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
1286         all_avail = root->fs_info->avail_data_alloc_bits |
1287                 root->fs_info->avail_system_alloc_bits |
1288                 root->fs_info->avail_metadata_alloc_bits;
1289
1290         if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) &&
1291             root->fs_info->fs_devices->num_devices <= 4) {
1292                 printk(KERN_ERR "btrfs: unable to go below four devices "
1293                        "on raid10\n");
1294                 ret = -EINVAL;
1295                 goto out;
1296         }
1297
1298         if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) &&
1299             root->fs_info->fs_devices->num_devices <= 2) {
1300                 printk(KERN_ERR "btrfs: unable to go below two "
1301                        "devices on raid1\n");
1302                 ret = -EINVAL;
1303                 goto out;
1304         }
1305
1306         if (strcmp(device_path, "missing") == 0) {
1307                 struct list_head *devices;
1308                 struct btrfs_device *tmp;
1309
1310                 device = NULL;
1311                 devices = &root->fs_info->fs_devices->devices;
1312                 /*
1313                  * It is safe to read the devices since the volume_mutex
1314                  * is held.
1315                  */
1316                 list_for_each_entry(tmp, devices, dev_list) {
1317                         if (tmp->in_fs_metadata && !tmp->bdev) {
1318                                 device = tmp;
1319                                 break;
1320                         }
1321                 }
1322                 bdev = NULL;
1323                 bh = NULL;
1324                 disk_super = NULL;
1325                 if (!device) {
1326                         printk(KERN_ERR "btrfs: no missing devices found to "
1327                                "remove\n");
1328                         goto out;
1329                 }
1330         } else {
1331                 bdev = blkdev_get_by_path(device_path, FMODE_READ | FMODE_EXCL,
1332                                           root->fs_info->bdev_holder);
1333                 if (IS_ERR(bdev)) {
1334                         ret = PTR_ERR(bdev);
1335                         goto out;
1336                 }
1337
1338                 set_blocksize(bdev, 4096);
1339                 bh = btrfs_read_dev_super(bdev);
1340                 if (!bh) {
1341                         ret = -EINVAL;
1342                         goto error_close;
1343                 }
1344                 disk_super = (struct btrfs_super_block *)bh->b_data;
1345                 devid = btrfs_stack_device_id(&disk_super->dev_item);
1346                 dev_uuid = disk_super->dev_item.uuid;
1347                 device = btrfs_find_device(root, devid, dev_uuid,
1348                                            disk_super->fsid);
1349                 if (!device) {
1350                         ret = -ENOENT;
1351                         goto error_brelse;
1352                 }
1353         }
1354
1355         if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1356                 printk(KERN_ERR "btrfs: unable to remove the only writeable "
1357                        "device\n");
1358                 ret = -EINVAL;
1359                 goto error_brelse;
1360         }
1361
1362         if (device->writeable) {
1363                 lock_chunks(root);
1364                 list_del_init(&device->dev_alloc_list);
1365                 unlock_chunks(root);
1366                 root->fs_info->fs_devices->rw_devices--;
1367                 clear_super = true;
1368         }
1369
1370         ret = btrfs_shrink_device(device, 0);
1371         if (ret)
1372                 goto error_undo;
1373
1374         ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1375         if (ret)
1376                 goto error_undo;
1377
1378         spin_lock(&root->fs_info->free_chunk_lock);
1379         root->fs_info->free_chunk_space = device->total_bytes -
1380                 device->bytes_used;
1381         spin_unlock(&root->fs_info->free_chunk_lock);
1382
1383         device->in_fs_metadata = 0;
1384         btrfs_scrub_cancel_dev(root, device);
1385
1386         /*
1387          * the device list mutex makes sure that we don't change
1388          * the device list while someone else is writing out all
1389          * the device supers.
1390          */
1391
1392         cur_devices = device->fs_devices;
1393         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1394         list_del_rcu(&device->dev_list);
1395
1396         device->fs_devices->num_devices--;
1397
1398         if (device->missing)
1399                 root->fs_info->fs_devices->missing_devices--;
1400
1401         next_device = list_entry(root->fs_info->fs_devices->devices.next,
1402                                  struct btrfs_device, dev_list);
1403         if (device->bdev == root->fs_info->sb->s_bdev)
1404                 root->fs_info->sb->s_bdev = next_device->bdev;
1405         if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1406                 root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1407
1408         if (device->bdev)
1409                 device->fs_devices->open_devices--;
1410
1411         call_rcu(&device->rcu, free_device);
1412         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1413
1414         num_devices = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
1415         btrfs_set_super_num_devices(root->fs_info->super_copy, num_devices);
1416
1417         if (cur_devices->open_devices == 0) {
1418                 struct btrfs_fs_devices *fs_devices;
1419                 fs_devices = root->fs_info->fs_devices;
1420                 while (fs_devices) {
1421                         if (fs_devices->seed == cur_devices)
1422                                 break;
1423                         fs_devices = fs_devices->seed;
1424                 }
1425                 fs_devices->seed = cur_devices->seed;
1426                 cur_devices->seed = NULL;
1427                 lock_chunks(root);
1428                 __btrfs_close_devices(cur_devices);
1429                 unlock_chunks(root);
1430                 free_fs_devices(cur_devices);
1431         }
1432
1433         /*
1434          * at this point, the device is zero sized.  We want to
1435          * remove it from the devices list and zero out the old super
1436          */
1437         if (clear_super) {
1438                 /* make sure this device isn't detected as part of
1439                  * the FS anymore
1440                  */
1441                 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1442                 set_buffer_dirty(bh);
1443                 sync_dirty_buffer(bh);
1444         }
1445
1446         ret = 0;
1447
1448 error_brelse:
1449         brelse(bh);
1450 error_close:
1451         if (bdev)
1452                 blkdev_put(bdev, FMODE_READ | FMODE_EXCL);
1453 out:
1454         mutex_unlock(&uuid_mutex);
1455         return ret;
1456 error_undo:
1457         if (device->writeable) {
1458                 lock_chunks(root);
1459                 list_add(&device->dev_alloc_list,
1460                          &root->fs_info->fs_devices->alloc_list);
1461                 unlock_chunks(root);
1462                 root->fs_info->fs_devices->rw_devices++;
1463         }
1464         goto error_brelse;
1465 }
1466
1467 /*
1468  * does all the dirty work required for changing file system's UUID.
1469  */
1470 static int btrfs_prepare_sprout(struct btrfs_trans_handle *trans,
1471                                 struct btrfs_root *root)
1472 {
1473         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
1474         struct btrfs_fs_devices *old_devices;
1475         struct btrfs_fs_devices *seed_devices;
1476         struct btrfs_super_block *disk_super = root->fs_info->super_copy;
1477         struct btrfs_device *device;
1478         u64 super_flags;
1479
1480         BUG_ON(!mutex_is_locked(&uuid_mutex));
1481         if (!fs_devices->seeding)
1482                 return -EINVAL;
1483
1484         seed_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
1485         if (!seed_devices)
1486                 return -ENOMEM;
1487
1488         old_devices = clone_fs_devices(fs_devices);
1489         if (IS_ERR(old_devices)) {
1490                 kfree(seed_devices);
1491                 return PTR_ERR(old_devices);
1492         }
1493
1494         list_add(&old_devices->list, &fs_uuids);
1495
1496         memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
1497         seed_devices->opened = 1;
1498         INIT_LIST_HEAD(&seed_devices->devices);
1499         INIT_LIST_HEAD(&seed_devices->alloc_list);
1500         mutex_init(&seed_devices->device_list_mutex);
1501
1502         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1503         list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
1504                               synchronize_rcu);
1505         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1506
1507         list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
1508         list_for_each_entry(device, &seed_devices->devices, dev_list) {
1509                 device->fs_devices = seed_devices;
1510         }
1511
1512         fs_devices->seeding = 0;
1513         fs_devices->num_devices = 0;
1514         fs_devices->open_devices = 0;
1515         fs_devices->seed = seed_devices;
1516
1517         generate_random_uuid(fs_devices->fsid);
1518         memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1519         memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1520         super_flags = btrfs_super_flags(disk_super) &
1521                       ~BTRFS_SUPER_FLAG_SEEDING;
1522         btrfs_set_super_flags(disk_super, super_flags);
1523
1524         return 0;
1525 }
1526
1527 /*
1528  * strore the expected generation for seed devices in device items.
1529  */
1530 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
1531                                struct btrfs_root *root)
1532 {
1533         struct btrfs_path *path;
1534         struct extent_buffer *leaf;
1535         struct btrfs_dev_item *dev_item;
1536         struct btrfs_device *device;
1537         struct btrfs_key key;
1538         u8 fs_uuid[BTRFS_UUID_SIZE];
1539         u8 dev_uuid[BTRFS_UUID_SIZE];
1540         u64 devid;
1541         int ret;
1542
1543         path = btrfs_alloc_path();
1544         if (!path)
1545                 return -ENOMEM;
1546
1547         root = root->fs_info->chunk_root;
1548         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1549         key.offset = 0;
1550         key.type = BTRFS_DEV_ITEM_KEY;
1551
1552         while (1) {
1553                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1554                 if (ret < 0)
1555                         goto error;
1556
1557                 leaf = path->nodes[0];
1558 next_slot:
1559                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1560                         ret = btrfs_next_leaf(root, path);
1561                         if (ret > 0)
1562                                 break;
1563                         if (ret < 0)
1564                                 goto error;
1565                         leaf = path->nodes[0];
1566                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1567                         btrfs_release_path(path);
1568                         continue;
1569                 }
1570
1571                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1572                 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
1573                     key.type != BTRFS_DEV_ITEM_KEY)
1574                         break;
1575
1576                 dev_item = btrfs_item_ptr(leaf, path->slots[0],
1577                                           struct btrfs_dev_item);
1578                 devid = btrfs_device_id(leaf, dev_item);
1579                 read_extent_buffer(leaf, dev_uuid,
1580                                    (unsigned long)btrfs_device_uuid(dev_item),
1581                                    BTRFS_UUID_SIZE);
1582                 read_extent_buffer(leaf, fs_uuid,
1583                                    (unsigned long)btrfs_device_fsid(dev_item),
1584                                    BTRFS_UUID_SIZE);
1585                 device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
1586                 BUG_ON(!device);
1587
1588                 if (device->fs_devices->seeding) {
1589                         btrfs_set_device_generation(leaf, dev_item,
1590                                                     device->generation);
1591                         btrfs_mark_buffer_dirty(leaf);
1592                 }
1593
1594                 path->slots[0]++;
1595                 goto next_slot;
1596         }
1597         ret = 0;
1598 error:
1599         btrfs_free_path(path);
1600         return ret;
1601 }
1602
1603 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
1604 {
1605         struct request_queue *q;
1606         struct btrfs_trans_handle *trans;
1607         struct btrfs_device *device;
1608         struct block_device *bdev;
1609         struct list_head *devices;
1610         struct super_block *sb = root->fs_info->sb;
1611         u64 total_bytes;
1612         int seeding_dev = 0;
1613         int ret = 0;
1614
1615         if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
1616                 return -EINVAL;
1617
1618         bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
1619                                   root->fs_info->bdev_holder);
1620         if (IS_ERR(bdev))
1621                 return PTR_ERR(bdev);
1622
1623         if (root->fs_info->fs_devices->seeding) {
1624                 seeding_dev = 1;
1625                 down_write(&sb->s_umount);
1626                 mutex_lock(&uuid_mutex);
1627         }
1628
1629         filemap_write_and_wait(bdev->bd_inode->i_mapping);
1630
1631         devices = &root->fs_info->fs_devices->devices;
1632         /*
1633          * we have the volume lock, so we don't need the extra
1634          * device list mutex while reading the list here.
