reg && device->pending_sync_bios.head) {
 216                 pending_bios = &device->pending_sync_bios;
 217                 force_reg = 1;
 218         } else {
 219                 pending_bios = &device->pending_bios;
 220                 force_reg = 0;
 221         }
 222
 223         pending = pending_bios->head;
 224         tail = pending_bios->tail;
 225         WARN_ON(pending && !tail);
 226
 227         /*
 228          * if pending was null this time around, no bios need processing
 229          * at all and we can stop.  Otherwise it'll loop back up again
 230          * and do an additional check so no bios are missed.
 231          *
 232          * device->running_pending is used to synchronize with the
 233          * schedule_bio code.
 234          */
 235         if (device->pending_sync_bios.head == NULL &&
 236             device->pending_bios.head == NULL) {
 237                 again = 0;
 238                 device->running_pending = 0;
 239         } else {
 240                 again = 1;
 241                 device->running_pending = 1;
 242         }
 243
 244         pending_bios->head = NULL;
 245         pending_bios->tail = NULL;
 246
 247         spin_unlock(&device->io_lock);
 248
 249         while (pending) {
 250
 251                 rmb();
 252                 /* we want to work on both lists, but do more bios on the
 253                  * sync list than the regular list
 254                  */
 255                 if ((num_run > 32 &&
 256                     pending_bios != &device->pending_sync_bios &&
 257                     device->pending_sync_bios.head) ||
 258                    (num_run > 64 && pending_bios == &device->pending_sync_bios &&
 259                     device->pending_bios.head)) {
 260                         spin_lock(&device->io_lock);
 261                         requeue_list(pending_bios, pending, tail);
 262                         goto loop_lock;
 263                 }
 264
 265                 cur = pending;
 266                 pending = pending->bi_next;
 267                 cur->bi_next = NULL;
 268
 269                 if (atomic_dec_return(&fs_info->nr_async_bios) < limit &&
 270                     waitqueue_active(&fs_info->async_submit_wait))
 271                         wake_up(&fs_info->async_submit_wait);
 272
 273                 BUG_ON(atomic_read(&cur->bi_cnt) == 0);
 274
 275                 /*
 276                  * if we're doing the sync list, record that our
 277                  * plug has some sync requests on it
 278                  *
 279                  * If we're doing the regular list and there are
 280                  * sync requests sitting around, unplug before
 281                  * we add more
 282                  */
 283                 if (pending_bios == &device->pending_sync_bios) {
 284                         sync_pending = 1;
 285                 } else if (sync_pending) {
 286                         blk_finish_plug(&plug);
 287                         blk_start_plug(&plug);
 288                         sync_pending = 0;
 289                 }
 290
 291                 btrfsic_submit_bio(cur->bi_rw, cur);
 292                 num_run++;
 293                 batch_run++;
 294                 if (need_resched())
 295                         cond_resched();
 296
 297                 /*
 298                  * we made progress, there is more work to do and the bdi
 299                  * is now congested.  Back off and let other work structs
 300                  * run instead
 301                  */
 302                 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
 303                     fs_info->fs_devices->open_devices > 1) {
 304                         struct io_context *ioc;
 305
 306                         ioc = current->io_context;
 307
 308                         /*
 309                          * the main goal here is that we don't want to
 310                          * block if we're going to be able to submit
 311                          * more requests without blocking.
 312                          *
 313                          * This code does two great things, it pokes into
 314                          * the elevator code from a filesystem _and_
 315                          * it makes assumptions about how batching works.
 316                          */
 317                         if (ioc && ioc->nr_batch_requests > 0 &&
 318                             time_before(jiffies, ioc->last_waited + HZ/50UL) &&
 319                             (last_waited == 0 ||
 320                              ioc->last_waited == last_waited)) {
 321                                 /*
 322                                  * we want to go through our batch of
 323                                  * requests and stop.  So, we copy out
 324                                  * the ioc->last_waited time and test
 325                                  * against it before looping
 326                                  */
 327                                 last_waited = ioc->last_waited;
 328                                 if (need_resched())
 329                                         cond_resched();
 330                                 continue;
 331                         }
 332                         spin_lock(&device->io_lock);
 333                         requeue_list(pending_bios, pending, tail);
 334                         device->running_pending = 1;
 335
 336                         spin_unlock(&device->io_lock);
 337                         btrfs_requeue_work(&device->work);
 338                         goto done;
 339                 }
 340                 /* unplug every 64 requests just for good measure */
 341                 if (batch_run % 64 == 0) {
 342                         blk_finish_plug(&plug);
 343                         blk_start_plug(&plug);
 344                         sync_pending = 0;
 345                 }
 346         }
 347
 348         cond_resched();
 349         if (again)
 350                 goto loop;
 351
 352         spin_lock(&device->io_lock);
 353         if (device->pending_bios.head || device->pending_sync_bios.head)
 354                 goto loop_lock;
 355         spin_unlock(&device->io_lock);
 356
 357 done:
 358         blk_finish_plug(&plug);
 359 }
 360
 361 static void pending_bios_fn(struct btrfs_work *work)
 362 {
 363         struct btrfs_device *device;
 364
 365         device = container_of(work, struct btrfs_device, work);
 366         run_scheduled_bios(device);
 367 }
 368
 369 static noinline int device_list_add(const char *path,
 370                            struct btrfs_super_block *disk_super,
 371                            u64 devid, struct btrfs_fs_devices **fs_devices_ret)
 372 {
 373         struct btrfs_device *device;
 374         struct btrfs_fs_devices *fs_devices;
 375         struct rcu_string *name;
 376         u64 found_transid = btrfs_super_generation(disk_super);
 377
 378         fs_devices = find_fsid(disk_super->fsid);
 379         if (!fs_devices) {
 380                 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
 381                 if (!fs_devices)
 382                         return -ENOMEM;
 383                 INIT_LIST_HEAD(&fs_devices->devices);
 384                 INIT_LIST_HEAD(&fs_devices->alloc_list);
 385                 list_add(&fs_devices->list, &fs_uuids);
 386                 memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
 387                 fs_devices->latest_devid = devid;
 388                 fs_devices->latest_trans = found_transid;
 389                 mutex_init(&fs_devices->device_list_mutex);
 390                 device = NULL;
 391         } else {
 392                 device = __find_device(&fs_devices->devices, devid,
 393                                        disk_super->dev_item.uuid);
 394         }
 395         if (!device) {
 396                 if (fs_devices->opened)
 397                         return -EBUSY;
 398
 399                 device = kzalloc(sizeof(*device), GFP_NOFS);
 400                 if (!device) {
 401                         /* we can safely leave the fs_devices entry around */
 402                         return -ENOMEM;
 403                 }
 404                 device->devid = devid;
 405                 device->dev_stats_valid = 0;
 406                 device->work.func = pending_bios_fn;
 407                 memcpy(device->uuid, disk_super->dev_item.uuid,
 408                        BTRFS_UUID_SIZE);
 409                 spin_lock_init(&device->io_lock);
 410
 411                 name = rcu_string_strdup(path, GFP_NOFS);
 412                 if (!name) {
 413                         kfree(device);
 414                         return -ENOMEM;
 415                 }
 416                 rcu_assign_pointer(device->name, name);
 417                 INIT_LIST_HEAD(&device->dev_alloc_list);
 418
 419                 /* init readahead state */
 420                 spin_lock_init(&device->reada_lock);
 421                 device->reada_curr_zone = NULL;
 422                 atomic_set(&device->reada_in_flight, 0);
 423                 device->reada_next = 0;
 424                 INIT_RADIX_TREE(&device->reada_zones, GFP_NOFS & ~__GFP_WAIT);
 425                 INIT_RADIX_TREE(&device->reada_extents, GFP_NOFS & ~__GFP_WAIT);
 426
 427                 mutex_lock(&fs_devices->device_list_mutex);
 428                 list_add_rcu(&device->dev_list, &fs_devices->devices);
 429                 mutex_unlock(&fs_devices->device_list_mutex);
 430
 431                 device->fs_devices = fs_devices;
 432                 fs_devices->num_devices++;
 433         } else if (!device->name || strcmp(device->name->str, path)) {
 434                 name = rcu_string_strdup(path, GFP_NOFS);
 435                 if (!name)
 436                         return -ENOMEM;
 437                 rcu_string_free(device->name);
 438                 rcu_assign_pointer(device->name, name);
 439                 if (device->missing) {
 440                         fs_devices->missing_devices--;
 441                         device->missing = 0;
 442                 }
 443         }
 444
 445         if (found_transid > fs_devices->latest_trans) {
 446                 fs_devices->latest_devid = devid;
 447                 fs_devices->latest_trans = found_transid;
 448         }
 449         *fs_devices_ret = fs_devices;
 450         return 0;
 451 }
 452
 453 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
 454 {
 455         struct btrfs_fs_devices *fs_devices;
 456         struct btrfs_device *device;
 457         struct btrfs_device *orig_dev;
 458
 459         fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
 460         if (!fs_devices)
 461                 return ERR_PTR(-ENOMEM);
 462
 463         INIT_LIST_HEAD(&fs_devices->devices);
 464         INIT_LIST_HEAD(&fs_devices->alloc_list);
 465         INIT_LIST_HEAD(&fs_devices->list);
 466         mutex_init(&fs_devices->device_list_mutex);
 467         fs_devices->latest_devid = orig->latest_devid;
 468         fs_devices->latest_trans = orig->latest_trans;
 469         fs_devices->total_devices = orig->total_devices;
 470         memcpy(fs_devices->fsid, orig->fsid, sizeof(fs_devices->fsid));
 471
 472         /* We have held the volume lock, it is safe to get the devices. */
 473         list_for_each_entry(orig_dev, &orig->devices, dev_list) {
 474                 struct rcu_string *name;
 475
 476                 device = kzalloc(sizeof(*device), GFP_NOFS);
 477                 if (!device)
 478                         goto error;
 479
 480                 /*
 481                  * This is ok to do without rcu read locked because we hold the
 482                  * uuid mutex so nothing we touch in here is going to disappear.
 483                  */
 484                 name = rcu_string_strdup(orig_dev->name->str, GFP_NOFS);
 485                 if (!name) {
 486                         kfree(device);
 487                         goto error;
 488                 }
 489                 rcu_assign_pointer(device->name, name);
 490
 491                 device->devid = orig_dev->devid;
 492                 device->work.func = pending_bios_fn;
 493                 memcpy(device->uuid, orig_dev->uuid, sizeof(device->uuid));
 494                 spin_lock_init(&device->io_lock);
 495                 INIT_LIST_HEAD(&device->dev_list);
 496                 INIT_LIST_HEAD(&device->dev_alloc_list);
 497
 498                 list_add(&device->dev_list, &fs_devices->devices);
 499                 device->fs_devices = fs_devices;
 500                 fs_devices->num_devices++;
 501         }
 502         return fs_devices;
 503 error:
 504         free_fs_devices(fs_devices);
 505         return ERR_PTR(-ENOMEM);
 506 }
 507
 508 void btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices)
 509 {
 510         struct btrfs_device *device, *next;
 511
 512         struct block_device *latest_bdev = NULL;
 513         u64 latest_devid = 0;
 514         u64 latest_transid = 0;
 515
 516         mutex_lock(&uuid_mutex);
 517 again:
 518         /* This is the initialized path, it is safe to release the devices. */
 519         list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
 520                 if (device->in_fs_metadata) {
 521                         if (!device->is_tgtdev_for_dev_replace &&
 522                             (!latest_transid ||
 523                              device->generation > latest_transid)) {
 524                                 latest_devid = device->devid;
 525                                 latest_transid = device->generation;
 526                                 latest_bdev = device->bdev;
 527                         }
 528                         continue;
 529                 }
 530
 531                 if (device->bdev) {
 532                         blkdev_put(device->bdev, device->mode);
 533                         device->bdev = NULL;
 534                         fs_devices->open_devices--;
 535                 }
 536                 if (device->writeable) {
 537                         list_del_init(&device->dev_alloc_list);
 538                         device->writeable = 0;
 539                         fs_devices->rw_devices--;
 540                 }
 541                 list_del_init(&device->dev_list);
 542                 fs_devices->num_devices--;
 543                 rcu_string_free(device->name);
 544                 kfree(device);
 545         }
 546
 547         if (fs_devices->seed) {
 548                 fs_devices = fs_devices->seed;
 549                 goto again;
 550         }
 551
 552         fs_devices->latest_bdev = latest_bdev;
 553         fs_devices->latest_devid = latest_devid;
 554         fs_devices->latest_trans = latest_transid;
 555
 556         mutex_unlock(&uuid_mutex);
 557 }
 558
 559 static void __free_device(struct work_struct *work)
 560 {
 561         struct btrfs_device *device;
 562
 563         device = container_of(work, struct btrfs_device, rcu_work);
 564
 565         if (device->bdev)
 566                 blkdev_put(device->bdev, device->mode);
 567
 568         rcu_string_free(device->name);
 569         kfree(device);
 570 }
 571
 572 static void free_device(struct rcu_head *head)
 573 {
 574         struct btrfs_device *device;
 575
 576         device = container_of(head, struct btrfs_device, rcu);
 577
 578         INIT_WORK(&device->rcu_work, __free_device);
 579         schedule_work(&device->rcu_work);
 580 }
 581
 582 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
 583 {
 584         struct btrfs_device *device;
 585
 586         if (--fs_devices->opened > 0)
 587                 return 0;
 588
 589         mutex_lock(&fs_devices->device_list_mutex);
 590         list_for_each_entry(device, &fs_devices->devices, dev_list) {
 591                 struct btrfs_device *new_device;
 592                 struct rcu_string *name;
 593
 594                 if (device->bdev)
 595                         fs_devices->open_devices--;
 596
 597                 if (device->writeable) {
 598                         list_del_init(&device->dev_alloc_list);
 599                         fs_devices->rw_devices--;
 600                 }
 601
 602                 if (device->can_discard)
 603                         fs_devices->num_can_discard--;
 604
 605                 new_device = kmalloc(sizeof(*new_device), GFP_NOFS);
 606                 BUG_ON(!new_device); /* -ENOMEM */
 607                 memcpy(new_device, device, sizeof(*new_device));
 608
 609                 /* Safe because we are under uuid_mutex */
 610                 if (device->name) {
 611                         name = rcu_string_strdup(device->name->str, GFP_NOFS);
 612                         BUG_ON(device->name && !name); /* -ENOMEM */
 613                         rcu_assign_pointer(new_device->name, name);
 614                 }
 615                 new_device->bdev = NULL;
 616                 new_device->writeable = 0;
 617                 new_device->in_fs_metadata = 0;
 618                 new_device->can_discard = 0;
 619                 list_replace_rcu(&device->dev_list, &new_device->dev_list);
 620
 621                 call_rcu(&device->rcu, free_device);
 622         }
 623         mutex_unlock(&fs_devices->device_list_mutex);
 624
 625         WARN_ON(fs_devices->open_devices);
 626         WARN_ON(fs_devices->rw_devices);
 627         fs_devices->opened = 0;
 628         fs_devices->seeding = 0;
 629
 630         return 0;
 631 }
 632
 633 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
 634 {
 635         struct btrfs_fs_devices *seed_devices = NULL;
 636         int ret;
 637
 638         mutex_lock(&uuid_mutex);
 639         ret = __btrfs_close_devices(fs_devices);
 640         if (!fs_devices->opened) {
 641                 seed_devices = fs_devices->seed;
 642                 fs_devices->seed = NULL;
 643         }
 644         mutex_unlock(&uuid_mutex);
 645
 646         while (seed_devices) {
 647                 fs_devices = seed_devices;
 648                 seed_devices = fs_devices->seed;
 649                 __btrfs_close_devices(fs_devices);
 650                 free_fs_devices(fs_devices);
 651         }
 652         return ret;
 653 }
 654
 655 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
 656                                 fmode_t flags, void *holder)
 657 {
 658         struct request_queue *q;
 659         struct block_device *bdev;
 660         struct list_head *head = &fs_devices->devices;
 661         struct btrfs_device *device;
 662         struct block_device *latest_bdev = NULL;
 663         struct buffer_head *bh;
 664         struct btrfs_super_block *disk_super;
 665         u64 latest_devid = 0;
 666         u64 latest_transid = 0;
 667         u64 devid;
 668         int seeding = 1;
 669         int ret = 0;
 670
 671         flags |= FMODE_EXCL;
 672
 673         list_for_each_entry(device, head, dev_list) {
 674                 if (device->bdev)
 675                         continue;
 676                 if (!device->name)
 677                         continue;
 678
 679                 ret = btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1,
 680                                             &bdev, &bh);
 681                 if (ret)
 682                         continue;
 683
 684                 disk_super = (struct btrfs_super_block *)bh->b_data;
 685                 devid = btrfs_stack_device_id(&disk_super->dev_item);
 686                 if (devid != device->devid)
 687                         goto error_brelse;
 688
 689                 if (memcmp(device->uuid, disk_super->dev_item.uuid,
 690                            BTRFS_UUID_SIZE))
 691                         goto error_brelse;
 692
 693                 device->generation = btrfs_super_generation(disk_super);
 694                 if (!latest_transid || device->generation > latest_transid) {
 695                         latest_devid = devid;
 696                         latest_transid = device->generation;
 697                         latest_bdev = bdev;
 698                 }
 699
 700                 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
 701                         device->writeable = 0;
 702                 } else {
 703                         device->writeable = !bdev_read_only(bdev);
 704                         seeding = 0;
 705                 }
 706
 707                 q = bdev_get_queue(bdev);
 708                 if (blk_queue_discard(q)) {
 709                         device->can_discard = 1;
 710                         fs_devices->num_can_discard++;
 711                 }
 712
 713                 device->bdev = bdev;
 714                 device->in_fs_metadata = 0;
 715                 device->mode = flags;
 716
 717                 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
 718                         fs_devices->rotating = 1;
 719
 720                 fs_devices->open_devices++;
 721                 if (device->writeable) {
 722                         fs_devices->rw_devices++;
 723                         list_add(&device->dev_alloc_list,
 724                                  &fs_devices->alloc_list);
 725                 }
 726                 brelse(bh);
 727                 continue;
 728
 729 error_brelse:
 730                 brelse(bh);
 731                 blkdev_put(bdev, flags);
 732                 continue;
 733         }
 734         if (fs_devices->open_devices == 0) {
 735                 ret = -EINVAL;
 736                 goto out;
 737         }
 738         fs_devices->seeding = seeding;
 739         fs_devices->opened = 1;
 740         fs_devices->latest_bdev = latest_bdev;
 741         fs_devices->latest_devid = latest_devid;
 742         fs_devices->latest_trans = latest_transid;
 743         fs_devices->total_rw_bytes = 0;
 744 out:
 745         return ret;
 746 }
 747
 748 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
 749                        fmode_t flags, void *holder)
 750 {
 751         int ret;
 752
 753         mutex_lock(&uuid_mutex);
 754         if (fs_devices->opened) {
 755                 fs_devices->opened++;
 756                 ret = 0;
 757         } else {
 758                 ret = __btrfs_open_devices(fs_devices, flags, holder);
 759         }
 760         mutex_unlock(&uuid_mutex);
 761         return ret;
 762 }
 763
 764 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
 765                           struct btrfs_fs_devices **fs_devices_ret)
 766 {
 767         struct btrfs_super_block *disk_super;
 768         struct block_device *bdev;
 769         struct buffer_head *bh;
 770         int ret;
 771         u64 devid;
 772         u64 transid;
 773         u64 total_devices;
 774
 775         flags |= FMODE_EXCL;
 776         mutex_lock(&uuid_mutex);
 777         ret = btrfs_get_bdev_and_sb(path, flags, holder, 0, &bdev, &bh);
 778         if (ret)
 779                 goto error;
 780         disk_super = (struct btrfs_super_block *)bh->b_data;
 781         devid = btrfs_stack_device_id(&disk_super->dev_item);
 782         transid = btrfs_super_generation(disk_super);
 783         total_devices = btrfs_super_num_devices(disk_super);
 784         if (disk_super->label[0]) {
 785                 if (disk_super->label[BTRFS_LABEL_SIZE - 1])
 786                         disk_super->label[BTRFS_LABEL_SIZE - 1] = '\0';
 787                 printk(KERN_INFO "device label %s ", disk_super->label);
 788         } else {
 789                 printk(KERN_INFO "device fsid %pU ", disk_super->fsid);
 790         }
 791         printk(KERN_CONT "devid %llu transid %llu %s\n",
 792                (unsigned long long)devid, (unsigned long long)transid, path);
 793         ret = device_list_add(path, disk_super, devid, fs_devices_ret);
 794         if (!ret && fs_devices_ret)
 795                 (*fs_devices_ret)->total_devices = total_devices;
 796         brelse(bh);
 797         blkdev_put(bdev, flags);
 798 error:
 799         mutex_unlock(&uuid_mutex);
 800         return ret;
 801 }
 802
 803 /* helper to account the used device space in the range */
 804 int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
 805                                    u64 end, u64 *length)
 806 {
 807         struct btrfs_key key;
 808         struct btrfs_root *root = device->dev_root;
 809         struct btrfs_dev_extent *dev_extent;
 810         struct btrfs_path *path;
 811         u64 extent_end;
 812         int ret;
 813         int slot;
 814         struct extent_buffer *l;
 815
 816         *length = 0;
 817
 818         if (start >= device->total_bytes || device->is_tgtdev_for_dev_replace)
 819                 return 0;
 820
 821         path = btrfs_alloc_path();
 822         if (!path)
 823                 return -ENOMEM;
 824         path->reada = 2;
 825
 826         key.objectid = device->devid;
 827         key.offset = start;
 828         key.type = BTRFS_DEV_EXTENT_KEY;
 829
 830         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 831         if (ret < 0)
 832                 goto out;
 833         if (ret > 0) {
 834                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
 835                 if (ret < 0)
 836                         goto out;
 837         }
 838
 839         while (1) {
 840                 l = path->nodes[0];
 841                 slot = path->slots[0];
 842                 if (slot >= btrfs_header_nritems(l)) {
 843                         ret = btrfs_next_leaf(root, path);
 844                         if (ret == 0)
 845                                 continue;
 846                         if (ret < 0)
 847                                 goto out;
 848
 849                         break;
 850                 }
 851                 btrfs_item_key_to_cpu(l, &key, slot);
 852
 853                 if (key.objectid < device->devid)
 854                         goto next;
 855
 856                 if (key.objectid > device->devid)
 857                         break;
 858
 859                 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
 860                         goto next;
 861
 862                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
 863                 extent_end = key.offset + btrfs_dev_extent_length(l,
 864                                                                   dev_extent);
 865                 if (key.offset <= start && extent_end > end) {
 866                         *length = end - start + 1;
 867                         break;
 868                 } else if (key.offset <= start && extent_end > start)
 869                         *length += extent_end - start;
 870                 else if (key.offset > start && extent_end <= end)
 871                         *length += extent_end - key.offset;
 872                 else if (key.offset > start && key.offset <= end) {
 873                         *length += end - key.offset + 1;
 874                         break;
 875                 } else if (key.offset > end)
 876                         break;
 877
 878 next:
 879                 path->slots[0]++;
 880         }
 881         ret = 0;
 882 out:
 883         btrfs_free_path(path);
 884         return ret;
 885 }
 886
 887 /*
 888  * find_free_dev_extent - find free space in the specified device
 889  * @device:     the device which we search the free space in
 890  * @num_bytes:  the size of the free space that we need
 891  * @start:      store the start of the free space.
