fs/btrfs/volumes.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18 #include <linux/sched.h>
  19 #include <linux/bio.h>
  20 #include <linux/slab.h>
  21 #include <linux/buffer_head.h>
  22 #include <linux/blkdev.h>
  23 #include <linux/random.h>
  24 #include <linux/iocontext.h>
  25 #include <linux/capability.h>
  26 #include <linux/ratelimit.h>
  27 #include <linux/kthread.h>
  28 #include <linux/raid/pq.h>
  29 #include <linux/semaphore.h>
  30 #include <asm/div64.h>
  31 #include "compat.h"
  32 #include "ctree.h"
  33 #include "extent_map.h"
  34 #include "disk-io.h"
  35 #include "transaction.h"
  36 #include "print-tree.h"
  37 #include "volumes.h"
  38 #include "raid56.h"
  39 #include "async-thread.h"
  40 #include "check-integrity.h"
  41 #include "rcu-string.h"
  42 #include "math.h"
  43 #include "dev-replace.h"
  44
  45 static int init_first_rw_device(struct btrfs_trans_handle *trans,
  46                                 struct btrfs_root *root,
  47                                 struct btrfs_device *device);
  48 static int btrfs_relocate_sys_chunks(struct btrfs_root *root);
  49 static void __btrfs_reset_dev_stats(struct btrfs_device *dev);
  50 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev);
  51 static void btrfs_dev_stat_print_on_load(struct btrfs_device *device);
  52
  53 static DEFINE_MUTEX(uuid_mutex);
  54 static LIST_HEAD(fs_uuids);
  55
  56 static void lock_chunks(struct btrfs_root *root)
  57 {
  58         mutex_lock(&root->fs_info->chunk_mutex);
  59 }
  60
  61 static void unlock_chunks(struct btrfs_root *root)
  62 {
  63         mutex_unlock(&root->fs_info->chunk_mutex);
  64 }
  65
  66 static struct btrfs_fs_devices *__alloc_fs_devices(void)
  67 {
  68         struct btrfs_fs_devices *fs_devs;
  69
  70         fs_devs = kzalloc(sizeof(*fs_devs), GFP_NOFS);
  71         if (!fs_devs)
  72                 return ERR_PTR(-ENOMEM);
  73
  74         mutex_init(&fs_devs->device_list_mutex);
  75
  76         INIT_LIST_HEAD(&fs_devs->devices);
  77         INIT_LIST_HEAD(&fs_devs->alloc_list);
  78         INIT_LIST_HEAD(&fs_devs->list);
  79
  80         return fs_devs;
  81 }
  82
  83 /**
  84  * alloc_fs_devices - allocate struct btrfs_fs_devices
  85  * @fsid:       a pointer to UUID for this FS.  If NULL a new UUID is
  86  *              generated.
  87  *
  88  * Return: a pointer to a new &struct btrfs_fs_devices on success;
  89  * ERR_PTR() on error.  Returned struct is not linked onto any lists and
  90  * can be destroyed with kfree() right away.
  91  */
  92 static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid)
  93 {
  94         struct btrfs_fs_devices *fs_devs;
  95
  96         fs_devs = __alloc_fs_devices();
  97         if (IS_ERR(fs_devs))
  98                 return fs_devs;
  99
 100         if (fsid)
 101                 memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE);
 102         else
 103                 generate_random_uuid(fs_devs->fsid);
 104
 105         return fs_devs;
 106 }
 107
 108 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
 109 {
 110         struct btrfs_device *device;
 111         WARN_ON(fs_devices->opened);
 112         while (!list_empty(&fs_devices->devices)) {
 113                 device = list_entry(fs_devices->devices.next,
 114                                     struct btrfs_device, dev_list);
 115                 list_del(&device->dev_list);
 116                 rcu_string_free(device->name);
 117                 kfree(device);
 118         }
 119         kfree(fs_devices);
 120 }
 121
 122 static void btrfs_kobject_uevent(struct block_device *bdev,
 123                                  enum kobject_action action)
 124 {
 125         int ret;
 126
 127         ret = kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, action);
 128         if (ret)
 129                 pr_warn("Sending event '%d' to kobject: '%s' (%p): failed\n",
 130                         action,
 131                         kobject_name(&disk_to_dev(bdev->bd_disk)->kobj),
 132                         &disk_to_dev(bdev->bd_disk)->kobj);
 133 }
 134
 135 void btrfs_cleanup_fs_uuids(void)
 136 {
 137         struct btrfs_fs_devices *fs_devices;
 138
 139         while (!list_empty(&fs_uuids)) {
 140                 fs_devices = list_entry(fs_uuids.next,
 141                                         struct btrfs_fs_devices, list);
 142                 list_del(&fs_devices->list);
 143                 free_fs_devices(fs_devices);
 144         }
 145 }
 146
 147 static struct btrfs_device *__alloc_device(void)
 148 {
 149         struct btrfs_device *dev;
 150
 151         dev = kzalloc(sizeof(*dev), GFP_NOFS);
 152         if (!dev)
 153                 return ERR_PTR(-ENOMEM);
 154
 155         INIT_LIST_HEAD(&dev->dev_list);
 156         INIT_LIST_HEAD(&dev->dev_alloc_list);
 157
 158         spin_lock_init(&dev->io_lock);
 159
 160         spin_lock_init(&dev->reada_lock);
 161         atomic_set(&dev->reada_in_flight, 0);
 162         INIT_RADIX_TREE(&dev->reada_zones, GFP_NOFS & ~__GFP_WAIT);
 163         INIT_RADIX_TREE(&dev->reada_extents, GFP_NOFS & ~__GFP_WAIT);
 164
 165         return dev;
 166 }
 167
 168 static noinline struct btrfs_device *__find_device(struct list_head *head,
 169                                                    u64 devid, u8 *uuid)
 170 {
 171         struct btrfs_device *dev;
 172
 173         list_for_each_entry(dev, head, dev_list) {
 174                 if (dev->devid == devid &&
 175                     (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
 176                         return dev;
 177                 }
 178         }
 179         return NULL;
 180 }
 181
 182 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
 183 {
 184         struct btrfs_fs_devices *fs_devices;
 185
 186         list_for_each_entry(fs_devices, &fs_uuids, list) {
 187                 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
 188                         return fs_devices;
 189         }
 190         return NULL;
 191 }
 192
 193 static int
 194 btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder,
 195                       int flush, struct block_device **bdev,
 196                       struct buffer_head **bh)
 197 {
 198         int ret;
 199
 200         *bdev = blkdev_get_by_path(device_path, flags, holder);
 201
 202         if (IS_ERR(*bdev)) {
 203                 ret = PTR_ERR(*bdev);
 204                 printk(KERN_INFO "btrfs: open %s failed\n", device_path);
 205                 goto error;
 206         }
 207
 208         if (flush)
 209                 filemap_write_and_wait((*bdev)->bd_inode->i_mapping);
 210         ret = set_blocksize(*bdev, 4096);
 211         if (ret) {
 212                 blkdev_put(*bdev, flags);
 213                 goto error;
 214         }
 215         invalidate_bdev(*bdev);
 216         *bh = btrfs_read_dev_super(*bdev);
 217         if (!*bh) {
 218                 ret = -EINVAL;
 219                 blkdev_put(*bdev, flags);
 220                 goto error;
 221         }
 222
 223         return 0;
 224
 225 error:
 226         *bdev = NULL;
 227         *bh = NULL;
 228         return ret;
 229 }
 230
 231 static void requeue_list(struct btrfs_pending_bios *pending_bios,
 232                         struct bio *head, struct bio *tail)
 233 {
 234
 235         struct bio *old_head;
 236
 237         old_head = pending_bios->head;
 238         pending_bios->head = head;
 239         if (pending_bios->tail)
 240                 tail->bi_next = old_head;
 241         else
 242                 pending_bios->tail = tail;
 243 }
 244
 245 /*
 246  * we try to collect pending bios for a device so we don't get a large
 247  * number of procs sending bios down to the same device.  This greatly
 248  * improves the schedulers ability to collect and merge the bios.
 249  *
 250  * But, it also turns into a long list of bios to process and that is sure
 251  * to eventually make the worker thread block.  The solution here is to
 252  * make some progress and then put this work struct back at the end of
 253  * the list if the block device is congested.  This way, multiple devices
 254  * can make progress from a single worker thread.
 255  */
 256 static noinline void run_scheduled_bios(struct btrfs_device *device)
 257 {
 258         struct bio *pending;
 259         struct backing_dev_info *bdi;
 260         struct btrfs_fs_info *fs_info;
 261         struct btrfs_pending_bios *pending_bios;
 262         struct bio *tail;
 263         struct bio *cur;
 264         int again = 0;
 265         unsigned long num_run;
 266         unsigned long batch_run = 0;
 267         unsigned long limit;
 268         unsigned long last_waited = 0;
 269         int force_reg = 0;
 270         int sync_pending = 0;
 271         struct blk_plug plug;
 272
 273         /*
 274          * this function runs all the bios we've collected for
 275          * a particular device.  We don't want to wander off to
 276          * another device without first sending all of these down.
 277          * So, setup a plug here and finish it off before we return
 278          */
 279         blk_start_plug(&plug);
 280
 281         bdi = blk_get_backing_dev_info(device->bdev);
 282         fs_info = device->dev_root->fs_info;
 283         limit = btrfs_async_submit_limit(fs_info);
 284         limit = limit * 2 / 3;
 285
 286 loop:
 287         spin_lock(&device->io_lock);
 288
 289 loop_lock:
 290         num_run = 0;
 291
 292         /* take all the bios off the list at once and process them
 293          * later on (without the lock held).  But, remember the
 294          * tail and other pointers so the bios can be properly reinserted
 295          * into the list if we hit congestion
 296          */
 297         if (!force_reg && device->pending_sync_bios.head) {
 298                 pending_bios = &device->pending_sync_bios;
 299                 force_reg = 1;
 300         } else {
 301                 pending_bios = &device->pending_bios;
 302                 force_reg = 0;
 303         }
 304
 305         pending = pending_bios->head;
 306         tail = pending_bios->tail;
 307         WARN_ON(pending && !tail);
 308
 309         /*
 310          * if pending was null this time around, no bios need processing
 311          * at all and we can stop.  Otherwise it'll loop back up again
 312          * and do an additional check so no bios are missed.
 313          *
 314          * device->running_pending is used to synchronize with the
 315          * schedule_bio code.
 316          */
 317         if (device->pending_sync_bios.head == NULL &&
 318             device->pending_bios.head == NULL) {
 319                 again = 0;
 320                 device->running_pending = 0;
 321         } else {
 322                 again = 1;
 323                 device->running_pending = 1;
 324         }
 325
 326         pending_bios->head = NULL;
 327         pending_bios->tail = NULL;
 328
 329         spin_unlock(&device->io_lock);
 330
 331         while (pending) {
 332
 333                 rmb();
 334                 /* we want to work on both lists, but do more bios on the
 335                  * sync list than the regular list
 336                  */
 337                 if ((num_run > 32 &&
 338                     pending_bios != &device->pending_sync_bios &&
 339                     device->pending_sync_bios.head) ||
 340                    (num_run > 64 && pending_bios == &device->pending_sync_bios &&
 341                     device->pending_bios.head)) {
 342                         spin_lock(&device->io_lock);
 343                         requeue_list(pending_bios, pending, tail);
 344                         goto loop_lock;
 345                 }
 346
 347                 cur = pending;
 348                 pending = pending->bi_next;
 349                 cur->bi_next = NULL;
 350
 351                 if (atomic_dec_return(&fs_info->nr_async_bios) < limit &&
 352                     waitqueue_active(&fs_info->async_submit_wait))
 353                         wake_up(&fs_info->async_submit_wait);
 354
 355                 BUG_ON(atomic_read(&cur->bi_cnt) == 0);
 356
 357                 /*
 358                  * if we're doing the sync list, record that our
 359                  * plug has some sync requests on it
 360                  *
 361                  * If we're doing the regular list and there are
 362                  * sync requests sitting around, unplug before
 363                  * we add more
 364                  */
 365                 if (pending_bios == &device->pending_sync_bios) {
 366                         sync_pending = 1;
 367                 } else if (sync_pending) {
 368                         blk_finish_plug(&plug);
 369                         blk_start_plug(&plug);
 370                         sync_pending = 0;
 371                 }
 372
 373                 btrfsic_submit_bio(cur->bi_rw, cur);
 374                 num_run++;
 375                 batch_run++;
 376                 if (need_resched())
 377                         cond_resched();
 378
 379                 /*
 380                  * we made progress, there is more work to do and the bdi
 381                  * is now congested.  Back off and let other work structs
 382                  * run instead
 383                  */
 384                 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
 385                     fs_info->fs_devices->open_devices > 1) {
 386                         struct io_context *ioc;
 387
 388                         ioc = current->io_context;
 389
 390                         /*
 391                          * the main goal here is that we don't want to
 392                          * block if we're going to be able to submit
 393                          * more requests without blocking.
 394                          *
 395                          * This code does two great things, it pokes into
 396                          * the elevator code from a filesystem _and_
 397                          * it makes assumptions about how batching works.
 398                          */
 399                         if (ioc && ioc->nr_batch_requests > 0 &&
 400                             time_before(jiffies, ioc->last_waited + HZ/50UL) &&
 401                             (last_waited == 0 ||
 402                              ioc->last_waited == last_waited)) {
 403                                 /*
 404                                  * we want to go through our batch of
 405                                  * requests and stop.  So, we copy out
 406                                  * the ioc->last_waited time and test
 407                                  * against it before looping
 408                                  */
 409                                 last_waited = ioc->last_waited;
 410                                 if (need_resched())
 411                                         cond_resched();
 412                                 continue;
 413                         }
 414                         spin_lock(&device->io_lock);
 415                         requeue_list(pending_bios, pending, tail);
 416                         device->running_pending = 1;
 417
 418                         spin_unlock(&device->io_lock);
 419                         btrfs_requeue_work(&device->work);
 420                         goto done;
 421                 }
 422                 /* unplug every 64 requests just for good measure */
 423                 if (batch_run % 64 == 0) {
 424                         blk_finish_plug(&plug);
 425                         blk_start_plug(&plug);
 426                         sync_pending = 0;
 427                 }
 428         }
 429
 430         cond_resched();
 431         if (again)
 432                 goto loop;
 433
 434         spin_lock(&device->io_lock);
 435         if (device->pending_bios.head || device->pending_sync_bios.head)
 436                 goto loop_lock;
 437         spin_unlock(&device->io_lock);
 438
 439 done:
 440         blk_finish_plug(&plug);
 441 }
 442
 443 static void pending_bios_fn(struct btrfs_work *work)
 444 {
 445         struct btrfs_device *device;
 446
 447         device = container_of(work, struct btrfs_device, work);
 448         run_scheduled_bios(device);
 449 }
 450
 451 static noinline int device_list_add(const char *path,
 452                            struct btrfs_super_block *disk_super,
 453                            u64 devid, struct btrfs_fs_devices **fs_devices_ret)
 454 {
 455         struct btrfs_device *device;
 456         struct btrfs_fs_devices *fs_devices;
 457         struct rcu_string *name;
 458         u64 found_transid = btrfs_super_generation(disk_super);
 459
 460         fs_devices = find_fsid(disk_super->fsid);
 461         if (!fs_devices) {
 462                 fs_devices = alloc_fs_devices(disk_super->fsid);
 463                 if (IS_ERR(fs_devices))
 464                         return PTR_ERR(fs_devices);
 465
 466                 list_add(&fs_devices->list, &fs_uuids);
 467                 fs_devices->latest_devid = devid;
 468                 fs_devices->latest_trans = found_transid;
 469
 470                 device = NULL;
 471         } else {
 472                 device = __find_device(&fs_devices->devices, devid,
 473                                        disk_super->dev_item.uuid);
 474         }
 475         if (!device) {
 476                 if (fs_devices->opened)
 477                         return -EBUSY;
 478
 479                 device = btrfs_alloc_device(NULL, &devid,
 480                                             disk_super->dev_item.uuid);
 481                 if (IS_ERR(device)) {
 482                         /* we can safely leave the fs_devices entry around */
 483                         return PTR_ERR(device);
 484                 }
 485
 486                 name = rcu_string_strdup(path, GFP_NOFS);
 487                 if (!name) {
 488                         kfree(device);
 489                         return -ENOMEM;
 490                 }
 491                 rcu_assign_pointer(device->name, name);
 492
 493                 mutex_lock(&fs_devices->device_list_mutex);
 494                 list_add_rcu(&device->dev_list, &fs_devices->devices);
 495                 fs_devices->num_devices++;
 496                 mutex_unlock(&fs_devices->device_list_mutex);
 497
 498                 device->fs_devices = fs_devices;
 499         } else if (!device->name || strcmp(device->name->str, path)) {
 500                 name = rcu_string_strdup(path, GFP_NOFS);
 501                 if (!name)
 502                         return -ENOMEM;
 503                 rcu_string_free(device->name);
 504                 rcu_assign_pointer(device->name, name);
 505                 if (device->missing) {
 506                         fs_devices->missing_devices--;
 507                         device->missing = 0;
 508                 }
 509         }
 510
 511         if (found_transid > fs_devices->latest_trans) {
 512                 fs_devices->latest_devid = devid;
 513                 fs_devices->latest_trans = found_transid;
 514         }
 515         *fs_devices_ret = fs_devices;
 516         return 0;
 517 }
 518
 519 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
 520 {
 521         struct btrfs_fs_devices *fs_devices;
 522         struct btrfs_device *device;
 523         struct btrfs_device *orig_dev;
 524
 525         fs_devices = alloc_fs_devices(orig->fsid);
 526         if (IS_ERR(fs_devices))
 527                 return fs_devices;
 528
 529         fs_devices->latest_devid = orig->latest_devid;
 530         fs_devices->latest_trans = orig->latest_trans;
 531         fs_devices->total_devices = orig->total_devices;
 532
 533         /* We have held the volume lock, it is safe to get the devices. */
 534         list_for_each_entry(orig_dev, &orig->devices, dev_list) {
 535                 struct rcu_string *name;
 536
 537                 device = btrfs_alloc_device(NULL, &orig_dev->devid,
 538                                             orig_dev->uuid);
 539                 if (IS_ERR(device))
 540                         goto error;
 541
 542                 /*
 543                  * This is ok to do without rcu read locked because we hold the
 544                  * uuid mutex so nothing we touch in here is going to disappear.
 545                  */
 546                 name = rcu_string_strdup(orig_dev->name->str, GFP_NOFS);
 547                 if (!name) {
 548                         kfree(device);
 549                         goto error;
 550                 }
 551                 rcu_assign_pointer(device->name, name);
 552
 553                 list_add(&device->dev_list, &fs_devices->devices);
 554                 device->fs_devices = fs_devices;
 555                 fs_devices->num_devices++;
 556         }
 557         return fs_devices;
 558 error:
 559         free_fs_devices(fs_devices);
 560         return ERR_PTR(-ENOMEM);
 561 }
 562
 563 void btrfs_close_extra_devices(struct btrfs_fs_info *fs_info,
 564                                struct btrfs_fs_devices *fs_devices, int step)
 565 {
 566         struct btrfs_device *device, *next;
 567
 568         struct block_device *latest_bdev = NULL;
 569         u64 latest_devid = 0;
 570         u64 latest_transid = 0;
 571
 572         mutex_lock(&uuid_mutex);
 573 again:
 574         /* This is the initialized path, it is safe to release the devices. */
 575         list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
 576                 if (device->in_fs_metadata) {
 577                         if (!device->is_tgtdev_for_dev_replace &&
 578                             (!latest_transid ||
 579                              device->generation > latest_transid)) {
 580                                 latest_devid = device->devid;
 581                                 latest_transid = device->generation;
 582                                 latest_bdev = device->bdev;
 583                         }
 584                         continue;
 585                 }
 586
 587                 if (device->devid == BTRFS_DEV_REPLACE_DEVID) {
 588                         /*
 589                          * In the first step, keep the device which has
 590                          * the correct fsid and the devid that is used
 591                          * for the dev_replace procedure.
