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 #include "dev-replace.h"
  40
  41 static int init_first_rw_device(struct btrfs_trans_handle *trans,
  42                                 struct btrfs_root *root,
  43                                 struct btrfs_device *device);
  44 static int btrfs_relocate_sys_chunks(struct btrfs_root *root);
  45 static void __btrfs_reset_dev_stats(struct btrfs_device *dev);
  46 static void btrfs_dev_stat_print_on_load(struct btrfs_device *device);
  47
  48 static DEFINE_MUTEX(uuid_mutex);
  49 static LIST_HEAD(fs_uuids);
  50
  51 static void lock_chunks(struct btrfs_root *root)
  52 {
  53         mutex_lock(&root->fs_info->chunk_mutex);
  54 }
  55
  56 static void unlock_chunks(struct btrfs_root *root)
  57 {
  58         mutex_unlock(&root->fs_info->chunk_mutex);
  59 }
  60
  61 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
  62 {
  63         struct btrfs_device *device;
  64         WARN_ON(fs_devices->opened);
  65         while (!list_empty(&fs_devices->devices)) {
  66                 device = list_entry(fs_devices->devices.next,
  67                                     struct btrfs_device, dev_list);
  68                 list_del(&device->dev_list);
  69                 rcu_string_free(device->name);
  70                 kfree(device);
  71         }
  72         kfree(fs_devices);
  73 }
  74
  75 void btrfs_cleanup_fs_uuids(void)
  76 {
  77         struct btrfs_fs_devices *fs_devices;
  78
  79         while (!list_empty(&fs_uuids)) {
  80                 fs_devices = list_entry(fs_uuids.next,
  81                                         struct btrfs_fs_devices, list);
  82                 list_del(&fs_devices->list);
  83                 free_fs_devices(fs_devices);
  84         }
  85 }
  86
  87 static noinline struct btrfs_device *__find_device(struct list_head *head,
  88                                                    u64 devid, u8 *uuid)
  89 {
  90         struct btrfs_device *dev;
  91
  92         list_for_each_entry(dev, head, dev_list) {
  93                 if (dev->devid == devid &&
  94                     (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
  95                         return dev;
  96                 }
  97         }
  98         return NULL;
  99 }
 100
 101 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
 102 {
 103         struct btrfs_fs_devices *fs_devices;
 104
 105         list_for_each_entry(fs_devices, &fs_uuids, list) {
 106                 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
 107                         return fs_devices;
 108         }
 109         return NULL;
 110 }
 111
 112 static int
 113 btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder,
 114                       int flush, struct block_device **bdev,
 115                       struct buffer_head **bh)
 116 {
 117         int ret;
 118
 119         *bdev = blkdev_get_by_path(device_path, flags, holder);
 120
 121         if (IS_ERR(*bdev)) {
 122                 ret = PTR_ERR(*bdev);
 123                 printk(KERN_INFO "btrfs: open %s failed\n", device_path);
 124                 goto error;
 125         }
 126
 127         if (flush)
 128                 filemap_write_and_wait((*bdev)->bd_inode->i_mapping);
 129         ret = set_blocksize(*bdev, 4096);
 130         if (ret) {
 131                 blkdev_put(*bdev, flags);
 132                 goto error;
 133         }
 134         invalidate_bdev(*bdev);
 135         *bh = btrfs_read_dev_super(*bdev);
 136         if (!*bh) {
 137                 ret = -EINVAL;
 138                 blkdev_put(*bdev, flags);
 139                 goto error;
 140         }
 141
 142         return 0;
 143
 144 error:
 145         *bdev = NULL;
 146         *bh = NULL;
 147         return ret;
 148 }
 149
 150 static void requeue_list(struct btrfs_pending_bios *pending_bios,
 151                         struct bio *head, struct bio *tail)
 152 {
 153
 154         struct bio *old_head;
 155
 156         old_head = pending_bios->head;
 157         pending_bios->head = head;
 158         if (pending_bios->tail)
 159                 tail->bi_next = old_head;
 160         else
 161                 pending_bios->tail = tail;
 162 }
 163
 164 /*
 165  * we try to collect pending bios for a device so we don't get a large
 166  * number of procs sending bios down to the same device.  This greatly
 167  * improves the schedulers ability to collect and merge the bios.
 168  *
 169  * But, it also turns into a long list of bios to process and that is sure
 170  * to eventually make the worker thread block.  The solution here is to
 171  * make some progress and then put this work struct back at the end of
 172  * the list if the block device is congested.  This way, multiple devices
 173  * can make progress from a single worker thread.
 174  */
 175 static noinline void run_scheduled_bios(struct btrfs_device *device)
 176 {
 177         struct bio *pending;
 178         struct backing_dev_info *bdi;
 179         struct btrfs_fs_info *fs_info;
 180         struct btrfs_pending_bios *pending_bios;
 181         struct bio *tail;
 182         struct bio *cur;
 183         int again = 0;
 184         unsigned long num_run;
 185         unsigned long batch_run = 0;
 186         unsigned long limit;
 187         unsigned long last_waited = 0;
 188         int force_reg = 0;
 189         int sync_pending = 0;
 190         struct blk_plug plug;
 191
 192         /*
 193          * this function runs all the bios we've collected for
 194          * a particular device.  We don't want to wander off to
 195          * another device without first sending all of these down.
 196          * So, setup a plug here and finish it off before we return
 197          */
 198         blk_start_plug(&plug);
 199
 200         bdi = blk_get_backing_dev_info(device->bdev);
 201         fs_info = device->dev_root->fs_info;
 202         limit = btrfs_async_submit_limit(fs_info);
 203         limit = limit * 2 / 3;
 204
 205 loop:
 206         spin_lock(&device->io_lock);
 207
 208 loop_lock:
 209         num_run = 0;
 210
 211         /* take all the bios off the list at once and process them
 212          * later on (without the lock held).  But, remember the
 213          * tail and other pointers so the bios can be properly reinserted
 214          * into the list if we hit congestion
 215          */
 216         if (!force_reg && device->pending_sync_bios.head) {
 217                 pending_bios = &device->pending_sync_bios;
 218                 force_reg = 1;
 219         } else {
 220                 pending_bios = &device->pending_bios;
 221                 force_reg = 0;
 222         }
 223
 224         pending = pending_bios->head;
 225         tail = pending_bios->tail;
 226         WARN_ON(pending && !tail);
 227
 228         /*
 229          * if pending was null this time around, no bios need processing
 230          * at all and we can stop.  Otherwise it'll loop back up again
 231          * and do an additional check so no bios are missed.
 232          *
 233          * device->running_pending is used to synchronize with the
 234          * schedule_bio code.
 235          */
 236         if (device->pending_sync_bios.head == NULL &&
 237             device->pending_bios.head == NULL) {
 238                 again = 0;
 239                 device->running_pending = 0;
 240         } else {
 241                 again = 1;
 242                 device->running_pending = 1;
 243         }
 244
 245         pending_bios->head = NULL;
 246         pending_bios->tail = NULL;
 247
 248         spin_unlock(&device->io_lock);
 249
 250         while (pending) {
 251
 252                 rmb();
 253                 /* we want to work on both lists, but do more bios on the
 254                  * sync list than the regular list
 255                  */
 256                 if ((num_run > 32 &&
 257                     pending_bios != &device->pending_sync_bios &&
 258                     device->pending_sync_bios.head) ||
 259                    (num_run > 64 && pending_bios == &device->pending_sync_bios &&
 260                     device->pending_bios.head)) {
 261                         spin_lock(&device->io_lock);
 262                         requeue_list(pending_bios, pending, tail);
 263                         goto loop_lock;
 264                 }
 265
 266                 cur = pending;
 267                 pending = pending->bi_next;
 268                 cur->bi_next = NULL;
 269
 270                 if (atomic_dec_return(&fs_info->nr_async_bios) < limit &&
 271                     waitqueue_active(&fs_info->async_submit_wait))
 272                         wake_up(&fs_info->async_submit_wait);
 273
 274                 BUG_ON(atomic_read(&cur->bi_cnt) == 0);
 275
 276                 /*
 277                  * if we're doing the sync list, record that our
 278                  * plug has some sync requests on it
 279                  *
 280                  * If we're doing the regular list and there are
 281                  * sync requests sitting around, unplug before
 282                  * we add more
 283                  */
 284                 if (pending_bios == &device->pending_sync_bios) {
 285                         sync_pending = 1;
 286                 } else if (sync_pending) {
 287                         blk_finish_plug(&plug);
 288                         blk_start_plug(&plug);
 289                         sync_pending = 0;
 290                 }
 291
 292                 btrfsic_submit_bio(cur->bi_rw, cur);
 293                 num_run++;
 294                 batch_run++;
 295                 if (need_resched())
 296                         cond_resched();
 297
 298                 /*
 299                  * we made progress, there is more work to do and the bdi
 300                  * is now congested.  Back off and let other work structs
 301                  * run instead
 302                  */
 303                 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
 304                     fs_info->fs_devices->open_devices > 1) {
 305                         struct io_context *ioc;
 306
 307                         ioc = current->io_context;
 308
 309                         /*
 310                          * the main goal here is that we don't want to
 311                          * block if we're going to be able to submit
 312                          * more requests without blocking.
 313                          *
 314                          * This code does two great things, it pokes into
 315                          * the elevator code from a filesystem _and_
 316                          * it makes assumptions about how batching works.
 317                          */
 318                         if (ioc && ioc->nr_batch_requests > 0 &&
 319                             time_before(jiffies, ioc->last_waited + HZ/50UL) &&
 320                             (last_waited == 0 ||
 321                              ioc->last_waited == last_waited)) {
 322                                 /*
 323                                  * we want to go through our batch of
 324                                  * requests and stop.  So, we copy out
 325                                  * the ioc->last_waited time and test
 326                                  * against it before looping
 327                                  */
 328                                 last_waited = ioc->last_waited;
 329                                 if (need_resched())
 330                                         cond_resched();
 331                                 continue;
 332                         }
 333                         spin_lock(&device->io_lock);
 334                         requeue_list(pending_bios, pending, tail);
 335                         device->running_pending = 1;
 336
 337                         spin_unlock(&device->io_lock);
 338                         btrfs_requeue_work(&device->work);
 339                         goto done;
 340                 }
 341                 /* unplug every 64 requests just for good measure */
 342                 if (batch_run % 64 == 0) {
 343                         blk_finish_plug(&plug);
 344                         blk_start_plug(&plug);
 345                         sync_pending = 0;
 346                 }
 347         }
 348
 349         cond_resched();
 350         if (again)
 351                 goto loop;
 352
 353         spin_lock(&device->io_lock);
 354         if (device->pending_bios.head || device->pending_sync_bios.head)
 355                 goto loop_lock;
 356         spin_unlock(&device->io_lock);
 357
 358 done:
 359         blk_finish_plug(&plug);
 360 }
 361
 362 static void pending_bios_fn(struct btrfs_work *work)
 363 {
 364         struct btrfs_device *device;
 365
 366         device = container_of(work, struct btrfs_device, work);
 367         run_scheduled_bios(device);
 368 }
 369
 370 static noinline int device_list_add(const char *path,
 371                            struct btrfs_super_block *disk_super,
 372                            u64 devid, struct btrfs_fs_devices **fs_devices_ret)
 373 {
 374         struct btrfs_device *device;
 375         struct btrfs_fs_devices *fs_devices;
 376         struct rcu_string *name;
 377         u64 found_transid = btrfs_super_generation(disk_super);
 378
 379         fs_devices = find_fsid(disk_super->fsid);
 380         if (!fs_devices) {
 381                 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
 382                 if (!fs_devices)
 383                         return -ENOMEM;
 384                 INIT_LIST_HEAD(&fs_devices->devices);
 385                 INIT_LIST_HEAD(&fs_devices->alloc_list);
 386                 list_add(&fs_devices->list, &fs_uuids);
 387                 memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
 388                 fs_devices->latest_devid = devid;
 389                 fs_devices->latest_trans = found_transid;
 390                 mutex_init(&fs_devices->device_list_mutex);
 391                 device = NULL;
 392         } else {
 393                 device = __find_device(&fs_devices->devices, devid,
 394                                        disk_super->dev_item.uuid);
 395         }
 396         if (!device) {
 397                 if (fs_devices->opened)
 398                         return -EBUSY;
 399
 400                 device = kzalloc(sizeof(*device), GFP_NOFS);
 401                 if (!device) {
 402                         /* we can safely leave the fs_devices entry around */
 403                         return -ENOMEM;
 404                 }
 405                 device->devid = devid;
 406                 device->dev_stats_valid = 0;
 407                 device->work.func = pending_bios_fn;
 408                 memcpy(device->uuid, disk_super->dev_item.uuid,
 409                        BTRFS_UUID_SIZE);
 410                 spin_lock_init(&device->io_lock);
 411
 412                 name = rcu_string_strdup(path, GFP_NOFS);
 413                 if (!name) {
 414                         kfree(device);
 415                         return -ENOMEM;
 416                 }
 417                 rcu_assign_pointer(device->name, name);
 418                 INIT_LIST_HEAD(&device->dev_alloc_list);
 419
 420                 /* init readahead state */
 421                 spin_lock_init(&device->reada_lock);
 422                 device->reada_curr_zone = NULL;
 423                 atomic_set(&device->reada_in_flight, 0);
 424                 device->reada_next = 0;
 425                 INIT_RADIX_TREE(&device->reada_zones, GFP_NOFS & ~__GFP_WAIT);
 426                 INIT_RADIX_TREE(&device->reada_extents, GFP_NOFS & ~__GFP_WAIT);
 427
 428                 mutex_lock(&fs_devices->device_list_mutex);
 429                 list_add_rcu(&device->dev_list, &fs_devices->devices);
 430                 mutex_unlock(&fs_devices->device_list_mutex);
 431
 432                 device->fs_devices = fs_devices;
 433                 fs_devices->num_devices++;
 434         } else if (!device->name || strcmp(device->name->str, path)) {
 435                 name = rcu_string_strdup(path, GFP_NOFS);
 436                 if (!name)
 437                         return -ENOMEM;
 438                 rcu_string_free(device->name);
 439                 rcu_assign_pointer(device->name, name);
 440                 if (device->missing) {
 441                         fs_devices->missing_devices--;
 442                         device->missing = 0;
 443                 }
 444         }
 445
 446         if (found_transid > fs_devices->latest_trans) {
 447                 fs_devices->latest_devid = devid;
 448                 fs_devices->latest_trans = found_transid;
 449         }
 450         *fs_devices_ret = fs_devices;
 451         return 0;
 452 }
 453
 454 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
 455 {
 456         struct btrfs_fs_devices *fs_devices;
 457         struct btrfs_device *device;
 458         struct btrfs_device *orig_dev;
 459
 460         fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
 461         if (!fs_devices)
 462                 return ERR_PTR(-ENOMEM);
 463
 464         INIT_LIST_HEAD(&fs_devices->devices);
 465         INIT_LIST_HEAD(&fs_devices->alloc_list);
 466         INIT_LIST_HEAD(&fs_devices->list);
 467         mutex_init(&fs_devices->device_list_mutex);
 468         fs_devices->latest_devid = orig->latest_devid;
 469         fs_devices->latest_trans = orig->latest_trans;
 470         fs_devices->total_devices = orig->total_devices;
 471         memcpy(fs_devices->fsid, orig->fsid, sizeof(fs_devices->fsid));
 472
 473         /* We have held the volume lock, it is safe to get the devices. */
 474         list_for_each_entry(orig_dev, &orig->devices, dev_list) {
 475                 struct rcu_string *name;
 476
 477                 device = kzalloc(sizeof(*device), GFP_NOFS);
 478                 if (!device)
 479                         goto error;
 480
 481                 /*
 482                  * This is ok to do without rcu read locked because we hold the
 483                  * uuid mutex so nothing we touch in here is going to disappear.
 484                  */
 485                 name = rcu_string_strdup(orig_dev->name->str, GFP_NOFS);
 486                 if (!name) {
 487                         kfree(device);
 488                         goto error;
 489                 }
 490                 rcu_assign_pointer(device->name, name);
 491
 492                 device->devid = orig_dev->devid;
 493                 device->work.func = pending_bios_fn;
 494                 memcpy(device->uuid, orig_dev->uuid, sizeof(device->uuid));
 495                 spin_lock_init(&device->io_lock);
 496                 INIT_LIST_HEAD(&device->dev_list);
 497                 INIT_LIST_HEAD(&device->dev_alloc_list);
 498
 499                 list_add(&device->dev_list, &fs_devices->devices);
 500                 device->fs_devices = fs_devices;
 501                 fs_devices->num_devices++;
 502         }
 503         return fs_devices;
 504 error:
 505         free_fs_devices(fs_devices);
 506         return ERR_PTR(-ENOMEM);
 507 }
 508
 509 void btrfs_close_extra_devices(struct btrfs_fs_info *fs_info,
 510                                struct btrfs_fs_devices *fs_devices, int step)
 511 {
 512         struct btrfs_device *device, *next;
 513
 514         struct block_device *latest_bdev = NULL;
 515         u64 latest_devid = 0;
 516         u64 latest_transid = 0;
 517
 518         mutex_lock(&uuid_mutex);
 519 again:
 520         /* This is the initialized path, it is safe to release the devices. */
 521         list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
 522                 if (device->in_fs_metadata) {
 523                         if (!device->is_tgtdev_for_dev_replace &&
 524                             (!latest_transid ||
 525                              device->generation > latest_transid)) {
 526                                 latest_devid = device->devid;
 527                                 latest_transid = device->generation;
 528                                 latest_bdev = device->bdev;
 529                         }
 530                         continue;
 531                 }
 532
 533                 if (device->devid == BTRFS_DEV_REPLACE_DEVID) {
 534                         /*
 535                          * In the first step, keep the device which has
 536                          * the correct fsid and the devid that is used
 537                          * for the dev_replace procedure.
 538                          * In the second step, the dev_replace state is
 539                          * read from the device tree and it is known
 540                          * whether the procedure is really active or
 541                          * not, which means whether this device is
 542                          * used or whether it should be removed.