1635          */
1636         list_for_each_entry(device, devices, dev_list) {
1637                 if (device->bdev == bdev) {
1638                         ret = -EEXIST;
1639                         goto error;
1640                 }
1641         }
1642
1643         device = kzalloc(sizeof(*device), GFP_NOFS);
1644         if (!device) {
1645                 /* we can safely leave the fs_devices entry around */
1646                 ret = -ENOMEM;
1647                 goto error;
1648         }
1649
1650         device->name = kstrdup(device_path, GFP_NOFS);
1651         if (!device->name) {
1652                 kfree(device);
1653                 ret = -ENOMEM;
1654                 goto error;
1655         }
1656
1657         ret = find_next_devid(root, &device->devid);
1658         if (ret) {
1659                 kfree(device->name);
1660                 kfree(device);
1661                 goto error;
1662         }
1663
1664         trans = btrfs_start_transaction(root, 0);
1665         if (IS_ERR(trans)) {
1666                 kfree(device->name);
1667                 kfree(device);
1668                 ret = PTR_ERR(trans);
1669                 goto error;
1670         }
1671
1672         lock_chunks(root);
1673
1674         q = bdev_get_queue(bdev);
1675         if (blk_queue_discard(q))
1676                 device->can_discard = 1;
1677         device->writeable = 1;
1678         device->work.func = pending_bios_fn;
1679         generate_random_uuid(device->uuid);
1680         spin_lock_init(&device->io_lock);
1681         device->generation = trans->transid;
1682         device->io_width = root->sectorsize;
1683         device->io_align = root->sectorsize;
1684         device->sector_size = root->sectorsize;
1685         device->total_bytes = i_size_read(bdev->bd_inode);
1686         device->disk_total_bytes = device->total_bytes;
1687         device->dev_root = root->fs_info->dev_root;
1688         device->bdev = bdev;
1689         device->in_fs_metadata = 1;
1690         device->mode = FMODE_EXCL;
1691         set_blocksize(device->bdev, 4096);
1692
1693         if (seeding_dev) {
1694                 sb->s_flags &= ~MS_RDONLY;
1695                 ret = btrfs_prepare_sprout(trans, root);
1696                 BUG_ON(ret);
1697         }
1698
1699         device->fs_devices = root->fs_info->fs_devices;
1700
1701         /*
1702          * we don't want write_supers to jump in here with our device
1703          * half setup
1704          */
1705         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1706         list_add_rcu(&device->dev_list, &root->fs_info->fs_devices->devices);
1707         list_add(&device->dev_alloc_list,
1708                  &root->fs_info->fs_devices->alloc_list);
1709         root->fs_info->fs_devices->num_devices++;
1710         root->fs_info->fs_devices->open_devices++;
1711         root->fs_info->fs_devices->rw_devices++;
1712         if (device->can_discard)
1713                 root->fs_info->fs_devices->num_can_discard++;
1714         root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
1715
1716         spin_lock(&root->fs_info->free_chunk_lock);
1717         root->fs_info->free_chunk_space += device->total_bytes;
1718         spin_unlock(&root->fs_info->free_chunk_lock);
1719
1720         if (!blk_queue_nonrot(bdev_get_queue(bdev)))
1721                 root->fs_info->fs_devices->rotating = 1;
1722
1723         total_bytes = btrfs_super_total_bytes(root->fs_info->super_copy);
1724         btrfs_set_super_total_bytes(root->fs_info->super_copy,
1725                                     total_bytes + device->total_bytes);
1726
1727         total_bytes = btrfs_super_num_devices(root->fs_info->super_copy);
1728         btrfs_set_super_num_devices(root->fs_info->super_copy,
1729                                     total_bytes + 1);
1730         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1731
1732         if (seeding_dev) {
1733                 ret = init_first_rw_device(trans, root, device);
1734                 BUG_ON(ret);
1735                 ret = btrfs_finish_sprout(trans, root);
1736                 BUG_ON(ret);
1737         } else {
1738                 ret = btrfs_add_device(trans, root, device);
1739         }
1740
1741         /*
1742          * we've got more storage, clear any full flags on the space
1743          * infos
1744          */
1745         btrfs_clear_space_info_full(root->fs_info);
1746
1747         unlock_chunks(root);
1748         btrfs_commit_transaction(trans, root);
1749
1750         if (seeding_dev) {
1751                 mutex_unlock(&uuid_mutex);
1752                 up_write(&sb->s_umount);
1753
1754                 ret = btrfs_relocate_sys_chunks(root);
1755                 BUG_ON(ret);
1756         }
1757
1758         return ret;
1759 error:
1760         blkdev_put(bdev, FMODE_EXCL);
1761         if (seeding_dev) {
1762                 mutex_unlock(&uuid_mutex);
1763                 up_write(&sb->s_umount);
1764         }
1765         return ret;
1766 }
1767
1768 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
1769                                         struct btrfs_device *device)
1770 {
1771         int ret;
1772         struct btrfs_path *path;
1773         struct btrfs_root *root;
1774         struct btrfs_dev_item *dev_item;
1775         struct extent_buffer *leaf;
1776         struct btrfs_key key;
1777
1778         root = device->dev_root->fs_info->chunk_root;
1779
1780         path = btrfs_alloc_path();
1781         if (!path)
1782                 return -ENOMEM;
1783
1784         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1785         key.type = BTRFS_DEV_ITEM_KEY;
1786         key.offset = device->devid;
1787
1788         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1789         if (ret < 0)
1790                 goto out;
1791
1792         if (ret > 0) {
1793                 ret = -ENOENT;
1794                 goto out;
1795         }
1796
1797         leaf = path->nodes[0];
1798         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1799
1800         btrfs_set_device_id(leaf, dev_item, device->devid);
1801         btrfs_set_device_type(leaf, dev_item, device->type);
1802         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1803         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1804         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1805         btrfs_set_device_total_bytes(leaf, dev_item, device->disk_total_bytes);
1806         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1807         btrfs_mark_buffer_dirty(leaf);
1808
1809 out:
1810         btrfs_free_path(path);
1811         return ret;
1812 }
1813
1814 static int __btrfs_grow_device(struct btrfs_trans_handle *trans,
1815                       struct btrfs_device *device, u64 new_size)
1816 {
1817         struct btrfs_super_block *super_copy =
1818                 device->dev_root->fs_info->super_copy;
1819         u64 old_total = btrfs_super_total_bytes(super_copy);
1820         u64 diff = new_size - device->total_bytes;
1821
1822         if (!device->writeable)
1823                 return -EACCES;
1824         if (new_size <= device->total_bytes)
1825                 return -EINVAL;
1826
1827         btrfs_set_super_total_bytes(super_copy, old_total + diff);
1828         device->fs_devices->total_rw_bytes += diff;
1829
1830         device->total_bytes = new_size;
1831         device->disk_total_bytes = new_size;
1832         btrfs_clear_space_info_full(device->dev_root->fs_info);
1833
1834         return btrfs_update_device(trans, device);
1835 }
1836
1837 int btrfs_grow_device(struct btrfs_trans_handle *trans,
1838                       struct btrfs_device *device, u64 new_size)
1839 {
1840         int ret;
1841         lock_chunks(device->dev_root);
1842         ret = __btrfs_grow_device(trans, device, new_size);
1843         unlock_chunks(device->dev_root);
1844         return ret;
1845 }
1846
1847 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
1848                             struct btrfs_root *root,
1849                             u64 chunk_tree, u64 chunk_objectid,
1850                             u64 chunk_offset)
1851 {
1852         int ret;
1853         struct btrfs_path *path;
1854         struct btrfs_key key;
1855
1856         root = root->fs_info->chunk_root;
1857         path = btrfs_alloc_path();
1858         if (!path)
1859                 return -ENOMEM;
1860
1861         key.objectid = chunk_objectid;
1862         key.offset = chunk_offset;
1863         key.type = BTRFS_CHUNK_ITEM_KEY;
1864
1865         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1866         BUG_ON(ret);
1867
1868         ret = btrfs_del_item(trans, root, path);
1869
1870         btrfs_free_path(path);
1871         return ret;
1872 }
1873
1874 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
1875                         chunk_offset)
1876 {
1877         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
1878         struct btrfs_disk_key *disk_key;
1879         struct btrfs_chunk *chunk;
1880         u8 *ptr;
1881         int ret = 0;
1882         u32 num_stripes;
1883         u32 array_size;
1884         u32 len = 0;
1885         u32 cur;
1886         struct btrfs_key key;
1887
1888         array_size = btrfs_super_sys_array_size(super_copy);
1889
1890         ptr = super_copy->sys_chunk_array;
1891         cur = 0;
1892
1893         while (cur < array_size) {
1894                 disk_key = (struct btrfs_disk_key *)ptr;
1895                 btrfs_disk_key_to_cpu(&key, disk_key);
1896
1897                 len = sizeof(*disk_key);
1898
1899                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
1900                         chunk = (struct btrfs_chunk *)(ptr + len);
1901                         num_stripes = btrfs_stack_chunk_num_stripes(chunk);
1902                         len += btrfs_chunk_item_size(num_stripes);
1903                 } else {
1904                         ret = -EIO;
1905                         break;
1906                 }
1907                 if (key.objectid == chunk_objectid &&
1908                     key.offset == chunk_offset) {
1909                         memmove(ptr, ptr + len, array_size - (cur + len));
1910                         array_size -= len;
1911                         btrfs_set_super_sys_array_size(super_copy, array_size);
1912                 } else {
1913                         ptr += len;
1914                         cur += len;
1915                 }
1916         }
1917         return ret;
1918 }
1919
1920 static int btrfs_relocate_chunk(struct btrfs_root *root,
1921                          u64 chunk_tree, u64 chunk_objectid,
1922                          u64 chunk_offset)
1923 {
1924         struct extent_map_tree *em_tree;
1925         struct btrfs_root *extent_root;
1926         struct btrfs_trans_handle *trans;
1927         struct extent_map *em;
1928         struct map_lookup *map;
1929         int ret;
1930         int i;
1931
1932         root = root->fs_info->chunk_root;
1933         extent_root = root->fs_info->extent_root;
1934         em_tree = &root->fs_info->mapping_tree.map_tree;
1935
1936         ret = btrfs_can_relocate(extent_root, chunk_offset);
1937         if (ret)
1938                 return -ENOSPC;
1939
1940         /* step one, relocate all the extents inside this chunk */
1941         ret = btrfs_relocate_block_group(extent_root, chunk_offset);
1942         if (ret)
1943                 return ret;
1944
1945         trans = btrfs_start_transaction(root, 0);
1946         BUG_ON(IS_ERR(trans));
1947
1948         lock_chunks(root);
1949
1950         /*
1951          * step two, delete the device extents and the
1952          * chunk tree entries
1953          */
1954         read_lock(&em_tree->lock);
1955         em = lookup_extent_mapping(em_tree, chunk_offset, 1);
1956         read_unlock(&em_tree->lock);
1957
1958         BUG_ON(em->start > chunk_offset ||
1959                em->start + em->len < chunk_offset);
1960         map = (struct map_lookup *)em->bdev;
1961
1962         for (i = 0; i < map->num_stripes; i++) {
1963                 ret = btrfs_free_dev_extent(trans, map->stripes[i].dev,
1964                                             map->stripes[i].physical);
1965                 BUG_ON(ret);
1966
1967                 if (map->stripes[i].dev) {
1968                         ret = btrfs_update_device(trans, map->stripes[i].dev);
1969                         BUG_ON(ret);
1970                 }
1971         }
1972         ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
1973                                chunk_offset);
1974
1975         BUG_ON(ret);
1976
1977         trace_btrfs_chunk_free(root, map, chunk_offset, em->len);
1978
1979         if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
1980                 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
1981                 BUG_ON(ret);
1982         }
1983
1984         ret = btrfs_remove_block_group(trans, extent_root, chunk_offset);
1985         BUG_ON(ret);
1986
1987         write_lock(&em_tree->lock);
1988         remove_extent_mapping(em_tree, em);
1989         write_unlock(&em_tree->lock);
1990
1991         kfree(map);
1992         em->bdev = NULL;
1993
1994         /* once for the tree */
1995         free_extent_map(em);
1996         /* once for us */
1997         free_extent_map(em);
1998
1999         unlock_chunks(root);
2000         btrfs_end_transaction(trans, root);
2001         return 0;
2002 }
2003
2004 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
2005 {
2006         struct btrfs_root *chunk_root = root->fs_info->chunk_root;
2007         struct btrfs_path *path;
2008         struct extent_buffer *leaf;
2009         struct btrfs_chunk *chunk;
2010         struct btrfs_key key;
2011         struct btrfs_key found_key;
2012         u64 chunk_tree = chunk_root->root_key.objectid;
2013         u64 chunk_type;
2014         bool retried = false;
2015         int failed = 0;
2016         int ret;
2017
2018         path = btrfs_alloc_path();
2019         if (!path)
2020                 return -ENOMEM;
2021
2022 again:
2023         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2024         key.offset = (u64)-1;
2025         key.type = BTRFS_CHUNK_ITEM_KEY;
2026
2027         while (1) {
2028                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2029                 if (ret < 0)
2030                         goto error;
2031                 BUG_ON(ret == 0);
2032
2033                 ret = btrfs_previous_item(chunk_root, path, key.objectid,
2034                                           key.type);
2035                 if (ret < 0)
2036                         goto error;
2037                 if (ret > 0)
2038                         break;
2039
2040                 leaf = path->nodes[0];
2041                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2042
2043                 chunk = btrfs_item_ptr(leaf, path->slots[0],
2044                                        struct btrfs_chunk);
2045                 chunk_type = btrfs_chunk_type(leaf, chunk);
2046                 btrfs_release_path(path);
2047
2048                 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
2049                         ret = btrfs_relocate_chunk(chunk_root, chunk_tree,
2050                                                    found_key.objectid,
2051                                                    found_key.offset);
2052                         if (ret == -ENOSPC)
2053                                 failed++;
2054                         else if (ret)
2055                                 BUG();
2056                 }
2057
2058                 if (found_key.offset == 0)
2059                         break;
2060                 key.offset = found_key.offset - 1;
2061         }
2062         ret = 0;
2063         if (failed && !retried) {
2064                 failed = 0;
2065                 retried = true;
2066                 goto again;
2067         } else if (failed && retried) {
2068                 WARN_ON(1);
2069                 ret = -ENOSPC;
2070         }
2071 error:
2072         btrfs_free_path(path);
2073         return ret;
2074 }
2075
2076 static int insert_balance_item(struct btrfs_root *root,
2077                                struct btrfs_balance_control *bctl)
2078 {
2079         struct btrfs_trans_handle *trans;
2080         struct btrfs_balance_item *item;