 892  * @len:        the size of the free space. that we find, or the size of the max
 893  *              free space if we don't find suitable free space
 894  *
 895  * this uses a pretty simple search, the expectation is that it is
 896  * called very infrequently and that a given device has a small number
 897  * of extents
 898  *
 899  * @start is used to store the start of the free space if we find. But if we
 900  * don't find suitable free space, it will be used to store the start position
 901  * of the max free space.
 902  *
 903  * @len is used to store the size of the free space that we find.
 904  * But if we don't find suitable free space, it is used to store the size of
 905  * the max free space.
 906  */
 907 int find_free_dev_extent(struct btrfs_device *device, u64 num_bytes,
 908                          u64 *start, u64 *len)
 909 {
 910         struct btrfs_key key;
 911         struct btrfs_root *root = device->dev_root;
 912         struct btrfs_dev_extent *dev_extent;
 913         struct btrfs_path *path;
 914         u64 hole_size;
 915         u64 max_hole_start;
 916         u64 max_hole_size;
 917         u64 extent_end;
 918         u64 search_start;
 919         u64 search_end = device->total_bytes;
 920         int ret;
 921         int slot;
 922         struct extent_buffer *l;
 923
 924         /* FIXME use last free of some kind */
 925
 926         /* we don't want to overwrite the superblock on the drive,
 927          * so we make sure to start at an offset of at least 1MB
 928          */
 929         search_start = max(root->fs_info->alloc_start, 1024ull * 1024);
 930
 931         max_hole_start = search_start;
 932         max_hole_size = 0;
 933         hole_size = 0;
 934
 935         if (search_start >= search_end || device->is_tgtdev_for_dev_replace) {
 936                 ret = -ENOSPC;
 937                 goto error;
 938         }
 939
 940         path = btrfs_alloc_path();
 941         if (!path) {
 942                 ret = -ENOMEM;
 943                 goto error;
 944         }
 945         path->reada = 2;
 946
 947         key.objectid = device->devid;
 948         key.offset = search_start;
 949         key.type = BTRFS_DEV_EXTENT_KEY;
 950
 951         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 952         if (ret < 0)
 953                 goto out;
 954         if (ret > 0) {
 955                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
 956                 if (ret < 0)
 957                         goto out;
 958         }
 959
 960         while (1) {
 961                 l = path->nodes[0];
 962                 slot = path->slots[0];
 963                 if (slot >= btrfs_header_nritems(l)) {
 964                         ret = btrfs_next_leaf(root, path);
 965                         if (ret == 0)
 966                                 continue;
 967                         if (ret < 0)
 968                                 goto out;
 969
 970                         break;
 971                 }
 972                 btrfs_item_key_to_cpu(l, &key, slot);
 973
 974                 if (key.objectid < device->devid)
 975                         goto next;
 976
 977                 if (key.objectid > device->devid)
 978                         break;
 979
 980                 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
 981                         goto next;
 982
 983                 if (key.offset > search_start) {
 984                         hole_size = key.offset - search_start;
 985
 986                         if (hole_size > max_hole_size) {
 987                                 max_hole_start = search_start;
 988                                 max_hole_size = hole_size;
 989                         }
 990
 991                         /*
 992                          * If this free space is greater than which we need,
 993                          * it must be the max free space that we have found
 994                          * until now, so max_hole_start must point to the start
 995                          * of this free space and the length of this free space
 996                          * is stored in max_hole_size. Thus, we return
 997                          * max_hole_start and max_hole_size and go back to the
 998                          * caller.
 999                          */
1000                         if (hole_size >= num_bytes) {
1001                                 ret = 0;
1002                                 goto out;
1003                         }
1004                 }
1005
1006                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1007                 extent_end = key.offset + btrfs_dev_extent_length(l,
1008                                                                   dev_extent);
1009                 if (extent_end > search_start)
1010                         search_start = extent_end;
1011 next:
1012                 path->slots[0]++;
1013                 cond_resched();
1014         }
1015
1016         /*
1017          * At this point, search_start should be the end of
1018          * allocated dev extents, and when shrinking the device,
1019          * search_end may be smaller than search_start.
1020          */
1021         if (search_end > search_start)
1022                 hole_size = search_end - search_start;
1023
1024         if (hole_size > max_hole_size) {
1025                 max_hole_start = search_start;
1026                 max_hole_size = hole_size;
1027         }
1028
1029         /* See above. */
1030         if (hole_size < num_bytes)
1031                 ret = -ENOSPC;
1032         else
1033                 ret = 0;
1034
1035 out:
1036         btrfs_free_path(path);
1037 error:
1038         *start = max_hole_start;
1039         if (len)
1040                 *len = max_hole_size;
1041         return ret;
1042 }
1043
1044 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1045                           struct btrfs_device *device,
1046                           u64 start)
1047 {
1048         int ret;
1049         struct btrfs_path *path;
1050         struct btrfs_root *root = device->dev_root;
1051         struct btrfs_key key;
1052         struct btrfs_key found_key;
1053         struct extent_buffer *leaf = NULL;
1054         struct btrfs_dev_extent *extent = NULL;
1055
1056         path = btrfs_alloc_path();
1057         if (!path)
1058                 return -ENOMEM;
1059
1060         key.objectid = device->devid;
1061         key.offset = start;
1062         key.type = BTRFS_DEV_EXTENT_KEY;
1063 again:
1064         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1065         if (ret > 0) {
1066                 ret = btrfs_previous_item(root, path, key.objectid,
1067                                           BTRFS_DEV_EXTENT_KEY);
1068                 if (ret)
1069                         goto out;
1070                 leaf = path->nodes[0];
1071                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1072                 extent = btrfs_item_ptr(leaf, path->slots[0],
1073                                         struct btrfs_dev_extent);
1074                 BUG_ON(found_key.offset > start || found_key.offset +
1075                        btrfs_dev_extent_length(leaf, extent) < start);
1076                 key = found_key;
1077                 btrfs_release_path(path);
1078                 goto again;
1079         } else if (ret == 0) {
1080                 leaf = path->nodes[0];
1081                 extent = btrfs_item_ptr(leaf, path->slots[0],
1082                                         struct btrfs_dev_extent);
1083         } else {
1084                 btrfs_error(root->fs_info, ret, "Slot search failed");
1085                 goto out;
1086         }
1087
1088         if (device->bytes_used > 0) {
1089                 u64 len = btrfs_dev_extent_length(leaf, extent);
1090                 device->bytes_used -= len;
1091                 spin_lock(&root->fs_info->free_chunk_lock);
1092                 root->fs_info->free_chunk_space += len;
1093                 spin_unlock(&root->fs_info->free_chunk_lock);
1094         }
1095         ret = btrfs_del_item(trans, root, path);
1096         if (ret) {
1097                 btrfs_error(root->fs_info, ret,
1098                             "Failed to remove dev extent item");
1099         }
1100 out:
1101         btrfs_free_path(path);
1102         return ret;
1103 }
1104
1105 int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1106                            struct btrfs_device *device,
1107                            u64 chunk_tree, u64 chunk_objectid,
1108                            u64 chunk_offset, u64 start, u64 num_bytes)
1109 {
1110         int ret;
1111         struct btrfs_path *path;
1112         struct btrfs_root *root = device->dev_root;
1113         struct btrfs_dev_extent *extent;
1114         struct extent_buffer *leaf;
1115         struct btrfs_key key;
1116
1117         WARN_ON(!device->in_fs_metadata);
1118         WARN_ON(device->is_tgtdev_for_dev_replace);
1119         path = btrfs_alloc_path();
1120         if (!path)
1121                 return -ENOMEM;
1122
1123         key.objectid = device->devid;
1124         key.offset = start;
1125         key.type = BTRFS_DEV_EXTENT_KEY;
1126         ret = btrfs_insert_empty_item(trans, root, path, &key,
1127                                       sizeof(*extent));
1128         if (ret)
1129                 goto out;
1130
1131         leaf = path->nodes[0];
1132         extent = btrfs_item_ptr(leaf, path->slots[0],
1133                                 struct btrfs_dev_extent);
1134         btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
1135         btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
1136         btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1137
1138         write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
1139                     (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
1140                     BTRFS_UUID_SIZE);
1141
1142         btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1143         btrfs_mark_buffer_dirty(leaf);
1144 out:
1145         btrfs_free_path(path);
1146         return ret;
1147 }
1148
1149 static noinline int find_next_chunk(struct btrfs_root *root,
1150                                     u64 objectid, u64 *offset)
1151 {
1152         struct btrfs_path *path;
1153         int ret;
1154         struct btrfs_key key;
1155         struct btrfs_chunk *chunk;
1156         struct btrfs_key found_key;
1157
1158         path = btrfs_alloc_path();
1159         if (!path)
1160                 return -ENOMEM;
1161
1162         key.objectid = objectid;
1163         key.offset = (u64)-1;
1164         key.type = BTRFS_CHUNK_ITEM_KEY;
1165
1166         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1167         if (ret < 0)
1168                 goto error;
1169
1170         BUG_ON(ret == 0); /* Corruption */
1171
1172         ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
1173         if (ret) {
1174                 *offset = 0;
1175         } else {
1176                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1177                                       path->slots[0]);
1178                 if (found_key.objectid != objectid)
1179                         *offset = 0;
1180                 else {
1181                         chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
1182                                                struct btrfs_chunk);
1183                         *offset = found_key.offset +
1184                                 btrfs_chunk_length(path->nodes[0], chunk);
1185                 }
1186         }
1187         ret = 0;
1188 error:
1189         btrfs_free_path(path);
1190         return ret;
1191 }
1192
1193 static noinline int find_next_devid(struct btrfs_root *root, u64 *objectid)
1194 {
1195         int ret;
1196         struct btrfs_key key;
1197         struct btrfs_key found_key;
1198         struct btrfs_path *path;
1199
1200         root = root->fs_info->chunk_root;
1201
1202         path = btrfs_alloc_path();
1203         if (!path)
1204                 return -ENOMEM;
1205
1206         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1207         key.type = BTRFS_DEV_ITEM_KEY;
1208         key.offset = (u64)-1;
1209
1210         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1211         if (ret < 0)
1212                 goto error;
1213
1214         BUG_ON(ret == 0); /* Corruption */
1215
1216         ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
1217                                   BTRFS_DEV_ITEM_KEY);
1218         if (ret) {
1219                 *objectid = 1;
1220         } else {
1221                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1222                                       path->slots[0]);
1223                 *objectid = found_key.offset + 1;
1224         }
1225         ret = 0;
1226 error:
1227         btrfs_free_path(path);
1228         return ret;
1229 }
1230
1231 /*
1232  * the device information is stored in the chunk root
1233  * the btrfs_device struct should be fully filled in
1234  */
1235 int btrfs_add_device(struct btrfs_trans_handle *trans,
1236                      struct btrfs_root *root,
1237                      struct btrfs_device *device)
1238 {
1239         int ret;
1240         struct btrfs_path *path;
1241         struct btrfs_dev_item *dev_item;
1242         struct extent_buffer *leaf;
1243         struct btrfs_key key;
1244         unsigned long ptr;
1245
1246         root = root->fs_info->chunk_root;
1247
1248         path = btrfs_alloc_path();
1249         if (!path)
1250                 return -ENOMEM;
1251
1252         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1253         key.type = BTRFS_DEV_ITEM_KEY;
1254         key.offset = device->devid;
1255
1256         ret = btrfs_insert_empty_item(trans, root, path, &key,
1257                                       sizeof(*dev_item));
1258         if (ret)
1259                 goto out;
1260
1261         leaf = path->nodes[0];
1262         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1263
1264         btrfs_set_device_id(leaf, dev_item, device->devid);
1265         btrfs_set_device_generation(leaf, dev_item, 0);
1266         btrfs_set_device_type(leaf, dev_item, device->type);
1267         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1268         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1269         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1270         btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
1271         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1272         btrfs_set_device_group(leaf, dev_item, 0);
1273         btrfs_set_device_seek_speed(leaf, dev_item, 0);
1274         btrfs_set_device_bandwidth(leaf, dev_item, 0);
1275         btrfs_set_device_start_offset(leaf, dev_item, 0);
1276
1277         ptr = (unsigned long)btrfs_device_uuid(dev_item);
1278         write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1279         ptr = (unsigned long)btrfs_device_fsid(dev_item);
1280         write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1281         btrfs_mark_buffer_dirty(leaf);
1282
1283         ret = 0;
1284 out:
1285         btrfs_free_path(path);
1286         return ret;
1287 }
1288
1289 static int btrfs_rm_dev_item(struct btrfs_root *root,
1290                              struct btrfs_device *device)
1291 {
1292         int ret;
1293         struct btrfs_path *path;
1294         struct btrfs_key key;
1295         struct btrfs_trans_handle *trans;
1296
1297         root = root->fs_info->chunk_root;
1298
1299         path = btrfs_alloc_path();
1300         if (!path)
1301                 return -ENOMEM;
1302
1303         trans = btrfs_start_transaction(root, 0);
1304         if (IS_ERR(trans)) {
1305                 btrfs_free_path(path);
1306                 return PTR_ERR(trans);
1307         }
1308         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1309         key.type = BTRFS_DEV_ITEM_KEY;
1310         key.offset = device->devid;
1311         lock_chunks(root);
1312
1313         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1314         if (ret < 0)
1315                 goto out;
1316
1317         if (ret > 0) {
1318                 ret = -ENOENT;
1319                 goto out;
1320         }
1321
1322         ret = btrfs_del_item(trans, root, path);
1323         if (ret)
1324                 goto out;
1325 out:
1326         btrfs_free_path(path);
1327         unlock_chunks(root);
1328         btrfs_commit_transaction(trans, root);
1329         return ret;
1330 }
1331
1332 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
1333 {
1334         struct btrfs_device *device;
1335         struct btrfs_device *next_device;
1336         struct block_device *bdev;
1337         struct buffer_head *bh = NULL;
1338         struct btrfs_super_block *disk_super;
1339         struct btrfs_fs_devices *cur_devices;
1340         u64 all_avail;
1341         u64 devid;
1342         u64 num_devices;
1343         u8 *dev_uuid;
1344         int ret = 0;
1345         bool clear_super = false;
1346
1347         mutex_lock(&uuid_mutex);
1348
1349         all_avail = root->fs_info->avail_data_alloc_bits |
1350                 root->fs_info->avail_system_alloc_bits |
1351                 root->fs_info->avail_metadata_alloc_bits;
1352
1353         if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) &&
1354             root->fs_info->fs_devices->num_devices <= 4) {
1355                 printk(KERN_ERR "btrfs: unable to go below four devices "
1356                        "on raid10\n");
1357                 ret = -EINVAL;
1358                 goto out;
1359         }
1360
1361         if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) &&
1362             root->fs_info->fs_devices->num_devices <= 2) {
1363                 printk(KERN_ERR "btrfs: unable to go below two "
1364                        "devices on raid1\n");
1365                 ret = -EINVAL;
1366                 goto out;
1367         }
1368
1369         if (strcmp(device_path, "missing") == 0) {
1370                 struct list_head *devices;
1371                 struct btrfs_device *tmp;
1372
1373                 device = NULL;
1374                 devices = &root->fs_info->fs_devices->devices;
1375                 /*
1376                  * It is safe to read the devices since the volume_mutex
1377                  * is held.
1378                  */
1379                 list_for_each_entry(tmp, devices, dev_list) {
1380                         if (tmp->in_fs_metadata &&
1381                             !tmp->is_tgtdev_for_dev_replace &&
1382                             !tmp->bdev) {
1383                                 device = tmp;
1384                                 break;
1385                         }
1386                 }
1387                 bdev = NULL;
1388                 bh = NULL;
1389                 disk_super = NULL;
1390                 if (!device) {
1391                         printk(KERN_ERR "btrfs: no missing devices found to "
1392                                "remove\n");
1393                         goto out;
1394                 }
1395         } else {
1396                 ret = btrfs_get_bdev_and_sb(device_path,
1397                                             FMODE_READ | FMODE_EXCL,
1398                                             root->fs_info->bdev_holder, 0,
1399                                             &bdev, &bh);
1400                 if (ret)
1401                         goto out;
1402                 disk_super = (struct btrfs_super_block *)bh->b_data;
1403                 devid = btrfs_stack_device_id(&disk_super->dev_item);
1404                 dev_uuid = disk_super->dev_item.uuid;
1405                 device = btrfs_find_device(root->fs_info, devid, dev_uuid,
1406                                            disk_super->fsid);
1407                 if (!device) {
1408                         ret = -ENOENT;
1409                         goto error_brelse;
1410                 }
1411         }
1412
1413         if (device->is_tgtdev_for_dev_replace) {
1414                 pr_err("btrfs: unable to remove the dev_replace target dev\n");
1415                 ret = -EINVAL;
1416                 goto error_brelse;
1417         }
1418
1419         if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1420                 printk(KERN_ERR "btrfs: unable to remove the only writeable "
1421                        "device\n");
1422                 ret = -EINVAL;
1423                 goto error_brelse;
1424         }
1425
1426         if (device->writeable) {
1427                 lock_chunks(root);
1428                 list_del_init(&device->dev_alloc_list);
1429                 unlock_chunks(root);
1430                 root->fs_info->fs_devices->rw_devices--;
1431                 clear_super = true;
1432         }
1433
1434         ret = btrfs_shrink_device(device, 0);
1435         if (ret)
1436                 goto error_undo;
1437
1438         /*
1439          * TODO: the superblock still includes this device in its num_devices
1440          * counter although write_all_supers() is not locked out. This
1441          * could give a filesystem state which requires a degraded mount.
1442          */
1443         ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1444         if (ret)
1445                 goto error_undo;
1446
1447         spin_lock(&root->fs_info->free_chunk_lock);
1448         root->fs_info->free_chunk_space = device->total_bytes -
1449                 device->bytes_used;
1450         spin_unlock(&root->fs_info->free_chunk_lock);
1451
1452         device->in_fs_metadata = 0;
1453         btrfs_scrub_cancel_dev(root->fs_info, device);
1454
1455         /*
1456          * the device list mutex makes sure that we don't change
1457          * the device list while someone else is writing out all
1458          * the device supers.
