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 <asm/div64.h>
  27 #include "compat.h"
  28 #include "ctree.h"
  29 #include "extent_map.h"
  30 #include "disk-io.h"
  31 #include "transaction.h"
  32 #include "print-tree.h"
  33 #include "volumes.h"
  34 #include "async-thread.h"
  35
  36 static int init_first_rw_device(struct btrfs_trans_handle *trans,
  37                                 struct btrfs_root *root,
  38                                 struct btrfs_device *device);
  39 static int btrfs_relocate_sys_chunks(struct btrfs_root *root);
  40
  41 static DEFINE_MUTEX(uuid_mutex);
  42 static LIST_HEAD(fs_uuids);
  43
  44 static void lock_chunks(struct btrfs_root *root)
  45 {
  46         mutex_lock(&root->fs_info->chunk_mutex);
  47 }
  48
  49 static void unlock_chunks(struct btrfs_root *root)
  50 {
  51         mutex_unlock(&root->fs_info->chunk_mutex);
  52 }
  53
  54 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
  55 {
  56         struct btrfs_device *device;
  57         WARN_ON(fs_devices->opened);
  58         while (!list_empty(&fs_devices->devices)) {
  59                 device = list_entry(fs_devices->devices.next,
  60                                     struct btrfs_device, dev_list);
  61                 list_del(&device->dev_list);
  62                 kfree(device->name);
  63                 kfree(device);
  64         }
  65         kfree(fs_devices);
  66 }
  67
  68 int btrfs_cleanup_fs_uuids(void)
  69 {
  70         struct btrfs_fs_devices *fs_devices;
  71
  72         while (!list_empty(&fs_uuids)) {
  73                 fs_devices = list_entry(fs_uuids.next,
  74                                         struct btrfs_fs_devices, list);
  75                 list_del(&fs_devices->list);
  76                 free_fs_devices(fs_devices);
  77         }
  78         return 0;
  79 }
  80
  81 static noinline struct btrfs_device *__find_device(struct list_head *head,
  82                                                    u64 devid, u8 *uuid)
  83 {
  84         struct btrfs_device *dev;
  85
  86         list_for_each_entry(dev, head, dev_list) {
  87                 if (dev->devid == devid &&
  88                     (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
  89                         return dev;
  90                 }
  91         }
  92         return NULL;
  93 }
  94
  95 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
  96 {
  97         struct btrfs_fs_devices *fs_devices;
  98
  99         list_for_each_entry(fs_devices, &fs_uuids, list) {
 100                 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
 101                         return fs_devices;
 102         }
 103         return NULL;
 104 }
 105
 106 static void requeue_list(struct btrfs_pending_bios *pending_bios,
 107                         struct bio *head, struct bio *tail)
 108 {
 109
 110         struct bio *old_head;
 111
 112         old_head = pending_bios->head;
 113         pending_bios->head = head;
 114         if (pending_bios->tail)
 115                 tail->bi_next = old_head;
 116         else
 117                 pending_bios->tail = tail;
 118 }
 119
 120 /*
 121  * we try to collect pending bios for a device so we don't get a large
 122  * number of procs sending bios down to the same device.  This greatly
 123  * improves the schedulers ability to collect and merge the bios.
 124  *
 125  * But, it also turns into a long list of bios to process and that is sure
 126  * to eventually make the worker thread block.  The solution here is to
 127  * make some progress and then put this work struct back at the end of
 128  * the list if the block device is congested.  This way, multiple devices
 129  * can make progress from a single worker thread.
 130  */
 131 static noinline int run_scheduled_bios(struct btrfs_device *device)
 132 {
 133         struct bio *pending;
 134         struct backing_dev_info *bdi;
 135         struct btrfs_fs_info *fs_info;
 136         struct btrfs_pending_bios *pending_bios;
 137         struct bio *tail;
 138         struct bio *cur;
 139         int again = 0;
 140         unsigned long num_run;
 141         unsigned long batch_run = 0;
 142         unsigned long limit;
 143         unsigned long last_waited = 0;
 144         int force_reg = 0;
 145         int sync_pending = 0;
 146         struct blk_plug plug;
 147
 148         /*
 149          * this function runs all the bios we've collected for
 150          * a particular device.  We don't want to wander off to
 151          * another device without first sending all of these down.
 152          * So, setup a plug here and finish it off before we return
 153          */
 154         blk_start_plug(&plug);
 155
 156         bdi = blk_get_backing_dev_info(device->bdev);
 157         fs_info = device->dev_root->fs_info;
 158         limit = btrfs_async_submit_limit(fs_info);
 159         limit = limit * 2 / 3;
 160
 161 loop:
 162         spin_lock(&device->io_lock);
 163
 164 loop_lock:
 165         num_run = 0;
 166
 167         /* take all the bios off the list at once and process them
 168          * later on (without the lock held).  But, remember the
 169          * tail and other pointers so the bios can be properly reinserted
 170          * into the list if we hit congestion
 171          */
 172         if (!force_reg && device->pending_sync_bios.head) {
 173                 pending_bios = &device->pending_sync_bios;
 174                 force_reg = 1;
 175         } else {
 176                 pending_bios = &device->pending_bios;
 177                 force_reg = 0;
 178         }
 179
 180         pending = pending_bios->head;
 181         tail = pending_bios->tail;
 182         WARN_ON(pending && !tail);
 183
 184         /*
 185          * if pending was null this time around, no bios need processing
 186          * at all and we can stop.  Otherwise it'll loop back up again
 187          * and do an additional check so no bios are missed.
 188          *
 189          * device->running_pending is used to synchronize with the
 190          * schedule_bio code.
 191          */
 192         if (device->pending_sync_bios.head == NULL &&
 193             device->pending_bios.head == NULL) {
 194                 again = 0;
 195                 device->running_pending = 0;
 196         } else {
 197                 again = 1;
 198                 device->running_pending = 1;
 199         }
 200
 201         pending_bios->head = NULL;
 202         pending_bios->tail = NULL;
 203
 204         spin_unlock(&device->io_lock);
 205
 206         while (pending) {
 207
 208                 rmb();
 209                 /* we want to work on both lists, but do more bios on the
 210                  * sync list than the regular list
 211                  */
 212                 if ((num_run > 32 &&
 213                     pending_bios != &device->pending_sync_bios &&
 214                     device->pending_sync_bios.head) ||
 215                    (num_run > 64 && pending_bios == &device->pending_sync_bios &&
 216                     device->pending_bios.head)) {
 217                         spin_lock(&device->io_lock);
 218                         requeue_list(pending_bios, pending, tail);
 219                         goto loop_lock;
 220                 }
 221
 222                 cur = pending;
 223                 pending = pending->bi_next;
 224                 cur->bi_next = NULL;
 225                 atomic_dec(&fs_info->nr_async_bios);
 226
 227                 if (atomic_read(&fs_info->nr_async_bios) < limit &&
 228                     waitqueue_active(&fs_info->async_submit_wait))
 229                         wake_up(&fs_info->async_submit_wait);
 230
 231                 BUG_ON(atomic_read(&cur->bi_cnt) == 0);
 232
 233                 /*
 234                  * if we're doing the sync list, record that our
 235                  * plug has some sync requests on it
 236                  *
 237                  * If we're doing the regular list and there are
 238                  * sync requests sitting around, unplug before
 239                  * we add more
 240                  */
 241                 if (pending_bios == &device->pending_sync_bios) {
 242                         sync_pending = 1;
 243                 } else if (sync_pending) {
 244                         blk_finish_plug(&plug);
 245                         blk_start_plug(&plug);
 246                         sync_pending = 0;
 247                 }
 248
 249                 submit_bio(cur->bi_rw, cur);
 250                 num_run++;
 251                 batch_run++;
 252                 if (need_resched())
 253                         cond_resched();
 254
 255                 /*
 256                  * we made progress, there is more work to do and the bdi
 257                  * is now congested.  Back off and let other work structs
 258                  * run instead
 259                  */
 260                 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
 261                     fs_info->fs_devices->open_devices > 1) {
 262                         struct io_context *ioc;
 263
 264                         ioc = current->io_context;
 265
 266                         /*
 267                          * the main goal here is that we don't want to
 268                          * block if we're going to be able to submit
 269                          * more requests without blocking.
 270                          *
 271                          * This code does two great things, it pokes into
 272                          * the elevator code from a filesystem _and_
 273                          * it makes assumptions about how batching works.
 274                          */
 275                         if (ioc && ioc->nr_batch_requests > 0 &&
 276                             time_before(jiffies, ioc->last_waited + HZ/50UL) &&
 277                             (last_waited == 0 ||
 278                              ioc->last_waited == last_waited)) {
 279                                 /*
 280                                  * we want to go through our batch of
 281                                  * requests and stop.  So, we copy out
 282                                  * the ioc->last_waited time and test
 283                                  * against it before looping
 284                                  */
 285                                 last_waited = ioc->last_waited;
 286                                 if (need_resched())
 287                                         cond_resched();
 288                                 continue;
 289                         }
 290                         spin_lock(&device->io_lock);
 291                         requeue_list(pending_bios, pending, tail);
 292                         device->running_pending = 1;
 293
 294                         spin_unlock(&device->io_lock);
 295                         btrfs_requeue_work(&device->work);
 296                         goto done;
 297                 }
 298                 /* unplug every 64 requests just for good measure */
 299                 if (batch_run % 64 == 0) {
 300                         blk_finish_plug(&plug);
 301                         blk_start_plug(&plug);
 302                         sync_pending = 0;
 303                 }
 304         }
 305
 306         cond_resched();
 307         if (again)
 308                 goto loop;
 309
 310         spin_lock(&device->io_lock);
 311         if (device->pending_bios.head || device->pending_sync_bios.head)
 312                 goto loop_lock;
 313         spin_unlock(&device->io_lock);
 314
 315 done:
 316         blk_finish_plug(&plug);
 317         return 0;
 318 }
 319
 320 static void pending_bios_fn(struct btrfs_work *work)
 321 {
 322         struct btrfs_device *device;
 323
 324         device = container_of(work, struct btrfs_device, work);
 325         run_scheduled_bios(device);
 326 }
 327
 328 static noinline int device_list_add(const char *path,
 329                            struct btrfs_super_block *disk_super,
 330                            u64 devid, struct btrfs_fs_devices **fs_devices_ret)
 331 {
 332         struct btrfs_device *device;
 333         struct btrfs_fs_devices *fs_devices;
 334         u64 found_transid = btrfs_super_generation(disk_super);
 335         char *name;
 336
 337         fs_devices = find_fsid(disk_super->fsid);
 338         if (!fs_devices) {
 339                 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
 340                 if (!fs_devices)
 341                         return -ENOMEM;
 342                 INIT_LIST_HEAD(&fs_devices->devices);
 343                 INIT_LIST_HEAD(&fs_devices->alloc_list);
 344                 list_add(&fs_devices->list, &fs_uuids);
 345                 memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
 346                 fs_devices->latest_devid = devid;
 347                 fs_devices->latest_trans = found_transid;
 348                 mutex_init(&fs_devices->device_list_mutex);
 349                 device = NULL;
 350         } else {
 351                 device = __find_device(&fs_devices->devices, devid,
 352                                        disk_super->dev_item.uuid);
 353         }
 354         if (!device) {
 355                 if (fs_devices->opened)
 356                         return -EBUSY;
 357
 358                 device = kzalloc(sizeof(*device), GFP_NOFS);
 359                 if (!device) {
 360                         /* we can safely leave the fs_devices entry around */
 361                         return -ENOMEM;
 362                 }
 363                 device->devid = devid;
 364                 device->work.func = pending_bios_fn;
 365                 memcpy(device->uuid, disk_super->dev_item.uuid,
 366                        BTRFS_UUID_SIZE);
 367                 spin_lock_init(&device->io_lock);
 368                 device->name = kstrdup(path, GFP_NOFS);
 369                 if (!device->name) {
 370                         kfree(device);
 371                         return -ENOMEM;
 372                 }
 373                 INIT_LIST_HEAD(&device->dev_alloc_list);
 374
 375                 /* init readahead state */
 376                 spin_lock_init(&device->reada_lock);
 377                 device->reada_curr_zone = NULL;
 378                 atomic_set(&device->reada_in_flight, 0);
 379                 device->reada_next = 0;
 380                 INIT_RADIX_TREE(&device->reada_zones, GFP_NOFS & ~__GFP_WAIT);
 381                 INIT_RADIX_TREE(&device->reada_extents, GFP_NOFS & ~__GFP_WAIT);
 382
 383                 mutex_lock(&fs_devices->device_list_mutex);
 384                 list_add_rcu(&device->dev_list, &fs_devices->devices);
 385                 mutex_unlock(&fs_devices->device_list_mutex);
 386
 387                 device->fs_devices = fs_devices;
 388                 fs_devices->num_devices++;
 389         } else if (!device->name || strcmp(device->name, path)) {
 390                 name = kstrdup(path, GFP_NOFS);
 391                 if (!name)
 392                         return -ENOMEM;
 393                 kfree(device->name);
 394                 device->name = name;
 395                 if (device->missing) {
 396                         fs_devices->missing_devices--;
 397                         device->missing = 0;
 398                 }
 399         }
 400
 401         if (found_transid > fs_devices->latest_trans) {
 402                 fs_devices->latest_devid = devid;
 403                 fs_devices->latest_trans = found_transid;
 404         }
 405         *fs_devices_ret = fs_devices;
 406         return 0;
 407 }
 408
 409 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
 410 {
 411         struct btrfs_fs_devices *fs_devices;
 412         struct btrfs_device *device;
 413         struct btrfs_device *orig_dev;
 414
 415         fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
 416         if (!fs_devices)
 417                 return ERR_PTR(-ENOMEM);
 418
 419         INIT_LIST_HEAD(&fs_devices->devices);
 420         INIT_LIST_HEAD(&fs_devices->alloc_list);
 421         INIT_LIST_HEAD(&fs_devices->list);
 422         mutex_init(&fs_devices->device_list_mutex);
 423         fs_devices->latest_devid = orig->latest_devid;
 424         fs_devices->latest_trans = orig->latest_trans;
 425         memcpy(fs_devices->fsid, orig->fsid, sizeof(fs_devices->fsid));
 426
 427         /* We have held the volume lock, it is safe to get the devices. */
 428         list_for_each_entry(orig_dev, &orig->devices, dev_list) {
 429                 device = kzalloc(sizeof(*device), GFP_NOFS);
 430                 if (!device)
 431                         goto error;
 432
 433                 device->name = kstrdup(orig_dev->name, GFP_NOFS);
 434                 if (!device->name) {
 435                         kfree(device);
 436                         goto error;
 437                 }
 438
 439                 device->devid = orig_dev->devid;
 440                 device->work.func = pending_bios_fn;
 441                 memcpy(device->uuid, orig_dev->uuid, sizeof(device->uuid));
 442                 spin_lock_init(&device->io_lock);
 443                 INIT_LIST_HEAD(&device->dev_list);
 444                 INIT_LIST_HEAD(&device->dev_alloc_list);
 445
 446                 list_add(&device->dev_list, &fs_devices->devices);
 447                 device->fs_devices = fs_devices;
 448                 fs_devices->num_devices++;
 449         }
 450         return fs_devices;
 451 error:
 452         free_fs_devices(fs_devices);
 453         return ERR_PTR(-ENOMEM);
 454 }
 455
 456 int btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices)
 457 {
 458         struct btrfs_device *device, *next;
 459
 460         mutex_lock(&uuid_mutex);
 461 again:
 462         /* This is the initialized path, it is safe to release the devices. */
 463         list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
 464                 if (device->in_fs_metadata)
 465                         continue;
 466
 467                 if (device->bdev) {
 468                         blkdev_put(device->bdev, device->mode);
 469                         device->bdev = NULL;
 470                         fs_devices->open_devices--;
 471                 }
 472                 if (device->writeable) {
 473                         list_del_init(&device->dev_alloc_list);