2081         struct btrfs_disk_balance_args disk_bargs;
2082         struct btrfs_path *path;
2083         struct extent_buffer *leaf;
2084         struct btrfs_key key;
2085         int ret, err;
2086
2087         path = btrfs_alloc_path();
2088         if (!path)
2089                 return -ENOMEM;
2090
2091         trans = btrfs_start_transaction(root, 0);
2092         if (IS_ERR(trans)) {
2093                 btrfs_free_path(path);
2094                 return PTR_ERR(trans);
2095         }
2096
2097         key.objectid = BTRFS_BALANCE_OBJECTID;
2098         key.type = BTRFS_BALANCE_ITEM_KEY;
2099         key.offset = 0;
2100
2101         ret = btrfs_insert_empty_item(trans, root, path, &key,
2102                                       sizeof(*item));
2103         if (ret)
2104                 goto out;
2105
2106         leaf = path->nodes[0];
2107         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
2108
2109         memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
2110
2111         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
2112         btrfs_set_balance_data(leaf, item, &disk_bargs);
2113         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
2114         btrfs_set_balance_meta(leaf, item, &disk_bargs);
2115         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
2116         btrfs_set_balance_sys(leaf, item, &disk_bargs);
2117
2118         btrfs_set_balance_flags(leaf, item, bctl->flags);
2119
2120         btrfs_mark_buffer_dirty(leaf);
2121 out:
2122         btrfs_free_path(path);
2123         err = btrfs_commit_transaction(trans, root);
2124         if (err && !ret)
2125                 ret = err;
2126         return ret;
2127 }
2128
2129 static int del_balance_item(struct btrfs_root *root)
2130 {
2131         struct btrfs_trans_handle *trans;
2132         struct btrfs_path *path;
2133         struct btrfs_key key;
2134         int ret, err;
2135
2136         path = btrfs_alloc_path();
2137         if (!path)
2138                 return -ENOMEM;
2139
2140         trans = btrfs_start_transaction(root, 0);
2141         if (IS_ERR(trans)) {
2142                 btrfs_free_path(path);
2143                 return PTR_ERR(trans);
2144         }
2145
2146         key.objectid = BTRFS_BALANCE_OBJECTID;
2147         key.type = BTRFS_BALANCE_ITEM_KEY;
2148         key.offset = 0;
2149
2150         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2151         if (ret < 0)
2152                 goto out;
2153         if (ret > 0) {
2154                 ret = -ENOENT;
2155                 goto out;
2156         }
2157
2158         ret = btrfs_del_item(trans, root, path);
2159 out:
2160         btrfs_free_path(path);
2161         err = btrfs_commit_transaction(trans, root);
2162         if (err && !ret)
2163                 ret = err;
2164         return ret;
2165 }
2166
2167 /*
2168  * Should be called with both balance and volume mutexes held to
2169  * serialize other volume operations (add_dev/rm_dev/resize) with
2170  * restriper.  Same goes for unset_balance_control.
2171  */
2172 static void set_balance_control(struct btrfs_balance_control *bctl)
2173 {
2174         struct btrfs_fs_info *fs_info = bctl->fs_info;
2175
2176         BUG_ON(fs_info->balance_ctl);
2177
2178         spin_lock(&fs_info->balance_lock);
2179         fs_info->balance_ctl = bctl;
2180         spin_unlock(&fs_info->balance_lock);
2181 }
2182
2183 static void unset_balance_control(struct btrfs_fs_info *fs_info)
2184 {
2185         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
2186
2187         BUG_ON(!fs_info->balance_ctl);
2188
2189         spin_lock(&fs_info->balance_lock);
2190         fs_info->balance_ctl = NULL;
2191         spin_unlock(&fs_info->balance_lock);
2192
2193         kfree(bctl);
2194 }
2195
2196 /*
2197  * Balance filters.  Return 1 if chunk should be filtered out
2198  * (should not be balanced).
2199  */
2200 static int chunk_profiles_filter(u64 chunk_profile,
2201                                  struct btrfs_balance_args *bargs)
2202 {
2203         chunk_profile &= BTRFS_BLOCK_GROUP_PROFILE_MASK;
2204
2205         if (chunk_profile == 0)
2206                 chunk_profile = BTRFS_AVAIL_ALLOC_BIT_SINGLE;
2207
2208         if (bargs->profiles & chunk_profile)
2209                 return 0;
2210
2211         return 1;
2212 }
2213
2214 static u64 div_factor_fine(u64 num, int factor)
2215 {
2216         if (factor <= 0)
2217                 return 0;
2218         if (factor >= 100)
2219                 return num;
2220
2221         num *= factor;
2222         do_div(num, 100);
2223         return num;
2224 }
2225
2226 static int chunk_usage_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
2227                               struct btrfs_balance_args *bargs)
2228 {
2229         struct btrfs_block_group_cache *cache;
2230         u64 chunk_used, user_thresh;
2231         int ret = 1;
2232
2233         cache = btrfs_lookup_block_group(fs_info, chunk_offset);
2234         chunk_used = btrfs_block_group_used(&cache->item);
2235
2236         user_thresh = div_factor_fine(cache->key.offset, bargs->usage);
2237         if (chunk_used < user_thresh)
2238                 ret = 0;
2239
2240         btrfs_put_block_group(cache);
2241         return ret;
2242 }
2243
2244 static int chunk_devid_filter(struct extent_buffer *leaf,
2245                               struct btrfs_chunk *chunk,
2246                               struct btrfs_balance_args *bargs)
2247 {
2248         struct btrfs_stripe *stripe;
2249         int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2250         int i;
2251
2252         for (i = 0; i < num_stripes; i++) {
2253                 stripe = btrfs_stripe_nr(chunk, i);
2254                 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
2255                         return 0;
2256         }
2257
2258         return 1;
2259 }
2260
2261 /* [pstart, pend) */
2262 static int chunk_drange_filter(struct extent_buffer *leaf,
2263                                struct btrfs_chunk *chunk,
2264                                u64 chunk_offset,
2265                                struct btrfs_balance_args *bargs)
2266 {
2267         struct btrfs_stripe *stripe;
2268         int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2269         u64 stripe_offset;
2270         u64 stripe_length;
2271         int factor;
2272         int i;
2273
2274         if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
2275                 return 0;
2276
2277         if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP |
2278              BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10))
2279                 factor = 2;
2280         else
2281                 factor = 1;
2282         factor = num_stripes / factor;
2283
2284         for (i = 0; i < num_stripes; i++) {
2285                 stripe = btrfs_stripe_nr(chunk, i);
2286                 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
2287                         continue;
2288
2289                 stripe_offset = btrfs_stripe_offset(leaf, stripe);
2290                 stripe_length = btrfs_chunk_length(leaf, chunk);
2291                 do_div(stripe_length, factor);
2292
2293                 if (stripe_offset < bargs->pend &&
2294                     stripe_offset + stripe_length > bargs->pstart)
2295                         return 0;
2296         }
2297
2298         return 1;
2299 }
2300
2301 /* [vstart, vend) */
2302 static int chunk_vrange_filter(struct extent_buffer *leaf,
2303                                struct btrfs_chunk *chunk,
2304                                u64 chunk_offset,
2305                                struct btrfs_balance_args *bargs)
2306 {
2307         if (chunk_offset < bargs->vend &&
2308             chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
2309                 /* at least part of the chunk is inside this vrange */
2310                 return 0;
2311
2312         return 1;
2313 }
2314
2315 static int chunk_soft_convert_filter(u64 chunk_profile,
2316                                      struct btrfs_balance_args *bargs)
2317 {
2318         if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
2319                 return 0;
2320
2321         chunk_profile &= BTRFS_BLOCK_GROUP_PROFILE_MASK;
2322
2323         if (chunk_profile == 0)
2324                 chunk_profile = BTRFS_AVAIL_ALLOC_BIT_SINGLE;
2325
2326         if (bargs->target & chunk_profile)
2327                 return 1;
2328
2329         return 0;
2330 }
2331
2332 static int should_balance_chunk(struct btrfs_root *root,
2333                                 struct extent_buffer *leaf,
2334                                 struct btrfs_chunk *chunk, u64 chunk_offset)
2335 {
2336         struct btrfs_balance_control *bctl = root->fs_info->balance_ctl;
2337         struct btrfs_balance_args *bargs = NULL;
2338         u64 chunk_type = btrfs_chunk_type(leaf, chunk);
2339
2340         /* type filter */
2341         if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
2342               (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
2343                 return 0;
2344         }
2345
2346         if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
2347                 bargs = &bctl->data;
2348         else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
2349                 bargs = &bctl->sys;
2350         else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
2351                 bargs = &bctl->meta;
2352
2353         /* profiles filter */
2354         if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
2355             chunk_profiles_filter(chunk_type, bargs)) {
2356                 return 0;
2357         }
2358
2359         /* usage filter */
2360         if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
2361             chunk_usage_filter(bctl->fs_info, chunk_offset, bargs)) {
2362                 return 0;
2363         }
2364
2365         /* devid filter */
2366         if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
2367             chunk_devid_filter(leaf, chunk, bargs)) {
2368                 return 0;
2369         }
2370
2371         /* drange filter, makes sense only with devid filter */
2372         if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
2373             chunk_drange_filter(leaf, chunk, chunk_offset, bargs)) {
2374                 return 0;
2375         }
2376
2377         /* vrange filter */
2378         if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
2379             chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
2380                 return 0;
2381         }
2382
2383         /* soft profile changing mode */
2384         if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
2385             chunk_soft_convert_filter(chunk_type, bargs)) {
2386                 return 0;
2387         }
2388
2389         return 1;
2390 }
2391
2392 static u64 div_factor(u64 num, int factor)
2393 {
2394         if (factor == 10)
2395                 return num;
2396         num *= factor;
2397         do_div(num, 10);
2398         return num;
2399 }
2400
2401 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
2402 {
2403         struct btrfs_root *chunk_root = fs_info->chunk_root;
2404         struct btrfs_root *dev_root = fs_info->dev_root;
2405         struct list_head *devices;
2406         struct btrfs_device *device;
2407         u64 old_size;
2408         u64 size_to_free;
2409         struct btrfs_chunk *chunk;
2410         struct btrfs_path *path;
2411         struct btrfs_key key;
2412         struct btrfs_key found_key;
2413         struct btrfs_trans_handle *trans;
2414         struct extent_buffer *leaf;
2415         int slot;
2416         int ret;
2417         int enospc_errors = 0;
2418
2419         /* step one make some room on all the devices */
2420         devices = &fs_info->fs_devices->devices;
2421         list_for_each_entry(device, devices, dev_list) {
2422                 old_size = device->total_bytes;
2423                 size_to_free = div_factor(old_size, 1);
2424                 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
2425                 if (!device->writeable ||
2426                     device->total_bytes - device->bytes_used > size_to_free)
2427                         continue;
2428
2429                 ret = btrfs_shrink_device(device, old_size - size_to_free);
2430                 if (ret == -ENOSPC)
2431                         break;
2432                 BUG_ON(ret);
2433
2434                 trans = btrfs_start_transaction(dev_root, 0);
2435                 BUG_ON(IS_ERR(trans));
2436
2437                 ret = btrfs_grow_device(trans, device, old_size);
2438                 BUG_ON(ret);
2439
2440                 btrfs_end_transaction(trans, dev_root);
2441         }
2442
2443         /* step two, relocate all the chunks */
2444         path = btrfs_alloc_path();
2445         if (!path) {
2446                 ret = -ENOMEM;
2447                 goto error;
2448         }
2449         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2450         key.offset = (u64)-1;
2451         key.type = BTRFS_CHUNK_ITEM_KEY;
2452
2453         while (1) {
2454                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2455                 if (ret < 0)
2456                         goto error;
2457
2458                 /*
2459                  * this shouldn't happen, it means the last relocate
2460                  * failed
2461                  */
2462                 if (ret == 0)
2463                         BUG(); /* FIXME break ? */
2464
2465                 ret = btrfs_previous_item(chunk_root, path, 0,
2466                                           BTRFS_CHUNK_ITEM_KEY);
2467                 if (ret) {
2468                         ret = 0;
2469                         break;
2470                 }
2471
2472                 leaf = path->nodes[0];
2473                 slot = path->slots[0];
2474                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2475
2476                 if (found_key.objectid != key.objectid)
2477                         break;
2478
2479                 /* chunk zero is special */
2480                 if (found_key.offset == 0)
2481                         break;
2482
2483                 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
2484
2485                 ret = should_balance_chunk(chunk_root, leaf, chunk,
2486                                            found_key.offset);
2487                 btrfs_release_path(path);
2488                 if (!ret)
2489                         goto loop;
2490
2491                 ret = btrfs_relocate_chunk(chunk_root,
2492                                            chunk_root->root_key.objectid,
2493                                            found_key.objectid,
2494                                            found_key.offset);
2495                 if (ret && ret != -ENOSPC)
2496                         goto error;
2497                 if (ret == -ENOSPC)
2498                         enospc_errors++;
2499 loop:
2500                 key.offset = found_key.offset - 1;
2501         }
2502
2503 error:
2504         btrfs_free_path(path);
2505         if (enospc_errors) {
2506                 printk(KERN_INFO "btrfs: %d enospc errors during balance\n",
2507                        enospc_errors);
2508                 if (!ret)
2509                         ret = -ENOSPC;
2510         }
2511
2512         return ret;
2513 }
2514
2515 static void __cancel_balance(struct btrfs_fs_info *fs_info)
2516 {
2517         int ret;
2518
2519         unset_balance_control(fs_info);
2520         ret = del_balance_item(fs_info->tree_root);
2521         BUG_ON(ret);
2522 }
2523
2524 void update_ioctl_balance_args(struct btrfs_fs_info *fs_info,
2525                                struct btrfs_ioctl_balance_args *bargs);
2526
2527 /*
2528  * Should be called with both balance and volume mutexes held
2529  */
2530 int btrfs_balance(struct btrfs_balance_control *bctl,
2531                   struct btrfs_ioctl_balance_args *bargs)
2532 {
2533         struct btrfs_fs_info *fs_info = bctl->fs_info;
2534         u64 allowed;
2535         int ret;
2536
2537         if (btrfs_fs_closing(fs_info)) {
2538                 ret = -EINVAL;
2539                 goto out;
2540         }
2541
2542         /*
2543          * In case of mixed groups both data and meta should be picked,
2544          * and identical options should be given for both of them.