1459          */
1460
1461         cur_devices = device->fs_devices;
1462         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1463         list_del_rcu(&device->dev_list);
1464
1465         device->fs_devices->num_devices--;
1466         device->fs_devices->total_devices--;
1467
1468         if (device->missing)
1469                 root->fs_info->fs_devices->missing_devices--;
1470
1471         next_device = list_entry(root->fs_info->fs_devices->devices.next,
1472                                  struct btrfs_device, dev_list);
1473         if (device->bdev == root->fs_info->sb->s_bdev)
1474                 root->fs_info->sb->s_bdev = next_device->bdev;
1475         if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1476                 root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1477
1478         if (device->bdev)
1479                 device->fs_devices->open_devices--;
1480
1481         call_rcu(&device->rcu, free_device);
1482         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1483
1484         num_devices = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
1485         btrfs_set_super_num_devices(root->fs_info->super_copy, num_devices);
1486
1487         if (cur_devices->open_devices == 0) {
1488                 struct btrfs_fs_devices *fs_devices;
1489                 fs_devices = root->fs_info->fs_devices;
1490                 while (fs_devices) {
1491                         if (fs_devices->seed == cur_devices)
1492                                 break;
1493                         fs_devices = fs_devices->seed;
1494                 }
1495                 fs_devices->seed = cur_devices->seed;
1496                 cur_devices->seed = NULL;
1497                 lock_chunks(root);
1498                 __btrfs_close_devices(cur_devices);
1499                 unlock_chunks(root);
1500                 free_fs_devices(cur_devices);
1501         }
1502
1503         root->fs_info->num_tolerated_disk_barrier_failures =
1504                 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
1505
1506         /*
1507          * at this point, the device is zero sized.  We want to
1508          * remove it from the devices list and zero out the old super
1509          */
1510         if (clear_super && disk_super) {
1511                 /* make sure this device isn't detected as part of
1512                  * the FS anymore
1513                  */
1514                 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1515                 set_buffer_dirty(bh);
1516                 sync_dirty_buffer(bh);
1517         }
1518
1519         ret = 0;
1520
1521 error_brelse:
1522         brelse(bh);
1523 error_close:
1524         if (bdev)
1525                 blkdev_put(bdev, FMODE_READ | FMODE_EXCL);
1526 out:
1527         mutex_unlock(&uuid_mutex);
1528         return ret;
1529 error_undo:
1530         if (device->writeable) {
1531                 lock_chunks(root);
1532                 list_add(&device->dev_alloc_list,
1533                          &root->fs_info->fs_devices->alloc_list);
1534                 unlock_chunks(root);
1535                 root->fs_info->fs_devices->rw_devices++;
1536         }
1537         goto error_brelse;
1538 }
1539
1540 void btrfs_rm_dev_replace_srcdev(struct btrfs_fs_info *fs_info,
1541                                  struct btrfs_device *srcdev)
1542 {
1543         WARN_ON(!mutex_is_locked(&fs_info->fs_devices->device_list_mutex));
1544         list_del_rcu(&srcdev->dev_list);
1545         list_del_rcu(&srcdev->dev_alloc_list);
1546         fs_info->fs_devices->num_devices--;
1547         if (srcdev->missing) {
1548                 fs_info->fs_devices->missing_devices--;
1549                 fs_info->fs_devices->rw_devices++;
1550         }
1551         if (srcdev->can_discard)
1552                 fs_info->fs_devices->num_can_discard--;
1553         if (srcdev->bdev)
1554                 fs_info->fs_devices->open_devices--;
1555
1556         call_rcu(&srcdev->rcu, free_device);
1557 }
1558
1559 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
1560                                       struct btrfs_device *tgtdev)
1561 {
1562         struct btrfs_device *next_device;
1563
1564         WARN_ON(!tgtdev);
1565         mutex_lock(&fs_info->fs_devices->device_list_mutex);
1566         if (tgtdev->bdev) {
1567                 btrfs_scratch_superblock(tgtdev);
1568                 fs_info->fs_devices->open_devices--;
1569         }
1570         fs_info->fs_devices->num_devices--;
1571         if (tgtdev->can_discard)
1572                 fs_info->fs_devices->num_can_discard++;
1573
1574         next_device = list_entry(fs_info->fs_devices->devices.next,
1575                                  struct btrfs_device, dev_list);
1576         if (tgtdev->bdev == fs_info->sb->s_bdev)
1577                 fs_info->sb->s_bdev = next_device->bdev;
1578         if (tgtdev->bdev == fs_info->fs_devices->latest_bdev)
1579                 fs_info->fs_devices->latest_bdev = next_device->bdev;
1580         list_del_rcu(&tgtdev->dev_list);
1581
1582         call_rcu(&tgtdev->rcu, free_device);
1583
1584         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
1585 }
1586
1587 int btrfs_find_device_by_path(struct btrfs_root *root, char *device_path,
1588                               struct btrfs_device **device)
1589 {
1590         int ret = 0;
1591         struct btrfs_super_block *disk_super;
1592         u64 devid;
1593         u8 *dev_uuid;
1594         struct block_device *bdev;
1595         struct buffer_head *bh;
1596
1597         *device = NULL;
1598         ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
1599                                     root->fs_info->bdev_holder, 0, &bdev, &bh);
1600         if (ret)
1601                 return ret;
1602         disk_super = (struct btrfs_super_block *)bh->b_data;
1603         devid = btrfs_stack_device_id(&disk_super->dev_item);
1604         dev_uuid = disk_super->dev_item.uuid;
1605         *device = btrfs_find_device(root->fs_info, devid, dev_uuid,
1606                                     disk_super->fsid);
1607         brelse(bh);
1608         if (!*device)
1609                 ret = -ENOENT;
1610         blkdev_put(bdev, FMODE_READ);
1611         return ret;
1612 }
1613
1614 int btrfs_find_device_missing_or_by_path(struct btrfs_root *root,
1615                                          char *device_path,
1616                                          struct btrfs_device **device)
1617 {
1618         *device = NULL;
1619         if (strcmp(device_path, "missing") == 0) {
1620                 struct list_head *devices;
1621                 struct btrfs_device *tmp;
1622
1623                 devices = &root->fs_info->fs_devices->devices;
1624                 /*
1625                  * It is safe to read the devices since the volume_mutex
1626                  * is held by the caller.
1627                  */
1628                 list_for_each_entry(tmp, devices, dev_list) {
1629                         if (tmp->in_fs_metadata && !tmp->bdev) {
1630                                 *device = tmp;
1631                                 break;
1632                         }
1633                 }
1634
1635                 if (!*device) {
1636                         pr_err("btrfs: no missing device found\n");
1637                         return -ENOENT;
1638                 }
1639
1640                 return 0;
1641         } else {
1642                 return btrfs_find_device_by_path(root, device_path, device);
1643         }
1644 }
1645
1646 /*
1647  * does all the dirty work required for changing file system's UUID.
1648  */
1649 static int btrfs_prepare_sprout(struct btrfs_root *root)
1650 {
1651         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
1652         struct btrfs_fs_devices *old_devices;
1653         struct btrfs_fs_devices *seed_devices;
1654         struct btrfs_super_block *disk_super = root->fs_info->super_copy;
1655         struct btrfs_device *device;
1656         u64 super_flags;
1657
1658         BUG_ON(!mutex_is_locked(&uuid_mutex));
1659         if (!fs_devices->seeding)
1660                 return -EINVAL;
1661
1662         seed_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
1663         if (!seed_devices)
1664                 return -ENOMEM;
1665
1666         old_devices = clone_fs_devices(fs_devices);
1667         if (IS_ERR(old_devices)) {
1668                 kfree(seed_devices);
1669                 return PTR_ERR(old_devices);
1670         }
1671
1672         list_add(&old_devices->list, &fs_uuids);
1673
1674         memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
1675         seed_devices->opened = 1;
1676         INIT_LIST_HEAD(&seed_devices->devices);
1677         INIT_LIST_HEAD(&seed_devices->alloc_list);
1678         mutex_init(&seed_devices->device_list_mutex);
1679
1680         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1681         list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
1682                               synchronize_rcu);
1683         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1684
1685         list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
1686         list_for_each_entry(device, &seed_devices->devices, dev_list) {
1687                 device->fs_devices = seed_devices;
1688         }
1689
1690         fs_devices->seeding = 0;
1691         fs_devices->num_devices = 0;
1692         fs_devices->open_devices = 0;
1693         fs_devices->total_devices = 0;
1694         fs_devices->seed = seed_devices;
1695
1696         generate_random_uuid(fs_devices->fsid);
1697         memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1698         memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1699         super_flags = btrfs_super_flags(disk_super) &
1700                       ~BTRFS_SUPER_FLAG_SEEDING;
1701         btrfs_set_super_flags(disk_super, super_flags);
1702
1703         return 0;
1704 }
1705
1706 /*
1707  * strore the expected generation for seed devices in device items.
1708  */
1709 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
1710                                struct btrfs_root *root)
1711 {
1712         struct btrfs_path *path;
1713         struct extent_buffer *leaf;
1714         struct btrfs_dev_item *dev_item;
1715         struct btrfs_device *device;
1716         struct btrfs_key key;
1717         u8 fs_uuid[BTRFS_UUID_SIZE];
1718         u8 dev_uuid[BTRFS_UUID_SIZE];
1719         u64 devid;
1720         int ret;
1721
1722         path = btrfs_alloc_path();
1723         if (!path)
1724                 return -ENOMEM;
1725
1726         root = root->fs_info->chunk_root;
1727         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1728         key.offset = 0;
1729         key.type = BTRFS_DEV_ITEM_KEY;
1730
1731         while (1) {
1732                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1733                 if (ret < 0)
1734                         goto error;
1735
1736                 leaf = path->nodes[0];
1737 next_slot:
1738                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1739                         ret = btrfs_next_leaf(root, path);
1740                         if (ret > 0)
1741                                 break;
1742                         if (ret < 0)
1743                                 goto error;
1744                         leaf = path->nodes[0];
1745                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1746                         btrfs_release_path(path);
1747                         continue;
1748                 }
1749
1750                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1751                 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
1752                     key.type != BTRFS_DEV_ITEM_KEY)
1753                         break;
1754
1755                 dev_item = btrfs_item_ptr(leaf, path->slots[0],
1756                                           struct btrfs_dev_item);
1757                 devid = btrfs_device_id(leaf, dev_item);
1758                 read_extent_buffer(leaf, dev_uuid,
1759                                    (unsigned long)btrfs_device_uuid(dev_item),
1760                                    BTRFS_UUID_SIZE);
1761                 read_extent_buffer(leaf, fs_uuid,
1762                                    (unsigned long)btrfs_device_fsid(dev_item),
1763                                    BTRFS_UUID_SIZE);
1764                 device = btrfs_find_device(root->fs_info, devid, dev_uuid,
1765                                            fs_uuid);
1766                 BUG_ON(!device); /* Logic error */
1767
1768                 if (device->fs_devices->seeding) {
1769                         btrfs_set_device_generation(leaf, dev_item,
1770                                                     device->generation);
1771                         btrfs_mark_buffer_dirty(leaf);
1772                 }
1773
1774                 path->slots[0]++;
1775                 goto next_slot;
1776         }
1777         ret = 0;
1778 error:
1779         btrfs_free_path(path);
1780         return ret;
1781 }
1782
1783 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
1784 {
1785         struct request_queue *q;
1786         struct btrfs_trans_handle *trans;
1787         struct btrfs_device *device;
1788         struct block_device *bdev;
1789         struct list_head *devices;
1790         struct super_block *sb = root->fs_info->sb;
1791         struct rcu_string *name;
1792         u64 total_bytes;
1793         int seeding_dev = 0;
1794         int ret = 0;
1795
1796         if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
1797                 return -EROFS;
1798
1799         bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
1800                                   root->fs_info->bdev_holder);
1801         if (IS_ERR(bdev))
1802                 return PTR_ERR(bdev);
1803
1804         if (root->fs_info->fs_devices->seeding) {
1805                 seeding_dev = 1;
1806                 down_write(&sb->s_umount);
1807                 mutex_lock(&uuid_mutex);
1808         }
1809
1810         filemap_write_and_wait(bdev->bd_inode->i_mapping);
1811
1812         devices = &root->fs_info->fs_devices->devices;
1813
1814         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1815         list_for_each_entry(device, devices, dev_list) {
1816                 if (device->bdev == bdev) {
1817                         ret = -EEXIST;
1818                         mutex_unlock(
1819                                 &root->fs_info->fs_devices->device_list_mutex);
1820                         goto error;
1821                 }
1822         }
1823         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1824
1825         device = kzalloc(sizeof(*device), GFP_NOFS);
1826         if (!device) {
1827                 /* we can safely leave the fs_devices entry around */
1828                 ret = -ENOMEM;
1829                 goto error;
1830         }
1831
1832         name = rcu_string_strdup(device_path, GFP_NOFS);
1833         if (!name) {
1834                 kfree(device);
1835                 ret = -ENOMEM;
1836                 goto error;
1837         }
1838         rcu_assign_pointer(device->name, name);
1839
1840         ret = find_next_devid(root, &device->devid);
1841         if (ret) {
1842                 rcu_string_free(device->name);
1843                 kfree(device);
1844                 goto error;
1845         }
1846
1847         trans = btrfs_start_transaction(root, 0);
1848         if (IS_ERR(trans)) {
1849                 rcu_string_free(device->name);
1850                 kfree(device);
1851                 ret = PTR_ERR(trans);
1852                 goto error;
1853         }
1854
1855         lock_chunks(root);
1856
1857         q = bdev_get_queue(bdev);
1858         if (blk_queue_discard(q))
1859                 device->can_discard = 1;
1860         device->writeable = 1;
1861         device->work.func = pending_bios_fn;
1862         generate_random_uuid(device->uuid);
1863         spin_lock_init(&device->io_lock);
1864         device->generation = trans->transid;
1865         device->io_width = root->sectorsize;
1866         device->io_align = root->sectorsize;
1867         device->sector_size = root->sectorsize;
1868         device->total_bytes = i_size_read(bdev->bd_inode);
1869         device->disk_total_bytes = device->total_bytes;
1870         device->dev_root = root->fs_info->dev_root;
1871         device->bdev = bdev;
1872         device->in_fs_metadata = 1;
1873         device->is_tgtdev_for_dev_replace = 0;
1874         device->mode = FMODE_EXCL;
1875         set_blocksize(device->bdev, 4096);
1876
1877         if (seeding_dev) {
1878                 sb->s_flags &= ~MS_RDONLY;
1879                 ret = btrfs_prepare_sprout(root);
1880                 BUG_ON(ret); /* -ENOMEM */
1881         }
1882
1883         device->fs_devices = root->fs_info->fs_devices;
1884
1885         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1886         list_add_rcu(&device->dev_list, &root->fs_info->fs_devices->devices);
1887         list_add(&device->dev_alloc_list,
1888                  &root->fs_info->fs_devices->alloc_list);
1889         root->fs_info->fs_devices->num_devices++;
1890         root->fs_info->fs_devices->open_devices++;
1891         root->fs_info->fs_devices->rw_devices++;
1892         root->fs_info->fs_devices->total_devices++;
1893         if (device->can_discard)
1894                 root->fs_info->fs_devices->num_can_discard++;
1895         root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
1896
1897         spin_lock(&root->fs_info->free_chunk_lock);
1898         root->fs_info->free_chunk_space += device->total_bytes;
1899         spin_unlock(&root->fs_info->free_chunk_lock);
1900
1901         if (!blk_queue_nonrot(bdev_get_queue(bdev)))
1902                 root->fs_info->fs_devices->rotating = 1;
1903
1904         total_bytes = btrfs_super_total_bytes(root->fs_info->super_copy);
1905         btrfs_set_super_total_bytes(root->fs_info->super_copy,
1906                                     total_bytes + device->total_bytes);
1907
1908         total_bytes = btrfs_super_num_devices(root->fs_info->super_copy);
1909         btrfs_set_super_num_devices(root->fs_info->super_copy,
1910                                     total_bytes + 1);
1911         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1912
1913         if (seeding_dev) {
1914                 ret = init_first_rw_device(trans, root, device);
1915                 if (ret) {
1916                         btrfs_abort_transaction(trans, root, ret);
1917                         goto error_trans;
1918                 }
1919                 ret = btrfs_finish_sprout(trans, root);
1920                 if (ret) {
1921                         btrfs_abort_transaction(trans, root, ret);
1922                         goto error_trans;
1923                 }
1924         } else {
1925                 ret = btrfs_add_device(trans, root, device);
1926                 if (ret) {
1927                         btrfs_abort_transaction(trans, root, ret);
1928                         goto error_trans;
1929                 }
1930         }
1931
1932         /*
1933          * we've got more storage, clear any full flags on the space
1934          * infos
1935          */
1936         btrfs_clear_space_info_full(root->fs_info);
1937
1938         unlock_chunks(root);
1939         root->fs_info->num_tolerated_disk_barrier_failures =
1940                 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
1941         ret = btrfs_commit_transaction(trans, root);
1942
1943         if (seeding_dev) {
1944                 mutex_unlock(&uuid_mutex);
1945                 up_write(&sb->s_umount);
1946
1947                 if (ret) /* transaction commit */
1948                         return ret;
1949
1950                 ret = btrfs_relocate_sys_chunks(root);
1951                 if (ret < 0)
1952                         btrfs_error(root->fs_info, ret,
1953                                     "Failed to relocate sys chunks after "
1954                                     "device initialization. This can be fixed "
1955                                     "using the \"btrfs balance\" command.");
1956                 trans = btrfs_attach_transaction(root);
1957                 if (IS_ERR(trans)) {
1958                         if (PTR_ERR(trans) == -ENOENT)
1959                                 return 0;
1960                         return PTR_ERR(trans);
1961                 }
1962                 ret = btrfs_commit_transaction(trans, root);
1963         }
1964
1965         return ret;
1966
1967 error_trans:
1968         unlock_chunks(root);
1969         btrfs_end_transaction(trans, root);
1970         rcu_string_free(device->name);
1971         kfree(device);
1972 error:
1973         blkdev_put(bdev, FMODE_EXCL);
1974         if (seeding_dev) {
1975                 mutex_unlock(&uuid_mutex);
1976                 up_write(&sb->s_umount);
1977         }
1978         return ret;
1979 }
1980
1981 int btrfs_init_dev_replace_tgtdev(struct btrfs_root *root, char *device_path,
1982                                   struct btrfs_device **device_out)
1983 {
1984         struct request_queue *q;
1985         struct btrfs_device *device;
1986         struct block_device *bdev;
1987         struct btrfs_fs_info *fs_info = root->fs_info;
1988         struct list_head *devices;
1989         struct rcu_string *name;
1990         int ret = 0;
1991
1992         *device_out = NULL;
1993         if (fs_info->fs_devices->seeding)
1994                 return -EINVAL;
1995
1996         bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
1997                                   fs_info->bdev_holder);
1998         if (IS_ERR(bdev))
1999                 return PTR_ERR(bdev);
2000
2001         filemap_write_and_wait(bdev->bd_inode->i_mapping);
2002
2003         devices = &fs_info->fs_devices->devices;
2004         list_for_each_entry(device, devices, dev_list) {
2005                 if (device->bdev == bdev) {
2006                         ret = -EEXIST;
2007                         goto error;
2008                 }
2009         }
2010
2011         device = kzalloc(sizeof(*device), GFP_NOFS);
2012         if (!device) {
2013                 ret = -ENOMEM;
2014                 goto error;
2015         }
2016
2017         name = rcu_string_strdup(device_path, GFP_NOFS);
2018         if (!name) {
2019                 kfree(device);
2020                 ret = -ENOMEM;
2021                 goto error;
2022         }
2023         rcu_assign_pointer(device->name, name);
2024
2025         q = bdev_get_queue(bdev);
2026         if (blk_queue_discard(q))
2027                 device->can_discard = 1;
2028         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2029         device->writeable = 1;
2030         device->work.func = pending_bios_fn;
2031         generate_random_uuid(device->uuid);
2032         device->devid = BTRFS_DEV_REPLACE_DEVID;
2033         spin_lock_init(&device->io_lock);
2034         device->generation = 0;
2035         device->io_width = root->sectorsize;
2036         device->io_align = root->sectorsize;
2037         device->sector_size = root->sectorsize;
2038         device->total_bytes = i_size_read(bdev->bd_inode);
2039         device->disk_total_bytes = device->total_bytes;
2040         device->dev_root = fs_info->dev_root;
2041         device->bdev = bdev;
2042         device->in_fs_metadata = 1;
2043         device->is_tgtdev_for_dev_replace = 1;
2044         device->mode = FMODE_EXCL;
2045         set_blocksize(device->bdev, 4096);
2046         device->fs_devices = fs_info->fs_devices;
2047         list_add(&device->dev_list, &fs_info->fs_devices->devices);
2048         fs_info->fs_devices->num_devices++;
2049         fs_info->fs_devices->open_devices++;
2050         if (device->can_discard)
2051                 fs_info->fs_devices->num_can_discard++;
2052         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2053
2054         *device_out = device;
2055         return ret;
2056
2057 error:
2058         blkdev_put(bdev, FMODE_EXCL);
2059         return ret;
2060 }
2061
2062 void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info *fs_info,
2063                                               struct btrfs_device *tgtdev)
2064 {
2065         WARN_ON(fs_info->fs_devices->rw_devices == 0);
2066         tgtdev->io_width = fs_info->dev_root->sectorsize;
2067         tgtdev->io_align = fs_info->dev_root->sectorsize;
2068         tgtdev->sector_size = fs_info->dev_root->sectorsize;
2069         tgtdev->dev_root = fs_info->dev_root;
2070         tgtdev->in_fs_metadata = 1;
2071 }
2072
2073 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2074                                         struct btrfs_device *device)
2075 {
2076         int ret;
2077         struct btrfs_path *path;
2078         struct btrfs_root *root;
2079         struct btrfs_dev_item *dev_item;
2080         struct extent_buffer *leaf;
2081         struct btrfs_key key;
2082
2083         root = device->dev_root->fs_info->chunk_root;
2084
2085         path = btrfs_alloc_path();
2086         if (!path)
2087                 return -ENOMEM;
2088
2089         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2090         key.type = BTRFS_DEV_ITEM_KEY;
2091         key.offset = device->devid;
2092
2093         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2094         if (ret < 0)
2095                 goto out;
2096
2097         if (ret > 0) {
2098                 ret = -ENOENT;
2099                 goto out;
2100         }
2101
2102         leaf = path->nodes[0];
2103         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2104
2105         btrfs_set_device_id(leaf, dev_item, device->devid);
2106         btrfs_set_device_type(leaf, dev_item, device->type);
2107         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2108         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2109         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2110         btrfs_set_device_total_bytes(leaf, dev_item, device->disk_total_bytes);
2111         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
2112         btrfs_mark_buffer_dirty(leaf);
2113
2114 out:
2115         btrfs_free_path(path);
2116         return ret;
2117 }
2118
2119 static int __btrfs_grow_device(struct btrfs_trans_handle *trans,
2120                       struct btrfs_device *device, u64 new_size)
2121 {
2122         struct btrfs_super_block *super_copy =
2123                 device->dev_root->fs_info->super_copy;
2124         u64 old_total = btrfs_super_total_bytes(super_copy);
2125         u64 diff = new_size - device->total_bytes;
2126
2127         if (!device->writeable)
2128                 return -EACCES;
2129         if (new_size <= device->total_bytes ||
2130             device->is_tgtdev_for_dev_replace)
2131                 return -EINVAL;
2132
2133         btrfs_set_super_total_bytes(super_copy, old_total + diff);
2134         device->fs_devices->total_rw_bytes += diff;
2135
2136         device->total_bytes = new_size;
2137         device->disk_total_bytes = new_size;
2138         btrfs_clear_space_info_full(device->dev_root->fs_info);
2139
2140         return btrfs_update_device(trans, device);
2141 }
2142
2143 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2144                       struct btrfs_device *device, u64 new_size)
2145 {
2146         int ret;
2147         lock_chunks(device->dev_root);
2148         ret = __btrfs_grow_device(trans, device, new_size);
2149         unlock_chunks(device->dev_root);
2150         return ret;
2151 }
2152
2153 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
2154                             struct btrfs_root *root,
2155                             u64 chunk_tree, u64 chunk_objectid,
2156                             u64 chunk_offset)
2157 {
2158         int ret;
2159         struct btrfs_path *path;
2160         struct btrfs_key key;
2161
2162         root = root->fs_info->chunk_root;
2163         path = btrfs_alloc_path();
2164         if (!path)
2165                 return -ENOMEM;
2166
2167         key.objectid = chunk_objectid;
2168         key.offset = chunk_offset;
2169         key.type = BTRFS_CHUNK_ITEM_KEY;
2170
2171         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2172         if (ret < 0)
2173                 goto out;
2174         else if (ret > 0) { /* Logic error or corruption */
2175                 btrfs_error(root->fs_info, -ENOENT,
2176                             "Failed lookup while freeing chunk.");
2177                 ret = -ENOENT;
2178                 goto out;
2179         }
2180
2181         ret = btrfs_del_item(trans, root, path);
2182         if (ret < 0)