 474                         device->writeable = 0;
 475                         fs_devices->rw_devices--;
 476                 }
 477                 list_del_init(&device->dev_list);
 478                 fs_devices->num_devices--;
 479                 kfree(device->name);
 480                 kfree(device);
 481         }
 482
 483         if (fs_devices->seed) {
 484                 fs_devices = fs_devices->seed;
 485                 goto again;
 486         }
 487
 488         mutex_unlock(&uuid_mutex);
 489         return 0;
 490 }
 491
 492 static void __free_device(struct work_struct *work)
 493 {
 494         struct btrfs_device *device;
 495
 496         device = container_of(work, struct btrfs_device, rcu_work);
 497
 498         if (device->bdev)
 499                 blkdev_put(device->bdev, device->mode);
 500
 501         kfree(device->name);
 502         kfree(device);
 503 }
 504
 505 static void free_device(struct rcu_head *head)
 506 {
 507         struct btrfs_device *device;
 508
 509         device = container_of(head, struct btrfs_device, rcu);
 510
 511         INIT_WORK(&device->rcu_work, __free_device);
 512         schedule_work(&device->rcu_work);
 513 }
 514
 515 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
 516 {
 517         struct btrfs_device *device;
 518
 519         if (--fs_devices->opened > 0)
 520                 return 0;
 521
 522         mutex_lock(&fs_devices->device_list_mutex);
 523         list_for_each_entry(device, &fs_devices->devices, dev_list) {
 524                 struct btrfs_device *new_device;
 525
 526                 if (device->bdev)
 527                         fs_devices->open_devices--;
 528
 529                 if (device->writeable) {
 530                         list_del_init(&device->dev_alloc_list);
 531                         fs_devices->rw_devices--;
 532                 }
 533
 534                 if (device->can_discard)
 535                         fs_devices->num_can_discard--;
 536
 537                 new_device = kmalloc(sizeof(*new_device), GFP_NOFS);
 538                 BUG_ON(!new_device);
 539                 memcpy(new_device, device, sizeof(*new_device));
 540                 new_device->name = kstrdup(device->name, GFP_NOFS);
 541                 BUG_ON(device->name && !new_device->name);
 542                 new_device->bdev = NULL;
 543                 new_device->writeable = 0;
 544                 new_device->in_fs_metadata = 0;
 545                 new_device->can_discard = 0;
 546                 list_replace_rcu(&device->dev_list, &new_device->dev_list);
 547
 548                 call_rcu(&device->rcu, free_device);
 549         }
 550         mutex_unlock(&fs_devices->device_list_mutex);
 551
 552         WARN_ON(fs_devices->open_devices);
 553         WARN_ON(fs_devices->rw_devices);
 554         fs_devices->opened = 0;
 555         fs_devices->seeding = 0;
 556
 557         return 0;
 558 }
 559
 560 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
 561 {
 562         struct btrfs_fs_devices *seed_devices = NULL;
 563         int ret;
 564
 565         mutex_lock(&uuid_mutex);
 566         ret = __btrfs_close_devices(fs_devices);
 567         if (!fs_devices->opened) {
 568                 seed_devices = fs_devices->seed;
 569                 fs_devices->seed = NULL;
 570         }
 571         mutex_unlock(&uuid_mutex);
 572
 573         while (seed_devices) {
 574                 fs_devices = seed_devices;
 575                 seed_devices = fs_devices->seed;
 576                 __btrfs_close_devices(fs_devices);
 577                 free_fs_devices(fs_devices);
 578         }
 579         return ret;
 580 }
 581
 582 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
 583                                 fmode_t flags, void *holder)
 584 {
 585         struct request_queue *q;
 586         struct block_device *bdev;
 587         struct list_head *head = &fs_devices->devices;
 588         struct btrfs_device *device;
 589         struct block_device *latest_bdev = NULL;
 590         struct buffer_head *bh;
 591         struct btrfs_super_block *disk_super;
 592         u64 latest_devid = 0;
 593         u64 latest_transid = 0;
 594         u64 devid;
 595         int seeding = 1;
 596         int ret = 0;
 597
 598         flags |= FMODE_EXCL;
 599
 600         list_for_each_entry(device, head, dev_list) {
 601                 if (device->bdev)
 602                         continue;
 603                 if (!device->name)
 604                         continue;
 605
 606                 bdev = blkdev_get_by_path(device->name, flags, holder);
 607                 if (IS_ERR(bdev)) {
 608                         printk(KERN_INFO "open %s failed\n", device->name);
 609                         goto error;
 610                 }
 611                 set_blocksize(bdev, 4096);
 612
 613                 bh = btrfs_read_dev_super(bdev);
 614                 if (!bh)
 615                         goto error_close;
 616
 617                 disk_super = (struct btrfs_super_block *)bh->b_data;
 618                 devid = btrfs_stack_device_id(&disk_super->dev_item);
 619                 if (devid != device->devid)
 620                         goto error_brelse;
 621
 622                 if (memcmp(device->uuid, disk_super->dev_item.uuid,
 623                            BTRFS_UUID_SIZE))
 624                         goto error_brelse;
 625
 626                 device->generation = btrfs_super_generation(disk_super);
 627                 if (!latest_transid || device->generation > latest_transid) {
 628                         latest_devid = devid;
 629                         latest_transid = device->generation;
 630                         latest_bdev = bdev;
 631                 }
 632
 633                 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
 634                         device->writeable = 0;
 635                 } else {
 636                         device->writeable = !bdev_read_only(bdev);
 637                         seeding = 0;
 638                 }
 639
 640                 q = bdev_get_queue(bdev);
 641                 if (blk_queue_discard(q)) {
 642                         device->can_discard = 1;
 643                         fs_devices->num_can_discard++;
 644                 }
 645
 646                 device->bdev = bdev;
 647                 device->in_fs_metadata = 0;
 648                 device->mode = flags;
 649
 650                 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
 651                         fs_devices->rotating = 1;
 652
 653                 fs_devices->open_devices++;
 654                 if (device->writeable) {
 655                         fs_devices->rw_devices++;
 656                         list_add(&device->dev_alloc_list,
 657                                  &fs_devices->alloc_list);
 658                 }
 659                 brelse(bh);
 660                 continue;
 661
 662 error_brelse:
 663                 brelse(bh);
 664 error_close:
 665                 blkdev_put(bdev, flags);
 666 error:
 667                 continue;
 668         }
 669         if (fs_devices->open_devices == 0) {
 670                 ret = -EINVAL;
 671                 goto out;
 672         }
 673         fs_devices->seeding = seeding;
 674         fs_devices->opened = 1;
 675         fs_devices->latest_bdev = latest_bdev;
 676         fs_devices->latest_devid = latest_devid;
 677         fs_devices->latest_trans = latest_transid;
 678         fs_devices->total_rw_bytes = 0;
 679 out:
 680         return ret;
 681 }
 682
 683 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
 684                        fmode_t flags, void *holder)
 685 {
 686         int ret;
 687
 688         mutex_lock(&uuid_mutex);
 689         if (fs_devices->opened) {
 690                 fs_devices->opened++;
 691                 ret = 0;
 692         } else {
 693                 ret = __btrfs_open_devices(fs_devices, flags, holder);
 694         }
 695         mutex_unlock(&uuid_mutex);
 696         return ret;
 697 }
 698
 699 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
 700                           struct btrfs_fs_devices **fs_devices_ret)
 701 {
 702         struct btrfs_super_block *disk_super;
 703         struct block_device *bdev;
 704         struct buffer_head *bh;
 705         int ret;
 706         u64 devid;
 707         u64 transid;
 708
 709         mutex_lock(&uuid_mutex);
 710
 711         flags |= FMODE_EXCL;
 712         bdev = blkdev_get_by_path(path, flags, holder);
 713
 714         if (IS_ERR(bdev)) {
 715                 ret = PTR_ERR(bdev);
 716                 goto error;
 717         }
 718
 719         ret = set_blocksize(bdev, 4096);
 720         if (ret)
 721                 goto error_close;
 722         bh = btrfs_read_dev_super(bdev);
 723         if (!bh) {
 724                 ret = -EINVAL;
 725                 goto error_close;
 726         }
 727         disk_super = (struct btrfs_super_block *)bh->b_data;
 728         devid = btrfs_stack_device_id(&disk_super->dev_item);
 729         transid = btrfs_super_generation(disk_super);
 730         if (disk_super->label[0])
 731                 printk(KERN_INFO "device label %s ", disk_super->label);
 732         else
 733                 printk(KERN_INFO "device fsid %pU ", disk_super->fsid);
 734         printk(KERN_CONT "devid %llu transid %llu %s\n",
 735                (unsigned long long)devid, (unsigned long long)transid, path);
 736         ret = device_list_add(path, disk_super, devid, fs_devices_ret);
 737
 738         brelse(bh);
 739 error_close:
 740         blkdev_put(bdev, flags);
 741 error:
 742         mutex_unlock(&uuid_mutex);
 743         return ret;
 744 }
 745
 746 /* helper to account the used device space in the range */
 747 int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
 748                                    u64 end, u64 *length)
 749 {
 750         struct btrfs_key key;
 751         struct btrfs_root *root = device->dev_root;
 752         struct btrfs_dev_extent *dev_extent;
 753         struct btrfs_path *path;
 754         u64 extent_end;
 755         int ret;
 756         int slot;
 757         struct extent_buffer *l;
 758
 759         *length = 0;
 760
 761         if (start >= device->total_bytes)
 762                 return 0;
 763
 764         path = btrfs_alloc_path();
 765         if (!path)
 766                 return -ENOMEM;
 767         path->reada = 2;
 768
 769         key.objectid = device->devid;
 770         key.offset = start;
 771         key.type = BTRFS_DEV_EXTENT_KEY;
 772
 773         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 774         if (ret < 0)
 775                 goto out;
 776         if (ret > 0) {
 777                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
 778                 if (ret < 0)
 779                         goto out;
 780         }
 781
 782         while (1) {
 783                 l = path->nodes[0];
 784                 slot = path->slots[0];
 785                 if (slot >= btrfs_header_nritems(l)) {
 786                         ret = btrfs_next_leaf(root, path);
 787                         if (ret == 0)
 788                                 continue;
 789                         if (ret < 0)
 790                                 goto out;
 791
 792                         break;
 793                 }
 794                 btrfs_item_key_to_cpu(l, &key, slot);
 795
 796                 if (key.objectid < device->devid)
 797                         goto next;
 798
 799                 if (key.objectid > device->devid)
 800                         break;
 801
 802                 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
 803                         goto next;
 804
 805                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
 806                 extent_end = key.offset + btrfs_dev_extent_length(l,
 807                                                                   dev_extent);
 808                 if (key.offset <= start && extent_end > end) {
 809                         *length = end - start + 1;
 810                         break;
 811                 } else if (key.offset <= start && extent_end > start)
 812                         *length += extent_end - start;
 813                 else if (key.offset > start && extent_end <= end)
 814                         *length += extent_end - key.offset;
 815                 else if (key.offset > start && key.offset <= end) {
 816                         *length += end - key.offset + 1;
 817                         break;
 818                 } else if (key.offset > end)
 819                         break;
 820
 821 next:
 822                 path->slots[0]++;
 823         }
 824         ret = 0;
 825 out:
 826         btrfs_free_path(path);
 827         return ret;
 828 }
 829
 830 /*
 831  * find_free_dev_extent - find free space in the specified device
 832  * @trans:      transaction handler
 833  * @device:     the device which we search the free space in
 834  * @num_bytes:  the size of the free space that we need
 835  * @start:      store the start of the free space.
 836  * @len:        the size of the free space. that we find, or the size of the max
 837  *              free space if we don't find suitable free space
 838  *
 839  * this uses a pretty simple search, the expectation is that it is
 840  * called very infrequently and that a given device has a small number
 841  * of extents
 842  *
 843  * @start is used to store the start of the free space if we find. But if we
 844  * don't find suitable free space, it will be used to store the start position
 845  * of the max free space.
 846  *
 847  * @len is used to store the size of the free space that we find.
 848  * But if we don't find suitable free space, it is used to store the size of
 849  * the max free space.
 850  */
 851 int find_free_dev_extent(struct btrfs_trans_handle *trans,
 852                          struct btrfs_device *device, u64 num_bytes,
 853                          u64 *start, u64 *len)
 854 {
 855         struct btrfs_key key;
 856         struct btrfs_root *root = device->dev_root;
 857         struct btrfs_dev_extent *dev_extent;
 858         struct btrfs_path *path;
 859         u64 hole_size;
 860         u64 max_hole_start;
 861         u64 max_hole_size;
 862         u64 extent_end;
 863         u64 search_start;
 864         u64 search_end = device->total_bytes;
 865         int ret;
 866         int slot;
 867         struct extent_buffer *l;
 868
 869         /* FIXME use last free of some kind */
 870
 871         /* we don't want to overwrite the superblock on the drive,
 872          * so we make sure to start at an offset of at least 1MB
 873          */
 874         search_start = max(root->fs_info->alloc_start, 1024ull * 1024);
 875
 876         max_hole_start = search_start;
 877         max_hole_size = 0;
 878         hole_size = 0;
 879
 880         if (search_start >= search_end) {
 881                 ret = -ENOSPC;
 882                 goto error;
 883         }
 884
 885         path = btrfs_alloc_path();
 886         if (!path) {
 887                 ret = -ENOMEM;
 888                 goto error;
 889         }
 890         path->reada = 2;
 891
 892         key.objectid = device->devid;
 893         key.offset = search_start;
 894         key.type = BTRFS_DEV_EXTENT_KEY;
 895
 896         ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
 897         if (ret < 0)
 898                 goto out;
 899         if (ret > 0) {
 900                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
 901                 if (ret < 0)
 902                         goto out;
 903         }
 904
 905         while (1) {
 906                 l = path->nodes[0];
 907                 slot = path->slots[0];
 908                 if (slot >= btrfs_header_nritems(l)) {
 909                         ret = btrfs_next_leaf(root, path);
 910                         if (ret == 0)
 911                                 continue;
 912                         if (ret < 0)
 913                                 goto out;
 914
 915                         break;
 916                 }
 917                 btrfs_item_key_to_cpu(l, &key, slot);
 918
 919                 if (key.objectid < device->devid)
 920                         goto next;
 921
 922                 if (key.objectid > device->devid)
 923                         break;
 924
 925                 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
 926                         goto next;
 927
 928                 if (key.offset > search_start) {
 929                         hole_size = key.offset - search_start;
 930
 931                         if (hole_size > max_hole_size) {
 932                                 max_hole_start = search_start;
 933                                 max_hole_size = hole_size;
 934                         }
 935
 936                         /*
 937                          * If this free space is greater than which we need,
 938                          * it must be the max free space that we have found
 939                          * until now, so max_hole_start must point to the start
 940                          * of this free space and the length of this free space
 941                          * is stored in max_hole_size. Thus, we return
 942                          * max_hole_start and max_hole_size and go back to the
 943                          * caller.
 944                          */
 945                         if (hole_size >= num_bytes) {
 946                                 ret = 0;
 947                                 goto out;
 948                         }
 949                 }
 950
 951                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
 952                 extent_end = key.offset + btrfs_dev_extent_length(l,
 953                                                                   dev_extent);
 954                 if (extent_end > search_start)
 955                         search_start = extent_end;
 956 next:
 957                 path->slots[0]++;
 958                 cond_resched();
 959         }
 960
 961         /*
 962          * At this point, search_start should be the end of
 963          * allocated dev extents, and when shrinking the device,
 964          * search_end may be smaller than search_start.