2545          */
2546         allowed = btrfs_super_incompat_flags(fs_info->super_copy);
2547         if ((allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2548             (bctl->flags & (BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA))) {
2549                 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
2550                     !(bctl->flags & BTRFS_BALANCE_METADATA) ||
2551                     memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
2552                         printk(KERN_ERR "btrfs: with mixed groups data and "
2553                                "metadata balance options must be the same\n");
2554                         ret = -EINVAL;
2555                         goto out;
2556                 }
2557         }
2558
2559         /*
2560          * Profile changing sanity checks.  Skip them if a simple
2561          * balance is requested.
2562          */
2563         if (!((bctl->data.flags | bctl->sys.flags | bctl->meta.flags) &
2564               BTRFS_BALANCE_ARGS_CONVERT))
2565                 goto do_balance;
2566
2567         allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE;
2568         if (fs_info->fs_devices->num_devices == 1)
2569                 allowed |= BTRFS_BLOCK_GROUP_DUP;
2570         else if (fs_info->fs_devices->num_devices < 4)
2571                 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1);
2572         else
2573                 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
2574                                 BTRFS_BLOCK_GROUP_RAID10);
2575
2576         if (!profile_is_valid(bctl->data.target, 1) ||
2577             bctl->data.target & ~allowed) {
2578                 printk(KERN_ERR "btrfs: unable to start balance with target "
2579                        "data profile %llu\n",
2580                        (unsigned long long)bctl->data.target);
2581                 ret = -EINVAL;
2582                 goto out;
2583         }
2584         if (!profile_is_valid(bctl->meta.target, 1) ||
2585             bctl->meta.target & ~allowed) {
2586                 printk(KERN_ERR "btrfs: unable to start balance with target "
2587                        "metadata profile %llu\n",
2588                        (unsigned long long)bctl->meta.target);
2589                 ret = -EINVAL;
2590                 goto out;
2591         }
2592         if (!profile_is_valid(bctl->sys.target, 1) ||
2593             bctl->sys.target & ~allowed) {
2594                 printk(KERN_ERR "btrfs: unable to start balance with target "
2595                        "system profile %llu\n",
2596                        (unsigned long long)bctl->sys.target);
2597                 ret = -EINVAL;
2598                 goto out;
2599         }
2600
2601         if (bctl->data.target & BTRFS_BLOCK_GROUP_DUP) {
2602                 printk(KERN_ERR "btrfs: dup for data is not allowed\n");
2603                 ret = -EINVAL;
2604                 goto out;
2605         }
2606
2607         /* allow to reduce meta or sys integrity only if force set */
2608         allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
2609                         BTRFS_BLOCK_GROUP_RAID10;
2610         if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
2611              (fs_info->avail_system_alloc_bits & allowed) &&
2612              !(bctl->sys.target & allowed)) ||
2613             ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
2614              (fs_info->avail_metadata_alloc_bits & allowed) &&
2615              !(bctl->meta.target & allowed))) {
2616                 if (bctl->flags & BTRFS_BALANCE_FORCE) {
2617                         printk(KERN_INFO "btrfs: force reducing metadata "
2618                                "integrity\n");
2619                 } else {
2620                         printk(KERN_ERR "btrfs: balance will reduce metadata "
2621                                "integrity, use force if you want this\n");
2622                         ret = -EINVAL;
2623                         goto out;
2624                 }
2625         }
2626
2627 do_balance:
2628         ret = insert_balance_item(fs_info->tree_root, bctl);
2629         if (ret)
2630                 goto out;
2631
2632         set_balance_control(bctl);
2633
2634         mutex_unlock(&fs_info->balance_mutex);
2635
2636         ret = __btrfs_balance(fs_info);
2637
2638         mutex_lock(&fs_info->balance_mutex);
2639
2640         if (bargs) {
2641                 memset(bargs, 0, sizeof(*bargs));
2642                 update_ioctl_balance_args(fs_info, bargs);
2643         }
2644
2645         __cancel_balance(fs_info);
2646
2647         return ret;
2648 out:
2649         kfree(bctl);
2650         return ret;
2651 }
2652
2653 /*
2654  * shrinking a device means finding all of the device extents past
2655  * the new size, and then following the back refs to the chunks.
2656  * The chunk relocation code actually frees the device extent
2657  */
2658 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
2659 {
2660         struct btrfs_trans_handle *trans;
2661         struct btrfs_root *root = device->dev_root;
2662         struct btrfs_dev_extent *dev_extent = NULL;
2663         struct btrfs_path *path;
2664         u64 length;
2665         u64 chunk_tree;
2666         u64 chunk_objectid;
2667         u64 chunk_offset;
2668         int ret;
2669         int slot;
2670         int failed = 0;
2671         bool retried = false;
2672         struct extent_buffer *l;
2673         struct btrfs_key key;
2674         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
2675         u64 old_total = btrfs_super_total_bytes(super_copy);
2676         u64 old_size = device->total_bytes;
2677         u64 diff = device->total_bytes - new_size;
2678
2679         if (new_size >= device->total_bytes)
2680                 return -EINVAL;
2681
2682         path = btrfs_alloc_path();
2683         if (!path)
2684                 return -ENOMEM;
2685
2686         path->reada = 2;
2687
2688         lock_chunks(root);
2689
2690         device->total_bytes = new_size;
2691         if (device->writeable) {
2692                 device->fs_devices->total_rw_bytes -= diff;
2693                 spin_lock(&root->fs_info->free_chunk_lock);
2694                 root->fs_info->free_chunk_space -= diff;
2695                 spin_unlock(&root->fs_info->free_chunk_lock);
2696         }
2697         unlock_chunks(root);
2698
2699 again:
2700         key.objectid = device->devid;
2701         key.offset = (u64)-1;
2702         key.type = BTRFS_DEV_EXTENT_KEY;
2703
2704         while (1) {
2705                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2706                 if (ret < 0)
2707                         goto done;
2708
2709                 ret = btrfs_previous_item(root, path, 0, key.type);
2710                 if (ret < 0)
2711                         goto done;
2712                 if (ret) {
2713                         ret = 0;
2714                         btrfs_release_path(path);
2715                         break;
2716                 }
2717
2718                 l = path->nodes[0];
2719                 slot = path->slots[0];
2720                 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
2721
2722                 if (key.objectid != device->devid) {
2723                         btrfs_release_path(path);
2724                         break;
2725                 }
2726
2727                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
2728                 length = btrfs_dev_extent_length(l, dev_extent);
2729
2730                 if (key.offset + length <= new_size) {
2731                         btrfs_release_path(path);
2732                         break;
2733                 }
2734
2735                 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
2736                 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
2737                 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
2738                 btrfs_release_path(path);
2739
2740                 ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
2741                                            chunk_offset);
2742                 if (ret && ret != -ENOSPC)
2743                         goto done;
2744                 if (ret == -ENOSPC)
2745                         failed++;
2746                 key.offset -= 1;
2747         }
2748
2749         if (failed && !retried) {
2750                 failed = 0;
2751                 retried = true;
2752                 goto again;
2753         } else if (failed && retried) {
2754                 ret = -ENOSPC;
2755                 lock_chunks(root);
2756
2757                 device->total_bytes = old_size;
2758                 if (device->writeable)
2759                         device->fs_devices->total_rw_bytes += diff;
2760                 spin_lock(&root->fs_info->free_chunk_lock);
2761                 root->fs_info->free_chunk_space += diff;
2762                 spin_unlock(&root->fs_info->free_chunk_lock);
2763                 unlock_chunks(root);
2764                 goto done;
2765         }
2766
2767         /* Shrinking succeeded, else we would be at "done". */
2768         trans = btrfs_start_transaction(root, 0);
2769         if (IS_ERR(trans)) {
2770                 ret = PTR_ERR(trans);
2771                 goto done;
2772         }
2773
2774         lock_chunks(root);
2775
2776         device->disk_total_bytes = new_size;
2777         /* Now btrfs_update_device() will change the on-disk size. */
2778         ret = btrfs_update_device(trans, device);
2779         if (ret) {
2780                 unlock_chunks(root);
2781                 btrfs_end_transaction(trans, root);
2782                 goto done;
2783         }
2784         WARN_ON(diff > old_total);
2785         btrfs_set_super_total_bytes(super_copy, old_total - diff);
2786         unlock_chunks(root);
2787         btrfs_end_transaction(trans, root);
2788 done:
2789         btrfs_free_path(path);
2790         return ret;
2791 }
2792
2793 static int btrfs_add_system_chunk(struct btrfs_trans_handle *trans,
2794                            struct btrfs_root *root,
2795                            struct btrfs_key *key,
2796                            struct btrfs_chunk *chunk, int item_size)
2797 {
2798         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
2799         struct btrfs_disk_key disk_key;
2800         u32 array_size;
2801         u8 *ptr;
2802
2803         array_size = btrfs_super_sys_array_size(super_copy);
2804         if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
2805                 return -EFBIG;
2806
2807         ptr = super_copy->sys_chunk_array + array_size;
2808         btrfs_cpu_key_to_disk(&disk_key, key);
2809         memcpy(ptr, &disk_key, sizeof(disk_key));
2810         ptr += sizeof(disk_key);
2811         memcpy(ptr, chunk, item_size);
2812         item_size += sizeof(disk_key);
2813         btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
2814         return 0;
2815 }
2816
2817 /*
2818  * sort the devices in descending order by max_avail, total_avail
2819  */
2820 static int btrfs_cmp_device_info(const void *a, const void *b)
2821 {
2822         const struct btrfs_device_info *di_a = a;
2823         const struct btrfs_device_info *di_b = b;
2824
2825         if (di_a->max_avail > di_b->max_avail)
2826                 return -1;
2827         if (di_a->max_avail < di_b->max_avail)
2828                 return 1;
2829         if (di_a->total_avail > di_b->total_avail)
2830                 return -1;
2831         if (di_a->total_avail < di_b->total_avail)
2832                 return 1;
2833         return 0;
2834 }
2835
2836 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
2837                                struct btrfs_root *extent_root,
2838                                struct map_lookup **map_ret,
2839                                u64 *num_bytes_out, u64 *stripe_size_out,
2840                                u64 start, u64 type)
2841 {
2842         struct btrfs_fs_info *info = extent_root->fs_info;
2843         struct btrfs_fs_devices *fs_devices = info->fs_devices;
2844         struct list_head *cur;
2845         struct map_lookup *map = NULL;
2846         struct extent_map_tree *em_tree;
2847         struct extent_map *em;
2848         struct btrfs_device_info *devices_info = NULL;
2849         u64 total_avail;
2850         int num_stripes;        /* total number of stripes to allocate */
2851         int sub_stripes;        /* sub_stripes info for map */
2852         int dev_stripes;        /* stripes per dev */
2853         int devs_max;           /* max devs to use */
2854         int devs_min;           /* min devs needed */
2855         int devs_increment;     /* ndevs has to be a multiple of this */
2856         int ncopies;            /* how many copies to data has */
2857         int ret;
2858         u64 max_stripe_size;
2859         u64 max_chunk_size;
2860         u64 stripe_size;
2861         u64 num_bytes;
2862         int ndevs;
2863         int i;
2864         int j;
2865
2866         if ((type & BTRFS_BLOCK_GROUP_RAID1) &&
2867             (type & BTRFS_BLOCK_GROUP_DUP)) {
2868                 WARN_ON(1);
2869                 type &= ~BTRFS_BLOCK_GROUP_DUP;
2870         }
2871
2872         if (list_empty(&fs_devices->alloc_list))
2873                 return -ENOSPC;
2874
2875         sub_stripes = 1;
2876         dev_stripes = 1;
2877         devs_increment = 1;
2878         ncopies = 1;
2879         devs_max = 0;   /* 0 == as many as possible */
2880         devs_min = 1;
2881
2882         /*
2883          * define the properties of each RAID type.