 543                          */
 544                         if (step == 0 || device->is_tgtdev_for_dev_replace) {
 545                                 continue;
 546                         }
 547                 }
 548                 if (device->bdev) {
 549                         blkdev_put(device->bdev, device->mode);
 550                         device->bdev = NULL;
 551                         fs_devices->open_devices--;
 552                 }
 553                 if (device->writeable) {
 554                         list_del_init(&device->dev_alloc_list);
 555                         device->writeable = 0;
 556                         if (!device->is_tgtdev_for_dev_replace)
 557                                 fs_devices->rw_devices--;
 558                 }
 559                 list_del_init(&device->dev_list);
 560                 fs_devices->num_devices--;
 561                 rcu_string_free(device->name);
 562                 kfree(device);
 563         }
 564
 565         if (fs_devices->seed) {
 566                 fs_devices = fs_devices->seed;
 567                 goto again;
 568         }
 569
 570         fs_devices->latest_bdev = latest_bdev;
 571         fs_devices->latest_devid = latest_devid;
 572         fs_devices->latest_trans = latest_transid;
 573
 574         mutex_unlock(&uuid_mutex);
 575 }
 576
 577 static void __free_device(struct work_struct *work)
 578 {
 579         struct btrfs_device *device;
 580
 581         device = container_of(work, struct btrfs_device, rcu_work);
 582
 583         if (device->bdev)
 584                 blkdev_put(device->bdev, device->mode);
 585
 586         rcu_string_free(device->name);
 587         kfree(device);
 588 }
 589
 590 static void free_device(struct rcu_head *head)
 591 {
 592         struct btrfs_device *device;
 593
 594         device = container_of(head, struct btrfs_device, rcu);
 595
 596         INIT_WORK(&device->rcu_work, __free_device);
 597         schedule_work(&device->rcu_work);
 598 }
 599
 600 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
 601 {
 602         struct btrfs_device *device;
 603
 604         if (--fs_devices->opened > 0)
 605                 return 0;
 606
 607         mutex_lock(&fs_devices->device_list_mutex);
 608         list_for_each_entry(device, &fs_devices->devices, dev_list) {
 609                 struct btrfs_device *new_device;
 610                 struct rcu_string *name;
 611
 612                 if (device->bdev)
 613                         fs_devices->open_devices--;
 614
 615                 if (device->writeable && !device->is_tgtdev_for_dev_replace) {
 616                         list_del_init(&device->dev_alloc_list);
 617                         fs_devices->rw_devices--;
 618                 }
 619
 620                 if (device->can_discard)
 621                         fs_devices->num_can_discard--;
 622
 623                 new_device = kmalloc(sizeof(*new_device), GFP_NOFS);
 624                 BUG_ON(!new_device); /* -ENOMEM */
 625                 memcpy(new_device, device, sizeof(*new_device));
 626
 627                 /* Safe because we are under uuid_mutex */
 628                 if (device->name) {
 629                         name = rcu_string_strdup(device->name->str, GFP_NOFS);
 630                         BUG_ON(device->name && !name); /* -ENOMEM */
 631                         rcu_assign_pointer(new_device->name, name);
 632                 }
 633                 new_device->bdev = NULL;
 634                 new_device->writeable = 0;
 635                 new_device->in_fs_metadata = 0;
 636                 new_device->can_discard = 0;
 637                 list_replace_rcu(&device->dev_list, &new_device->dev_list);
 638
 639                 call_rcu(&device->rcu, free_device);
 640         }
 641         mutex_unlock(&fs_devices->device_list_mutex);
 642
 643         WARN_ON(fs_devices->open_devices);
 644         WARN_ON(fs_devices->rw_devices);
 645         fs_devices->opened = 0;
 646         fs_devices->seeding = 0;
 647
 648         return 0;
 649 }
 650
 651 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
 652 {
 653         struct btrfs_fs_devices *seed_devices = NULL;
 654         int ret;
 655
 656         mutex_lock(&uuid_mutex);
 657         ret = __btrfs_close_devices(fs_devices);
 658         if (!fs_devices->opened) {
 659                 seed_devices = fs_devices->seed;
 660                 fs_devices->seed = NULL;
 661         }
 662         mutex_unlock(&uuid_mutex);
 663
 664         while (seed_devices) {
 665                 fs_devices = seed_devices;
 666                 seed_devices = fs_devices->seed;
 667                 __btrfs_close_devices(fs_devices);
 668                 free_fs_devices(fs_devices);
 669         }
 670         return ret;
 671 }
 672
 673 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
 674                                 fmode_t flags, void *holder)
 675 {
 676         struct request_queue *q;
 677         struct block_device *bdev;
 678         struct list_head *head = &fs_devices->devices;
 679         struct btrfs_device *device;
 680         struct block_device *latest_bdev = NULL;
 681         struct buffer_head *bh;
 682         struct btrfs_super_block *disk_super;
 683         u64 latest_devid = 0;
 684         u64 latest_transid = 0;
 685         u64 devid;
 686         int seeding = 1;
 687         int ret = 0;
 688
 689         flags |= FMODE_EXCL;
 690
 691         list_for_each_entry(device, head, dev_list) {
 692                 if (device->bdev)
 693                         continue;
 694                 if (!device->name)
 695                         continue;
 696
 697                 ret = btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1,
 698                                             &bdev, &bh);
 699                 if (ret)
 700                         continue;
 701
 702                 disk_super = (struct btrfs_super_block *)bh->b_data;
 703                 devid = btrfs_stack_device_id(&disk_super->dev_item);
 704                 if (devid != device->devid)
 705                         goto error_brelse;
 706
 707                 if (memcmp(device->uuid, disk_super->dev_item.uuid,
 708                            BTRFS_UUID_SIZE))
 709                         goto error_brelse;
 710
 711                 device->generation = btrfs_super_generation(disk_super);
 712                 if (!latest_transid || device->generation > latest_transid) {
 713                         latest_devid = devid;
 714                         latest_transid = device->generation;
 715                         latest_bdev = bdev;
 716                 }
 717
 718                 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
 719                         device->writeable = 0;
 720                 } else {
 721                         device->writeable = !bdev_read_only(bdev);
 722                         seeding = 0;
 723                 }
 724
 725                 q = bdev_get_queue(bdev);
 726                 if (blk_queue_discard(q)) {
 727                         device->can_discard = 1;
 728                         fs_devices->num_can_discard++;
 729                 }
 730
 731                 device->bdev = bdev;
 732                 device->in_fs_metadata = 0;
 733                 device->mode = flags;
 734
 735                 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
 736                         fs_devices->rotating = 1;
 737
 738                 fs_devices->open_devices++;
 739                 if (device->writeable && !device->is_tgtdev_for_dev_replace) {
 740                         fs_devices->rw_devices++;
 741                         list_add(&device->dev_alloc_list,
 742                                  &fs_devices->alloc_list);
 743                 }
 744                 brelse(bh);
 745                 continue;
 746
 747 error_brelse:
 748                 brelse(bh);
 749                 blkdev_put(bdev, flags);
 750                 continue;
 751         }
 752         if (fs_devices->open_devices == 0) {
 753                 ret = -EINVAL;
 754                 goto out;
 755         }
 756         fs_devices->seeding = seeding;
 757         fs_devices->opened = 1;
 758         fs_devices->latest_bdev = latest_bdev;
 759         fs_devices->latest_devid = latest_devid;
 760         fs_devices->latest_trans = latest_transid;
 761         fs_devices->total_rw_bytes = 0;
 762 out:
 763         return ret;
 764 }
 765
 766 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
 767                        fmode_t flags, void *holder)
 768 {
 769         int ret;
 770
 771         mutex_lock(&uuid_mutex);
 772         if (fs_devices->opened) {
 773                 fs_devices->opened++;
 774                 ret = 0;
 775         } else {
 776                 ret = __btrfs_open_devices(fs_devices, flags, holder);
 777         }
 778         mutex_unlock(&uuid_mutex);
 779         return ret;
 780 }
 781
 782 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
 783                           struct btrfs_fs_devices **fs_devices_ret)
 784 {
 785         struct btrfs_super_block *disk_super;
 786         struct block_device *bdev;
 787         struct buffer_head *bh;
 788         int ret;
 789         u64 devid;
 790         u64 transid;
 791         u64 total_devices;
 792
 793         flags |= FMODE_EXCL;
 794         mutex_lock(&uuid_mutex);
 795         ret = btrfs_get_bdev_and_sb(path, flags, holder, 0, &bdev, &bh);
 796         if (ret)
 797                 goto error;
 798         disk_super = (struct btrfs_super_block *)bh->b_data;
 799         devid = btrfs_stack_device_id(&disk_super->dev_item);
 800         transid = btrfs_super_generation(disk_super);
 801         total_devices = btrfs_super_num_devices(disk_super);
 802         if (disk_super->label[0]) {
 803                 if (disk_super->label[BTRFS_LABEL_SIZE - 1])
 804                         disk_super->label[BTRFS_LABEL_SIZE - 1] = '\0';
 805                 printk(KERN_INFO "device label %s ", disk_super->label);
 806         } else {
 807                 printk(KERN_INFO "device fsid %pU ", disk_super->fsid);
 808         }
 809         printk(KERN_CONT "devid %llu transid %llu %s\n",
 810                (unsigned long long)devid, (unsigned long long)transid, path);
 811         ret = device_list_add(path, disk_super, devid, fs_devices_ret);
 812         if (!ret && fs_devices_ret)
 813                 (*fs_devices_ret)->total_devices = total_devices;
 814         brelse(bh);
 815         blkdev_put(bdev, flags);
 816 error:
 817         mutex_unlock(&uuid_mutex);
 818         return ret;
 819 }
 820
 821 /* helper to account the used device space in the range */
 822 int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
 823                                    u64 end, u64 *length)
 824 {
 825         struct btrfs_key key;
 826         struct btrfs_root *root = device->dev_root;
 827         struct btrfs_dev_extent *dev_extent;
 828         struct btrfs_path *path;
 829         u64 extent_end;
 830         int ret;
 831         int slot;
 832         struct extent_buffer *l;
 833
 834         *length = 0;
 835
 836         if (start >= device->total_bytes || device->is_tgtdev_for_dev_replace)
 837                 return 0;
 838
 839         path = btrfs_alloc_path();
 840         if (!path)
 841                 return -ENOMEM;
 842         path->reada = 2;
 843
 844         key.objectid = device->devid;
 845         key.offset = start;
 846         key.type = BTRFS_DEV_EXTENT_KEY;
 847
 848         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 849         if (ret < 0)
 850                 goto out;
 851         if (ret > 0) {
 852                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
 853                 if (ret < 0)
 854                         goto out;
 855         }
 856
 857         while (1) {
 858                 l = path->nodes[0];
 859                 slot = path->slots[0];
 860                 if (slot >= btrfs_header_nritems(l)) {
 861                         ret = btrfs_next_leaf(root, path);
 862                         if (ret == 0)
 863                                 continue;
 864                         if (ret < 0)
 865                                 goto out;
 866
 867                         break;
 868                 }
 869                 btrfs_item_key_to_cpu(l, &key, slot);
 870
 871                 if (key.objectid < device->devid)
 872                         goto next;
 873
 874                 if (key.objectid > device->devid)
 875                         break;
 876
 877                 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
 878                         goto next;
 879
 880                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
 881                 extent_end = key.offset + btrfs_dev_extent_length(l,
 882                                                                   dev_extent);
 883                 if (key.offset <= start && extent_end > end) {
 884                         *length = end - start + 1;
 885                         break;
 886                 } else if (key.offset <= start && extent_end > start)
 887                         *length += extent_end - start;
 888                 else if (key.offset > start && extent_end <= end)
 889                         *length += extent_end - key.offset;
 890                 else if (key.offset > start && key.offset <= end) {
 891                         *length += end - key.offset + 1;
 892                         break;
 893                 } else if (key.offset > end)
 894                         break;
 895
 896 next:
 897                 path->slots[0]++;
 898         }
 899         ret = 0;
 900 out:
 901         btrfs_free_path(path);
 902         return ret;
 903 }
 904
 905 /*
 906  * find_free_dev_extent - find free space in the specified device
 907  * @device:     the device which we search the free space in
 908  * @num_bytes:  the size of the free space that we need
 909  * @start:      store the start of the free space.
 910  * @len:        the size of the free space. that we find, or the size of the max
 911  *              free space if we don't find suitable free space
 912  *
 913  * this uses a pretty simple search, the expectation is that it is
 914  * called very infrequently and that a given device has a small number
 915  * of extents
 916  *
 917  * @start is used to store the start of the free space if we find. But if we
 918  * don't find suitable free space, it will be used to store the start position
 919  * of the max free space.
 920  *
 921  * @len is used to store the size of the free space that we find.
 922  * But if we don't find suitable free space, it is used to store the size of
 923  * the max free space.
 924  */
 925 int find_free_dev_extent(struct btrfs_device *device, u64 num_bytes,
 926                          u64 *start, u64 *len)
 927 {
 928         struct btrfs_key key;
 929         struct btrfs_root *root = device->dev_root;
 930         struct btrfs_dev_extent *dev_extent;
 931         struct btrfs_path *path;
 932         u64 hole_size;
 933         u64 max_hole_start;
 934         u64 max_hole_size;
 935         u64 extent_end;
 936         u64 search_start;
 937         u64 search_end = device->total_bytes;
 938         int ret;
 939         int slot;
 940         struct extent_buffer *l;
 941
 942         /* FIXME use last free of some kind */
 943
 944         /* we don't want to overwrite the superblock on the drive,
 945          * so we make sure to start at an offset of at least 1MB
 946          */
 947         search_start = max(root->fs_info->alloc_start, 1024ull * 1024);
 948
 949         max_hole_start = search_start;
 950         max_hole_size = 0;
 951         hole_size = 0;
 952
 953         if (search_start >= search_end || device->is_tgtdev_for_dev_replace) {
 954                 ret = -ENOSPC;
 955                 goto error;
 956         }
 957
 958         path = btrfs_alloc_path();
 959         if (!path) {
 960                 ret = -ENOMEM;
 961                 goto error;
 962         }
 963         path->reada = 2;
 964
 965         key.objectid = device->devid;
 966         key.offset = search_start;
 967         key.type = BTRFS_DEV_EXTENT_KEY;
 968
 969         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 970         if (ret < 0)
 971                 goto out;
 972         if (ret > 0) {
 973                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
 974                 if (ret < 0)
 975                         goto out;
 976         }
 977
 978         while (1) {
 979                 l = path->nodes[0];
 980                 slot = path->slots[0];
 981                 if (slot >= btrfs_header_nritems(l)) {
 982                         ret = btrfs_next_leaf(root, path);
 983                         if (ret == 0)
 984                                 continue;
 985                         if (ret < 0)
 986                                 goto out;
 987
 988                         break;
 989                 }
 990                 btrfs_item_key_to_cpu(l, &key, slot);
 991
 992                 if (key.objectid < device->devid)
 993                         goto next;
 994
 995                 if (key.objectid > device->devid)
 996                         break;
 997
 998                 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
 999                         goto next;
1000
1001                 if (key.offset > search_start) {
1002                         hole_size = key.offset - search_start;
1003
1004                         if (hole_size > max_hole_size) {
1005                                 max_hole_start = search_start;
1006                                 max_hole_size = hole_size;
1007                         }
1008
1009                         /*
1010                          * If this free space is greater than which we need,
1011                          * it must be the max free space that we have found
1012                          * until now, so max_hole_start must point to the start
1013                          * of this free space and the length of this free space
1014                          * is stored in max_hole_size. Thus, we return
1015                          * max_hole_start and max_hole_size and go back to the
1016                          * caller.
1017                          */
1018                         if (hole_size >= num_bytes) {
1019                                 ret = 0;
1020                                 goto out;
1021                         }
1022                 }
1023
1024                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1025                 extent_end = key.offset + btrfs_dev_extent_length(l,
1026                                                                   dev_extent);
1027                 if (extent_end > search_start)
1028                         search_start = extent_end;
1029 next:
1030                 path->slots[0]++;
1031                 cond_resched();
1032         }
1033
1034         /*
1035          * At this point, search_start should be the end of
1036          * allocated dev extents, and when shrinking the device,
1037          * search_end may be smaller than search_start.
1038          */
1039         if (search_end > search_start)
1040                 hole_size = search_end - search_start;
1041
1042         if (hole_size > max_hole_size) {
1043                 max_hole_start = search_start;
1044                 max_hole_size = hole_size;
1045         }
1046
1047         /* See above. */
1048         if (hole_size < num_bytes)
1049                 ret = -ENOSPC;
1050         else
1051                 ret = 0;
1052
1053 out:
1054         btrfs_free_path(path);
1055 error:
1056         *start = max_hole_start;
1057         if (len)
1058                 *len = max_hole_size;
1059         return ret;
1060 }
1061
1062 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1063                           struct btrfs_device *device,
1064                           u64 start)
1065 {
1066         int ret;
1067         struct btrfs_path *path;
1068         struct btrfs_root *root = device->dev_root;
1069         struct btrfs_key key;
1070         struct btrfs_key found_key;
1071         struct extent_buffer *leaf = NULL;
1072         struct btrfs_dev_extent *extent = NULL;
1073
1074         path = btrfs_alloc_path();
1075         if (!path)
1076                 return -ENOMEM;
1077
1078         key.objectid = device->devid;
1079         key.offset = start;
1080         key.type = BTRFS_DEV_EXTENT_KEY;
1081 again:
1082         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1083         if (ret > 0) {
1084                 ret = btrfs_previous_item(root, path, key.objectid,
1085                                           BTRFS_DEV_EXTENT_KEY);
1086                 if (ret)
1087                         goto out;
1088                 leaf = path->nodes[0];
1089                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1090                 extent = btrfs_item_ptr(leaf, path->slots[0],
1091                                         struct btrfs_dev_extent);
1092                 BUG_ON(found_key.offset > start || found_key.offset +
1093                        btrfs_dev_extent_length(leaf, extent) < start);
1094                 key = found_key;
1095                 btrfs_release_path(path);
1096                 goto again;
1097         } else if (ret == 0) {
1098                 leaf = path->nodes[0];
1099                 extent = btrfs_item_ptr(leaf, path->slots[0],
1100                                         struct btrfs_dev_extent);
1101         } else {
1102                 btrfs_error(root->fs_info, ret, "Slot search failed");
1103                 goto out;
1104         }
1105
1106         if (device->bytes_used > 0) {
1107                 u64 len = btrfs_dev_extent_length(leaf, extent);
1108                 device->bytes_used -= len;
1109                 spin_lock(&root->fs_info->free_chunk_lock);
1110                 root->fs_info->free_chunk_space += len;
1111                 spin_unlock(&root->fs_info->free_chunk_lock);
1112         }
1113         ret = btrfs_del_item(trans, root, path);
1114         if (ret) {
1115                 btrfs_error(root->fs_info, ret,
1116                             "Failed to remove dev extent item");
1117         }
1118 out:
1119         btrfs_free_path(path);
1120         return ret;
1121 }
1122
1123 int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1124                            struct btrfs_device *device,
1125                            u64 chunk_tree, u64 chunk_objectid,
1126                            u64 chunk_offset, u64 start, u64 num_bytes)
1127 {
1128         int ret;
1129         struct btrfs_path *path;
1130         struct btrfs_root *root = device->dev_root;
1131         struct btrfs_dev_extent *extent;
1132         struct extent_buffer *leaf;
1133         struct btrfs_key key;
1134
1135         WARN_ON(!device->in_fs_metadata);
1136         WARN_ON(device->is_tgtdev_for_dev_replace);
1137         path = btrfs_alloc_path();
1138         if (!path)
1139                 return -ENOMEM;
1140
1141         key.objectid = device->devid;
1142         key.offset = start;
1143         key.type = BTRFS_DEV_EXTENT_KEY;
1144         ret = btrfs_insert_empty_item(trans, root, path, &key,
1145                                       sizeof(*extent));
1146         if (ret)
1147                 goto out;
1148
1149         leaf = path->nodes[0];
1150         extent = btrfs_item_ptr(leaf, path->slots[0],
1151                                 struct btrfs_dev_extent);
1152         btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
1153         btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
1154         btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1155
1156         write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
1157                     (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
1158                     BTRFS_UUID_SIZE);
1159
1160         btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1161         btrfs_mark_buffer_dirty(leaf);
1162 out:
1163         btrfs_free_path(path);
1164         return ret;
1165 }
1166
1167 static noinline int find_next_chunk(struct btrfs_root *root,
1168                                     u64 objectid, u64 *offset)
1169 {
1170         struct btrfs_path *path;
1171         int ret;
1172         struct btrfs_key key;
1173         struct btrfs_chunk *chunk;
1174         struct btrfs_key found_key;
1175
1176         path = btrfs_alloc_path();
1177         if (!path)
1178                 return -ENOMEM;
1179
1180         key.objectid = objectid;
1181         key.offset = (u64)-1;
1182         key.type = BTRFS_CHUNK_ITEM_KEY;
1183
1184         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1185         if (ret < 0)
1186                 goto error;
1187
1188         BUG_ON(ret == 0); /* Corruption */
1189
1190         ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
1191         if (ret) {
1192                 *offset = 0;
1193         } else {
1194                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1195                                       path->slots[0]);
1196                 if (found_key.objectid != objectid)
1197                         *offset = 0;
1198                 else {
1199                         chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
1200                                                struct btrfs_chunk);
1201                         *offset = found_key.offset +
1202                                 btrfs_chunk_length(path->nodes[0], chunk);
1203                 }
1204         }
1205         ret = 0;
1206 error:
1207         btrfs_free_path(path);
1208         return ret;
1209 }
1210
1211 static noinline int find_next_devid(struct btrfs_root *root, u64 *objectid)
1212 {
1213         int ret;
1214         struct btrfs_key key;
1215         struct btrfs_key found_key;
1216         struct btrfs_path *path;
1217
1218         root = root->fs_info->chunk_root;
1219
1220         path = btrfs_alloc_path();
1221         if (!path)
1222                 return -ENOMEM;
1223
1224         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1225         key.type = BTRFS_DEV_ITEM_KEY;
1226         key.offset = (u64)-1;
1227
1228         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1229         if (ret < 0)
1230                 goto error;
1231
1232         BUG_ON(ret == 0); /* Corruption */
1233
1234         ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
1235                                   BTRFS_DEV_ITEM_KEY);
1236         if (ret) {
1237                 *objectid = 1;
1238         } else {
1239                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1240                                       path->slots[0]);
1241                 *objectid = found_key.offset + 1;
1242         }
1243         ret = 0;
1244 error:
1245         btrfs_free_path(path);
1246         return ret;
1247 }
1248
1249 /*
1250  * the device information is stored in the chunk root
1251  * the btrfs_device struct should be fully filled in
1252  */
1253 int btrfs_add_device(struct btrfs_trans_handle *trans,
1254                      struct btrfs_root *root,
1255                      struct btrfs_device *device)
1256 {
1257         int ret;
1258         struct btrfs_path *path;
1259         struct btrfs_dev_item *dev_item;
1260         struct extent_buffer *leaf;
1261         struct btrfs_key key;
1262         unsigned long ptr;
1263
1264         root = root->fs_info->chunk_root;
1265
1266         path = btrfs_alloc_path();
1267         if (!path)
1268                 return -ENOMEM;
1269
1270         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1271         key.type = BTRFS_DEV_ITEM_KEY;
1272         key.offset = device->devid;
1273
1274         ret = btrfs_insert_empty_item(trans, root, path, &key,
1275                                       sizeof(*dev_item));
1276         if (ret)
1277                 goto out;
1278
1279         leaf = path->nodes[0];
1280         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1281
1282         btrfs_set_device_id(leaf, dev_item, device->devid);
1283         btrfs_set_device_generation(leaf, dev_item, 0);
1284         btrfs_set_device_type(leaf, dev_item, device->type);
1285         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1286         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1287         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1288         btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
1289         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1290         btrfs_set_device_group(leaf, dev_item, 0);
1291         btrfs_set_device_seek_speed(leaf, dev_item, 0);
1292         btrfs_set_device_bandwidth(leaf, dev_item, 0);
1293         btrfs_set_device_start_offset(leaf, dev_item, 0);
1294
1295         ptr = (unsigned long)btrfs_device_uuid(dev_item);
1296         write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1297         ptr = (unsigned long)btrfs_device_fsid(dev_item);
1298         write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1299         btrfs_mark_buffer_dirty(leaf);
1300
1301         ret = 0;
1302 out:
1303         btrfs_free_path(path);
1304         return ret;
1305 }
1306
1307 static int btrfs_rm_dev_item(struct btrfs_root *root,
1308                              struct btrfs_device *device)
1309 {
1310         int ret;
1311         struct btrfs_path *path;
1312         struct btrfs_key key;
1313         struct btrfs_trans_handle *trans;
1314
1315         root = root->fs_info->chunk_root;
1316
1317         path = btrfs_alloc_path();
1318         if (!path)
1319                 return -ENOMEM;
1320
1321         trans = btrfs_start_transaction(root, 0);
1322         if (IS_ERR(trans)) {
1323                 btrfs_free_path(path);
1324                 return PTR_ERR(trans);
1325         }
1326         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1327         key.type = BTRFS_DEV_ITEM_KEY;
1328         key.offset = device->devid;
1329         lock_chunks(root);
1330
1331         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1332         if (ret < 0)
1333                 goto out;
1334
1335         if (ret > 0) {
1336                 ret = -ENOENT;
1337                 goto out;
1338         }
1339
1340         ret = btrfs_del_item(trans, root, path);
1341         if (ret)
1342                 goto out;
1343 out:
1344         btrfs_free_path(path);
1345         unlock_chunks(root);
1346         btrfs_commit_transaction(trans, root);
1347         return ret;
1348 }
1349
1350 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
1351 {
1352         struct btrfs_device *device;
1353         struct btrfs_device *next_device;
1354         struct block_device *bdev;
1355         struct buffer_head *bh = NULL;
1356         struct btrfs_super_block *disk_super;
1357         struct btrfs_fs_devices *cur_devices;
1358         u64 all_avail;
1359         u64 devid;
1360         u64 num_devices;
1361         u8 *dev_uuid;
1362         int ret = 0;
1363         bool clear_super = false;
1364
1365         mutex_lock(&uuid_mutex);
1366
1367         all_avail = root->fs_info->avail_data_alloc_bits |
1368                 root->fs_info->avail_system_alloc_bits |
1369                 root->fs_info->avail_metadata_alloc_bits;
1370
1371         num_devices = root->fs_info->fs_devices->num_devices;
1372         btrfs_dev_replace_lock(&root->fs_info->dev_replace);
1373         if (btrfs_dev_replace_is_ongoing(&root->fs_info->dev_replace)) {
1374                 WARN_ON(num_devices < 1);
1375                 num_devices--;
1376         }
1377         btrfs_dev_replace_unlock(&root->fs_info->dev_replace);
1378
1379         if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) && num_devices <= 4) {
1380                 printk(KERN_ERR "btrfs: unable to go below four devices "
1381                        "on raid10\n");
1382                 ret = -EINVAL;
1383                 goto out;
1384         }
1385
1386         if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) && num_devices <= 2) {
1387                 printk(KERN_ERR "btrfs: unable to go below two "
1388                        "devices on raid1\n");
1389                 ret = -EINVAL;
1390                 goto out;
1391         }
1392
1393         if (strcmp(device_path, "missing") == 0) {
1394                 struct list_head *devices;
1395                 struct btrfs_device *tmp;
1396
1397                 device = NULL;
1398                 devices = &root->fs_info->fs_devices->devices;
1399                 /*
1400                  * It is safe to read the devices since the volume_mutex
1401                  * is held.