 965          */
 966         if (search_end > search_start)
 967                 hole_size = search_end - search_start;
 968
 969         if (hole_size > max_hole_size) {
 970                 max_hole_start = search_start;
 971                 max_hole_size = hole_size;
 972         }
 973
 974         /* See above. */
 975         if (hole_size < num_bytes)
 976                 ret = -ENOSPC;
 977         else
 978                 ret = 0;
 979
 980 out:
 981         btrfs_free_path(path);
 982 error:
 983         *start = max_hole_start;
 984         if (len)
 985                 *len = max_hole_size;
 986         return ret;
 987 }
 988
 989 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
 990                           struct btrfs_device *device,
 991                           u64 start)
 992 {
 993         int ret;
 994         struct btrfs_path *path;
 995         struct btrfs_root *root = device->dev_root;
 996         struct btrfs_key key;
 997         struct btrfs_key found_key;
 998         struct extent_buffer *leaf = NULL;
 999         struct btrfs_dev_extent *extent = NULL;
1000
1001         path = btrfs_alloc_path();
1002         if (!path)
1003                 return -ENOMEM;
1004
1005         key.objectid = device->devid;
1006         key.offset = start;
1007         key.type = BTRFS_DEV_EXTENT_KEY;
1008 again:
1009         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1010         if (ret > 0) {
1011                 ret = btrfs_previous_item(root, path, key.objectid,
1012                                           BTRFS_DEV_EXTENT_KEY);
1013                 if (ret)
1014                         goto out;
1015                 leaf = path->nodes[0];
1016                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1017                 extent = btrfs_item_ptr(leaf, path->slots[0],
1018                                         struct btrfs_dev_extent);
1019                 BUG_ON(found_key.offset > start || found_key.offset +
1020                        btrfs_dev_extent_length(leaf, extent) < start);
1021                 key = found_key;
1022                 btrfs_release_path(path);
1023                 goto again;
1024         } else if (ret == 0) {
1025                 leaf = path->nodes[0];
1026                 extent = btrfs_item_ptr(leaf, path->slots[0],
1027                                         struct btrfs_dev_extent);
1028         }
1029         BUG_ON(ret);
1030
1031         if (device->bytes_used > 0) {
1032                 u64 len = btrfs_dev_extent_length(leaf, extent);
1033                 device->bytes_used -= len;
1034                 spin_lock(&root->fs_info->free_chunk_lock);
1035                 root->fs_info->free_chunk_space += len;
1036                 spin_unlock(&root->fs_info->free_chunk_lock);
1037         }
1038         ret = btrfs_del_item(trans, root, path);
1039
1040 out:
1041         btrfs_free_path(path);
1042         return ret;
1043 }
1044
1045 int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1046                            struct btrfs_device *device,
1047                            u64 chunk_tree, u64 chunk_objectid,
1048                            u64 chunk_offset, u64 start, u64 num_bytes)
1049 {
1050         int ret;
1051         struct btrfs_path *path;
1052         struct btrfs_root *root = device->dev_root;
1053         struct btrfs_dev_extent *extent;
1054         struct extent_buffer *leaf;
1055         struct btrfs_key key;
1056
1057         WARN_ON(!device->in_fs_metadata);
1058         path = btrfs_alloc_path();
1059         if (!path)
1060                 return -ENOMEM;
1061
1062         key.objectid = device->devid;
1063         key.offset = start;
1064         key.type = BTRFS_DEV_EXTENT_KEY;
1065         ret = btrfs_insert_empty_item(trans, root, path, &key,
1066                                       sizeof(*extent));
1067         BUG_ON(ret);
1068
1069         leaf = path->nodes[0];
1070         extent = btrfs_item_ptr(leaf, path->slots[0],
1071                                 struct btrfs_dev_extent);
1072         btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
1073         btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
1074         btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1075
1076         write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
1077                     (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
1078                     BTRFS_UUID_SIZE);
1079
1080         btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1081         btrfs_mark_buffer_dirty(leaf);
1082         btrfs_free_path(path);
1083         return ret;
1084 }
1085
1086 static noinline int find_next_chunk(struct btrfs_root *root,
1087                                     u64 objectid, u64 *offset)
1088 {
1089         struct btrfs_path *path;
1090         int ret;
1091         struct btrfs_key key;
1092         struct btrfs_chunk *chunk;
1093         struct btrfs_key found_key;
1094
1095         path = btrfs_alloc_path();
1096         if (!path)
1097                 return -ENOMEM;
1098
1099         key.objectid = objectid;
1100         key.offset = (u64)-1;
1101         key.type = BTRFS_CHUNK_ITEM_KEY;
1102
1103         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1104         if (ret < 0)
1105                 goto error;
1106
1107         BUG_ON(ret == 0);
1108
1109         ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
1110         if (ret) {
1111                 *offset = 0;
1112         } else {
1113                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1114                                       path->slots[0]);
1115                 if (found_key.objectid != objectid)
1116                         *offset = 0;
1117                 else {
1118                         chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
1119                                                struct btrfs_chunk);
1120                         *offset = found_key.offset +
1121                                 btrfs_chunk_length(path->nodes[0], chunk);
1122                 }
1123         }
1124         ret = 0;
1125 error:
1126         btrfs_free_path(path);
1127         return ret;
1128 }
1129
1130 static noinline int find_next_devid(struct btrfs_root *root, u64 *objectid)
1131 {
1132         int ret;
1133         struct btrfs_key key;
1134         struct btrfs_key found_key;
1135         struct btrfs_path *path;
1136
1137         root = root->fs_info->chunk_root;
1138
1139         path = btrfs_alloc_path();
1140         if (!path)
1141                 return -ENOMEM;
1142
1143         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1144         key.type = BTRFS_DEV_ITEM_KEY;
1145         key.offset = (u64)-1;
1146
1147         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1148         if (ret < 0)
1149                 goto error;
1150
1151         BUG_ON(ret == 0);
1152
1153         ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
1154                                   BTRFS_DEV_ITEM_KEY);
1155         if (ret) {
1156                 *objectid = 1;
1157         } else {
1158                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1159                                       path->slots[0]);
1160                 *objectid = found_key.offset + 1;
1161         }
1162         ret = 0;
1163 error:
1164         btrfs_free_path(path);
1165         return ret;
1166 }
1167
1168 /*
1169  * the device information is stored in the chunk root
1170  * the btrfs_device struct should be fully filled in
1171  */
1172 int btrfs_add_device(struct btrfs_trans_handle *trans,
1173                      struct btrfs_root *root,
1174                      struct btrfs_device *device)
1175 {
1176         int ret;
1177         struct btrfs_path *path;
1178         struct btrfs_dev_item *dev_item;
1179         struct extent_buffer *leaf;
1180         struct btrfs_key key;
1181         unsigned long ptr;
1182
1183         root = root->fs_info->chunk_root;
1184
1185         path = btrfs_alloc_path();
1186         if (!path)
1187                 return -ENOMEM;
1188
1189         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1190         key.type = BTRFS_DEV_ITEM_KEY;
1191         key.offset = device->devid;
1192
1193         ret = btrfs_insert_empty_item(trans, root, path, &key,
1194                                       sizeof(*dev_item));
1195         if (ret)
1196                 goto out;
1197
1198         leaf = path->nodes[0];
1199         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1200
1201         btrfs_set_device_id(leaf, dev_item, device->devid);
1202         btrfs_set_device_generation(leaf, dev_item, 0);
1203         btrfs_set_device_type(leaf, dev_item, device->type);
1204         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1205         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1206         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1207         btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
1208         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1209         btrfs_set_device_group(leaf, dev_item, 0);
1210         btrfs_set_device_seek_speed(leaf, dev_item, 0);
1211         btrfs_set_device_bandwidth(leaf, dev_item, 0);
1212         btrfs_set_device_start_offset(leaf, dev_item, 0);
1213
1214         ptr = (unsigned long)btrfs_device_uuid(dev_item);
1215         write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1216         ptr = (unsigned long)btrfs_device_fsid(dev_item);
1217         write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1218         btrfs_mark_buffer_dirty(leaf);
1219
1220         ret = 0;
1221 out:
1222         btrfs_free_path(path);
1223         return ret;
1224 }
1225
1226 static int btrfs_rm_dev_item(struct btrfs_root *root,
1227                              struct btrfs_device *device)
1228 {
1229         int ret;
1230         struct btrfs_path *path;
1231         struct btrfs_key key;
1232         struct btrfs_trans_handle *trans;
1233
1234         root = root->fs_info->chunk_root;
1235
1236         path = btrfs_alloc_path();
1237         if (!path)
1238                 return -ENOMEM;
1239
1240         trans = btrfs_start_transaction(root, 0);
1241         if (IS_ERR(trans)) {
1242                 btrfs_free_path(path);
1243                 return PTR_ERR(trans);
1244         }
1245         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1246         key.type = BTRFS_DEV_ITEM_KEY;
1247         key.offset = device->devid;
1248         lock_chunks(root);
1249
1250         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1251         if (ret < 0)
1252                 goto out;
1253
1254         if (ret > 0) {
1255                 ret = -ENOENT;
1256                 goto out;
1257         }
1258
1259         ret = btrfs_del_item(trans, root, path);
1260         if (ret)
1261                 goto out;
1262 out:
1263         btrfs_free_path(path);
1264         unlock_chunks(root);
1265         btrfs_commit_transaction(trans, root);
1266         return ret;
1267 }
1268
1269 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
1270 {
1271         struct btrfs_device *device;
1272         struct btrfs_device *next_device;
1273         struct block_device *bdev;
1274         struct buffer_head *bh = NULL;
1275         struct btrfs_super_block *disk_super;
1276         struct btrfs_fs_devices *cur_devices;
1277         u64 all_avail;
1278         u64 devid;
1279         u64 num_devices;
1280         u8 *dev_uuid;
1281         int ret = 0;
1282         bool clear_super = false;
1283
1284         mutex_lock(&uuid_mutex);
1285         mutex_lock(&root->fs_info->volume_mutex);
1286
1287         all_avail = root->fs_info->avail_data_alloc_bits |
1288                 root->fs_info->avail_system_alloc_bits |
1289                 root->fs_info->avail_metadata_alloc_bits;
1290
1291         if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) &&
1292             root->fs_info->fs_devices->num_devices <= 4) {
1293                 printk(KERN_ERR "btrfs: unable to go below four devices "
1294                        "on raid10\n");
1295                 ret = -EINVAL;
1296                 goto out;
1297         }
1298
1299         if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) &&
1300             root->fs_info->fs_devices->num_devices <= 2) {
1301                 printk(KERN_ERR "btrfs: unable to go below two "
1302                        "devices on raid1\n");
1303                 ret = -EINVAL;
1304                 goto out;
1305         }
1306
1307         if (strcmp(device_path, "missing") == 0) {
1308                 struct list_head *devices;
1309                 struct btrfs_device *tmp;
1310
1311                 device = NULL;
1312                 devices = &root->fs_info->fs_devices->devices;
1313                 /*
1314                  * It is safe to read the devices since the volume_mutex
1315                  * is held.
1316                  */
1317                 list_for_each_entry(tmp, devices, dev_list) {
1318                         if (tmp->in_fs_metadata && !tmp->bdev) {
1319                                 device = tmp;
1320                                 break;
1321                         }
1322                 }
1323                 bdev = NULL;
1324                 bh = NULL;
1325                 disk_super = NULL;
1326                 if (!device) {
1327                         printk(KERN_ERR "btrfs: no missing devices found to "
1328                                "remove\n");
1329                         goto out;
1330                 }
1331         } else {
1332                 bdev = blkdev_get_by_path(device_path, FMODE_READ | FMODE_EXCL,
1333                                           root->fs_info->bdev_holder);
1334                 if (IS_ERR(bdev)) {
1335                         ret = PTR_ERR(bdev);
1336                         goto out;
1337                 }
1338
1339                 set_blocksize(bdev, 4096);
1340                 bh = btrfs_read_dev_super(bdev);
1341                 if (!bh) {
1342                         ret = -EINVAL;
1343                         goto error_close;
1344                 }
1345                 disk_super = (struct btrfs_super_block *)bh->b_data;
1346                 devid = btrfs_stack_device_id(&disk_super->dev_item);
1347                 dev_uuid = disk_super->dev_item.uuid;
1348                 device = btrfs_find_device(root, devid, dev_uuid,
1349                                            disk_super->fsid);
1350                 if (!device) {
1351                         ret = -ENOENT;
1352                         goto error_brelse;
1353                 }
1354         }
1355
1356         if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1357                 printk(KERN_ERR "btrfs: unable to remove the only writeable "
1358                        "device\n");
1359                 ret = -EINVAL;
1360                 goto error_brelse;
1361         }
1362
1363         if (device->writeable) {
1364                 lock_chunks(root);
1365                 list_del_init(&device->dev_alloc_list);
1366                 unlock_chunks(root);
1367                 root->fs_info->fs_devices->rw_devices--;
1368                 clear_super = true;
1369         }
1370
1371         ret = btrfs_shrink_device(device, 0);
1372         if (ret)
1373                 goto error_undo;
1374
1375         ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1376         if (ret)
1377                 goto error_undo;
1378
1379         spin_lock(&root->fs_info->free_chunk_lock);
1380         root->fs_info->free_chunk_space = device->total_bytes -
1381                 device->bytes_used;
1382         spin_unlock(&root->fs_info->free_chunk_lock);
1383
1384         device->in_fs_metadata = 0;
1385         btrfs_scrub_cancel_dev(root, device);
1386
1387         /*
1388          * the device list mutex makes sure that we don't change
1389          * the device list while someone else is writing out all
1390          * the device supers.
1391          */
1392
1393         cur_devices = device->fs_devices;
1394         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1395         list_del_rcu(&device->dev_list);
1396
1397         device->fs_devices->num_devices--;
1398
1399         if (device->missing)
1400                 root->fs_info->fs_devices->missing_devices--;
1401
1402         next_device = list_entry(root->fs_info->fs_devices->devices.next,
1403                                  struct btrfs_device, dev_list);
1404         if (device->bdev == root->fs_info->sb->s_bdev)
1405                 root->fs_info->sb->s_bdev = next_device->bdev;
1406         if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1407                 root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1408
1409         if (device->bdev)
1410                 device->fs_devices->open_devices--;
1411
1412         call_rcu(&device->rcu, free_device);
1413         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1414
1415         num_devices = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
1416         btrfs_set_super_num_devices(root->fs_info->super_copy, num_devices);
1417
1418         if (cur_devices->open_devices == 0) {
1419                 struct btrfs_fs_devices *fs_devices;
1420                 fs_devices = root->fs_info->fs_devices;
1421                 while (fs_devices) {
1422                         if (fs_devices->seed == cur_devices)
1423                                 break;
1424                         fs_devices = fs_devices->seed;
1425                 }
1426                 fs_devices->seed = cur_devices->seed;
1427                 cur_devices->seed = NULL;
1428                 lock_chunks(root);
1429                 __btrfs_close_devices(cur_devices);
1430                 unlock_chunks(root);
1431                 free_fs_devices(cur_devices);
1432         }
1433
1434         /*
1435          * at this point, the device is zero sized.  We want to
1436          * remove it from the devices list and zero out the old super
1437          */
1438         if (clear_super) {
1439                 /* make sure this device isn't detected as part of
1440                  * the FS anymore
1441                  */
1442                 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1443                 set_buffer_dirty(bh);
1444                 sync_dirty_buffer(bh);
1445         }
1446
1447         ret = 0;
1448
1449 error_brelse:
1450         brelse(bh);
1451 error_close:
1452         if (bdev)
1453                 blkdev_put(bdev, FMODE_READ | FMODE_EXCL);
1454 out:
1455         mutex_unlock(&root->fs_info->volume_mutex);
1456         mutex_unlock(&uuid_mutex);
1457         return ret;
1458 error_undo:
1459         if (device->writeable) {
1460                 lock_chunks(root);
1461                 list_add(&device->dev_alloc_list,
1462                          &root->fs_info->fs_devices->alloc_list);
1463                 unlock_chunks(root);
1464                 root->fs_info->fs_devices->rw_devices++;
1465         }
1466         goto error_brelse;
1467 }
1468
1469 /*
1470  * does all the dirty work required for changing file system's UUID.