 592                          * In the second step, the dev_replace state is
 593                          * read from the device tree and it is known
 594                          * whether the procedure is really active or
 595                          * not, which means whether this device is
 596                          * used or whether it should be removed.
 597                          */
 598                         if (step == 0 || device->is_tgtdev_for_dev_replace) {
 599                                 continue;
 600                         }
 601                 }
 602                 if (device->bdev) {
 603                         blkdev_put(device->bdev, device->mode);
 604                         device->bdev = NULL;
 605                         fs_devices->open_devices--;
 606                 }
 607                 if (device->writeable) {
 608                         list_del_init(&device->dev_alloc_list);
 609                         device->writeable = 0;
 610                         if (!device->is_tgtdev_for_dev_replace)
 611                                 fs_devices->rw_devices--;
 612                 }
 613                 list_del_init(&device->dev_list);
 614                 fs_devices->num_devices--;
 615                 rcu_string_free(device->name);
 616                 kfree(device);
 617         }
 618
 619         if (fs_devices->seed) {
 620                 fs_devices = fs_devices->seed;
 621                 goto again;
 622         }
 623
 624         fs_devices->latest_bdev = latest_bdev;
 625         fs_devices->latest_devid = latest_devid;
 626         fs_devices->latest_trans = latest_transid;
 627
 628         mutex_unlock(&uuid_mutex);
 629 }
 630
 631 static void __free_device(struct work_struct *work)
 632 {
 633         struct btrfs_device *device;
 634
 635         device = container_of(work, struct btrfs_device, rcu_work);
 636
 637         if (device->bdev)
 638                 blkdev_put(device->bdev, device->mode);
 639
 640         rcu_string_free(device->name);
 641         kfree(device);
 642 }
 643
 644 static void free_device(struct rcu_head *head)
 645 {
 646         struct btrfs_device *device;
 647
 648         device = container_of(head, struct btrfs_device, rcu);
 649
 650         INIT_WORK(&device->rcu_work, __free_device);
 651         schedule_work(&device->rcu_work);
 652 }
 653
 654 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
 655 {
 656         struct btrfs_device *device;
 657
 658         if (--fs_devices->opened > 0)
 659                 return 0;
 660
 661         mutex_lock(&fs_devices->device_list_mutex);
 662         list_for_each_entry(device, &fs_devices->devices, dev_list) {
 663                 struct btrfs_device *new_device;
 664                 struct rcu_string *name;
 665
 666                 if (device->bdev)
 667                         fs_devices->open_devices--;
 668
 669                 if (device->writeable && !device->is_tgtdev_for_dev_replace) {
 670                         list_del_init(&device->dev_alloc_list);
 671                         fs_devices->rw_devices--;
 672                 }
 673
 674                 if (device->can_discard)
 675                         fs_devices->num_can_discard--;
 676                 if (device->missing)
 677                         fs_devices->missing_devices--;
 678
 679                 new_device = btrfs_alloc_device(NULL, &device->devid,
 680                                                 device->uuid);
 681                 BUG_ON(IS_ERR(new_device)); /* -ENOMEM */
 682
 683                 /* Safe because we are under uuid_mutex */
 684                 if (device->name) {
 685                         name = rcu_string_strdup(device->name->str, GFP_NOFS);
 686                         BUG_ON(!name); /* -ENOMEM */
 687                         rcu_assign_pointer(new_device->name, name);
 688                 }
 689
 690                 list_replace_rcu(&device->dev_list, &new_device->dev_list);
 691                 new_device->fs_devices = device->fs_devices;
 692
 693                 call_rcu(&device->rcu, free_device);
 694         }
 695         mutex_unlock(&fs_devices->device_list_mutex);
 696
 697         WARN_ON(fs_devices->open_devices);
 698         WARN_ON(fs_devices->rw_devices);
 699         fs_devices->opened = 0;
 700         fs_devices->seeding = 0;
 701
 702         return 0;
 703 }
 704
 705 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
 706 {
 707         struct btrfs_fs_devices *seed_devices = NULL;
 708         int ret;
 709
 710         mutex_lock(&uuid_mutex);
 711         ret = __btrfs_close_devices(fs_devices);
 712         if (!fs_devices->opened) {
 713                 seed_devices = fs_devices->seed;
 714                 fs_devices->seed = NULL;
 715         }
 716         mutex_unlock(&uuid_mutex);
 717
 718         while (seed_devices) {
 719                 fs_devices = seed_devices;
 720                 seed_devices = fs_devices->seed;
 721                 __btrfs_close_devices(fs_devices);
 722                 free_fs_devices(fs_devices);
 723         }
 724         /*
 725          * Wait for rcu kworkers under __btrfs_close_devices
 726          * to finish all blkdev_puts so device is really
 727          * free when umount is done.
 728          */
 729         rcu_barrier();
 730         return ret;
 731 }
 732
 733 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
 734                                 fmode_t flags, void *holder)
 735 {
 736         struct request_queue *q;
 737         struct block_device *bdev;
 738         struct list_head *head = &fs_devices->devices;
 739         struct btrfs_device *device;
 740         struct block_device *latest_bdev = NULL;
 741         struct buffer_head *bh;
 742         struct btrfs_super_block *disk_super;
 743         u64 latest_devid = 0;
 744         u64 latest_transid = 0;
 745         u64 devid;
 746         int seeding = 1;
 747         int ret = 0;
 748
 749         flags |= FMODE_EXCL;
 750
 751         list_for_each_entry(device, head, dev_list) {
 752                 if (device->bdev)
 753                         continue;
 754                 if (!device->name)
 755                         continue;
 756
 757                 /* Just open everything we can; ignore failures here */
 758                 if (btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1,
 759                                             &bdev, &bh))
 760                         continue;
 761
 762                 disk_super = (struct btrfs_super_block *)bh->b_data;
 763                 devid = btrfs_stack_device_id(&disk_super->dev_item);
 764                 if (devid != device->devid)
 765                         goto error_brelse;
 766
 767                 if (memcmp(device->uuid, disk_super->dev_item.uuid,
 768                            BTRFS_UUID_SIZE))
 769                         goto error_brelse;
 770
 771                 device->generation = btrfs_super_generation(disk_super);
 772                 if (!latest_transid || device->generation > latest_transid) {
 773                         latest_devid = devid;
 774                         latest_transid = device->generation;
 775                         latest_bdev = bdev;
 776                 }
 777
 778                 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
 779                         device->writeable = 0;
 780                 } else {
 781                         device->writeable = !bdev_read_only(bdev);
 782                         seeding = 0;
 783                 }
 784
 785                 q = bdev_get_queue(bdev);
 786                 if (blk_queue_discard(q)) {
 787                         device->can_discard = 1;
 788                         fs_devices->num_can_discard++;
 789                 }
 790
 791                 device->bdev = bdev;
 792                 device->in_fs_metadata = 0;
 793                 device->mode = flags;
 794
 795                 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
 796                         fs_devices->rotating = 1;
 797
 798                 fs_devices->open_devices++;
 799                 if (device->writeable && !device->is_tgtdev_for_dev_replace) {
 800                         fs_devices->rw_devices++;
 801                         list_add(&device->dev_alloc_list,
 802                                  &fs_devices->alloc_list);
 803                 }
 804                 brelse(bh);
 805                 continue;
 806
 807 error_brelse:
 808                 brelse(bh);
 809                 blkdev_put(bdev, flags);
 810                 continue;
 811         }
 812         if (fs_devices->open_devices == 0) {
 813                 ret = -EINVAL;
 814                 goto out;
 815         }
 816         fs_devices->seeding = seeding;
 817         fs_devices->opened = 1;
 818         fs_devices->latest_bdev = latest_bdev;
 819         fs_devices->latest_devid = latest_devid;
 820         fs_devices->latest_trans = latest_transid;
 821         fs_devices->total_rw_bytes = 0;
 822 out:
 823         return ret;
 824 }
 825
 826 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
 827                        fmode_t flags, void *holder)
 828 {
 829         int ret;
 830
 831         mutex_lock(&uuid_mutex);
 832         if (fs_devices->opened) {
 833                 fs_devices->opened++;
 834                 ret = 0;
 835         } else {
 836                 ret = __btrfs_open_devices(fs_devices, flags, holder);
 837         }
 838         mutex_unlock(&uuid_mutex);
 839         return ret;
 840 }
 841
 842 /*
 843  * Look for a btrfs signature on a device. This may be called out of the mount path
 844  * and we are not allowed to call set_blocksize during the scan. The superblock
 845  * is read via pagecache
 846  */
 847 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
 848                           struct btrfs_fs_devices **fs_devices_ret)
 849 {
 850         struct btrfs_super_block *disk_super;
 851         struct block_device *bdev;
 852         struct page *page;
 853         void *p;
 854         int ret = -EINVAL;
 855         u64 devid;
 856         u64 transid;
 857         u64 total_devices;
 858         u64 bytenr;
 859         pgoff_t index;
 860
 861         /*
 862          * we would like to check all the supers, but that would make
 863          * a btrfs mount succeed after a mkfs from a different FS.
 864          * So, we need to add a special mount option to scan for
 865          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
 866          */
 867         bytenr = btrfs_sb_offset(0);
 868         flags |= FMODE_EXCL;
 869         mutex_lock(&uuid_mutex);
 870
 871         bdev = blkdev_get_by_path(path, flags, holder);
 872
 873         if (IS_ERR(bdev)) {
 874                 ret = PTR_ERR(bdev);
 875                 goto error;
 876         }
 877
 878         /* make sure our super fits in the device */
 879         if (bytenr + PAGE_CACHE_SIZE >= i_size_read(bdev->bd_inode))
 880                 goto error_bdev_put;
 881
 882         /* make sure our super fits in the page */
 883         if (sizeof(*disk_super) > PAGE_CACHE_SIZE)
 884                 goto error_bdev_put;
 885
 886         /* make sure our super doesn't straddle pages on disk */
 887         index = bytenr >> PAGE_CACHE_SHIFT;
 888         if ((bytenr + sizeof(*disk_super) - 1) >> PAGE_CACHE_SHIFT != index)
 889                 goto error_bdev_put;
 890
 891         /* pull in the page with our super */
 892         page = read_cache_page_gfp(bdev->bd_inode->i_mapping,
 893                                    index, GFP_NOFS);
 894
 895         if (IS_ERR_OR_NULL(page))
 896                 goto error_bdev_put;
 897
 898         p = kmap(page);
 899
 900         /* align our pointer to the offset of the super block */
 901         disk_super = p + (bytenr & ~PAGE_CACHE_MASK);
 902
 903         if (btrfs_super_bytenr(disk_super) != bytenr ||
 904             btrfs_super_magic(disk_super) != BTRFS_MAGIC)
 905                 goto error_unmap;
 906
 907         devid = btrfs_stack_device_id(&disk_super->dev_item);
 908         transid = btrfs_super_generation(disk_super);
 909         total_devices = btrfs_super_num_devices(disk_super);
 910
 911         if (disk_super->label[0]) {
 912                 if (disk_super->label[BTRFS_LABEL_SIZE - 1])
 913                         disk_super->label[BTRFS_LABEL_SIZE - 1] = '\0';
 914                 printk(KERN_INFO "device label %s ", disk_super->label);
 915         } else {
 916                 printk(KERN_INFO "device fsid %pU ", disk_super->fsid);
 917         }
 918
 919         printk(KERN_CONT "devid %llu transid %llu %s\n", devid, transid, path);
 920
 921         ret = device_list_add(path, disk_super, devid, fs_devices_ret);
 922         if (!ret && fs_devices_ret)
 923                 (*fs_devices_ret)->total_devices = total_devices;
 924
 925 error_unmap:
 926         kunmap(page);
 927         page_cache_release(page);
 928
 929 error_bdev_put:
 930         blkdev_put(bdev, flags);
 931 error:
 932         mutex_unlock(&uuid_mutex);
 933         return ret;
 934 }
 935
 936 /* helper to account the used device space in the range */
 937 int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
 938                                    u64 end, u64 *length)
 939 {
 940         struct btrfs_key key;
 941         struct btrfs_root *root = device->dev_root;
 942         struct btrfs_dev_extent *dev_extent;
 943         struct btrfs_path *path;
 944         u64 extent_end;
 945         int ret;
 946         int slot;
 947         struct extent_buffer *l;
 948
 949         *length = 0;
 950
 951         if (start >= device->total_bytes || device->is_tgtdev_for_dev_replace)
 952                 return 0;
 953
 954         path = btrfs_alloc_path();
 955         if (!path)
 956                 return -ENOMEM;
 957         path->reada = 2;
 958
 959         key.objectid = device->devid;
 960         key.offset = start;
 961         key.type = BTRFS_DEV_EXTENT_KEY;
 962
 963         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 964         if (ret < 0)
 965                 goto out;
 966         if (ret > 0) {
 967                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
 968                 if (ret < 0)
 969                         goto out;
 970         }
 971
 972         while (1) {
 973                 l = path->nodes[0];
 974                 slot = path->slots[0];
 975                 if (slot >= btrfs_header_nritems(l)) {
 976                         ret = btrfs_next_leaf(root, path);
 977                         if (ret == 0)
 978                                 continue;
 979                         if (ret < 0)
 980                                 goto out;
 981
 982                         break;
 983                 }
 984                 btrfs_item_key_to_cpu(l, &key, slot);
 985
 986                 if (key.objectid < device->devid)
 987                         goto next;
 988
 989                 if (key.objectid > device->devid)
 990                         break;
 991
 992                 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
 993                         goto next;
 994
 995                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
 996                 extent_end = key.offset + btrfs_dev_extent_length(l,
 997                                                                   dev_extent);
 998                 if (key.offset <= start && extent_end > end) {
 999                         *length = end - start + 1;
1000                         break;
1001                 } else if (key.offset <= start && extent_end > start)
1002                         *length += extent_end - start;
1003                 else if (key.offset > start && extent_end <= end)
1004                         *length += extent_end - key.offset;
1005                 else if (key.offset > start && key.offset <= end) {
1006                         *length += end - key.offset + 1;
1007                         break;
1008                 } else if (key.offset > end)
1009                         break;
1010
1011 next:
1012                 path->slots[0]++;
1013         }
1014         ret = 0;
1015 out:
1016         btrfs_free_path(path);
1017         return ret;
1018 }
1019
1020 static int contains_pending_extent(struct btrfs_trans_handle *trans,
1021                                    struct btrfs_device *device,
1022                                    u64 *start, u64 len)
1023 {
1024         struct extent_map *em;
1025         int ret = 0;
1026
1027         list_for_each_entry(em, &trans->transaction->pending_chunks, list) {
1028                 struct map_lookup *map;
1029                 int i;
1030
1031                 map = (struct map_lookup *)em->bdev;
1032                 for (i = 0; i < map->num_stripes; i++) {
1033                         if (map->stripes[i].dev != device)
1034                                 continue;
1035                         if (map->stripes[i].physical >= *start + len ||
1036                             map->stripes[i].physical + em->orig_block_len <=
1037                             *start)
1038                                 continue;
1039                         *start = map->stripes[i].physical +
1040                                 em->orig_block_len;
1041                         ret = 1;
1042                 }
1043         }
1044
1045         return ret;
1046 }
1047
1048
1049 /*
1050  * find_free_dev_extent - find free space in the specified device
1051  * @device:     the device which we search the free space in
1052  * @num_bytes:  the size of the free space that we need
1053  * @start:      store the start of the free space.
1054  * @len:        the size of the free space. that we find, or the size of the max
1055  *              free space if we don't find suitable free space
1056  *
1057  * this uses a pretty simple search, the expectation is that it is
1058  * called very infrequently and that a given device has a small number
1059  * of extents
1060  *
1061  * @start is used to store the start of the free space if we find. But if we
1062  * don't find suitable free space, it will be used to store the start position
1063  * of the max free space.
1064  *
1065  * @len is used to store the size of the free space that we find.
1066  * But if we don't find suitable free space, it is used to store the size of
1067  * the max free space.
1068  */
1069 int find_free_dev_extent(struct btrfs_trans_handle *trans,
1070                          struct btrfs_device *device, u64 num_bytes,
1071                          u64 *start, u64 *len)
1072 {
1073         struct btrfs_key key;
1074         struct btrfs_root *root = device->dev_root;
1075         struct btrfs_dev_extent *dev_extent;
1076         struct btrfs_path *path;
1077         u64 hole_size;
1078         u64 max_hole_start;
1079         u64 max_hole_size;
1080         u64 extent_end;
1081         u64 search_start;
1082         u64 search_end = device->total_bytes;
1083         int ret;
1084         int slot;
1085         struct extent_buffer *l;
1086
1087         /* FIXME use last free of some kind */
1088
1089         /* we don't want to overwrite the superblock on the drive,
1090          * so we make sure to start at an offset of at least 1MB
1091          */
1092         search_start = max(root->fs_info->alloc_start, 1024ull * 1024);
1093
1094         path = btrfs_alloc_path();
1095         if (!path)
1096                 return -ENOMEM;
1097 again:
1098         max_hole_start = search_start;
1099         max_hole_size = 0;
1100         hole_size = 0;
1101
1102         if (search_start >= search_end || device->is_tgtdev_for_dev_replace) {
1103                 ret = -ENOSPC;
1104                 goto out;
1105         }
1106
1107         path->reada = 2;
1108         path->search_commit_root = 1;
1109         path->skip_locking = 1;
1110
1111         key.objectid = device->devid;
1112         key.offset = search_start;
1113         key.type = BTRFS_DEV_EXTENT_KEY;
1114
1115         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1116         if (ret < 0)
1117                 goto out;
1118         if (ret > 0) {
1119                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1120                 if (ret < 0)
1121                         goto out;
1122         }
1123
1124         while (1) {
1125                 l = path->nodes[0];
1126                 slot = path->slots[0];
1127                 if (slot >= btrfs_header_nritems(l)) {
1128                         ret = btrfs_next_leaf(root, path);
1129                         if (ret == 0)
1130                                 continue;
1131                         if (ret < 0)
1132                                 goto out;
1133
1134                         break;
1135                 }
1136                 btrfs_item_key_to_cpu(l, &key, slot);
1137
1138                 if (key.objectid < device->devid)
1139                         goto next;
1140
1141                 if (key.objectid > device->devid)
1142                         break;
1143
1144                 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
1145                         goto next;
1146
1147                 if (key.offset > search_start) {
1148                         hole_size = key.offset - search_start;
1149
1150                         /*
1151                          * Have to check before we set max_hole_start, otherwise
1152                          * we could end up sending back this offset anyway.
1153                          */
1154                         if (contains_pending_extent(trans, device,
1155                                                     &search_start,
1156                                                     hole_size))
1157                                 hole_size = 0;
1158
1159                         if (hole_size > max_hole_size) {
1160                                 max_hole_start = search_start;
1161                                 max_hole_size = hole_size;
1162                         }
1163
1164                         /*
1165                          * If this free space is greater than which we need,
1166                          * it must be the max free space that we have found
1167                          * until now, so max_hole_start must point to the start
1168                          * of this free space and the length of this free space
1169                          * is stored in max_hole_size. Thus, we return
1170                          * max_hole_start and max_hole_size and go back to the
1171                          * caller.
1172                          */
1173                         if (hole_size >= num_bytes) {
1174                                 ret = 0;
1175                                 goto out;
1176                         }
1177                 }
1178
1179                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1180                 extent_end = key.offset + btrfs_dev_extent_length(l,
1181                                                                   dev_extent);
1182                 if (extent_end > search_start)
1183                         search_start = extent_end;
1184 next:
1185                 path->slots[0]++;
1186                 cond_resched();
1187         }
1188
1189         /*
1190          * At this point, search_start should be the end of
1191          * allocated dev extents, and when shrinking the device,
1192          * search_end may be smaller than search_start.