1402                  */
1403                 list_for_each_entry(tmp, devices, dev_list) {
1404                         if (tmp->in_fs_metadata &&
1405                             !tmp->is_tgtdev_for_dev_replace &&
1406                             !tmp->bdev) {
1407                                 device = tmp;
1408                                 break;
1409                         }
1410                 }
1411                 bdev = NULL;
1412                 bh = NULL;
1413                 disk_super = NULL;
1414                 if (!device) {
1415                         printk(KERN_ERR "btrfs: no missing devices found to "
1416                                "remove\n");
1417                         goto out;
1418                 }
1419         } else {
1420                 ret = btrfs_get_bdev_and_sb(device_path,
1421                                             FMODE_READ | FMODE_EXCL,
1422                                             root->fs_info->bdev_holder, 0,
1423                                             &bdev, &bh);
1424                 if (ret)
1425                         goto out;
1426                 disk_super = (struct btrfs_super_block *)bh->b_data;
1427                 devid = btrfs_stack_device_id(&disk_super->dev_item);
1428                 dev_uuid = disk_super->dev_item.uuid;
1429                 device = btrfs_find_device(root->fs_info, devid, dev_uuid,
1430                                            disk_super->fsid);
1431                 if (!device) {
1432                         ret = -ENOENT;
1433                         goto error_brelse;
1434                 }
1435         }
1436
1437         if (device->is_tgtdev_for_dev_replace) {
1438                 pr_err("btrfs: unable to remove the dev_replace target dev\n");
1439                 ret = -EINVAL;
1440                 goto error_brelse;
1441         }
1442
1443         if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1444                 printk(KERN_ERR "btrfs: unable to remove the only writeable "
1445                        "device\n");
1446                 ret = -EINVAL;
1447                 goto error_brelse;
1448         }
1449
1450         if (device->writeable) {
1451                 lock_chunks(root);
1452                 list_del_init(&device->dev_alloc_list);
1453                 unlock_chunks(root);
1454                 root->fs_info->fs_devices->rw_devices--;
1455                 clear_super = true;
1456         }
1457
1458         ret = btrfs_shrink_device(device, 0);
1459         if (ret)
1460                 goto error_undo;
1461
1462         /*
1463          * TODO: the superblock still includes this device in its num_devices
1464          * counter although write_all_supers() is not locked out. This
1465          * could give a filesystem state which requires a degraded mount.
1466          */
1467         ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1468         if (ret)
1469                 goto error_undo;
1470
1471         spin_lock(&root->fs_info->free_chunk_lock);
1472         root->fs_info->free_chunk_space = device->total_bytes -
1473                 device->bytes_used;
1474         spin_unlock(&root->fs_info->free_chunk_lock);
1475
1476         device->in_fs_metadata = 0;
1477         btrfs_scrub_cancel_dev(root->fs_info, device);
1478
1479         /*
1480          * the device list mutex makes sure that we don't change
1481          * the device list while someone else is writing out all
1482          * the device supers.
1483          */
1484
1485         cur_devices = device->fs_devices;
1486         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1487         list_del_rcu(&device->dev_list);
1488
1489         device->fs_devices->num_devices--;
1490         device->fs_devices->total_devices--;
1491
1492         if (device->missing)
1493                 root->fs_info->fs_devices->missing_devices--;
1494
1495         next_device = list_entry(root->fs_info->fs_devices->devices.next,
1496                                  struct btrfs_device, dev_list);
1497         if (device->bdev == root->fs_info->sb->s_bdev)
1498                 root->fs_info->sb->s_bdev = next_device->bdev;
1499         if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1500                 root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1501
1502         if (device->bdev)
1503                 device->fs_devices->open_devices--;
1504
1505         call_rcu(&device->rcu, free_device);
1506         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1507
1508         num_devices = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
1509         btrfs_set_super_num_devices(root->fs_info->super_copy, num_devices);
1510
1511         if (cur_devices->open_devices == 0) {
1512                 struct btrfs_fs_devices *fs_devices;
1513                 fs_devices = root->fs_info->fs_devices;
1514                 while (fs_devices) {
1515                         if (fs_devices->seed == cur_devices)
1516                                 break;
1517                         fs_devices = fs_devices->seed;
1518                 }
1519                 fs_devices->seed = cur_devices->seed;
1520                 cur_devices->seed = NULL;
1521                 lock_chunks(root);
1522                 __btrfs_close_devices(cur_devices);
1523                 unlock_chunks(root);
1524                 free_fs_devices(cur_devices);
1525         }
1526
1527         root->fs_info->num_tolerated_disk_barrier_failures =
1528                 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
1529
1530         /*
1531          * at this point, the device is zero sized.  We want to
1532          * remove it from the devices list and zero out the old super
1533          */
1534         if (clear_super && disk_super) {
1535                 /* make sure this device isn't detected as part of
1536                  * the FS anymore
1537                  */
1538                 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1539                 set_buffer_dirty(bh);
1540                 sync_dirty_buffer(bh);
1541         }
1542
1543         ret = 0;
1544
1545 error_brelse:
1546         brelse(bh);
1547 error_close:
1548         if (bdev)
1549                 blkdev_put(bdev, FMODE_READ | FMODE_EXCL);
1550 out:
1551         mutex_unlock(&uuid_mutex);
1552         return ret;
1553 error_undo:
1554         if (device->writeable) {
1555                 lock_chunks(root);
1556                 list_add(&device->dev_alloc_list,
1557                          &root->fs_info->fs_devices->alloc_list);
1558                 unlock_chunks(root);
1559                 root->fs_info->fs_devices->rw_devices++;
1560         }
1561         goto error_brelse;
1562 }
1563
1564 void btrfs_rm_dev_replace_srcdev(struct btrfs_fs_info *fs_info,
1565                                  struct btrfs_device *srcdev)
1566 {
1567         WARN_ON(!mutex_is_locked(&fs_info->fs_devices->device_list_mutex));
1568         list_del_rcu(&srcdev->dev_list);
1569         list_del_rcu(&srcdev->dev_alloc_list);
1570         fs_info->fs_devices->num_devices--;
1571         if (srcdev->missing) {
1572                 fs_info->fs_devices->missing_devices--;
1573                 fs_info->fs_devices->rw_devices++;
1574         }
1575         if (srcdev->can_discard)
1576                 fs_info->fs_devices->num_can_discard--;
1577         if (srcdev->bdev)
1578                 fs_info->fs_devices->open_devices--;
1579
1580         call_rcu(&srcdev->rcu, free_device);
1581 }
1582
1583 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
1584                                       struct btrfs_device *tgtdev)
1585 {
1586         struct btrfs_device *next_device;
1587
1588         WARN_ON(!tgtdev);
1589         mutex_lock(&fs_info->fs_devices->device_list_mutex);
1590         if (tgtdev->bdev) {
1591                 btrfs_scratch_superblock(tgtdev);
1592                 fs_info->fs_devices->open_devices--;
1593         }
1594         fs_info->fs_devices->num_devices--;
1595         if (tgtdev->can_discard)
1596                 fs_info->fs_devices->num_can_discard++;
1597
1598         next_device = list_entry(fs_info->fs_devices->devices.next,
1599                                  struct btrfs_device, dev_list);
1600         if (tgtdev->bdev == fs_info->sb->s_bdev)
1601                 fs_info->sb->s_bdev = next_device->bdev;
1602         if (tgtdev->bdev == fs_info->fs_devices->latest_bdev)
1603                 fs_info->fs_devices->latest_bdev = next_device->bdev;
1604         list_del_rcu(&tgtdev->dev_list);
1605
1606         call_rcu(&tgtdev->rcu, free_device);
1607
1608         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
1609 }
1610
1611 int btrfs_find_device_by_path(struct btrfs_root *root, char *device_path,
1612                               struct btrfs_device **device)
1613 {
1614         int ret = 0;
1615         struct btrfs_super_block *disk_super;
1616         u64 devid;
1617         u8 *dev_uuid;
1618         struct block_device *bdev;
1619         struct buffer_head *bh;
1620
1621         *device = NULL;
1622         ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
1623                                     root->fs_info->bdev_holder, 0, &bdev, &bh);
1624         if (ret)
1625                 return ret;
1626         disk_super = (struct btrfs_super_block *)bh->b_data;
1627         devid = btrfs_stack_device_id(&disk_super->dev_item);
1628         dev_uuid = disk_super->dev_item.uuid;
1629         *device = btrfs_find_device(root->fs_info, devid, dev_uuid,
1630                                     disk_super->fsid);
1631         brelse(bh);
1632         if (!*device)
1633                 ret = -ENOENT;
1634         blkdev_put(bdev, FMODE_READ);
1635         return ret;
1636 }
1637
1638 int btrfs_find_device_missing_or_by_path(struct btrfs_root *root,
1639                                          char *device_path,
1640                                          struct btrfs_device **device)
1641 {
1642         *device = NULL;
1643         if (strcmp(device_path, "missing") == 0) {
1644                 struct list_head *devices;
1645                 struct btrfs_device *tmp;
1646
1647                 devices = &root->fs_info->fs_devices->devices;
1648                 /*
1649                  * It is safe to read the devices since the volume_mutex
1650                  * is held by the caller.
1651                  */
1652                 list_for_each_entry(tmp, devices, dev_list) {
1653                         if (tmp->in_fs_metadata && !tmp->bdev) {
1654                                 *device = tmp;
1655                                 break;
1656                         }
1657                 }
1658
1659                 if (!*device) {
1660                         pr_err("btrfs: no missing device found\n");
1661                         return -ENOENT;
1662                 }
1663
1664                 return 0;
1665         } else {
1666                 return btrfs_find_device_by_path(root, device_path, device);
1667         }
1668 }
1669
1670 /*
1671  * does all the dirty work required for changing file system's UUID.
1672  */
1673 static int btrfs_prepare_sprout(struct btrfs_root *root)
1674 {
1675         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
1676         struct btrfs_fs_devices *old_devices;
1677         struct btrfs_fs_devices *seed_devices;
1678         struct btrfs_super_block *disk_super = root->fs_info->super_copy;
1679         struct btrfs_device *device;
1680         u64 super_flags;
1681
1682         BUG_ON(!mutex_is_locked(&uuid_mutex));
1683         if (!fs_devices->seeding)
1684                 return -EINVAL;
1685
1686         seed_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
1687         if (!seed_devices)
1688                 return -ENOMEM;
1689
1690         old_devices = clone_fs_devices(fs_devices);
1691         if (IS_ERR(old_devices)) {
1692                 kfree(seed_devices);
1693                 return PTR_ERR(old_devices);
1694         }
1695
1696         list_add(&old_devices->list, &fs_uuids);
1697
1698         memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
1699         seed_devices->opened = 1;
1700         INIT_LIST_HEAD(&seed_devices->devices);
1701         INIT_LIST_HEAD(&seed_devices->alloc_list);
1702         mutex_init(&seed_devices->device_list_mutex);
1703
1704         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1705         list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
1706                               synchronize_rcu);
1707         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1708
1709         list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
1710         list_for_each_entry(device, &seed_devices->devices, dev_list) {
1711                 device->fs_devices = seed_devices;
1712         }
1713
1714         fs_devices->seeding = 0;
1715         fs_devices->num_devices = 0;
1716         fs_devices->open_devices = 0;
1717         fs_devices->total_devices = 0;
1718         fs_devices->seed = seed_devices;
1719
1720         generate_random_uuid(fs_devices->fsid);
1721         memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1722         memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1723         super_flags = btrfs_super_flags(disk_super) &
1724                       ~BTRFS_SUPER_FLAG_SEEDING;
1725         btrfs_set_super_flags(disk_super, super_flags);
1726
1727         return 0;
1728 }
1729
1730 /*
1731  * strore the expected generation for seed devices in device items.
1732  */
1733 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
1734                                struct btrfs_root *root)
1735 {
1736         struct btrfs_path *path;
1737         struct extent_buffer *leaf;
1738         struct btrfs_dev_item *dev_item;
1739         struct btrfs_device *device;
1740         struct btrfs_key key;
1741         u8 fs_uuid[BTRFS_UUID_SIZE];
1742         u8 dev_uuid[BTRFS_UUID_SIZE];
1743         u64 devid;
1744         int ret;
1745
1746         path = btrfs_alloc_path();
1747         if (!path)
1748                 return -ENOMEM;
1749
1750         root = root->fs_info->chunk_root;
1751         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1752         key.offset = 0;
1753         key.type = BTRFS_DEV_ITEM_KEY;
1754
1755         while (1) {
1756                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1757                 if (ret < 0)
1758                         goto error;
1759
1760                 leaf = path->nodes[0];
1761 next_slot:
1762                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1763                         ret = btrfs_next_leaf(root, path);
1764                         if (ret > 0)
1765                                 break;
1766                         if (ret < 0)
1767                                 goto error;
1768                         leaf = path->nodes[0];
1769                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1770                         btrfs_release_path(path);
1771                         continue;
1772                 }
1773
1774                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1775                 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
1776                     key.type != BTRFS_DEV_ITEM_KEY)
1777                         break;
1778
1779                 dev_item = btrfs_item_ptr(leaf, path->slots[0],
1780                                           struct btrfs_dev_item);
1781                 devid = btrfs_device_id(leaf, dev_item);
1782                 read_extent_buffer(leaf, dev_uuid,
1783                                    (unsigned long)btrfs_device_uuid(dev_item),
1784                                    BTRFS_UUID_SIZE);
1785                 read_extent_buffer(leaf, fs_uuid,
1786                                    (unsigned long)btrfs_device_fsid(dev_item),
1787                                    BTRFS_UUID_SIZE);
1788                 device = btrfs_find_device(root->fs_info, devid, dev_uuid,
1789                                            fs_uuid);
1790                 BUG_ON(!device); /* Logic error */
1791
1792                 if (device->fs_devices->seeding) {
1793                         btrfs_set_device_generation(leaf, dev_item,
1794                                                     device->generation);
1795                         btrfs_mark_buffer_dirty(leaf);
1796                 }
1797
1798                 path->slots[0]++;
1799                 goto next_slot;
1800         }
1801         ret = 0;
1802 error:
1803         btrfs_free_path(path);
1804         return ret;
1805 }
1806
1807 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
1808 {
1809         struct request_queue *q;
1810         struct btrfs_trans_handle *trans;
1811         struct btrfs_device *device;
1812         struct block_device *bdev;
1813         struct list_head *devices;
1814         struct super_block *sb = root->fs_info->sb;
1815         struct rcu_string *name;
1816         u64 total_bytes;
1817         int seeding_dev = 0;
1818         int ret = 0;
1819
1820         if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
1821                 return -EROFS;
1822
1823         bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
1824                                   root->fs_info->bdev_holder);
1825         if (IS_ERR(bdev))
1826                 return PTR_ERR(bdev);
1827
1828         if (root->fs_info->fs_devices->seeding) {
1829                 seeding_dev = 1;
1830                 down_write(&sb->s_umount);
1831                 mutex_lock(&uuid_mutex);
1832         }
1833
1834         filemap_write_and_wait(bdev->bd_inode->i_mapping);
1835
1836         devices = &root->fs_info->fs_devices->devices;
1837
1838         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1839         list_for_each_entry(device, devices, dev_list) {
1840                 if (device->bdev == bdev) {
1841                         ret = -EEXIST;
1842                         mutex_unlock(
1843                                 &root->fs_info->fs_devices->device_list_mutex);
1844                         goto error;
1845                 }
1846         }
1847         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1848
1849         device = kzalloc(sizeof(*device), GFP_NOFS);
1850         if (!device) {
1851                 /* we can safely leave the fs_devices entry around */
1852                 ret = -ENOMEM;
1853                 goto error;
1854         }
1855
1856         name = rcu_string_strdup(device_path, GFP_NOFS);
1857         if (!name) {
1858                 kfree(device);
1859                 ret = -ENOMEM;
1860                 goto error;
1861         }
1862         rcu_assign_pointer(device->name, name);
1863
1864         ret = find_next_devid(root, &device->devid);
1865         if (ret) {
1866                 rcu_string_free(device->name);
1867                 kfree(device);
1868                 goto error;
1869         }
1870
1871         trans = btrfs_start_transaction(root, 0);
1872         if (IS_ERR(trans)) {
1873                 rcu_string_free(device->name);
1874                 kfree(device);
1875                 ret = PTR_ERR(trans);
1876                 goto error;
1877         }
1878
1879         lock_chunks(root);
1880
1881         q = bdev_get_queue(bdev);
1882         if (blk_queue_discard(q))
1883                 device->can_discard = 1;
1884         device->writeable = 1;
1885         device->work.func = pending_bios_fn;
1886         generate_random_uuid(device->uuid);
1887         spin_lock_init(&device->io_lock);
1888         device->generation = trans->transid;
1889         device->io_width = root->sectorsize;
1890         device->io_align = root->sectorsize;
1891         device->sector_size = root->sectorsize;
1892         device->total_bytes = i_size_read(bdev->bd_inode);
1893         device->disk_total_bytes = device->total_bytes;
1894         device->dev_root = root->fs_info->dev_root;
1895         device->bdev = bdev;
1896         device->in_fs_metadata = 1;
1897         device->is_tgtdev_for_dev_replace = 0;
1898         device->mode = FMODE_EXCL;
1899         set_blocksize(device->bdev, 4096);
1900
1901         if (seeding_dev) {
1902                 sb->s_flags &= ~MS_RDONLY;
1903                 ret = btrfs_prepare_sprout(root);
1904                 BUG_ON(ret); /* -ENOMEM */
1905         }
1906
1907         device->fs_devices = root->fs_info->fs_devices;
1908
1909         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1910         list_add_rcu(&device->dev_list, &root->fs_info->fs_devices->devices);
1911         list_add(&device->dev_alloc_list,
1912                  &root->fs_info->fs_devices->alloc_list);
1913         root->fs_info->fs_devices->num_devices++;
1914         root->fs_info->fs_devices->open_devices++;
1915         root->fs_info->fs_devices->rw_devices++;
1916         root->fs_info->fs_devices->total_devices++;
1917         if (device->can_discard)
1918                 root->fs_info->fs_devices->num_can_discard++;
1919         root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
1920
1921         spin_lock(&root->fs_info->free_chunk_lock);
1922         root->fs_info->free_chunk_space += device->total_bytes;
1923         spin_unlock(&root->fs_info->free_chunk_lock);
1924
1925         if (!blk_queue_nonrot(bdev_get_queue(bdev)))
1926                 root->fs_info->fs_devices->rotating = 1;
1927
1928         total_bytes = btrfs_super_total_bytes(root->fs_info->super_copy);
1929         btrfs_set_super_total_bytes(root->fs_info->super_copy,
1930                                     total_bytes + device->total_bytes);
1931
1932         total_bytes = btrfs_super_num_devices(root->fs_info->super_copy);
1933         btrfs_set_super_num_devices(root->fs_info->super_copy,
1934                                     total_bytes + 1);
1935         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1936
1937         if (seeding_dev) {
1938                 ret = init_first_rw_device(trans, root, device);
1939                 if (ret) {
1940                         btrfs_abort_transaction(trans, root, ret);
1941                         goto error_trans;
1942                 }
1943                 ret = btrfs_finish_sprout(trans, root);
1944                 if (ret) {
1945                         btrfs_abort_transaction(trans, root, ret);
1946                         goto error_trans;
1947                 }
1948         } else {
1949                 ret = btrfs_add_device(trans, root, device);
1950                 if (ret) {
1951                         btrfs_abort_transaction(trans, root, ret);
1952                         goto error_trans;
1953                 }
1954         }
1955
1956         /*
1957          * we've got more storage, clear any full flags on the space
1958          * infos
1959          */
1960         btrfs_clear_space_info_full(root->fs_info);
1961
1962         unlock_chunks(root);
1963         root->fs_info->num_tolerated_disk_barrier_failures =