2884          * FIXME: move this to a global table and use it in all RAID
2885          * calculation code
2886          */
2887         if (type & (BTRFS_BLOCK_GROUP_DUP)) {
2888                 dev_stripes = 2;
2889                 ncopies = 2;
2890                 devs_max = 1;
2891         } else if (type & (BTRFS_BLOCK_GROUP_RAID0)) {
2892                 devs_min = 2;
2893         } else if (type & (BTRFS_BLOCK_GROUP_RAID1)) {
2894                 devs_increment = 2;
2895                 ncopies = 2;
2896                 devs_max = 2;
2897                 devs_min = 2;
2898         } else if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
2899                 sub_stripes = 2;
2900                 devs_increment = 2;
2901                 ncopies = 2;
2902                 devs_min = 4;
2903         } else {
2904                 devs_max = 1;
2905         }
2906
2907         if (type & BTRFS_BLOCK_GROUP_DATA) {
2908                 max_stripe_size = 1024 * 1024 * 1024;
2909                 max_chunk_size = 10 * max_stripe_size;
2910         } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
2911                 max_stripe_size = 256 * 1024 * 1024;
2912                 max_chunk_size = max_stripe_size;
2913         } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
2914                 max_stripe_size = 8 * 1024 * 1024;
2915                 max_chunk_size = 2 * max_stripe_size;
2916         } else {
2917                 printk(KERN_ERR "btrfs: invalid chunk type 0x%llx requested\n",
2918                        type);
2919                 BUG_ON(1);
2920         }
2921
2922         /* we don't want a chunk larger than 10% of writeable space */
2923         max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
2924                              max_chunk_size);
2925
2926         devices_info = kzalloc(sizeof(*devices_info) * fs_devices->rw_devices,
2927                                GFP_NOFS);
2928         if (!devices_info)
2929                 return -ENOMEM;
2930
2931         cur = fs_devices->alloc_list.next;
2932
2933         /*
2934          * in the first pass through the devices list, we gather information
2935          * about the available holes on each device.
2936          */
2937         ndevs = 0;
2938         while (cur != &fs_devices->alloc_list) {
2939                 struct btrfs_device *device;
2940                 u64 max_avail;
2941                 u64 dev_offset;
2942
2943                 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
2944
2945                 cur = cur->next;
2946
2947                 if (!device->writeable) {
2948                         printk(KERN_ERR
2949                                "btrfs: read-only device in alloc_list\n");
2950                         WARN_ON(1);
2951                         continue;
2952                 }
2953
2954                 if (!device->in_fs_metadata)
2955                         continue;
2956
2957                 if (device->total_bytes > device->bytes_used)
2958                         total_avail = device->total_bytes - device->bytes_used;
2959                 else
2960                         total_avail = 0;
2961
2962                 /* If there is no space on this device, skip it. */
2963                 if (total_avail == 0)
2964                         continue;
2965
2966                 ret = find_free_dev_extent(trans, device,
2967                                            max_stripe_size * dev_stripes,
2968                                            &dev_offset, &max_avail);
2969                 if (ret && ret != -ENOSPC)
2970                         goto error;
2971
2972                 if (ret == 0)
2973                         max_avail = max_stripe_size * dev_stripes;
2974
2975                 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes)
2976                         continue;
2977
2978                 devices_info[ndevs].dev_offset = dev_offset;
2979                 devices_info[ndevs].max_avail = max_avail;
2980                 devices_info[ndevs].total_avail = total_avail;
2981                 devices_info[ndevs].dev = device;
2982                 ++ndevs;
2983         }
2984
2985         /*
2986          * now sort the devices by hole size / available space
2987          */
2988         sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
2989              btrfs_cmp_device_info, NULL);
2990
2991         /* round down to number of usable stripes */
2992         ndevs -= ndevs % devs_increment;
2993
2994         if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) {
2995                 ret = -ENOSPC;
2996                 goto error;
2997         }
2998
2999         if (devs_max && ndevs > devs_max)
3000                 ndevs = devs_max;
3001         /*
3002          * the primary goal is to maximize the number of stripes, so use as many
3003          * devices as possible, even if the stripes are not maximum sized.
3004          */
3005         stripe_size = devices_info[ndevs-1].max_avail;
3006         num_stripes = ndevs * dev_stripes;
3007
3008         if (stripe_size * num_stripes > max_chunk_size * ncopies) {
3009                 stripe_size = max_chunk_size * ncopies;
3010                 do_div(stripe_size, num_stripes);
3011         }
3012
3013         do_div(stripe_size, dev_stripes);
3014         do_div(stripe_size, BTRFS_STRIPE_LEN);
3015         stripe_size *= BTRFS_STRIPE_LEN;
3016
3017         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
3018         if (!map) {
3019                 ret = -ENOMEM;
3020                 goto error;
3021         }
3022         map->num_stripes = num_stripes;
3023
3024         for (i = 0; i < ndevs; ++i) {
3025                 for (j = 0; j < dev_stripes; ++j) {
3026                         int s = i * dev_stripes + j;
3027                         map->stripes[s].dev = devices_info[i].dev;
3028                         map->stripes[s].physical = devices_info[i].dev_offset +
3029                                                    j * stripe_size;
3030                 }
3031         }
3032         map->sector_size = extent_root->sectorsize;
3033         map->stripe_len = BTRFS_STRIPE_LEN;
3034         map->io_align = BTRFS_STRIPE_LEN;
3035         map->io_width = BTRFS_STRIPE_LEN;
3036         map->type = type;
3037         map->sub_stripes = sub_stripes;
3038
3039         *map_ret = map;
3040         num_bytes = stripe_size * (num_stripes / ncopies);
3041
3042         *stripe_size_out = stripe_size;
3043         *num_bytes_out = num_bytes;
3044
3045         trace_btrfs_chunk_alloc(info->chunk_root, map, start, num_bytes);
3046
3047         em = alloc_extent_map();
3048         if (!em) {
3049                 ret = -ENOMEM;
3050                 goto error;
3051         }
3052         em->bdev = (struct block_device *)map;
3053         em->start = start;
3054         em->len = num_bytes;
3055         em->block_start = 0;
3056         em->block_len = em->len;
3057
3058         em_tree = &extent_root->fs_info->mapping_tree.map_tree;
3059         write_lock(&em_tree->lock);
3060         ret = add_extent_mapping(em_tree, em);
3061         write_unlock(&em_tree->lock);
3062         BUG_ON(ret);
3063         free_extent_map(em);
3064
3065         ret = btrfs_make_block_group(trans, extent_root, 0, type,
3066                                      BTRFS_FIRST_CHUNK_TREE_OBJECTID,
3067                                      start, num_bytes);
3068         BUG_ON(ret);
3069
3070         for (i = 0; i < map->num_stripes; ++i) {
3071                 struct btrfs_device *device;
3072                 u64 dev_offset;
3073
3074                 device = map->stripes[i].dev;
3075                 dev_offset = map->stripes[i].physical;
3076
3077                 ret = btrfs_alloc_dev_extent(trans, device,
3078                                 info->chunk_root->root_key.objectid,
3079                                 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
3080                                 start, dev_offset, stripe_size);
3081                 BUG_ON(ret);
3082         }
3083
3084         kfree(devices_info);
3085         return 0;
3086
3087 error:
3088         kfree(map);
3089         kfree(devices_info);
3090         return ret;
3091 }
3092
3093 static int __finish_chunk_alloc(struct btrfs_trans_handle *trans,
3094                                 struct btrfs_root *extent_root,
3095                                 struct map_lookup *map, u64 chunk_offset,
3096                                 u64 chunk_size, u64 stripe_size)
3097 {
3098         u64 dev_offset;
3099         struct btrfs_key key;
3100         struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
3101         struct btrfs_device *device;
3102         struct btrfs_chunk *chunk;
3103         struct btrfs_stripe *stripe;
3104         size_t item_size = btrfs_chunk_item_size(map->num_stripes);
3105         int index = 0;
3106         int ret;
3107
3108         chunk = kzalloc(item_size, GFP_NOFS);
3109         if (!chunk)
3110                 return -ENOMEM;
3111
3112         index = 0;
3113         while (index < map->num_stripes) {
3114                 device = map->stripes[index].dev;
3115                 device->bytes_used += stripe_size;
3116                 ret = btrfs_update_device(trans, device);
3117                 BUG_ON(ret);
3118                 index++;
3119         }
3120
3121         spin_lock(&extent_root->fs_info->free_chunk_lock);
3122         extent_root->fs_info->free_chunk_space -= (stripe_size *
3123                                                    map->num_stripes);
3124         spin_unlock(&extent_root->fs_info->free_chunk_lock);
3125
3126         index = 0;
3127         stripe = &chunk->stripe;
3128         while (index < map->num_stripes) {
3129                 device = map->stripes[index].dev;
3130                 dev_offset = map->stripes[index].physical;
3131
3132                 btrfs_set_stack_stripe_devid(stripe, device->devid);
3133                 btrfs_set_stack_stripe_offset(stripe, dev_offset);
3134                 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
3135                 stripe++;
3136                 index++;
3137         }
3138
3139         btrfs_set_stack_chunk_length(chunk, chunk_size);
3140         btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
3141         btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
3142         btrfs_set_stack_chunk_type(chunk, map->type);
3143         btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
3144         btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
3145         btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
3146         btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
3147         btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
3148
3149         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3150         key.type = BTRFS_CHUNK_ITEM_KEY;
3151         key.offset = chunk_offset;
3152
3153         ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
3154         BUG_ON(ret);
3155
3156         if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
3157                 ret = btrfs_add_system_chunk(trans, chunk_root, &key, chunk,
3158                                              item_size);
3159                 BUG_ON(ret);
3160         }
3161
3162         kfree(chunk);
3163         return 0;
3164 }
3165
3166 /*
3167  * Chunk allocation falls into two parts. The first part does works
3168  * that make the new allocated chunk useable, but not do any operation
3169  * that modifies the chunk tree. The second part does the works that
3170  * require modifying the chunk tree. This division is important for the
3171  * bootstrap process of adding storage to a seed btrfs.