1193          */
1194         if (search_end > search_start)
1195                 hole_size = search_end - search_start;
1196
1197         if (hole_size > max_hole_size) {
1198                 max_hole_start = search_start;
1199                 max_hole_size = hole_size;
1200         }
1201
1202         if (contains_pending_extent(trans, device, &search_start, hole_size)) {
1203                 btrfs_release_path(path);
1204                 goto again;
1205         }
1206
1207         /* See above. */
1208         if (hole_size < num_bytes)
1209                 ret = -ENOSPC;
1210         else
1211                 ret = 0;
1212
1213 out:
1214         btrfs_free_path(path);
1215         *start = max_hole_start;
1216         if (len)
1217                 *len = max_hole_size;
1218         return ret;
1219 }
1220
1221 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1222                           struct btrfs_device *device,
1223                           u64 start)
1224 {
1225         int ret;
1226         struct btrfs_path *path;
1227         struct btrfs_root *root = device->dev_root;
1228         struct btrfs_key key;
1229         struct btrfs_key found_key;
1230         struct extent_buffer *leaf = NULL;
1231         struct btrfs_dev_extent *extent = NULL;
1232
1233         path = btrfs_alloc_path();
1234         if (!path)
1235                 return -ENOMEM;
1236
1237         key.objectid = device->devid;
1238         key.offset = start;
1239         key.type = BTRFS_DEV_EXTENT_KEY;
1240 again:
1241         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1242         if (ret > 0) {
1243                 ret = btrfs_previous_item(root, path, key.objectid,
1244                                           BTRFS_DEV_EXTENT_KEY);
1245                 if (ret)
1246                         goto out;
1247                 leaf = path->nodes[0];
1248                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1249                 extent = btrfs_item_ptr(leaf, path->slots[0],
1250                                         struct btrfs_dev_extent);
1251                 BUG_ON(found_key.offset > start || found_key.offset +
1252                        btrfs_dev_extent_length(leaf, extent) < start);
1253                 key = found_key;
1254                 btrfs_release_path(path);
1255                 goto again;
1256         } else if (ret == 0) {
1257                 leaf = path->nodes[0];
1258                 extent = btrfs_item_ptr(leaf, path->slots[0],
1259                                         struct btrfs_dev_extent);
1260         } else {
1261                 btrfs_error(root->fs_info, ret, "Slot search failed");
1262                 goto out;
1263         }
1264
1265         if (device->bytes_used > 0) {
1266                 u64 len = btrfs_dev_extent_length(leaf, extent);
1267                 device->bytes_used -= len;
1268                 spin_lock(&root->fs_info->free_chunk_lock);
1269                 root->fs_info->free_chunk_space += len;
1270                 spin_unlock(&root->fs_info->free_chunk_lock);
1271         }
1272         ret = btrfs_del_item(trans, root, path);
1273         if (ret) {
1274                 btrfs_error(root->fs_info, ret,
1275                             "Failed to remove dev extent item");
1276         }
1277 out:
1278         btrfs_free_path(path);
1279         return ret;
1280 }
1281
1282 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1283                                   struct btrfs_device *device,
1284                                   u64 chunk_tree, u64 chunk_objectid,
1285                                   u64 chunk_offset, u64 start, u64 num_bytes)
1286 {
1287         int ret;
1288         struct btrfs_path *path;
1289         struct btrfs_root *root = device->dev_root;
1290         struct btrfs_dev_extent *extent;
1291         struct extent_buffer *leaf;
1292         struct btrfs_key key;
1293
1294         WARN_ON(!device->in_fs_metadata);
1295         WARN_ON(device->is_tgtdev_for_dev_replace);
1296         path = btrfs_alloc_path();
1297         if (!path)
1298                 return -ENOMEM;
1299
1300         key.objectid = device->devid;
1301         key.offset = start;
1302         key.type = BTRFS_DEV_EXTENT_KEY;
1303         ret = btrfs_insert_empty_item(trans, root, path, &key,
1304                                       sizeof(*extent));
1305         if (ret)
1306                 goto out;
1307
1308         leaf = path->nodes[0];
1309         extent = btrfs_item_ptr(leaf, path->slots[0],
1310                                 struct btrfs_dev_extent);
1311         btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
1312         btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
1313         btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1314
1315         write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
1316                     btrfs_dev_extent_chunk_tree_uuid(extent), BTRFS_UUID_SIZE);
1317
1318         btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1319         btrfs_mark_buffer_dirty(leaf);
1320 out:
1321         btrfs_free_path(path);
1322         return ret;
1323 }
1324
1325 static u64 find_next_chunk(struct btrfs_fs_info *fs_info)
1326 {
1327         struct extent_map_tree *em_tree;
1328         struct extent_map *em;
1329         struct rb_node *n;
1330         u64 ret = 0;
1331
1332         em_tree = &fs_info->mapping_tree.map_tree;
1333         read_lock(&em_tree->lock);
1334         n = rb_last(&em_tree->map);
1335         if (n) {
1336                 em = rb_entry(n, struct extent_map, rb_node);
1337                 ret = em->start + em->len;
1338         }
1339         read_unlock(&em_tree->lock);
1340
1341         return ret;
1342 }
1343
1344 static noinline int find_next_devid(struct btrfs_fs_info *fs_info,
1345                                     u64 *devid_ret)
1346 {
1347         int ret;
1348         struct btrfs_key key;
1349         struct btrfs_key found_key;
1350         struct btrfs_path *path;
1351
1352         path = btrfs_alloc_path();
1353         if (!path)
1354                 return -ENOMEM;
1355
1356         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1357         key.type = BTRFS_DEV_ITEM_KEY;
1358         key.offset = (u64)-1;
1359
1360         ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
1361         if (ret < 0)
1362                 goto error;
1363
1364         BUG_ON(ret == 0); /* Corruption */
1365
1366         ret = btrfs_previous_item(fs_info->chunk_root, path,
1367                                   BTRFS_DEV_ITEMS_OBJECTID,
1368                                   BTRFS_DEV_ITEM_KEY);
1369         if (ret) {
1370                 *devid_ret = 1;
1371         } else {
1372                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1373                                       path->slots[0]);
1374                 *devid_ret = found_key.offset + 1;
1375         }
1376         ret = 0;
1377 error:
1378         btrfs_free_path(path);
1379         return ret;
1380 }
1381
1382 /*
1383  * the device information is stored in the chunk root
1384  * the btrfs_device struct should be fully filled in
1385  */
1386 static int btrfs_add_device(struct btrfs_trans_handle *trans,
1387                             struct btrfs_root *root,
1388                             struct btrfs_device *device)
1389 {
1390         int ret;
1391         struct btrfs_path *path;
1392         struct btrfs_dev_item *dev_item;
1393         struct extent_buffer *leaf;
1394         struct btrfs_key key;
1395         unsigned long ptr;
1396
1397         root = root->fs_info->chunk_root;
1398
1399         path = btrfs_alloc_path();
1400         if (!path)
1401                 return -ENOMEM;
1402
1403         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1404         key.type = BTRFS_DEV_ITEM_KEY;
1405         key.offset = device->devid;
1406
1407         ret = btrfs_insert_empty_item(trans, root, path, &key,
1408                                       sizeof(*dev_item));
1409         if (ret)
1410                 goto out;
1411
1412         leaf = path->nodes[0];
1413         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1414
1415         btrfs_set_device_id(leaf, dev_item, device->devid);
1416         btrfs_set_device_generation(leaf, dev_item, 0);
1417         btrfs_set_device_type(leaf, dev_item, device->type);
1418         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1419         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1420         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1421         btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
1422         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1423         btrfs_set_device_group(leaf, dev_item, 0);
1424         btrfs_set_device_seek_speed(leaf, dev_item, 0);
1425         btrfs_set_device_bandwidth(leaf, dev_item, 0);
1426         btrfs_set_device_start_offset(leaf, dev_item, 0);
1427
1428         ptr = btrfs_device_uuid(dev_item);
1429         write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1430         ptr = btrfs_device_fsid(dev_item);
1431         write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1432         btrfs_mark_buffer_dirty(leaf);
1433
1434         ret = 0;
1435 out:
1436         btrfs_free_path(path);
1437         return ret;
1438 }
1439
1440 static int btrfs_rm_dev_item(struct btrfs_root *root,
1441                              struct btrfs_device *device)
1442 {
1443         int ret;
1444         struct btrfs_path *path;
1445         struct btrfs_key key;
1446         struct btrfs_trans_handle *trans;
1447
1448         root = root->fs_info->chunk_root;
1449
1450         path = btrfs_alloc_path();
1451         if (!path)
1452                 return -ENOMEM;
1453
1454         trans = btrfs_start_transaction(root, 0);
1455         if (IS_ERR(trans)) {
1456                 btrfs_free_path(path);
1457                 return PTR_ERR(trans);
1458         }
1459         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1460         key.type = BTRFS_DEV_ITEM_KEY;
1461         key.offset = device->devid;
1462         lock_chunks(root);
1463
1464         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1465         if (ret < 0)
1466                 goto out;
1467
1468         if (ret > 0) {
1469                 ret = -ENOENT;
1470                 goto out;
1471         }
1472
1473         ret = btrfs_del_item(trans, root, path);
1474         if (ret)
1475                 goto out;
1476 out:
1477         btrfs_free_path(path);
1478         unlock_chunks(root);
1479         btrfs_commit_transaction(trans, root);
1480         return ret;
1481 }
1482
1483 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
1484 {
1485         struct btrfs_device *device;
1486         struct btrfs_device *next_device;
1487         struct block_device *bdev;
1488         struct buffer_head *bh = NULL;
1489         struct btrfs_super_block *disk_super;
1490         struct btrfs_fs_devices *cur_devices;
1491         u64 all_avail;
1492         u64 devid;
1493         u64 num_devices;
1494         u8 *dev_uuid;
1495         unsigned seq;
1496         int ret = 0;
1497         bool clear_super = false;
1498
1499         mutex_lock(&uuid_mutex);
1500
1501         do {
1502                 seq = read_seqbegin(&root->fs_info->profiles_lock);
1503
1504                 all_avail = root->fs_info->avail_data_alloc_bits |
1505                             root->fs_info->avail_system_alloc_bits |
1506                             root->fs_info->avail_metadata_alloc_bits;
1507         } while (read_seqretry(&root->fs_info->profiles_lock, seq));
1508
1509         num_devices = root->fs_info->fs_devices->num_devices;
1510         btrfs_dev_replace_lock(&root->fs_info->dev_replace);
1511         if (btrfs_dev_replace_is_ongoing(&root->fs_info->dev_replace)) {
1512                 WARN_ON(num_devices < 1);
1513                 num_devices--;
1514         }
1515         btrfs_dev_replace_unlock(&root->fs_info->dev_replace);
1516
1517         if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) && num_devices <= 4) {
1518                 ret = BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET;
1519                 goto out;
1520         }
1521
1522         if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) && num_devices <= 2) {
1523                 ret = BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET;
1524                 goto out;
1525         }
1526
1527         if ((all_avail & BTRFS_BLOCK_GROUP_RAID5) &&
1528             root->fs_info->fs_devices->rw_devices <= 2) {
1529                 ret = BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET;
1530                 goto out;
1531         }
1532         if ((all_avail & BTRFS_BLOCK_GROUP_RAID6) &&
1533             root->fs_info->fs_devices->rw_devices <= 3) {
1534                 ret = BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET;
1535                 goto out;
1536         }
1537
1538         if (strcmp(device_path, "missing") == 0) {
1539                 struct list_head *devices;
1540                 struct btrfs_device *tmp;
1541
1542                 device = NULL;
1543                 devices = &root->fs_info->fs_devices->devices;
1544                 /*
1545                  * It is safe to read the devices since the volume_mutex
1546                  * is held.
1547                  */
1548                 list_for_each_entry(tmp, devices, dev_list) {
1549                         if (tmp->in_fs_metadata &&
1550                             !tmp->is_tgtdev_for_dev_replace &&
1551                             !tmp->bdev) {
1552                                 device = tmp;
1553                                 break;
1554                         }
1555                 }
1556                 bdev = NULL;
1557                 bh = NULL;
1558                 disk_super = NULL;
1559                 if (!device) {
1560                         ret = BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
1561                         goto out;
1562                 }
1563         } else {
1564                 ret = btrfs_get_bdev_and_sb(device_path,
1565                                             FMODE_WRITE | FMODE_EXCL,
1566                                             root->fs_info->bdev_holder, 0,
1567                                             &bdev, &bh);
1568                 if (ret)
1569                         goto out;
1570                 disk_super = (struct btrfs_super_block *)bh->b_data;
1571                 devid = btrfs_stack_device_id(&disk_super->dev_item);
1572                 dev_uuid = disk_super->dev_item.uuid;
1573                 device = btrfs_find_device(root->fs_info, devid, dev_uuid,
1574                                            disk_super->fsid);
1575                 if (!device) {
1576                         ret = -ENOENT;
1577                         goto error_brelse;
1578                 }
1579         }
1580
1581         if (device->is_tgtdev_for_dev_replace) {
1582                 ret = BTRFS_ERROR_DEV_TGT_REPLACE;
1583                 goto error_brelse;
1584         }
1585
1586         if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1587                 ret = BTRFS_ERROR_DEV_ONLY_WRITABLE;
1588                 goto error_brelse;
1589         }
1590
1591         if (device->writeable) {
1592                 lock_chunks(root);
1593                 list_del_init(&device->dev_alloc_list);
1594                 unlock_chunks(root);
1595                 root->fs_info->fs_devices->rw_devices--;
1596                 clear_super = true;
1597         }
1598
1599         mutex_unlock(&uuid_mutex);
1600         ret = btrfs_shrink_device(device, 0);
1601         mutex_lock(&uuid_mutex);
1602         if (ret)
1603                 goto error_undo;
1604
1605         /*
1606          * TODO: the superblock still includes this device in its num_devices
1607          * counter although write_all_supers() is not locked out. This
1608          * could give a filesystem state which requires a degraded mount.
1609          */
1610         ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1611         if (ret)
1612                 goto error_undo;
1613
1614         spin_lock(&root->fs_info->free_chunk_lock);
1615         root->fs_info->free_chunk_space = device->total_bytes -
1616                 device->bytes_used;
1617         spin_unlock(&root->fs_info->free_chunk_lock);
1618
1619         device->in_fs_metadata = 0;
1620         btrfs_scrub_cancel_dev(root->fs_info, device);
1621
1622         /*
1623          * the device list mutex makes sure that we don't change
1624          * the device list while someone else is writing out all
1625          * the device supers. Whoever is writing all supers, should
1626          * lock the device list mutex before getting the number of
1627          * devices in the super block (super_copy). Conversely,
1628          * whoever updates the number of devices in the super block
1629          * (super_copy) should hold the device list mutex.
1630          */
1631
1632         cur_devices = device->fs_devices;
1633         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1634         list_del_rcu(&device->dev_list);
1635
1636         device->fs_devices->num_devices--;
1637         device->fs_devices->total_devices--;
1638
1639         if (device->missing)
1640                 root->fs_info->fs_devices->missing_devices--;
1641
1642         next_device = list_entry(root->fs_info->fs_devices->devices.next,
1643                                  struct btrfs_device, dev_list);
1644         if (device->bdev == root->fs_info->sb->s_bdev)
1645                 root->fs_info->sb->s_bdev = next_device->bdev;
1646         if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1647                 root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1648
1649         if (device->bdev)
1650                 device->fs_devices->open_devices--;
1651
1652         call_rcu(&device->rcu, free_device);
1653
1654         num_devices = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
1655         btrfs_set_super_num_devices(root->fs_info->super_copy, num_devices);
1656         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1657
1658         if (cur_devices->open_devices == 0) {
1659                 struct btrfs_fs_devices *fs_devices;
1660                 fs_devices = root->fs_info->fs_devices;
1661                 while (fs_devices) {
1662                         if (fs_devices->seed == cur_devices)
1663                                 break;
1664                         fs_devices = fs_devices->seed;
1665                 }
1666                 fs_devices->seed = cur_devices->seed;
1667                 cur_devices->seed = NULL;
1668                 lock_chunks(root);
1669                 __btrfs_close_devices(cur_devices);
1670                 unlock_chunks(root);
1671                 free_fs_devices(cur_devices);
1672         }
1673
1674         root->fs_info->num_tolerated_disk_barrier_failures =
1675                 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
1676
1677         /*
1678          * at this point, the device is zero sized.  We want to
1679          * remove it from the devices list and zero out the old super
1680          */
1681         if (clear_super && disk_super) {
1682                 /* make sure this device isn't detected as part of
1683                  * the FS anymore
1684                  */
1685                 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1686                 set_buffer_dirty(bh);
1687                 sync_dirty_buffer(bh);
1688         }
1689
1690         ret = 0;
1691
1692         /* Notify udev that device has changed */
1693         if (bdev)
1694                 btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
1695
1696 error_brelse:
1697         brelse(bh);
1698         if (bdev)
1699                 blkdev_put(bdev, FMODE_READ | FMODE_EXCL);
1700 out:
1701         mutex_unlock(&uuid_mutex);
1702         return ret;
1703 error_undo:
1704         if (device->writeable) {
1705                 lock_chunks(root);
1706                 list_add(&device->dev_alloc_list,
1707                          &root->fs_info->fs_devices->alloc_list);
1708                 unlock_chunks(root);
1709                 root->fs_info->fs_devices->rw_devices++;
1710         }
1711         goto error_brelse;
1712 }
1713
1714 void btrfs_rm_dev_replace_srcdev(struct btrfs_fs_info *fs_info,
1715                                  struct btrfs_device *srcdev)
1716 {
1717         WARN_ON(!mutex_is_locked(&fs_info->fs_devices->device_list_mutex));
1718         list_del_rcu(&srcdev->dev_list);
1719         list_del_rcu(&srcdev->dev_alloc_list);
1720         fs_info->fs_devices->num_devices--;
1721         if (srcdev->missing) {
1722                 fs_info->fs_devices->missing_devices--;
1723                 fs_info->fs_devices->rw_devices++;
1724         }
1725         if (srcdev->can_discard)
1726                 fs_info->fs_devices->num_can_discard--;
1727         if (srcdev->bdev)
1728                 fs_info->fs_devices->open_devices--;
1729
1730         call_rcu(&srcdev->rcu, free_device);
1731 }
1732
1733 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
1734                                       struct btrfs_device *tgtdev)
1735 {
1736         struct btrfs_device *next_device;
1737
1738         WARN_ON(!tgtdev);
1739         mutex_lock(&fs_info->fs_devices->device_list_mutex);
1740         if (tgtdev->bdev) {
1741                 btrfs_scratch_superblock(tgtdev);
1742                 fs_info->fs_devices->open_devices--;
1743         }
1744         fs_info->fs_devices->num_devices--;
1745         if (tgtdev->can_discard)
1746                 fs_info->fs_devices->num_can_discard++;
1747
1748         next_device = list_entry(fs_info->fs_devices->devices.next,
1749                                  struct btrfs_device, dev_list);
1750         if (tgtdev->bdev == fs_info->sb->s_bdev)
1751                 fs_info->sb->s_bdev = next_device->bdev;
1752         if (tgtdev->bdev == fs_info->fs_devices->latest_bdev)
1753                 fs_info->fs_devices->latest_bdev = next_device->bdev;
1754         list_del_rcu(&tgtdev->dev_list);
1755
1756         call_rcu(&tgtdev->rcu, free_device);
1757
1758         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
1759 }
1760
1761 static int btrfs_find_device_by_path(struct btrfs_root *root, char *device_path,
1762                                      struct btrfs_device **device)
1763 {
1764         int ret = 0;
1765         struct btrfs_super_block *disk_super;
1766         u64 devid;
1767         u8 *dev_uuid;
1768         struct block_device *bdev;
1769         struct buffer_head *bh;
1770
1771         *device = NULL;
1772         ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
1773                                     root->fs_info->bdev_holder, 0, &bdev, &bh);
1774         if (ret)
1775                 return ret;
1776         disk_super = (struct btrfs_super_block *)bh->b_data;
1777         devid = btrfs_stack_device_id(&disk_super->dev_item);
1778         dev_uuid = disk_super->dev_item.uuid;
1779         *device = btrfs_find_device(root->fs_info, devid, dev_uuid,
1780                                     disk_super->fsid);
1781         brelse(bh);
1782         if (!*device)
1783                 ret = -ENOENT;
1784         blkdev_put(bdev, FMODE_READ);
1785         return ret;
1786 }
1787
1788 int btrfs_find_device_missing_or_by_path(struct btrfs_root *root,
1789                                          char *device_path,
1790                                          struct btrfs_device **device)
1791 {
1792         *device = NULL;
1793         if (strcmp(device_path, "missing") == 0) {
1794                 struct list_head *devices;
1795                 struct btrfs_device *tmp;
1796
1797                 devices = &root->fs_info->fs_devices->devices;
1798                 /*
1799                  * It is safe to read the devices since the volume_mutex
1800                  * is held by the caller.