1964                 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
1965         ret = btrfs_commit_transaction(trans, root);
1966
1967         if (seeding_dev) {
1968                 mutex_unlock(&uuid_mutex);
1969                 up_write(&sb->s_umount);
1970
1971                 if (ret) /* transaction commit */
1972                         return ret;
1973
1974                 ret = btrfs_relocate_sys_chunks(root);
1975                 if (ret < 0)
1976                         btrfs_error(root->fs_info, ret,
1977                                     "Failed to relocate sys chunks after "
1978                                     "device initialization. This can be fixed "
1979                                     "using the \"btrfs balance\" command.");
1980                 trans = btrfs_attach_transaction(root);
1981                 if (IS_ERR(trans)) {
1982                         if (PTR_ERR(trans) == -ENOENT)
1983                                 return 0;
1984                         return PTR_ERR(trans);
1985                 }
1986                 ret = btrfs_commit_transaction(trans, root);
1987         }
1988
1989         return ret;
1990
1991 error_trans:
1992         unlock_chunks(root);
1993         btrfs_end_transaction(trans, root);
1994         rcu_string_free(device->name);
1995         kfree(device);
1996 error:
1997         blkdev_put(bdev, FMODE_EXCL);
1998         if (seeding_dev) {
1999                 mutex_unlock(&uuid_mutex);
2000                 up_write(&sb->s_umount);
2001         }
2002         return ret;
2003 }
2004
2005 int btrfs_init_dev_replace_tgtdev(struct btrfs_root *root, char *device_path,
2006                                   struct btrfs_device **device_out)
2007 {
2008         struct request_queue *q;
2009         struct btrfs_device *device;
2010         struct block_device *bdev;
2011         struct btrfs_fs_info *fs_info = root->fs_info;
2012         struct list_head *devices;
2013         struct rcu_string *name;
2014         int ret = 0;
2015
2016         *device_out = NULL;
2017         if (fs_info->fs_devices->seeding)
2018                 return -EINVAL;
2019
2020         bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2021                                   fs_info->bdev_holder);
2022         if (IS_ERR(bdev))
2023                 return PTR_ERR(bdev);
2024
2025         filemap_write_and_wait(bdev->bd_inode->i_mapping);
2026
2027         devices = &fs_info->fs_devices->devices;
2028         list_for_each_entry(device, devices, dev_list) {
2029                 if (device->bdev == bdev) {
2030                         ret = -EEXIST;
2031                         goto error;
2032                 }
2033         }
2034
2035         device = kzalloc(sizeof(*device), GFP_NOFS);
2036         if (!device) {
2037                 ret = -ENOMEM;
2038                 goto error;
2039         }
2040
2041         name = rcu_string_strdup(device_path, GFP_NOFS);
2042         if (!name) {
2043                 kfree(device);
2044                 ret = -ENOMEM;
2045                 goto error;
2046         }
2047         rcu_assign_pointer(device->name, name);
2048
2049         q = bdev_get_queue(bdev);
2050         if (blk_queue_discard(q))
2051                 device->can_discard = 1;
2052         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2053         device->writeable = 1;
2054         device->work.func = pending_bios_fn;
2055         generate_random_uuid(device->uuid);
2056         device->devid = BTRFS_DEV_REPLACE_DEVID;
2057         spin_lock_init(&device->io_lock);
2058         device->generation = 0;
2059         device->io_width = root->sectorsize;
2060         device->io_align = root->sectorsize;
2061         device->sector_size = root->sectorsize;
2062         device->total_bytes = i_size_read(bdev->bd_inode);
2063         device->disk_total_bytes = device->total_bytes;
2064         device->dev_root = fs_info->dev_root;
2065         device->bdev = bdev;
2066         device->in_fs_metadata = 1;
2067         device->is_tgtdev_for_dev_replace = 1;
2068         device->mode = FMODE_EXCL;
2069         set_blocksize(device->bdev, 4096);
2070         device->fs_devices = fs_info->fs_devices;
2071         list_add(&device->dev_list, &fs_info->fs_devices->devices);
2072         fs_info->fs_devices->num_devices++;
2073         fs_info->fs_devices->open_devices++;
2074         if (device->can_discard)
2075                 fs_info->fs_devices->num_can_discard++;
2076         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2077
2078         *device_out = device;
2079         return ret;
2080
2081 error:
2082         blkdev_put(bdev, FMODE_EXCL);
2083         return ret;
2084 }
2085
2086 void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info *fs_info,
2087                                               struct btrfs_device *tgtdev)
2088 {
2089         WARN_ON(fs_info->fs_devices->rw_devices == 0);
2090         tgtdev->io_width = fs_info->dev_root->sectorsize;
2091         tgtdev->io_align = fs_info->dev_root->sectorsize;
2092         tgtdev->sector_size = fs_info->dev_root->sectorsize;
2093         tgtdev->dev_root = fs_info->dev_root;
2094         tgtdev->in_fs_metadata = 1;
2095 }
2096
2097 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2098                                         struct btrfs_device *device)
2099 {
2100         int ret;
2101         struct btrfs_path *path;
2102         struct btrfs_root *root;
2103         struct btrfs_dev_item *dev_item;
2104         struct extent_buffer *leaf;
2105         struct btrfs_key key;
2106
2107         root = device->dev_root->fs_info->chunk_root;
2108
2109         path = btrfs_alloc_path();
2110         if (!path)
2111                 return -ENOMEM;
2112
2113         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2114         key.type = BTRFS_DEV_ITEM_KEY;
2115         key.offset = device->devid;
2116
2117         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2118         if (ret < 0)
2119                 goto out;
2120
2121         if (ret > 0) {
2122                 ret = -ENOENT;
2123                 goto out;
2124         }
2125
2126         leaf = path->nodes[0];
2127         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2128
2129         btrfs_set_device_id(leaf, dev_item, device->devid);
2130         btrfs_set_device_type(leaf, dev_item, device->type);
2131         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2132         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2133         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2134         btrfs_set_device_total_bytes(leaf, dev_item, device->disk_total_bytes);
2135         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
2136         btrfs_mark_buffer_dirty(leaf);
2137
2138 out:
2139         btrfs_free_path(path);
2140         return ret;
2141 }
2142
2143 static int __btrfs_grow_device(struct btrfs_trans_handle *trans,
2144                       struct btrfs_device *device, u64 new_size)
2145 {
2146         struct btrfs_super_block *super_copy =
2147                 device->dev_root->fs_info->super_copy;
2148         u64 old_total = btrfs_super_total_bytes(super_copy);
2149         u64 diff = new_size - device->total_bytes;
2150
2151         if (!device->writeable)
2152                 return -EACCES;
2153         if (new_size <= device->total_bytes ||
2154             device->is_tgtdev_for_dev_replace)
2155                 return -EINVAL;
2156
2157         btrfs_set_super_total_bytes(super_copy, old_total + diff);
2158         device->fs_devices->total_rw_bytes += diff;
2159
2160         device->total_bytes = new_size;
2161         device->disk_total_bytes = new_size;
2162         btrfs_clear_space_info_full(device->dev_root->fs_info);
2163
2164         return btrfs_update_device(trans, device);
2165 }
2166
2167 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2168                       struct btrfs_device *device, u64 new_size)
2169 {
2170         int ret;
2171         lock_chunks(device->dev_root);
2172         ret = __btrfs_grow_device(trans, device, new_size);
2173         unlock_chunks(device->dev_root);
2174         return ret;
2175 }
2176
2177 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
2178                             struct btrfs_root *root,
2179                             u64 chunk_tree, u64 chunk_objectid,
2180                             u64 chunk_offset)
2181 {
2182         int ret;
2183         struct btrfs_path *path;
2184         struct btrfs_key key;
2185
2186         root = root->fs_info->chunk_root;
2187         path = btrfs_alloc_path();
2188         if (!path)
2189                 return -ENOMEM;
2190
2191         key.objectid = chunk_objectid;
2192         key.offset = chunk_offset;
2193         key.type = BTRFS_CHUNK_ITEM_KEY;
2194
2195         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2196         if (ret < 0)
2197                 goto out;
2198         else if (ret > 0) { /* Logic error or corruption */
2199                 btrfs_error(root->fs_info, -ENOENT,
2200                             "Failed lookup while freeing chunk.");
2201                 ret = -ENOENT;
2202                 goto out;
2203         }
2204
2205         ret = btrfs_del_item(trans, root, path);
2206         if (ret < 0)
2207                 btrfs_error(root->fs_info, ret,
2208                             "Failed to delete chunk item.");
2209 out:
2210         btrfs_free_path(path);
2211         return ret;
2212 }
2213
2214 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
2215                         chunk_offset)
2216 {
2217         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
2218         struct btrfs_disk_key *disk_key;
2219         struct btrfs_chunk *chunk;
2220         u8 *ptr;
2221         int ret = 0;
2222         u32 num_stripes;
2223         u32 array_size;
2224         u32 len = 0;
2225         u32 cur;
2226         struct btrfs_key key;
2227
2228         array_size = btrfs_super_sys_array_size(super_copy);
2229
2230         ptr = super_copy->sys_chunk_array;
2231         cur = 0;
2232
2233         while (cur < array_size) {
2234                 disk_key = (struct btrfs_disk_key *)ptr;
2235                 btrfs_disk_key_to_cpu(&key, disk_key);
2236
2237                 len = sizeof(*disk_key);
2238
2239                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2240                         chunk = (struct btrfs_chunk *)(ptr + len);
2241                         num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2242                         len += btrfs_chunk_item_size(num_stripes);
2243                 } else {
2244                         ret = -EIO;
2245                         break;
2246                 }
2247                 if (key.objectid == chunk_objectid &&
2248                     key.offset == chunk_offset) {
2249                         memmove(ptr, ptr + len, array_size - (cur + len));
2250                         array_size -= len;
2251                         btrfs_set_super_sys_array_size(super_copy, array_size);
2252                 } else {
2253                         ptr += len;
2254                         cur += len;
2255                 }
2256         }
2257         return ret;
2258 }
2259
2260 static int btrfs_relocate_chunk(struct btrfs_root *root,
2261                          u64 chunk_tree, u64 chunk_objectid,
2262                          u64 chunk_offset)
2263 {
2264         struct extent_map_tree *em_tree;
2265         struct btrfs_root *extent_root;
2266         struct btrfs_trans_handle *trans;
2267         struct extent_map *em;
2268         struct map_lookup *map;
2269         int ret;
2270         int i;
2271
2272         root = root->fs_info->chunk_root;
2273         extent_root = root->fs_info->extent_root;
2274         em_tree = &root->fs_info->mapping_tree.map_tree;
2275
2276         ret = btrfs_can_relocate(extent_root, chunk_offset);
2277         if (ret)
2278                 return -ENOSPC;
2279
2280         /* step one, relocate all the extents inside this chunk */
2281         ret = btrfs_relocate_block_group(extent_root, chunk_offset);
2282         if (ret)
2283                 return ret;
2284
2285         trans = btrfs_start_transaction(root, 0);
2286         BUG_ON(IS_ERR(trans));
2287
2288         lock_chunks(root);
2289
2290         /*
2291          * step two, delete the device extents and the
2292          * chunk tree entries
2293          */
2294         read_lock(&em_tree->lock);
2295         em = lookup_extent_mapping(em_tree, chunk_offset, 1);
2296         read_unlock(&em_tree->lock);
2297
2298         BUG_ON(!em || em->start > chunk_offset ||
2299                em->start + em->len < chunk_offset);
2300         map = (struct map_lookup *)em->bdev;
2301
2302         for (i = 0; i < map->num_stripes; i++) {
2303                 ret = btrfs_free_dev_extent(trans, map->stripes[i].dev,
2304                                             map->stripes[i].physical);
2305                 BUG_ON(ret);
2306
2307                 if (map->stripes[i].dev) {
2308                         ret = btrfs_update_device(trans, map->stripes[i].dev);
2309                         BUG_ON(ret);
2310                 }
2311         }
2312         ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
2313                                chunk_offset);
2314
2315         BUG_ON(ret);
2316
2317         trace_btrfs_chunk_free(root, map, chunk_offset, em->len);
2318
2319         if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2320                 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
2321                 BUG_ON(ret);
2322         }
2323
2324         ret = btrfs_remove_block_group(trans, extent_root, chunk_offset);
2325         BUG_ON(ret);
2326
2327         write_lock(&em_tree->lock);
2328         remove_extent_mapping(em_tree, em);
2329         write_unlock(&em_tree->lock);
2330
2331         kfree(map);
2332         em->bdev = NULL;
2333
2334         /* once for the tree */
2335         free_extent_map(em);
2336         /* once for us */
2337         free_extent_map(em);
2338
2339         unlock_chunks(root);
2340         btrfs_end_transaction(trans, root);
2341         return 0;
2342 }
2343
2344 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
2345 {
2346         struct btrfs_root *chunk_root = root->fs_info->chunk_root;
2347         struct btrfs_path *path;
2348         struct extent_buffer *leaf;
2349         struct btrfs_chunk *chunk;
2350         struct btrfs_key key;
2351         struct btrfs_key found_key;
2352         u64 chunk_tree = chunk_root->root_key.objectid;
2353         u64 chunk_type;
2354         bool retried = false;
2355         int failed = 0;
2356         int ret;
2357
2358         path = btrfs_alloc_path();
2359         if (!path)
2360                 return -ENOMEM;
2361
2362 again:
2363         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2364         key.offset = (u64)-1;
2365         key.type = BTRFS_CHUNK_ITEM_KEY;
2366
2367         while (1) {
2368                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2369                 if (ret < 0)
2370                         goto error;
2371                 BUG_ON(ret == 0); /* Corruption */
2372
2373                 ret = btrfs_previous_item(chunk_root, path, key.objectid,
2374                                           key.type);
2375                 if (ret < 0)
2376                         goto error;
2377                 if (ret > 0)
2378                         break;
2379
2380                 leaf = path->nodes[0];
2381                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2382
2383                 chunk = btrfs_item_ptr(leaf, path->slots[0],
2384                                        struct btrfs_chunk);
2385                 chunk_type = btrfs_chunk_type(leaf, chunk);
2386                 btrfs_release_path(path);
2387
2388                 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
2389                         ret = btrfs_relocate_chunk(chunk_root, chunk_tree,
2390                                                    found_key.objectid,
2391                                                    found_key.offset);
2392                         if (ret == -ENOSPC)
2393                                 failed++;
2394                         else if (ret)
2395                                 BUG();
2396                 }
2397
2398                 if (found_key.offset == 0)
2399                         break;
2400                 key.offset = found_key.offset - 1;
2401         }
2402         ret = 0;
2403         if (failed && !retried) {
2404                 failed = 0;
2405                 retried = true;
2406                 goto again;
2407         } else if (failed && retried) {
2408                 WARN_ON(1);
2409                 ret = -ENOSPC;
2410         }
2411 error:
2412         btrfs_free_path(path);
2413         return ret;
2414 }
2415
2416 static int insert_balance_item(struct btrfs_root *root,
2417                                struct btrfs_balance_control *bctl)
2418 {
2419         struct btrfs_trans_handle *trans;
2420         struct btrfs_balance_item *item;
2421         struct btrfs_disk_balance_args disk_bargs;
2422         struct btrfs_path *path;
2423         struct extent_buffer *leaf;
2424         struct btrfs_key key;
2425         int ret, err;
2426
2427         path = btrfs_alloc_path();
2428         if (!path)
2429                 return -ENOMEM;
2430
2431         trans = btrfs_start_transaction(root, 0);
2432         if (IS_ERR(trans)) {
2433                 btrfs_free_path(path);
2434                 return PTR_ERR(trans);
2435         }
2436
2437         key.objectid = BTRFS_BALANCE_OBJECTID;
2438         key.type = BTRFS_BALANCE_ITEM_KEY;
2439         key.offset = 0;
2440
2441         ret = btrfs_insert_empty_item(trans, root, path, &key,
2442                                       sizeof(*item));
2443         if (ret)
2444                 goto out;
2445
2446         leaf = path->nodes[0];
2447         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
2448
2449         memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
2450
2451         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
2452         btrfs_set_balance_data(leaf, item, &disk_bargs);
2453         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
2454         btrfs_set_balance_meta(leaf, item, &disk_bargs);
2455         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
2456         btrfs_set_balance_sys(leaf, item, &disk_bargs);
2457
2458         btrfs_set_balance_flags(leaf, item, bctl->flags);
2459
2460         btrfs_mark_buffer_dirty(leaf);
2461 out:
2462         btrfs_free_path(path);
2463         err = btrfs_commit_transaction(trans, root);
2464         if (err && !ret)
2465                 ret = err;
2466         return ret;
2467 }
2468
2469 static int del_balance_item(struct btrfs_root *root)
2470 {
2471         struct btrfs_trans_handle *trans;
2472         struct btrfs_path *path;
2473         struct btrfs_key key;
2474         int ret, err;
2475
2476         path = btrfs_alloc_path();
2477         if (!path)
2478                 return -ENOMEM;
2479
2480         trans = btrfs_start_transaction(root, 0);
2481         if (IS_ERR(trans)) {
2482                 btrfs_free_path(path);
2483                 return PTR_ERR(trans);
2484         }
2485
2486         key.objectid = BTRFS_BALANCE_OBJECTID;
2487         key.type = BTRFS_BALANCE_ITEM_KEY;
2488         key.offset = 0;
2489
2490         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2491         if (ret < 0)
2492                 goto out;
2493         if (ret > 0) {
2494                 ret = -ENOENT;
2495                 goto out;
2496         }
2497
2498         ret = btrfs_del_item(trans, root, path);
2499 out:
2500         btrfs_free_path(path);
2501         err = btrfs_commit_transaction(trans, root);
2502         if (err && !ret)
2503                 ret = err;
2504         return ret;
2505 }
2506
2507 /*
2508  * This is a heuristic used to reduce the number of chunks balanced on
2509  * resume after balance was interrupted.
2510  */
2511 static void update_balance_args(struct btrfs_balance_control *bctl)
2512 {
2513         /*
2514          * Turn on soft mode for chunk types that were being converted.
2515          */
2516         if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
2517                 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
2518         if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
2519                 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
2520         if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
2521                 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
2522
2523         /*
2524          * Turn on usage filter if is not already used.  The idea is
2525          * that chunks that we have already balanced should be
2526          * reasonably full.  Don't do it for chunks that are being
2527          * converted - that will keep us from relocating unconverted
2528          * (albeit full) chunks.
2529          */
2530         if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2531             !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2532                 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
2533                 bctl->data.usage = 90;
2534         }
2535         if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2536             !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2537                 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
2538                 bctl->sys.usage = 90;
2539         }
2540         if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2541             !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2542                 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
2543                 bctl->meta.usage = 90;
2544         }
2545 }
2546
2547 /*
2548  * Should be called with both balance and volume mutexes held to
2549  * serialize other volume operations (add_dev/rm_dev/resize) with
2550  * restriper.  Same goes for unset_balance_control.
2551  */
2552 static void set_balance_control(struct btrfs_balance_control *bctl)
2553 {
2554         struct btrfs_fs_info *fs_info = bctl->fs_info;
2555
2556         BUG_ON(fs_info->balance_ctl);
2557
2558         spin_lock(&fs_info->balance_lock);
2559         fs_info->balance_ctl = bctl;
2560         spin_unlock(&fs_info->balance_lock);
2561 }
2562
2563 static void unset_balance_control(struct btrfs_fs_info *fs_info)
2564 {
2565         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
2566
2567         BUG_ON(!fs_info->balance_ctl);
2568
2569         spin_lock(&fs_info->balance_lock);
2570         fs_info->balance_ctl = NULL;
2571         spin_unlock(&fs_info->balance_lock);
2572
2573         kfree(bctl);
2574 }
2575
2576 /*
2577  * Balance filters.  Return 1 if chunk should be filtered out
2578  * (should not be balanced).