2183                 btrfs_error(root->fs_info, ret,
2184                             "Failed to delete chunk item.");
2185 out:
2186         btrfs_free_path(path);
2187         return ret;
2188 }
2189
2190 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
2191                         chunk_offset)
2192 {
2193         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
2194         struct btrfs_disk_key *disk_key;
2195         struct btrfs_chunk *chunk;
2196         u8 *ptr;
2197         int ret = 0;
2198         u32 num_stripes;
2199         u32 array_size;
2200         u32 len = 0;
2201         u32 cur;
2202         struct btrfs_key key;
2203
2204         array_size = btrfs_super_sys_array_size(super_copy);
2205
2206         ptr = super_copy->sys_chunk_array;
2207         cur = 0;
2208
2209         while (cur < array_size) {
2210                 disk_key = (struct btrfs_disk_key *)ptr;
2211                 btrfs_disk_key_to_cpu(&key, disk_key);
2212
2213                 len = sizeof(*disk_key);
2214
2215                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2216                         chunk = (struct btrfs_chunk *)(ptr + len);
2217                         num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2218                         len += btrfs_chunk_item_size(num_stripes);
2219                 } else {
2220                         ret = -EIO;
2221                         break;
2222                 }
2223                 if (key.objectid == chunk_objectid &&
2224                     key.offset == chunk_offset) {
2225                         memmove(ptr, ptr + len, array_size - (cur + len));
2226                         array_size -= len;
2227                         btrfs_set_super_sys_array_size(super_copy, array_size);
2228                 } else {
2229                         ptr += len;
2230                         cur += len;
2231                 }
2232         }
2233         return ret;
2234 }
2235
2236 static int btrfs_relocate_chunk(struct btrfs_root *root,
2237                          u64 chunk_tree, u64 chunk_objectid,
2238                          u64 chunk_offset)
2239 {
2240         struct extent_map_tree *em_tree;
2241         struct btrfs_root *extent_root;
2242         struct btrfs_trans_handle *trans;
2243         struct extent_map *em;
2244         struct map_lookup *map;
2245         int ret;
2246         int i;
2247
2248         root = root->fs_info->chunk_root;
2249         extent_root = root->fs_info->extent_root;
2250         em_tree = &root->fs_info->mapping_tree.map_tree;
2251
2252         ret = btrfs_can_relocate(extent_root, chunk_offset);
2253         if (ret)
2254                 return -ENOSPC;
2255
2256         /* step one, relocate all the extents inside this chunk */
2257         ret = btrfs_relocate_block_group(extent_root, chunk_offset);
2258         if (ret)
2259                 return ret;
2260
2261         trans = btrfs_start_transaction(root, 0);
2262         BUG_ON(IS_ERR(trans));
2263
2264         lock_chunks(root);
2265
2266         /*
2267          * step two, delete the device extents and the
2268          * chunk tree entries
2269          */
2270         read_lock(&em_tree->lock);
2271         em = lookup_extent_mapping(em_tree, chunk_offset, 1);
2272         read_unlock(&em_tree->lock);
2273
2274         BUG_ON(!em || em->start > chunk_offset ||
2275                em->start + em->len < chunk_offset);
2276         map = (struct map_lookup *)em->bdev;
2277
2278         for (i = 0; i < map->num_stripes; i++) {
2279                 ret = btrfs_free_dev_extent(trans, map->stripes[i].dev,
2280                                             map->stripes[i].physical);
2281                 BUG_ON(ret);
2282
2283                 if (map->stripes[i].dev) {
2284                         ret = btrfs_update_device(trans, map->stripes[i].dev);
2285                         BUG_ON(ret);
2286                 }
2287         }
2288         ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
2289                                chunk_offset);
2290
2291         BUG_ON(ret);
2292
2293         trace_btrfs_chunk_free(root, map, chunk_offset, em->len);
2294
2295         if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2296                 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
2297                 BUG_ON(ret);
2298         }
2299
2300         ret = btrfs_remove_block_group(trans, extent_root, chunk_offset);
2301         BUG_ON(ret);
2302
2303         write_lock(&em_tree->lock);
2304         remove_extent_mapping(em_tree, em);
2305         write_unlock(&em_tree->lock);
2306
2307         kfree(map);
2308         em->bdev = NULL;
2309
2310         /* once for the tree */
2311         free_extent_map(em);
2312         /* once for us */
2313         free_extent_map(em);
2314
2315         unlock_chunks(root);
2316         btrfs_end_transaction(trans, root);
2317         return 0;
2318 }
2319
2320 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
2321 {
2322         struct btrfs_root *chunk_root = root->fs_info->chunk_root;
2323         struct btrfs_path *path;
2324         struct extent_buffer *leaf;
2325         struct btrfs_chunk *chunk;
2326         struct btrfs_key key;
2327         struct btrfs_key found_key;
2328         u64 chunk_tree = chunk_root->root_key.objectid;
2329         u64 chunk_type;
2330         bool retried = false;
2331         int failed = 0;
2332         int ret;
2333
2334         path = btrfs_alloc_path();
2335         if (!path)
2336                 return -ENOMEM;
2337
2338 again:
2339         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2340         key.offset = (u64)-1;
2341         key.type = BTRFS_CHUNK_ITEM_KEY;
2342
2343         while (1) {
2344                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2345                 if (ret < 0)
2346                         goto error;
2347                 BUG_ON(ret == 0); /* Corruption */
2348
2349                 ret = btrfs_previous_item(chunk_root, path, key.objectid,
2350                                           key.type);
2351                 if (ret < 0)
2352                         goto error;
2353                 if (ret > 0)
2354                         break;
2355
2356                 leaf = path->nodes[0];
2357                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2358
2359                 chunk = btrfs_item_ptr(leaf, path->slots[0],
2360                                        struct btrfs_chunk);
2361                 chunk_type = btrfs_chunk_type(leaf, chunk);
2362                 btrfs_release_path(path);
2363
2364                 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
2365                         ret = btrfs_relocate_chunk(chunk_root, chunk_tree,
2366                                                    found_key.objectid,
2367                                                    found_key.offset);
2368                         if (ret == -ENOSPC)
2369                                 failed++;
2370                         else if (ret)
2371                                 BUG();
2372                 }
2373
2374                 if (found_key.offset == 0)
2375                         break;
2376                 key.offset = found_key.offset - 1;
2377         }
2378         ret = 0;
2379         if (failed && !retried) {
2380                 failed = 0;
2381                 retried = true;
2382                 goto again;
2383         } else if (failed && retried) {
2384                 WARN_ON(1);
2385                 ret = -ENOSPC;
2386         }
2387 error:
2388         btrfs_free_path(path);
2389         return ret;
2390 }
2391
2392 static int insert_balance_item(struct btrfs_root *root,
2393                                struct btrfs_balance_control *bctl)
2394 {
2395         struct btrfs_trans_handle *trans;
2396         struct btrfs_balance_item *item;
2397         struct btrfs_disk_balance_args disk_bargs;
2398         struct btrfs_path *path;
2399         struct extent_buffer *leaf;
2400         struct btrfs_key key;
2401         int ret, err;
2402
2403         path = btrfs_alloc_path();
2404         if (!path)
2405                 return -ENOMEM;
2406
2407         trans = btrfs_start_transaction(root, 0);
2408         if (IS_ERR(trans)) {
2409                 btrfs_free_path(path);
2410                 return PTR_ERR(trans);
2411         }
2412
2413         key.objectid = BTRFS_BALANCE_OBJECTID;
2414         key.type = BTRFS_BALANCE_ITEM_KEY;
2415         key.offset = 0;
2416
2417         ret = btrfs_insert_empty_item(trans, root, path, &key,
2418                                       sizeof(*item));
2419         if (ret)
2420                 goto out;
2421
2422         leaf = path->nodes[0];
2423         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
2424
2425         memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
2426
2427         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
2428         btrfs_set_balance_data(leaf, item, &disk_bargs);
2429         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
2430         btrfs_set_balance_meta(leaf, item, &disk_bargs);
2431         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
2432         btrfs_set_balance_sys(leaf, item, &disk_bargs);
2433
2434         btrfs_set_balance_flags(leaf, item, bctl->flags);
2435
2436         btrfs_mark_buffer_dirty(leaf);
2437 out:
2438         btrfs_free_path(path);
2439         err = btrfs_commit_transaction(trans, root);
2440         if (err && !ret)
2441                 ret = err;
2442         return ret;
2443 }
2444
2445 static int del_balance_item(struct btrfs_root *root)
2446 {
2447         struct btrfs_trans_handle *trans;
2448         struct btrfs_path *path;
2449         struct btrfs_key key;
2450         int ret, err;
2451
2452         path = btrfs_alloc_path();
2453         if (!path)
2454                 return -ENOMEM;
2455
2456         trans = btrfs_start_transaction(root, 0);
2457         if (IS_ERR(trans)) {
2458                 btrfs_free_path(path);
2459                 return PTR_ERR(trans);
2460         }
2461
2462         key.objectid = BTRFS_BALANCE_OBJECTID;
2463         key.type = BTRFS_BALANCE_ITEM_KEY;
2464         key.offset = 0;
2465
2466         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2467         if (ret < 0)
2468                 goto out;
2469         if (ret > 0) {
2470                 ret = -ENOENT;
2471                 goto out;
2472         }
2473
2474         ret = btrfs_del_item(trans, root, path);
2475 out:
2476         btrfs_free_path(path);
2477         err = btrfs_commit_transaction(trans, root);
2478         if (err && !ret)
2479                 ret = err;
2480         return ret;
2481 }
2482
2483 /*
2484  * This is a heuristic used to reduce the number of chunks balanced on
2485  * resume after balance was interrupted.
2486  */
2487 static void update_balance_args(struct btrfs_balance_control *bctl)
2488 {
2489         /*
2490          * Turn on soft mode for chunk types that were being converted.
2491          */
2492         if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
2493                 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
2494         if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
2495                 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
2496         if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
2497                 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
2498
2499         /*
2500          * Turn on usage filter if is not already used.  The idea is
2501          * that chunks that we have already balanced should be
2502          * reasonably full.  Don't do it for chunks that are being
2503          * converted - that will keep us from relocating unconverted
2504          * (albeit full) chunks.
2505          */
2506         if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2507             !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2508                 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
2509                 bctl->data.usage = 90;
2510         }
2511         if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2512             !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2513                 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
2514                 bctl->sys.usage = 90;
2515         }
2516         if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2517             !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2518                 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
2519                 bctl->meta.usage = 90;
2520         }
2521 }
2522
2523 /*
2524  * Should be called with both balance and volume mutexes held to
2525  * serialize other volume operations (add_dev/rm_dev/resize) with
2526  * restriper.  Same goes for unset_balance_control.
2527  */
2528 static void set_balance_control(struct btrfs_balance_control *bctl)
2529 {
2530         struct btrfs_fs_info *fs_info = bctl->fs_info;
2531
2532         BUG_ON(fs_info->balance_ctl);
2533
2534         spin_lock(&fs_info->balance_lock);
2535         fs_info->balance_ctl = bctl;
2536         spin_unlock(&fs_info->balance_lock);
2537 }
2538
2539 static void unset_balance_control(struct btrfs_fs_info *fs_info)
2540 {
2541         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
2542
2543         BUG_ON(!fs_info->balance_ctl);
2544
2545         spin_lock(&fs_info->balance_lock);
2546         fs_info->balance_ctl = NULL;
2547         spin_unlock(&fs_info->balance_lock);
2548
2549         kfree(bctl);
2550 }
2551
2552 /*
2553  * Balance filters.  Return 1 if chunk should be filtered out
2554  * (should not be balanced).
2555  */
2556 static int chunk_profiles_filter(u64 chunk_type,
2557                                  struct btrfs_balance_args *bargs)
2558 {
2559         chunk_type = chunk_to_extended(chunk_type) &
2560                                 BTRFS_EXTENDED_PROFILE_MASK;
2561
2562         if (bargs->profiles & chunk_type)
2563                 return 0;
2564
2565         return 1;
2566 }
2567
2568 static int chunk_usage_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
2569                               struct btrfs_balance_args *bargs)
2570 {
2571         struct btrfs_block_group_cache *cache;
2572         u64 chunk_used, user_thresh;
2573         int ret = 1;
2574
2575         cache = btrfs_lookup_block_group(fs_info, chunk_offset);
2576         chunk_used = btrfs_block_group_used(&cache->item);
2577
2578         user_thresh = div_factor_fine(cache->key.offset, bargs->usage);
2579         if (chunk_used < user_thresh)
2580                 ret = 0;
2581
2582         btrfs_put_block_group(cache);
2583         return ret;
2584 }
2585
2586 static int chunk_devid_filter(struct extent_buffer *leaf,
2587                               struct btrfs_chunk *chunk,
2588                               struct btrfs_balance_args *bargs)
2589 {
2590         struct btrfs_stripe *stripe;
2591         int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2592         int i;
2593
2594         for (i = 0; i < num_stripes; i++) {
2595                 stripe = btrfs_stripe_nr(chunk, i);
2596                 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
2597                         return 0;
2598         }
2599
2600         return 1;
2601 }
2602
2603 /* [pstart, pend) */
2604 static int chunk_drange_filter(struct extent_buffer *leaf,
2605                                struct btrfs_chunk *chunk,
2606                                u64 chunk_offset,
2607                                struct btrfs_balance_args *bargs)
2608 {
2609         struct btrfs_stripe *stripe;
2610         int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2611         u64 stripe_offset;
2612         u64 stripe_length;
2613         int factor;
2614         int i;
2615
2616         if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
2617                 return 0;
2618
2619         if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP |
2620              BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10))
2621                 factor = 2;
2622         else
2623                 factor = 1;
2624         factor = num_stripes / factor;
2625
2626         for (i = 0; i < num_stripes; i++) {
2627                 stripe = btrfs_stripe_nr(chunk, i);
2628                 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
2629                         continue;
2630
2631                 stripe_offset = btrfs_stripe_offset(leaf, stripe);
2632                 stripe_length = btrfs_chunk_length(leaf, chunk);
2633                 do_div(stripe_length, factor);
2634
2635                 if (stripe_offset < bargs->pend &&
2636                     stripe_offset + stripe_length > bargs->pstart)
2637                         return 0;
2638         }
2639
2640         return 1;
2641 }
2642
2643 /* [vstart, vend) */
2644 static int chunk_vrange_filter(struct extent_buffer *leaf,
2645                                struct btrfs_chunk *chunk,
2646                                u64 chunk_offset,
2647                                struct btrfs_balance_args *bargs)
2648 {
2649         if (chunk_offset < bargs->vend &&
2650             chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
2651                 /* at least part of the chunk is inside this vrange */
2652                 return 0;
2653
2654         return 1;
2655 }
2656
2657 static int chunk_soft_convert_filter(u64 chunk_type,
2658                                      struct btrfs_balance_args *bargs)
2659 {
2660         if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
2661                 return 0;
2662
2663         chunk_type = chunk_to_extended(chunk_type) &
2664                                 BTRFS_EXTENDED_PROFILE_MASK;
2665
2666         if (bargs->target == chunk_type)
2667                 return 1;
2668
2669         return 0;
2670 }
2671
2672 static int should_balance_chunk(struct btrfs_root *root,
2673                                 struct extent_buffer *leaf,
2674                                 struct btrfs_chunk *chunk, u64 chunk_offset)
2675 {
2676         struct btrfs_balance_control *bctl = root->fs_info->balance_ctl;
2677         struct btrfs_balance_args *bargs = NULL;
2678         u64 chunk_type = btrfs_chunk_type(leaf, chunk);
2679
2680         /* type filter */
2681         if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
2682               (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
2683                 return 0;
2684         }
2685
2686         if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
2687                 bargs = &bctl->data;
2688         else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
2689                 bargs = &bctl->sys;
2690         else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
2691                 bargs = &bctl->meta;
2692
2693         /* profiles filter */
2694         if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
2695             chunk_profiles_filter(chunk_type, bargs)) {
2696                 return 0;
2697         }
2698
2699         /* usage filter */
2700         if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
2701             chunk_usage_filter(bctl->fs_info, chunk_offset, bargs)) {
2702                 return 0;
2703         }
2704
2705         /* devid filter */
2706         if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
2707             chunk_devid_filter(leaf, chunk, bargs)) {
2708                 return 0;
2709         }
2710
2711         /* drange filter, makes sense only with devid filter */
2712         if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
2713             chunk_drange_filter(leaf, chunk, chunk_offset, bargs)) {
2714                 return 0;
2715         }
2716
2717         /* vrange filter */
2718         if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
2719             chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
2720                 return 0;
2721         }
2722
2723         /* soft profile changing mode */
2724         if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
2725             chunk_soft_convert_filter(chunk_type, bargs)) {
2726                 return 0;
2727         }
2728
2729         return 1;
2730 }
2731
2732 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
2733 {
2734         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
2735         struct btrfs_root *chunk_root = fs_info->chunk_root;
2736         struct btrfs_root *dev_root = fs_info->dev_root;
2737         struct list_head *devices;
2738         struct btrfs_device *device;
2739         u64 old_size;
2740         u64 size_to_free;
2741         struct btrfs_chunk *chunk;
2742         struct btrfs_path *path;
2743         struct btrfs_key key;
2744         struct btrfs_key found_key;
2745         struct btrfs_trans_handle *trans;
2746         struct extent_buffer *leaf;
2747         int slot;
2748         int ret;
2749         int enospc_errors = 0;
2750         bool counting = true;
2751
2752         /* step one make some room on all the devices */
2753         devices = &fs_info->fs_devices->devices;
2754         list_for_each_entry(device, devices, dev_list) {
2755                 old_size = device->total_bytes;
2756                 size_to_free = div_factor(old_size, 1);
2757                 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
2758                 if (!device->writeable ||
2759                     device->total_bytes - device->bytes_used > size_to_free ||
2760                     device->is_tgtdev_for_dev_replace)
2761                         continue;
2762
2763                 ret = btrfs_shrink_device(device, old_size - size_to_free);
2764                 if (ret == -ENOSPC)
2765                         break;
2766                 BUG_ON(ret);
2767
2768                 trans = btrfs_start_transaction(dev_root, 0);
2769                 BUG_ON(IS_ERR(trans));
2770
2771                 ret = btrfs_grow_device(trans, device, old_size);
2772                 BUG_ON(ret);
2773
2774                 btrfs_end_transaction(trans, dev_root);
2775         }
2776
2777         /* step two, relocate all the chunks */
2778         path = btrfs_alloc_path();
2779         if (!path) {
2780                 ret = -ENOMEM;
2781                 goto error;
2782         }
2783
2784         /* zero out stat counters */
2785         spin_lock(&fs_info->balance_lock);
2786         memset(&bctl->stat, 0, sizeof(bctl->stat));
2787         spin_unlock(&fs_info->balance_lock);
2788 again:
2789         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2790         key.offset = (u64)-1;
2791         key.type = BTRFS_CHUNK_ITEM_KEY;
2792
2793         while (1) {
2794                 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
2795                     atomic_read(&fs_info->balance_cancel_req)) {
2796                         ret = -ECANCELED;
2797                         goto error;
2798                 }
2799
2800                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2801                 if (ret < 0)
2802                         goto error;
2803
2804                 /*
2805                  * this shouldn't happen, it means the last relocate
2806                  * failed
2807                  */
2808                 if (ret == 0)
2809                         BUG(); /* FIXME break ? */
2810
2811                 ret = btrfs_previous_item(chunk_root, path, 0,
2812                                           BTRFS_CHUNK_ITEM_KEY);
2813                 if (ret) {
2814                         ret = 0;
2815                         break;
2816                 }
2817
2818                 leaf = path->nodes[0];
2819                 slot = path->slots[0];
2820                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2821
2822                 if (found_key.objectid != key.objectid)
2823                         break;
2824
2825                 /* chunk zero is special */
2826                 if (found_key.offset == 0)
2827                         break;
2828
2829                 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
2830
2831                 if (!counting) {
2832                         spin_lock(&fs_info->balance_lock);
2833                         bctl->stat.considered++;
2834                         spin_unlock(&fs_info->balance_lock);
2835                 }
2836
2837                 ret = should_balance_chunk(chunk_root, leaf, chunk,
2838                                            found_key.offset);
2839                 btrfs_release_path(path);
2840                 if (!ret)
2841                         goto loop;
2842
2843                 if (counting) {
2844                         spin_lock(&fs_info->balance_lock);
2845                         bctl->stat.expected++;
2846                         spin_unlock(&fs_info->balance_lock);
2847                         goto loop;
2848                 }
2849
2850                 ret = btrfs_relocate_chunk(chunk_root,
2851                                            chunk_root->root_key.objectid,
2852                                            found_key.objectid,
2853                                            found_key.offset);
2854                 if (ret && ret != -ENOSPC)
2855                         goto error;
2856                 if (ret == -ENOSPC) {
2857                         enospc_errors++;
2858                 } else {
2859                         spin_lock(&fs_info->balance_lock);
2860                         bctl->stat.completed++;
2861                         spin_unlock(&fs_info->balance_lock);
2862                 }
2863 loop:
2864                 key.offset = found_key.offset - 1;
2865         }
2866
2867         if (counting) {
2868                 btrfs_release_path(path);
2869                 counting = false;
2870                 goto again;
2871         }
2872 error:
2873         btrfs_free_path(path);
2874         if (enospc_errors) {
2875                 printk(KERN_INFO "btrfs: %d enospc errors during balance\n",
2876                        enospc_errors);
2877                 if (!ret)
2878                         ret = -ENOSPC;
2879         }
2880
2881         return ret;
2882 }
2883
2884 /**
2885  * alloc_profile_is_valid - see if a given profile is valid and reduced
2886  * @flags: profile to validate
2887  * @extended: if true @flags is treated as an extended profile
2888  */
2889 static int alloc_profile_is_valid(u64 flags, int extended)
2890 {
2891         u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
2892                                BTRFS_BLOCK_GROUP_PROFILE_MASK);
2893
2894         flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
2895
2896         /* 1) check that all other bits are zeroed */
2897         if (flags & ~mask)
2898                 return 0;
2899
2900         /* 2) see if profile is reduced */
2901         if (flags == 0)
2902                 return !extended; /* "0" is valid for usual profiles */
2903
2904         /* true if exactly one bit set */
2905         return (flags & (flags - 1)) == 0;
2906 }
2907
2908 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
2909 {
2910         /* cancel requested || normal exit path */
2911         return atomic_read(&fs_info->balance_cancel_req) ||
2912                 (atomic_read(&fs_info->balance_pause_req) == 0 &&
2913                  atomic_read(&fs_info->balance_cancel_req) == 0);
2914 }
2915
2916 static void __cancel_balance(struct btrfs_fs_info *fs_info)
2917 {
2918         int ret;
2919
2920         unset_balance_control(fs_info);
2921         ret = del_balance_item(fs_info->tree_root);
2922         BUG_ON(ret);
2923 }
2924
2925 void update_ioctl_balance_args(struct btrfs_fs_info *fs_info, int lock,
2926                                struct btrfs_ioctl_balance_args *bargs);
2927
2928 /*
2929  * Should be called with both balance and volume mutexes held
2930  */
2931 int btrfs_balance(struct btrfs_balance_control *bctl,
2932                   struct btrfs_ioctl_balance_args *bargs)
2933 {
2934         struct btrfs_fs_info *fs_info = bctl->fs_info;
2935         u64 allowed;
2936         int mixed = 0;
2937         int ret;
2938
2939         if (btrfs_fs_closing(fs_info) ||
2940             atomic_read(&fs_info->balance_pause_req) ||
2941             atomic_read(&fs_info->balance_cancel_req)) {
2942                 ret = -EINVAL;
2943                 goto out;
2944         }
2945
2946         allowed = btrfs_super_incompat_flags(fs_info->super_copy);
2947         if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
2948                 mixed = 1;
2949
2950         /*
2951          * In case of mixed groups both data and meta should be picked,
2952          * and identical options should be given for both of them.