1471  */
1472 static int btrfs_prepare_sprout(struct btrfs_trans_handle *trans,
1473                                 struct btrfs_root *root)
1474 {
1475         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
1476         struct btrfs_fs_devices *old_devices;
1477         struct btrfs_fs_devices *seed_devices;
1478         struct btrfs_super_block *disk_super = root->fs_info->super_copy;
1479         struct btrfs_device *device;
1480         u64 super_flags;
1481
1482         BUG_ON(!mutex_is_locked(&uuid_mutex));
1483         if (!fs_devices->seeding)
1484                 return -EINVAL;
1485
1486         seed_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
1487         if (!seed_devices)
1488                 return -ENOMEM;
1489
1490         old_devices = clone_fs_devices(fs_devices);
1491         if (IS_ERR(old_devices)) {
1492                 kfree(seed_devices);
1493                 return PTR_ERR(old_devices);
1494         }
1495
1496         list_add(&old_devices->list, &fs_uuids);
1497
1498         memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
1499         seed_devices->opened = 1;
1500         INIT_LIST_HEAD(&seed_devices->devices);
1501         INIT_LIST_HEAD(&seed_devices->alloc_list);
1502         mutex_init(&seed_devices->device_list_mutex);
1503
1504         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1505         list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
1506                               synchronize_rcu);
1507         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1508
1509         list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
1510         list_for_each_entry(device, &seed_devices->devices, dev_list) {
1511                 device->fs_devices = seed_devices;
1512         }
1513
1514         fs_devices->seeding = 0;
1515         fs_devices->num_devices = 0;
1516         fs_devices->open_devices = 0;
1517         fs_devices->seed = seed_devices;
1518
1519         generate_random_uuid(fs_devices->fsid);
1520         memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1521         memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1522         super_flags = btrfs_super_flags(disk_super) &
1523                       ~BTRFS_SUPER_FLAG_SEEDING;
1524         btrfs_set_super_flags(disk_super, super_flags);
1525
1526         return 0;
1527 }
1528
1529 /*
1530  * strore the expected generation for seed devices in device items.
1531  */
1532 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
1533                                struct btrfs_root *root)
1534 {
1535         struct btrfs_path *path;
1536         struct extent_buffer *leaf;
1537         struct btrfs_dev_item *dev_item;
1538         struct btrfs_device *device;
1539         struct btrfs_key key;
1540         u8 fs_uuid[BTRFS_UUID_SIZE];
1541         u8 dev_uuid[BTRFS_UUID_SIZE];
1542         u64 devid;
1543         int ret;
1544
1545         path = btrfs_alloc_path();
1546         if (!path)
1547                 return -ENOMEM;
1548
1549         root = root->fs_info->chunk_root;
1550         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1551         key.offset = 0;
1552         key.type = BTRFS_DEV_ITEM_KEY;
1553
1554         while (1) {
1555                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1556                 if (ret < 0)
1557                         goto error;
1558
1559                 leaf = path->nodes[0];
1560 next_slot:
1561                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1562                         ret = btrfs_next_leaf(root, path);
1563                         if (ret > 0)
1564                                 break;
1565                         if (ret < 0)
1566                                 goto error;
1567                         leaf = path->nodes[0];
1568                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1569                         btrfs_release_path(path);
1570                         continue;
1571                 }
1572
1573                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1574                 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
1575                     key.type != BTRFS_DEV_ITEM_KEY)
1576                         break;
1577
1578                 dev_item = btrfs_item_ptr(leaf, path->slots[0],
1579                                           struct btrfs_dev_item);
1580                 devid = btrfs_device_id(leaf, dev_item);
1581                 read_extent_buffer(leaf, dev_uuid,
1582                                    (unsigned long)btrfs_device_uuid(dev_item),
1583                                    BTRFS_UUID_SIZE);
1584                 read_extent_buffer(leaf, fs_uuid,
1585                                    (unsigned long)btrfs_device_fsid(dev_item),
1586                                    BTRFS_UUID_SIZE);
1587                 device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
1588                 BUG_ON(!device);
1589
1590                 if (device->fs_devices->seeding) {
1591                         btrfs_set_device_generation(leaf, dev_item,
1592                                                     device->generation);
1593                         btrfs_mark_buffer_dirty(leaf);
1594                 }
1595
1596                 path->slots[0]++;
1597                 goto next_slot;
1598         }
1599         ret = 0;
1600 error:
1601         btrfs_free_path(path);
1602         return ret;
1603 }
1604
1605 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
1606 {
1607         struct request_queue *q;
1608         struct btrfs_trans_handle *trans;
1609         struct btrfs_device *device;
1610         struct block_device *bdev;
1611         struct list_head *devices;
1612         struct super_block *sb = root->fs_info->sb;
1613         u64 total_bytes;
1614         int seeding_dev = 0;
1615         int ret = 0;
1616
1617         if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
1618                 return -EINVAL;
1619
1620         bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
1621                                   root->fs_info->bdev_holder);
1622         if (IS_ERR(bdev))
1623                 return PTR_ERR(bdev);
1624
1625         if (root->fs_info->fs_devices->seeding) {
1626                 seeding_dev = 1;
1627                 down_write(&sb->s_umount);
1628                 mutex_lock(&uuid_mutex);
1629         }
1630
1631         filemap_write_and_wait(bdev->bd_inode->i_mapping);
1632         mutex_lock(&root->fs_info->volume_mutex);
1633
1634         devices = &root->fs_info->fs_devices->devices;
1635         /*
1636          * we have the volume lock, so we don't need the extra
1637          * device list mutex while reading the list here.
1638          */
1639         list_for_each_entry(device, devices, dev_list) {
1640                 if (device->bdev == bdev) {
1641                         ret = -EEXIST;
1642                         goto error;
1643                 }
1644         }
1645
1646         device = kzalloc(sizeof(*device), GFP_NOFS);
1647         if (!device) {
1648                 /* we can safely leave the fs_devices entry around */
1649                 ret = -ENOMEM;
1650                 goto error;
1651         }
1652
1653         device->name = kstrdup(device_path, GFP_NOFS);
1654         if (!device->name) {
1655                 kfree(device);
1656                 ret = -ENOMEM;
1657                 goto error;
1658         }
1659
1660         ret = find_next_devid(root, &device->devid);
1661         if (ret) {
1662                 kfree(device->name);
1663                 kfree(device);
1664                 goto error;
1665         }
1666
1667         trans = btrfs_start_transaction(root, 0);
1668         if (IS_ERR(trans)) {
1669                 kfree(device->name);
1670                 kfree(device);
1671                 ret = PTR_ERR(trans);
1672                 goto error;
1673         }
1674
1675         lock_chunks(root);
1676
1677         q = bdev_get_queue(bdev);
1678         if (blk_queue_discard(q))
1679                 device->can_discard = 1;
1680         device->writeable = 1;
1681         device->work.func = pending_bios_fn;
1682         generate_random_uuid(device->uuid);
1683         spin_lock_init(&device->io_lock);
1684         device->generation = trans->transid;
1685         device->io_width = root->sectorsize;
1686         device->io_align = root->sectorsize;
1687         device->sector_size = root->sectorsize;
1688         device->total_bytes = i_size_read(bdev->bd_inode);
1689         device->disk_total_bytes = device->total_bytes;
1690         device->dev_root = root->fs_info->dev_root;
1691         device->bdev = bdev;
1692         device->in_fs_metadata = 1;
1693         device->mode = FMODE_EXCL;
1694         set_blocksize(device->bdev, 4096);
1695
1696         if (seeding_dev) {
1697                 sb->s_flags &= ~MS_RDONLY;
1698                 ret = btrfs_prepare_sprout(trans, root);
1699                 BUG_ON(ret);
1700         }
1701
1702         device->fs_devices = root->fs_info->fs_devices;
1703
1704         /*
1705          * we don't want write_supers to jump in here with our device
1706          * half setup
1707          */
1708         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1709         list_add_rcu(&device->dev_list, &root->fs_info->fs_devices->devices);
1710         list_add(&device->dev_alloc_list,
1711                  &root->fs_info->fs_devices->alloc_list);
1712         root->fs_info->fs_devices->num_devices++;
1713         root->fs_info->fs_devices->open_devices++;
1714         root->fs_info->fs_devices->rw_devices++;
1715         if (device->can_discard)
1716                 root->fs_info->fs_devices->num_can_discard++;
1717         root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
1718
1719         spin_lock(&root->fs_info->free_chunk_lock);
1720         root->fs_info->free_chunk_space += device->total_bytes;
1721         spin_unlock(&root->fs_info->free_chunk_lock);
1722
1723         if (!blk_queue_nonrot(bdev_get_queue(bdev)))
1724                 root->fs_info->fs_devices->rotating = 1;
1725
1726         total_bytes = btrfs_super_total_bytes(root->fs_info->super_copy);
1727         btrfs_set_super_total_bytes(root->fs_info->super_copy,
1728                                     total_bytes + device->total_bytes);
1729
1730         total_bytes = btrfs_super_num_devices(root->fs_info->super_copy);
1731         btrfs_set_super_num_devices(root->fs_info->super_copy,
1732                                     total_bytes + 1);
1733         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1734
1735         if (seeding_dev) {
1736                 ret = init_first_rw_device(trans, root, device);
1737                 BUG_ON(ret);
1738                 ret = btrfs_finish_sprout(trans, root);
1739                 BUG_ON(ret);
1740         } else {
1741                 ret = btrfs_add_device(trans, root, device);
1742         }
1743
1744         /*
1745          * we've got more storage, clear any full flags on the space
1746          * infos
1747          */
1748         btrfs_clear_space_info_full(root->fs_info);
1749
1750         unlock_chunks(root);
1751         btrfs_commit_transaction(trans, root);
1752
1753         if (seeding_dev) {
1754                 mutex_unlock(&uuid_mutex);
1755                 up_write(&sb->s_umount);
1756
1757                 ret = btrfs_relocate_sys_chunks(root);
1758                 BUG_ON(ret);
1759         }
1760 out:
1761         mutex_unlock(&root->fs_info->volume_mutex);
1762         return ret;
1763 error:
1764         blkdev_put(bdev, FMODE_EXCL);
1765         if (seeding_dev) {
1766                 mutex_unlock(&uuid_mutex);
1767                 up_write(&sb->s_umount);
1768         }
1769         goto out;
1770 }
1771
1772 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
1773                                         struct btrfs_device *device)
1774 {
1775         int ret;
1776         struct btrfs_path *path;
1777         struct btrfs_root *root;
1778         struct btrfs_dev_item *dev_item;
1779         struct extent_buffer *leaf;
1780         struct btrfs_key key;
1781
1782         root = device->dev_root->fs_info->chunk_root;
1783
1784         path = btrfs_alloc_path();
1785         if (!path)
1786                 return -ENOMEM;
1787
1788         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1789         key.type = BTRFS_DEV_ITEM_KEY;
1790         key.offset = device->devid;
1791
1792         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1793         if (ret < 0)
1794                 goto out;
1795
1796         if (ret > 0) {
1797                 ret = -ENOENT;
1798                 goto out;
1799         }
1800
1801         leaf = path->nodes[0];
1802         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1803
1804         btrfs_set_device_id(leaf, dev_item, device->devid);
1805         btrfs_set_device_type(leaf, dev_item, device->type);
1806         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1807         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1808         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1809         btrfs_set_device_total_bytes(leaf, dev_item, device->disk_total_bytes);
1810         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1811         btrfs_mark_buffer_dirty(leaf);
1812
1813 out:
1814         btrfs_free_path(path);
1815         return ret;
1816 }
1817
1818 static int __btrfs_grow_device(struct btrfs_trans_handle *trans,
1819                       struct btrfs_device *device, u64 new_size)
1820 {
1821         struct btrfs_super_block *super_copy =
1822                 device->dev_root->fs_info->super_copy;
1823         u64 old_total = btrfs_super_total_bytes(super_copy);
1824         u64 diff = new_size - device->total_bytes;
1825
1826         if (!device->writeable)
1827                 return -EACCES;
1828         if (new_size <= device->total_bytes)
1829                 return -EINVAL;
1830
1831         btrfs_set_super_total_bytes(super_copy, old_total + diff);
1832         device->fs_devices->total_rw_bytes += diff;
1833
1834         device->total_bytes = new_size;
1835         device->disk_total_bytes = new_size;
1836         btrfs_clear_space_info_full(device->dev_root->fs_info);
1837
1838         return btrfs_update_device(trans, device);
1839 }
1840
1841 int btrfs_grow_device(struct btrfs_trans_handle *trans,
1842                       struct btrfs_device *device, u64 new_size)
1843 {
1844         int ret;
1845         lock_chunks(device->dev_root);
1846         ret = __btrfs_grow_device(trans, device, new_size);
1847         unlock_chunks(device->dev_root);
1848         return ret;
1849 }
1850
1851 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
1852                             struct btrfs_root *root,
1853                             u64 chunk_tree, u64 chunk_objectid,
1854                             u64 chunk_offset)
1855 {
1856         int ret;
1857         struct btrfs_path *path;
1858         struct btrfs_key key;
1859
1860         root = root->fs_info->chunk_root;
1861         path = btrfs_alloc_path();
1862         if (!path)
1863                 return -ENOMEM;
1864
1865         key.objectid = chunk_objectid;
1866         key.offset = chunk_offset;
1867         key.type = BTRFS_CHUNK_ITEM_KEY;
1868
1869         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1870         BUG_ON(ret);
1871
1872         ret = btrfs_del_item(trans, root, path);
1873
1874         btrfs_free_path(path);
1875         return ret;
1876 }
1877
1878 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
1879                         chunk_offset)
1880 {
1881         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
1882         struct btrfs_disk_key *disk_key;
1883         struct btrfs_chunk *chunk;
1884         u8 *ptr;
1885         int ret = 0;
1886         u32 num_stripes;
1887         u32 array_size;
1888         u32 len = 0;
1889         u32 cur;
1890         struct btrfs_key key;
1891
1892         array_size = btrfs_super_sys_array_size(super_copy);
1893
1894         ptr = super_copy->sys_chunk_array;
1895         cur = 0;
1896
1897         while (cur < array_size) {
1898                 disk_key = (struct btrfs_disk_key *)ptr;
1899                 btrfs_disk_key_to_cpu(&key, disk_key);
1900
1901                 len = sizeof(*disk_key);
1902
1903                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
1904                         chunk = (struct btrfs_chunk *)(ptr + len);
1905                         num_stripes = btrfs_stack_chunk_num_stripes(chunk);
1906                         len += btrfs_chunk_item_size(num_stripes);
1907                 } else {
1908                         ret = -EIO;
1909                         break;
1910                 }
1911                 if (key.objectid == chunk_objectid &&
1912                     key.offset == chunk_offset) {
1913                         memmove(ptr, ptr + len, array_size - (cur + len));
1914                         array_size -= len;
1915                         btrfs_set_super_sys_array_size(super_copy, array_size);
1916                 } else {
1917                         ptr += len;
1918                         cur += len;
1919                 }
1920         }
1921         return ret;
1922 }
1923
1924 static int btrfs_relocate_chunk(struct btrfs_root *root,
1925                          u64 chunk_tree, u64 chunk_objectid,
1926                          u64 chunk_offset)
1927 {
1928         struct extent_map_tree *em_tree;
1929         struct btrfs_root *extent_root;
1930         struct btrfs_trans_handle *trans;
1931         struct extent_map *em;
1932         struct map_lookup *map;
1933         int ret;
1934         int i;
1935
1936         root = root->fs_info->chunk_root;
1937         extent_root = root->fs_info->extent_root;
1938         em_tree = &root->fs_info->mapping_tree.map_tree;
1939