2579  */
2580 static int chunk_profiles_filter(u64 chunk_type,
2581                                  struct btrfs_balance_args *bargs)
2582 {
2583         chunk_type = chunk_to_extended(chunk_type) &
2584                                 BTRFS_EXTENDED_PROFILE_MASK;
2585
2586         if (bargs->profiles & chunk_type)
2587                 return 0;
2588
2589         return 1;
2590 }
2591
2592 static int chunk_usage_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
2593                               struct btrfs_balance_args *bargs)
2594 {
2595         struct btrfs_block_group_cache *cache;
2596         u64 chunk_used, user_thresh;
2597         int ret = 1;
2598
2599         cache = btrfs_lookup_block_group(fs_info, chunk_offset);
2600         chunk_used = btrfs_block_group_used(&cache->item);
2601
2602         user_thresh = div_factor_fine(cache->key.offset, bargs->usage);
2603         if (chunk_used < user_thresh)
2604                 ret = 0;
2605
2606         btrfs_put_block_group(cache);
2607         return ret;
2608 }
2609
2610 static int chunk_devid_filter(struct extent_buffer *leaf,
2611                               struct btrfs_chunk *chunk,
2612                               struct btrfs_balance_args *bargs)
2613 {
2614         struct btrfs_stripe *stripe;
2615         int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2616         int i;
2617
2618         for (i = 0; i < num_stripes; i++) {
2619                 stripe = btrfs_stripe_nr(chunk, i);
2620                 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
2621                         return 0;
2622         }
2623
2624         return 1;
2625 }
2626
2627 /* [pstart, pend) */
2628 static int chunk_drange_filter(struct extent_buffer *leaf,
2629                                struct btrfs_chunk *chunk,
2630                                u64 chunk_offset,
2631                                struct btrfs_balance_args *bargs)
2632 {
2633         struct btrfs_stripe *stripe;
2634         int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2635         u64 stripe_offset;
2636         u64 stripe_length;
2637         int factor;
2638         int i;
2639
2640         if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
2641                 return 0;
2642
2643         if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP |
2644              BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10))
2645                 factor = 2;
2646         else
2647                 factor = 1;
2648         factor = num_stripes / factor;
2649
2650         for (i = 0; i < num_stripes; i++) {
2651                 stripe = btrfs_stripe_nr(chunk, i);
2652                 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
2653                         continue;
2654
2655                 stripe_offset = btrfs_stripe_offset(leaf, stripe);
2656                 stripe_length = btrfs_chunk_length(leaf, chunk);
2657                 do_div(stripe_length, factor);
2658
2659                 if (stripe_offset < bargs->pend &&
2660                     stripe_offset + stripe_length > bargs->pstart)
2661                         return 0;
2662         }
2663
2664         return 1;
2665 }
2666
2667 /* [vstart, vend) */
2668 static int chunk_vrange_filter(struct extent_buffer *leaf,
2669                                struct btrfs_chunk *chunk,
2670                                u64 chunk_offset,
2671                                struct btrfs_balance_args *bargs)
2672 {
2673         if (chunk_offset < bargs->vend &&
2674             chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
2675                 /* at least part of the chunk is inside this vrange */
2676                 return 0;
2677
2678         return 1;
2679 }
2680
2681 static int chunk_soft_convert_filter(u64 chunk_type,
2682                                      struct btrfs_balance_args *bargs)
2683 {
2684         if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
2685                 return 0;
2686
2687         chunk_type = chunk_to_extended(chunk_type) &
2688                                 BTRFS_EXTENDED_PROFILE_MASK;
2689
2690         if (bargs->target == chunk_type)
2691                 return 1;
2692
2693         return 0;
2694 }
2695
2696 static int should_balance_chunk(struct btrfs_root *root,
2697                                 struct extent_buffer *leaf,
2698                                 struct btrfs_chunk *chunk, u64 chunk_offset)
2699 {
2700         struct btrfs_balance_control *bctl = root->fs_info->balance_ctl;
2701         struct btrfs_balance_args *bargs = NULL;
2702         u64 chunk_type = btrfs_chunk_type(leaf, chunk);
2703
2704         /* type filter */
2705         if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
2706               (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
2707                 return 0;
2708         }
2709
2710         if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
2711                 bargs = &bctl->data;
2712         else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
2713                 bargs = &bctl->sys;
2714         else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
2715                 bargs = &bctl->meta;
2716
2717         /* profiles filter */
2718         if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
2719             chunk_profiles_filter(chunk_type, bargs)) {
2720                 return 0;
2721         }
2722
2723         /* usage filter */
2724         if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
2725             chunk_usage_filter(bctl->fs_info, chunk_offset, bargs)) {
2726                 return 0;
2727         }
2728
2729         /* devid filter */
2730         if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
2731             chunk_devid_filter(leaf, chunk, bargs)) {
2732                 return 0;
2733         }
2734
2735         /* drange filter, makes sense only with devid filter */
2736         if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
2737             chunk_drange_filter(leaf, chunk, chunk_offset, bargs)) {
2738                 return 0;
2739         }
2740
2741         /* vrange filter */
2742         if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
2743             chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
2744                 return 0;
2745         }
2746
2747         /* soft profile changing mode */
2748         if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
2749             chunk_soft_convert_filter(chunk_type, bargs)) {
2750                 return 0;
2751         }
2752
2753         return 1;
2754 }
2755
2756 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
2757 {
2758         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
2759         struct btrfs_root *chunk_root = fs_info->chunk_root;
2760         struct btrfs_root *dev_root = fs_info->dev_root;
2761         struct list_head *devices;
2762         struct btrfs_device *device;
2763         u64 old_size;
2764         u64 size_to_free;
2765         struct btrfs_chunk *chunk;
2766         struct btrfs_path *path;
2767         struct btrfs_key key;
2768         struct btrfs_key found_key;
2769         struct btrfs_trans_handle *trans;
2770         struct extent_buffer *leaf;
2771         int slot;
2772         int ret;
2773         int enospc_errors = 0;
2774         bool counting = true;
2775
2776         /* step one make some room on all the devices */
2777         devices = &fs_info->fs_devices->devices;
2778         list_for_each_entry(device, devices, dev_list) {
2779                 old_size = device->total_bytes;
2780                 size_to_free = div_factor(old_size, 1);
2781                 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
2782                 if (!device->writeable ||
2783                     device->total_bytes - device->bytes_used > size_to_free ||
2784                     device->is_tgtdev_for_dev_replace)
2785                         continue;
2786
2787                 ret = btrfs_shrink_device(device, old_size - size_to_free);
2788                 if (ret == -ENOSPC)
2789                         break;
2790                 BUG_ON(ret);
2791
2792                 trans = btrfs_start_transaction(dev_root, 0);
2793                 BUG_ON(IS_ERR(trans));
2794
2795                 ret = btrfs_grow_device(trans, device, old_size);
2796                 BUG_ON(ret);
2797
2798                 btrfs_end_transaction(trans, dev_root);
2799         }
2800
2801         /* step two, relocate all the chunks */
2802         path = btrfs_alloc_path();
2803         if (!path) {
2804                 ret = -ENOMEM;
2805                 goto error;
2806         }
2807
2808         /* zero out stat counters */
2809         spin_lock(&fs_info->balance_lock);
2810         memset(&bctl->stat, 0, sizeof(bctl->stat));
2811         spin_unlock(&fs_info->balance_lock);
2812 again:
2813         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2814         key.offset = (u64)-1;
2815         key.type = BTRFS_CHUNK_ITEM_KEY;
2816
2817         while (1) {
2818                 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
2819                     atomic_read(&fs_info->balance_cancel_req)) {
2820                         ret = -ECANCELED;
2821                         goto error;
2822                 }
2823
2824                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2825                 if (ret < 0)
2826                         goto error;
2827
2828                 /*
2829                  * this shouldn't happen, it means the last relocate
2830                  * failed
2831                  */
2832                 if (ret == 0)
2833                         BUG(); /* FIXME break ? */
2834
2835                 ret = btrfs_previous_item(chunk_root, path, 0,
2836                                           BTRFS_CHUNK_ITEM_KEY);
2837                 if (ret) {
2838                         ret = 0;
2839                         break;
2840                 }
2841
2842                 leaf = path->nodes[0];
2843                 slot = path->slots[0];
2844                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2845
2846                 if (found_key.objectid != key.objectid)
2847                         break;
2848
2849                 /* chunk zero is special */
2850                 if (found_key.offset == 0)
2851                         break;
2852
2853                 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
2854
2855                 if (!counting) {
2856                         spin_lock(&fs_info->balance_lock);
2857                         bctl->stat.considered++;
2858                         spin_unlock(&fs_info->balance_lock);
2859                 }
2860
2861                 ret = should_balance_chunk(chunk_root, leaf, chunk,
2862                                            found_key.offset);
2863                 btrfs_release_path(path);
2864                 if (!ret)
2865                         goto loop;
2866
2867                 if (counting) {
2868                         spin_lock(&fs_info->balance_lock);
2869                         bctl->stat.expected++;
2870                         spin_unlock(&fs_info->balance_lock);
2871                         goto loop;
2872                 }
2873
2874                 ret = btrfs_relocate_chunk(chunk_root,
2875                                            chunk_root->root_key.objectid,
2876                                            found_key.objectid,
2877                                            found_key.offset);
2878                 if (ret && ret != -ENOSPC)
2879                         goto error;
2880                 if (ret == -ENOSPC) {
2881                         enospc_errors++;
2882                 } else {
2883                         spin_lock(&fs_info->balance_lock);
2884                         bctl->stat.completed++;
2885                         spin_unlock(&fs_info->balance_lock);
2886                 }
2887 loop:
2888                 key.offset = found_key.offset - 1;
2889         }
2890
2891         if (counting) {
2892                 btrfs_release_path(path);
2893                 counting = false;
2894                 goto again;
2895         }
2896 error:
2897         btrfs_free_path(path);
2898         if (enospc_errors) {
2899                 printk(KERN_INFO "btrfs: %d enospc errors during balance\n",
2900                        enospc_errors);
2901                 if (!ret)
2902                         ret = -ENOSPC;
2903         }
2904
2905         return ret;
2906 }
2907
2908 /**
2909  * alloc_profile_is_valid - see if a given profile is valid and reduced
2910  * @flags: profile to validate
2911  * @extended: if true @flags is treated as an extended profile
2912  */
2913 static int alloc_profile_is_valid(u64 flags, int extended)
2914 {
2915         u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
2916                                BTRFS_BLOCK_GROUP_PROFILE_MASK);
2917
2918         flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
2919
2920         /* 1) check that all other bits are zeroed */
2921         if (flags & ~mask)
2922                 return 0;
2923
2924         /* 2) see if profile is reduced */
2925         if (flags == 0)
2926                 return !extended; /* "0" is valid for usual profiles */
2927
2928         /* true if exactly one bit set */
2929         return (flags & (flags - 1)) == 0;
2930 }
2931
2932 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
2933 {
2934         /* cancel requested || normal exit path */
2935         return atomic_read(&fs_info->balance_cancel_req) ||
2936                 (atomic_read(&fs_info->balance_pause_req) == 0 &&
2937                  atomic_read(&fs_info->balance_cancel_req) == 0);
2938 }
2939
2940 static void __cancel_balance(struct btrfs_fs_info *fs_info)
2941 {
2942         int ret;
2943
2944         unset_balance_control(fs_info);
2945         ret = del_balance_item(fs_info->tree_root);
2946         BUG_ON(ret);
2947 }
2948
2949 void update_ioctl_balance_args(struct btrfs_fs_info *fs_info, int lock,
2950                                struct btrfs_ioctl_balance_args *bargs);
2951
2952 /*
2953  * Should be called with both balance and volume mutexes held
2954  */
2955 int btrfs_balance(struct btrfs_balance_control *bctl,
2956                   struct btrfs_ioctl_balance_args *bargs)
2957 {
2958         struct btrfs_fs_info *fs_info = bctl->fs_info;
2959         u64 allowed;
2960         int mixed = 0;
2961         int ret;
2962         u64 num_devices;
2963
2964         if (btrfs_fs_closing(fs_info) ||
2965             atomic_read(&fs_info->balance_pause_req) ||
2966             atomic_read(&fs_info->balance_cancel_req)) {
2967                 ret = -EINVAL;
2968                 goto out;
2969         }
2970
2971         allowed = btrfs_super_incompat_flags(fs_info->super_copy);
2972         if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
2973                 mixed = 1;
2974
2975         /*
2976          * In case of mixed groups both data and meta should be picked,
2977          * and identical options should be given for both of them.
2978          */
2979         allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
2980         if (mixed && (bctl->flags & allowed)) {
2981                 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
2982                     !(bctl->flags & BTRFS_BALANCE_METADATA) ||
2983                     memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
2984                         printk(KERN_ERR "btrfs: with mixed groups data and "
2985                                "metadata balance options must be the same\n");
2986                         ret = -EINVAL;
2987                         goto out;
2988                 }
2989         }
2990
2991         num_devices = fs_info->fs_devices->num_devices;
2992         btrfs_dev_replace_lock(&fs_info->dev_replace);
2993         if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
2994                 BUG_ON(num_devices < 1);
2995                 num_devices--;
2996         }
2997         btrfs_dev_replace_unlock(&fs_info->dev_replace);
2998         allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE;
2999         if (num_devices == 1)
3000                 allowed |= BTRFS_BLOCK_GROUP_DUP;
3001         else if (num_devices < 4)
3002                 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1);
3003         else
3004                 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
3005                                 BTRFS_BLOCK_GROUP_RAID10);
3006
3007         if ((bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3008             (!alloc_profile_is_valid(bctl->data.target, 1) ||
3009              (bctl->data.target & ~allowed))) {
3010                 printk(KERN_ERR "btrfs: unable to start balance with target "
3011                        "data profile %llu\n",
3012                        (unsigned long long)bctl->data.target);
3013                 ret = -EINVAL;
3014                 goto out;
3015         }
3016         if ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3017             (!alloc_profile_is_valid(bctl->meta.target, 1) ||
3018              (bctl->meta.target & ~allowed))) {
3019                 printk(KERN_ERR "btrfs: unable to start balance with target "
3020                        "metadata profile %llu\n",
3021                        (unsigned long long)bctl->meta.target);
3022                 ret = -EINVAL;
3023                 goto out;
3024         }
3025         if ((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3026             (!alloc_profile_is_valid(bctl->sys.target, 1) ||
3027              (bctl->sys.target & ~allowed))) {
3028                 printk(KERN_ERR "btrfs: unable to start balance with target "
3029                        "system profile %llu\n",
3030                        (unsigned long long)bctl->sys.target);
3031                 ret = -EINVAL;
3032                 goto out;
3033         }
3034
3035         /* allow dup'ed data chunks only in mixed mode */
3036         if (!mixed && (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3037             (bctl->data.target & BTRFS_BLOCK_GROUP_DUP)) {
3038                 printk(KERN_ERR "btrfs: dup for data is not allowed\n");
3039                 ret = -EINVAL;
3040                 goto out;
3041         }
3042
3043         /* allow to reduce meta or sys integrity only if force set */
3044         allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3045                         BTRFS_BLOCK_GROUP_RAID10;
3046         if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3047              (fs_info->avail_system_alloc_bits & allowed) &&
3048              !(bctl->sys.target & allowed)) ||
3049             ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3050              (fs_info->avail_metadata_alloc_bits & allowed) &&
3051              !(bctl->meta.target & allowed))) {
3052                 if (bctl->flags & BTRFS_BALANCE_FORCE) {
3053                         printk(KERN_INFO "btrfs: force reducing metadata "
3054                                "integrity\n");
3055                 } else {
3056                         printk(KERN_ERR "btrfs: balance will reduce metadata "
3057                                "integrity, use force if you want this\n");
3058                         ret = -EINVAL;
3059                         goto out;
3060                 }
3061         }
3062
3063         if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3064                 int num_tolerated_disk_barrier_failures;
3065                 u64 target = bctl->sys.target;
3066
3067                 num_tolerated_disk_barrier_failures =
3068                         btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3069                 if (num_tolerated_disk_barrier_failures > 0 &&
3070                     (target &
3071                      (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID0 |
3072                       BTRFS_AVAIL_ALLOC_BIT_SINGLE)))
3073                         num_tolerated_disk_barrier_failures = 0;
3074                 else if (num_tolerated_disk_barrier_failures > 1 &&
3075                          (target &
3076                           (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)))
3077                         num_tolerated_disk_barrier_failures = 1;
3078
3079                 fs_info->num_tolerated_disk_barrier_failures =
3080                         num_tolerated_disk_barrier_failures;
3081         }
3082
3083         ret = insert_balance_item(fs_info->tree_root, bctl);
3084         if (ret && ret != -EEXIST)
3085                 goto out;
3086
3087         if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
3088                 BUG_ON(ret == -EEXIST);
3089                 set_balance_control(bctl);
3090         } else {
3091                 BUG_ON(ret != -EEXIST);
3092                 spin_lock(&fs_info->balance_lock);
3093                 update_balance_args(bctl);
3094                 spin_unlock(&fs_info->balance_lock);
3095         }
3096
3097         atomic_inc(&fs_info->balance_running);
3098         mutex_unlock(&fs_info->balance_mutex);
3099
3100         ret = __btrfs_balance(fs_info);
3101
3102         mutex_lock(&fs_info->balance_mutex);
3103         atomic_dec(&fs_info->balance_running);
3104
3105         if (bargs) {
3106                 memset(bargs, 0, sizeof(*bargs));
3107                 update_ioctl_balance_args(fs_info, 0, bargs);
3108         }
3109
3110         if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
3111             balance_need_close(fs_info)) {
3112                 __cancel_balance(fs_info);
3113         }
3114
3115         if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3116                 fs_info->num_tolerated_disk_barrier_failures =
3117                         btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3118         }
3119
3120         wake_up(&fs_info->balance_wait_q);
3121
3122         return ret;
3123 out:
3124         if (bctl->flags & BTRFS_BALANCE_RESUME)
3125                 __cancel_balance(fs_info);
3126         else
3127                 kfree(bctl);
3128         return ret;
3129 }
3130
3131 static int balance_kthread(void *data)
3132 {
3133         struct btrfs_fs_info *fs_info = data;
3134         int ret = 0;
3135
3136         mutex_lock(&fs_info->volume_mutex);
3137         mutex_lock(&fs_info->balance_mutex);
3138
3139         if (fs_info->balance_ctl) {
3140                 printk(KERN_INFO "btrfs: continuing balance\n");
3141                 ret = btrfs_balance(fs_info->balance_ctl, NULL);
3142         }
3143
3144         atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3145         mutex_unlock(&fs_info->balance_mutex);
3146         mutex_unlock(&fs_info->volume_mutex);
3147
3148         return ret;
3149 }
3150
3151 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
3152 {
3153         struct task_struct *tsk;
3154
3155         spin_lock(&fs_info->balance_lock);
3156         if (!fs_info->balance_ctl) {
3157                 spin_unlock(&fs_info->balance_lock);
3158                 return 0;
3159         }
3160         spin_unlock(&fs_info->balance_lock);
3161
3162         if (btrfs_test_opt(fs_info->tree_root, SKIP_BALANCE)) {
3163                 printk(KERN_INFO "btrfs: force skipping balance\n");
3164                 return 0;
3165         }
3166
3167         WARN_ON(atomic_xchg(&fs_info->mutually_exclusive_operation_running, 1));
3168         tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
3169         if (IS_ERR(tsk))
3170                 return PTR_ERR(tsk);
3171
3172         return 0;
3173 }
3174
3175 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
3176 {
3177         struct btrfs_balance_control *bctl;
3178         struct btrfs_balance_item *item;
3179         struct btrfs_disk_balance_args disk_bargs;
3180         struct btrfs_path *path;
3181         struct extent_buffer *leaf;
3182         struct btrfs_key key;
3183         int ret;
3184
3185         path = btrfs_alloc_path();
3186         if (!path)
3187                 return -ENOMEM;
3188
3189         key.objectid = BTRFS_BALANCE_OBJECTID;
3190         key.type = BTRFS_BALANCE_ITEM_KEY;
3191         key.offset = 0;
3192
3193         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
3194         if (ret < 0)
3195                 goto out;
3196         if (ret > 0) { /* ret = -ENOENT; */
3197                 ret = 0;
3198                 goto out;
3199         }
3200
3201         bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
3202         if (!bctl) {
3203                 ret = -ENOMEM;
3204                 goto out;
3205         }
3206
3207         leaf = path->nodes[0];
3208         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3209
3210         bctl->fs_info = fs_info;
3211         bctl->flags = btrfs_balance_flags(leaf, item);
3212         bctl->flags |= BTRFS_BALANCE_RESUME;
3213
3214         btrfs_balance_data(leaf, item, &disk_bargs);
3215         btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
3216         btrfs_balance_meta(leaf, item, &disk_bargs);
3217         btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
3218         btrfs_balance_sys(leaf, item, &disk_bargs);
3219         btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
3220
3221         mutex_lock(&fs_info->volume_mutex);
3222         mutex_lock(&fs_info->balance_mutex);
3223
3224         set_balance_control(bctl);
3225
3226         mutex_unlock(&fs_info->balance_mutex);
3227         mutex_unlock(&fs_info->volume_mutex);
3228 out:
3229         btrfs_free_path(path);
3230         return ret;
3231 }
3232
3233 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
3234 {
3235         int ret = 0;
3236
3237         mutex_lock(&fs_info->balance_mutex);
3238         if (!fs_info->balance_ctl) {
3239                 mutex_unlock(&fs_info->balance_mutex);
3240                 return -ENOTCONN;
3241         }
3242
3243         if (atomic_read(&fs_info->balance_running)) {
3244                 atomic_inc(&fs_info->balance_pause_req);
3245                 mutex_unlock(&fs_info->balance_mutex);
3246
3247                 wait_event(fs_info->balance_wait_q,
3248                            atomic_read(&fs_info->balance_running) == 0);
3249
3250                 mutex_lock(&fs_info->balance_mutex);
3251                 /* we are good with balance_ctl ripped off from under us */
3252                 BUG_ON(atomic_read(&fs_info->balance_running));
3253                 atomic_dec(&fs_info->balance_pause_req);
3254         } else {
3255                 ret = -ENOTCONN;
3256         }
3257
3258         mutex_unlock(&fs_info->balance_mutex);
3259         return ret;
3260 }
3261
3262 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
3263 {
3264         mutex_lock(&fs_info->balance_mutex);
3265         if (!fs_info->balance_ctl) {
3266                 mutex_unlock(&fs_info->balance_mutex);
3267                 return -ENOTCONN;
3268         }
3269
3270         atomic_inc(&fs_info->balance_cancel_req);
3271         /*
3272          * if we are running just wait and return, balance item is
3273          * deleted in btrfs_balance in this case
3274          */
3275         if (atomic_read(&fs_info->balance_running)) {
3276                 mutex_unlock(&fs_info->balance_mutex);
3277                 wait_event(fs_info->balance_wait_q,
3278                            atomic_read(&fs_info->balance_running) == 0);
3279                 mutex_lock(&fs_info->balance_mutex);
3280         } else {
3281                 /* __cancel_balance needs volume_mutex */
3282                 mutex_unlock(&fs_info->balance_mutex);
3283                 mutex_lock(&fs_info->volume_mutex);
3284                 mutex_lock(&fs_info->balance_mutex);
3285
3286                 if (fs_info->balance_ctl)
3287                         __cancel_balance(fs_info);
3288
3289                 mutex_unlock(&fs_info->volume_mutex);
3290         }
3291
3292         BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running));
3293         atomic_dec(&fs_info->balance_cancel_req);
3294         mutex_unlock(&fs_info->balance_mutex);
3295         return 0;
3296 }
3297
3298 /*
3299  * shrinking a device means finding all of the device extents past
3300  * the new size, and then following the back refs to the chunks.