3172  */
3173 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
3174                       struct btrfs_root *extent_root, u64 type)
3175 {
3176         u64 chunk_offset;
3177         u64 chunk_size;
3178         u64 stripe_size;
3179         struct map_lookup *map;
3180         struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
3181         int ret;
3182
3183         ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
3184                               &chunk_offset);
3185         if (ret)
3186                 return ret;
3187
3188         ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
3189                                   &stripe_size, chunk_offset, type);
3190         if (ret)
3191                 return ret;
3192
3193         ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
3194                                    chunk_size, stripe_size);
3195         BUG_ON(ret);
3196         return 0;
3197 }
3198
3199 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
3200                                          struct btrfs_root *root,
3201                                          struct btrfs_device *device)
3202 {
3203         u64 chunk_offset;
3204         u64 sys_chunk_offset;
3205         u64 chunk_size;
3206         u64 sys_chunk_size;
3207         u64 stripe_size;
3208         u64 sys_stripe_size;
3209         u64 alloc_profile;
3210         struct map_lookup *map;
3211         struct map_lookup *sys_map;
3212         struct btrfs_fs_info *fs_info = root->fs_info;
3213         struct btrfs_root *extent_root = fs_info->extent_root;
3214         int ret;
3215
3216         ret = find_next_chunk(fs_info->chunk_root,
3217                               BTRFS_FIRST_CHUNK_TREE_OBJECTID, &chunk_offset);
3218         if (ret)
3219                 return ret;
3220
3221         alloc_profile = BTRFS_BLOCK_GROUP_METADATA |
3222                                 fs_info->avail_metadata_alloc_bits;
3223         alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
3224
3225         ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
3226                                   &stripe_size, chunk_offset, alloc_profile);
3227         BUG_ON(ret);
3228
3229         sys_chunk_offset = chunk_offset + chunk_size;
3230
3231         alloc_profile = BTRFS_BLOCK_GROUP_SYSTEM |
3232                                 fs_info->avail_system_alloc_bits;
3233         alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
3234
3235         ret = __btrfs_alloc_chunk(trans, extent_root, &sys_map,
3236                                   &sys_chunk_size, &sys_stripe_size,
3237                                   sys_chunk_offset, alloc_profile);
3238         BUG_ON(ret);
3239
3240         ret = btrfs_add_device(trans, fs_info->chunk_root, device);
3241         BUG_ON(ret);
3242
3243         /*
3244          * Modifying chunk tree needs allocating new blocks from both
3245          * system block group and metadata block group. So we only can
3246          * do operations require modifying the chunk tree after both
3247          * block groups were created.
3248          */
3249         ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
3250                                    chunk_size, stripe_size);
3251         BUG_ON(ret);
3252
3253         ret = __finish_chunk_alloc(trans, extent_root, sys_map,
3254                                    sys_chunk_offset, sys_chunk_size,
3255                                    sys_stripe_size);
3256         BUG_ON(ret);
3257         return 0;
3258 }
3259
3260 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
3261 {
3262         struct extent_map *em;
3263         struct map_lookup *map;
3264         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
3265         int readonly = 0;
3266         int i;
3267
3268         read_lock(&map_tree->map_tree.lock);
3269         em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
3270         read_unlock(&map_tree->map_tree.lock);
3271         if (!em)
3272                 return 1;
3273
3274         if (btrfs_test_opt(root, DEGRADED)) {
3275                 free_extent_map(em);
3276                 return 0;
3277         }
3278
3279         map = (struct map_lookup *)em->bdev;
3280         for (i = 0; i < map->num_stripes; i++) {
3281                 if (!map->stripes[i].dev->writeable) {
3282                         readonly = 1;
3283                         break;
3284                 }
3285         }
3286         free_extent_map(em);
3287         return readonly;
3288 }
3289
3290 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
3291 {
3292         extent_map_tree_init(&tree->map_tree);
3293 }
3294
3295 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
3296 {
3297         struct extent_map *em;
3298
3299         while (1) {
3300                 write_lock(&tree->map_tree.lock);
3301                 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
3302                 if (em)
3303                         remove_extent_mapping(&tree->map_tree, em);
3304                 write_unlock(&tree->map_tree.lock);
3305                 if (!em)
3306                         break;
3307                 kfree(em->bdev);
3308                 /* once for us */
3309                 free_extent_map(em);
3310                 /* once for the tree */
3311                 free_extent_map(em);
3312         }
3313 }
3314
3315 int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len)
3316 {
3317         struct extent_map *em;
3318         struct map_lookup *map;
3319         struct extent_map_tree *em_tree = &map_tree->map_tree;
3320         int ret;
3321
3322         read_lock(&em_tree->lock);
3323         em = lookup_extent_mapping(em_tree, logical, len);
3324         read_unlock(&em_tree->lock);
3325         BUG_ON(!em);
3326
3327         BUG_ON(em->start > logical || em->start + em->len < logical);
3328         map = (struct map_lookup *)em->bdev;
3329         if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
3330                 ret = map->num_stripes;
3331         else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
3332                 ret = map->sub_stripes;
3333         else
3334                 ret = 1;
3335         free_extent_map(em);
3336         return ret;
3337 }
3338
3339 static int find_live_mirror(struct map_lookup *map, int first, int num,
3340                             int optimal)
3341 {
3342         int i;
3343         if (map->stripes[optimal].dev->bdev)
3344                 return optimal;
3345         for (i = first; i < first + num; i++) {
3346                 if (map->stripes[i].dev->bdev)
3347                         return i;
3348         }
3349         /* we couldn't find one that doesn't fail.  Just return something
3350          * and the io error handling code will clean up eventually
3351          */
3352         return optimal;
3353 }
3354
3355 static int __btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
3356                              u64 logical, u64 *length,
3357                              struct btrfs_bio **bbio_ret,
3358                              int mirror_num)
3359 {
3360         struct extent_map *em;
3361         struct map_lookup *map;
3362         struct extent_map_tree *em_tree = &map_tree->map_tree;
3363         u64 offset;
3364         u64 stripe_offset;
3365         u64 stripe_end_offset;
3366         u64 stripe_nr;
3367         u64 stripe_nr_orig;
3368         u64 stripe_nr_end;
3369         int stripes_allocated = 8;
3370         int stripes_required = 1;
3371         int stripe_index;
3372         int i;
3373         int num_stripes;
3374         int max_errors = 0;
3375         struct btrfs_bio *bbio = NULL;
3376
3377         if (bbio_ret && !(rw & (REQ_WRITE | REQ_DISCARD)))
3378                 stripes_allocated = 1;
3379 again:
3380         if (bbio_ret) {
3381                 bbio = kzalloc(btrfs_bio_size(stripes_allocated),
3382                                 GFP_NOFS);
3383                 if (!bbio)
3384                         return -ENOMEM;
3385
3386                 atomic_set(&bbio->error, 0);
3387         }
3388
3389         read_lock(&em_tree->lock);
3390         em = lookup_extent_mapping(em_tree, logical, *length);
3391         read_unlock(&em_tree->lock);
3392
3393         if (!em) {
3394                 printk(KERN_CRIT "unable to find logical %llu len %llu\n",
3395                        (unsigned long long)logical,
3396                        (unsigned long long)*length);
3397                 BUG();
3398         }
3399
3400         BUG_ON(em->start > logical || em->start + em->len < logical);
3401         map = (struct map_lookup *)em->bdev;
3402         offset = logical - em->start;
3403
3404         if (mirror_num > map->num_stripes)
3405                 mirror_num = 0;
3406
3407         /* if our btrfs_bio struct is too small, back off and try again */
3408         if (rw & REQ_WRITE) {
3409                 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
3410                                  BTRFS_BLOCK_GROUP_DUP)) {
3411                         stripes_required = map->num_stripes;
3412                         max_errors = 1;
3413                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
3414                         stripes_required = map->sub_stripes;
3415                         max_errors = 1;
3416                 }
3417         }
3418         if (rw & REQ_DISCARD) {
3419                 if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK)
3420                         stripes_required = map->num_stripes;
3421         }
3422         if (bbio_ret && (rw & (REQ_WRITE | REQ_DISCARD)) &&
3423             stripes_allocated < stripes_required) {
3424                 stripes_allocated = map->num_stripes;
3425                 free_extent_map(em);
3426                 kfree(bbio);
3427                 goto again;
3428         }
3429         stripe_nr = offset;
3430         /*
3431          * stripe_nr counts the total number of stripes we have to stride
3432          * to get to this block
3433          */
3434         do_div(stripe_nr, map->stripe_len);
3435
3436         stripe_offset = stripe_nr * map->stripe_len;
3437         BUG_ON(offset < stripe_offset);
3438
3439         /* stripe_offset is the offset of this block in its stripe*/
3440         stripe_offset = offset - stripe_offset;
3441
3442         if (rw & REQ_DISCARD)
3443                 *length = min_t(u64, em->len - offset, *length);
3444         else if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
3445                 /* we limit the length of each bio to what fits in a stripe */
3446                 *length = min_t(u64, em->len - offset,
3447                                 map->stripe_len - stripe_offset);
3448         } else {
3449                 *length = em->len - offset;
3450         }
3451
3452         if (!bbio_ret)
3453                 goto out;
3454
3455         num_stripes = 1;
3456         stripe_index = 0;
3457         stripe_nr_orig = stripe_nr;
3458         stripe_nr_end = (offset + *length + map->stripe_len - 1) &
3459                         (~(map->stripe_len - 1));
3460         do_div(stripe_nr_end, map->stripe_len);
3461         stripe_end_offset = stripe_nr_end * map->stripe_len -
3462                             (offset + *length);
3463         if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
3464                 if (rw & REQ_DISCARD)
3465                         num_stripes = min_t(u64, map->num_stripes,
3466                                             stripe_nr_end - stripe_nr_orig);
3467                 stripe_index = do_div(stripe_nr, map->num_stripes);
3468         } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
3469                 if (rw & (REQ_WRITE | REQ_DISCARD))
3470                         num_stripes = map->num_stripes;
3471                 else if (mirror_num)
3472                         stripe_index = mirror_num - 1;
3473                 else {
3474                         stripe_index = find_live_mirror(map, 0,
3475                                             map->num_stripes,
3476                                             current->pid % map->num_stripes);
3477                         mirror_num = stripe_index + 1;
3478                 }
3479
3480         } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
3481                 if (rw & (REQ_WRITE | REQ_DISCARD)) {
3482                         num_stripes = map->num_stripes;
3483                 } else if (mirror_num) {
3484                         stripe_index = mirror_num - 1;
3485                 } else {
3486                         mirror_num = 1;
3487                 }
3488
3489         } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
3490                 int factor = map->num_stripes / map->sub_stripes;
3491
3492                 stripe_index = do_div(stripe_nr, factor);
3493                 stripe_index *= map->sub_stripes;
3494
3495                 if (rw & REQ_WRITE)
3496                         num_stripes = map->sub_stripes;
3497                 else if (rw & REQ_DISCARD)
3498                         num_stripes = min_t(u64, map->sub_stripes *
3499                                             (stripe_nr_end - stripe_nr_orig),
3500                                             map->num_stripes);
3501                 else if (mirror_num)
3502                         stripe_index += mirror_num - 1;
3503                 else {
3504                         stripe_index = find_live_mirror(map, stripe_index,
3505                                               map->sub_stripes, stripe_index +
3506                                               current->pid % map->sub_stripes);
3507                         mirror_num = stripe_index + 1;
3508                 }
3509         } else {
3510                 /*
3511                  * after this do_div call, stripe_nr is the number of stripes
3512                  * on this device we have to walk to find the data, and
3513                  * stripe_index is the number of our device in the stripe array
3514                  */
3515                 stripe_index = do_div(stripe_nr, map->num_stripes);
3516                 mirror_num = stripe_index + 1;
3517         }
3518         BUG_ON(stripe_index >= map->num_stripes);
3519
3520         if (rw & REQ_DISCARD) {
3521                 for (i = 0; i < num_stripes; i++) {
3522                         bbio->stripes[i].physical =
3523                                 map->stripes[stripe_index].physical +
3524                                 stripe_offset + stripe_nr * map->stripe_len;
3525                         bbio->stripes[i].dev = map->stripes[stripe_index].dev;
3526
3527                         if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
3528                                 u64 stripes;
3529                                 u32 last_stripe = 0;
3530                                 int j;
3531
3532                                 div_u64_rem(stripe_nr_end - 1,
3533                                             map->num_stripes,
3534                                             &last_stripe);
3535
3536                                 for (j = 0; j < map->num_stripes; j++) {
3537                                         u32 test;