1801                  */
1802                 list_for_each_entry(tmp, devices, dev_list) {
1803                         if (tmp->in_fs_metadata && !tmp->bdev) {
1804                                 *device = tmp;
1805                                 break;
1806                         }
1807                 }
1808
1809                 if (!*device) {
1810                         pr_err("btrfs: no missing device found\n");
1811                         return -ENOENT;
1812                 }
1813
1814                 return 0;
1815         } else {
1816                 return btrfs_find_device_by_path(root, device_path, device);
1817         }
1818 }
1819
1820 /*
1821  * does all the dirty work required for changing file system's UUID.
1822  */
1823 static int btrfs_prepare_sprout(struct btrfs_root *root)
1824 {
1825         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
1826         struct btrfs_fs_devices *old_devices;
1827         struct btrfs_fs_devices *seed_devices;
1828         struct btrfs_super_block *disk_super = root->fs_info->super_copy;
1829         struct btrfs_device *device;
1830         u64 super_flags;
1831
1832         BUG_ON(!mutex_is_locked(&uuid_mutex));
1833         if (!fs_devices->seeding)
1834                 return -EINVAL;
1835
1836         seed_devices = __alloc_fs_devices();
1837         if (IS_ERR(seed_devices))
1838                 return PTR_ERR(seed_devices);
1839
1840         old_devices = clone_fs_devices(fs_devices);
1841         if (IS_ERR(old_devices)) {
1842                 kfree(seed_devices);
1843                 return PTR_ERR(old_devices);
1844         }
1845
1846         list_add(&old_devices->list, &fs_uuids);
1847
1848         memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
1849         seed_devices->opened = 1;
1850         INIT_LIST_HEAD(&seed_devices->devices);
1851         INIT_LIST_HEAD(&seed_devices->alloc_list);
1852         mutex_init(&seed_devices->device_list_mutex);
1853
1854         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1855         list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
1856                               synchronize_rcu);
1857
1858         list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
1859         list_for_each_entry(device, &seed_devices->devices, dev_list) {
1860                 device->fs_devices = seed_devices;
1861         }
1862
1863         fs_devices->seeding = 0;
1864         fs_devices->num_devices = 0;
1865         fs_devices->open_devices = 0;
1866         fs_devices->total_devices = 0;
1867         fs_devices->seed = seed_devices;
1868
1869         generate_random_uuid(fs_devices->fsid);
1870         memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1871         memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1872         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1873
1874         super_flags = btrfs_super_flags(disk_super) &
1875                       ~BTRFS_SUPER_FLAG_SEEDING;
1876         btrfs_set_super_flags(disk_super, super_flags);
1877
1878         return 0;
1879 }
1880
1881 /*
1882  * strore the expected generation for seed devices in device items.
1883  */
1884 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
1885                                struct btrfs_root *root)
1886 {
1887         struct btrfs_path *path;
1888         struct extent_buffer *leaf;
1889         struct btrfs_dev_item *dev_item;
1890         struct btrfs_device *device;
1891         struct btrfs_key key;
1892         u8 fs_uuid[BTRFS_UUID_SIZE];
1893         u8 dev_uuid[BTRFS_UUID_SIZE];
1894         u64 devid;
1895         int ret;
1896
1897         path = btrfs_alloc_path();
1898         if (!path)
1899                 return -ENOMEM;
1900
1901         root = root->fs_info->chunk_root;
1902         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1903         key.offset = 0;
1904         key.type = BTRFS_DEV_ITEM_KEY;
1905
1906         while (1) {
1907                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1908                 if (ret < 0)
1909                         goto error;
1910
1911                 leaf = path->nodes[0];
1912 next_slot:
1913                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1914                         ret = btrfs_next_leaf(root, path);
1915                         if (ret > 0)
1916                                 break;
1917                         if (ret < 0)
1918                                 goto error;
1919                         leaf = path->nodes[0];
1920                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1921                         btrfs_release_path(path);
1922                         continue;
1923                 }
1924
1925                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1926                 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
1927                     key.type != BTRFS_DEV_ITEM_KEY)
1928                         break;
1929
1930                 dev_item = btrfs_item_ptr(leaf, path->slots[0],
1931                                           struct btrfs_dev_item);
1932                 devid = btrfs_device_id(leaf, dev_item);
1933                 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
1934                                    BTRFS_UUID_SIZE);
1935                 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
1936                                    BTRFS_UUID_SIZE);
1937                 device = btrfs_find_device(root->fs_info, devid, dev_uuid,
1938                                            fs_uuid);
1939                 BUG_ON(!device); /* Logic error */
1940
1941                 if (device->fs_devices->seeding) {
1942                         btrfs_set_device_generation(leaf, dev_item,
1943                                                     device->generation);
1944                         btrfs_mark_buffer_dirty(leaf);
1945                 }
1946
1947                 path->slots[0]++;
1948                 goto next_slot;
1949         }
1950         ret = 0;
1951 error:
1952         btrfs_free_path(path);
1953         return ret;
1954 }
1955
1956 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
1957 {
1958         struct request_queue *q;
1959         struct btrfs_trans_handle *trans;
1960         struct btrfs_device *device;
1961         struct block_device *bdev;
1962         struct list_head *devices;
1963         struct super_block *sb = root->fs_info->sb;
1964         struct rcu_string *name;
1965         u64 total_bytes;
1966         int seeding_dev = 0;
1967         int ret = 0;
1968
1969         if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
1970                 return -EROFS;
1971
1972         bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
1973                                   root->fs_info->bdev_holder);
1974         if (IS_ERR(bdev))
1975                 return PTR_ERR(bdev);
1976
1977         if (root->fs_info->fs_devices->seeding) {
1978                 seeding_dev = 1;
1979                 down_write(&sb->s_umount);
1980                 mutex_lock(&uuid_mutex);
1981         }
1982
1983         filemap_write_and_wait(bdev->bd_inode->i_mapping);
1984
1985         devices = &root->fs_info->fs_devices->devices;
1986
1987         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1988         list_for_each_entry(device, devices, dev_list) {
1989                 if (device->bdev == bdev) {
1990                         ret = -EEXIST;
1991                         mutex_unlock(
1992                                 &root->fs_info->fs_devices->device_list_mutex);
1993                         goto error;
1994                 }
1995         }
1996         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1997
1998         device = btrfs_alloc_device(root->fs_info, NULL, NULL);
1999         if (IS_ERR(device)) {
2000                 /* we can safely leave the fs_devices entry around */
2001                 ret = PTR_ERR(device);
2002                 goto error;
2003         }
2004
2005         name = rcu_string_strdup(device_path, GFP_NOFS);
2006         if (!name) {
2007                 kfree(device);
2008                 ret = -ENOMEM;
2009                 goto error;
2010         }
2011         rcu_assign_pointer(device->name, name);
2012
2013         trans = btrfs_start_transaction(root, 0);
2014         if (IS_ERR(trans)) {
2015                 rcu_string_free(device->name);
2016                 kfree(device);
2017                 ret = PTR_ERR(trans);
2018                 goto error;
2019         }
2020
2021         lock_chunks(root);
2022
2023         q = bdev_get_queue(bdev);
2024         if (blk_queue_discard(q))
2025                 device->can_discard = 1;
2026         device->writeable = 1;
2027         device->generation = trans->transid;
2028         device->io_width = root->sectorsize;
2029         device->io_align = root->sectorsize;
2030         device->sector_size = root->sectorsize;
2031         device->total_bytes = i_size_read(bdev->bd_inode);
2032         device->disk_total_bytes = device->total_bytes;
2033         device->dev_root = root->fs_info->dev_root;
2034         device->bdev = bdev;
2035         device->in_fs_metadata = 1;
2036         device->is_tgtdev_for_dev_replace = 0;
2037         device->mode = FMODE_EXCL;
2038         set_blocksize(device->bdev, 4096);
2039
2040         if (seeding_dev) {
2041                 sb->s_flags &= ~MS_RDONLY;
2042                 ret = btrfs_prepare_sprout(root);
2043                 BUG_ON(ret); /* -ENOMEM */
2044         }
2045
2046         device->fs_devices = root->fs_info->fs_devices;
2047
2048         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2049         list_add_rcu(&device->dev_list, &root->fs_info->fs_devices->devices);
2050         list_add(&device->dev_alloc_list,
2051                  &root->fs_info->fs_devices->alloc_list);
2052         root->fs_info->fs_devices->num_devices++;
2053         root->fs_info->fs_devices->open_devices++;
2054         root->fs_info->fs_devices->rw_devices++;
2055         root->fs_info->fs_devices->total_devices++;
2056         if (device->can_discard)
2057                 root->fs_info->fs_devices->num_can_discard++;
2058         root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
2059
2060         spin_lock(&root->fs_info->free_chunk_lock);
2061         root->fs_info->free_chunk_space += device->total_bytes;
2062         spin_unlock(&root->fs_info->free_chunk_lock);
2063
2064         if (!blk_queue_nonrot(bdev_get_queue(bdev)))
2065                 root->fs_info->fs_devices->rotating = 1;
2066
2067         total_bytes = btrfs_super_total_bytes(root->fs_info->super_copy);
2068         btrfs_set_super_total_bytes(root->fs_info->super_copy,
2069                                     total_bytes + device->total_bytes);
2070
2071         total_bytes = btrfs_super_num_devices(root->fs_info->super_copy);
2072         btrfs_set_super_num_devices(root->fs_info->super_copy,
2073                                     total_bytes + 1);
2074         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2075
2076         if (seeding_dev) {
2077                 ret = init_first_rw_device(trans, root, device);
2078                 if (ret) {
2079                         btrfs_abort_transaction(trans, root, ret);
2080                         goto error_trans;
2081                 }
2082                 ret = btrfs_finish_sprout(trans, root);
2083                 if (ret) {
2084                         btrfs_abort_transaction(trans, root, ret);
2085                         goto error_trans;
2086                 }
2087         } else {
2088                 ret = btrfs_add_device(trans, root, device);
2089                 if (ret) {
2090                         btrfs_abort_transaction(trans, root, ret);
2091                         goto error_trans;
2092                 }
2093         }
2094
2095         /*
2096          * we've got more storage, clear any full flags on the space
2097          * infos
2098          */
2099         btrfs_clear_space_info_full(root->fs_info);
2100
2101         unlock_chunks(root);
2102         root->fs_info->num_tolerated_disk_barrier_failures =
2103                 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
2104         ret = btrfs_commit_transaction(trans, root);
2105
2106         if (seeding_dev) {
2107                 mutex_unlock(&uuid_mutex);
2108                 up_write(&sb->s_umount);
2109
2110                 if (ret) /* transaction commit */
2111                         return ret;
2112
2113                 ret = btrfs_relocate_sys_chunks(root);
2114                 if (ret < 0)
2115                         btrfs_error(root->fs_info, ret,
2116                                     "Failed to relocate sys chunks after "
2117                                     "device initialization. This can be fixed "
2118                                     "using the \"btrfs balance\" command.");
2119                 trans = btrfs_attach_transaction(root);
2120                 if (IS_ERR(trans)) {
2121                         if (PTR_ERR(trans) == -ENOENT)
2122                                 return 0;
2123                         return PTR_ERR(trans);
2124                 }
2125                 ret = btrfs_commit_transaction(trans, root);
2126         }
2127
2128         return ret;
2129
2130 error_trans:
2131         unlock_chunks(root);
2132         btrfs_end_transaction(trans, root);
2133         rcu_string_free(device->name);
2134         kfree(device);
2135 error:
2136         blkdev_put(bdev, FMODE_EXCL);
2137         if (seeding_dev) {
2138                 mutex_unlock(&uuid_mutex);
2139                 up_write(&sb->s_umount);
2140         }
2141         return ret;
2142 }
2143
2144 int btrfs_init_dev_replace_tgtdev(struct btrfs_root *root, char *device_path,
2145                                   struct btrfs_device **device_out)
2146 {
2147         struct request_queue *q;
2148         struct btrfs_device *device;
2149         struct block_device *bdev;
2150         struct btrfs_fs_info *fs_info = root->fs_info;
2151         struct list_head *devices;
2152         struct rcu_string *name;
2153         u64 devid = BTRFS_DEV_REPLACE_DEVID;
2154         int ret = 0;
2155
2156         *device_out = NULL;
2157         if (fs_info->fs_devices->seeding)
2158                 return -EINVAL;
2159
2160         bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2161                                   fs_info->bdev_holder);
2162         if (IS_ERR(bdev))
2163                 return PTR_ERR(bdev);
2164
2165         filemap_write_and_wait(bdev->bd_inode->i_mapping);
2166
2167         devices = &fs_info->fs_devices->devices;
2168         list_for_each_entry(device, devices, dev_list) {
2169                 if (device->bdev == bdev) {
2170                         ret = -EEXIST;
2171                         goto error;
2172                 }
2173         }
2174
2175         device = btrfs_alloc_device(NULL, &devid, NULL);
2176         if (IS_ERR(device)) {
2177                 ret = PTR_ERR(device);
2178                 goto error;
2179         }
2180
2181         name = rcu_string_strdup(device_path, GFP_NOFS);
2182         if (!name) {
2183                 kfree(device);
2184                 ret = -ENOMEM;
2185                 goto error;
2186         }
2187         rcu_assign_pointer(device->name, name);
2188
2189         q = bdev_get_queue(bdev);
2190         if (blk_queue_discard(q))
2191                 device->can_discard = 1;
2192         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2193         device->writeable = 1;
2194         device->generation = 0;
2195         device->io_width = root->sectorsize;
2196         device->io_align = root->sectorsize;
2197         device->sector_size = root->sectorsize;
2198         device->total_bytes = i_size_read(bdev->bd_inode);
2199         device->disk_total_bytes = device->total_bytes;
2200         device->dev_root = fs_info->dev_root;
2201         device->bdev = bdev;
2202         device->in_fs_metadata = 1;
2203         device->is_tgtdev_for_dev_replace = 1;
2204         device->mode = FMODE_EXCL;
2205         set_blocksize(device->bdev, 4096);
2206         device->fs_devices = fs_info->fs_devices;
2207         list_add(&device->dev_list, &fs_info->fs_devices->devices);
2208         fs_info->fs_devices->num_devices++;
2209         fs_info->fs_devices->open_devices++;
2210         if (device->can_discard)
2211                 fs_info->fs_devices->num_can_discard++;
2212         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2213
2214         *device_out = device;
2215         return ret;
2216
2217 error:
2218         blkdev_put(bdev, FMODE_EXCL);
2219         return ret;
2220 }
2221
2222 void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info *fs_info,
2223                                               struct btrfs_device *tgtdev)
2224 {
2225         WARN_ON(fs_info->fs_devices->rw_devices == 0);
2226         tgtdev->io_width = fs_info->dev_root->sectorsize;
2227         tgtdev->io_align = fs_info->dev_root->sectorsize;
2228         tgtdev->sector_size = fs_info->dev_root->sectorsize;
2229         tgtdev->dev_root = fs_info->dev_root;
2230         tgtdev->in_fs_metadata = 1;
2231 }
2232
2233 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2234                                         struct btrfs_device *device)
2235 {
2236         int ret;
2237         struct btrfs_path *path;
2238         struct btrfs_root *root;
2239         struct btrfs_dev_item *dev_item;
2240         struct extent_buffer *leaf;
2241         struct btrfs_key key;
2242
2243         root = device->dev_root->fs_info->chunk_root;
2244
2245         path = btrfs_alloc_path();
2246         if (!path)
2247                 return -ENOMEM;
2248
2249         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2250         key.type = BTRFS_DEV_ITEM_KEY;
2251         key.offset = device->devid;
2252
2253         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2254         if (ret < 0)
2255                 goto out;
2256
2257         if (ret > 0) {
2258                 ret = -ENOENT;
2259                 goto out;
2260         }
2261
2262         leaf = path->nodes[0];
2263         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2264
2265         btrfs_set_device_id(leaf, dev_item, device->devid);
2266         btrfs_set_device_type(leaf, dev_item, device->type);
2267         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2268         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2269         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2270         btrfs_set_device_total_bytes(leaf, dev_item, device->disk_total_bytes);
2271         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
2272         btrfs_mark_buffer_dirty(leaf);
2273
2274 out:
2275         btrfs_free_path(path);
2276         return ret;
2277 }
2278
2279 static int __btrfs_grow_device(struct btrfs_trans_handle *trans,
2280                       struct btrfs_device *device, u64 new_size)
2281 {
2282         struct btrfs_super_block *super_copy =
2283                 device->dev_root->fs_info->super_copy;
2284         u64 old_total = btrfs_super_total_bytes(super_copy);
2285         u64 diff = new_size - device->total_bytes;
2286
2287         if (!device->writeable)
2288                 return -EACCES;
2289         if (new_size <= device->total_bytes ||
2290             device->is_tgtdev_for_dev_replace)
2291                 return -EINVAL;
2292
2293         btrfs_set_super_total_bytes(super_copy, old_total + diff);
2294         device->fs_devices->total_rw_bytes += diff;
2295
2296         device->total_bytes = new_size;
2297         device->disk_total_bytes = new_size;
2298         btrfs_clear_space_info_full(device->dev_root->fs_info);
2299
2300         return btrfs_update_device(trans, device);
2301 }
2302
2303 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2304                       struct btrfs_device *device, u64 new_size)
2305 {
2306         int ret;
2307         lock_chunks(device->dev_root);
2308         ret = __btrfs_grow_device(trans, device, new_size);
2309         unlock_chunks(device->dev_root);
2310         return ret;
2311 }
2312
2313 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
2314                             struct btrfs_root *root,
2315                             u64 chunk_tree, u64 chunk_objectid,
2316                             u64 chunk_offset)
2317 {
2318         int ret;
2319         struct btrfs_path *path;
2320         struct btrfs_key key;
2321
2322         root = root->fs_info->chunk_root;
2323         path = btrfs_alloc_path();
2324         if (!path)
2325                 return -ENOMEM;
2326
2327         key.objectid = chunk_objectid;
2328         key.offset = chunk_offset;
2329         key.type = BTRFS_CHUNK_ITEM_KEY;
2330
2331         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2332         if (ret < 0)
2333                 goto out;
2334         else if (ret > 0) { /* Logic error or corruption */
2335                 btrfs_error(root->fs_info, -ENOENT,
2336                             "Failed lookup while freeing chunk.");
2337                 ret = -ENOENT;
2338                 goto out;
2339         }
2340
2341         ret = btrfs_del_item(trans, root, path);
2342         if (ret < 0)
2343                 btrfs_error(root->fs_info, ret,
2344                             "Failed to delete chunk item.");
2345 out:
2346         btrfs_free_path(path);
2347         return ret;
2348 }
2349
2350 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
2351                         chunk_offset)
2352 {
2353         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
2354         struct btrfs_disk_key *disk_key;
2355         struct btrfs_chunk *chunk;
2356         u8 *ptr;
2357         int ret = 0;
2358         u32 num_stripes;
2359         u32 array_size;
2360         u32 len = 0;
2361         u32 cur;
2362         struct btrfs_key key;
2363
2364         array_size = btrfs_super_sys_array_size(super_copy);
2365
2366         ptr = super_copy->sys_chunk_array;
2367         cur = 0;
2368
2369         while (cur < array_size) {
2370                 disk_key = (struct btrfs_disk_key *)ptr;
2371                 btrfs_disk_key_to_cpu(&key, disk_key);
2372
2373                 len = sizeof(*disk_key);
2374
2375                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2376                         chunk = (struct btrfs_chunk *)(ptr + len);
2377                         num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2378                         len += btrfs_chunk_item_size(num_stripes);
2379                 } else {
2380                         ret = -EIO;
2381                         break;
2382                 }
2383                 if (key.objectid == chunk_objectid &&
2384                     key.offset == chunk_offset) {
2385                         memmove(ptr, ptr + len, array_size - (cur + len));
2386                         array_size -= len;
2387                         btrfs_set_super_sys_array_size(super_copy, array_size);
2388                 } else {
2389                         ptr += len;
2390                         cur += len;
2391                 }
2392         }
2393         return ret;
2394 }
2395
2396 static int btrfs_relocate_chunk(struct btrfs_root *root,
2397                          u64 chunk_tree, u64 chunk_objectid,
2398                          u64 chunk_offset)
2399 {
2400         struct extent_map_tree *em_tree;
2401         struct btrfs_root *extent_root;
2402         struct btrfs_trans_handle *trans;
2403         struct extent_map *em;
2404         struct map_lookup *map;
2405         int ret;
2406         int i;
2407
2408         root = root->fs_info->chunk_root;
2409         extent_root = root->fs_info->extent_root;
2410         em_tree = &root->fs_info->mapping_tree.map_tree;
2411
2412         ret = btrfs_can_relocate(extent_root, chunk_offset);
2413         if (ret)
2414                 return -ENOSPC;
2415
2416         /* step one, relocate all the extents inside this chunk */
2417         ret = btrfs_relocate_block_group(extent_root, chunk_offset);
2418         if (ret)
2419                 return ret;
2420
2421         trans = btrfs_start_transaction(root, 0);
2422         if (IS_ERR(trans)) {
2423                 ret = PTR_ERR(trans);
2424                 btrfs_std_error(root->fs_info, ret);
2425                 return ret;
2426         }
2427
2428         lock_chunks(root);
2429
2430         /*
2431          * step two, delete the device extents and the
2432          * chunk tree entries
2433          */
2434         read_lock(&em_tree->lock);
2435         em = lookup_extent_mapping(em_tree, chunk_offset, 1);
2436         read_unlock(&em_tree->lock);
2437
2438         BUG_ON(!em || em->start > chunk_offset ||
2439                em->start + em->len < chunk_offset);
2440         map = (struct map_lookup *)em->bdev;
2441
2442         for (i = 0; i < map->num_stripes; i++) {
2443                 ret = btrfs_free_dev_extent(trans, map->stripes[i].dev,
2444                                             map->stripes[i].physical);
2445                 BUG_ON(ret);
2446
2447                 if (map->stripes[i].dev) {
2448                         ret = btrfs_update_device(trans, map->stripes[i].dev);
2449                         BUG_ON(ret);