2953          */
2954         allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
2955         if (mixed && (bctl->flags & allowed)) {
2956                 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
2957                     !(bctl->flags & BTRFS_BALANCE_METADATA) ||
2958                     memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
2959                         printk(KERN_ERR "btrfs: with mixed groups data and "
2960                                "metadata balance options must be the same\n");
2961                         ret = -EINVAL;
2962                         goto out;
2963                 }
2964         }
2965
2966         allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE;
2967         if (fs_info->fs_devices->num_devices == 1)
2968                 allowed |= BTRFS_BLOCK_GROUP_DUP;
2969         else if (fs_info->fs_devices->num_devices < 4)
2970                 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1);
2971         else
2972                 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
2973                                 BTRFS_BLOCK_GROUP_RAID10);
2974
2975         if ((bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
2976             (!alloc_profile_is_valid(bctl->data.target, 1) ||
2977              (bctl->data.target & ~allowed))) {
2978                 printk(KERN_ERR "btrfs: unable to start balance with target "
2979                        "data profile %llu\n",
2980                        (unsigned long long)bctl->data.target);
2981                 ret = -EINVAL;
2982                 goto out;
2983         }
2984         if ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
2985             (!alloc_profile_is_valid(bctl->meta.target, 1) ||
2986              (bctl->meta.target & ~allowed))) {
2987                 printk(KERN_ERR "btrfs: unable to start balance with target "
2988                        "metadata profile %llu\n",
2989                        (unsigned long long)bctl->meta.target);
2990                 ret = -EINVAL;
2991                 goto out;
2992         }
2993         if ((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
2994             (!alloc_profile_is_valid(bctl->sys.target, 1) ||
2995              (bctl->sys.target & ~allowed))) {
2996                 printk(KERN_ERR "btrfs: unable to start balance with target "
2997                        "system profile %llu\n",
2998                        (unsigned long long)bctl->sys.target);
2999                 ret = -EINVAL;
3000                 goto out;
3001         }
3002
3003         /* allow dup'ed data chunks only in mixed mode */
3004         if (!mixed && (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3005             (bctl->data.target & BTRFS_BLOCK_GROUP_DUP)) {
3006                 printk(KERN_ERR "btrfs: dup for data is not allowed\n");
3007                 ret = -EINVAL;
3008                 goto out;
3009         }
3010
3011         /* allow to reduce meta or sys integrity only if force set */
3012         allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3013                         BTRFS_BLOCK_GROUP_RAID10;
3014         if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3015              (fs_info->avail_system_alloc_bits & allowed) &&
3016              !(bctl->sys.target & allowed)) ||
3017             ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3018              (fs_info->avail_metadata_alloc_bits & allowed) &&
3019              !(bctl->meta.target & allowed))) {
3020                 if (bctl->flags & BTRFS_BALANCE_FORCE) {
3021                         printk(KERN_INFO "btrfs: force reducing metadata "
3022                                "integrity\n");
3023                 } else {
3024                         printk(KERN_ERR "btrfs: balance will reduce metadata "
3025                                "integrity, use force if you want this\n");
3026                         ret = -EINVAL;
3027                         goto out;
3028                 }
3029         }
3030
3031         if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3032                 int num_tolerated_disk_barrier_failures;
3033                 u64 target = bctl->sys.target;
3034
3035                 num_tolerated_disk_barrier_failures =
3036                         btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3037                 if (num_tolerated_disk_barrier_failures > 0 &&
3038                     (target &
3039                      (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID0 |
3040                       BTRFS_AVAIL_ALLOC_BIT_SINGLE)))
3041                         num_tolerated_disk_barrier_failures = 0;
3042                 else if (num_tolerated_disk_barrier_failures > 1 &&
3043                          (target &
3044                           (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)))
3045                         num_tolerated_disk_barrier_failures = 1;
3046
3047                 fs_info->num_tolerated_disk_barrier_failures =
3048                         num_tolerated_disk_barrier_failures;
3049         }
3050
3051         ret = insert_balance_item(fs_info->tree_root, bctl);
3052         if (ret && ret != -EEXIST)
3053                 goto out;
3054
3055         if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
3056                 BUG_ON(ret == -EEXIST);
3057                 set_balance_control(bctl);
3058         } else {
3059                 BUG_ON(ret != -EEXIST);
3060                 spin_lock(&fs_info->balance_lock);
3061                 update_balance_args(bctl);
3062                 spin_unlock(&fs_info->balance_lock);
3063         }
3064
3065         atomic_inc(&fs_info->balance_running);
3066         mutex_unlock(&fs_info->balance_mutex);
3067
3068         ret = __btrfs_balance(fs_info);
3069
3070         mutex_lock(&fs_info->balance_mutex);
3071         atomic_dec(&fs_info->balance_running);
3072
3073         if (bargs) {
3074                 memset(bargs, 0, sizeof(*bargs));
3075                 update_ioctl_balance_args(fs_info, 0, bargs);
3076         }
3077
3078         if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
3079             balance_need_close(fs_info)) {
3080                 __cancel_balance(fs_info);
3081         }
3082
3083         if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3084                 fs_info->num_tolerated_disk_barrier_failures =
3085                         btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3086         }
3087
3088         wake_up(&fs_info->balance_wait_q);
3089
3090         return ret;
3091 out:
3092         if (bctl->flags & BTRFS_BALANCE_RESUME)
3093                 __cancel_balance(fs_info);
3094         else
3095                 kfree(bctl);
3096         return ret;
3097 }
3098
3099 static int balance_kthread(void *data)
3100 {
3101         struct btrfs_fs_info *fs_info = data;
3102         int ret = 0;
3103
3104         mutex_lock(&fs_info->volume_mutex);
3105         mutex_lock(&fs_info->balance_mutex);
3106
3107         if (fs_info->balance_ctl) {
3108                 printk(KERN_INFO "btrfs: continuing balance\n");
3109                 ret = btrfs_balance(fs_info->balance_ctl, NULL);
3110         }
3111
3112         atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3113         mutex_unlock(&fs_info->balance_mutex);
3114         mutex_unlock(&fs_info->volume_mutex);
3115
3116         return ret;
3117 }
3118
3119 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
3120 {
3121         struct task_struct *tsk;
3122
3123         spin_lock(&fs_info->balance_lock);
3124         if (!fs_info->balance_ctl) {
3125                 spin_unlock(&fs_info->balance_lock);
3126                 return 0;
3127         }
3128         spin_unlock(&fs_info->balance_lock);
3129
3130         if (btrfs_test_opt(fs_info->tree_root, SKIP_BALANCE)) {
3131                 printk(KERN_INFO "btrfs: force skipping balance\n");
3132                 return 0;
3133         }
3134
3135         WARN_ON(atomic_xchg(&fs_info->mutually_exclusive_operation_running, 1));
3136         tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
3137         if (IS_ERR(tsk))
3138                 return PTR_ERR(tsk);
3139
3140         return 0;
3141 }
3142
3143 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
3144 {
3145         struct btrfs_balance_control *bctl;
3146         struct btrfs_balance_item *item;
3147         struct btrfs_disk_balance_args disk_bargs;
3148         struct btrfs_path *path;
3149         struct extent_buffer *leaf;
3150         struct btrfs_key key;
3151         int ret;
3152
3153         path = btrfs_alloc_path();
3154         if (!path)
3155                 return -ENOMEM;
3156
3157         key.objectid = BTRFS_BALANCE_OBJECTID;
3158         key.type = BTRFS_BALANCE_ITEM_KEY;
3159         key.offset = 0;
3160
3161         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
3162         if (ret < 0)
3163                 goto out;
3164         if (ret > 0) { /* ret = -ENOENT; */
3165                 ret = 0;
3166                 goto out;
3167         }
3168
3169         bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
3170         if (!bctl) {
3171                 ret = -ENOMEM;
3172                 goto out;
3173         }
3174
3175         leaf = path->nodes[0];
3176         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3177
3178         bctl->fs_info = fs_info;
3179         bctl->flags = btrfs_balance_flags(leaf, item);
3180         bctl->flags |= BTRFS_BALANCE_RESUME;
3181
3182         btrfs_balance_data(leaf, item, &disk_bargs);
3183         btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
3184         btrfs_balance_meta(leaf, item, &disk_bargs);
3185         btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
3186         btrfs_balance_sys(leaf, item, &disk_bargs);
3187         btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
3188
3189         mutex_lock(&fs_info->volume_mutex);
3190         mutex_lock(&fs_info->balance_mutex);
3191
3192         set_balance_control(bctl);
3193
3194         mutex_unlock(&fs_info->balance_mutex);
3195         mutex_unlock(&fs_info->volume_mutex);
3196 out:
3197         btrfs_free_path(path);
3198         return ret;
3199 }
3200
3201 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
3202 {
3203         int ret = 0;
3204
3205         mutex_lock(&fs_info->balance_mutex);
3206         if (!fs_info->balance_ctl) {
3207                 mutex_unlock(&fs_info->balance_mutex);
3208                 return -ENOTCONN;
3209         }
3210
3211         if (atomic_read(&fs_info->balance_running)) {
3212                 atomic_inc(&fs_info->balance_pause_req);
3213                 mutex_unlock(&fs_info->balance_mutex);
3214
3215                 wait_event(fs_info->balance_wait_q,
3216                            atomic_read(&fs_info->balance_running) == 0);
3217
3218                 mutex_lock(&fs_info->balance_mutex);
3219                 /* we are good with balance_ctl ripped off from under us */
3220                 BUG_ON(atomic_read(&fs_info->balance_running));
3221                 atomic_dec(&fs_info->balance_pause_req);
3222         } else {
3223                 ret = -ENOTCONN;
3224         }
3225
3226         mutex_unlock(&fs_info->balance_mutex);
3227         return ret;
3228 }
3229
3230 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
3231 {
3232         mutex_lock(&fs_info->balance_mutex);
3233         if (!fs_info->balance_ctl) {
3234                 mutex_unlock(&fs_info->balance_mutex);
3235                 return -ENOTCONN;
3236         }
3237
3238         atomic_inc(&fs_info->balance_cancel_req);
3239         /*
3240          * if we are running just wait and return, balance item is
3241          * deleted in btrfs_balance in this case
3242          */
3243         if (atomic_read(&fs_info->balance_running)) {
3244                 mutex_unlock(&fs_info->balance_mutex);
3245                 wait_event(fs_info->balance_wait_q,
3246                            atomic_read(&fs_info->balance_running) == 0);
3247                 mutex_lock(&fs_info->balance_mutex);
3248         } else {
3249                 /* __cancel_balance needs volume_mutex */
3250                 mutex_unlock(&fs_info->balance_mutex);
3251                 mutex_lock(&fs_info->volume_mutex);
3252                 mutex_lock(&fs_info->balance_mutex);
3253
3254                 if (fs_info->balance_ctl)
3255                         __cancel_balance(fs_info);
3256
3257                 mutex_unlock(&fs_info->volume_mutex);
3258         }
3259
3260         BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running));
3261         atomic_dec(&fs_info->balance_cancel_req);
3262         mutex_unlock(&fs_info->balance_mutex);
3263         return 0;
3264 }
3265
3266 /*
3267  * shrinking a device means finding all of the device extents past
3268  * the new size, and then following the back refs to the chunks.
3269  * The chunk relocation code actually frees the device extent
3270  */
3271 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
3272 {
3273         struct btrfs_trans_handle *trans;
3274         struct btrfs_root *root = device->dev_root;
3275         struct btrfs_dev_extent *dev_extent = NULL;
3276         struct btrfs_path *path;
3277         u64 length;
3278         u64 chunk_tree;
3279         u64 chunk_objectid;
3280         u64 chunk_offset;
3281         int ret;
3282         int slot;
3283         int failed = 0;
3284         bool retried = false;
3285         struct extent_buffer *l;
3286         struct btrfs_key key;
3287         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
3288         u64 old_total = btrfs_super_total_bytes(super_copy);
3289         u64 old_size = device->total_bytes;
3290         u64 diff = device->total_bytes - new_size;
3291
3292         if (device->is_tgtdev_for_dev_replace)
3293                 return -EINVAL;
3294
3295         path = btrfs_alloc_path();
3296         if (!path)
3297                 return -ENOMEM;
3298
3299         path->reada = 2;
3300
3301         lock_chunks(root);
3302
3303         device->total_bytes = new_size;
3304         if (device->writeable) {
3305                 device->fs_devices->total_rw_bytes -= diff;
3306                 spin_lock(&root->fs_info->free_chunk_lock);
3307                 root->fs_info->free_chunk_space -= diff;
3308                 spin_unlock(&root->fs_info->free_chunk_lock);
3309         }
3310         unlock_chunks(root);
3311
3312 again:
3313         key.objectid = device->devid;
3314         key.offset = (u64)-1;
3315         key.type = BTRFS_DEV_EXTENT_KEY;
3316
3317         do {
3318                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3319                 if (ret < 0)
3320                         goto done;
3321
3322                 ret = btrfs_previous_item(root, path, 0, key.type);
3323                 if (ret < 0)
3324                         goto done;
3325                 if (ret) {
3326                         ret = 0;
3327                         btrfs_release_path(path);
3328                         break;
3329                 }
3330
3331                 l = path->nodes[0];
3332                 slot = path->slots[0];
3333                 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
3334
3335                 if (key.objectid != device->devid) {
3336                         btrfs_release_path(path);
3337                         break;
3338                 }
3339
3340                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
3341                 length = btrfs_dev_extent_length(l, dev_extent);
3342
3343                 if (key.offset + length <= new_size) {
3344                         btrfs_release_path(path);
3345                         break;
3346                 }
3347
3348                 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
3349                 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
3350                 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
3351                 btrfs_release_path(path);
3352
3353                 ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
3354                                            chunk_offset);
3355                 if (ret && ret != -ENOSPC)
3356                         goto done;
3357                 if (ret == -ENOSPC)
3358                         failed++;
3359         } while (key.offset-- > 0);
3360
3361         if (failed && !retried) {
3362                 failed = 0;
3363                 retried = true;
3364                 goto again;
3365         } else if (failed && retried) {
3366                 ret = -ENOSPC;
3367                 lock_chunks(root);
3368
3369                 device->total_bytes = old_size;
3370                 if (device->writeable)
3371                         device->fs_devices->total_rw_bytes += diff;
3372                 spin_lock(&root->fs_info->free_chunk_lock);
3373                 root->fs_info->free_chunk_space += diff;
3374                 spin_unlock(&root->fs_info->free_chunk_lock);
3375                 unlock_chunks(root);
3376                 goto done;
3377         }
3378
3379         /* Shrinking succeeded, else we would be at "done". */
3380         trans = btrfs_start_transaction(root, 0);
3381         if (IS_ERR(trans)) {
3382                 ret = PTR_ERR(trans);
3383                 goto done;
3384         }
3385
3386         lock_chunks(root);
3387
3388         device->disk_total_bytes = new_size;
3389         /* Now btrfs_update_device() will change the on-disk size. */
3390         ret = btrfs_update_device(trans, device);
3391         if (ret) {
3392                 unlock_chunks(root);
3393                 btrfs_end_transaction(trans, root);
3394                 goto done;
3395         }
3396         WARN_ON(diff > old_total);
3397         btrfs_set_super_total_bytes(super_copy, old_total - diff);
3398         unlock_chunks(root);
3399         btrfs_end_transaction(trans, root);
3400 done:
3401         btrfs_free_path(path);
3402         return ret;
3403 }
3404
3405 static int btrfs_add_system_chunk(struct btrfs_root *root,
3406                            struct btrfs_key *key,
3407                            struct btrfs_chunk *chunk, int item_size)
3408 {
3409         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
3410         struct btrfs_disk_key disk_key;
3411         u32 array_size;
3412         u8 *ptr;
3413
3414         array_size = btrfs_super_sys_array_size(super_copy);
3415         if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
3416                 return -EFBIG;
3417
3418         ptr = super_copy->sys_chunk_array + array_size;
3419         btrfs_cpu_key_to_disk(&disk_key, key);
3420         memcpy(ptr, &disk_key, sizeof(disk_key));
3421         ptr += sizeof(disk_key);
3422         memcpy(ptr, chunk, item_size);
3423         item_size += sizeof(disk_key);
3424         btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
3425         return 0;
3426 }
3427
3428 /*
3429  * sort the devices in descending order by max_avail, total_avail
3430  */
3431 static int btrfs_cmp_device_info(const void *a, const void *b)
3432 {
3433         const struct btrfs_device_info *di_a = a;
3434         const struct btrfs_device_info *di_b = b;
3435
3436         if (di_a->max_avail > di_b->max_avail)
3437                 return -1;
3438         if (di_a->max_avail < di_b->max_avail)
3439                 return 1;
3440         if (di_a->total_avail > di_b->total_avail)
3441                 return -1;
3442         if (di_a->total_avail < di_b->total_avail)
3443                 return 1;
3444         return 0;
3445 }
3446
3447 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
3448                                struct btrfs_root *extent_root,
3449                                struct map_lookup **map_ret,
3450                                u64 *num_bytes_out, u64 *stripe_size_out,
3451                                u64 start, u64 type)
3452 {
3453         struct btrfs_fs_info *info = extent_root->fs_info;
3454         struct btrfs_fs_devices *fs_devices = info->fs_devices;
3455         struct list_head *cur;
3456         struct map_lookup *map = NULL;
3457         struct extent_map_tree *em_tree;
3458         struct extent_map *em;
3459         struct btrfs_device_info *devices_info = NULL;
3460         u64 total_avail;
3461         int num_stripes;        /* total number of stripes to allocate */
3462         int sub_stripes;        /* sub_stripes info for map */
3463         int dev_stripes;        /* stripes per dev */
3464         int devs_max;           /* max devs to use */
3465         int devs_min;           /* min devs needed */
3466         int devs_increment;     /* ndevs has to be a multiple of this */
3467         int ncopies;            /* how many copies to data has */
3468         int ret;
3469         u64 max_stripe_size;
3470         u64 max_chunk_size;
3471         u64 stripe_size;
3472         u64 num_bytes;
3473         int ndevs;
3474         int i;
3475         int j;
3476
3477         BUG_ON(!alloc_profile_is_valid(type, 0));
3478
3479         if (list_empty(&fs_devices->alloc_list))
3480                 return -ENOSPC;
3481
3482         sub_stripes = 1;
3483         dev_stripes = 1;
3484         devs_increment = 1;
3485         ncopies = 1;
3486         devs_max = 0;   /* 0 == as many as possible */
3487         devs_min = 1;
3488
3489         /*
3490          * define the properties of each RAID type.