3538
3539                                         div_u64_rem(stripe_nr_end - 1 - j,
3540                                                     map->num_stripes, &test);
3541                                         if (test == stripe_index)
3542                                                 break;
3543                                 }
3544                                 stripes = stripe_nr_end - 1 - j;
3545                                 do_div(stripes, map->num_stripes);
3546                                 bbio->stripes[i].length = map->stripe_len *
3547                                         (stripes - stripe_nr + 1);
3548
3549                                 if (i == 0) {
3550                                         bbio->stripes[i].length -=
3551                                                 stripe_offset;
3552                                         stripe_offset = 0;
3553                                 }
3554                                 if (stripe_index == last_stripe)
3555                                         bbio->stripes[i].length -=
3556                                                 stripe_end_offset;
3557                         } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
3558                                 u64 stripes;
3559                                 int j;
3560                                 int factor = map->num_stripes /
3561                                              map->sub_stripes;
3562                                 u32 last_stripe = 0;
3563
3564                                 div_u64_rem(stripe_nr_end - 1,
3565                                             factor, &last_stripe);
3566                                 last_stripe *= map->sub_stripes;
3567
3568                                 for (j = 0; j < factor; j++) {
3569                                         u32 test;
3570
3571                                         div_u64_rem(stripe_nr_end - 1 - j,
3572                                                     factor, &test);
3573
3574                                         if (test ==
3575                                             stripe_index / map->sub_stripes)
3576                                                 break;
3577                                 }
3578                                 stripes = stripe_nr_end - 1 - j;
3579                                 do_div(stripes, factor);
3580                                 bbio->stripes[i].length = map->stripe_len *
3581                                         (stripes - stripe_nr + 1);
3582
3583                                 if (i < map->sub_stripes) {
3584                                         bbio->stripes[i].length -=
3585                                                 stripe_offset;
3586                                         if (i == map->sub_stripes - 1)
3587                                                 stripe_offset = 0;
3588                                 }
3589                                 if (stripe_index >= last_stripe &&
3590                                     stripe_index <= (last_stripe +
3591                                                      map->sub_stripes - 1)) {
3592                                         bbio->stripes[i].length -=
3593                                                 stripe_end_offset;
3594                                 }
3595                         } else
3596                                 bbio->stripes[i].length = *length;
3597
3598                         stripe_index++;
3599                         if (stripe_index == map->num_stripes) {
3600                                 /* This could only happen for RAID0/10 */
3601                                 stripe_index = 0;
3602                                 stripe_nr++;
3603                         }
3604                 }
3605         } else {
3606                 for (i = 0; i < num_stripes; i++) {
3607                         bbio->stripes[i].physical =
3608                                 map->stripes[stripe_index].physical +
3609                                 stripe_offset +
3610                                 stripe_nr * map->stripe_len;
3611                         bbio->stripes[i].dev =
3612                                 map->stripes[stripe_index].dev;
3613                         stripe_index++;
3614                 }
3615         }
3616         if (bbio_ret) {
3617                 *bbio_ret = bbio;
3618                 bbio->num_stripes = num_stripes;
3619                 bbio->max_errors = max_errors;
3620                 bbio->mirror_num = mirror_num;
3621         }
3622 out:
3623         free_extent_map(em);
3624         return 0;
3625 }
3626
3627 int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
3628                       u64 logical, u64 *length,
3629                       struct btrfs_bio **bbio_ret, int mirror_num)
3630 {
3631         return __btrfs_map_block(map_tree, rw, logical, length, bbio_ret,
3632                                  mirror_num);
3633 }
3634
3635 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
3636                      u64 chunk_start, u64 physical, u64 devid,
3637                      u64 **logical, int *naddrs, int *stripe_len)
3638 {
3639         struct extent_map_tree *em_tree = &map_tree->map_tree;
3640         struct extent_map *em;
3641         struct map_lookup *map;
3642         u64 *buf;
3643         u64 bytenr;
3644         u64 length;
3645         u64 stripe_nr;
3646         int i, j, nr = 0;
3647
3648         read_lock(&em_tree->lock);
3649         em = lookup_extent_mapping(em_tree, chunk_start, 1);
3650         read_unlock(&em_tree->lock);
3651
3652         BUG_ON(!em || em->start != chunk_start);
3653         map = (struct map_lookup *)em->bdev;
3654
3655         length = em->len;
3656         if (map->type & BTRFS_BLOCK_GROUP_RAID10)
3657                 do_div(length, map->num_stripes / map->sub_stripes);
3658         else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
3659                 do_div(length, map->num_stripes);
3660
3661         buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
3662         BUG_ON(!buf);
3663
3664         for (i = 0; i < map->num_stripes; i++) {
3665                 if (devid && map->stripes[i].dev->devid != devid)
3666                         continue;
3667                 if (map->stripes[i].physical > physical ||
3668                     map->stripes[i].physical + length <= physical)
3669                         continue;
3670
3671                 stripe_nr = physical - map->stripes[i].physical;
3672                 do_div(stripe_nr, map->stripe_len);
3673
3674                 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
3675                         stripe_nr = stripe_nr * map->num_stripes + i;
3676                         do_div(stripe_nr, map->sub_stripes);
3677                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
3678                         stripe_nr = stripe_nr * map->num_stripes + i;
3679                 }
3680                 bytenr = chunk_start + stripe_nr * map->stripe_len;
3681                 WARN_ON(nr >= map->num_stripes);
3682                 for (j = 0; j < nr; j++) {
3683                         if (buf[j] == bytenr)
3684                                 break;
3685                 }
3686                 if (j == nr) {
3687                         WARN_ON(nr >= map->num_stripes);
3688                         buf[nr++] = bytenr;
3689                 }
3690         }
3691
3692         *logical = buf;
3693         *naddrs = nr;
3694         *stripe_len = map->stripe_len;
3695
3696         free_extent_map(em);
3697         return 0;
3698 }
3699
3700 static void btrfs_end_bio(struct bio *bio, int err)
3701 {
3702         struct btrfs_bio *bbio = bio->bi_private;
3703         int is_orig_bio = 0;
3704
3705         if (err)
3706                 atomic_inc(&bbio->error);
3707
3708         if (bio == bbio->orig_bio)
3709                 is_orig_bio = 1;
3710
3711         if (atomic_dec_and_test(&bbio->stripes_pending)) {
3712                 if (!is_orig_bio) {
3713                         bio_put(bio);
3714                         bio = bbio->orig_bio;
3715                 }
3716                 bio->bi_private = bbio->private;
3717                 bio->bi_end_io = bbio->end_io;
3718                 bio->bi_bdev = (struct block_device *)
3719                                         (unsigned long)bbio->mirror_num;
3720                 /* only send an error to the higher layers if it is
3721                  * beyond the tolerance of the multi-bio
3722                  */
3723                 if (atomic_read(&bbio->error) > bbio->max_errors) {
3724                         err = -EIO;
3725                 } else {
3726                         /*
3727                          * this bio is actually up to date, we didn't
3728                          * go over the max number of errors
3729                          */
3730                         set_bit(BIO_UPTODATE, &bio->bi_flags);
3731                         err = 0;
3732                 }
3733                 kfree(bbio);
3734
3735                 bio_endio(bio, err);
3736         } else if (!is_orig_bio) {
3737                 bio_put(bio);
3738         }
3739 }
3740
3741 struct async_sched {
3742         struct bio *bio;
3743         int rw;
3744         struct btrfs_fs_info *info;
3745         struct btrfs_work work;
3746 };
3747
3748 /*
3749  * see run_scheduled_bios for a description of why bios are collected for
3750  * async submit.
3751  *
3752  * This will add one bio to the pending list for a device and make sure
3753  * the work struct is scheduled.
3754  */
3755 static noinline int schedule_bio(struct btrfs_root *root,
3756                                  struct btrfs_device *device,
3757                                  int rw, struct bio *bio)
3758 {
3759         int should_queue = 1;
3760         struct btrfs_pending_bios *pending_bios;
3761
3762         /* don't bother with additional async steps for reads, right now */
3763         if (!(rw & REQ_WRITE)) {
3764                 bio_get(bio);
3765                 submit_bio(rw, bio);
3766                 bio_put(bio);
3767                 return 0;
3768         }
3769
3770         /*
3771          * nr_async_bios allows us to reliably return congestion to the
3772          * higher layers.  Otherwise, the async bio makes it appear we have
3773          * made progress against dirty pages when we've really just put it
3774          * on a queue for later
3775          */
3776         atomic_inc(&root->fs_info->nr_async_bios);
3777         WARN_ON(bio->bi_next);
3778         bio->bi_next = NULL;
3779         bio->bi_rw |= rw;
3780
3781         spin_lock(&device->io_lock);
3782         if (bio->bi_rw & REQ_SYNC)
3783                 pending_bios = &device->pending_sync_bios;
3784         else
3785                 pending_bios = &device->pending_bios;
3786
3787         if (pending_bios->tail)
3788                 pending_bios->tail->bi_next = bio;
3789
3790         pending_bios->tail = bio;
3791         if (!pending_bios->head)
3792                 pending_bios->head = bio;
3793         if (device->running_pending)
3794                 should_queue = 0;
3795
3796         spin_unlock(&device->io_lock);
3797
3798         if (should_queue)
3799                 btrfs_queue_worker(&root->fs_info->submit_workers,
3800                                    &device->work);
3801         return 0;
3802 }
3803
3804 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
3805                   int mirror_num, int async_submit)
3806 {
3807         struct btrfs_mapping_tree *map_tree;
3808         struct btrfs_device *dev;
3809         struct bio *first_bio = bio;
3810         u64 logical = (u64)bio->bi_sector << 9;
3811         u64 length = 0;
3812         u64 map_length;
3813         int ret;
3814         int dev_nr = 0;
3815         int total_devs = 1;
3816         struct btrfs_bio *bbio = NULL;
3817
3818         length = bio->bi_size;
3819         map_tree = &root->fs_info->mapping_tree;
3820         map_length = length;
3821
3822         ret = btrfs_map_block(map_tree, rw, logical, &map_length, &bbio,
3823                               mirror_num);
3824         BUG_ON(ret);
3825
3826         total_devs = bbio->num_stripes;
3827         if (map_length < length) {
3828                 printk(KERN_CRIT "mapping failed logical %llu bio len %llu "
3829                        "len %llu\n", (unsigned long long)logical,
3830                        (unsigned long long)length,
3831                        (unsigned long long)map_length);
3832                 BUG();
3833         }
3834
3835         bbio->orig_bio = first_bio;
3836         bbio->private = first_bio->bi_private;
3837         bbio->end_io = first_bio->bi_end_io;
3838         atomic_set(&bbio->stripes_pending, bbio->num_stripes);
3839
3840         while (dev_nr < total_devs) {
3841                 if (dev_nr < total_devs - 1) {
3842                         bio = bio_clone(first_bio, GFP_NOFS);
3843                         BUG_ON(!bio);
3844                 } else {
3845                         bio = first_bio;
3846                 }
3847                 bio->bi_private = bbio;
3848                 bio->bi_end_io = btrfs_end_bio;
3849                 bio->bi_sector = bbio->stripes[dev_nr].physical >> 9;
3850                 dev = bbio->stripes[dev_nr].dev;
3851                 if (dev && dev->bdev && (rw != WRITE || dev->writeable)) {
3852                         pr_debug("btrfs_map_bio: rw %d, secor=%llu, dev=%lu "
3853                                  "(%s id %llu), size=%u\n", rw,
3854                                  (u64)bio->bi_sector, (u_long)dev->bdev->bd_dev,
3855                                  dev->name, dev->devid, bio->bi_size);
3856                         bio->bi_bdev = dev->bdev;
3857                         if (async_submit)
3858                                 schedule_bio(root, dev, rw, bio);
3859                         else
3860                                 submit_bio(rw, bio);
3861                 } else {
3862                         bio->bi_bdev = root->fs_info->fs_devices->latest_bdev;
3863                         bio->bi_sector = logical >> 9;
3864                         bio_endio(bio, -EIO);
3865                 }
3866                 dev_nr++;
3867         }
3868         return 0;
3869 }
3870
3871 struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid,
3872                                        u8 *uuid, u8 *fsid)
3873 {
3874         struct btrfs_device *device;
3875         struct btrfs_fs_devices *cur_devices;
3876
3877         cur_devices = root->fs_info->fs_devices;
3878         while (cur_devices) {
3879                 if (!fsid ||
3880                     !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
3881                         device = __find_device(&cur_devices->devices,
3882                                                devid, uuid);
3883                         if (device)
3884                                 return device;
3885                 }
3886                 cur_devices = cur_devices->seed;
3887         }
3888         return NULL;
3889 }
3890
3891 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