1940         ret = btrfs_can_relocate(extent_root, chunk_offset);
1941         if (ret)
1942                 return -ENOSPC;
1943
1944         /* step one, relocate all the extents inside this chunk */
1945         ret = btrfs_relocate_block_group(extent_root, chunk_offset);
1946         if (ret)
1947                 return ret;
1948
1949         trans = btrfs_start_transaction(root, 0);
1950         BUG_ON(IS_ERR(trans));
1951
1952         lock_chunks(root);
1953
1954         /*
1955          * step two, delete the device extents and the
1956          * chunk tree entries
1957          */
1958         read_lock(&em_tree->lock);
1959         em = lookup_extent_mapping(em_tree, chunk_offset, 1);
1960         read_unlock(&em_tree->lock);
1961
1962         BUG_ON(em->start > chunk_offset ||
1963                em->start + em->len < chunk_offset);
1964         map = (struct map_lookup *)em->bdev;
1965
1966         for (i = 0; i < map->num_stripes; i++) {
1967                 ret = btrfs_free_dev_extent(trans, map->stripes[i].dev,
1968                                             map->stripes[i].physical);
1969                 BUG_ON(ret);
1970
1971                 if (map->stripes[i].dev) {
1972                         ret = btrfs_update_device(trans, map->stripes[i].dev);
1973                         BUG_ON(ret);
1974                 }
1975         }
1976         ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
1977                                chunk_offset);
1978
1979         BUG_ON(ret);
1980
1981         trace_btrfs_chunk_free(root, map, chunk_offset, em->len);
1982
1983         if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
1984                 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
1985                 BUG_ON(ret);
1986         }
1987
1988         ret = btrfs_remove_block_group(trans, extent_root, chunk_offset);
1989         BUG_ON(ret);
1990
1991         write_lock(&em_tree->lock);
1992         remove_extent_mapping(em_tree, em);
1993         write_unlock(&em_tree->lock);
1994
1995         kfree(map);
1996         em->bdev = NULL;
1997
1998         /* once for the tree */
1999         free_extent_map(em);
2000         /* once for us */
2001         free_extent_map(em);
2002
2003         unlock_chunks(root);
2004         btrfs_end_transaction(trans, root);
2005         return 0;
2006 }
2007
2008 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
2009 {
2010         struct btrfs_root *chunk_root = root->fs_info->chunk_root;
2011         struct btrfs_path *path;
2012         struct extent_buffer *leaf;
2013         struct btrfs_chunk *chunk;
2014         struct btrfs_key key;
2015         struct btrfs_key found_key;
2016         u64 chunk_tree = chunk_root->root_key.objectid;
2017         u64 chunk_type;
2018         bool retried = false;
2019         int failed = 0;
2020         int ret;
2021
2022         path = btrfs_alloc_path();
2023         if (!path)
2024                 return -ENOMEM;
2025
2026 again:
2027         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2028         key.offset = (u64)-1;
2029         key.type = BTRFS_CHUNK_ITEM_KEY;
2030
2031         while (1) {
2032                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2033                 if (ret < 0)
2034                         goto error;
2035                 BUG_ON(ret == 0);
2036
2037                 ret = btrfs_previous_item(chunk_root, path, key.objectid,
2038                                           key.type);
2039                 if (ret < 0)
2040                         goto error;
2041                 if (ret > 0)
2042                         break;
2043
2044                 leaf = path->nodes[0];
2045                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2046
2047                 chunk = btrfs_item_ptr(leaf, path->slots[0],
2048                                        struct btrfs_chunk);
2049                 chunk_type = btrfs_chunk_type(leaf, chunk);
2050                 btrfs_release_path(path);
2051
2052                 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
2053                         ret = btrfs_relocate_chunk(chunk_root, chunk_tree,
2054                                                    found_key.objectid,
2055                                                    found_key.offset);
2056                         if (ret == -ENOSPC)
2057                                 failed++;
2058                         else if (ret)
2059                                 BUG();
2060                 }
2061
2062                 if (found_key.offset == 0)
2063                         break;
2064                 key.offset = found_key.offset - 1;
2065         }
2066         ret = 0;
2067         if (failed && !retried) {
2068                 failed = 0;
2069                 retried = true;
2070                 goto again;
2071         } else if (failed && retried) {
2072                 WARN_ON(1);
2073                 ret = -ENOSPC;
2074         }
2075 error:
2076         btrfs_free_path(path);
2077         return ret;
2078 }
2079
2080 static u64 div_factor(u64 num, int factor)
2081 {
2082         if (factor == 10)
2083                 return num;
2084         num *= factor;
2085         do_div(num, 10);
2086         return num;
2087 }
2088
2089 int btrfs_balance(struct btrfs_root *dev_root)
2090 {
2091         int ret;
2092         struct list_head *devices = &dev_root->fs_info->fs_devices->devices;
2093         struct btrfs_device *device;
2094         u64 old_size;
2095         u64 size_to_free;
2096         struct btrfs_path *path;
2097         struct btrfs_key key;
2098         struct btrfs_root *chunk_root = dev_root->fs_info->chunk_root;
2099         struct btrfs_trans_handle *trans;
2100         struct btrfs_key found_key;
2101
2102         if (dev_root->fs_info->sb->s_flags & MS_RDONLY)
2103                 return -EROFS;
2104
2105         if (!capable(CAP_SYS_ADMIN))
2106                 return -EPERM;
2107
2108         mutex_lock(&dev_root->fs_info->volume_mutex);
2109         dev_root = dev_root->fs_info->dev_root;
2110
2111         /* step one make some room on all the devices */
2112         list_for_each_entry(device, devices, dev_list) {
2113                 old_size = device->total_bytes;
2114                 size_to_free = div_factor(old_size, 1);
2115                 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
2116                 if (!device->writeable ||
2117                     device->total_bytes - device->bytes_used > size_to_free)
2118                         continue;
2119
2120                 ret = btrfs_shrink_device(device, old_size - size_to_free);
2121                 if (ret == -ENOSPC)
2122                         break;
2123                 BUG_ON(ret);
2124
2125                 trans = btrfs_start_transaction(dev_root, 0);
2126                 BUG_ON(IS_ERR(trans));
2127
2128                 ret = btrfs_grow_device(trans, device, old_size);
2129                 BUG_ON(ret);
2130
2131                 btrfs_end_transaction(trans, dev_root);
2132         }
2133
2134         /* step two, relocate all the chunks */
2135         path = btrfs_alloc_path();
2136         if (!path) {
2137                 ret = -ENOMEM;
2138                 goto error;
2139         }
2140         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2141         key.offset = (u64)-1;
2142         key.type = BTRFS_CHUNK_ITEM_KEY;
2143
2144         while (1) {
2145                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2146                 if (ret < 0)
2147                         goto error;
2148
2149                 /*
2150                  * this shouldn't happen, it means the last relocate
2151                  * failed
2152                  */
2153                 if (ret == 0)
2154                         break;
2155
2156                 ret = btrfs_previous_item(chunk_root, path, 0,
2157                                           BTRFS_CHUNK_ITEM_KEY);
2158                 if (ret)
2159                         break;
2160
2161                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2162                                       path->slots[0]);
2163                 if (found_key.objectid != key.objectid)
2164                         break;
2165
2166                 /* chunk zero is special */
2167                 if (found_key.offset == 0)
2168                         break;
2169
2170                 btrfs_release_path(path);
2171                 ret = btrfs_relocate_chunk(chunk_root,
2172                                            chunk_root->root_key.objectid,
2173                                            found_key.objectid,
2174                                            found_key.offset);
2175                 if (ret && ret != -ENOSPC)
2176                         goto error;
2177                 key.offset = found_key.offset - 1;
2178         }
2179         ret = 0;
2180 error:
2181         btrfs_free_path(path);
2182         mutex_unlock(&dev_root->fs_info->volume_mutex);
2183         return ret;
2184 }
2185
2186 /*
2187  * shrinking a device means finding all of the device extents past
2188  * the new size, and then following the back refs to the chunks.
2189  * The chunk relocation code actually frees the device extent
2190  */
2191 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
2192 {
2193         struct btrfs_trans_handle *trans;
2194         struct btrfs_root *root = device->dev_root;
2195         struct btrfs_dev_extent *dev_extent = NULL;
2196         struct btrfs_path *path;
2197         u64 length;
2198         u64 chunk_tree;
2199         u64 chunk_objectid;
2200         u64 chunk_offset;
2201         int ret;
2202         int slot;
2203         int failed = 0;
2204         bool retried = false;
2205         struct extent_buffer *l;
2206         struct btrfs_key key;
2207         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
2208         u64 old_total = btrfs_super_total_bytes(super_copy);
2209         u64 old_size = device->total_bytes;
2210         u64 diff = device->total_bytes - new_size;
2211
2212         if (new_size >= device->total_bytes)
2213                 return -EINVAL;
2214
2215         path = btrfs_alloc_path();
2216         if (!path)
2217                 return -ENOMEM;
2218
2219         path->reada = 2;
2220
2221         lock_chunks(root);
2222
2223         device->total_bytes = new_size;
2224         if (device->writeable) {
2225                 device->fs_devices->total_rw_bytes -= diff;
2226                 spin_lock(&root->fs_info->free_chunk_lock);
2227                 root->fs_info->free_chunk_space -= diff;
2228                 spin_unlock(&root->fs_info->free_chunk_lock);
2229         }
2230         unlock_chunks(root);
2231
2232 again:
2233         key.objectid = device->devid;
2234         key.offset = (u64)-1;
2235         key.type = BTRFS_DEV_EXTENT_KEY;
2236
2237         while (1) {
2238                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2239                 if (ret < 0)
2240                         goto done;
2241
2242                 ret = btrfs_previous_item(root, path, 0, key.type);
2243                 if (ret < 0)
2244                         goto done;
2245                 if (ret) {
2246                         ret = 0;
2247                         btrfs_release_path(path);
2248                         break;
2249                 }
2250
2251                 l = path->nodes[0];
2252                 slot = path->slots[0];
2253                 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
2254
2255                 if (key.objectid != device->devid) {
2256                         btrfs_release_path(path);
2257                         break;
2258                 }
2259
2260                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
2261                 length = btrfs_dev_extent_length(l, dev_extent);
2262
2263                 if (key.offset + length <= new_size) {
2264                         btrfs_release_path(path);
2265                         break;
2266                 }
2267
2268                 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
2269                 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
2270                 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
2271                 btrfs_release_path(path);
2272
2273                 ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
2274                                            chunk_offset);
2275                 if (ret && ret != -ENOSPC)
2276                         goto done;
2277                 if (ret == -ENOSPC)
2278                         failed++;
2279                 key.offset -= 1;
2280         }
2281
2282         if (failed && !retried) {
2283                 failed = 0;
2284                 retried = true;
2285                 goto again;
2286         } else if (failed && retried) {
2287                 ret = -ENOSPC;
2288                 lock_chunks(root);
2289
2290                 device->total_bytes = old_size;
2291                 if (device->writeable)
2292                         device->fs_devices->total_rw_bytes += diff;
2293                 spin_lock(&root->fs_info->free_chunk_lock);
2294                 root->fs_info->free_chunk_space += diff;
2295                 spin_unlock(&root->fs_info->free_chunk_lock);
2296                 unlock_chunks(root);
2297                 goto done;
2298         }
2299
2300         /* Shrinking succeeded, else we would be at "done". */
2301         trans = btrfs_start_transaction(root, 0);
2302         if (IS_ERR(trans)) {
2303                 ret = PTR_ERR(trans);
2304                 goto done;
2305         }
2306
2307         lock_chunks(root);
2308
2309         device->disk_total_bytes = new_size;
2310         /* Now btrfs_update_device() will change the on-disk size. */
2311         ret = btrfs_update_device(trans, device);
2312         if (ret) {
2313                 unlock_chunks(root);
2314                 btrfs_end_transaction(trans, root);
2315                 goto done;
2316         }
2317         WARN_ON(diff > old_total);
2318         btrfs_set_super_total_bytes(super_copy, old_total - diff);
2319         unlock_chunks(root);
2320         btrfs_end_transaction(trans, root);
2321 done:
2322         btrfs_free_path(path);
2323         return ret;
2324 }
2325
2326 static int btrfs_add_system_chunk(struct btrfs_trans_handle *trans,
2327                            struct btrfs_root *root,
2328                            struct btrfs_key *key,
2329                            struct btrfs_chunk *chunk, int item_size)
2330 {
2331         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
2332         struct btrfs_disk_key disk_key;
2333         u32 array_size;
2334         u8 *ptr;
2335
2336         array_size = btrfs_super_sys_array_size(super_copy);
2337         if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
2338                 return -EFBIG;
2339
2340         ptr = super_copy->sys_chunk_array + array_size;
2341         btrfs_cpu_key_to_disk(&disk_key, key);
2342         memcpy(ptr, &disk_key, sizeof(disk_key));
2343         ptr += sizeof(disk_key);
2344         memcpy(ptr, chunk, item_size);
2345         item_size += sizeof(disk_key);
2346         btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
2347         return 0;
2348 }
2349
2350 /*
2351  * sort the devices in descending order by max_avail, total_avail
2352  */
2353 static int btrfs_cmp_device_info(const void *a, const void *b)
2354 {
2355         const struct btrfs_device_info *di_a = a;
2356         const struct btrfs_device_info *di_b = b;
2357
2358         if (di_a->max_avail > di_b->max_avail)
2359                 return -1;
2360         if (di_a->max_avail < di_b->max_avail)
2361                 return 1;
2362         if (di_a->total_avail > di_b->total_avail)
2363                 return -1;
2364         if (di_a->total_avail < di_b->total_avail)
2365                 return 1;
2366         return 0;
2367 }
2368
2369 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
2370                                struct btrfs_root *extent_root,
2371                                struct map_lookup **map_ret,
2372                                u64 *num_bytes_out, u64 *stripe_size_out,
2373                                u64 start, u64 type)
2374 {
2375         struct btrfs_fs_info *info = extent_root->fs_info;
2376         struct btrfs_fs_devices *fs_devices = info->fs_devices;
2377         struct list_head *cur;
2378         struct map_lookup *map = NULL;
2379         struct extent_map_tree *em_tree;
2380         struct extent_map *em;
2381         struct btrfs_device_info *devices_info = NULL;
2382         u64 total_avail;
2383         int num_stripes;        /* total number of stripes to allocate */
2384         int sub_stripes;        /* sub_stripes info for map */
2385         int dev_stripes;        /* stripes per dev */
2386         int devs_max;           /* max devs to use */
2387         int devs_min;           /* min devs needed */
2388         int devs_increment;     /* ndevs has to be a multiple of this */
2389         int ncopies;            /* how many copies to data has */
2390         int ret;
2391         u64 max_stripe_size;
2392         u64 max_chunk_size;
2393         u64 stripe_size;
2394         u64 num_bytes;
2395         int ndevs;
2396         int i;
2397         int j;
2398
2399         if ((type & BTRFS_BLOCK_GROUP_RAID1) &&
2400             (type & BTRFS_BLOCK_GROUP_DUP)) {
2401                 WARN_ON(1);
2402                 type &= ~BTRFS_BLOCK_GROUP_DUP;
2403         }
2404
2405         if (list_empty(&fs_devices->alloc_list))
2406                 return -ENOSPC;
2407
2408         sub_stripes = 1;
2409         dev_stripes = 1;
2410         devs_increment = 1;
2411         ncopies = 1;
2412         devs_max = 0;   /* 0 == as many as possible */
2413         devs_min = 1;
2414
2415         /*
2416          * define the properties of each RAID type.