3491          * FIXME: move this to a global table and use it in all RAID
3492          * calculation code
3493          */
3494         if (type & (BTRFS_BLOCK_GROUP_DUP)) {
3495                 dev_stripes = 2;
3496                 ncopies = 2;
3497                 devs_max = 1;
3498         } else if (type & (BTRFS_BLOCK_GROUP_RAID0)) {
3499                 devs_min = 2;
3500         } else if (type & (BTRFS_BLOCK_GROUP_RAID1)) {
3501                 devs_increment = 2;
3502                 ncopies = 2;
3503                 devs_max = 2;
3504                 devs_min = 2;
3505         } else if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
3506                 sub_stripes = 2;
3507                 devs_increment = 2;
3508                 ncopies = 2;
3509                 devs_min = 4;
3510         } else {
3511                 devs_max = 1;
3512         }
3513
3514         if (type & BTRFS_BLOCK_GROUP_DATA) {
3515                 max_stripe_size = 1024 * 1024 * 1024;
3516                 max_chunk_size = 10 * max_stripe_size;
3517         } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
3518                 /* for larger filesystems, use larger metadata chunks */
3519                 if (fs_devices->total_rw_bytes > 50ULL * 1024 * 1024 * 1024)
3520                         max_stripe_size = 1024 * 1024 * 1024;
3521                 else
3522                         max_stripe_size = 256 * 1024 * 1024;
3523                 max_chunk_size = max_stripe_size;
3524         } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
3525                 max_stripe_size = 32 * 1024 * 1024;
3526                 max_chunk_size = 2 * max_stripe_size;
3527         } else {
3528                 printk(KERN_ERR "btrfs: invalid chunk type 0x%llx requested\n",
3529                        type);
3530                 BUG_ON(1);
3531         }
3532
3533         /* we don't want a chunk larger than 10% of writeable space */
3534         max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
3535                              max_chunk_size);
3536
3537         devices_info = kzalloc(sizeof(*devices_info) * fs_devices->rw_devices,
3538                                GFP_NOFS);
3539         if (!devices_info)
3540                 return -ENOMEM;
3541
3542         cur = fs_devices->alloc_list.next;
3543
3544         /*
3545          * in the first pass through the devices list, we gather information
3546          * about the available holes on each device.
3547          */
3548         ndevs = 0;
3549         while (cur != &fs_devices->alloc_list) {
3550                 struct btrfs_device *device;
3551                 u64 max_avail;
3552                 u64 dev_offset;
3553
3554                 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
3555
3556                 cur = cur->next;
3557
3558                 if (!device->writeable) {
3559                         WARN(1, KERN_ERR
3560                                "btrfs: read-only device in alloc_list\n");
3561                         continue;
3562                 }
3563
3564                 if (!device->in_fs_metadata ||
3565                     device->is_tgtdev_for_dev_replace)
3566                         continue;
3567
3568                 if (device->total_bytes > device->bytes_used)
3569                         total_avail = device->total_bytes - device->bytes_used;
3570                 else
3571                         total_avail = 0;
3572
3573                 /* If there is no space on this device, skip it. */
3574                 if (total_avail == 0)
3575                         continue;
3576
3577                 ret = find_free_dev_extent(device,
3578                                            max_stripe_size * dev_stripes,
3579                                            &dev_offset, &max_avail);
3580                 if (ret && ret != -ENOSPC)
3581                         goto error;
3582
3583                 if (ret == 0)
3584                         max_avail = max_stripe_size * dev_stripes;
3585
3586                 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes)
3587                         continue;
3588
3589                 devices_info[ndevs].dev_offset = dev_offset;
3590                 devices_info[ndevs].max_avail = max_avail;
3591                 devices_info[ndevs].total_avail = total_avail;
3592                 devices_info[ndevs].dev = device;
3593                 ++ndevs;
3594         }
3595
3596         /*
3597          * now sort the devices by hole size / available space
3598          */
3599         sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
3600              btrfs_cmp_device_info, NULL);
3601
3602         /* round down to number of usable stripes */
3603         ndevs -= ndevs % devs_increment;
3604
3605         if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) {
3606                 ret = -ENOSPC;
3607                 goto error;
3608         }
3609
3610         if (devs_max && ndevs > devs_max)
3611                 ndevs = devs_max;
3612         /*
3613          * the primary goal is to maximize the number of stripes, so use as many
3614          * devices as possible, even if the stripes are not maximum sized.
3615          */
3616         stripe_size = devices_info[ndevs-1].max_avail;
3617         num_stripes = ndevs * dev_stripes;
3618
3619         if (stripe_size * ndevs > max_chunk_size * ncopies) {
3620                 stripe_size = max_chunk_size * ncopies;
3621                 do_div(stripe_size, ndevs);
3622         }
3623
3624         do_div(stripe_size, dev_stripes);
3625
3626         /* align to BTRFS_STRIPE_LEN */
3627         do_div(stripe_size, BTRFS_STRIPE_LEN);
3628         stripe_size *= BTRFS_STRIPE_LEN;
3629
3630         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
3631         if (!map) {
3632                 ret = -ENOMEM;
3633                 goto error;
3634         }
3635         map->num_stripes = num_stripes;
3636
3637         for (i = 0; i < ndevs; ++i) {
3638                 for (j = 0; j < dev_stripes; ++j) {
3639                         int s = i * dev_stripes + j;
3640                         map->stripes[s].dev = devices_info[i].dev;
3641                         map->stripes[s].physical = devices_info[i].dev_offset +
3642                                                    j * stripe_size;
3643                 }
3644         }
3645         map->sector_size = extent_root->sectorsize;
3646         map->stripe_len = BTRFS_STRIPE_LEN;
3647         map->io_align = BTRFS_STRIPE_LEN;
3648         map->io_width = BTRFS_STRIPE_LEN;
3649         map->type = type;
3650         map->sub_stripes = sub_stripes;
3651
3652         *map_ret = map;
3653         num_bytes = stripe_size * (num_stripes / ncopies);
3654
3655         *stripe_size_out = stripe_size;
3656         *num_bytes_out = num_bytes;
3657
3658         trace_btrfs_chunk_alloc(info->chunk_root, map, start, num_bytes);
3659
3660         em = alloc_extent_map();
3661         if (!em) {
3662                 ret = -ENOMEM;
3663                 goto error;
3664         }
3665         em->bdev = (struct block_device *)map;
3666         em->start = start;
3667         em->len = num_bytes;
3668         em->block_start = 0;
3669         em->block_len = em->len;
3670
3671         em_tree = &extent_root->fs_info->mapping_tree.map_tree;
3672         write_lock(&em_tree->lock);
3673         ret = add_extent_mapping(em_tree, em);
3674         write_unlock(&em_tree->lock);
3675         free_extent_map(em);
3676         if (ret)
3677                 goto error;
3678
3679         ret = btrfs_make_block_group(trans, extent_root, 0, type,
3680                                      BTRFS_FIRST_CHUNK_TREE_OBJECTID,
3681                                      start, num_bytes);
3682         if (ret)
3683                 goto error;
3684
3685         for (i = 0; i < map->num_stripes; ++i) {
3686                 struct btrfs_device *device;
3687                 u64 dev_offset;
3688
3689                 device = map->stripes[i].dev;
3690                 dev_offset = map->stripes[i].physical;
3691
3692                 ret = btrfs_alloc_dev_extent(trans, device,
3693                                 info->chunk_root->root_key.objectid,
3694                                 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
3695                                 start, dev_offset, stripe_size);
3696                 if (ret) {
3697                         btrfs_abort_transaction(trans, extent_root, ret);
3698                         goto error;
3699                 }
3700         }
3701
3702         kfree(devices_info);
3703         return 0;
3704
3705 error:
3706         kfree(map);
3707         kfree(devices_info);
3708         return ret;
3709 }
3710
3711 static int __finish_chunk_alloc(struct btrfs_trans_handle *trans,
3712                                 struct btrfs_root *extent_root,
3713                                 struct map_lookup *map, u64 chunk_offset,
3714                                 u64 chunk_size, u64 stripe_size)
3715 {
3716         u64 dev_offset;
3717         struct btrfs_key key;
3718         struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
3719         struct btrfs_device *device;
3720         struct btrfs_chunk *chunk;
3721         struct btrfs_stripe *stripe;
3722         size_t item_size = btrfs_chunk_item_size(map->num_stripes);
3723         int index = 0;
3724         int ret;
3725
3726         chunk = kzalloc(item_size, GFP_NOFS);
3727         if (!chunk)
3728                 return -ENOMEM;
3729
3730         index = 0;
3731         while (index < map->num_stripes) {
3732                 device = map->stripes[index].dev;
3733                 device->bytes_used += stripe_size;
3734                 ret = btrfs_update_device(trans, device);
3735                 if (ret)
3736                         goto out_free;
3737                 index++;
3738         }
3739
3740         spin_lock(&extent_root->fs_info->free_chunk_lock);
3741         extent_root->fs_info->free_chunk_space -= (stripe_size *
3742                                                    map->num_stripes);
3743         spin_unlock(&extent_root->fs_info->free_chunk_lock);
3744
3745         index = 0;
3746         stripe = &chunk->stripe;
3747         while (index < map->num_stripes) {
3748                 device = map->stripes[index].dev;
3749                 dev_offset = map->stripes[index].physical;
3750
3751                 btrfs_set_stack_stripe_devid(stripe, device->devid);
3752                 btrfs_set_stack_stripe_offset(stripe, dev_offset);
3753                 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
3754                 stripe++;
3755                 index++;
3756         }
3757
3758         btrfs_set_stack_chunk_length(chunk, chunk_size);
3759         btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
3760         btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
3761         btrfs_set_stack_chunk_type(chunk, map->type);
3762         btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
3763         btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
3764         btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
3765         btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
3766         btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
3767
3768         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3769         key.type = BTRFS_CHUNK_ITEM_KEY;
3770         key.offset = chunk_offset;
3771
3772         ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
3773
3774         if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
3775                 /*
3776                  * TODO: Cleanup of inserted chunk root in case of
3777                  * failure.
3778                  */
3779                 ret = btrfs_add_system_chunk(chunk_root, &key, chunk,
3780                                              item_size);
3781         }
3782
3783 out_free:
3784         kfree(chunk);
3785         return ret;
3786 }
3787
3788 /*
3789  * Chunk allocation falls into two parts. The first part does works
3790  * that make the new allocated chunk useable, but not do any operation
3791  * that modifies the chunk tree. The second part does the works that
3792  * require modifying the chunk tree. This division is important for the
3793  * bootstrap process of adding storage to a seed btrfs.
3794  */
3795 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
3796                       struct btrfs_root *extent_root, u64 type)
3797 {
3798         u64 chunk_offset;
3799         u64 chunk_size;
3800         u64 stripe_size;
3801         struct map_lookup *map;
3802         struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
3803         int ret;
3804
3805         ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
3806                               &chunk_offset);
3807         if (ret)
3808                 return ret;
3809
3810         ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
3811                                   &stripe_size, chunk_offset, type);
3812         if (ret)
3813                 return ret;
3814
3815         ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
3816                                    chunk_size, stripe_size);
3817         if (ret)
3818                 return ret;
3819         return 0;
3820 }
3821
3822 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
3823                                          struct btrfs_root *root,
3824                                          struct btrfs_device *device)
3825 {
3826         u64 chunk_offset;
3827         u64 sys_chunk_offset;
3828         u64 chunk_size;
3829         u64 sys_chunk_size;
3830         u64 stripe_size;
3831         u64 sys_stripe_size;
3832         u64 alloc_profile;
3833         struct map_lookup *map;
3834         struct map_lookup *sys_map;
3835         struct btrfs_fs_info *fs_info = root->fs_info;
3836         struct btrfs_root *extent_root = fs_info->extent_root;
3837         int ret;
3838
3839         ret = find_next_chunk(fs_info->chunk_root,
3840                               BTRFS_FIRST_CHUNK_TREE_OBJECTID, &chunk_offset);
3841         if (ret)
3842                 return ret;
3843
3844         alloc_profile = BTRFS_BLOCK_GROUP_METADATA |
3845                                 fs_info->avail_metadata_alloc_bits;
3846         alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
3847
3848         ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
3849                                   &stripe_size, chunk_offset, alloc_profile);
3850         if (ret)
3851                 return ret;
3852
3853         sys_chunk_offset = chunk_offset + chunk_size;
3854
3855         alloc_profile = BTRFS_BLOCK_GROUP_SYSTEM |
3856                                 fs_info->avail_system_alloc_bits;
3857         alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
3858
3859         ret = __btrfs_alloc_chunk(trans, extent_root, &sys_map,
3860                                   &sys_chunk_size, &sys_stripe_size,
3861                                   sys_chunk_offset, alloc_profile);
3862         if (ret) {
3863                 btrfs_abort_transaction(trans, root, ret);
3864                 goto out;
3865         }
3866
3867         ret = btrfs_add_device(trans, fs_info->chunk_root, device);
3868         if (ret) {
3869                 btrfs_abort_transaction(trans, root, ret);
3870                 goto out;
3871         }
3872
3873         /*
3874          * Modifying chunk tree needs allocating new blocks from both
3875          * system block group and metadata block group. So we only can
3876          * do operations require modifying the chunk tree after both
3877          * block groups were created.