3301  * The chunk relocation code actually frees the device extent
3302  */
3303 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
3304 {
3305         struct btrfs_trans_handle *trans;
3306         struct btrfs_root *root = device->dev_root;
3307         struct btrfs_dev_extent *dev_extent = NULL;
3308         struct btrfs_path *path;
3309         u64 length;
3310         u64 chunk_tree;
3311         u64 chunk_objectid;
3312         u64 chunk_offset;
3313         int ret;
3314         int slot;
3315         int failed = 0;
3316         bool retried = false;
3317         struct extent_buffer *l;
3318         struct btrfs_key key;
3319         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
3320         u64 old_total = btrfs_super_total_bytes(super_copy);
3321         u64 old_size = device->total_bytes;
3322         u64 diff = device->total_bytes - new_size;
3323
3324         if (device->is_tgtdev_for_dev_replace)
3325                 return -EINVAL;
3326
3327         path = btrfs_alloc_path();
3328         if (!path)
3329                 return -ENOMEM;
3330
3331         path->reada = 2;
3332
3333         lock_chunks(root);
3334
3335         device->total_bytes = new_size;
3336         if (device->writeable) {
3337                 device->fs_devices->total_rw_bytes -= diff;
3338                 spin_lock(&root->fs_info->free_chunk_lock);
3339                 root->fs_info->free_chunk_space -= diff;
3340                 spin_unlock(&root->fs_info->free_chunk_lock);
3341         }
3342         unlock_chunks(root);
3343
3344 again:
3345         key.objectid = device->devid;
3346         key.offset = (u64)-1;
3347         key.type = BTRFS_DEV_EXTENT_KEY;
3348
3349         do {
3350                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3351                 if (ret < 0)
3352                         goto done;
3353
3354                 ret = btrfs_previous_item(root, path, 0, key.type);
3355                 if (ret < 0)
3356                         goto done;
3357                 if (ret) {
3358                         ret = 0;
3359                         btrfs_release_path(path);
3360                         break;
3361                 }
3362
3363                 l = path->nodes[0];
3364                 slot = path->slots[0];
3365                 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
3366
3367                 if (key.objectid != device->devid) {
3368                         btrfs_release_path(path);
3369                         break;
3370                 }
3371
3372                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
3373                 length = btrfs_dev_extent_length(l, dev_extent);
3374
3375                 if (key.offset + length <= new_size) {
3376                         btrfs_release_path(path);
3377                         break;
3378                 }
3379
3380                 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
3381                 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
3382                 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
3383                 btrfs_release_path(path);
3384
3385                 ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
3386                                            chunk_offset);
3387                 if (ret && ret != -ENOSPC)
3388                         goto done;
3389                 if (ret == -ENOSPC)
3390                         failed++;
3391         } while (key.offset-- > 0);
3392
3393         if (failed && !retried) {
3394                 failed = 0;
3395                 retried = true;
3396                 goto again;
3397         } else if (failed && retried) {
3398                 ret = -ENOSPC;
3399                 lock_chunks(root);
3400
3401                 device->total_bytes = old_size;
3402                 if (device->writeable)
3403                         device->fs_devices->total_rw_bytes += diff;
3404                 spin_lock(&root->fs_info->free_chunk_lock);
3405                 root->fs_info->free_chunk_space += diff;
3406                 spin_unlock(&root->fs_info->free_chunk_lock);
3407                 unlock_chunks(root);
3408                 goto done;
3409         }
3410
3411         /* Shrinking succeeded, else we would be at "done". */
3412         trans = btrfs_start_transaction(root, 0);
3413         if (IS_ERR(trans)) {
3414                 ret = PTR_ERR(trans);
3415                 goto done;
3416         }
3417
3418         lock_chunks(root);
3419
3420         device->disk_total_bytes = new_size;
3421         /* Now btrfs_update_device() will change the on-disk size. */
3422         ret = btrfs_update_device(trans, device);
3423         if (ret) {
3424                 unlock_chunks(root);
3425                 btrfs_end_transaction(trans, root);
3426                 goto done;
3427         }
3428         WARN_ON(diff > old_total);
3429         btrfs_set_super_total_bytes(super_copy, old_total - diff);
3430         unlock_chunks(root);
3431         btrfs_end_transaction(trans, root);
3432 done:
3433         btrfs_free_path(path);
3434         return ret;
3435 }
3436
3437 static int btrfs_add_system_chunk(struct btrfs_root *root,
3438                            struct btrfs_key *key,
3439                            struct btrfs_chunk *chunk, int item_size)
3440 {
3441         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
3442         struct btrfs_disk_key disk_key;
3443         u32 array_size;
3444         u8 *ptr;
3445
3446         array_size = btrfs_super_sys_array_size(super_copy);
3447         if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
3448                 return -EFBIG;
3449
3450         ptr = super_copy->sys_chunk_array + array_size;
3451         btrfs_cpu_key_to_disk(&disk_key, key);
3452         memcpy(ptr, &disk_key, sizeof(disk_key));
3453         ptr += sizeof(disk_key);
3454         memcpy(ptr, chunk, item_size);
3455         item_size += sizeof(disk_key);
3456         btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
3457         return 0;
3458 }
3459
3460 /*
3461  * sort the devices in descending order by max_avail, total_avail
3462  */
3463 static int btrfs_cmp_device_info(const void *a, const void *b)
3464 {
3465         const struct btrfs_device_info *di_a = a;
3466         const struct btrfs_device_info *di_b = b;
3467
3468         if (di_a->max_avail > di_b->max_avail)
3469                 return -1;
3470         if (di_a->max_avail < di_b->max_avail)
3471                 return 1;
3472         if (di_a->total_avail > di_b->total_avail)
3473                 return -1;
3474         if (di_a->total_avail < di_b->total_avail)
3475                 return 1;
3476         return 0;
3477 }
3478
3479 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
3480                                struct btrfs_root *extent_root,
3481                                struct map_lookup **map_ret,
3482                                u64 *num_bytes_out, u64 *stripe_size_out,
3483                                u64 start, u64 type)
3484 {
3485         struct btrfs_fs_info *info = extent_root->fs_info;
3486         struct btrfs_fs_devices *fs_devices = info->fs_devices;
3487         struct list_head *cur;
3488         struct map_lookup *map = NULL;
3489         struct extent_map_tree *em_tree;
3490         struct extent_map *em;
3491         struct btrfs_device_info *devices_info = NULL;
3492         u64 total_avail;
3493         int num_stripes;        /* total number of stripes to allocate */
3494         int sub_stripes;        /* sub_stripes info for map */
3495         int dev_stripes;        /* stripes per dev */
3496         int devs_max;           /* max devs to use */
3497         int devs_min;           /* min devs needed */
3498         int devs_increment;     /* ndevs has to be a multiple of this */
3499         int ncopies;            /* how many copies to data has */
3500         int ret;
3501         u64 max_stripe_size;
3502         u64 max_chunk_size;
3503         u64 stripe_size;
3504         u64 num_bytes;
3505         int ndevs;
3506         int i;
3507         int j;
3508
3509         BUG_ON(!alloc_profile_is_valid(type, 0));
3510
3511         if (list_empty(&fs_devices->alloc_list))
3512                 return -ENOSPC;
3513
3514         sub_stripes = 1;
3515         dev_stripes = 1;
3516         devs_increment = 1;
3517         ncopies = 1;
3518         devs_max = 0;   /* 0 == as many as possible */
3519         devs_min = 1;
3520
3521         /*
3522          * define the properties of each RAID type.
3523          * FIXME: move this to a global table and use it in all RAID
3524          * calculation code
3525          */
3526         if (type & (BTRFS_BLOCK_GROUP_DUP)) {
3527                 dev_stripes = 2;
3528                 ncopies = 2;
3529                 devs_max = 1;
3530         } else if (type & (BTRFS_BLOCK_GROUP_RAID0)) {
3531                 devs_min = 2;
3532         } else if (type & (BTRFS_BLOCK_GROUP_RAID1)) {
3533                 devs_increment = 2;
3534                 ncopies = 2;
3535                 devs_max = 2;
3536                 devs_min = 2;
3537         } else if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
3538                 sub_stripes = 2;
3539                 devs_increment = 2;
3540                 ncopies = 2;
3541                 devs_min = 4;
3542         } else {
3543                 devs_max = 1;
3544         }
3545
3546         if (type & BTRFS_BLOCK_GROUP_DATA) {
3547                 max_stripe_size = 1024 * 1024 * 1024;
3548                 max_chunk_size = 10 * max_stripe_size;
3549         } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
3550                 /* for larger filesystems, use larger metadata chunks */
3551                 if (fs_devices->total_rw_bytes > 50ULL * 1024 * 1024 * 1024)
3552                         max_stripe_size = 1024 * 1024 * 1024;
3553                 else
3554                         max_stripe_size = 256 * 1024 * 1024;
3555                 max_chunk_size = max_stripe_size;
3556         } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
3557                 max_stripe_size = 32 * 1024 * 1024;
3558                 max_chunk_size = 2 * max_stripe_size;
3559         } else {
3560                 printk(KERN_ERR "btrfs: invalid chunk type 0x%llx requested\n",
3561                        type);
3562                 BUG_ON(1);
3563         }
3564
3565         /* we don't want a chunk larger than 10% of writeable space */
3566         max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
3567                              max_chunk_size);
3568
3569         devices_info = kzalloc(sizeof(*devices_info) * fs_devices->rw_devices,
3570                                GFP_NOFS);
3571         if (!devices_info)
3572                 return -ENOMEM;
3573
3574         cur = fs_devices->alloc_list.next;
3575
3576         /*
3577          * in the first pass through the devices list, we gather information
3578          * about the available holes on each device.
3579          */
3580         ndevs = 0;
3581         while (cur != &fs_devices->alloc_list) {
3582                 struct btrfs_device *device;
3583                 u64 max_avail;
3584                 u64 dev_offset;
3585
3586                 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
3587
3588                 cur = cur->next;
3589
3590                 if (!device->writeable) {
3591                         WARN(1, KERN_ERR
3592                                "btrfs: read-only device in alloc_list\n");
3593                         continue;
3594                 }
3595
3596                 if (!device->in_fs_metadata ||
3597                     device->is_tgtdev_for_dev_replace)
3598                         continue;
3599
3600                 if (device->total_bytes > device->bytes_used)
3601                         total_avail = device->total_bytes - device->bytes_used;
3602                 else
3603                         total_avail = 0;
3604
3605                 /* If there is no space on this device, skip it. */
3606                 if (total_avail == 0)
3607                         continue;
3608
3609                 ret = find_free_dev_extent(device,
3610                                            max_stripe_size * dev_stripes,
3611                                            &dev_offset, &max_avail);
3612                 if (ret && ret != -ENOSPC)
3613                         goto error;
3614
3615                 if (ret == 0)
3616                         max_avail = max_stripe_size * dev_stripes;
3617
3618                 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes)
3619                         continue;
3620
3621                 devices_info[ndevs].dev_offset = dev_offset;
3622                 devices_info[ndevs].max_avail = max_avail;
3623                 devices_info[ndevs].total_avail = total_avail;
3624                 devices_info[ndevs].dev = device;
3625                 ++ndevs;
3626                 WARN_ON(ndevs > fs_devices->rw_devices);
3627         }
3628
3629         /*
3630          * now sort the devices by hole size / available space
3631          */
3632         sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
3633              btrfs_cmp_device_info, NULL);
3634
3635         /* round down to number of usable stripes */
3636         ndevs -= ndevs % devs_increment;
3637
3638         if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) {
3639                 ret = -ENOSPC;
3640                 goto error;
3641         }
3642
3643         if (devs_max && ndevs > devs_max)
3644                 ndevs = devs_max;
3645         /*
3646          * the primary goal is to maximize the number of stripes, so use as many
3647          * devices as possible, even if the stripes are not maximum sized.
3648          */
3649         stripe_size = devices_info[ndevs-1].max_avail;
3650         num_stripes = ndevs * dev_stripes;
3651
3652         if (stripe_size * ndevs > max_chunk_size * ncopies) {
3653                 stripe_size = max_chunk_size * ncopies;
3654                 do_div(stripe_size, ndevs);
3655         }
3656
3657         do_div(stripe_size, dev_stripes);
3658
3659         /* align to BTRFS_STRIPE_LEN */
3660         do_div(stripe_size, BTRFS_STRIPE_LEN);
3661         stripe_size *= BTRFS_STRIPE_LEN;
3662
3663         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
3664         if (!map) {
3665                 ret = -ENOMEM;
3666                 goto error;
3667         }
3668         map->num_stripes = num_stripes;
3669
3670         for (i = 0; i < ndevs; ++i) {
3671                 for (j = 0; j < dev_stripes; ++j) {
3672                         int s = i * dev_stripes + j;
3673                         map->stripes[s].dev = devices_info[i].dev;
3674                         map->stripes[s].physical = devices_info[i].dev_offset +
3675                                                    j * stripe_size;
3676                 }
3677         }
3678         map->sector_size = extent_root->sectorsize;
3679         map->stripe_len = BTRFS_STRIPE_LEN;
3680         map->io_align = BTRFS_STRIPE_LEN;
3681         map->io_width = BTRFS_STRIPE_LEN;
3682         map->type = type;
3683         map->sub_stripes = sub_stripes;
3684
3685         *map_ret = map;
3686         num_bytes = stripe_size * (num_stripes / ncopies);
3687
3688         *stripe_size_out = stripe_size;
3689         *num_bytes_out = num_bytes;
3690
3691         trace_btrfs_chunk_alloc(info->chunk_root, map, start, num_bytes);
3692
3693         em = alloc_extent_map();
3694         if (!em) {
3695                 ret = -ENOMEM;
3696                 goto error;
3697         }
3698         em->bdev = (struct block_device *)map;
3699         em->start = start;
3700         em->len = num_bytes;
3701         em->block_start = 0;
3702         em->block_len = em->len;
3703
3704         em_tree = &extent_root->fs_info->mapping_tree.map_tree;
3705         write_lock(&em_tree->lock);
3706         ret = add_extent_mapping(em_tree, em);
3707         write_unlock(&em_tree->lock);
3708         free_extent_map(em);
3709         if (ret)
3710                 goto error;
3711
3712         ret = btrfs_make_block_group(trans, extent_root, 0, type,
3713                                      BTRFS_FIRST_CHUNK_TREE_OBJECTID,
3714                                      start, num_bytes);
3715         if (ret)
3716                 goto error;
3717
3718         for (i = 0; i < map->num_stripes; ++i) {
3719                 struct btrfs_device *device;
3720                 u64 dev_offset;
3721
3722                 device = map->stripes[i].dev;
3723                 dev_offset = map->stripes[i].physical;
3724
3725                 ret = btrfs_alloc_dev_extent(trans, device,
3726                                 info->chunk_root->root_key.objectid,
3727                                 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
3728                                 start, dev_offset, stripe_size);
3729                 if (ret) {
3730                         btrfs_abort_transaction(trans, extent_root, ret);
3731                         goto error;
3732                 }
3733         }
3734
3735         kfree(devices_info);
3736         return 0;
3737
3738 error:
3739         kfree(map);
3740         kfree(devices_info);
3741         return ret;
3742 }
3743
3744 static int __finish_chunk_alloc(struct btrfs_trans_handle *trans,
3745                                 struct btrfs_root *extent_root,
3746                                 struct map_lookup *map, u64 chunk_offset,
3747                                 u64 chunk_size, u64 stripe_size)
3748 {
3749         u64 dev_offset;
3750         struct btrfs_key key;
3751         struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
3752         struct btrfs_device *device;
3753         struct btrfs_chunk *chunk;
3754         struct btrfs_stripe *stripe;
3755         size_t item_size = btrfs_chunk_item_size(map->num_stripes);
3756         int index = 0;
3757         int ret;
3758
3759         chunk = kzalloc(item_size, GFP_NOFS);
3760         if (!chunk)
3761                 return -ENOMEM;
3762
3763         index = 0;
3764         while (index < map->num_stripes) {
3765                 device = map->stripes[index].dev;
3766                 device->bytes_used += stripe_size;
3767                 ret = btrfs_update_device(trans, device);
3768                 if (ret)
3769                         goto out_free;
3770                 index++;
3771         }
3772
3773         spin_lock(&extent_root->fs_info->free_chunk_lock);
3774         extent_root->fs_info->free_chunk_space -= (stripe_size *
3775                                                    map->num_stripes);
3776         spin_unlock(&extent_root->fs_info->free_chunk_lock);
3777
3778         index = 0;
3779         stripe = &chunk->stripe;
3780         while (index < map->num_stripes) {
3781                 device = map->stripes[index].dev;
3782                 dev_offset = map->stripes[index].physical;
3783
3784                 btrfs_set_stack_stripe_devid(stripe, device->devid);
3785                 btrfs_set_stack_stripe_offset(stripe, dev_offset);
3786                 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
3787                 stripe++;
3788                 index++;
3789         }
3790
3791         btrfs_set_stack_chunk_length(chunk, chunk_size);
3792         btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
3793         btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
3794         btrfs_set_stack_chunk_type(chunk, map->type);
3795         btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
3796         btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
3797         btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
3798         btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
3799         btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
3800
3801         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3802         key.type = BTRFS_CHUNK_ITEM_KEY;
3803         key.offset = chunk_offset;
3804
3805         ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
3806
3807         if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
3808                 /*
3809                  * TODO: Cleanup of inserted chunk root in case of
3810                  * failure.
3811                  */
3812                 ret = btrfs_add_system_chunk(chunk_root, &key, chunk,
3813                                              item_size);
3814         }
3815
3816 out_free:
3817         kfree(chunk);
3818         return ret;
3819 }
3820
3821 /*
3822  * Chunk allocation falls into two parts. The first part does works
3823  * that make the new allocated chunk useable, but not do any operation
3824  * that modifies the chunk tree. The second part does the works that
3825  * require modifying the chunk tree. This division is important for the
3826  * bootstrap process of adding storage to a seed btrfs.
3827  */
3828 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
3829                       struct btrfs_root *extent_root, u64 type)
3830 {
3831         u64 chunk_offset;
3832         u64 chunk_size;
3833         u64 stripe_size;
3834         struct map_lookup *map;
3835         struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
3836         int ret;
3837
3838         ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
3839                               &chunk_offset);
3840         if (ret)
3841                 return ret;
3842
3843         ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
3844                                   &stripe_size, chunk_offset, type);
3845         if (ret)
3846                 return ret;
3847
3848         ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
3849                                    chunk_size, stripe_size);
3850         if (ret)
3851                 return ret;
3852         return 0;
3853 }
3854
3855 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
3856                                          struct btrfs_root *root,
3857                                          struct btrfs_device *device)
3858 {
3859         u64 chunk_offset;
3860         u64 sys_chunk_offset;
3861         u64 chunk_size;
3862         u64 sys_chunk_size;
3863         u64 stripe_size;
3864         u64 sys_stripe_size;
3865         u64 alloc_profile;
3866         struct map_lookup *map;
3867         struct map_lookup *sys_map;
3868         struct btrfs_fs_info *fs_info = root->fs_info;
3869         struct btrfs_root *extent_root = fs_info->extent_root;
3870         int ret;
3871
3872         ret = find_next_chunk(fs_info->chunk_root,
3873                               BTRFS_FIRST_CHUNK_TREE_OBJECTID, &chunk_offset);
3874         if (ret)
3875                 return ret;
3876
3877         alloc_profile = BTRFS_BLOCK_GROUP_METADATA |
3878                                 fs_info->avail_metadata_alloc_bits;
3879         alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
3880
3881         ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
3882                                   &stripe_size, chunk_offset, alloc_profile);
3883         if (ret)
3884                 return ret;
3885
3886         sys_chunk_offset = chunk_offset + chunk_size;
3887
3888         alloc_profile = BTRFS_BLOCK_GROUP_SYSTEM |
3889                                 fs_info->avail_system_alloc_bits;
3890         alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
3891
3892         ret = __btrfs_alloc_chunk(trans, extent_root, &sys_map,
3893                                   &sys_chunk_size, &sys_stripe_size,
3894                                   sys_chunk_offset, alloc_profile);
3895         if (ret) {
3896                 btrfs_abort_transaction(trans, root, ret);
3897                 goto out;
3898         }
3899
3900         ret = btrfs_add_device(trans, fs_info->chunk_root, device);
3901         if (ret) {
3902                 btrfs_abort_transaction(trans, root, ret);
3903                 goto out;
3904         }
3905
3906         /*
3907          * Modifying chunk tree needs allocating new blocks from both
3908          * system block group and metadata block group. So we only can
3909          * do operations require modifying the chunk tree after both
3910          * block groups were created.