2417          * FIXME: move this to a global table and use it in all RAID
2418          * calculation code
2419          */
2420         if (type & (BTRFS_BLOCK_GROUP_DUP)) {
2421                 dev_stripes = 2;
2422                 ncopies = 2;
2423                 devs_max = 1;
2424         } else if (type & (BTRFS_BLOCK_GROUP_RAID0)) {
2425                 devs_min = 2;
2426         } else if (type & (BTRFS_BLOCK_GROUP_RAID1)) {
2427                 devs_increment = 2;
2428                 ncopies = 2;
2429                 devs_max = 2;
2430                 devs_min = 2;
2431         } else if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
2432                 sub_stripes = 2;
2433                 devs_increment = 2;
2434                 ncopies = 2;
2435                 devs_min = 4;
2436         } else {
2437                 devs_max = 1;
2438         }
2439
2440         if (type & BTRFS_BLOCK_GROUP_DATA) {
2441                 max_stripe_size = 1024 * 1024 * 1024;
2442                 max_chunk_size = 10 * max_stripe_size;
2443         } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
2444                 max_stripe_size = 256 * 1024 * 1024;
2445                 max_chunk_size = max_stripe_size;
2446         } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
2447                 max_stripe_size = 8 * 1024 * 1024;
2448                 max_chunk_size = 2 * max_stripe_size;
2449         } else {
2450                 printk(KERN_ERR "btrfs: invalid chunk type 0x%llx requested\n",
2451                        type);
2452                 BUG_ON(1);
2453         }
2454
2455         /* we don't want a chunk larger than 10% of writeable space */
2456         max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
2457                              max_chunk_size);
2458
2459         devices_info = kzalloc(sizeof(*devices_info) * fs_devices->rw_devices,
2460                                GFP_NOFS);
2461         if (!devices_info)
2462                 return -ENOMEM;
2463
2464         cur = fs_devices->alloc_list.next;
2465
2466         /*
2467          * in the first pass through the devices list, we gather information
2468          * about the available holes on each device.
2469          */
2470         ndevs = 0;
2471         while (cur != &fs_devices->alloc_list) {
2472                 struct btrfs_device *device;
2473                 u64 max_avail;
2474                 u64 dev_offset;
2475
2476                 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
2477
2478                 cur = cur->next;
2479
2480                 if (!device->writeable) {
2481                         printk(KERN_ERR
2482                                "btrfs: read-only device in alloc_list\n");
2483                         WARN_ON(1);
2484                         continue;
2485                 }
2486
2487                 if (!device->in_fs_metadata)
2488                         continue;
2489
2490                 if (device->total_bytes > device->bytes_used)
2491                         total_avail = device->total_bytes - device->bytes_used;
2492                 else
2493                         total_avail = 0;
2494
2495                 /* If there is no space on this device, skip it. */
2496                 if (total_avail == 0)
2497                         continue;
2498
2499                 ret = find_free_dev_extent(trans, device,
2500                                            max_stripe_size * dev_stripes,
2501                                            &dev_offset, &max_avail);
2502                 if (ret && ret != -ENOSPC)
2503                         goto error;
2504
2505                 if (ret == 0)
2506                         max_avail = max_stripe_size * dev_stripes;
2507
2508                 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes)
2509                         continue;
2510
2511                 devices_info[ndevs].dev_offset = dev_offset;
2512                 devices_info[ndevs].max_avail = max_avail;
2513                 devices_info[ndevs].total_avail = total_avail;
2514                 devices_info[ndevs].dev = device;
2515                 ++ndevs;
2516         }
2517
2518         /*
2519          * now sort the devices by hole size / available space
2520          */
2521         sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
2522              btrfs_cmp_device_info, NULL);
2523
2524         /* round down to number of usable stripes */
2525         ndevs -= ndevs % devs_increment;
2526
2527         if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) {
2528                 ret = -ENOSPC;
2529                 goto error;
2530         }
2531
2532         if (devs_max && ndevs > devs_max)
2533                 ndevs = devs_max;
2534         /*
2535          * the primary goal is to maximize the number of stripes, so use as many
2536          * devices as possible, even if the stripes are not maximum sized.
2537          */
2538         stripe_size = devices_info[ndevs-1].max_avail;
2539         num_stripes = ndevs * dev_stripes;
2540
2541         if (stripe_size * num_stripes > max_chunk_size * ncopies) {
2542                 stripe_size = max_chunk_size * ncopies;
2543                 do_div(stripe_size, num_stripes);
2544         }
2545
2546         do_div(stripe_size, dev_stripes);
2547         do_div(stripe_size, BTRFS_STRIPE_LEN);
2548         stripe_size *= BTRFS_STRIPE_LEN;
2549
2550         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
2551         if (!map) {
2552                 ret = -ENOMEM;
2553                 goto error;
2554         }
2555         map->num_stripes = num_stripes;
2556
2557         for (i = 0; i < ndevs; ++i) {
2558                 for (j = 0; j < dev_stripes; ++j) {
2559                         int s = i * dev_stripes + j;
2560                         map->stripes[s].dev = devices_info[i].dev;
2561                         map->stripes[s].physical = devices_info[i].dev_offset +
2562                                                    j * stripe_size;
2563                 }
2564         }
2565         map->sector_size = extent_root->sectorsize;
2566         map->stripe_len = BTRFS_STRIPE_LEN;
2567         map->io_align = BTRFS_STRIPE_LEN;
2568         map->io_width = BTRFS_STRIPE_LEN;
2569         map->type = type;
2570         map->sub_stripes = sub_stripes;
2571
2572         *map_ret = map;
2573         num_bytes = stripe_size * (num_stripes / ncopies);
2574
2575         *stripe_size_out = stripe_size;
2576         *num_bytes_out = num_bytes;
2577
2578         trace_btrfs_chunk_alloc(info->chunk_root, map, start, num_bytes);
2579
2580         em = alloc_extent_map();
2581         if (!em) {
2582                 ret = -ENOMEM;
2583                 goto error;
2584         }
2585         em->bdev = (struct block_device *)map;
2586         em->start = start;
2587         em->len = num_bytes;
2588         em->block_start = 0;
2589         em->block_len = em->len;
2590
2591         em_tree = &extent_root->fs_info->mapping_tree.map_tree;
2592         write_lock(&em_tree->lock);
2593         ret = add_extent_mapping(em_tree, em);
2594         write_unlock(&em_tree->lock);
2595         BUG_ON(ret);
2596         free_extent_map(em);
2597
2598         ret = btrfs_make_block_group(trans, extent_root, 0, type,
2599                                      BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2600                                      start, num_bytes);
2601         BUG_ON(ret);
2602
2603         for (i = 0; i < map->num_stripes; ++i) {
2604                 struct btrfs_device *device;
2605                 u64 dev_offset;
2606
2607                 device = map->stripes[i].dev;
2608                 dev_offset = map->stripes[i].physical;
2609
2610                 ret = btrfs_alloc_dev_extent(trans, device,
2611                                 info->chunk_root->root_key.objectid,
2612                                 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2613                                 start, dev_offset, stripe_size);
2614                 BUG_ON(ret);
2615         }
2616
2617         kfree(devices_info);
2618         return 0;
2619
2620 error:
2621         kfree(map);
2622         kfree(devices_info);
2623         return ret;
2624 }
2625
2626 static int __finish_chunk_alloc(struct btrfs_trans_handle *trans,
2627                                 struct btrfs_root *extent_root,
2628                                 struct map_lookup *map, u64 chunk_offset,
2629                                 u64 chunk_size, u64 stripe_size)
2630 {
2631         u64 dev_offset;
2632         struct btrfs_key key;
2633         struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
2634         struct btrfs_device *device;
2635         struct btrfs_chunk *chunk;
2636         struct btrfs_stripe *stripe;
2637         size_t item_size = btrfs_chunk_item_size(map->num_stripes);
2638         int index = 0;
2639         int ret;
2640
2641         chunk = kzalloc(item_size, GFP_NOFS);
2642         if (!chunk)
2643                 return -ENOMEM;
2644
2645         index = 0;
2646         while (index < map->num_stripes) {
2647                 device = map->stripes[index].dev;
2648                 device->bytes_used += stripe_size;
2649                 ret = btrfs_update_device(trans, device);
2650                 BUG_ON(ret);
2651                 index++;
2652         }
2653
2654         spin_lock(&extent_root->fs_info->free_chunk_lock);
2655         extent_root->fs_info->free_chunk_space -= (stripe_size *
2656                                                    map->num_stripes);
2657         spin_unlock(&extent_root->fs_info->free_chunk_lock);
2658
2659         index = 0;
2660         stripe = &chunk->stripe;
2661         while (index < map->num_stripes) {
2662                 device = map->stripes[index].dev;
2663                 dev_offset = map->stripes[index].physical;
2664
2665                 btrfs_set_stack_stripe_devid(stripe, device->devid);
2666                 btrfs_set_stack_stripe_offset(stripe, dev_offset);
2667                 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
2668                 stripe++;
2669                 index++;
2670         }
2671
2672         btrfs_set_stack_chunk_length(chunk, chunk_size);
2673         btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
2674         btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
2675         btrfs_set_stack_chunk_type(chunk, map->type);
2676         btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
2677         btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
2678         btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
2679         btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
2680         btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
2681
2682         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2683         key.type = BTRFS_CHUNK_ITEM_KEY;
2684         key.offset = chunk_offset;
2685
2686         ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
2687         BUG_ON(ret);
2688
2689         if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2690                 ret = btrfs_add_system_chunk(trans, chunk_root, &key, chunk,
2691                                              item_size);
2692                 BUG_ON(ret);
2693         }
2694
2695         kfree(chunk);
2696         return 0;
2697 }
2698
2699 /*
2700  * Chunk allocation falls into two parts. The first part does works
2701  * that make the new allocated chunk useable, but not do any operation
2702  * that modifies the chunk tree. The second part does the works that
2703  * require modifying the chunk tree. This division is important for the
2704  * bootstrap process of adding storage to a seed btrfs.
2705  */
2706 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
2707                       struct btrfs_root *extent_root, u64 type)
2708 {
2709         u64 chunk_offset;
2710         u64 chunk_size;
2711         u64 stripe_size;
2712         struct map_lookup *map;
2713         struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
2714         int ret;
2715
2716         ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2717                               &chunk_offset);
2718         if (ret)
2719                 return ret;
2720
2721         ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
2722                                   &stripe_size, chunk_offset, type);
2723         if (ret)
2724                 return ret;
2725
2726         ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
2727                                    chunk_size, stripe_size);
2728         BUG_ON(ret);
2729         return 0;
2730 }
2731
2732 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
2733                                          struct btrfs_root *root,
2734                                          struct btrfs_device *device)
2735 {
2736         u64 chunk_offset;
2737         u64 sys_chunk_offset;
2738         u64 chunk_size;
2739         u64 sys_chunk_size;
2740         u64 stripe_size;
2741         u64 sys_stripe_size;
2742         u64 alloc_profile;
2743         struct map_lookup *map;
2744         struct map_lookup *sys_map;
2745         struct btrfs_fs_info *fs_info = root->fs_info;
2746         struct btrfs_root *extent_root = fs_info->extent_root;
2747         int ret;
2748
2749         ret = find_next_chunk(fs_info->chunk_root,
2750                               BTRFS_FIRST_CHUNK_TREE_OBJECTID, &chunk_offset);
2751         if (ret)
2752                 return ret;
2753
2754         alloc_profile = BTRFS_BLOCK_GROUP_METADATA |
2755                         (fs_info->metadata_alloc_profile &
2756                          fs_info->avail_metadata_alloc_bits);
2757         alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
2758
2759         ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
2760                                   &stripe_size, chunk_offset, alloc_profile);
2761         BUG_ON(ret);
2762
2763         sys_chunk_offset = chunk_offset + chunk_size;
2764
2765         alloc_profile = BTRFS_BLOCK_GROUP_SYSTEM |
2766                         (fs_info->system_alloc_profile &
2767                          fs_info->avail_system_alloc_bits);
2768         alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
2769
2770         ret = __btrfs_alloc_chunk(trans, extent_root, &sys_map,
2771                                   &sys_chunk_size, &sys_stripe_size,
2772                                   sys_chunk_offset, alloc_profile);
2773         BUG_ON(ret);
2774
2775         ret = btrfs_add_device(trans, fs_info->chunk_root, device);
2776         BUG_ON(ret);
2777
2778         /*
2779          * Modifying chunk tree needs allocating new blocks from both
2780          * system block group and metadata block group. So we only can
2781          * do operations require modifying the chunk tree after both
2782          * block groups were created.