3878          */
3879         ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
3880                                    chunk_size, stripe_size);
3881         if (ret) {
3882                 btrfs_abort_transaction(trans, root, ret);
3883                 goto out;
3884         }
3885
3886         ret = __finish_chunk_alloc(trans, extent_root, sys_map,
3887                                    sys_chunk_offset, sys_chunk_size,
3888                                    sys_stripe_size);
3889         if (ret)
3890                 btrfs_abort_transaction(trans, root, ret);
3891
3892 out:
3893
3894         return ret;
3895 }
3896
3897 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
3898 {
3899         struct extent_map *em;
3900         struct map_lookup *map;
3901         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
3902         int readonly = 0;
3903         int i;
3904
3905         read_lock(&map_tree->map_tree.lock);
3906         em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
3907         read_unlock(&map_tree->map_tree.lock);
3908         if (!em)
3909                 return 1;
3910
3911         if (btrfs_test_opt(root, DEGRADED)) {
3912                 free_extent_map(em);
3913                 return 0;
3914         }
3915
3916         map = (struct map_lookup *)em->bdev;
3917         for (i = 0; i < map->num_stripes; i++) {
3918                 if (!map->stripes[i].dev->writeable) {
3919                         readonly = 1;
3920                         break;
3921                 }
3922         }
3923         free_extent_map(em);
3924         return readonly;
3925 }
3926
3927 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
3928 {
3929         extent_map_tree_init(&tree->map_tree);
3930 }
3931
3932 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
3933 {
3934         struct extent_map *em;
3935
3936         while (1) {
3937                 write_lock(&tree->map_tree.lock);
3938                 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
3939                 if (em)
3940                         remove_extent_mapping(&tree->map_tree, em);
3941                 write_unlock(&tree->map_tree.lock);
3942                 if (!em)
3943                         break;
3944                 kfree(em->bdev);
3945                 /* once for us */
3946                 free_extent_map(em);
3947                 /* once for the tree */
3948                 free_extent_map(em);
3949         }
3950 }
3951
3952 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
3953 {
3954         struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
3955         struct extent_map *em;
3956         struct map_lookup *map;
3957         struct extent_map_tree *em_tree = &map_tree->map_tree;
3958         int ret;
3959
3960         read_lock(&em_tree->lock);
3961         em = lookup_extent_mapping(em_tree, logical, len);
3962         read_unlock(&em_tree->lock);
3963         BUG_ON(!em);
3964
3965         BUG_ON(em->start > logical || em->start + em->len < logical);
3966         map = (struct map_lookup *)em->bdev;
3967         if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
3968                 ret = map->num_stripes;
3969         else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
3970                 ret = map->sub_stripes;
3971         else
3972                 ret = 1;
3973         free_extent_map(em);
3974         return ret;
3975 }
3976
3977 static int find_live_mirror(struct map_lookup *map, int first, int num,
3978                             int optimal)
3979 {
3980         int i;
3981         if (map->stripes[optimal].dev->bdev)
3982                 return optimal;
3983         for (i = first; i < first + num; i++) {
3984                 if (map->stripes[i].dev->bdev)
3985                         return i;
3986         }
3987         /* we couldn't find one that doesn't fail.  Just return something
3988          * and the io error handling code will clean up eventually
3989          */
3990         return optimal;
3991 }
3992
3993 static int __btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
3994                              u64 logical, u64 *length,
3995                              struct btrfs_bio **bbio_ret,
3996                              int mirror_num)
3997 {
3998         struct extent_map *em;
3999         struct map_lookup *map;
4000         struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
4001         struct extent_map_tree *em_tree = &map_tree->map_tree;
4002         u64 offset;
4003         u64 stripe_offset;
4004         u64 stripe_end_offset;
4005         u64 stripe_nr;
4006         u64 stripe_nr_orig;
4007         u64 stripe_nr_end;
4008         int stripe_index;
4009         int i;
4010         int ret = 0;
4011         int num_stripes;
4012         int max_errors = 0;
4013         struct btrfs_bio *bbio = NULL;
4014
4015         read_lock(&em_tree->lock);
4016         em = lookup_extent_mapping(em_tree, logical, *length);
4017         read_unlock(&em_tree->lock);
4018
4019         if (!em) {
4020                 printk(KERN_CRIT "btrfs: unable to find logical %llu len %llu\n",
4021                        (unsigned long long)logical,
4022                        (unsigned long long)*length);
4023                 BUG();
4024         }
4025
4026         BUG_ON(em->start > logical || em->start + em->len < logical);
4027         map = (struct map_lookup *)em->bdev;
4028         offset = logical - em->start;
4029
4030         if (mirror_num > map->num_stripes)
4031                 mirror_num = 0;
4032
4033         stripe_nr = offset;
4034         /*
4035          * stripe_nr counts the total number of stripes we have to stride
4036          * to get to this block
4037          */
4038         do_div(stripe_nr, map->stripe_len);
4039
4040         stripe_offset = stripe_nr * map->stripe_len;
4041         BUG_ON(offset < stripe_offset);
4042
4043         /* stripe_offset is the offset of this block in its stripe*/
4044         stripe_offset = offset - stripe_offset;
4045
4046         if (rw & REQ_DISCARD)
4047                 *length = min_t(u64, em->len - offset, *length);
4048         else if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
4049                 /* we limit the length of each bio to what fits in a stripe */
4050                 *length = min_t(u64, em->len - offset,
4051                                 map->stripe_len - stripe_offset);
4052         } else {
4053                 *length = em->len - offset;
4054         }
4055
4056         if (!bbio_ret)
4057                 goto out;
4058
4059         num_stripes = 1;
4060         stripe_index = 0;
4061         stripe_nr_orig = stripe_nr;
4062         stripe_nr_end = (offset + *length + map->stripe_len - 1) &
4063                         (~(map->stripe_len - 1));
4064         do_div(stripe_nr_end, map->stripe_len);
4065         stripe_end_offset = stripe_nr_end * map->stripe_len -
4066                             (offset + *length);
4067         if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
4068                 if (rw & REQ_DISCARD)
4069                         num_stripes = min_t(u64, map->num_stripes,
4070                                             stripe_nr_end - stripe_nr_orig);
4071                 stripe_index = do_div(stripe_nr, map->num_stripes);
4072         } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
4073                 if (rw & (REQ_WRITE | REQ_DISCARD))
4074                         num_stripes = map->num_stripes;
4075                 else if (mirror_num)
4076                         stripe_index = mirror_num - 1;
4077                 else {
4078                         stripe_index = find_live_mirror(map, 0,
4079                                             map->num_stripes,
4080                                             current->pid % map->num_stripes);
4081                         mirror_num = stripe_index + 1;
4082                 }
4083
4084         } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
4085                 if (rw & (REQ_WRITE | REQ_DISCARD)) {
4086                         num_stripes = map->num_stripes;
4087                 } else if (mirror_num) {
4088                         stripe_index = mirror_num - 1;
4089                 } else {
4090                         mirror_num = 1;
4091                 }
4092
4093         } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
4094                 int factor = map->num_stripes / map->sub_stripes;
4095
4096                 stripe_index = do_div(stripe_nr, factor);
4097                 stripe_index *= map->sub_stripes;
4098
4099                 if (rw & REQ_WRITE)
4100                         num_stripes = map->sub_stripes;
4101                 else if (rw & REQ_DISCARD)
4102                         num_stripes = min_t(u64, map->sub_stripes *
4103                                             (stripe_nr_end - stripe_nr_orig),
4104                                             map->num_stripes);
4105                 else if (mirror_num)
4106                         stripe_index += mirror_num - 1;
4107                 else {
4108                         int old_stripe_index = stripe_index;
4109                         stripe_index = find_live_mirror(map, stripe_index,
4110                                               map->sub_stripes, stripe_index +
4111                                               current->pid % map->sub_stripes);
4112                         mirror_num = stripe_index - old_stripe_index + 1;
4113                 }
4114         } else {
4115                 /*
4116                  * after this do_div call, stripe_nr is the number of stripes
4117                  * on this device we have to walk to find the data, and
4118                  * stripe_index is the number of our device in the stripe array
4119                  */
4120                 stripe_index = do_div(stripe_nr, map->num_stripes);
4121                 mirror_num = stripe_index + 1;
4122         }
4123         BUG_ON(stripe_index >= map->num_stripes);
4124
4125         bbio = kzalloc(btrfs_bio_size(num_stripes), GFP_NOFS);
4126         if (!bbio) {
4127                 ret = -ENOMEM;
4128                 goto out;
4129         }
4130         atomic_set(&bbio->error, 0);
4131
4132         if (rw & REQ_DISCARD) {
4133                 int factor = 0;
4134                 int sub_stripes = 0;
4135                 u64 stripes_per_dev = 0;
4136                 u32 remaining_stripes = 0;
4137                 u32 last_stripe = 0;
4138
4139                 if (map->type &
4140                     (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID10)) {
4141                         if (map->type & BTRFS_BLOCK_GROUP_RAID0)
4142                                 sub_stripes = 1;
4143                         else
4144                                 sub_stripes = map->sub_stripes;
4145
4146                         factor = map->num_stripes / sub_stripes;
4147                         stripes_per_dev = div_u64_rem(stripe_nr_end -
4148                                                       stripe_nr_orig,
4149                                                       factor,
4150                                                       &remaining_stripes);
4151                         div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
4152                         last_stripe *= sub_stripes;
4153                 }
4154
4155                 for (i = 0; i < num_stripes; i++) {
4156                         bbio->stripes[i].physical =
4157                                 map->stripes[stripe_index].physical +
4158                                 stripe_offset + stripe_nr * map->stripe_len;
4159                         bbio->stripes[i].dev = map->stripes[stripe_index].dev;
4160
4161                         if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
4162                                          BTRFS_BLOCK_GROUP_RAID10)) {
4163                                 bbio->stripes[i].length = stripes_per_dev *
4164                                                           map->stripe_len;
4165
4166                                 if (i / sub_stripes < remaining_stripes)
4167                                         bbio->stripes[i].length +=
4168                                                 map->stripe_len;
4169
4170                                 /*
4171                                  * Special for the first stripe and
4172                                  * the last stripe:
4173                                  *
4174                                  * |-------|...|-------|
4175                                  *     |----------|
4176                                  *    off     end_off
4177                                  */
4178                                 if (i < sub_stripes)
4179                                         bbio->stripes[i].length -=
4180                                                 stripe_offset;
4181
4182                                 if (stripe_index >= last_stripe &&
4183                                     stripe_index <= (last_stripe +
4184                                                      sub_stripes - 1))
4185                                         bbio->stripes[i].length -=
4186                                                 stripe_end_offset;
4187
4188                                 if (i == sub_stripes - 1)
4189                                         stripe_offset = 0;
4190                         } else
4191                                 bbio->stripes[i].length = *length;
4192
4193                         stripe_index++;
4194                         if (stripe_index == map->num_stripes) {
4195                                 /* This could only happen for RAID0/10 */
4196                                 stripe_index = 0;
4197                                 stripe_nr++;
4198                         }
4199                 }
4200         } else {
4201                 for (i = 0; i < num_stripes; i++) {
4202                         bbio->stripes[i].physical =
4203                                 map->stripes[stripe_index].physical +
4204                                 stripe_offset +
4205                                 stripe_nr * map->stripe_len;
4206                         bbio->stripes[i].dev =
4207                                 map->stripes[stripe_index].dev;
4208                         stripe_index++;
4209                 }
4210         }
4211
4212         if (rw & REQ_WRITE) {
4213                 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
4214                                  BTRFS_BLOCK_GROUP_RAID10 |
4215                                  BTRFS_BLOCK_GROUP_DUP)) {
4216                         max_errors = 1;
4217                 }
4218         }
4219
4220         *bbio_ret = bbio;
4221         bbio->num_stripes = num_stripes;
4222         bbio->max_errors = max_errors;
4223         bbio->mirror_num = mirror_num;
4224 out:
4225         free_extent_map(em);
4226         return ret;
4227 }
4228
4229 int btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
4230                       u64 logical, u64 *length,
4231                       struct btrfs_bio **bbio_ret, int mirror_num)
4232 {
4233         return __btrfs_map_block(fs_info, rw, logical, length, bbio_ret,
4234                                  mirror_num);
4235 }
4236
4237 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
4238                      u64 chunk_start, u64 physical, u64 devid,
4239                      u64 **logical, int *naddrs, int *stripe_len)
4240 {
4241         struct extent_map_tree *em_tree = &map_tree->map_tree;
4242         struct extent_map *em;
4243         struct map_lookup *map;
4244         u64 *buf;
4245         u64 bytenr;
4246         u64 length;
4247         u64 stripe_nr;
4248         int i, j, nr = 0;
4249
4250         read_lock(&em_tree->lock);
4251         em = lookup_extent_mapping(em_tree, chunk_start, 1);
4252         read_unlock(&em_tree->lock);
4253
4254         BUG_ON(!em || em->start != chunk_start);
4255         map = (struct map_lookup *)em->bdev;
4256
4257         length = em->len;
4258         if (map->type & BTRFS_BLOCK_GROUP_RAID10)
4259                 do_div(length, map->num_stripes / map->sub_stripes);
4260         else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
4261                 do_div(length, map->num_stripes);
4262
4263         buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
4264         BUG_ON(!buf); /* -ENOMEM */
4265
4266         for (i = 0; i < map->num_stripes; i++) {
4267                 if (devid && map->stripes[i].dev->devid != devid)
4268                         continue;
4269                 if (map->stripes[i].physical > physical ||
4270                     map->stripes[i].physical + length <= physical)
4271                         continue;
4272
4273                 stripe_nr = physical - map->stripes[i].physical;
4274                 do_div(stripe_nr, map->stripe_len);
4275
4276                 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
4277                         stripe_nr = stripe_nr * map->num_stripes + i;
4278                         do_div(stripe_nr, map->sub_stripes);
4279                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
4280                         stripe_nr = stripe_nr * map->num_stripes + i;
4281                 }
4282                 bytenr = chunk_start + stripe_nr * map->stripe_len;
4283                 WARN_ON(nr >= map->num_stripes);
4284                 for (j = 0; j < nr; j++) {
4285                         if (buf[j] == bytenr)
4286                                 break;
4287                 }
4288                 if (j == nr) {
4289                         WARN_ON(nr >= map->num_stripes);
4290                         buf[nr++] = bytenr;
4291                 }
4292         }
4293
4294         *logical = buf;
4295         *naddrs = nr;
4296         *stripe_len = map->stripe_len;
4297
4298         free_extent_map(em);
4299         return 0;
4300 }
4301
4302 static void *merge_stripe_index_into_bio_private(void *bi_private,
4303                                                  unsigned int stripe_index)
4304 {
4305         /*
4306          * with single, dup, RAID0, RAID1 and RAID10, stripe_index is
4307          * at most 1.
4308          * The alternative solution (instead of stealing bits from the
4309          * pointer) would be to allocate an intermediate structure
4310          * that contains the old private pointer plus the stripe_index.
4311          */
4312         BUG_ON((((uintptr_t)bi_private) & 3) != 0);
4313         BUG_ON(stripe_index > 3);
4314         return (void *)(((uintptr_t)bi_private) | stripe_index);
4315 }
4316
4317 static struct btrfs_bio *extract_bbio_from_bio_private(void *bi_private)
4318 {
4319         return (struct btrfs_bio *)(((uintptr_t)bi_private) & ~((uintptr_t)3));
4320 }
4321
4322 static unsigned int extract_stripe_index_from_bio_private(void *bi_private)
4323 {
4324         return (unsigned int)((uintptr_t)bi_private) & 3;
4325 }
4326
4327 static void btrfs_end_bio(struct bio *bio, int err)
4328 {
4329         struct btrfs_bio *bbio = extract_bbio_from_bio_private(bio->bi_private);
4330         int is_orig_bio = 0;
4331
4332         if (err) {
4333                 atomic_inc(&bbio->error);
4334                 if (err == -EIO || err == -EREMOTEIO) {
4335                         unsigned int stripe_index =
4336                                 extract_stripe_index_from_bio_private(
4337                                         bio->bi_private);
4338                         struct btrfs_device *dev;
4339
4340                         BUG_ON(stripe_index >= bbio->num_stripes);
4341                         dev = bbio->stripes[stripe_index].dev;
4342                         if (dev->bdev) {
4343                                 if (bio->bi_rw & WRITE)
4344                                         btrfs_dev_stat_inc(dev,
4345                                                 BTRFS_DEV_STAT_WRITE_ERRS);
4346                                 else
4347                                         btrfs_dev_stat_inc(dev,
4348                                                 BTRFS_DEV_STAT_READ_ERRS);
4349                                 if ((bio->bi_rw & WRITE_FLUSH) == WRITE_FLUSH)
4350                                         btrfs_dev_stat_inc(dev,
4351                                                 BTRFS_DEV_STAT_FLUSH_ERRS);
4352                                 btrfs_dev_stat_print_on_error(dev);
4353                         }
4354                 }
4355         }
4356
4357         if (bio == bbio->orig_bio)
4358                 is_orig_bio = 1;
4359
4360         if (atomic_dec_and_test(&bbio->stripes_pending)) {
4361                 if (!is_orig_bio) {
4362                         bio_put(bio);
4363                         bio = bbio->orig_bio;
4364                 }
4365                 bio->bi_private = bbio->private;
4366                 bio->bi_end_io = bbio->end_io;
4367                 bio->bi_bdev = (struct block_device *)
4368                                         (unsigned long)bbio->mirror_num;
4369                 /* only send an error to the higher layers if it is
4370                  * beyond the tolerance of the multi-bio
4371                  */
4372                 if (atomic_read(&bbio->error) > bbio->max_errors) {
4373                         err = -EIO;
4374                 } else {
4375                         /*
4376                          * this bio is actually up to date, we didn't
4377                          * go over the max number of errors
4378                          */
4379                         set_bit(BIO_UPTODATE, &bio->bi_flags);
4380                         err = 0;
4381                 }
4382                 kfree(bbio);
4383
4384                 bio_endio(bio, err);
4385         } else if (!is_orig_bio) {
4386                 bio_put(bio);
4387         }
4388 }
4389
4390 struct async_sched {
4391         struct bio *bio;
4392         int rw;
4393         struct btrfs_fs_info *info;
4394         struct btrfs_work work;
4395 };
4396
4397 /*
4398  * see run_scheduled_bios for a description of why bios are collected for
4399  * async submit.
4400  *
4401  * This will add one bio to the pending list for a device and make sure
4402  * the work struct is scheduled.
4403  */
4404 static noinline void schedule_bio(struct btrfs_root *root,
4405                                  struct btrfs_device *device,
4406                                  int rw, struct bio *bio)
4407 {
4408         int should_queue = 1;
4409         struct btrfs_pending_bios *pending_bios;
4410
4411         /* don't bother with additional async steps for reads, right now */
4412         if (!(rw & REQ_WRITE)) {
4413                 bio_get(bio);
4414                 btrfsic_submit_bio(rw, bio);
4415                 bio_put(bio);
4416                 return;
4417         }
4418
4419         /*
4420          * nr_async_bios allows us to reliably return congestion to the
4421          * higher layers.  Otherwise, the async bio makes it appear we have
4422          * made progress against dirty pages when we've really just put it
4423          * on a queue for later
4424          */
4425         atomic_inc(&root->fs_info->nr_async_bios);
4426         WARN_ON(bio->bi_next);
4427         bio->bi_next = NULL;
4428         bio->bi_rw |= rw;
4429
4430         spin_lock(&device->io_lock);
4431         if (bio->bi_rw & REQ_SYNC)
4432                 pending_bios = &device->pending_sync_bios;
4433         else
4434                 pending_bios = &device->pending_bios;
4435
4436         if (pending_bios->tail)
4437                 pending_bios->tail->bi_next = bio;
4438
4439         pending_bios->tail = bio;
4440         if (!pending_bios->head)
4441                 pending_bios->head = bio;
4442         if (device->running_pending)
4443                 should_queue = 0;
4444
4445         spin_unlock(&device->io_lock);
4446
4447         if (should_queue)
4448                 btrfs_queue_worker(&root->fs_info->submit_workers,
4449                                    &device->work);
4450 }
4451
4452 static int bio_size_ok(struct block_device *bdev, struct bio *bio,
4453                        sector_t sector)
4454 {
4455         struct bio_vec *prev;
4456         struct request_queue *q = bdev_get_queue(bdev);
4457         unsigned short max_sectors = queue_max_sectors(q);
4458         struct bvec_merge_data bvm = {
4459                 .bi_bdev = bdev,
4460                 .bi_sector = sector,
4461                 .bi_rw = bio->bi_rw,
4462         };
4463
4464         if (bio->bi_vcnt == 0) {
4465                 WARN_ON(1);
4466                 return 1;
4467         }
4468
4469         prev = &bio->bi_io_vec[bio->bi_vcnt - 1];
4470         if ((bio->bi_size >> 9) > max_sectors)
4471                 return 0;
4472
4473         if (!q->merge_bvec_fn)
4474                 return 1;
4475
4476         bvm.bi_size = bio->bi_size - prev->bv_len;
4477         if (q->merge_bvec_fn(q, &bvm, prev) < prev->bv_len)
4478                 return 0;
4479         return 1;
4480 }
4481
4482 static void submit_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio,
4483                               struct bio *bio, u64 physical, int dev_nr,
4484                               int rw, int async)
4485 {
4486         struct btrfs_device *dev = bbio->stripes[dev_nr].dev;
4487
4488         bio->bi_private = bbio;
4489         bio->bi_private = merge_stripe_index_into_bio_private(
4490                         bio->bi_private, (unsigned int)dev_nr);
4491         bio->bi_end_io = btrfs_end_bio;
4492         bio->bi_sector = physical >> 9;
4493 #ifdef DEBUG
4494         {
4495                 struct rcu_string *name;
4496
4497                 rcu_read_lock();
4498                 name = rcu_dereference(dev->name);
4499                 pr_debug("btrfs_map_bio: rw %d, sector=%llu, dev=%lu "
4500                          "(%s id %llu), size=%u\n", rw,
4501                          (u64)bio->bi_sector, (u_long)dev->bdev->bd_dev,
4502                          name->str, dev->devid, bio->bi_size);
4503                 rcu_read_unlock();
4504         }
4505 #endif
4506         bio->bi_bdev = dev->bdev;
4507         if (async)
4508                 schedule_bio(root, dev, rw, bio);
4509         else
4510                 btrfsic_submit_bio(rw, bio);
4511 }
4512
4513 static int breakup_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio,
4514                               struct bio *first_bio, struct btrfs_device *dev,
4515                               int dev_nr, int rw, int async)
4516 {
4517         struct bio_vec *bvec = first_bio->bi_io_vec;
4518         struct bio *bio;
4519         int nr_vecs = bio_get_nr_vecs(dev->bdev);
4520         u64 physical = bbio->stripes[dev_nr].physical;
4521
4522 again:
4523         bio = btrfs_bio_alloc(dev->bdev, physical >> 9, nr_vecs, GFP_NOFS);
4524         if (!bio)
4525                 return -ENOMEM;
4526
4527         while (bvec <= (first_bio->bi_io_vec + first_bio->bi_vcnt - 1)) {
4528                 if (bio_add_page(bio, bvec->bv_page, bvec->bv_len,
4529                                  bvec->bv_offset) < bvec->bv_len) {
4530                         u64 len = bio->bi_size;
4531
4532                         atomic_inc(&bbio->stripes_pending);
4533                         submit_stripe_bio(root, bbio, bio, physical, dev_nr,
4534                                           rw, async);
4535                         physical += len;
4536                         goto again;
4537                 }
4538                 bvec++;
4539         }
4540
4541         submit_stripe_bio(root, bbio, bio, physical, dev_nr, rw, async);
4542         return 0;
4543 }
4544
4545 static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical)
4546 {
4547         atomic_inc(&bbio->error);
4548         if (atomic_dec_and_test(&bbio->stripes_pending)) {
4549                 bio->bi_private = bbio->private;
4550                 bio->bi_end_io = bbio->end_io;
4551                 bio->bi_bdev = (struct block_device *)
4552                         (unsigned long)bbio->mirror_num;
4553                 bio->bi_sector = logical >> 9;
4554                 kfree(bbio);
4555                 bio_endio(bio, -EIO);
4556         }
4557 }
4558
4559 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
4560                   int mirror_num, int async_submit)
4561 {
4562         struct btrfs_device *dev;
4563         struct bio *first_bio = bio;
4564         u64 logical = (u64)bio->bi_sector << 9;
4565         u64 length = 0;
4566         u64 map_length;
4567         int ret;
4568         int dev_nr = 0;
4569         int total_devs = 1;
4570         struct btrfs_bio *bbio = NULL;
4571
4572         length = bio->bi_size;
4573         map_length = length;
4574
4575         ret = btrfs_map_block(root->fs_info, rw, logical, &map_length, &bbio,
4576                               mirror_num);
4577         if (ret)
4578                 return ret;
4579
4580         total_devs = bbio->num_stripes;
4581         if (map_length < length) {
4582                 printk(KERN_CRIT "btrfs: mapping failed logical %llu bio len %llu "
4583                        "len %llu\n", (unsigned long long)logical,
4584                        (unsigned long long)length,
4585                        (unsigned long long)map_length);
4586                 BUG();
4587         }
4588
4589         bbio->orig_bio = first_bio;
4590         bbio->private = first_bio->bi_private;
4591         bbio->end_io = first_bio->bi_end_io;
4592         atomic_set(&bbio->stripes_pending, bbio->num_stripes);
4593
4594         while (dev_nr < total_devs) {
4595                 dev = bbio->stripes[dev_nr].dev;
4596                 if (!dev || !dev->bdev || (rw & WRITE && !dev->writeable)) {
4597                         bbio_error(bbio, first_bio, logical);
4598                         dev_nr++;
4599                         continue;
4600                 }
4601
4602                 /*
4603                  * Check and see if we're ok with this bio based on it's size
4604                  * and offset with the given device.