3911          */
3912         ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
3913                                    chunk_size, stripe_size);
3914         if (ret) {
3915                 btrfs_abort_transaction(trans, root, ret);
3916                 goto out;
3917         }
3918
3919         ret = __finish_chunk_alloc(trans, extent_root, sys_map,
3920                                    sys_chunk_offset, sys_chunk_size,
3921                                    sys_stripe_size);
3922         if (ret)
3923                 btrfs_abort_transaction(trans, root, ret);
3924
3925 out:
3926
3927         return ret;
3928 }
3929
3930 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
3931 {
3932         struct extent_map *em;
3933         struct map_lookup *map;
3934         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
3935         int readonly = 0;
3936         int i;
3937
3938         read_lock(&map_tree->map_tree.lock);
3939         em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
3940         read_unlock(&map_tree->map_tree.lock);
3941         if (!em)
3942                 return 1;
3943
3944         if (btrfs_test_opt(root, DEGRADED)) {
3945                 free_extent_map(em);
3946                 return 0;
3947         }
3948
3949         map = (struct map_lookup *)em->bdev;
3950         for (i = 0; i < map->num_stripes; i++) {
3951                 if (!map->stripes[i].dev->writeable) {
3952                         readonly = 1;
3953                         break;
3954                 }
3955         }
3956         free_extent_map(em);
3957         return readonly;
3958 }
3959
3960 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
3961 {
3962         extent_map_tree_init(&tree->map_tree);
3963 }
3964
3965 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
3966 {
3967         struct extent_map *em;
3968
3969         while (1) {
3970                 write_lock(&tree->map_tree.lock);
3971                 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
3972                 if (em)
3973                         remove_extent_mapping(&tree->map_tree, em);
3974                 write_unlock(&tree->map_tree.lock);
3975                 if (!em)
3976                         break;
3977                 kfree(em->bdev);
3978                 /* once for us */
3979                 free_extent_map(em);
3980                 /* once for the tree */
3981                 free_extent_map(em);
3982         }
3983 }
3984
3985 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
3986 {
3987         struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
3988         struct extent_map *em;
3989         struct map_lookup *map;
3990         struct extent_map_tree *em_tree = &map_tree->map_tree;
3991         int ret;
3992
3993         read_lock(&em_tree->lock);
3994         em = lookup_extent_mapping(em_tree, logical, len);
3995         read_unlock(&em_tree->lock);
3996         BUG_ON(!em);
3997
3998         BUG_ON(em->start > logical || em->start + em->len < logical);
3999         map = (struct map_lookup *)em->bdev;
4000         if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
4001                 ret = map->num_stripes;
4002         else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
4003                 ret = map->sub_stripes;
4004         else
4005                 ret = 1;
4006         free_extent_map(em);
4007         return ret;
4008 }
4009
4010 static int find_live_mirror(struct map_lookup *map, int first, int num,
4011                             int optimal)
4012 {
4013         int i;
4014         if (map->stripes[optimal].dev->bdev)
4015                 return optimal;
4016         for (i = first; i < first + num; i++) {
4017                 if (map->stripes[i].dev->bdev)
4018                         return i;
4019         }
4020         /* we couldn't find one that doesn't fail.  Just return something
4021          * and the io error handling code will clean up eventually
4022          */
4023         return optimal;
4024 }
4025
4026 static int __btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
4027                              u64 logical, u64 *length,
4028                              struct btrfs_bio **bbio_ret,
4029                              int mirror_num)
4030 {
4031         struct extent_map *em;
4032         struct map_lookup *map;
4033         struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
4034         struct extent_map_tree *em_tree = &map_tree->map_tree;
4035         u64 offset;
4036         u64 stripe_offset;
4037         u64 stripe_end_offset;
4038         u64 stripe_nr;
4039         u64 stripe_nr_orig;
4040         u64 stripe_nr_end;
4041         int stripe_index;
4042         int i;
4043         int ret = 0;
4044         int num_stripes;
4045         int max_errors = 0;
4046         struct btrfs_bio *bbio = NULL;
4047
4048         read_lock(&em_tree->lock);
4049         em = lookup_extent_mapping(em_tree, logical, *length);
4050         read_unlock(&em_tree->lock);
4051
4052         if (!em) {
4053                 printk(KERN_CRIT "btrfs: unable to find logical %llu len %llu\n",
4054                        (unsigned long long)logical,
4055                        (unsigned long long)*length);
4056                 BUG();
4057         }
4058
4059         BUG_ON(em->start > logical || em->start + em->len < logical);
4060         map = (struct map_lookup *)em->bdev;
4061         offset = logical - em->start;
4062
4063         if (mirror_num > map->num_stripes)
4064                 mirror_num = 0;
4065
4066         stripe_nr = offset;
4067         /*
4068          * stripe_nr counts the total number of stripes we have to stride
4069          * to get to this block
4070          */
4071         do_div(stripe_nr, map->stripe_len);
4072
4073         stripe_offset = stripe_nr * map->stripe_len;
4074         BUG_ON(offset < stripe_offset);
4075
4076         /* stripe_offset is the offset of this block in its stripe*/
4077         stripe_offset = offset - stripe_offset;
4078
4079         if (rw & REQ_DISCARD)
4080                 *length = min_t(u64, em->len - offset, *length);
4081         else if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
4082                 /* we limit the length of each bio to what fits in a stripe */
4083                 *length = min_t(u64, em->len - offset,
4084                                 map->stripe_len - stripe_offset);
4085         } else {
4086                 *length = em->len - offset;
4087         }
4088
4089         if (!bbio_ret)
4090                 goto out;
4091
4092         num_stripes = 1;
4093         stripe_index = 0;
4094         stripe_nr_orig = stripe_nr;
4095         stripe_nr_end = (offset + *length + map->stripe_len - 1) &
4096                         (~(map->stripe_len - 1));
4097         do_div(stripe_nr_end, map->stripe_len);
4098         stripe_end_offset = stripe_nr_end * map->stripe_len -
4099                             (offset + *length);
4100         if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
4101                 if (rw & REQ_DISCARD)
4102                         num_stripes = min_t(u64, map->num_stripes,
4103                                             stripe_nr_end - stripe_nr_orig);
4104                 stripe_index = do_div(stripe_nr, map->num_stripes);
4105         } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
4106                 if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS))
4107                         num_stripes = map->num_stripes;
4108                 else if (mirror_num)
4109                         stripe_index = mirror_num - 1;
4110                 else {
4111                         stripe_index = find_live_mirror(map, 0,
4112                                             map->num_stripes,
4113                                             current->pid % map->num_stripes);
4114                         mirror_num = stripe_index + 1;
4115                 }
4116
4117         } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
4118                 if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) {
4119                         num_stripes = map->num_stripes;
4120                 } else if (mirror_num) {
4121                         stripe_index = mirror_num - 1;
4122                 } else {
4123                         mirror_num = 1;
4124                 }
4125
4126         } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
4127                 int factor = map->num_stripes / map->sub_stripes;
4128
4129                 stripe_index = do_div(stripe_nr, factor);
4130                 stripe_index *= map->sub_stripes;
4131
4132                 if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS))
4133                         num_stripes = map->sub_stripes;
4134                 else if (rw & REQ_DISCARD)
4135                         num_stripes = min_t(u64, map->sub_stripes *
4136                                             (stripe_nr_end - stripe_nr_orig),
4137                                             map->num_stripes);
4138                 else if (mirror_num)
4139                         stripe_index += mirror_num - 1;
4140                 else {
4141                         int old_stripe_index = stripe_index;
4142                         stripe_index = find_live_mirror(map, stripe_index,
4143                                               map->sub_stripes, stripe_index +
4144                                               current->pid % map->sub_stripes);
4145                         mirror_num = stripe_index - old_stripe_index + 1;
4146                 }
4147         } else {
4148                 /*
4149                  * after this do_div call, stripe_nr is the number of stripes
4150                  * on this device we have to walk to find the data, and
4151                  * stripe_index is the number of our device in the stripe array
4152                  */
4153                 stripe_index = do_div(stripe_nr, map->num_stripes);
4154                 mirror_num = stripe_index + 1;
4155         }
4156         BUG_ON(stripe_index >= map->num_stripes);
4157
4158         bbio = kzalloc(btrfs_bio_size(num_stripes), GFP_NOFS);
4159         if (!bbio) {
4160                 ret = -ENOMEM;
4161                 goto out;
4162         }
4163         atomic_set(&bbio->error, 0);
4164
4165         if (rw & REQ_DISCARD) {
4166                 int factor = 0;
4167                 int sub_stripes = 0;
4168                 u64 stripes_per_dev = 0;
4169                 u32 remaining_stripes = 0;
4170                 u32 last_stripe = 0;
4171
4172                 if (map->type &
4173                     (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID10)) {
4174                         if (map->type & BTRFS_BLOCK_GROUP_RAID0)
4175                                 sub_stripes = 1;
4176                         else
4177                                 sub_stripes = map->sub_stripes;
4178
4179                         factor = map->num_stripes / sub_stripes;
4180                         stripes_per_dev = div_u64_rem(stripe_nr_end -
4181                                                       stripe_nr_orig,
4182                                                       factor,
4183                                                       &remaining_stripes);
4184                         div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
4185                         last_stripe *= sub_stripes;
4186                 }
4187
4188                 for (i = 0; i < num_stripes; i++) {
4189                         bbio->stripes[i].physical =
4190                                 map->stripes[stripe_index].physical +
4191                                 stripe_offset + stripe_nr * map->stripe_len;
4192                         bbio->stripes[i].dev = map->stripes[stripe_index].dev;
4193
4194                         if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
4195                                          BTRFS_BLOCK_GROUP_RAID10)) {
4196                                 bbio->stripes[i].length = stripes_per_dev *
4197                                                           map->stripe_len;
4198
4199                                 if (i / sub_stripes < remaining_stripes)
4200                                         bbio->stripes[i].length +=
4201                                                 map->stripe_len;
4202
4203                                 /*
4204                                  * Special for the first stripe and
4205                                  * the last stripe:
4206                                  *
4207                                  * |-------|...|-------|
4208                                  *     |----------|
4209                                  *    off     end_off
4210                                  */
4211                                 if (i < sub_stripes)
4212                                         bbio->stripes[i].length -=
4213                                                 stripe_offset;
4214
4215                                 if (stripe_index >= last_stripe &&
4216                                     stripe_index <= (last_stripe +
4217                                                      sub_stripes - 1))
4218                                         bbio->stripes[i].length -=
4219                                                 stripe_end_offset;
4220
4221                                 if (i == sub_stripes - 1)
4222                                         stripe_offset = 0;
4223                         } else
4224                                 bbio->stripes[i].length = *length;
4225
4226                         stripe_index++;
4227                         if (stripe_index == map->num_stripes) {
4228                                 /* This could only happen for RAID0/10 */
4229                                 stripe_index = 0;
4230                                 stripe_nr++;
4231                         }
4232                 }
4233         } else {
4234                 for (i = 0; i < num_stripes; i++) {
4235                         bbio->stripes[i].physical =
4236                                 map->stripes[stripe_index].physical +
4237                                 stripe_offset +
4238                                 stripe_nr * map->stripe_len;
4239                         bbio->stripes[i].dev =
4240                                 map->stripes[stripe_index].dev;
4241                         stripe_index++;
4242                 }
4243         }
4244
4245         if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS)) {
4246                 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
4247                                  BTRFS_BLOCK_GROUP_RAID10 |
4248                                  BTRFS_BLOCK_GROUP_DUP)) {
4249                         max_errors = 1;
4250                 }
4251         }
4252
4253         *bbio_ret = bbio;
4254         bbio->num_stripes = num_stripes;
4255         bbio->max_errors = max_errors;
4256         bbio->mirror_num = mirror_num;
4257 out:
4258         free_extent_map(em);
4259         return ret;
4260 }
4261
4262 int btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
4263                       u64 logical, u64 *length,
4264                       struct btrfs_bio **bbio_ret, int mirror_num)
4265 {
4266         return __btrfs_map_block(fs_info, rw, logical, length, bbio_ret,
4267                                  mirror_num);
4268 }
4269
4270 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
4271                      u64 chunk_start, u64 physical, u64 devid,
4272                      u64 **logical, int *naddrs, int *stripe_len)
4273 {
4274         struct extent_map_tree *em_tree = &map_tree->map_tree;
4275         struct extent_map *em;
4276         struct map_lookup *map;
4277         u64 *buf;
4278         u64 bytenr;
4279         u64 length;
4280         u64 stripe_nr;
4281         int i, j, nr = 0;
4282
4283         read_lock(&em_tree->lock);
4284         em = lookup_extent_mapping(em_tree, chunk_start, 1);
4285         read_unlock(&em_tree->lock);
4286
4287         BUG_ON(!em || em->start != chunk_start);
4288         map = (struct map_lookup *)em->bdev;
4289
4290         length = em->len;
4291         if (map->type & BTRFS_BLOCK_GROUP_RAID10)
4292                 do_div(length, map->num_stripes / map->sub_stripes);
4293         else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
4294                 do_div(length, map->num_stripes);
4295
4296         buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
4297         BUG_ON(!buf); /* -ENOMEM */
4298
4299         for (i = 0; i < map->num_stripes; i++) {
4300                 if (devid && map->stripes[i].dev->devid != devid)
4301                         continue;
4302                 if (map->stripes[i].physical > physical ||
4303                     map->stripes[i].physical + length <= physical)
4304                         continue;
4305
4306                 stripe_nr = physical - map->stripes[i].physical;
4307                 do_div(stripe_nr, map->stripe_len);
4308
4309                 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
4310                         stripe_nr = stripe_nr * map->num_stripes + i;
4311                         do_div(stripe_nr, map->sub_stripes);
4312                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
4313                         stripe_nr = stripe_nr * map->num_stripes + i;
4314                 }
4315                 bytenr = chunk_start + stripe_nr * map->stripe_len;
4316                 WARN_ON(nr >= map->num_stripes);
4317                 for (j = 0; j < nr; j++) {
4318                         if (buf[j] == bytenr)
4319                                 break;
4320                 }
4321                 if (j == nr) {
4322                         WARN_ON(nr >= map->num_stripes);
4323                         buf[nr++] = bytenr;
4324                 }
4325         }
4326
4327         *logical = buf;
4328         *naddrs = nr;
4329         *stripe_len = map->stripe_len;
4330
4331         free_extent_map(em);
4332         return 0;
4333 }
4334
4335 static void *merge_stripe_index_into_bio_private(void *bi_private,
4336                                                  unsigned int stripe_index)
4337 {
4338         /*
4339          * with single, dup, RAID0, RAID1 and RAID10, stripe_index is
4340          * at most 1.
4341          * The alternative solution (instead of stealing bits from the
4342          * pointer) would be to allocate an intermediate structure
4343          * that contains the old private pointer plus the stripe_index.
4344          */
4345         BUG_ON((((uintptr_t)bi_private) & 3) != 0);
4346         BUG_ON(stripe_index > 3);
4347         return (void *)(((uintptr_t)bi_private) | stripe_index);
4348 }
4349
4350 static struct btrfs_bio *extract_bbio_from_bio_private(void *bi_private)
4351 {
4352         return (struct btrfs_bio *)(((uintptr_t)bi_private) & ~((uintptr_t)3));
4353 }
4354
4355 static unsigned int extract_stripe_index_from_bio_private(void *bi_private)
4356 {
4357         return (unsigned int)((uintptr_t)bi_private) & 3;
4358 }
4359
4360 static void btrfs_end_bio(struct bio *bio, int err)
4361 {
4362         struct btrfs_bio *bbio = extract_bbio_from_bio_private(bio->bi_private);
4363         int is_orig_bio = 0;
4364
4365         if (err) {
4366                 atomic_inc(&bbio->error);
4367                 if (err == -EIO || err == -EREMOTEIO) {
4368                         unsigned int stripe_index =
4369                                 extract_stripe_index_from_bio_private(
4370                                         bio->bi_private);
4371                         struct btrfs_device *dev;
4372
4373                         BUG_ON(stripe_index >= bbio->num_stripes);
4374                         dev = bbio->stripes[stripe_index].dev;
4375                         if (dev->bdev) {
4376                                 if (bio->bi_rw & WRITE)
4377                                         btrfs_dev_stat_inc(dev,
4378                                                 BTRFS_DEV_STAT_WRITE_ERRS);
4379                                 else
4380                                         btrfs_dev_stat_inc(dev,
4381                                                 BTRFS_DEV_STAT_READ_ERRS);
4382                                 if ((bio->bi_rw & WRITE_FLUSH) == WRITE_FLUSH)
4383                                         btrfs_dev_stat_inc(dev,
4384                                                 BTRFS_DEV_STAT_FLUSH_ERRS);
4385                                 btrfs_dev_stat_print_on_error(dev);
4386                         }
4387                 }
4388         }
4389
4390         if (bio == bbio->orig_bio)
4391                 is_orig_bio = 1;
4392
4393         if (atomic_dec_and_test(&bbio->stripes_pending)) {
4394                 if (!is_orig_bio) {
4395                         bio_put(bio);
4396                         bio = bbio->orig_bio;
4397                 }
4398                 bio->bi_private = bbio->private;
4399                 bio->bi_end_io = bbio->end_io;
4400                 bio->bi_bdev = (struct block_device *)
4401                                         (unsigned long)bbio->mirror_num;
4402                 /* only send an error to the higher layers if it is
4403                  * beyond the tolerance of the multi-bio
4404                  */
4405                 if (atomic_read(&bbio->error) > bbio->max_errors) {
4406                         err = -EIO;
4407                 } else {
4408                         /*
4409                          * this bio is actually up to date, we didn't
4410                          * go over the max number of errors
4411                          */
4412                         set_bit(BIO_UPTODATE, &bio->bi_flags);
4413                         err = 0;
4414                 }
4415                 kfree(bbio);
4416
4417                 bio_endio(bio, err);
4418         } else if (!is_orig_bio) {
4419                 bio_put(bio);
4420         }
4421 }
4422
4423 struct async_sched {
4424         struct bio *bio;
4425         int rw;
4426         struct btrfs_fs_info *info;
4427         struct btrfs_work work;
4428 };
4429
4430 /*
4431  * see run_scheduled_bios for a description of why bios are collected for
4432  * async submit.
4433  *
4434  * This will add one bio to the pending list for a device and make sure
4435  * the work struct is scheduled.
4436  */
4437 static noinline void schedule_bio(struct btrfs_root *root,
4438                                  struct btrfs_device *device,
4439                                  int rw, struct bio *bio)
4440 {
4441         int should_queue = 1;
4442         struct btrfs_pending_bios *pending_bios;
4443
4444         /* don't bother with additional async steps for reads, right now */
4445         if (!(rw & REQ_WRITE)) {
4446                 bio_get(bio);
4447                 btrfsic_submit_bio(rw, bio);
4448                 bio_put(bio);
4449                 return;
4450         }
4451
4452         /*
4453          * nr_async_bios allows us to reliably return congestion to the
4454          * higher layers.  Otherwise, the async bio makes it appear we have
4455          * made progress against dirty pages when we've really just put it
4456          * on a queue for later
4457          */
4458         atomic_inc(&root->fs_info->nr_async_bios);
4459         WARN_ON(bio->bi_next);
4460         bio->bi_next = NULL;
4461         bio->bi_rw |= rw;
4462
4463         spin_lock(&device->io_lock);
4464         if (bio->bi_rw & REQ_SYNC)
4465                 pending_bios = &device->pending_sync_bios;
4466         else
4467                 pending_bios = &device->pending_bios;
4468
4469         if (pending_bios->tail)
4470                 pending_bios->tail->bi_next = bio;
4471
4472         pending_bios->tail = bio;
4473         if (!pending_bios->head)
4474                 pending_bios->head = bio;
4475         if (device->running_pending)
4476                 should_queue = 0;
4477
4478         spin_unlock(&device->io_lock);
4479
4480         if (should_queue)
4481                 btrfs_queue_worker(&root->fs_info->submit_workers,
4482                                    &device->work);
4483 }
4484
4485 static int bio_size_ok(struct block_device *bdev, struct bio *bio,
4486                        sector_t sector)
4487 {
4488         struct bio_vec *prev;
4489         struct request_queue *q = bdev_get_queue(bdev);
4490         unsigned short max_sectors = queue_max_sectors(q);
4491         struct bvec_merge_data bvm = {
4492                 .bi_bdev = bdev,
4493                 .bi_sector = sector,
4494                 .bi_rw = bio->bi_rw,
4495         };
4496
4497         if (bio->bi_vcnt == 0) {
4498                 WARN_ON(1);
4499                 return 1;
4500         }
4501
4502         prev = &bio->bi_io_vec[bio->bi_vcnt - 1];
4503         if ((bio->bi_size >> 9) > max_sectors)
4504                 return 0;
4505
4506         if (!q->merge_bvec_fn)
4507                 return 1;
4508
4509         bvm.bi_size = bio->bi_size - prev->bv_len;
4510         if (q->merge_bvec_fn(q, &bvm, prev) < prev->bv_len)
4511                 return 0;
4512         return 1;
4513 }
4514
4515 static void submit_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio,
4516                               struct bio *bio, u64 physical, int dev_nr,
4517                               int rw, int async)
4518 {
4519         struct btrfs_device *dev = bbio->stripes[dev_nr].dev;
4520
4521         bio->bi_private = bbio;
4522         bio->bi_private = merge_stripe_index_into_bio_private(
4523                         bio->bi_private, (unsigned int)dev_nr);
4524         bio->bi_end_io = btrfs_end_bio;
4525         bio->bi_sector = physical >> 9;
4526 #ifdef DEBUG
4527         {
4528                 struct rcu_string *name;
4529
4530                 rcu_read_lock();
4531                 name = rcu_dereference(dev->name);
4532                 pr_debug("btrfs_map_bio: rw %d, sector=%llu, dev=%lu "
4533                          "(%s id %llu), size=%u\n", rw,
4534                          (u64)bio->bi_sector, (u_long)dev->bdev->bd_dev,
4535                          name->str, dev->devid, bio->bi_size);
4536                 rcu_read_unlock();
4537         }
4538 #endif
4539         bio->bi_bdev = dev->bdev;
4540         if (async)
4541                 schedule_bio(root, dev, rw, bio);
4542         else
4543                 btrfsic_submit_bio(rw, bio);
4544 }
4545
4546 static int breakup_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio,
4547                               struct bio *first_bio, struct btrfs_device *dev,
4548                               int dev_nr, int rw, int async)
4549 {
4550         struct bio_vec *bvec = first_bio->bi_io_vec;
4551         struct bio *bio;
4552         int nr_vecs = bio_get_nr_vecs(dev->bdev);
4553         u64 physical = bbio->stripes[dev_nr].physical;
4554
4555 again:
4556         bio = btrfs_bio_alloc(dev->bdev, physical >> 9, nr_vecs, GFP_NOFS);
4557         if (!bio)
4558                 return -ENOMEM;
4559
4560         while (bvec <= (first_bio->bi_io_vec + first_bio->bi_vcnt - 1)) {
4561                 if (bio_add_page(bio, bvec->bv_page, bvec->bv_len,
4562                                  bvec->bv_offset) < bvec->bv_len) {
4563                         u64 len = bio->bi_size;
4564
4565                         atomic_inc(&bbio->stripes_pending);
4566                         submit_stripe_bio(root, bbio, bio, physical, dev_nr,
4567                                           rw, async);
4568                         physical += len;
4569                         goto again;
4570                 }
4571                 bvec++;
4572         }
4573
4574         submit_stripe_bio(root, bbio, bio, physical, dev_nr, rw, async);
4575         return 0;
4576 }
4577
4578 static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical)
4579 {
4580         atomic_inc(&bbio->error);
4581         if (atomic_dec_and_test(&bbio->stripes_pending)) {
4582                 bio->bi_private = bbio->private;
4583                 bio->bi_end_io = bbio->end_io;
4584                 bio->bi_bdev = (struct block_device *)
4585                         (unsigned long)bbio->mirror_num;
4586                 bio->bi_sector = logical >> 9;
4587                 kfree(bbio);
4588                 bio_endio(bio, -EIO);
4589         }
4590 }
4591
4592 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
4593                   int mirror_num, int async_submit)
4594 {
4595         struct btrfs_device *dev;
4596         struct bio *first_bio = bio;
4597         u64 logical = (u64)bio->bi_sector << 9;
4598         u64 length = 0;
4599         u64 map_length;
4600         int ret;
4601         int dev_nr = 0;
4602         int total_devs = 1;
4603         struct btrfs_bio *bbio = NULL;
4604
4605         length = bio->bi_size;
4606         map_length = length;
4607
4608         ret = btrfs_map_block(root->fs_info, rw, logical, &map_length, &bbio,
4609                               mirror_num);
4610         if (ret)
4611                 return ret;
4612
4613         total_devs = bbio->num_stripes;
4614         if (map_length < length) {
4615                 printk(KERN_CRIT "btrfs: mapping failed logical %llu bio len %llu "
4616                        "len %llu\n", (unsigned long long)logical,
4617                        (unsigned long long)length,
4618                        (unsigned long long)map_length);
4619                 BUG();
4620         }
4621
4622         bbio->orig_bio = first_bio;
4623         bbio->private = first_bio->bi_private;
4624         bbio->end_io = first_bio->bi_end_io;
4625         atomic_set(&bbio->stripes_pending, bbio->num_stripes);
4626
4627         while (dev_nr < total_devs) {
4628                 dev = bbio->stripes[dev_nr].dev;
4629                 if (!dev || !dev->bdev || (rw & WRITE && !dev->writeable)) {
4630                         bbio_error(bbio, first_bio, logical);
4631                         dev_nr++;
4632                         continue;
4633                 }
4634
4635                 /*
4636                  * Check and see if we're ok with this bio based on it's size
4637                  * and offset with the given device.