2783          */
2784         ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
2785                                    chunk_size, stripe_size);
2786         BUG_ON(ret);
2787
2788         ret = __finish_chunk_alloc(trans, extent_root, sys_map,
2789                                    sys_chunk_offset, sys_chunk_size,
2790                                    sys_stripe_size);
2791         BUG_ON(ret);
2792         return 0;
2793 }
2794
2795 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
2796 {
2797         struct extent_map *em;
2798         struct map_lookup *map;
2799         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
2800         int readonly = 0;
2801         int i;
2802
2803         read_lock(&map_tree->map_tree.lock);
2804         em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
2805         read_unlock(&map_tree->map_tree.lock);
2806         if (!em)
2807                 return 1;
2808
2809         if (btrfs_test_opt(root, DEGRADED)) {
2810                 free_extent_map(em);
2811                 return 0;
2812         }
2813
2814         map = (struct map_lookup *)em->bdev;
2815         for (i = 0; i < map->num_stripes; i++) {
2816                 if (!map->stripes[i].dev->writeable) {
2817                         readonly = 1;
2818                         break;
2819                 }
2820         }
2821         free_extent_map(em);
2822         return readonly;
2823 }
2824
2825 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
2826 {
2827         extent_map_tree_init(&tree->map_tree);
2828 }
2829
2830 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
2831 {
2832         struct extent_map *em;
2833
2834         while (1) {
2835                 write_lock(&tree->map_tree.lock);
2836                 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
2837                 if (em)
2838                         remove_extent_mapping(&tree->map_tree, em);
2839                 write_unlock(&tree->map_tree.lock);
2840                 if (!em)
2841                         break;
2842                 kfree(em->bdev);
2843                 /* once for us */
2844                 free_extent_map(em);
2845                 /* once for the tree */
2846                 free_extent_map(em);
2847         }
2848 }
2849
2850 int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len)
2851 {
2852         struct extent_map *em;
2853         struct map_lookup *map;
2854         struct extent_map_tree *em_tree = &map_tree->map_tree;
2855         int ret;
2856
2857         read_lock(&em_tree->lock);
2858         em = lookup_extent_mapping(em_tree, logical, len);
2859         read_unlock(&em_tree->lock);
2860         BUG_ON(!em);
2861
2862         BUG_ON(em->start > logical || em->start + em->len < logical);
2863         map = (struct map_lookup *)em->bdev;
2864         if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
2865                 ret = map->num_stripes;
2866         else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
2867                 ret = map->sub_stripes;
2868         else
2869                 ret = 1;
2870         free_extent_map(em);
2871         return ret;
2872 }
2873
2874 static int find_live_mirror(struct map_lookup *map, int first, int num,
2875                             int optimal)
2876 {
2877         int i;
2878         if (map->stripes[optimal].dev->bdev)
2879                 return optimal;
2880         for (i = first; i < first + num; i++) {
2881                 if (map->stripes[i].dev->bdev)
2882                         return i;
2883         }
2884         /* we couldn't find one that doesn't fail.  Just return something
2885          * and the io error handling code will clean up eventually
2886          */
2887         return optimal;
2888 }
2889
2890 static int __btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
2891                              u64 logical, u64 *length,
2892                              struct btrfs_bio **bbio_ret,
2893                              int mirror_num)
2894 {
2895         struct extent_map *em;
2896         struct map_lookup *map;
2897         struct extent_map_tree *em_tree = &map_tree->map_tree;
2898         u64 offset;
2899         u64 stripe_offset;
2900         u64 stripe_end_offset;
2901         u64 stripe_nr;
2902         u64 stripe_nr_orig;
2903         u64 stripe_nr_end;
2904         int stripes_allocated = 8;
2905         int stripes_required = 1;
2906         int stripe_index;
2907         int i;
2908         int num_stripes;
2909         int max_errors = 0;
2910         struct btrfs_bio *bbio = NULL;
2911
2912         if (bbio_ret && !(rw & (REQ_WRITE | REQ_DISCARD)))
2913                 stripes_allocated = 1;
2914 again:
2915         if (bbio_ret) {
2916                 bbio = kzalloc(btrfs_bio_size(stripes_allocated),
2917                                 GFP_NOFS);
2918                 if (!bbio)
2919                         return -ENOMEM;
2920
2921                 atomic_set(&bbio->error, 0);
2922         }
2923
2924         read_lock(&em_tree->lock);
2925         em = lookup_extent_mapping(em_tree, logical, *length);
2926         read_unlock(&em_tree->lock);
2927
2928         if (!em) {
2929                 printk(KERN_CRIT "unable to find logical %llu len %llu\n",
2930                        (unsigned long long)logical,
2931                        (unsigned long long)*length);
2932                 BUG();
2933         }
2934
2935         BUG_ON(em->start > logical || em->start + em->len < logical);
2936         map = (struct map_lookup *)em->bdev;
2937         offset = logical - em->start;
2938
2939         if (mirror_num > map->num_stripes)
2940                 mirror_num = 0;
2941
2942         /* if our btrfs_bio struct is too small, back off and try again */
2943         if (rw & REQ_WRITE) {
2944                 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
2945                                  BTRFS_BLOCK_GROUP_DUP)) {
2946                         stripes_required = map->num_stripes;
2947                         max_errors = 1;
2948                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2949                         stripes_required = map->sub_stripes;
2950                         max_errors = 1;
2951                 }
2952         }
2953         if (rw & REQ_DISCARD) {
2954                 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
2955                                  BTRFS_BLOCK_GROUP_RAID1 |
2956                                  BTRFS_BLOCK_GROUP_DUP |
2957                                  BTRFS_BLOCK_GROUP_RAID10)) {
2958                         stripes_required = map->num_stripes;
2959                 }
2960         }
2961         if (bbio_ret && (rw & (REQ_WRITE | REQ_DISCARD)) &&
2962             stripes_allocated < stripes_required) {
2963                 stripes_allocated = map->num_stripes;
2964                 free_extent_map(em);
2965                 kfree(bbio);
2966                 goto again;
2967         }
2968         stripe_nr = offset;
2969         /*
2970          * stripe_nr counts the total number of stripes we have to stride
2971          * to get to this block
2972          */
2973         do_div(stripe_nr, map->stripe_len);
2974
2975         stripe_offset = stripe_nr * map->stripe_len;
2976         BUG_ON(offset < stripe_offset);
2977
2978         /* stripe_offset is the offset of this block in its stripe*/
2979         stripe_offset = offset - stripe_offset;
2980
2981         if (rw & REQ_DISCARD)
2982                 *length = min_t(u64, em->len - offset, *length);
2983         else if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
2984                               BTRFS_BLOCK_GROUP_RAID1 |
2985                               BTRFS_BLOCK_GROUP_RAID10 |
2986                               BTRFS_BLOCK_GROUP_DUP)) {
2987                 /* we limit the length of each bio to what fits in a stripe */
2988                 *length = min_t(u64, em->len - offset,
2989                                 map->stripe_len - stripe_offset);
2990         } else {
2991                 *length = em->len - offset;
2992         }
2993
2994         if (!bbio_ret)
2995                 goto out;
2996
2997         num_stripes = 1;
2998         stripe_index = 0;
2999         stripe_nr_orig = stripe_nr;
3000         stripe_nr_end = (offset + *length + map->stripe_len - 1) &
3001                         (~(map->stripe_len - 1));
3002         do_div(stripe_nr_end, map->stripe_len);
3003         stripe_end_offset = stripe_nr_end * map->stripe_len -
3004                             (offset + *length);
3005         if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
3006                 if (rw & REQ_DISCARD)
3007                         num_stripes = min_t(u64, map->num_stripes,
3008                                             stripe_nr_end - stripe_nr_orig);
3009                 stripe_index = do_div(stripe_nr, map->num_stripes);
3010         } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
3011                 if (rw & (REQ_WRITE | REQ_DISCARD))
3012                         num_stripes = map->num_stripes;
3013                 else if (mirror_num)
3014                         stripe_index = mirror_num - 1;
3015                 else {
3016                         stripe_index = find_live_mirror(map, 0,
3017                                             map->num_stripes,
3018                                             current->pid % map->num_stripes);
3019                         mirror_num = stripe_index + 1;
3020                 }
3021
3022         } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
3023                 if (rw & (REQ_WRITE | REQ_DISCARD)) {
3024                         num_stripes = map->num_stripes;
3025                 } else if (mirror_num) {
3026                         stripe_index = mirror_num - 1;
3027                 } else {
3028                         mirror_num = 1;
3029                 }
3030
3031         } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
3032                 int factor = map->num_stripes / map->sub_stripes;
3033
3034                 stripe_index = do_div(stripe_nr, factor);
3035                 stripe_index *= map->sub_stripes;
3036
3037                 if (rw & REQ_WRITE)
3038                         num_stripes = map->sub_stripes;
3039                 else if (rw & REQ_DISCARD)
3040                         num_stripes = min_t(u64, map->sub_stripes *
3041                                             (stripe_nr_end - stripe_nr_orig),
3042                                             map->num_stripes);
3043                 else if (mirror_num)
3044                         stripe_index += mirror_num - 1;
3045                 else {
3046                         stripe_index = find_live_mirror(map, stripe_index,
3047                                               map->sub_stripes, stripe_index +
3048                                               current->pid % map->sub_stripes);
3049                         mirror_num = stripe_index + 1;
3050                 }
3051         } else {
3052                 /*
3053                  * after this do_div call, stripe_nr is the number of stripes
3054                  * on this device we have to walk to find the data, and
3055                  * stripe_index is the number of our device in the stripe array
3056                  */
3057                 stripe_index = do_div(stripe_nr, map->num_stripes);
3058                 mirror_num = stripe_index + 1;
3059         }
3060         BUG_ON(stripe_index >= map->num_stripes);
3061
3062         if (rw & REQ_DISCARD) {
3063                 for (i = 0; i < num_stripes; i++) {
3064                         bbio->stripes[i].physical =
3065                                 map->stripes[stripe_index].physical +
3066                                 stripe_offset + stripe_nr * map->stripe_len;
3067                         bbio->stripes[i].dev = map->stripes[stripe_index].dev;
3068
3069                         if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
3070                                 u64 stripes;
3071                                 u32 last_stripe = 0;
3072                                 int j;
3073
3074                                 div_u64_rem(stripe_nr_end - 1,
3075                                             map->num_stripes,
3076                                             &last_stripe);
3077
3078                                 for (j = 0; j < map->num_stripes; j++) {
3079                                         u32 test;
3080
3081                                         div_u64_rem(stripe_nr_end - 1 - j,
3082                                                     map->num_stripes, &test);
3083                                         if (test == stripe_index)
3084                                                 break;
3085                                 }
3086                                 stripes = stripe_nr_end - 1 - j;
3087                                 do_div(stripes, map->num_stripes);
3088                                 bbio->stripes[i].length = map->stripe_len *
3089                                         (stripes - stripe_nr + 1);
3090
3091                                 if (i == 0) {
3092                                         bbio->stripes[i].length -=
3093                                                 stripe_offset;
3094                                         stripe_offset = 0;
3095                                 }
3096                                 if (stripe_index == last_stripe)
3097                                         bbio->stripes[i].length -=
3098                                                 stripe_end_offset;
3099                         } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
3100                                 u64 stripes;
3101                                 int j;
3102                                 int factor = map->num_stripes /
3103                                              map->sub_stripes;
3104                                 u32 last_stripe = 0;
3105
3106                                 div_u64_rem(stripe_nr_end - 1,
3107                                             factor, &last_stripe);
3108                                 last_stripe *= map->sub_stripes;
3109
3110                                 for (j = 0; j < factor; j++) {
3111                                         u32 test;
3112
3113                                         div_u64_rem(stripe_nr_end - 1 - j,
3114                                                     factor, &test);
3115
3116                                         if (test ==
3117                                             stripe_index / map->sub_stripes)
3118                                                 break;
3119                                 }
3120                                 stripes = stripe_nr_end - 1 - j;
3121                                 do_div(stripes, factor);
3122                                 bbio->stripes[i].length = map->stripe_len *
3123                                         (stripes - stripe_nr + 1);
3124
3125                                 if (i < map->sub_stripes) {
3126                                         bbio->stripes[i].length -=
3127                                                 stripe_offset;
3128                                         if (i == map->sub_stripes - 1)
3129                                                 stripe_offset = 0;
3130                                 }
3131                                 if (stripe_index >= last_stripe &&
3132                                     stripe_index <= (last_stripe +
3133                                                      map->sub_stripes - 1)) {
3134                                         bbio->stripes[i].length -=
3135                                                 stripe_end_offset;
3136                                 }
3137                         } else
3138                                 bbio->stripes[i].length = *length;
3139
3140                         stripe_index++;
3141                         if (stripe_index == map->num_stripes) {
3142                                 /* This could only happen for RAID0/10 */
3143                                 stripe_index = 0;
3144                                 stripe_nr++;
3145                         }
3146                 }
3147         } else {
3148                 for (i = 0; i < num_stripes; i++) {
3149                         bbio->stripes[i].physical =
3150                                 map->stripes[stripe_index].physical +
3151                                 stripe_offset +
3152                                 stripe_nr * map->stripe_len;
3153                         bbio->stripes[i].dev =
3154                                 map->stripes[stripe_index].dev;
3155                         stripe_index++;
3156                 }
3157         }
3158         if (bbio_ret) {
3159                 *bbio_ret = bbio;
3160                 bbio->num_stripes = num_stripes;
3161                 bbio->max_errors = max_errors;
3162                 bbio->mirror_num = mirror_num;
3163         }
3164 out:
3165         free_extent_map(em);
3166         return 0;
3167 }
3168
3169 int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
3170                       u64 logical, u64 *length,
3171                       struct btrfs_bio **bbio_ret, int mirror_num)
3172 {
3173         return __btrfs_map_block(map_tree, rw, logical, length, bbio_ret,
3174                                  mirror_num);
3175 }
3176
3177 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
3178                      u64 chunk_start, u64 physical, u64 devid,
3179                      u64 **logical, int *naddrs, int *stripe_len)
3180 {
3181         struct extent_map_tree *em_tree = &map_tree->map_tree;
3182         struct extent_map *em;
3183         struct map_lookup *map;
3184         u64 *buf;
3185         u64 bytenr;
3186         u64 length;
3187         u64 stripe_nr;
3188         int i, j, nr = 0;
3189
3190         read_lock(&em_tree->lock);
3191         em = lookup_extent_mapping(em_tree, chunk_start, 1);
3192         read_unlock(&em_tree->lock);
3193
3194         BUG_ON(!em || em->start != chunk_start);
3195         map = (struct map_lookup *)em->bdev;
3196
3197         length = em->len;
3198         if (map->type & BTRFS_BLOCK_GROUP_RAID10)
3199                 do_div(length, map->num_stripes / map->sub_stripes);
3200         else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
3201                 do_div(length, map->num_stripes);
3202
3203         buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
3204         BUG_ON(!buf);
3205
3206         for (i = 0; i < map->num_stripes; i++) {
3207                 if (devid && map->stripes[i].dev->devid != devid)
3208                         continue;
3209                 if (map->stripes[i].physical > physical ||
3210                     map->stripes[i].physical + length <= physical)
3211                         continue;
3212
3213                 stripe_nr = physical - map->stripes[i].physical;
3214                 do_div(stripe_nr, map->stripe_len);
3215
3216                 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
3217                         stripe_nr = stripe_nr * map->num_stripes + i;
3218                         do_div(stripe_nr, map->sub_stripes);
3219                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
3220                         stripe_nr = stripe_nr * map->num_stripes + i;
3221                 }
3222                 bytenr = chunk_start + stripe_nr * map->stripe_len;
3223                 WARN_ON(nr >= map->num_stripes);
3224                 for (j = 0; j < nr; j++) {
3225                         if (buf[j] == bytenr)
3226                                 break;
3227                 }
3228                 if (j == nr) {
3229                         WARN_ON(nr >= map->num_stripes);
3230                         buf[nr++] = bytenr;
3231                 }
3232         }
3233
3234         *logical = buf;
3235         *naddrs = nr;
3236         *stripe_len = map->stripe_len;
3237
3238         free_extent_map(em);
3239         return 0;
3240 }
3241
3242 static void btrfs_end_bio(struct bio *bio, int err)
3243 {
3244         struct btrfs_bio *bbio = bio->bi_private;
3245         int is_orig_bio = 0;
3246
3247         if (err)
3248                 atomic_inc(&bbio->error);
3249
3250         if (bio == bbio->orig_bio)
3251                 is_orig_bio = 1;
3252
3253         if (atomic_dec_and_test(&bbio->stripes_pending)) {
3254                 if (!is_orig_bio) {
3255                         bio_put(bio);
3256                         bio = bbio->orig_bio;
3257                 }
3258                 bio->bi_private = bbio->private;
3259                 bio->bi_end_io = bbio->end_io;
3260                 bio->bi_bdev = (struct block_device *)
3261                                         (unsigned long)bbio->mirror_num;
3262                 /* only send an error to the higher layers if it is
3263                  * beyond the tolerance of the multi-bio
3264                  */
3265                 if (atomic_read(&bbio->error) > bbio->max_errors) {
3266                         err = -EIO;
3267                 } else {
3268                         /*
3269                          * this bio is actually up to date, we didn't
3270                          * go over the max number of errors
3271                          */
3272                         set_bit(BIO_UPTODATE, &bio->bi_flags);
3273                         err = 0;
3274                 }
3275                 kfree(bbio);
3276
3277                 bio_endio(bio, err);
3278         } else if (!is_orig_bio) {
3279                 bio_put(bio);
3280         }
3281 }
3282
3283 struct async_sched {
3284         struct bio *bio;
3285         int rw;
3286         struct btrfs_fs_info *info;
3287         struct btrfs_work work;
3288 };
3289
3290 /*
3291  * see run_scheduled_bios for a description of why bios are collected for
3292  * async submit.
3293  *
3294  * This will add one bio to the pending list for a device and make sure
3295  * the work struct is scheduled.