2450                 }
2451         }
2452         ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
2453                                chunk_offset);
2454
2455         BUG_ON(ret);
2456
2457         trace_btrfs_chunk_free(root, map, chunk_offset, em->len);
2458
2459         if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2460                 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
2461                 BUG_ON(ret);
2462         }
2463
2464         ret = btrfs_remove_block_group(trans, extent_root, chunk_offset);
2465         BUG_ON(ret);
2466
2467         write_lock(&em_tree->lock);
2468         remove_extent_mapping(em_tree, em);
2469         write_unlock(&em_tree->lock);
2470
2471         kfree(map);
2472         em->bdev = NULL;
2473
2474         /* once for the tree */
2475         free_extent_map(em);
2476         /* once for us */
2477         free_extent_map(em);
2478
2479         unlock_chunks(root);
2480         btrfs_end_transaction(trans, root);
2481         return 0;
2482 }
2483
2484 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
2485 {
2486         struct btrfs_root *chunk_root = root->fs_info->chunk_root;
2487         struct btrfs_path *path;
2488         struct extent_buffer *leaf;
2489         struct btrfs_chunk *chunk;
2490         struct btrfs_key key;
2491         struct btrfs_key found_key;
2492         u64 chunk_tree = chunk_root->root_key.objectid;
2493         u64 chunk_type;
2494         bool retried = false;
2495         int failed = 0;
2496         int ret;
2497
2498         path = btrfs_alloc_path();
2499         if (!path)
2500                 return -ENOMEM;
2501
2502 again:
2503         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2504         key.offset = (u64)-1;
2505         key.type = BTRFS_CHUNK_ITEM_KEY;
2506
2507         while (1) {
2508                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2509                 if (ret < 0)
2510                         goto error;
2511                 BUG_ON(ret == 0); /* Corruption */
2512
2513                 ret = btrfs_previous_item(chunk_root, path, key.objectid,
2514                                           key.type);
2515                 if (ret < 0)
2516                         goto error;
2517                 if (ret > 0)
2518                         break;
2519
2520                 leaf = path->nodes[0];
2521                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2522
2523                 chunk = btrfs_item_ptr(leaf, path->slots[0],
2524                                        struct btrfs_chunk);
2525                 chunk_type = btrfs_chunk_type(leaf, chunk);
2526                 btrfs_release_path(path);
2527
2528                 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
2529                         ret = btrfs_relocate_chunk(chunk_root, chunk_tree,
2530                                                    found_key.objectid,
2531                                                    found_key.offset);
2532                         if (ret == -ENOSPC)
2533                                 failed++;
2534                         else if (ret)
2535                                 BUG();
2536                 }
2537
2538                 if (found_key.offset == 0)
2539                         break;
2540                 key.offset = found_key.offset - 1;
2541         }
2542         ret = 0;
2543         if (failed && !retried) {
2544                 failed = 0;
2545                 retried = true;
2546                 goto again;
2547         } else if (failed && retried) {
2548                 WARN_ON(1);
2549                 ret = -ENOSPC;
2550         }
2551 error:
2552         btrfs_free_path(path);
2553         return ret;
2554 }
2555
2556 static int insert_balance_item(struct btrfs_root *root,
2557                                struct btrfs_balance_control *bctl)
2558 {
2559         struct btrfs_trans_handle *trans;
2560         struct btrfs_balance_item *item;
2561         struct btrfs_disk_balance_args disk_bargs;
2562         struct btrfs_path *path;
2563         struct extent_buffer *leaf;
2564         struct btrfs_key key;
2565         int ret, err;
2566
2567         path = btrfs_alloc_path();
2568         if (!path)
2569                 return -ENOMEM;
2570
2571         trans = btrfs_start_transaction(root, 0);
2572         if (IS_ERR(trans)) {
2573                 btrfs_free_path(path);
2574                 return PTR_ERR(trans);
2575         }
2576
2577         key.objectid = BTRFS_BALANCE_OBJECTID;
2578         key.type = BTRFS_BALANCE_ITEM_KEY;
2579         key.offset = 0;
2580
2581         ret = btrfs_insert_empty_item(trans, root, path, &key,
2582                                       sizeof(*item));
2583         if (ret)
2584                 goto out;
2585
2586         leaf = path->nodes[0];
2587         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
2588
2589         memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
2590
2591         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
2592         btrfs_set_balance_data(leaf, item, &disk_bargs);
2593         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
2594         btrfs_set_balance_meta(leaf, item, &disk_bargs);
2595         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
2596         btrfs_set_balance_sys(leaf, item, &disk_bargs);
2597
2598         btrfs_set_balance_flags(leaf, item, bctl->flags);
2599
2600         btrfs_mark_buffer_dirty(leaf);
2601 out:
2602         btrfs_free_path(path);
2603         err = btrfs_commit_transaction(trans, root);
2604         if (err && !ret)
2605                 ret = err;
2606         return ret;
2607 }
2608
2609 static int del_balance_item(struct btrfs_root *root)
2610 {
2611         struct btrfs_trans_handle *trans;
2612         struct btrfs_path *path;
2613         struct btrfs_key key;
2614         int ret, err;
2615
2616         path = btrfs_alloc_path();
2617         if (!path)
2618                 return -ENOMEM;
2619
2620         trans = btrfs_start_transaction(root, 0);
2621         if (IS_ERR(trans)) {
2622                 btrfs_free_path(path);
2623                 return PTR_ERR(trans);
2624         }
2625
2626         key.objectid = BTRFS_BALANCE_OBJECTID;
2627         key.type = BTRFS_BALANCE_ITEM_KEY;
2628         key.offset = 0;
2629
2630         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2631         if (ret < 0)
2632                 goto out;
2633         if (ret > 0) {
2634                 ret = -ENOENT;
2635                 goto out;
2636         }
2637
2638         ret = btrfs_del_item(trans, root, path);
2639 out:
2640         btrfs_free_path(path);
2641         err = btrfs_commit_transaction(trans, root);
2642         if (err && !ret)
2643                 ret = err;
2644         return ret;
2645 }
2646
2647 /*
2648  * This is a heuristic used to reduce the number of chunks balanced on
2649  * resume after balance was interrupted.
2650  */
2651 static void update_balance_args(struct btrfs_balance_control *bctl)
2652 {
2653         /*
2654          * Turn on soft mode for chunk types that were being converted.
2655          */
2656         if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
2657                 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
2658         if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
2659                 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
2660         if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
2661                 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
2662
2663         /*
2664          * Turn on usage filter if is not already used.  The idea is
2665          * that chunks that we have already balanced should be
2666          * reasonably full.  Don't do it for chunks that are being
2667          * converted - that will keep us from relocating unconverted
2668          * (albeit full) chunks.
2669          */
2670         if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2671             !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2672                 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
2673                 bctl->data.usage = 90;
2674         }
2675         if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2676             !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2677                 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
2678                 bctl->sys.usage = 90;
2679         }
2680         if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2681             !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2682                 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
2683                 bctl->meta.usage = 90;
2684         }
2685 }
2686
2687 /*
2688  * Should be called with both balance and volume mutexes held to
2689  * serialize other volume operations (add_dev/rm_dev/resize) with
2690  * restriper.  Same goes for unset_balance_control.
2691  */
2692 static void set_balance_control(struct btrfs_balance_control *bctl)
2693 {
2694         struct btrfs_fs_info *fs_info = bctl->fs_info;
2695
2696         BUG_ON(fs_info->balance_ctl);
2697
2698         spin_lock(&fs_info->balance_lock);
2699         fs_info->balance_ctl = bctl;
2700         spin_unlock(&fs_info->balance_lock);
2701 }
2702
2703 static void unset_balance_control(struct btrfs_fs_info *fs_info)
2704 {
2705         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
2706
2707         BUG_ON(!fs_info->balance_ctl);
2708
2709         spin_lock(&fs_info->balance_lock);
2710         fs_info->balance_ctl = NULL;
2711         spin_unlock(&fs_info->balance_lock);
2712
2713         kfree(bctl);
2714 }
2715
2716 /*
2717  * Balance filters.  Return 1 if chunk should be filtered out
2718  * (should not be balanced).
2719  */
2720 static int chunk_profiles_filter(u64 chunk_type,
2721                                  struct btrfs_balance_args *bargs)
2722 {
2723         chunk_type = chunk_to_extended(chunk_type) &
2724                                 BTRFS_EXTENDED_PROFILE_MASK;
2725
2726         if (bargs->profiles & chunk_type)
2727                 return 0;
2728
2729         return 1;
2730 }
2731
2732 static int chunk_usage_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
2733                               struct btrfs_balance_args *bargs)
2734 {
2735         struct btrfs_block_group_cache *cache;
2736         u64 chunk_used, user_thresh;
2737         int ret = 1;
2738
2739         cache = btrfs_lookup_block_group(fs_info, chunk_offset);
2740         chunk_used = btrfs_block_group_used(&cache->item);
2741
2742         if (bargs->usage == 0)
2743                 user_thresh = 1;
2744         else if (bargs->usage > 100)
2745                 user_thresh = cache->key.offset;
2746         else
2747                 user_thresh = div_factor_fine(cache->key.offset,
2748                                               bargs->usage);
2749
2750         if (chunk_used < user_thresh)
2751                 ret = 0;
2752
2753         btrfs_put_block_group(cache);
2754         return ret;
2755 }
2756
2757 static int chunk_devid_filter(struct extent_buffer *leaf,
2758                               struct btrfs_chunk *chunk,
2759                               struct btrfs_balance_args *bargs)
2760 {
2761         struct btrfs_stripe *stripe;
2762         int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2763         int i;
2764
2765         for (i = 0; i < num_stripes; i++) {
2766                 stripe = btrfs_stripe_nr(chunk, i);
2767                 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
2768                         return 0;
2769         }
2770
2771         return 1;
2772 }
2773
2774 /* [pstart, pend) */
2775 static int chunk_drange_filter(struct extent_buffer *leaf,
2776                                struct btrfs_chunk *chunk,
2777                                u64 chunk_offset,
2778                                struct btrfs_balance_args *bargs)
2779 {
2780         struct btrfs_stripe *stripe;
2781         int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2782         u64 stripe_offset;
2783         u64 stripe_length;
2784         int factor;
2785         int i;
2786
2787         if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
2788                 return 0;
2789
2790         if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP |
2791              BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) {
2792                 factor = num_stripes / 2;
2793         } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) {
2794                 factor = num_stripes - 1;
2795         } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) {
2796                 factor = num_stripes - 2;
2797         } else {
2798                 factor = num_stripes;
2799         }
2800
2801         for (i = 0; i < num_stripes; i++) {
2802                 stripe = btrfs_stripe_nr(chunk, i);
2803                 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
2804                         continue;
2805
2806                 stripe_offset = btrfs_stripe_offset(leaf, stripe);
2807                 stripe_length = btrfs_chunk_length(leaf, chunk);
2808                 do_div(stripe_length, factor);
2809
2810                 if (stripe_offset < bargs->pend &&
2811                     stripe_offset + stripe_length > bargs->pstart)
2812                         return 0;
2813         }
2814
2815         return 1;
2816 }
2817
2818 /* [vstart, vend) */
2819 static int chunk_vrange_filter(struct extent_buffer *leaf,
2820                                struct btrfs_chunk *chunk,
2821                                u64 chunk_offset,
2822                                struct btrfs_balance_args *bargs)
2823 {
2824         if (chunk_offset < bargs->vend &&
2825             chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
2826                 /* at least part of the chunk is inside this vrange */
2827                 return 0;
2828
2829         return 1;
2830 }
2831
2832 static int chunk_soft_convert_filter(u64 chunk_type,
2833                                      struct btrfs_balance_args *bargs)
2834 {
2835         if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
2836                 return 0;
2837
2838         chunk_type = chunk_to_extended(chunk_type) &
2839                                 BTRFS_EXTENDED_PROFILE_MASK;
2840
2841         if (bargs->target == chunk_type)
2842                 return 1;
2843
2844         return 0;
2845 }
2846
2847 static int should_balance_chunk(struct btrfs_root *root,
2848                                 struct extent_buffer *leaf,
2849                                 struct btrfs_chunk *chunk, u64 chunk_offset)
2850 {
2851         struct btrfs_balance_control *bctl = root->fs_info->balance_ctl;
2852         struct btrfs_balance_args *bargs = NULL;
2853         u64 chunk_type = btrfs_chunk_type(leaf, chunk);
2854
2855         /* type filter */
2856         if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
2857               (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
2858                 return 0;
2859         }
2860
2861         if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
2862                 bargs = &bctl->data;
2863         else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
2864                 bargs = &bctl->sys;
2865         else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
2866                 bargs = &bctl->meta;
2867
2868         /* profiles filter */
2869         if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
2870             chunk_profiles_filter(chunk_type, bargs)) {
2871                 return 0;
2872         }
2873
2874         /* usage filter */
2875         if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
2876             chunk_usage_filter(bctl->fs_info, chunk_offset, bargs)) {
2877                 return 0;
2878         }
2879
2880         /* devid filter */
2881         if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
2882             chunk_devid_filter(leaf, chunk, bargs)) {
2883                 return 0;
2884         }
2885
2886         /* drange filter, makes sense only with devid filter */
2887         if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
2888             chunk_drange_filter(leaf, chunk, chunk_offset, bargs)) {
2889                 return 0;
2890         }
2891
2892         /* vrange filter */
2893         if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
2894             chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
2895                 return 0;
2896         }
2897
2898         /* soft profile changing mode */
2899         if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
2900             chunk_soft_convert_filter(chunk_type, bargs)) {
2901                 return 0;
2902         }
2903
2904         return 1;
2905 }
2906
2907 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
2908 {
2909         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
2910         struct btrfs_root *chunk_root = fs_info->chunk_root;
2911         struct btrfs_root *dev_root = fs_info->dev_root;
2912         struct list_head *devices;
2913         struct btrfs_device *device;
2914         u64 old_size;
2915         u64 size_to_free;
2916         struct btrfs_chunk *chunk;
2917         struct btrfs_path *path;
2918         struct btrfs_key key;
2919         struct btrfs_key found_key;
2920         struct btrfs_trans_handle *trans;
2921         struct extent_buffer *leaf;
2922         int slot;
2923         int ret;
2924         int enospc_errors = 0;
2925         bool counting = true;
2926
2927         /* step one make some room on all the devices */
2928         devices = &fs_info->fs_devices->devices;
2929         list_for_each_entry(device, devices, dev_list) {
2930                 old_size = device->total_bytes;
2931                 size_to_free = div_factor(old_size, 1);
2932                 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
2933                 if (!device->writeable ||
2934                     device->total_bytes - device->bytes_used > size_to_free ||
2935                     device->is_tgtdev_for_dev_replace)
2936                         continue;
2937
2938                 ret = btrfs_shrink_device(device, old_size - size_to_free);
2939                 if (ret == -ENOSPC)
2940                         break;
2941                 BUG_ON(ret);
2942
2943                 trans = btrfs_start_transaction(dev_root, 0);
2944                 BUG_ON(IS_ERR(trans));
2945
2946                 ret = btrfs_grow_device(trans, device, old_size);
2947                 BUG_ON(ret);
2948
2949                 btrfs_end_transaction(trans, dev_root);
2950         }
2951
2952         /* step two, relocate all the chunks */
2953         path = btrfs_alloc_path();
2954         if (!path) {
2955                 ret = -ENOMEM;
2956                 goto error;
2957         }
2958
2959         /* zero out stat counters */
2960         spin_lock(&fs_info->balance_lock);
2961         memset(&bctl->stat, 0, sizeof(bctl->stat));
2962         spin_unlock(&fs_info->balance_lock);
2963 again:
2964         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2965         key.offset = (u64)-1;
2966         key.type = BTRFS_CHUNK_ITEM_KEY;
2967
2968         while (1) {
2969                 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
2970                     atomic_read(&fs_info->balance_cancel_req)) {
2971                         ret = -ECANCELED;
2972                         goto error;
2973                 }
2974
2975                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2976                 if (ret < 0)
2977                         goto error;
2978
2979                 /*
2980                  * this shouldn't happen, it means the last relocate
2981                  * failed
2982                  */
2983                 if (ret == 0)
2984                         BUG(); /* FIXME break ? */
2985
2986                 ret = btrfs_previous_item(chunk_root, path, 0,
2987                                           BTRFS_CHUNK_ITEM_KEY);
2988                 if (ret) {
2989                         ret = 0;
2990                         break;
2991                 }
2992
2993                 leaf = path->nodes[0];
2994                 slot = path->slots[0];
2995                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2996
2997                 if (found_key.objectid != key.objectid)
2998                         break;
2999
3000                 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3001
3002                 if (!counting) {
3003                         spin_lock(&fs_info->balance_lock);
3004                         bctl->stat.considered++;
3005                         spin_unlock(&fs_info->balance_lock);
3006                 }
3007
3008                 ret = should_balance_chunk(chunk_root, leaf, chunk,
3009                                            found_key.offset);
3010                 btrfs_release_path(path);
3011                 if (!ret)
3012                         goto loop;
3013
3014                 if (counting) {
3015                         spin_lock(&fs_info->balance_lock);
3016                         bctl->stat.expected++;
3017                         spin_unlock(&fs_info->balance_lock);
3018                         goto loop;
3019                 }
3020
3021                 ret = btrfs_relocate_chunk(chunk_root,
3022                                            chunk_root->root_key.objectid,
3023                                            found_key.objectid,
3024                                            found_key.offset);
3025                 if (ret && ret != -ENOSPC)
3026                         goto error;
3027                 if (ret == -ENOSPC) {
3028                         enospc_errors++;
3029                 } else {
3030                         spin_lock(&fs_info->balance_lock);
3031                         bctl->stat.completed++;
3032                         spin_unlock(&fs_info->balance_lock);
3033                 }
3034 loop:
3035                 if (found_key.offset == 0)
3036                         break;
3037                 key.offset = found_key.offset - 1;
3038         }
3039
3040         if (counting) {
3041                 btrfs_release_path(path);
3042                 counting = false;
3043                 goto again;
3044         }
3045 error:
3046         btrfs_free_path(path);
3047         if (enospc_errors) {
3048                 printk(KERN_INFO "btrfs: %d enospc errors during balance\n",
3049                        enospc_errors);
3050                 if (!ret)
3051                         ret = -ENOSPC;
3052         }
3053
3054         return ret;
3055 }
3056
3057 /**
3058  * alloc_profile_is_valid - see if a given profile is valid and reduced
3059  * @flags: profile to validate
3060  * @extended: if true @flags is treated as an extended profile
3061  */
3062 static int alloc_profile_is_valid(u64 flags, int extended)
3063 {
3064         u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
3065                                BTRFS_BLOCK_GROUP_PROFILE_MASK);
3066
3067         flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
3068
3069         /* 1) check that all other bits are zeroed */
3070         if (flags & ~mask)
3071                 return 0;
3072
3073         /* 2) see if profile is reduced */
3074         if (flags == 0)
3075                 return !extended; /* "0" is valid for usual profiles */
3076
3077         /* true if exactly one bit set */
3078         return (flags & (flags - 1)) == 0;
3079 }
3080
3081 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
3082 {
3083         /* cancel requested || normal exit path */
3084         return atomic_read(&fs_info->balance_cancel_req) ||
3085                 (atomic_read(&fs_info->balance_pause_req) == 0 &&
3086                  atomic_read(&fs_info->balance_cancel_req) == 0);
3087 }
3088
3089 static void __cancel_balance(struct btrfs_fs_info *fs_info)
3090 {
3091         int ret;
3092
3093         unset_balance_control(fs_info);
3094         ret = del_balance_item(fs_info->tree_root);
3095         if (ret)
3096                 btrfs_std_error(fs_info, ret);
3097
3098         atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3099 }
3100
3101 /*
3102  * Should be called with both balance and volume mutexes held
3103  */
3104 int btrfs_balance(struct btrfs_balance_control *bctl,
3105                   struct btrfs_ioctl_balance_args *bargs)
3106 {
3107         struct btrfs_fs_info *fs_info = bctl->fs_info;
3108         u64 allowed;
3109         int mixed = 0;
3110         int ret;
3111         u64 num_devices;
3112         unsigned seq;
3113
3114         if (btrfs_fs_closing(fs_info) ||
3115             atomic_read(&fs_info->balance_pause_req) ||
3116             atomic_read(&fs_info->balance_cancel_req)) {
3117                 ret = -EINVAL;
3118                 goto out;
3119         }
3120
3121         allowed = btrfs_super_incompat_flags(fs_info->super_copy);
3122         if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
3123                 mixed = 1;
3124
3125         /*
3126          * In case of mixed groups both data and meta should be picked,
3127          * and identical options should be given for both of them.