4638                  */
4639                 if (!bio_size_ok(dev->bdev, first_bio,
4640                                  bbio->stripes[dev_nr].physical >> 9)) {
4641                         ret = breakup_stripe_bio(root, bbio, first_bio, dev,
4642                                                  dev_nr, rw, async_submit);
4643                         BUG_ON(ret);
4644                         dev_nr++;
4645                         continue;
4646                 }
4647
4648                 if (dev_nr < total_devs - 1) {
4649                         bio = bio_clone(first_bio, GFP_NOFS);
4650                         BUG_ON(!bio); /* -ENOMEM */
4651                 } else {
4652                         bio = first_bio;
4653                 }
4654
4655                 submit_stripe_bio(root, bbio, bio,
4656                                   bbio->stripes[dev_nr].physical, dev_nr, rw,
4657                                   async_submit);
4658                 dev_nr++;
4659         }
4660         return 0;
4661 }
4662
4663 struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid,
4664                                        u8 *uuid, u8 *fsid)
4665 {
4666         struct btrfs_device *device;
4667         struct btrfs_fs_devices *cur_devices;
4668
4669         cur_devices = fs_info->fs_devices;
4670         while (cur_devices) {
4671                 if (!fsid ||
4672                     !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
4673                         device = __find_device(&cur_devices->devices,
4674                                                devid, uuid);
4675                         if (device)
4676                                 return device;
4677                 }
4678                 cur_devices = cur_devices->seed;
4679         }
4680         return NULL;
4681 }
4682
4683 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
4684                                             u64 devid, u8 *dev_uuid)
4685 {
4686         struct btrfs_device *device;
4687         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
4688
4689         device = kzalloc(sizeof(*device), GFP_NOFS);
4690         if (!device)
4691                 return NULL;
4692         list_add(&device->dev_list,
4693                  &fs_devices->devices);
4694         device->dev_root = root->fs_info->dev_root;
4695         device->devid = devid;
4696         device->work.func = pending_bios_fn;
4697         device->fs_devices = fs_devices;
4698         device->missing = 1;
4699         fs_devices->num_devices++;
4700         fs_devices->missing_devices++;
4701         spin_lock_init(&device->io_lock);
4702         INIT_LIST_HEAD(&device->dev_alloc_list);
4703         memcpy(device->uuid, dev_uuid, BTRFS_UUID_SIZE);
4704         return device;
4705 }
4706
4707 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
4708                           struct extent_buffer *leaf,
4709                           struct btrfs_chunk *chunk)
4710 {
4711         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
4712         struct map_lookup *map;
4713         struct extent_map *em;
4714         u64 logical;
4715         u64 length;
4716         u64 devid;
4717         u8 uuid[BTRFS_UUID_SIZE];
4718         int num_stripes;
4719         int ret;
4720         int i;
4721
4722         logical = key->offset;
4723         length = btrfs_chunk_length(leaf, chunk);
4724
4725         read_lock(&map_tree->map_tree.lock);
4726         em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
4727         read_unlock(&map_tree->map_tree.lock);
4728
4729         /* already mapped? */
4730         if (em && em->start <= logical && em->start + em->len > logical) {
4731                 free_extent_map(em);
4732                 return 0;
4733         } else if (em) {
4734                 free_extent_map(em);
4735         }
4736
4737         em = alloc_extent_map();
4738         if (!em)
4739                 return -ENOMEM;
4740         num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
4741         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
4742         if (!map) {
4743                 free_extent_map(em);
4744                 return -ENOMEM;
4745         }
4746
4747         em->bdev = (struct block_device *)map;
4748         em->start = logical;
4749         em->len = length;
4750         em->block_start = 0;
4751         em->block_len = em->len;
4752
4753         map->num_stripes = num_stripes;
4754         map->io_width = btrfs_chunk_io_width(leaf, chunk);
4755         map->io_align = btrfs_chunk_io_align(leaf, chunk);
4756         map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
4757         map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
4758         map->type = btrfs_chunk_type(leaf, chunk);
4759         map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
4760         for (i = 0; i < num_stripes; i++) {
4761                 map->stripes[i].physical =
4762                         btrfs_stripe_offset_nr(leaf, chunk, i);
4763                 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
4764                 read_extent_buffer(leaf, uuid, (unsigned long)
4765                                    btrfs_stripe_dev_uuid_nr(chunk, i),
4766                                    BTRFS_UUID_SIZE);
4767                 map->stripes[i].dev = btrfs_find_device(root->fs_info, devid,
4768                                                         uuid, NULL);
4769                 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
4770                         kfree(map);
4771                         free_extent_map(em);
4772                         return -EIO;
4773                 }
4774                 if (!map->stripes[i].dev) {
4775                         map->stripes[i].dev =
4776                                 add_missing_dev(root, devid, uuid);
4777                         if (!map->stripes[i].dev) {
4778                                 kfree(map);
4779                                 free_extent_map(em);
4780                                 return -EIO;
4781                         }
4782                 }
4783                 map->stripes[i].dev->in_fs_metadata = 1;
4784         }
4785
4786         write_lock(&map_tree->map_tree.lock);
4787         ret = add_extent_mapping(&map_tree->map_tree, em);
4788         write_unlock(&map_tree->map_tree.lock);
4789         BUG_ON(ret); /* Tree corruption */
4790         free_extent_map(em);
4791
4792         return 0;
4793 }
4794
4795 static void fill_device_from_item(struct extent_buffer *leaf,
4796                                  struct btrfs_dev_item *dev_item,
4797                                  struct btrfs_device *device)
4798 {
4799         unsigned long ptr;
4800
4801         device->devid = btrfs_device_id(leaf, dev_item);
4802         device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
4803         device->total_bytes = device->disk_total_bytes;
4804         device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
4805         device->type = btrfs_device_type(leaf, dev_item);
4806         device->io_align = btrfs_device_io_align(leaf, dev_item);
4807         device->io_width = btrfs_device_io_width(leaf, dev_item);
4808         device->sector_size = btrfs_device_sector_size(leaf, dev_item);
4809         WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
4810         device->is_tgtdev_for_dev_replace = 0;
4811
4812         ptr = (unsigned long)btrfs_device_uuid(dev_item);
4813         read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
4814 }
4815
4816 static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
4817 {
4818         struct btrfs_fs_devices *fs_devices;
4819         int ret;
4820
4821         BUG_ON(!mutex_is_locked(&uuid_mutex));
4822
4823         fs_devices = root->fs_info->fs_devices->seed;
4824         while (fs_devices) {
4825                 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
4826                         ret = 0;
4827                         goto out;
4828                 }
4829                 fs_devices = fs_devices->seed;
4830         }
4831
4832         fs_devices = find_fsid(fsid);
4833         if (!fs_devices) {
4834                 ret = -ENOENT;
4835                 goto out;
4836         }
4837
4838         fs_devices = clone_fs_devices(fs_devices);
4839         if (IS_ERR(fs_devices)) {
4840                 ret = PTR_ERR(fs_devices);
4841                 goto out;
4842         }
4843
4844         ret = __btrfs_open_devices(fs_devices, FMODE_READ,
4845                                    root->fs_info->bdev_holder);
4846         if (ret) {
4847                 free_fs_devices(fs_devices);
4848                 goto out;
4849         }
4850
4851         if (!fs_devices->seeding) {
4852                 __btrfs_close_devices(fs_devices);
4853                 free_fs_devices(fs_devices);
4854                 ret = -EINVAL;
4855                 goto out;
4856         }
4857
4858         fs_devices->seed = root->fs_info->fs_devices->seed;
4859         root->fs_info->fs_devices->seed = fs_devices;
4860 out:
4861         return ret;
4862 }
4863
4864 static int read_one_dev(struct btrfs_root *root,
4865                         struct extent_buffer *leaf,
4866                         struct btrfs_dev_item *dev_item)
4867 {
4868         struct btrfs_device *device;
4869         u64 devid;
4870         int ret;
4871         u8 fs_uuid[BTRFS_UUID_SIZE];
4872         u8 dev_uuid[BTRFS_UUID_SIZE];
4873
4874         devid = btrfs_device_id(leaf, dev_item);
4875         read_extent_buffer(leaf, dev_uuid,
4876                            (unsigned long)btrfs_device_uuid(dev_item),
4877                            BTRFS_UUID_SIZE);
4878         read_extent_buffer(leaf, fs_uuid,
4879                            (unsigned long)btrfs_device_fsid(dev_item),
4880                            BTRFS_UUID_SIZE);
4881
4882         if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
4883                 ret = open_seed_devices(root, fs_uuid);
4884                 if (ret && !btrfs_test_opt(root, DEGRADED))
4885                         return ret;
4886         }
4887
4888         device = btrfs_find_device(root->fs_info, devid, dev_uuid, fs_uuid);
4889         if (!device || !device->bdev) {
4890                 if (!btrfs_test_opt(root, DEGRADED))
4891                         return -EIO;
4892
4893                 if (!device) {
4894                         printk(KERN_WARNING "warning devid %llu missing\n",
4895                                (unsigned long long)devid);
4896                         device = add_missing_dev(root, devid, dev_uuid);
4897                         if (!device)
4898                                 return -ENOMEM;
4899                 } else if (!device->missing) {
4900                         /*
4901                          * this happens when a device that was properly setup
4902                          * in the device info lists suddenly goes bad.
4903                          * device->bdev is NULL, and so we have to set
4904                          * device->missing to one here
4905                          */
4906                         root->fs_info->fs_devices->missing_devices++;
4907                         device->missing = 1;
4908                 }
4909         }
4910
4911         if (device->fs_devices != root->fs_info->fs_devices) {
4912                 BUG_ON(device->writeable);
4913                 if (device->generation !=
4914                     btrfs_device_generation(leaf, dev_item))
4915                         return -EINVAL;
4916         }
4917
4918         fill_device_from_item(leaf, dev_item, device);
4919         device->dev_root = root->fs_info->dev_root;
4920         device->in_fs_metadata = 1;
4921         if (device->writeable && !device->is_tgtdev_for_dev_replace) {
4922                 device->fs_devices->total_rw_bytes += device->total_bytes;
4923                 spin_lock(&root->fs_info->free_chunk_lock);
4924                 root->fs_info->free_chunk_space += device->total_bytes -
4925                         device->bytes_used;
4926                 spin_unlock(&root->fs_info->free_chunk_lock);
4927         }
4928         ret = 0;
4929         return ret;
4930 }
4931
4932 int btrfs_read_sys_array(struct btrfs_root *root)
4933 {
4934         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
4935         struct extent_buffer *sb;
4936         struct btrfs_disk_key *disk_key;
4937         struct btrfs_chunk *chunk;
4938         u8 *ptr;
4939         unsigned long sb_ptr;
4940         int ret = 0;
4941         u32 num_stripes;
4942         u32 array_size;
4943         u32 len = 0;
4944         u32 cur;
4945         struct btrfs_key key;
4946
4947         sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
4948                                           BTRFS_SUPER_INFO_SIZE);
4949         if (!sb)
4950                 return -ENOMEM;
4951         btrfs_set_buffer_uptodate(sb);
4952         btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
4953         /*
4954          * The sb extent buffer is artifical and just used to read the system array.
4955          * btrfs_set_buffer_uptodate() call does not properly mark all it's
4956          * pages up-to-date when the page is larger: extent does not cover the
4957          * whole page and consequently check_page_uptodate does not find all
4958          * the page's extents up-to-date (the hole beyond sb),
4959          * write_extent_buffer then triggers a WARN_ON.
4960          *
4961          * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
4962          * but sb spans only this function. Add an explicit SetPageUptodate call
4963          * to silence the warning eg. on PowerPC 64.
4964          */
4965         if (PAGE_CACHE_SIZE > BTRFS_SUPER_INFO_SIZE)
4966                 SetPageUptodate(sb->pages[0]);
4967
4968         write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
4969         array_size = btrfs_super_sys_array_size(super_copy);
4970
4971         ptr = super_copy->sys_chunk_array;
4972         sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
4973         cur = 0;
4974
4975         while (cur < array_size) {
4976                 disk_key = (struct btrfs_disk_key *)ptr;
4977                 btrfs_disk_key_to_cpu(&key, disk_key);
4978
4979                 len = sizeof(*disk_key); ptr += len;
4980                 sb_ptr += len;
4981                 cur += len;
4982
4983                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
4984                         chunk = (struct btrfs_chunk *)sb_ptr;
4985                         ret = read_one_chunk(root, &key, sb, chunk);
4986                         if (ret)
4987                                 break;
4988                         num_stripes = btrfs_chunk_num_stripes(sb, chunk);
4989                         len = btrfs_chunk_item_size(num_stripes);
4990                 } else {
4991                         ret = -EIO;
4992                         break;
4993                 }
4994                 ptr += len;
4995                 sb_ptr += len;
4996                 cur += len;
4997         }
4998         free_extent_buffer(sb);
4999         return ret;
5000 }
5001
5002 int btrfs_read_chunk_tree(struct btrfs_root *root)
5003 {
5004         struct btrfs_path *path;
5005         struct extent_buffer *leaf;
5006         struct btrfs_key key;
5007         struct btrfs_key found_key;
5008         int ret;
5009         int slot;
5010
5011         root = root->fs_info->chunk_root;
5012
5013         path = btrfs_alloc_path();
5014         if (!path)
5015                 return -ENOMEM;
5016
5017         mutex_lock(&uuid_mutex);
5018         lock_chunks(root);
5019
5020         /* first we search for all of the device items, and then we
5021          * read in all of the chunk items.  This way we can create chunk
5022          * mappings that reference all of the devices that are afound
5023          */
5024         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
5025         key.offset = 0;
5026         key.type = 0;
5027 again:
5028         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5029         if (ret < 0)
5030                 goto error;
5031         while (1) {
5032                 leaf = path->nodes[0];
5033                 slot = path->slots[0];
5034                 if (slot >= btrfs_header_nritems(leaf)) {
5035                         ret = btrfs_next_leaf(root, path);
5036                         if (ret == 0)
5037                                 continue;
5038                         if (ret < 0)
5039                                 goto error;
5040                         break;
5041                 }
5042                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5043                 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
5044                         if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID)
5045                                 break;
5046                         if (found_key.type == BTRFS_DEV_ITEM_KEY) {
5047                                 struct btrfs_dev_item *dev_item;
5048                                 dev_item = btrfs_item_ptr(leaf, slot,
5049                                                   struct btrfs_dev_item);
5050                                 ret = read_one_dev(root, leaf, dev_item);
5051                                 if (ret)
5052                                         goto error;
5053                         }
5054                 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
5055                         struct btrfs_chunk *chunk;
5056                         chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
5057                         ret = read_one_chunk(root, &found_key, leaf, chunk);
5058                         if (ret)
5059                                 goto error;
5060                 }
5061                 path->slots[0]++;
5062         }
5063         if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
5064                 key.objectid = 0;
5065                 btrfs_release_path(path);
5066                 goto again;
5067         }
5068         ret = 0;
5069 error:
5070         unlock_chunks(root);
5071         mutex_unlock(&uuid_mutex);
5072
5073         btrfs_free_path(path);
5074         return ret;
5075 }
5076
5077 static void __btrfs_reset_dev_stats(struct btrfs_device *dev)
5078 {
5079         int i;
5080
5081         for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
5082                 btrfs_dev_stat_reset(dev, i);
5083 }
5084
5085 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
5086 {
5087         struct btrfs_key key;
5088         struct btrfs_key found_key;
5089         struct btrfs_root *dev_root = fs_info->dev_root;
5090         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
5091         struct extent_buffer *eb;
5092         int slot;
5093         int ret = 0;
5094         struct btrfs_device *device;
5095         struct btrfs_path *path = NULL;
5096         int i;
5097
5098         path = btrfs_alloc_path();
5099         if (!path) {
5100                 ret = -ENOMEM;
5101                 goto out;
5102         }
5103
5104         mutex_lock(&fs_devices->device_list_mutex);
5105         list_for_each_entry(device, &fs_devices->devices, dev_list) {
5106                 int item_size;
5107                 struct btrfs_dev_stats_item *ptr;
5108
5109                 key.objectid = 0;
5110                 key.type = BTRFS_DEV_STATS_KEY;
5111                 key.offset = device->devid;
5112                 ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
5113                 if (ret) {
5114                         __btrfs_reset_dev_stats(device);
5115                         device->dev_stats_valid = 1;
5116                         btrfs_release_path(path);
5117                         continue;
5118                 }
5119                 slot = path->slots[0];
5120                 eb = path->nodes[0];
5121                 btrfs_item_key_to_cpu(eb, &found_key, slot);
5122                 item_size = btrfs_item_size_nr(eb, slot);
5123
5124                 ptr = btrfs_item_ptr(eb, slot,
5125                                      struct btrfs_dev_stats_item);
5126
5127                 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
5128                         if (item_size >= (1 + i) * sizeof(__le64))
5129                                 btrfs_dev_stat_set(device, i,
5130                                         btrfs_dev_stats_value(eb, ptr, i));
5131                         else
5132                                 btrfs_dev_stat_reset(device, i);
5133                 }
5134
5135                 device->dev_stats_valid = 1;
5136                 btrfs_dev_stat_print_on_load(device);
5137                 btrfs_release_path(path);
5138         }
5139         mutex_unlock(&fs_devices->device_list_mutex);
5140
5141 out:
5142         btrfs_free_path(path);
5143         return ret < 0 ? ret : 0;
5144 }
5145
5146 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
5147                                 struct btrfs_root *dev_root,
5148                                 struct btrfs_device *device)
5149 {
5150         struct btrfs_path *path;
5151         struct btrfs_key key;
5152         struct extent_buffer *eb;
5153         struct btrfs_dev_stats_item *ptr;
5154         int ret;
5155         int i;
5156
5157         key.objectid = 0;
5158         key.type = BTRFS_DEV_STATS_KEY;
5159         key.offset = device->devid;
5160
5161         path = btrfs_alloc_path();
5162         BUG_ON(!path);
5163         ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
5164         if (ret < 0) {
5165                 printk_in_rcu(KERN_WARNING "btrfs: error %d while searching for dev_stats item for device %s!\n",
5166                               ret, rcu_str_deref(device->name));
5167                 goto out;
5168         }
5169
5170         if (ret == 0 &&
5171             btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
5172                 /* need to delete old one and insert a new one */
5173                 ret = btrfs_del_item(trans, dev_root, path);
5174                 if (ret != 0) {
5175                         printk_in_rcu(KERN_WARNING "btrfs: delete too small dev_stats item for device %s failed %d!\n",
5176                                       rcu_str_deref(device->name), ret);
5177                         goto out;
5178                 }
5179                 ret = 1;
5180         }
5181
5182         if (ret == 1) {
5183                 /* need to insert a new item */
5184                 btrfs_release_path(path);
5185                 ret = btrfs_insert_empty_item(trans, dev_root, path,
5186                                               &key, sizeof(*ptr));
5187                 if (ret < 0) {
5188                         printk_in_rcu(KERN_WARNING "btrfs: insert dev_stats item for device %s failed %d!\n",
5189                                       rcu_str_deref(device->name), ret);
5190                         goto out;
5191                 }
5192         }
5193
5194         eb = path->nodes[0];
5195         ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
5196         for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
5197                 btrfs_set_dev_stats_value(eb, ptr, i,
5198                                           btrfs_dev_stat_read(device, i));
5199         btrfs_mark_buffer_dirty(eb);
5200
5201 out:
5202         btrfs_free_path(path);
5203         return ret;
5204 }
5205
5206 /*
5207  * called from commit_transaction. Writes all changed device stats to disk.
5208  */
5209 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans,
5210                         struct btrfs_fs_info *fs_info)
5211 {
5212         struct btrfs_root *dev_root = fs_info->dev_root;
5213         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
5214         struct btrfs_device *device;
5215         int ret = 0;
5216
5217         mutex_lock(&fs_devices->device_list_mutex);
5218         list_for_each_entry(device, &fs_devices->devices, dev_list) {
5219                 if (!device->dev_stats_valid || !device->dev_stats_dirty)
5220                         continue;
5221
5222                 ret = update_dev_stat_item(trans, dev_root, device);
5223                 if (!ret)
5224                         device->dev_stats_dirty = 0;
5225         }
5226         mutex_unlock(&fs_devices->device_list_mutex);
5227
5228         return ret;
5229 }
5230
5231 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
5232 {
5233         btrfs_dev_stat_inc(dev, index);
5234         btrfs_dev_stat_print_on_error(dev);
5235 }
5236
5237 void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
5238 {
5239         if (!dev->dev_stats_valid)
5240                 return;
5241         printk_ratelimited_in_rcu(KERN_ERR
5242                            "btrfs: bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
5243                            rcu_str_deref(dev->name),
5244                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
5245                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
5246                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
5247                            btrfs_dev_stat_read(dev,
5248                                                BTRFS_DEV_STAT_CORRUPTION_ERRS),
5249                            btrfs_dev_stat_read(dev,
5250                                                BTRFS_DEV_STAT_GENERATION_ERRS));
5251 }
5252
5253 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
5254 {
5255         int i;
5256
5257         for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
5258                 if (btrfs_dev_stat_read(dev, i) != 0)
5259                         break;
5260         if (i == BTRFS_DEV_STAT_VALUES_MAX)
5261                 return; /* all values == 0, suppress message */
5262
5263         printk_in_rcu(KERN_INFO "btrfs: bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
5264                rcu_str_deref(dev->name),
5265                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
5266                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
5267                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
5268                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
5269                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
5270 }
5271
5272 int btrfs_get_dev_stats(struct btrfs_root *root,
5273                         struct btrfs_ioctl_get_dev_stats *stats)
5274 {
5275         struct btrfs_device *dev;
5276         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
5277         int i;
5278
5279         mutex_lock(&fs_devices->device_list_mutex);
5280         dev = btrfs_find_device(root->fs_info, stats->devid, NULL, NULL);
5281         mutex_unlock(&fs_devices->device_list_mutex);
5282
5283         if (!dev) {
5284                 printk(KERN_WARNING
5285                        "btrfs: get dev_stats failed, device not found\n");
5286                 return -ENODEV;
5287         } else if (!dev->dev_stats_valid) {
5288                 printk(KERN_WARNING
5289                        "btrfs: get dev_stats failed, not yet valid\n");
5290                 return -ENODEV;
5291         } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
5292                 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
5293                         if (stats->nr_items > i)
5294                                 stats->values[i] =
5295                                         btrfs_dev_stat_read_and_reset(dev, i);
5296                         else
5297                                 btrfs_dev_stat_reset(dev, i);
5298                 }
5299         } else {
5300                 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
5301                         if (stats->nr_items > i)
5302                                 stats->values[i] = btrfs_dev_stat_read(dev, i);
5303         }
5304         if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
5305                 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
5306         return 0;
5307 }
5308
5309 int btrfs_scratch_superblock(struct btrfs_device *device)
5310 {
5311         struct buffer_head *bh;
5312         struct btrfs_super_block *disk_super;
5313
5314         bh = btrfs_read_dev_super(device->bdev);
5315         if (!bh)
5316                 return -EINVAL;
5317         disk_super = (struct btrfs_super_block *)bh->b_data;
5318
5319         memset(&disk_super->magic, 0, sizeof(disk_super->magic));
5320         set_buffer_dirty(bh);
5321         sync_dirty_buffer(bh);
5322         brelse(bh);
5323
5324         return 0;
5325 }