4605                  */
4606                 if (!bio_size_ok(dev->bdev, first_bio,
4607                                  bbio->stripes[dev_nr].physical >> 9)) {
4608                         ret = breakup_stripe_bio(root, bbio, first_bio, dev,
4609                                                  dev_nr, rw, async_submit);
4610                         BUG_ON(ret);
4611                         dev_nr++;
4612                         continue;
4613                 }
4614
4615                 if (dev_nr < total_devs - 1) {
4616                         bio = bio_clone(first_bio, GFP_NOFS);
4617                         BUG_ON(!bio); /* -ENOMEM */
4618                 } else {
4619                         bio = first_bio;
4620                 }
4621
4622                 submit_stripe_bio(root, bbio, bio,
4623                                   bbio->stripes[dev_nr].physical, dev_nr, rw,
4624                                   async_submit);
4625                 dev_nr++;
4626         }
4627         return 0;
4628 }
4629
4630 struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid,
4631                                        u8 *uuid, u8 *fsid)
4632 {
4633         struct btrfs_device *device;
4634         struct btrfs_fs_devices *cur_devices;
4635
4636         cur_devices = fs_info->fs_devices;
4637         while (cur_devices) {
4638                 if (!fsid ||
4639                     !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
4640                         device = __find_device(&cur_devices->devices,
4641                                                devid, uuid);
4642                         if (device)
4643                                 return device;
4644                 }
4645                 cur_devices = cur_devices->seed;
4646         }
4647         return NULL;
4648 }
4649
4650 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
4651                                             u64 devid, u8 *dev_uuid)
4652 {
4653         struct btrfs_device *device;
4654         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
4655
4656         device = kzalloc(sizeof(*device), GFP_NOFS);
4657         if (!device)
4658                 return NULL;
4659         list_add(&device->dev_list,
4660                  &fs_devices->devices);
4661         device->dev_root = root->fs_info->dev_root;
4662         device->devid = devid;
4663         device->work.func = pending_bios_fn;
4664         device->fs_devices = fs_devices;
4665         device->missing = 1;
4666         fs_devices->num_devices++;
4667         fs_devices->missing_devices++;
4668         spin_lock_init(&device->io_lock);
4669         INIT_LIST_HEAD(&device->dev_alloc_list);
4670         memcpy(device->uuid, dev_uuid, BTRFS_UUID_SIZE);
4671         return device;
4672 }
4673
4674 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
4675                           struct extent_buffer *leaf,
4676                           struct btrfs_chunk *chunk)
4677 {
4678         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
4679         struct map_lookup *map;
4680         struct extent_map *em;
4681         u64 logical;
4682         u64 length;
4683         u64 devid;
4684         u8 uuid[BTRFS_UUID_SIZE];
4685         int num_stripes;
4686         int ret;
4687         int i;
4688
4689         logical = key->offset;
4690         length = btrfs_chunk_length(leaf, chunk);
4691
4692         read_lock(&map_tree->map_tree.lock);
4693         em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
4694         read_unlock(&map_tree->map_tree.lock);
4695
4696         /* already mapped? */
4697         if (em && em->start <= logical && em->start + em->len > logical) {
4698                 free_extent_map(em);
4699                 return 0;
4700         } else if (em) {
4701                 free_extent_map(em);
4702         }
4703
4704         em = alloc_extent_map();
4705         if (!em)
4706                 return -ENOMEM;
4707         num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
4708         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
4709         if (!map) {
4710                 free_extent_map(em);
4711                 return -ENOMEM;
4712         }
4713
4714         em->bdev = (struct block_device *)map;
4715         em->start = logical;
4716         em->len = length;
4717         em->block_start = 0;
4718         em->block_len = em->len;
4719
4720         map->num_stripes = num_stripes;
4721         map->io_width = btrfs_chunk_io_width(leaf, chunk);
4722         map->io_align = btrfs_chunk_io_align(leaf, chunk);
4723         map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
4724         map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
4725         map->type = btrfs_chunk_type(leaf, chunk);
4726         map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
4727         for (i = 0; i < num_stripes; i++) {
4728                 map->stripes[i].physical =
4729                         btrfs_stripe_offset_nr(leaf, chunk, i);
4730                 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
4731                 read_extent_buffer(leaf, uuid, (unsigned long)
4732                                    btrfs_stripe_dev_uuid_nr(chunk, i),
4733                                    BTRFS_UUID_SIZE);
4734                 map->stripes[i].dev = btrfs_find_device(root->fs_info, devid,
4735                                                         uuid, NULL);
4736                 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
4737                         kfree(map);
4738                         free_extent_map(em);
4739                         return -EIO;
4740                 }
4741                 if (!map->stripes[i].dev) {
4742                         map->stripes[i].dev =
4743                                 add_missing_dev(root, devid, uuid);
4744                         if (!map->stripes[i].dev) {
4745                                 kfree(map);
4746                                 free_extent_map(em);
4747                                 return -EIO;
4748                         }
4749                 }
4750                 map->stripes[i].dev->in_fs_metadata = 1;
4751         }
4752
4753         write_lock(&map_tree->map_tree.lock);
4754         ret = add_extent_mapping(&map_tree->map_tree, em);
4755         write_unlock(&map_tree->map_tree.lock);
4756         BUG_ON(ret); /* Tree corruption */
4757         free_extent_map(em);
4758
4759         return 0;
4760 }
4761
4762 static void fill_device_from_item(struct extent_buffer *leaf,
4763                                  struct btrfs_dev_item *dev_item,
4764                                  struct btrfs_device *device)
4765 {
4766         unsigned long ptr;
4767
4768         device->devid = btrfs_device_id(leaf, dev_item);
4769         device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
4770         device->total_bytes = device->disk_total_bytes;
4771         device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
4772         device->type = btrfs_device_type(leaf, dev_item);
4773         device->io_align = btrfs_device_io_align(leaf, dev_item);
4774         device->io_width = btrfs_device_io_width(leaf, dev_item);
4775         device->sector_size = btrfs_device_sector_size(leaf, dev_item);
4776         device->is_tgtdev_for_dev_replace = 0;
4777
4778         ptr = (unsigned long)btrfs_device_uuid(dev_item);
4779         read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
4780 }
4781
4782 static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
4783 {
4784         struct btrfs_fs_devices *fs_devices;
4785         int ret;
4786
4787         BUG_ON(!mutex_is_locked(&uuid_mutex));
4788
4789         fs_devices = root->fs_info->fs_devices->seed;
4790         while (fs_devices) {
4791                 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
4792                         ret = 0;
4793                         goto out;
4794                 }
4795                 fs_devices = fs_devices->seed;
4796         }
4797
4798         fs_devices = find_fsid(fsid);
4799         if (!fs_devices) {
4800                 ret = -ENOENT;
4801                 goto out;
4802         }
4803
4804         fs_devices = clone_fs_devices(fs_devices);
4805         if (IS_ERR(fs_devices)) {
4806                 ret = PTR_ERR(fs_devices);
4807                 goto out;
4808         }
4809
4810         ret = __btrfs_open_devices(fs_devices, FMODE_READ,
4811                                    root->fs_info->bdev_holder);
4812         if (ret) {
4813                 free_fs_devices(fs_devices);
4814                 goto out;
4815         }
4816
4817         if (!fs_devices->seeding) {
4818                 __btrfs_close_devices(fs_devices);
4819                 free_fs_devices(fs_devices);
4820                 ret = -EINVAL;
4821                 goto out;
4822         }
4823
4824         fs_devices->seed = root->fs_info->fs_devices->seed;
4825         root->fs_info->fs_devices->seed = fs_devices;
4826 out:
4827         return ret;
4828 }
4829
4830 static int read_one_dev(struct btrfs_root *root,
4831                         struct extent_buffer *leaf,
4832                         struct btrfs_dev_item *dev_item)
4833 {
4834         struct btrfs_device *device;
4835         u64 devid;
4836         int ret;
4837         u8 fs_uuid[BTRFS_UUID_SIZE];
4838         u8 dev_uuid[BTRFS_UUID_SIZE];
4839
4840         devid = btrfs_device_id(leaf, dev_item);
4841         read_extent_buffer(leaf, dev_uuid,
4842                            (unsigned long)btrfs_device_uuid(dev_item),
4843                            BTRFS_UUID_SIZE);
4844         read_extent_buffer(leaf, fs_uuid,
4845                            (unsigned long)btrfs_device_fsid(dev_item),
4846                            BTRFS_UUID_SIZE);
4847
4848         if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
4849                 ret = open_seed_devices(root, fs_uuid);
4850                 if (ret && !btrfs_test_opt(root, DEGRADED))
4851                         return ret;
4852         }
4853
4854         device = btrfs_find_device(root->fs_info, devid, dev_uuid, fs_uuid);
4855         if (!device || !device->bdev) {
4856                 if (!btrfs_test_opt(root, DEGRADED))
4857                         return -EIO;
4858
4859                 if (!device) {
4860                         printk(KERN_WARNING "warning devid %llu missing\n",
4861                                (unsigned long long)devid);
4862                         device = add_missing_dev(root, devid, dev_uuid);
4863                         if (!device)
4864                                 return -ENOMEM;
4865                 } else if (!device->missing) {
4866                         /*
4867                          * this happens when a device that was properly setup
4868                          * in the device info lists suddenly goes bad.
4869                          * device->bdev is NULL, and so we have to set
4870                          * device->missing to one here
4871                          */
4872                         root->fs_info->fs_devices->missing_devices++;
4873                         device->missing = 1;
4874                 }
4875         }
4876
4877         if (device->fs_devices != root->fs_info->fs_devices) {
4878                 BUG_ON(device->writeable);
4879                 if (device->generation !=
4880                     btrfs_device_generation(leaf, dev_item))
4881                         return -EINVAL;
4882         }
4883
4884         fill_device_from_item(leaf, dev_item, device);
4885         device->dev_root = root->fs_info->dev_root;
4886         device->in_fs_metadata = 1;
4887         if (device->writeable && !device->is_tgtdev_for_dev_replace) {
4888                 device->fs_devices->total_rw_bytes += device->total_bytes;
4889                 spin_lock(&root->fs_info->free_chunk_lock);
4890                 root->fs_info->free_chunk_space += device->total_bytes -
4891                         device->bytes_used;
4892                 spin_unlock(&root->fs_info->free_chunk_lock);
4893         }
4894         ret = 0;
4895         return ret;
4896 }
4897
4898 int btrfs_read_sys_array(struct btrfs_root *root)
4899 {
4900         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
4901         struct extent_buffer *sb;
4902         struct btrfs_disk_key *disk_key;
4903         struct btrfs_chunk *chunk;
4904         u8 *ptr;
4905         unsigned long sb_ptr;
4906         int ret = 0;
4907         u32 num_stripes;
4908         u32 array_size;
4909         u32 len = 0;
4910         u32 cur;
4911         struct btrfs_key key;
4912
4913         sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
4914                                           BTRFS_SUPER_INFO_SIZE);
4915         if (!sb)
4916                 return -ENOMEM;
4917         btrfs_set_buffer_uptodate(sb);
4918         btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
4919         /*
4920          * The sb extent buffer is artifical and just used to read the system array.
4921          * btrfs_set_buffer_uptodate() call does not properly mark all it's
4922          * pages up-to-date when the page is larger: extent does not cover the
4923          * whole page and consequently check_page_uptodate does not find all
4924          * the page's extents up-to-date (the hole beyond sb),
4925          * write_extent_buffer then triggers a WARN_ON.
4926          *
4927          * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
4928          * but sb spans only this function. Add an explicit SetPageUptodate call
4929          * to silence the warning eg. on PowerPC 64.
4930          */
4931         if (PAGE_CACHE_SIZE > BTRFS_SUPER_INFO_SIZE)
4932                 SetPageUptodate(sb->pages[0]);
4933
4934         write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
4935         array_size = btrfs_super_sys_array_size(super_copy);
4936
4937         ptr = super_copy->sys_chunk_array;
4938         sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
4939         cur = 0;
4940
4941         while (cur < array_size) {
4942                 disk_key = (struct btrfs_disk_key *)ptr;
4943                 btrfs_disk_key_to_cpu(&key, disk_key);
4944
4945                 len = sizeof(*disk_key); ptr += len;
4946                 sb_ptr += len;
4947                 cur += len;
4948
4949                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
4950                         chunk = (struct btrfs_chunk *)sb_ptr;
4951                         ret = read_one_chunk(root, &key, sb, chunk);
4952                         if (ret)
4953                                 break;
4954                         num_stripes = btrfs_chunk_num_stripes(sb, chunk);
4955                         len = btrfs_chunk_item_size(num_stripes);
4956                 } else {
4957                         ret = -EIO;
4958                         break;
4959                 }
4960                 ptr += len;
4961                 sb_ptr += len;
4962                 cur += len;
4963         }
4964         free_extent_buffer(sb);
4965         return ret;
4966 }
4967
4968 int btrfs_read_chunk_tree(struct btrfs_root *root)
4969 {
4970         struct btrfs_path *path;
4971         struct extent_buffer *leaf;
4972         struct btrfs_key key;
4973         struct btrfs_key found_key;
4974         int ret;
4975         int slot;
4976
4977         root = root->fs_info->chunk_root;
4978
4979         path = btrfs_alloc_path();
4980         if (!path)
4981                 return -ENOMEM;
4982
4983         mutex_lock(&uuid_mutex);
4984         lock_chunks(root);
4985
4986         /* first we search for all of the device items, and then we
4987          * read in all of the chunk items.  This way we can create chunk
4988          * mappings that reference all of the devices that are afound
4989          */
4990         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
4991         key.offset = 0;
4992         key.type = 0;
4993 again:
4994         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4995         if (ret < 0)
4996                 goto error;
4997         while (1) {
4998                 leaf = path->nodes[0];
4999                 slot = path->slots[0];
5000                 if (slot >= btrfs_header_nritems(leaf)) {
5001                         ret = btrfs_next_leaf(root, path);
5002                         if (ret == 0)
5003                                 continue;
5004                         if (ret < 0)
5005                                 goto error;
5006                         break;
5007                 }
5008                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5009                 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
5010                         if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID)
5011                                 break;
5012                         if (found_key.type == BTRFS_DEV_ITEM_KEY) {
5013                                 struct btrfs_dev_item *dev_item;
5014                                 dev_item = btrfs_item_ptr(leaf, slot,
5015                                                   struct btrfs_dev_item);
5016                                 ret = read_one_dev(root, leaf, dev_item);
5017                                 if (ret)
5018                                         goto error;
5019                         }
5020                 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
5021                         struct btrfs_chunk *chunk;
5022                         chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
5023                         ret = read_one_chunk(root, &found_key, leaf, chunk);
5024                         if (ret)
5025                                 goto error;
5026                 }
5027                 path->slots[0]++;
5028         }
5029         if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
5030                 key.objectid = 0;
5031                 btrfs_release_path(path);
5032                 goto again;
5033         }
5034         ret = 0;
5035 error:
5036         unlock_chunks(root);
5037         mutex_unlock(&uuid_mutex);
5038
5039         btrfs_free_path(path);
5040         return ret;
5041 }
5042
5043 static void __btrfs_reset_dev_stats(struct btrfs_device *dev)
5044 {
5045         int i;
5046
5047         for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
5048                 btrfs_dev_stat_reset(dev, i);
5049 }
5050
5051 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
5052 {
5053         struct btrfs_key key;
5054         struct btrfs_key found_key;
5055         struct btrfs_root *dev_root = fs_info->dev_root;
5056         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
5057         struct extent_buffer *eb;
5058         int slot;
5059         int ret = 0;
5060         struct btrfs_device *device;
5061         struct btrfs_path *path = NULL;
5062         int i;
5063
5064         path = btrfs_alloc_path();
5065         if (!path) {
5066                 ret = -ENOMEM;
5067                 goto out;
5068         }
5069
5070         mutex_lock(&fs_devices->device_list_mutex);
5071         list_for_each_entry(device, &fs_devices->devices, dev_list) {
5072                 int item_size;
5073                 struct btrfs_dev_stats_item *ptr;
5074
5075                 key.objectid = 0;
5076                 key.type = BTRFS_DEV_STATS_KEY;
5077                 key.offset = device->devid;
5078                 ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
5079                 if (ret) {
5080                         __btrfs_reset_dev_stats(device);
5081                         device->dev_stats_valid = 1;
5082                         btrfs_release_path(path);
5083                         continue;
5084                 }
5085                 slot = path->slots[0];
5086                 eb = path->nodes[0];
5087                 btrfs_item_key_to_cpu(eb, &found_key, slot);
5088                 item_size = btrfs_item_size_nr(eb, slot);
5089
5090                 ptr = btrfs_item_ptr(eb, slot,
5091                                      struct btrfs_dev_stats_item);
5092
5093                 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
5094                         if (item_size >= (1 + i) * sizeof(__le64))
5095                                 btrfs_dev_stat_set(device, i,
5096                                         btrfs_dev_stats_value(eb, ptr, i));
5097                         else
5098                                 btrfs_dev_stat_reset(device, i);
5099                 }
5100
5101                 device->dev_stats_valid = 1;
5102                 btrfs_dev_stat_print_on_load(device);
5103                 btrfs_release_path(path);
5104         }
5105         mutex_unlock(&fs_devices->device_list_mutex);
5106
5107 out:
5108         btrfs_free_path(path);
5109         return ret < 0 ? ret : 0;
5110 }
5111
5112 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
5113                                 struct btrfs_root *dev_root,
5114                                 struct btrfs_device *device)
5115 {
5116         struct btrfs_path *path;
5117         struct btrfs_key key;
5118         struct extent_buffer *eb;
5119         struct btrfs_dev_stats_item *ptr;
5120         int ret;
5121         int i;
5122
5123         key.objectid = 0;
5124         key.type = BTRFS_DEV_STATS_KEY;
5125         key.offset = device->devid;
5126
5127         path = btrfs_alloc_path();
5128         BUG_ON(!path);
5129         ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
5130         if (ret < 0) {
5131                 printk_in_rcu(KERN_WARNING "btrfs: error %d while searching for dev_stats item for device %s!\n",
5132                               ret, rcu_str_deref(device->name));
5133                 goto out;
5134         }
5135
5136         if (ret == 0 &&
5137             btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
5138                 /* need to delete old one and insert a new one */
5139                 ret = btrfs_del_item(trans, dev_root, path);
5140                 if (ret != 0) {
5141                         printk_in_rcu(KERN_WARNING "btrfs: delete too small dev_stats item for device %s failed %d!\n",
5142                                       rcu_str_deref(device->name), ret);
5143                         goto out;
5144                 }
5145                 ret = 1;
5146         }
5147
5148         if (ret == 1) {
5149                 /* need to insert a new item */
5150                 btrfs_release_path(path);
5151                 ret = btrfs_insert_empty_item(trans, dev_root, path,
5152                                               &key, sizeof(*ptr));
5153                 if (ret < 0) {
5154                         printk_in_rcu(KERN_WARNING "btrfs: insert dev_stats item for device %s failed %d!\n",
5155                                       rcu_str_deref(device->name), ret);
5156                         goto out;
5157                 }
5158         }
5159
5160         eb = path->nodes[0];
5161         ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
5162         for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
5163                 btrfs_set_dev_stats_value(eb, ptr, i,
5164                                           btrfs_dev_stat_read(device, i));
5165         btrfs_mark_buffer_dirty(eb);
5166
5167 out:
5168         btrfs_free_path(path);
5169         return ret;
5170 }
5171
5172 /*
5173  * called from commit_transaction. Writes all changed device stats to disk.
5174  */
5175 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans,
5176                         struct btrfs_fs_info *fs_info)
5177 {
5178         struct btrfs_root *dev_root = fs_info->dev_root;
5179         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
5180         struct btrfs_device *device;
5181         int ret = 0;
5182
5183         mutex_lock(&fs_devices->device_list_mutex);
5184         list_for_each_entry(device, &fs_devices->devices, dev_list) {
5185                 if (!device->dev_stats_valid || !device->dev_stats_dirty)
5186                         continue;
5187
5188                 ret = update_dev_stat_item(trans, dev_root, device);
5189                 if (!ret)
5190                         device->dev_stats_dirty = 0;
5191         }
5192         mutex_unlock(&fs_devices->device_list_mutex);
5193
5194         return ret;
5195 }
5196
5197 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
5198 {
5199         btrfs_dev_stat_inc(dev, index);
5200         btrfs_dev_stat_print_on_error(dev);
5201 }
5202
5203 void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
5204 {
5205         if (!dev->dev_stats_valid)
5206                 return;
5207         printk_ratelimited_in_rcu(KERN_ERR
5208                            "btrfs: bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
5209                            rcu_str_deref(dev->name),
5210                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
5211                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
5212                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
5213                            btrfs_dev_stat_read(dev,
5214                                                BTRFS_DEV_STAT_CORRUPTION_ERRS),
5215                            btrfs_dev_stat_read(dev,
5216                                                BTRFS_DEV_STAT_GENERATION_ERRS));
5217 }
5218
5219 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
5220 {
5221         int i;
5222
5223         for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
5224                 if (btrfs_dev_stat_read(dev, i) != 0)
5225                         break;
5226         if (i == BTRFS_DEV_STAT_VALUES_MAX)
5227                 return; /* all values == 0, suppress message */
5228
5229         printk_in_rcu(KERN_INFO "btrfs: bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
5230                rcu_str_deref(dev->name),
5231                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
5232                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
5233                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
5234                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
5235                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
5236 }
5237
5238 int btrfs_get_dev_stats(struct btrfs_root *root,
5239                         struct btrfs_ioctl_get_dev_stats *stats)
5240 {
5241         struct btrfs_device *dev;
5242         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
5243         int i;
5244
5245         mutex_lock(&fs_devices->device_list_mutex);
5246         dev = btrfs_find_device(root->fs_info, stats->devid, NULL, NULL);
5247         mutex_unlock(&fs_devices->device_list_mutex);
5248
5249         if (!dev) {
5250                 printk(KERN_WARNING
5251                        "btrfs: get dev_stats failed, device not found\n");
5252                 return -ENODEV;
5253         } else if (!dev->dev_stats_valid) {
5254                 printk(KERN_WARNING
5255                        "btrfs: get dev_stats failed, not yet valid\n");
5256                 return -ENODEV;
5257         } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
5258                 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
5259                         if (stats->nr_items > i)
5260                                 stats->values[i] =
5261                                         btrfs_dev_stat_read_and_reset(dev, i);
5262                         else
5263                                 btrfs_dev_stat_reset(dev, i);
5264                 }
5265         } else {
5266                 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
5267                         if (stats->nr_items > i)
5268                                 stats->values[i] = btrfs_dev_stat_read(dev, i);
5269         }
5270         if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
5271                 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
5272         return 0;
5273 }
5274
5275 int btrfs_scratch_superblock(struct btrfs_device *device)
5276 {
5277         struct buffer_head *bh;
5278         struct btrfs_super_block *disk_super;
5279
5280         bh = btrfs_read_dev_super(device->bdev);
5281         if (!bh)
5282                 return -EINVAL;
5283         disk_super = (struct btrfs_super_block *)bh->b_data;
5284
5285         memset(&disk_super->magic, 0, sizeof(disk_super->magic));
5286         set_buffer_dirty(bh);
5287         sync_dirty_buffer(bh);
5288         brelse(bh);
5289
5290         return 0;
5291 }