3128          */
3129         allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
3130         if (mixed && (bctl->flags & allowed)) {
3131                 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
3132                     !(bctl->flags & BTRFS_BALANCE_METADATA) ||
3133                     memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
3134                         printk(KERN_ERR "btrfs: with mixed groups data and "
3135                                "metadata balance options must be the same\n");
3136                         ret = -EINVAL;
3137                         goto out;
3138                 }
3139         }
3140
3141         num_devices = fs_info->fs_devices->num_devices;
3142         btrfs_dev_replace_lock(&fs_info->dev_replace);
3143         if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
3144                 BUG_ON(num_devices < 1);
3145                 num_devices--;
3146         }
3147         btrfs_dev_replace_unlock(&fs_info->dev_replace);
3148         allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE;
3149         if (num_devices == 1)
3150                 allowed |= BTRFS_BLOCK_GROUP_DUP;
3151         else if (num_devices > 1)
3152                 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1);
3153         if (num_devices > 2)
3154                 allowed |= BTRFS_BLOCK_GROUP_RAID5;
3155         if (num_devices > 3)
3156                 allowed |= (BTRFS_BLOCK_GROUP_RAID10 |
3157                             BTRFS_BLOCK_GROUP_RAID6);
3158         if ((bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3159             (!alloc_profile_is_valid(bctl->data.target, 1) ||
3160              (bctl->data.target & ~allowed))) {
3161                 printk(KERN_ERR "btrfs: unable to start balance with target "
3162                        "data profile %llu\n",
3163                        bctl->data.target);
3164                 ret = -EINVAL;
3165                 goto out;
3166         }
3167         if ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3168             (!alloc_profile_is_valid(bctl->meta.target, 1) ||
3169              (bctl->meta.target & ~allowed))) {
3170                 printk(KERN_ERR "btrfs: unable to start balance with target "
3171                        "metadata profile %llu\n",
3172                        bctl->meta.target);
3173                 ret = -EINVAL;
3174                 goto out;
3175         }
3176         if ((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3177             (!alloc_profile_is_valid(bctl->sys.target, 1) ||
3178              (bctl->sys.target & ~allowed))) {
3179                 printk(KERN_ERR "btrfs: unable to start balance with target "
3180                        "system profile %llu\n",
3181                        bctl->sys.target);
3182                 ret = -EINVAL;
3183                 goto out;
3184         }
3185
3186         /* allow dup'ed data chunks only in mixed mode */
3187         if (!mixed && (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3188             (bctl->data.target & BTRFS_BLOCK_GROUP_DUP)) {
3189                 printk(KERN_ERR "btrfs: dup for data is not allowed\n");
3190                 ret = -EINVAL;
3191                 goto out;
3192         }
3193
3194         /* allow to reduce meta or sys integrity only if force set */
3195         allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3196                         BTRFS_BLOCK_GROUP_RAID10 |
3197                         BTRFS_BLOCK_GROUP_RAID5 |
3198                         BTRFS_BLOCK_GROUP_RAID6;
3199         do {
3200                 seq = read_seqbegin(&fs_info->profiles_lock);
3201
3202                 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3203                      (fs_info->avail_system_alloc_bits & allowed) &&
3204                      !(bctl->sys.target & allowed)) ||
3205                     ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3206                      (fs_info->avail_metadata_alloc_bits & allowed) &&
3207                      !(bctl->meta.target & allowed))) {
3208                         if (bctl->flags & BTRFS_BALANCE_FORCE) {
3209                                 printk(KERN_INFO "btrfs: force reducing metadata "
3210                                        "integrity\n");
3211                         } else {
3212                                 printk(KERN_ERR "btrfs: balance will reduce metadata "
3213                                        "integrity, use force if you want this\n");
3214                                 ret = -EINVAL;
3215                                 goto out;
3216                         }
3217                 }
3218         } while (read_seqretry(&fs_info->profiles_lock, seq));
3219
3220         if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3221                 int num_tolerated_disk_barrier_failures;
3222                 u64 target = bctl->sys.target;
3223
3224                 num_tolerated_disk_barrier_failures =
3225                         btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3226                 if (num_tolerated_disk_barrier_failures > 0 &&
3227                     (target &
3228                      (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID0 |
3229                       BTRFS_AVAIL_ALLOC_BIT_SINGLE)))
3230                         num_tolerated_disk_barrier_failures = 0;
3231                 else if (num_tolerated_disk_barrier_failures > 1 &&
3232                          (target &
3233                           (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)))
3234                         num_tolerated_disk_barrier_failures = 1;
3235
3236                 fs_info->num_tolerated_disk_barrier_failures =
3237                         num_tolerated_disk_barrier_failures;
3238         }
3239
3240         ret = insert_balance_item(fs_info->tree_root, bctl);
3241         if (ret && ret != -EEXIST)
3242                 goto out;
3243
3244         if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
3245                 BUG_ON(ret == -EEXIST);
3246                 set_balance_control(bctl);
3247         } else {
3248                 BUG_ON(ret != -EEXIST);
3249                 spin_lock(&fs_info->balance_lock);
3250                 update_balance_args(bctl);
3251                 spin_unlock(&fs_info->balance_lock);
3252         }
3253
3254         atomic_inc(&fs_info->balance_running);
3255         mutex_unlock(&fs_info->balance_mutex);
3256
3257         ret = __btrfs_balance(fs_info);
3258
3259         mutex_lock(&fs_info->balance_mutex);
3260         atomic_dec(&fs_info->balance_running);
3261
3262         if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3263                 fs_info->num_tolerated_disk_barrier_failures =
3264                         btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3265         }
3266
3267         if (bargs) {
3268                 memset(bargs, 0, sizeof(*bargs));
3269                 update_ioctl_balance_args(fs_info, 0, bargs);
3270         }
3271
3272         if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
3273             balance_need_close(fs_info)) {
3274                 __cancel_balance(fs_info);
3275         }
3276
3277         wake_up(&fs_info->balance_wait_q);
3278
3279         return ret;
3280 out:
3281         if (bctl->flags & BTRFS_BALANCE_RESUME)
3282                 __cancel_balance(fs_info);
3283         else {
3284                 kfree(bctl);
3285                 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3286         }
3287         return ret;
3288 }
3289
3290 static int balance_kthread(void *data)
3291 {
3292         struct btrfs_fs_info *fs_info = data;
3293         int ret = 0;
3294
3295         mutex_lock(&fs_info->volume_mutex);
3296         mutex_lock(&fs_info->balance_mutex);
3297
3298         if (fs_info->balance_ctl) {
3299                 printk(KERN_INFO "btrfs: continuing balance\n");
3300                 ret = btrfs_balance(fs_info->balance_ctl, NULL);
3301         }
3302
3303         mutex_unlock(&fs_info->balance_mutex);
3304         mutex_unlock(&fs_info->volume_mutex);
3305
3306         return ret;
3307 }
3308
3309 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
3310 {
3311         struct task_struct *tsk;
3312
3313         spin_lock(&fs_info->balance_lock);
3314         if (!fs_info->balance_ctl) {
3315                 spin_unlock(&fs_info->balance_lock);
3316                 return 0;
3317         }
3318         spin_unlock(&fs_info->balance_lock);
3319
3320         if (btrfs_test_opt(fs_info->tree_root, SKIP_BALANCE)) {
3321                 printk(KERN_INFO "btrfs: force skipping balance\n");
3322                 return 0;
3323         }
3324
3325         tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
3326         return PTR_ERR_OR_ZERO(tsk);
3327 }
3328
3329 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
3330 {
3331         struct btrfs_balance_control *bctl;
3332         struct btrfs_balance_item *item;
3333         struct btrfs_disk_balance_args disk_bargs;
3334         struct btrfs_path *path;
3335         struct extent_buffer *leaf;
3336         struct btrfs_key key;
3337         int ret;
3338
3339         path = btrfs_alloc_path();
3340         if (!path)
3341                 return -ENOMEM;
3342
3343         key.objectid = BTRFS_BALANCE_OBJECTID;
3344         key.type = BTRFS_BALANCE_ITEM_KEY;
3345         key.offset = 0;
3346
3347         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
3348         if (ret < 0)
3349                 goto out;
3350         if (ret > 0) { /* ret = -ENOENT; */
3351                 ret = 0;
3352                 goto out;
3353         }
3354
3355         bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
3356         if (!bctl) {
3357                 ret = -ENOMEM;
3358                 goto out;
3359         }
3360
3361         leaf = path->nodes[0];
3362         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3363
3364         bctl->fs_info = fs_info;
3365         bctl->flags = btrfs_balance_flags(leaf, item);
3366         bctl->flags |= BTRFS_BALANCE_RESUME;
3367
3368         btrfs_balance_data(leaf, item, &disk_bargs);
3369         btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
3370         btrfs_balance_meta(leaf, item, &disk_bargs);
3371         btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
3372         btrfs_balance_sys(leaf, item, &disk_bargs);
3373         btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
3374
3375         WARN_ON(atomic_xchg(&fs_info->mutually_exclusive_operation_running, 1));
3376
3377         mutex_lock(&fs_info->volume_mutex);
3378         mutex_lock(&fs_info->balance_mutex);
3379
3380         set_balance_control(bctl);
3381
3382         mutex_unlock(&fs_info->balance_mutex);
3383         mutex_unlock(&fs_info->volume_mutex);
3384 out:
3385         btrfs_free_path(path);
3386         return ret;
3387 }
3388
3389 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
3390 {
3391         int ret = 0;
3392
3393         mutex_lock(&fs_info->balance_mutex);
3394         if (!fs_info->balance_ctl) {
3395                 mutex_unlock(&fs_info->balance_mutex);
3396                 return -ENOTCONN;
3397         }
3398
3399         if (atomic_read(&fs_info->balance_running)) {
3400                 atomic_inc(&fs_info->balance_pause_req);
3401                 mutex_unlock(&fs_info->balance_mutex);
3402
3403                 wait_event(fs_info->balance_wait_q,
3404                            atomic_read(&fs_info->balance_running) == 0);
3405
3406                 mutex_lock(&fs_info->balance_mutex);
3407                 /* we are good with balance_ctl ripped off from under us */
3408                 BUG_ON(atomic_read(&fs_info->balance_running));
3409                 atomic_dec(&fs_info->balance_pause_req);
3410         } else {
3411                 ret = -ENOTCONN;
3412         }
3413
3414         mutex_unlock(&fs_info->balance_mutex);
3415         return ret;
3416 }
3417
3418 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
3419 {
3420         mutex_lock(&fs_info->balance_mutex);
3421         if (!fs_info->balance_ctl) {
3422                 mutex_unlock(&fs_info->balance_mutex);
3423                 return -ENOTCONN;
3424         }
3425
3426         atomic_inc(&fs_info->balance_cancel_req);
3427         /*
3428          * if we are running just wait and return, balance item is
3429          * deleted in btrfs_balance in this case
3430          */
3431         if (atomic_read(&fs_info->balance_running)) {
3432                 mutex_unlock(&fs_info->balance_mutex);
3433                 wait_event(fs_info->balance_wait_q,
3434                            atomic_read(&fs_info->balance_running) == 0);
3435                 mutex_lock(&fs_info->balance_mutex);
3436         } else {
3437                 /* __cancel_balance needs volume_mutex */
3438                 mutex_unlock(&fs_info->balance_mutex);
3439                 mutex_lock(&fs_info->volume_mutex);
3440                 mutex_lock(&fs_info->balance_mutex);
3441
3442                 if (fs_info->balance_ctl)
3443                         __cancel_balance(fs_info);
3444
3445                 mutex_unlock(&fs_info->volume_mutex);
3446         }
3447
3448         BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running));
3449         atomic_dec(&fs_info->balance_cancel_req);
3450         mutex_unlock(&fs_info->balance_mutex);
3451         return 0;
3452 }
3453
3454 static int btrfs_uuid_scan_kthread(void *data)
3455 {
3456         struct btrfs_fs_info *fs_info = data;
3457         struct btrfs_root *root = fs_info->tree_root;
3458         struct btrfs_key key;
3459         struct btrfs_key max_key;
3460         struct btrfs_path *path = NULL;
3461         int ret = 0;
3462         struct extent_buffer *eb;
3463         int slot;
3464         struct btrfs_root_item root_item;
3465         u32 item_size;
3466         struct btrfs_trans_handle *trans = NULL;
3467
3468         path = btrfs_alloc_path();
3469         if (!path) {
3470                 ret = -ENOMEM;
3471                 goto out;
3472         }
3473
3474         key.objectid = 0;
3475         key.type = BTRFS_ROOT_ITEM_KEY;
3476         key.offset = 0;
3477
3478         max_key.objectid = (u64)-1;
3479         max_key.type = BTRFS_ROOT_ITEM_KEY;
3480         max_key.offset = (u64)-1;
3481
3482         path->keep_locks = 1;
3483
3484         while (1) {
3485                 ret = btrfs_search_forward(root, &key, &max_key, path, 0);
3486                 if (ret) {
3487                         if (ret > 0)
3488                                 ret = 0;
3489                         break;
3490                 }
3491
3492                 if (key.type != BTRFS_ROOT_ITEM_KEY ||
3493                     (key.objectid < BTRFS_FIRST_FREE_OBJECTID &&
3494                      key.objectid != BTRFS_FS_TREE_OBJECTID) ||
3495                     key.objectid > BTRFS_LAST_FREE_OBJECTID)
3496                         goto skip;
3497
3498                 eb = path->nodes[0];
3499                 slot = path->slots[0];
3500                 item_size = btrfs_item_size_nr(eb, slot);
3501                 if (item_size < sizeof(root_item))
3502                         goto skip;
3503
3504                 read_extent_buffer(eb, &root_item,
3505                                    btrfs_item_ptr_offset(eb, slot),
3506                                    (int)sizeof(root_item));
3507                 if (btrfs_root_refs(&root_item) == 0)
3508                         goto skip;
3509
3510                 if (!btrfs_is_empty_uuid(root_item.uuid) ||
3511                     !btrfs_is_empty_uuid(root_item.received_uuid)) {
3512                         if (trans)
3513                                 goto update_tree;
3514
3515                         btrfs_release_path(path);
3516                         /*
3517                          * 1 - subvol uuid item
3518                          * 1 - received_subvol uuid item
3519                          */
3520                         trans = btrfs_start_transaction(fs_info->uuid_root, 2);
3521                         if (IS_ERR(trans)) {
3522                                 ret = PTR_ERR(trans);
3523                                 break;
3524                         }
3525                         continue;
3526                 } else {
3527                         goto skip;
3528                 }
3529 update_tree:
3530                 if (!btrfs_is_empty_uuid(root_item.uuid)) {
3531                         ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
3532                                                   root_item.uuid,
3533                                                   BTRFS_UUID_KEY_SUBVOL,
3534                                                   key.objectid);
3535                         if (ret < 0) {
3536                                 pr_warn("btrfs: uuid_tree_add failed %d\n",
3537                                         ret);
3538                                 break;
3539                         }
3540                 }
3541
3542                 if (!btrfs_is_empty_uuid(root_item.received_uuid)) {
3543                         ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
3544                                                   root_item.received_uuid,
3545                                                  BTRFS_UUID_KEY_RECEIVED_SUBVOL,
3546                                                   key.objectid);
3547                         if (ret < 0) {
3548                                 pr_warn("btrfs: uuid_tree_add failed %d\n",
3549                                         ret);
3550                                 break;
3551                         }
3552                 }
3553
3554 skip:
3555                 if (trans) {
3556                         ret = btrfs_end_transaction(trans, fs_info->uuid_root);
3557                         trans = NULL;
3558                         if (ret)
3559                                 break;
3560                 }
3561
3562                 btrfs_release_path(path);
3563                 if (key.offset < (u64)-1) {
3564                         key.offset++;
3565                 } else if (key.type < BTRFS_ROOT_ITEM_KEY) {
3566                         key.offset = 0;
3567                         key.type = BTRFS_ROOT_ITEM_KEY;
3568                 } else if (key.objectid < (u64)-1) {
3569                         key.offset = 0;
3570                         key.type = BTRFS_ROOT_ITEM_KEY;
3571                         key.objectid++;
3572                 } else {
3573                         break;
3574                 }
3575                 cond_resched();
3576         }
3577
3578 out:
3579         btrfs_free_path(path);
3580         if (trans && !IS_ERR(trans))
3581                 btrfs_end_transaction(trans, fs_info->uuid_root);
3582         if (ret)
3583                 pr_warn("btrfs: btrfs_uuid_scan_kthread failed %d\n", ret);
3584         else
3585                 fs_info->update_uuid_tree_gen = 1;
3586         up(&fs_info->uuid_tree_rescan_sem);
3587         return 0;
3588 }
3589
3590 /*
3591  * Callback for btrfs_uuid_tree_iterate().