3296  */
3297 static noinline int schedule_bio(struct btrfs_root *root,
3298                                  struct btrfs_device *device,
3299                                  int rw, struct bio *bio)
3300 {
3301         int should_queue = 1;
3302         struct btrfs_pending_bios *pending_bios;
3303
3304         /* don't bother with additional async steps for reads, right now */
3305         if (!(rw & REQ_WRITE)) {
3306                 bio_get(bio);
3307                 submit_bio(rw, bio);
3308                 bio_put(bio);
3309                 return 0;
3310         }
3311
3312         /*
3313          * nr_async_bios allows us to reliably return congestion to the
3314          * higher layers.  Otherwise, the async bio makes it appear we have
3315          * made progress against dirty pages when we've really just put it
3316          * on a queue for later
3317          */
3318         atomic_inc(&root->fs_info->nr_async_bios);
3319         WARN_ON(bio->bi_next);
3320         bio->bi_next = NULL;
3321         bio->bi_rw |= rw;
3322
3323         spin_lock(&device->io_lock);
3324         if (bio->bi_rw & REQ_SYNC)
3325                 pending_bios = &device->pending_sync_bios;
3326         else
3327                 pending_bios = &device->pending_bios;
3328
3329         if (pending_bios->tail)
3330                 pending_bios->tail->bi_next = bio;
3331
3332         pending_bios->tail = bio;
3333         if (!pending_bios->head)
3334                 pending_bios->head = bio;
3335         if (device->running_pending)
3336                 should_queue = 0;
3337
3338         spin_unlock(&device->io_lock);
3339
3340         if (should_queue)
3341                 btrfs_queue_worker(&root->fs_info->submit_workers,
3342                                    &device->work);
3343         return 0;
3344 }
3345
3346 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
3347                   int mirror_num, int async_submit)
3348 {
3349         struct btrfs_mapping_tree *map_tree;
3350         struct btrfs_device *dev;
3351         struct bio *first_bio = bio;
3352         u64 logical = (u64)bio->bi_sector << 9;
3353         u64 length = 0;
3354         u64 map_length;
3355         int ret;
3356         int dev_nr = 0;
3357         int total_devs = 1;
3358         struct btrfs_bio *bbio = NULL;
3359
3360         length = bio->bi_size;
3361         map_tree = &root->fs_info->mapping_tree;
3362         map_length = length;
3363
3364         ret = btrfs_map_block(map_tree, rw, logical, &map_length, &bbio,
3365                               mirror_num);
3366         BUG_ON(ret);
3367
3368         total_devs = bbio->num_stripes;
3369         if (map_length < length) {
3370                 printk(KERN_CRIT "mapping failed logical %llu bio len %llu "
3371                        "len %llu\n", (unsigned long long)logical,
3372                        (unsigned long long)length,
3373                        (unsigned long long)map_length);
3374                 BUG();
3375         }
3376
3377         bbio->orig_bio = first_bio;
3378         bbio->private = first_bio->bi_private;
3379         bbio->end_io = first_bio->bi_end_io;
3380         atomic_set(&bbio->stripes_pending, bbio->num_stripes);
3381
3382         while (dev_nr < total_devs) {
3383                 if (dev_nr < total_devs - 1) {
3384                         bio = bio_clone(first_bio, GFP_NOFS);
3385                         BUG_ON(!bio);
3386                 } else {
3387                         bio = first_bio;
3388                 }
3389                 bio->bi_private = bbio;
3390                 bio->bi_end_io = btrfs_end_bio;
3391                 bio->bi_sector = bbio->stripes[dev_nr].physical >> 9;
3392                 dev = bbio->stripes[dev_nr].dev;
3393                 if (dev && dev->bdev && (rw != WRITE || dev->writeable)) {
3394                         pr_debug("btrfs_map_bio: rw %d, secor=%llu, dev=%lu "
3395                                  "(%s id %llu), size=%u\n", rw,
3396                                  (u64)bio->bi_sector, (u_long)dev->bdev->bd_dev,
3397                                  dev->name, dev->devid, bio->bi_size);
3398                         bio->bi_bdev = dev->bdev;
3399                         if (async_submit)
3400                                 schedule_bio(root, dev, rw, bio);
3401                         else
3402                                 submit_bio(rw, bio);
3403                 } else {
3404                         bio->bi_bdev = root->fs_info->fs_devices->latest_bdev;
3405                         bio->bi_sector = logical >> 9;
3406                         bio_endio(bio, -EIO);
3407                 }
3408                 dev_nr++;
3409         }
3410         return 0;
3411 }
3412
3413 struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid,
3414                                        u8 *uuid, u8 *fsid)
3415 {
3416         struct btrfs_device *device;
3417         struct btrfs_fs_devices *cur_devices;
3418
3419         cur_devices = root->fs_info->fs_devices;
3420         while (cur_devices) {
3421                 if (!fsid ||
3422                     !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
3423                         device = __find_device(&cur_devices->devices,
3424                                                devid, uuid);
3425                         if (device)
3426                                 return device;
3427                 }
3428                 cur_devices = cur_devices->seed;
3429         }
3430         return NULL;
3431 }
3432
3433 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
3434                                             u64 devid, u8 *dev_uuid)
3435 {
3436         struct btrfs_device *device;
3437         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
3438
3439         device = kzalloc(sizeof(*device), GFP_NOFS);
3440         if (!device)
3441                 return NULL;
3442         list_add(&device->dev_list,
3443                  &fs_devices->devices);
3444         device->dev_root = root->fs_info->dev_root;
3445         device->devid = devid;
3446         device->work.func = pending_bios_fn;
3447         device->fs_devices = fs_devices;
3448         device->missing = 1;
3449         fs_devices->num_devices++;
3450         fs_devices->missing_devices++;
3451         spin_lock_init(&device->io_lock);
3452         INIT_LIST_HEAD(&device->dev_alloc_list);
3453         memcpy(device->uuid, dev_uuid, BTRFS_UUID_SIZE);
3454         return device;
3455 }
3456
3457 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
3458                           struct extent_buffer *leaf,
3459                           struct btrfs_chunk *chunk)
3460 {
3461         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
3462         struct map_lookup *map;
3463         struct extent_map *em;
3464         u64 logical;
3465         u64 length;
3466         u64 devid;
3467         u8 uuid[BTRFS_UUID_SIZE];
3468         int num_stripes;
3469         int ret;
3470         int i;
3471
3472         logical = key->offset;
3473         length = btrfs_chunk_length(leaf, chunk);
3474
3475         read_lock(&map_tree->map_tree.lock);
3476         em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
3477         read_unlock(&map_tree->map_tree.lock);
3478
3479         /* already mapped? */
3480         if (em && em->start <= logical && em->start + em->len > logical) {
3481                 free_extent_map(em);
3482                 return 0;
3483         } else if (em) {
3484                 free_extent_map(em);
3485         }
3486
3487         em = alloc_extent_map();
3488         if (!em)
3489                 return -ENOMEM;
3490         num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3491         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
3492         if (!map) {
3493                 free_extent_map(em);
3494                 return -ENOMEM;
3495         }
3496
3497         em->bdev = (struct block_device *)map;
3498         em->start = logical;
3499         em->len = length;
3500         em->block_start = 0;
3501         em->block_len = em->len;
3502
3503         map->num_stripes = num_stripes;
3504         map->io_width = btrfs_chunk_io_width(leaf, chunk);
3505         map->io_align = btrfs_chunk_io_align(leaf, chunk);
3506         map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
3507         map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
3508         map->type = btrfs_chunk_type(leaf, chunk);
3509         map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
3510         for (i = 0; i < num_stripes; i++) {
3511                 map->stripes[i].physical =
3512                         btrfs_stripe_offset_nr(leaf, chunk, i);
3513                 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
3514                 read_extent_buffer(leaf, uuid, (unsigned long)
3515                                    btrfs_stripe_dev_uuid_nr(chunk, i),
3516                                    BTRFS_UUID_SIZE);
3517                 map->stripes[i].dev = btrfs_find_device(root, devid, uuid,
3518                                                         NULL);
3519                 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
3520                         kfree(map);
3521                         free_extent_map(em);
3522                         return -EIO;
3523                 }
3524                 if (!map->stripes[i].dev) {
3525                         map->stripes[i].dev =
3526                                 add_missing_dev(root, devid, uuid);
3527                         if (!map->stripes[i].dev) {
3528                                 kfree(map);
3529                                 free_extent_map(em);
3530                                 return -EIO;
3531                         }
3532                 }
3533                 map->stripes[i].dev->in_fs_metadata = 1;
3534         }
3535
3536         write_lock(&map_tree->map_tree.lock);
3537         ret = add_extent_mapping(&map_tree->map_tree, em);
3538         write_unlock(&map_tree->map_tree.lock);
3539         BUG_ON(ret);
3540         free_extent_map(em);
3541
3542         return 0;
3543 }
3544
3545 static int fill_device_from_item(struct extent_buffer *leaf,
3546                                  struct btrfs_dev_item *dev_item,
3547                                  struct btrfs_device *device)
3548 {
3549         unsigned long ptr;
3550
3551         device->devid = btrfs_device_id(leaf, dev_item);
3552         device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
3553         device->total_bytes = device->disk_total_bytes;
3554         device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
3555         device->type = btrfs_device_type(leaf, dev_item);
3556         device->io_align = btrfs_device_io_align(leaf, dev_item);
3557         device->io_width = btrfs_device_io_width(leaf, dev_item);
3558         device->sector_size = btrfs_device_sector_size(leaf, dev_item);
3559
3560         ptr = (unsigned long)btrfs_device_uuid(dev_item);
3561         read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
3562
3563         return 0;
3564 }
3565
3566 static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
3567 {
3568         struct btrfs_fs_devices *fs_devices;
3569         int ret;
3570
3571         mutex_lock(&uuid_mutex);
3572
3573         fs_devices = root->fs_info->fs_devices->seed;
3574         while (fs_devices) {
3575                 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
3576                         ret = 0;
3577                         goto out;
3578                 }
3579                 fs_devices = fs_devices->seed;
3580         }
3581
3582         fs_devices = find_fsid(fsid);
3583         if (!fs_devices) {
3584                 ret = -ENOENT;
3585                 goto out;
3586         }
3587
3588         fs_devices = clone_fs_devices(fs_devices);
3589         if (IS_ERR(fs_devices)) {
3590                 ret = PTR_ERR(fs_devices);
3591                 goto out;
3592         }
3593
3594         ret = __btrfs_open_devices(fs_devices, FMODE_READ,
3595                                    root->fs_info->bdev_holder);
3596         if (ret)
3597                 goto out;
3598
3599         if (!fs_devices->seeding) {
3600                 __btrfs_close_devices(fs_devices);
3601                 free_fs_devices(fs_devices);
3602                 ret = -EINVAL;
3603                 goto out;
3604         }
3605
3606         fs_devices->seed = root->fs_info->fs_devices->seed;
3607         root->fs_info->fs_devices->seed = fs_devices;
3608 out:
3609         mutex_unlock(&uuid_mutex);
3610         return ret;
3611 }
3612
3613 static int read_one_dev(struct btrfs_root *root,
3614                         struct extent_buffer *leaf,
3615                         struct btrfs_dev_item *dev_item)
3616 {
3617         struct btrfs_device *device;
3618         u64 devid;
3619         int ret;
3620         u8 fs_uuid[BTRFS_UUID_SIZE];
3621         u8 dev_uuid[BTRFS_UUID_SIZE];
3622
3623         devid = btrfs_device_id(leaf, dev_item);
3624         read_extent_buffer(leaf, dev_uuid,
3625                            (unsigned long)btrfs_device_uuid(dev_item),
3626                            BTRFS_UUID_SIZE);
3627         read_extent_buffer(leaf, fs_uuid,
3628                            (unsigned long)btrfs_device_fsid(dev_item),
3629                            BTRFS_UUID_SIZE);
3630
3631         if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
3632                 ret = open_seed_devices(root, fs_uuid);
3633                 if (ret && !btrfs_test_opt(root, DEGRADED))
3634                         return ret;
3635         }
3636
3637         device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
3638         if (!device || !device->bdev) {
3639                 if (!btrfs_test_opt(root, DEGRADED))
3640                         return -EIO;
3641
3642                 if (!device) {
3643                         printk(KERN_WARNING "warning devid %llu missing\n",
3644                                (unsigned long long)devid);
3645                         device = add_missing_dev(root, devid, dev_uuid);
3646                         if (!device)
3647                                 return -ENOMEM;
3648                 } else if (!device->missing) {
3649                         /*
3650                          * this happens when a device that was properly setup
3651                          * in the device info lists suddenly goes bad.
3652                          * device->bdev is NULL, and so we have to set
3653                          * device->missing to one here
3654                          */
3655                         root->fs_info->fs_devices->missing_devices++;
3656                         device->missing = 1;
3657                 }
3658         }
3659
3660         if (device->fs_devices != root->fs_info->fs_devices) {
3661                 BUG_ON(device->writeable);
3662                 if (device->generation !=
3663                     btrfs_device_generation(leaf, dev_item))
3664                         return -EINVAL;
3665         }
3666
3667         fill_device_from_item(leaf, dev_item, device);
3668         device->dev_root = root->fs_info->dev_root;
3669         device->in_fs_metadata = 1;
3670         if (device->writeable) {
3671                 device->fs_devices->total_rw_bytes += device->total_bytes;
3672                 spin_lock(&root->fs_info->free_chunk_lock);
3673                 root->fs_info->free_chunk_space += device->total_bytes -
3674                         device->bytes_used;
3675                 spin_unlock(&root->fs_info->free_chunk_lock);
3676         }
3677         ret = 0;
3678         return ret;
3679 }
3680
3681 int btrfs_read_sys_array(struct btrfs_root *root)
3682 {
3683         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
3684         struct extent_buffer *sb;
3685         struct btrfs_disk_key *disk_key;
3686         struct btrfs_chunk *chunk;
3687         u8 *ptr;
3688         unsigned long sb_ptr;
3689         int ret = 0;
3690         u32 num_stripes;
3691         u32 array_size;
3692         u32 len = 0;
3693         u32 cur;
3694         struct btrfs_key key;
3695
3696         sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
3697                                           BTRFS_SUPER_INFO_SIZE);
3698         if (!sb)
3699                 return -ENOMEM;
3700         btrfs_set_buffer_uptodate(sb);
3701         btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
3702
3703         write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
3704         array_size = btrfs_super_sys_array_size(super_copy);
3705
3706         ptr = super_copy->sys_chunk_array;
3707         sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
3708         cur = 0;
3709
3710         while (cur < array_size) {
3711                 disk_key = (struct btrfs_disk_key *)ptr;
3712                 btrfs_disk_key_to_cpu(&key, disk_key);
3713
3714                 len = sizeof(*disk_key); ptr += len;
3715                 sb_ptr += len;
3716                 cur += len;
3717
3718                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
3719                         chunk = (struct btrfs_chunk *)sb_ptr;
3720                         ret = read_one_chunk(root, &key, sb, chunk);
3721                         if (ret)
3722                                 break;
3723                         num_stripes = btrfs_chunk_num_stripes(sb, chunk);
3724                         len = btrfs_chunk_item_size(num_stripes);
3725                 } else {
3726                         ret = -EIO;
3727                         break;
3728                 }
3729                 ptr += len;
3730                 sb_ptr += len;
3731                 cur += len;
3732         }
3733         free_extent_buffer(sb);
3734         return ret;
3735 }
3736
3737 int btrfs_read_chunk_tree(struct btrfs_root *root)
3738 {
3739         struct btrfs_path *path;
3740         struct extent_buffer *leaf;
3741         struct btrfs_key key;
3742         struct btrfs_key found_key;
3743         int ret;
3744         int slot;
3745
3746         root = root->fs_info->chunk_root;
3747
3748         path = btrfs_alloc_path();
3749         if (!path)
3750                 return -ENOMEM;
3751
3752         /* first we search for all of the device items, and then we
3753          * read in all of the chunk items.  This way we can create chunk
3754          * mappings that reference all of the devices that are afound
3755          */
3756         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
3757         key.offset = 0;
3758         key.type = 0;
3759 again:
3760         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3761         if (ret < 0)
3762                 goto error;
3763         while (1) {
3764                 leaf = path->nodes[0];
3765                 slot = path->slots[0];
3766                 if (slot >= btrfs_header_nritems(leaf)) {
3767                         ret = btrfs_next_leaf(root, path);
3768                         if (ret == 0)
3769                                 continue;
3770                         if (ret < 0)
3771                                 goto error;
3772                         break;
3773                 }
3774                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3775                 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
3776                         if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID)
3777                                 break;
3778                         if (found_key.type == BTRFS_DEV_ITEM_KEY) {
3779                                 struct btrfs_dev_item *dev_item;
3780                                 dev_item = btrfs_item_ptr(leaf, slot,
3781                                                   struct btrfs_dev_item);
3782                                 ret = read_one_dev(root, leaf, dev_item);
3783                                 if (ret)
3784                                         goto error;
3785                         }
3786                 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
3787                         struct btrfs_chunk *chunk;
3788                         chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3789                         ret = read_one_chunk(root, &found_key, leaf, chunk);
3790                         if (ret)
3791                                 goto error;
3792                 }
3793                 path->slots[0]++;
3794         }
3795         if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
3796                 key.objectid = 0;
3797                 btrfs_release_path(path);
3798                 goto again;
3799         }
3800         ret = 0;
3801 error:
3802         btrfs_free_path(path);
3803         return ret;
3804 }