3892                                             u64 devid, u8 *dev_uuid)
3893 {
3894         struct btrfs_device *device;
3895         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
3896
3897         device = kzalloc(sizeof(*device), GFP_NOFS);
3898         if (!device)
3899                 return NULL;
3900         list_add(&device->dev_list,
3901                  &fs_devices->devices);
3902         device->dev_root = root->fs_info->dev_root;
3903         device->devid = devid;
3904         device->work.func = pending_bios_fn;
3905         device->fs_devices = fs_devices;
3906         device->missing = 1;
3907         fs_devices->num_devices++;
3908         fs_devices->missing_devices++;
3909         spin_lock_init(&device->io_lock);
3910         INIT_LIST_HEAD(&device->dev_alloc_list);
3911         memcpy(device->uuid, dev_uuid, BTRFS_UUID_SIZE);
3912         return device;
3913 }
3914
3915 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
3916                           struct extent_buffer *leaf,
3917                           struct btrfs_chunk *chunk)
3918 {
3919         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
3920         struct map_lookup *map;
3921         struct extent_map *em;
3922         u64 logical;
3923         u64 length;
3924         u64 devid;
3925         u8 uuid[BTRFS_UUID_SIZE];
3926         int num_stripes;
3927         int ret;
3928         int i;
3929
3930         logical = key->offset;
3931         length = btrfs_chunk_length(leaf, chunk);
3932
3933         read_lock(&map_tree->map_tree.lock);
3934         em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
3935         read_unlock(&map_tree->map_tree.lock);
3936
3937         /* already mapped? */
3938         if (em && em->start <= logical && em->start + em->len > logical) {
3939                 free_extent_map(em);
3940                 return 0;
3941         } else if (em) {
3942                 free_extent_map(em);
3943         }
3944
3945         em = alloc_extent_map();
3946         if (!em)
3947                 return -ENOMEM;
3948         num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3949         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
3950         if (!map) {
3951                 free_extent_map(em);
3952                 return -ENOMEM;
3953         }
3954
3955         em->bdev = (struct block_device *)map;
3956         em->start = logical;
3957         em->len = length;
3958         em->block_start = 0;
3959         em->block_len = em->len;
3960
3961         map->num_stripes = num_stripes;
3962         map->io_width = btrfs_chunk_io_width(leaf, chunk);
3963         map->io_align = btrfs_chunk_io_align(leaf, chunk);
3964         map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
3965         map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
3966         map->type = btrfs_chunk_type(leaf, chunk);
3967         map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
3968         for (i = 0; i < num_stripes; i++) {
3969                 map->stripes[i].physical =
3970                         btrfs_stripe_offset_nr(leaf, chunk, i);
3971                 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
3972                 read_extent_buffer(leaf, uuid, (unsigned long)
3973                                    btrfs_stripe_dev_uuid_nr(chunk, i),
3974                                    BTRFS_UUID_SIZE);
3975                 map->stripes[i].dev = btrfs_find_device(root, devid, uuid,
3976                                                         NULL);
3977                 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
3978                         kfree(map);
3979                         free_extent_map(em);
3980                         return -EIO;
3981                 }
3982                 if (!map->stripes[i].dev) {
3983                         map->stripes[i].dev =
3984                                 add_missing_dev(root, devid, uuid);
3985                         if (!map->stripes[i].dev) {
3986                                 kfree(map);
3987                                 free_extent_map(em);
3988                                 return -EIO;
3989                         }
3990                 }
3991                 map->stripes[i].dev->in_fs_metadata = 1;
3992         }
3993
3994         write_lock(&map_tree->map_tree.lock);
3995         ret = add_extent_mapping(&map_tree->map_tree, em);
3996         write_unlock(&map_tree->map_tree.lock);
3997         BUG_ON(ret);
3998         free_extent_map(em);
3999
4000         return 0;
4001 }
4002
4003 static int fill_device_from_item(struct extent_buffer *leaf,
4004                                  struct btrfs_dev_item *dev_item,
4005                                  struct btrfs_device *device)
4006 {
4007         unsigned long ptr;
4008
4009         device->devid = btrfs_device_id(leaf, dev_item);
4010         device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
4011         device->total_bytes = device->disk_total_bytes;
4012         device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
4013         device->type = btrfs_device_type(leaf, dev_item);
4014         device->io_align = btrfs_device_io_align(leaf, dev_item);
4015         device->io_width = btrfs_device_io_width(leaf, dev_item);
4016         device->sector_size = btrfs_device_sector_size(leaf, dev_item);
4017
4018         ptr = (unsigned long)btrfs_device_uuid(dev_item);
4019         read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
4020
4021         return 0;
4022 }
4023
4024 static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
4025 {
4026         struct btrfs_fs_devices *fs_devices;
4027         int ret;
4028
4029         mutex_lock(&uuid_mutex);
4030
4031         fs_devices = root->fs_info->fs_devices->seed;
4032         while (fs_devices) {
4033                 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
4034                         ret = 0;
4035                         goto out;
4036                 }
4037                 fs_devices = fs_devices->seed;
4038         }
4039
4040         fs_devices = find_fsid(fsid);
4041         if (!fs_devices) {
4042                 ret = -ENOENT;
4043                 goto out;
4044         }
4045
4046         fs_devices = clone_fs_devices(fs_devices);
4047         if (IS_ERR(fs_devices)) {
4048                 ret = PTR_ERR(fs_devices);
4049                 goto out;
4050         }
4051
4052         ret = __btrfs_open_devices(fs_devices, FMODE_READ,
4053                                    root->fs_info->bdev_holder);
4054         if (ret)
4055                 goto out;
4056
4057         if (!fs_devices->seeding) {
4058                 __btrfs_close_devices(fs_devices);
4059                 free_fs_devices(fs_devices);
4060                 ret = -EINVAL;
4061                 goto out;
4062         }
4063
4064         fs_devices->seed = root->fs_info->fs_devices->seed;
4065         root->fs_info->fs_devices->seed = fs_devices;
4066 out:
4067         mutex_unlock(&uuid_mutex);
4068         return ret;
4069 }
4070
4071 static int read_one_dev(struct btrfs_root *root,
4072                         struct extent_buffer *leaf,
4073                         struct btrfs_dev_item *dev_item)
4074 {
4075         struct btrfs_device *device;
4076         u64 devid;
4077         int ret;
4078         u8 fs_uuid[BTRFS_UUID_SIZE];
4079         u8 dev_uuid[BTRFS_UUID_SIZE];
4080
4081         devid = btrfs_device_id(leaf, dev_item);
4082         read_extent_buffer(leaf, dev_uuid,
4083                            (unsigned long)btrfs_device_uuid(dev_item),
4084                            BTRFS_UUID_SIZE);
4085         read_extent_buffer(leaf, fs_uuid,
4086                            (unsigned long)btrfs_device_fsid(dev_item),
4087                            BTRFS_UUID_SIZE);
4088
4089         if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
4090                 ret = open_seed_devices(root, fs_uuid);
4091                 if (ret && !btrfs_test_opt(root, DEGRADED))
4092                         return ret;
4093         }
4094
4095         device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
4096         if (!device || !device->bdev) {
4097                 if (!btrfs_test_opt(root, DEGRADED))
4098                         return -EIO;
4099
4100                 if (!device) {
4101                         printk(KERN_WARNING "warning devid %llu missing\n",
4102                                (unsigned long long)devid);
4103                         device = add_missing_dev(root, devid, dev_uuid);
4104                         if (!device)
4105                                 return -ENOMEM;
4106                 } else if (!device->missing) {
4107                         /*
4108                          * this happens when a device that was properly setup
4109                          * in the device info lists suddenly goes bad.
4110                          * device->bdev is NULL, and so we have to set
4111                          * device->missing to one here
4112                          */
4113                         root->fs_info->fs_devices->missing_devices++;
4114                         device->missing = 1;
4115                 }
4116         }
4117
4118         if (device->fs_devices != root->fs_info->fs_devices) {
4119                 BUG_ON(device->writeable);
4120                 if (device->generation !=
4121                     btrfs_device_generation(leaf, dev_item))
4122                         return -EINVAL;
4123         }
4124
4125         fill_device_from_item(leaf, dev_item, device);
4126         device->dev_root = root->fs_info->dev_root;
4127         device->in_fs_metadata = 1;
4128         if (device->writeable) {
4129                 device->fs_devices->total_rw_bytes += device->total_bytes;
4130                 spin_lock(&root->fs_info->free_chunk_lock);
4131                 root->fs_info->free_chunk_space += device->total_bytes -
4132                         device->bytes_used;
4133                 spin_unlock(&root->fs_info->free_chunk_lock);
4134         }
4135         ret = 0;
4136         return ret;
4137 }
4138
4139 int btrfs_read_sys_array(struct btrfs_root *root)
4140 {
4141         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
4142         struct extent_buffer *sb;
4143         struct btrfs_disk_key *disk_key;
4144         struct btrfs_chunk *chunk;
4145         u8 *ptr;
4146         unsigned long sb_ptr;
4147         int ret = 0;
4148         u32 num_stripes;
4149         u32 array_size;
4150         u32 len = 0;
4151         u32 cur;
4152         struct btrfs_key key;
4153
4154         sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
4155                                           BTRFS_SUPER_INFO_SIZE);
4156         if (!sb)
4157                 return -ENOMEM;
4158         btrfs_set_buffer_uptodate(sb);
4159         btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
4160
4161         write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
4162         array_size = btrfs_super_sys_array_size(super_copy);
4163
4164         ptr = super_copy->sys_chunk_array;
4165         sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
4166         cur = 0;
4167
4168         while (cur < array_size) {
4169                 disk_key = (struct btrfs_disk_key *)ptr;
4170                 btrfs_disk_key_to_cpu(&key, disk_key);
4171
4172                 len = sizeof(*disk_key); ptr += len;
4173                 sb_ptr += len;
4174                 cur += len;
4175
4176                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
4177                         chunk = (struct btrfs_chunk *)sb_ptr;
4178                         ret = read_one_chunk(root, &key, sb, chunk);
4179                         if (ret)
4180                                 break;
4181                         num_stripes = btrfs_chunk_num_stripes(sb, chunk);
4182                         len = btrfs_chunk_item_size(num_stripes);
4183                 } else {
4184                         ret = -EIO;
4185                         break;
4186                 }
4187                 ptr += len;
4188                 sb_ptr += len;
4189                 cur += len;
4190         }
4191         free_extent_buffer(sb);
4192         return ret;
4193 }
4194
4195 int btrfs_read_chunk_tree(struct btrfs_root *root)
4196 {
4197         struct btrfs_path *path;
4198         struct extent_buffer *leaf;
4199         struct btrfs_key key;
4200         struct btrfs_key found_key;
4201         int ret;
4202         int slot;
4203
4204         root = root->fs_info->chunk_root;
4205
4206         path = btrfs_alloc_path();
4207         if (!path)
4208                 return -ENOMEM;
4209
4210         /* first we search for all of the device items, and then we
4211          * read in all of the chunk items.  This way we can create chunk
4212          * mappings that reference all of the devices that are afound
4213          */
4214         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
4215         key.offset = 0;
4216         key.type = 0;
4217 again:
4218         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4219         if (ret < 0)
4220                 goto error;
4221         while (1) {
4222                 leaf = path->nodes[0];
4223                 slot = path->slots[0];
4224                 if (slot >= btrfs_header_nritems(leaf)) {
4225                         ret = btrfs_next_leaf(root, path);
4226                         if (ret == 0)
4227                                 continue;
4228                         if (ret < 0)
4229                                 goto error;
4230                         break;
4231                 }
4232                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4233                 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
4234                         if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID)
4235                                 break;
4236                         if (found_key.type == BTRFS_DEV_ITEM_KEY) {
4237                                 struct btrfs_dev_item *dev_item;
4238                                 dev_item = btrfs_item_ptr(leaf, slot,
4239                                                   struct btrfs_dev_item);
4240                                 ret = read_one_dev(root, leaf, dev_item);
4241                                 if (ret)
4242                                         goto error;
4243                         }
4244                 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
4245                         struct btrfs_chunk *chunk;
4246                         chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
4247                         ret = read_one_chunk(root, &found_key, leaf, chunk);
4248                         if (ret)
4249                                 goto error;
4250                 }
4251                 path->slots[0]++;
4252         }
4253         if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
4254                 key.objectid = 0;
4255                 btrfs_release_path(path);
4256                 goto again;
4257         }
4258         ret = 0;
4259 error:
4260         btrfs_free_path(path);
4261         return ret;
4262 }