3592  * returns:
3593  * 0    check succeeded, the entry is not outdated.
3594  * < 0  if an error occured.
3595  * > 0  if the check failed, which means the caller shall remove the entry.
3596  */
3597 static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info *fs_info,
3598                                        u8 *uuid, u8 type, u64 subid)
3599 {
3600         struct btrfs_key key;
3601         int ret = 0;
3602         struct btrfs_root *subvol_root;
3603
3604         if (type != BTRFS_UUID_KEY_SUBVOL &&
3605             type != BTRFS_UUID_KEY_RECEIVED_SUBVOL)
3606                 goto out;
3607
3608         key.objectid = subid;
3609         key.type = BTRFS_ROOT_ITEM_KEY;
3610         key.offset = (u64)-1;
3611         subvol_root = btrfs_read_fs_root_no_name(fs_info, &key);
3612         if (IS_ERR(subvol_root)) {
3613                 ret = PTR_ERR(subvol_root);
3614                 if (ret == -ENOENT)
3615                         ret = 1;
3616                 goto out;
3617         }
3618
3619         switch (type) {
3620         case BTRFS_UUID_KEY_SUBVOL:
3621                 if (memcmp(uuid, subvol_root->root_item.uuid, BTRFS_UUID_SIZE))
3622                         ret = 1;
3623                 break;
3624         case BTRFS_UUID_KEY_RECEIVED_SUBVOL:
3625                 if (memcmp(uuid, subvol_root->root_item.received_uuid,
3626                            BTRFS_UUID_SIZE))
3627                         ret = 1;
3628                 break;
3629         }
3630
3631 out:
3632         return ret;
3633 }
3634
3635 static int btrfs_uuid_rescan_kthread(void *data)
3636 {
3637         struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data;
3638         int ret;
3639
3640         /*
3641          * 1st step is to iterate through the existing UUID tree and
3642          * to delete all entries that contain outdated data.
3643          * 2nd step is to add all missing entries to the UUID tree.
3644          */
3645         ret = btrfs_uuid_tree_iterate(fs_info, btrfs_check_uuid_tree_entry);
3646         if (ret < 0) {
3647                 pr_warn("btrfs: iterating uuid_tree failed %d\n", ret);
3648                 up(&fs_info->uuid_tree_rescan_sem);
3649                 return ret;
3650         }
3651         return btrfs_uuid_scan_kthread(data);
3652 }
3653
3654 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info)
3655 {
3656         struct btrfs_trans_handle *trans;
3657         struct btrfs_root *tree_root = fs_info->tree_root;
3658         struct btrfs_root *uuid_root;
3659         struct task_struct *task;
3660         int ret;
3661
3662         /*
3663          * 1 - root node
3664          * 1 - root item
3665          */
3666         trans = btrfs_start_transaction(tree_root, 2);
3667         if (IS_ERR(trans))
3668                 return PTR_ERR(trans);
3669
3670         uuid_root = btrfs_create_tree(trans, fs_info,
3671                                       BTRFS_UUID_TREE_OBJECTID);
3672         if (IS_ERR(uuid_root)) {
3673                 btrfs_abort_transaction(trans, tree_root,
3674                                         PTR_ERR(uuid_root));
3675                 return PTR_ERR(uuid_root);
3676         }
3677
3678         fs_info->uuid_root = uuid_root;
3679
3680         ret = btrfs_commit_transaction(trans, tree_root);
3681         if (ret)
3682                 return ret;
3683
3684         down(&fs_info->uuid_tree_rescan_sem);
3685         task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid");
3686         if (IS_ERR(task)) {
3687                 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
3688                 pr_warn("btrfs: failed to start uuid_scan task\n");
3689                 up(&fs_info->uuid_tree_rescan_sem);
3690                 return PTR_ERR(task);
3691         }
3692
3693         return 0;
3694 }
3695
3696 int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
3697 {
3698         struct task_struct *task;
3699
3700         down(&fs_info->uuid_tree_rescan_sem);
3701         task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
3702         if (IS_ERR(task)) {
3703                 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
3704                 pr_warn("btrfs: failed to start uuid_rescan task\n");
3705                 up(&fs_info->uuid_tree_rescan_sem);
3706                 return PTR_ERR(task);
3707         }
3708
3709         return 0;
3710 }
3711
3712 /*
3713  * shrinking a device means finding all of the device extents past
3714  * the new size, and then following the back refs to the chunks.
3715  * The chunk relocation code actually frees the device extent
3716  */
3717 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
3718 {
3719         struct btrfs_trans_handle *trans;
3720         struct btrfs_root *root = device->dev_root;
3721         struct btrfs_dev_extent *dev_extent = NULL;
3722         struct btrfs_path *path;
3723         u64 length;
3724         u64 chunk_tree;
3725         u64 chunk_objectid;
3726         u64 chunk_offset;
3727         int ret;
3728         int slot;
3729         int failed = 0;
3730         bool retried = false;
3731         struct extent_buffer *l;
3732         struct btrfs_key key;
3733         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
3734         u64 old_total = btrfs_super_total_bytes(super_copy);
3735         u64 old_size = device->total_bytes;
3736         u64 diff = device->total_bytes - new_size;
3737
3738         if (device->is_tgtdev_for_dev_replace)
3739                 return -EINVAL;
3740
3741         path = btrfs_alloc_path();
3742         if (!path)
3743                 return -ENOMEM;
3744
3745         path->reada = 2;
3746
3747         lock_chunks(root);
3748
3749         device->total_bytes = new_size;
3750         if (device->writeable) {
3751                 device->fs_devices->total_rw_bytes -= diff;
3752                 spin_lock(&root->fs_info->free_chunk_lock);
3753                 root->fs_info->free_chunk_space -= diff;
3754                 spin_unlock(&root->fs_info->free_chunk_lock);
3755         }
3756         unlock_chunks(root);
3757
3758 again:
3759         key.objectid = device->devid;
3760         key.offset = (u64)-1;
3761         key.type = BTRFS_DEV_EXTENT_KEY;
3762
3763         do {
3764                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3765                 if (ret < 0)
3766                         goto done;
3767
3768                 ret = btrfs_previous_item(root, path, 0, key.type);
3769                 if (ret < 0)
3770                         goto done;
3771                 if (ret) {
3772                         ret = 0;
3773                         btrfs_release_path(path);
3774                         break;
3775                 }
3776
3777                 l = path->nodes[0];
3778                 slot = path->slots[0];
3779                 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
3780
3781                 if (key.objectid != device->devid) {
3782                         btrfs_release_path(path);
3783                         break;
3784                 }
3785
3786                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
3787                 length = btrfs_dev_extent_length(l, dev_extent);
3788
3789                 if (key.offset + length <= new_size) {
3790                         btrfs_release_path(path);
3791                         break;
3792                 }
3793
3794                 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
3795                 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
3796                 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
3797                 btrfs_release_path(path);
3798
3799                 ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
3800                                            chunk_offset);
3801                 if (ret && ret != -ENOSPC)
3802                         goto done;
3803                 if (ret == -ENOSPC)
3804                         failed++;
3805         } while (key.offset-- > 0);
3806
3807         if (failed && !retried) {
3808                 failed = 0;
3809                 retried = true;
3810                 goto again;
3811         } else if (failed && retried) {
3812                 ret = -ENOSPC;
3813                 lock_chunks(root);
3814
3815                 device->total_bytes = old_size;
3816                 if (device->writeable)
3817                         device->fs_devices->total_rw_bytes += diff;
3818                 spin_lock(&root->fs_info->free_chunk_lock);
3819                 root->fs_info->free_chunk_space += diff;
3820                 spin_unlock(&root->fs_info->free_chunk_lock);
3821                 unlock_chunks(root);
3822                 goto done;
3823         }
3824
3825         /* Shrinking succeeded, else we would be at "done". */
3826         trans = btrfs_start_transaction(root, 0);
3827         if (IS_ERR(trans)) {
3828                 ret = PTR_ERR(trans);
3829                 goto done;
3830         }
3831
3832         lock_chunks(root);
3833
3834         device->disk_total_bytes = new_size;
3835         /* Now btrfs_update_device() will change the on-disk size. */
3836         ret = btrfs_update_device(trans, device);
3837         if (ret) {
3838                 unlock_chunks(root);
3839                 btrfs_end_transaction(trans, root);
3840                 goto done;
3841         }
3842         WARN_ON(diff > old_total);
3843         btrfs_set_super_total_bytes(super_copy, old_total - diff);
3844         unlock_chunks(root);
3845         btrfs_end_transaction(trans, root);
3846 done:
3847         btrfs_free_path(path);
3848         return ret;
3849 }
3850
3851 static int btrfs_add_system_chunk(struct btrfs_root *root,
3852                            struct btrfs_key *key,
3853                            struct btrfs_chunk *chunk, int item_size)
3854 {
3855         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
3856         struct btrfs_disk_key disk_key;
3857         u32 array_size;
3858         u8 *ptr;
3859
3860         array_size = btrfs_super_sys_array_size(super_copy);
3861         if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
3862                 return -EFBIG;
3863
3864         ptr = super_copy->sys_chunk_array + array_size;
3865         btrfs_cpu_key_to_disk(&disk_key, key);
3866         memcpy(ptr, &disk_key, sizeof(disk_key));
3867         ptr += sizeof(disk_key);
3868         memcpy(ptr, chunk, item_size);
3869         item_size += sizeof(disk_key);
3870         btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
3871         return 0;
3872 }
3873
3874 /*
3875  * sort the devices in descending order by max_avail, total_avail
3876  */
3877 static int btrfs_cmp_device_info(const void *a, const void *b)
3878 {
3879         const struct btrfs_device_info *di_a = a;
3880         const struct btrfs_device_info *di_b = b;
3881
3882         if (di_a->max_avail > di_b->max_avail)
3883                 return -1;
3884         if (di_a->max_avail < di_b->max_avail)
3885                 return 1;
3886         if (di_a->total_avail > di_b->total_avail)
3887                 return -1;
3888         if (di_a->total_avail < di_b->total_avail)
3889                 return 1;
3890         return 0;
3891 }
3892
3893 static struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = {
3894         [BTRFS_RAID_RAID10] = {
3895                 .sub_stripes    = 2,
3896                 .dev_stripes    = 1,
3897                 .devs_max       = 0,    /* 0 == as many as possible */
3898                 .devs_min       = 4,
3899                 .devs_increment = 2,
3900                 .ncopies        = 2,
3901         },
3902         [BTRFS_RAID_RAID1] = {
3903                 .sub_stripes    = 1,
3904                 .dev_stripes    = 1,
3905                 .devs_max       = 2,
3906                 .devs_min       = 2,
3907                 .devs_increment = 2,
3908                 .ncopies        = 2,
3909         },
3910         [BTRFS_RAID_DUP] = {
3911                 .sub_stripes    = 1,
3912                 .dev_stripes    = 2,
3913                 .devs_max       = 1,
3914                 .devs_min       = 1,
3915                 .devs_increment = 1,
3916                 .ncopies        = 2,
3917         },
3918         [BTRFS_RAID_RAID0] = {
3919                 .sub_stripes    = 1,
3920                 .dev_stripes    = 1,
3921                 .devs_max       = 0,
3922                 .devs_min       = 2,
3923                 .devs_increment = 1,
3924                 .ncopies        = 1,
3925         },
3926         [BTRFS_RAID_SINGLE] = {
3927                 .sub_stripes    = 1,
3928                 .dev_stripes    = 1,
3929                 .devs_max       = 1,
3930                 .devs_min       = 1,
3931                 .devs_increment = 1,
3932                 .ncopies        = 1,
3933         },
3934         [BTRFS_RAID_RAID5] = {
3935                 .sub_stripes    = 1,
3936                 .dev_stripes    = 1,
3937                 .devs_max       = 0,
3938                 .devs_min       = 2,
3939                 .devs_increment = 1,
3940                 .ncopies        = 2,
3941         },
3942         [BTRFS_RAID_RAID6] = {
3943                 .sub_stripes    = 1,
3944                 .dev_stripes    = 1,
3945                 .devs_max       = 0,
3946                 .devs_min       = 3,
3947                 .devs_increment = 1,
3948                 .ncopies        = 3,
3949         },
3950 };
3951
3952 static u32 find_raid56_stripe_len(u32 data_devices, u32 dev_stripe_target)
3953 {
3954         /* TODO allow them to set a preferred stripe size */
3955         return 64 * 1024;
3956 }
3957
3958 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
3959 {
3960         if (!(type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)))
3961                 return;
3962
3963         btrfs_set_fs_incompat(info, RAID56);
3964 }
3965
3966 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
3967                                struct btrfs_root *extent_root, u64 start,
3968                                u64 type)
3969 {
3970         struct btrfs_fs_info *info = extent_root->fs_info;
3971         struct btrfs_fs_devices *fs_devices = info->fs_devices;
3972         struct list_head *cur;
3973         struct map_lookup *map = NULL;
3974         struct extent_map_tree *em_tree;
3975         struct extent_map *em;
3976         struct btrfs_device_info *devices_info = NULL;
3977         u64 total_avail;
3978         int num_stripes;        /* total number of stripes to allocate */
3979         int data_stripes;       /* number of stripes that count for
3980                                    block group size */
3981         int sub_stripes;        /* sub_stripes info for map */
3982         int dev_stripes;        /* stripes per dev */
3983         int devs_max;           /* max devs to use */
3984         int devs_min;           /* min devs needed */
3985         int devs_increment;     /* ndevs has to be a multiple of this */
3986         int ncopies;            /* how many copies to data has */
3987         int ret;
3988         u64 max_stripe_size;
3989         u64 max_chunk_size;
3990         u64 stripe_size;
3991         u64 num_bytes;
3992         u64 raid_stripe_len = BTRFS_STRIPE_LEN;
3993         int ndevs;
3994         int i;
3995         int j;
3996         int index;
3997
3998         BUG_ON(!alloc_profile_is_valid(type, 0));
3999
4000         if (list_empty(&fs_devices->alloc_list))
4001                 return -ENOSPC;
4002
4003         index = __get_raid_index(type);
4004
4005         sub_stripes = btrfs_raid_array[index].sub_stripes;
4006         dev_stripes = btrfs_raid_array[index].dev_stripes;
4007         devs_max = btrfs_raid_array[index].devs_max;
4008         devs_min = btrfs_raid_array[index].devs_min;
4009         devs_increment = btrfs_raid_array[index].devs_increment;
4010         ncopies = btrfs_raid_array[index].ncopies;
4011
4012         if (type & BTRFS_BLOCK_GROUP_DATA) {
4013                 max_stripe_size = 1024 * 1024 * 1024;
4014                 max_chunk_size = 10 * max_stripe_size;
4015         } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
4016                 /* for larger filesystems, use larger metadata chunks */
4017                 if (fs_devices->total_rw_bytes > 50ULL * 1024 * 1024 * 1024)
4018                         max_stripe_size = 1024 * 1024 * 1024;
4019                 else
4020                         max_stripe_size = 256 * 1024 * 1024;
4021                 max_chunk_size = max_stripe_size;
4022         } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
4023                 max_stripe_size = 32 * 1024 * 1024;
4024                 max_chunk_size = 2 * max_stripe_size;
4025         } else {
4026                 printk(KERN_ERR "btrfs: invalid chunk type 0x%llx requested\n",
4027                        type);
4028                 BUG_ON(1);
4029         }
4030
4031         /* we don't want a chunk larger than 10% of writeable space */
4032         max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
4033                              max_chunk_size);
4034
4035         devices_info = kzalloc(sizeof(*devices_info) * fs_devices->rw_devices,
4036                                GFP_NOFS);
4037         if (!devices_info)
4038                 return -ENOMEM;
4039
4040         cur = fs_devices->alloc_list.next;
4041
4042         /*
4043          * in the first pass through the devices list, we gather information
4044          * about the available holes on each device.
4045          */
4046         ndevs = 0;
4047         while (cur != &fs_devices->alloc_list) {
4048                 struct btrfs_device *device;
4049                 u64 max_avail;
4050                 u64 dev_offset;
4051
4052                 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
4053
4054                 cur = cur->next;
4055
4056                 if (!device->writeable) {
4057                         WARN(1, KERN_ERR
4058                                "btrfs: read-only device in alloc_list\n");
4059                         continue;
4060                 }
4061
4062                 if (!device->in_fs_metadata ||
4063                     device->is_tgtdev_for_dev_replace)
4064                         continue;
4065
4066                 if (device->total_bytes > device->bytes_used)
4067                         total_avail = device->total_bytes - device->bytes_used;
4068                 else
4069                         total_avail = 0;
4070
4071                 /* If there is no space on this device, skip it. */
4072                 if (total_avail == 0)
4073                         continue;
4074
4075                 ret = find_free_dev_extent(trans, device,
4076                                            max_stripe_size * dev_stripes,
4077                                            &dev_offset, &max_avail);
4078                 if (ret && ret != -ENOSPC)
4079                         goto error;
4080
4081                 if (ret == 0)
4082                         max_avail = max_stripe_size * dev_stripes;
4083
4084                 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes)
4085                         continue;
4086
4087                 if (ndevs == fs_devices->rw_devices) {
4088                         WARN(1, "%s: found more than %llu devices\n",
4089                              __func__, fs_devices->rw_devices);
4090                         break;
4091                 }
4092                 devices_info[ndevs].dev_offset = dev_offset;
4093                 devices_info[ndevs].max_avail = max_avail;
4094                 devices_info[ndevs].total_avail = total_avail;
4095                 devices_info[ndevs].dev = device;
4096                 ++ndevs;
4097         }
4098
4099         /*
4100          * now sort the devices by hole size / available space
4101          */
4102         sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
4103              btrfs_cmp_device_info, NULL);
4104
4105         /* round down to number of usable stripes */
4106         ndevs -= ndevs % devs_increment;
4107
4108         if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) {
4109                 ret = -ENOSPC;
4110                 goto error;
4111         }
4112
4113         if (devs_max && ndevs > devs_max)
4114                 ndevs = devs_max;
4115         /*
4116          * the primary goal is to maximize the number of stripes, so use as many
4117          * devices as possible, even if the stripes are not maximum sized.
4118          */
4119         stripe_size = devices_info[ndevs-1].max_avail;
4120         num_stripes = ndevs * dev_stripes;
4121
4122         /*
4123          * this will have to be fixed for RAID1 and RAID10 over
4124          * more drives
4125          */
4126         data_stripes = num_stripes / ncopies;
4127
4128         if (type & BTRFS_BLOCK_GROUP_RAID5) {
4129                 raid_stripe_len = find_raid56_stripe_len(ndevs - 1,
4130                                  btrfs_super_stripesize(info->super_copy));
4131                 data_stripes = num_stripes - 1;
4132         }
4133         if (type & BTRFS_BLOCK_GROUP_RAID6) {
4134                 raid_stripe_len = find_raid56_stripe_len(ndevs - 2,
4135                                  btrfs_super_stripesize(info->super_copy));
4136                 data_stripes = num_stripes - 2;
4137         }
4138
4139         /*
4140          * Use the number of data stripes to figure out how big this chunk
4141          * is really going to be in terms of logical address space,
4142          * and compare that answer with the max chunk size
4143          */
4144         if (stripe_size * data_stripes > max_chunk_size) {
4145                 u64 mask = (1ULL << 24) - 1;
4146                 stripe_size = max_chunk_size;
4147                 do_div(stripe_size, data_stripes);
4148
4149                 /* bump the answer up to a 16MB boundary */
4150                 stripe_size = (stripe_size + mask) & ~mask;
4151
4152                 /* but don't go higher than the limits we found
4153                  * while searching for free extents
4154                  */
4155                 if (stripe_size > devices_info[ndevs-1].max_avail)
4156                         stripe_size = devices_info[ndevs-1].max_avail;
4157         }
4158
4159         do_div(stripe_size, dev_stripes);
4160
4161         /* align to BTRFS_STRIPE_LEN */
4162         do_div(stripe_size, raid_stripe_len);
4163         stripe_size *= raid_stripe_len;
4164
4165         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
4166         if (!map) {
4167                 ret = -ENOMEM;
4168                 goto error;
4169         }
4170         map->num_stripes = num_stripes;
4171
4172         for (i = 0; i < ndevs; ++i) {
4173                 for (j = 0; j < dev_stripes; ++j) {
4174                         int s = i * dev_stripes + j;
4175                         map->stripes[s].dev = devices_info[i].dev;
4176                         map->stripes[s].physical = devices_info[i].dev_offset +
4177                                                    j * stripe_size;
4178                 }
4179         }
4180         map->sector_size = extent_root->sectorsize;
4181         map->stripe_len = raid_stripe_len;
4182         map->io_align = raid_stripe_len;
4183         map->io_width = raid_stripe_len;
4184         map->type = type;
4185         map->sub_stripes = sub_stripes;
4186
4187         num_bytes = stripe_size * data_stripes;
4188
4189         trace_btrfs_chunk_alloc(info->chunk_root, map, start, num_bytes);
4190
4191         em = alloc_extent_map();
4192         if (!em) {
4193                 ret = -ENOMEM;
4194                 goto error;
4195         }
4196         em->bdev = (struct block_device *)map;
4197         em->start = start;
4198         em->len = num_bytes;
4199         em->block_start = 0;
4200         em->block_len = em->len;
4201         em->orig_block_len = stripe_size;
4202
4203         em_tree = &extent_root->fs_info->mapping_tree.map_tree;
4204         write_lock(&em_tree->lock);
4205         ret = add_extent_mapping(em_tree, em, 0);
4206         if (!ret) {
4207                 list_add_tail(&em->list, &trans->transaction->pending_chunks);
4208                 atomic_inc(&em->refs);
4209         }
4210         write_unlock(&em_tree->lock);
4211         if (ret) {
4212                 free_extent_map(em);
4213                 goto error;
4214         }
4215
4216         ret = btrfs_make_block_group(trans, extent_root, 0, type,
4217                                      BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4218                                      start, num_bytes);
4219         if (ret)
4220                 goto error_del_extent;
4221
4222         free_extent_map(em);
4223         check_raid56_incompat_flag(extent_root->fs_info, type);
4224
4225         kfree(devices_info);
4226         return 0;
4227
4228 error_del_extent:
4229         write_lock(&em_tree->lock);
4230         remove_extent_mapping(em_tree, em);
4231         write_unlock(&em_tree->lock);
4232
4233         /* One for our allocation */
4234         free_extent_map(em);
4235         /* One for the tree reference */
4236         free_extent_map(em);
4237 error:
4238         kfree(map);
4239         kfree(devices_info);
4240         return ret;
4241 }
4242
4243 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans,
4244                                 struct btrfs_root *extent_root,
4245                                 u64 chunk_offset, u64 chunk_size)
4246 {
4247         struct btrfs_key key;
4248         struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
4249         struct btrfs_device *device;
4250         struct btrfs_chunk *chunk;
4251         struct btrfs_stripe *stripe;
4252         struct extent_map_tree *em_tree;
4253         struct extent_map *em;
4254         struct map_lookup *map;
4255         size_t item_size;
4256         u64 dev_offset;
4257         u64 stripe_size;
4258         int i = 0;
4259         int ret;
4260
4261         em_tree = &extent_root->fs_info->mapping_tree.map_tree;
4262         read_lock(&em_tree->lock);
4263         em = lookup_extent_mapping(em_tree, chunk_offset, chunk_size);
4264         read_unlock(&em_tree->lock);
4265
4266         if (!em) {
4267                 btrfs_crit(extent_root->fs_info, "unable to find logical "
4268                            "%Lu len %Lu", chunk_offset, chunk_size);
4269                 return -EINVAL;
4270         }
4271
4272         if (em->start != chunk_offset || em->len != chunk_size) {
4273                 btrfs_crit(extent_root->fs_info, "found a bad mapping, wanted"
4274                           " %Lu-%Lu, found %Lu-%Lu\n", chunk_offset,
4275                           chunk_size, em->start, em->len);
4276                 free_extent_map(em);
4277                 return -EINVAL;
4278         }
4279
4280         map = (struct map_lookup *)em->bdev;
4281         item_size = btrfs_chunk_item_size(map->num_stripes);
4282         stripe_size = em->orig_block_len;
4283
4284         chunk = kzalloc(item_size, GFP_NOFS);
4285         if (!chunk) {
4286                 ret = -ENOMEM;
4287                 goto out;
4288         }
4289
4290         for (i = 0; i < map->num_stripes; i++) {
4291                 device = map->stripes[i].dev;
4292                 dev_offset = map->stripes[i].physical;
4293
4294                 device->bytes_used += stripe_size;
4295                 ret = btrfs_update_device(trans, device);
4296                 if (ret)
4297                         goto out;
4298                 ret = btrfs_alloc_dev_extent(trans, device,
4299                                              chunk_root->root_key.objectid,
4300                                              BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4301                                              chunk_offset, dev_offset,
4302                                              stripe_size);
4303                 if (ret)
4304                         goto out;
4305         }
4306
4307         spin_lock(&extent_root->fs_info->free_chunk_lock);
4308         extent_root->fs_info->free_chunk_space -= (stripe_size *
4309                                                    map->num_stripes);
4310         spin_unlock(&extent_root->fs_info->free_chunk_lock);
4311
4312         stripe = &chunk->stripe;
4313         for (i = 0; i < map->num_stripes; i++) {
4314                 device = map->stripes[i].dev;
4315                 dev_offset = map->stripes[i].physical;
4316
4317                 btrfs_set_stack_stripe_devid(stripe, device->devid);
4318                 btrfs_set_stack_stripe_offset(stripe, dev_offset);
4319                 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
4320                 stripe++;
4321         }
4322
4323         btrfs_set_stack_chunk_length(chunk, chunk_size);
4324         btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
4325         btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
4326         btrfs_set_stack_chunk_type(chunk, map->type);
4327         btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
4328         btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
4329         btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
4330         btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
4331         btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
4332
4333         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
4334         key.type = BTRFS_CHUNK_ITEM_KEY;
4335         key.offset = chunk_offset;
4336
4337         ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
4338         if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
4339                 /*
4340                  * TODO: Cleanup of inserted chunk root in case of
4341                  * failure.
4342                  */
4343                 ret = btrfs_add_system_chunk(chunk_root, &key, chunk,
4344                                              item_size);
4345         }
4346
4347 out:
4348         kfree(chunk);
4349         free_extent_map(em);
4350         return ret;
4351 }
4352
4353 /*
4354  * Chunk allocation falls into two parts. The first part does works
4355  * that make the new allocated chunk useable, but not do any operation
4356  * that modifies the chunk tree. The second part does the works that
4357  * require modifying the chunk tree. This division is important for the
4358  * bootstrap process of adding storage to a seed btrfs.
4359  */
4360 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4361                       struct btrfs_root *extent_root, u64 type)
4362 {
4363         u64 chunk_offset;
4364
4365         chunk_offset = find_next_chunk(extent_root->fs_info);
4366         return __btrfs_alloc_chunk(trans, extent_root, chunk_offset, type);
4367 }
4368
4369 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
4370                                          struct btrfs_root *root,
4371                                          struct btrfs_device *device)
4372 {
4373         u64 chunk_offset;
4374         u64 sys_chunk_offset;
4375         u64 alloc_profile;
4376         struct btrfs_fs_info *fs_info = root->fs_info;
4377         struct btrfs_root *extent_root = fs_info->extent_root;
4378         int ret;
4379
4380         chunk_offset = find_next_chunk(fs_info);
4381         alloc_profile = btrfs_get_alloc_profile(extent_root, 0);
4382         ret = __btrfs_alloc_chunk(trans, extent_root, chunk_offset,
4383                                   alloc_profile);
4384         if (ret)
4385                 return ret;
4386
4387         sys_chunk_offset = find_next_chunk(root->fs_info);
4388         alloc_profile = btrfs_get_alloc_profile(fs_info->chunk_root, 0);
4389         ret = __btrfs_alloc_chunk(trans, extent_root, sys_chunk_offset,
4390                                   alloc_profile);
4391         if (ret) {
4392                 btrfs_abort_transaction(trans, root, ret);
4393                 goto out;
4394         }
4395
4396         ret = btrfs_add_device(trans, fs_info->chunk_root, device);
4397         if (ret)
4398                 btrfs_abort_transaction(trans, root, ret);
4399 out:
4400         return ret;
4401 }
4402
4403 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
4404 {
4405         struct extent_map *em;
4406         struct map_lookup *map;
4407         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
4408         int readonly = 0;
4409         int i;
4410
4411         read_lock(&map_tree->map_tree.lock);
4412         em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
4413         read_unlock(&map_tree->map_tree.lock);
4414         if (!em)
4415                 return 1;
4416
4417         if (btrfs_test_opt(root, DEGRADED)) {
4418                 free_extent_map(em);
4419                 return 0;
4420         }
4421
4422         map = (struct map_lookup *)em->bdev;
4423         for (i = 0; i < map->num_stripes; i++) {
4424                 if (!map->stripes[i].dev->writeable) {
4425                         readonly = 1;
4426                         break;
4427                 }
4428         }
4429         free_extent_map(em);
4430         return readonly;
4431 }
4432
4433 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
4434 {
4435         extent_map_tree_init(&tree->map_tree);
4436 }
4437
4438 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
4439 {
4440         struct extent_map *em;
4441
4442         while (1) {
4443                 write_lock(&tree->map_tree.lock);
4444                 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
4445                 if (em)
4446                         remove_extent_mapping(&tree->map_tree, em);
4447                 write_unlock(&tree->map_tree.lock);
4448                 if (!em)
4449                         break;
4450                 kfree(em->bdev);
4451                 /* once for us */
4452                 free_extent_map(em);
4453                 /* once for the tree */
4454                 free_extent_map(em);
4455         }
4456 }
4457
4458 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
4459 {
4460         struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
4461         struct extent_map *em;
4462         struct map_lookup *map;
4463         struct extent_map_tree *em_tree = &map_tree->map_tree;
4464         int ret;
4465
4466         read_lock(&em_tree->lock);
4467         em = lookup_extent_mapping(em_tree, logical, len);
4468         read_unlock(&em_tree->lock);
4469
4470         /*
4471          * We could return errors for these cases, but that could get ugly and
4472          * we'd probably do the same thing which is just not do anything else
4473          * and exit, so return 1 so the callers don't try to use other copies.
4474          */
4475         if (!em) {
4476                 btrfs_crit(fs_info, "No mapping for %Lu-%Lu\n", logical,
4477                             logical+len);
4478                 return 1;
4479         }
4480
4481         if (em->start > logical || em->start + em->len < logical) {
4482                 btrfs_crit(fs_info, "Invalid mapping for %Lu-%Lu, got "
4483                             "%Lu-%Lu\n", logical, logical+len, em->start,
4484                             em->start + em->len);
4485                 return 1;
4486         }
4487
4488         map = (struct map_lookup *)em->bdev;
4489         if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
4490                 ret = map->num_stripes;
4491         else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
4492                 ret = map->sub_stripes;
4493         else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
4494                 ret = 2;
4495         else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
4496                 ret = 3;
4497         else
4498                 ret = 1;
4499         free_extent_map(em);
4500
4501         btrfs_dev_replace_lock(&fs_info->dev_replace);
4502         if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace))
4503                 ret++;
4504         btrfs_dev_replace_unlock(&fs_info->dev_replace);
4505
4506         return ret;
4507 }
4508
4509 unsigned long btrfs_full_stripe_len(struct btrfs_root *root,
4510                                     struct btrfs_mapping_tree *map_tree,
4511                                     u64 logical)
4512 {
4513         struct extent_map *em;
4514         struct map_lookup *map;
4515         struct extent_map_tree *em_tree = &map_tree->map_tree;
4516         unsigned long len = root->sectorsize;
4517
4518         read_lock(&em_tree->lock);
4519         em = lookup_extent_mapping(em_tree, logical, len);
4520         read_unlock(&em_tree->lock);
4521         BUG_ON(!em);
4522
4523         BUG_ON(em->start > logical || em->start + em->len < logical);
4524         map = (struct map_lookup *)em->bdev;
4525         if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
4526                          BTRFS_BLOCK_GROUP_RAID6)) {
4527                 len = map->stripe_len * nr_data_stripes(map);
4528         }
4529         free_extent_map(em);
4530         return len;
4531 }
4532
4533 int btrfs_is_parity_mirror(struct btrfs_mapping_tree *map_tree,
4534                            u64 logical, u64 len, int mirror_num)
4535 {
4536         struct extent_map *em;
4537         struct map_lookup *map;
4538         struct extent_map_tree *em_tree = &map_tree->map_tree;
4539         int ret = 0;
4540
4541         read_lock(&em_tree->lock);
4542         em = lookup_extent_mapping(em_tree, logical, len);
4543         read_unlock(&em_tree->lock);
4544         BUG_ON(!em);
4545
4546         BUG_ON(em->start > logical || em->start + em->len < logical);
4547         map = (struct map_lookup *)em->bdev;
4548         if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
4549                          BTRFS_BLOCK_GROUP_RAID6))
4550                 ret = 1;
4551         free_extent_map(em);
4552         return ret;
4553 }
4554
4555 static int find_live_mirror(struct btrfs_fs_info *fs_info,
4556                             struct map_lookup *map, int first, int num,
4557                             int optimal, int dev_replace_is_ongoing)
4558 {
4559         int i;
4560         int tolerance;
4561         struct btrfs_device *srcdev;
4562
4563         if (dev_replace_is_ongoing &&
4564             fs_info->dev_replace.cont_reading_from_srcdev_mode ==
4565              BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
4566                 srcdev = fs_info->dev_replace.srcdev;
4567         else
4568                 srcdev = NULL;
4569
4570         /*
4571          * try to avoid the drive that is the source drive for a
4572          * dev-replace procedure, only choose it if no other non-missing
4573          * mirror is available
4574          */
4575         for (tolerance = 0; tolerance < 2; tolerance++) {
4576                 if (map->stripes[optimal].dev->bdev &&
4577                     (tolerance || map->stripes[optimal].dev != srcdev))
4578                         return optimal;
4579                 for (i = first; i < first + num; i++) {
4580                         if (map->stripes[i].dev->bdev &&
4581                             (tolerance || map->stripes[i].dev != srcdev))
4582                                 return i;
4583                 }
4584         }
4585
4586         /* we couldn't find one that doesn't fail.  Just return something
4587          * and the io error handling code will clean up eventually
4588          */
4589         return optimal;
4590 }
4591
4592 static inline int parity_smaller(u64 a, u64 b)
4593 {
4594         return a > b;
4595 }
4596
4597 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
4598 static void sort_parity_stripes(struct btrfs_bio *bbio, u64 *raid_map)
4599 {
4600         struct btrfs_bio_stripe s;
4601         int i;
4602         u64 l;
4603         int again = 1;
4604
4605         while (again) {
4606                 again = 0;
4607                 for (i = 0; i < bbio->num_stripes - 1; i++) {
4608                         if (parity_smaller(raid_map[i], raid_map[i+1])) {
4609                                 s = bbio->stripes[i];
4610                                 l = raid_map[i];
4611                                 bbio->stripes[i] = bbio->stripes[i+1];
4612                                 raid_map[i] = raid_map[i+1];
4613                                 bbio->stripes[i+1] = s;
4614                                 raid_map[i+1] = l;
4615                                 again = 1;
4616                         }
4617                 }
4618         }
4619 }
4620
4621 static int __btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
4622                              u64 logical, u64 *length,
4623                              struct btrfs_bio **bbio_ret,
4624                              int mirror_num, u64 **raid_map_ret)
4625 {
4626         struct extent_map *em;
4627         struct map_lookup *map;
4628         struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
4629         struct extent_map_tree *em_tree = &map_tree->map_tree;
4630         u64 offset;
4631         u64 stripe_offset;
4632         u64 stripe_end_offset;
4633         u64 stripe_nr;
4634         u64 stripe_nr_orig;
4635         u64 stripe_nr_end;
4636         u64 stripe_len;
4637         u64 *raid_map = NULL;
4638         int stripe_index;
4639         int i;
4640         int ret = 0;
4641         int num_stripes;
4642         int max_errors = 0;
4643         struct btrfs_bio *bbio = NULL;
4644         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
4645         int dev_replace_is_ongoing = 0;
4646         int num_alloc_stripes;
4647         int patch_the_first_stripe_for_dev_replace = 0;
4648         u64 physical_to_patch_in_first_stripe = 0;
4649         u64 raid56_full_stripe_start = (u64)-1;
4650
4651         read_lock(&em_tree->lock);
4652         em = lookup_extent_mapping(em_tree, logical, *length);
4653         read_unlock(&em_tree->lock);
4654
4655         if (!em) {
4656                 btrfs_crit(fs_info, "unable to find logical %llu len %llu",
4657                         logical, *length);
4658                 return -EINVAL;
4659         }
4660
4661         if (em->start > logical || em->start + em->len < logical) {
4662                 btrfs_crit(fs_info, "found a bad mapping, wanted %Lu, "
4663                            "found %Lu-%Lu\n", logical, em->start,
4664                            em->start + em->len);
4665                 return -EINVAL;
4666         }
4667
4668         map = (struct map_lookup *)em->bdev;
4669         offset = logical - em->start;
4670
4671         stripe_len = map->stripe_len;
4672         stripe_nr = offset;
4673         /*
4674          * stripe_nr counts the total number of stripes we have to stride
4675          * to get to this block
4676          */
4677         do_div(stripe_nr, stripe_len);
4678
4679         stripe_offset = stripe_nr * stripe_len;
4680         BUG_ON(offset < stripe_offset);
4681
4682         /* stripe_offset is the offset of this block in its stripe*/
4683         stripe_offset = offset - stripe_offset;
4684
4685         /* if we're here for raid56, we need to know the stripe aligned start */
4686         if (map->type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)) {
4687                 unsigned long full_stripe_len = stripe_len * nr_data_stripes(map);
4688                 raid56_full_stripe_start = offset;
4689
4690                 /* allow a write of a full stripe, but make sure we don't
4691                  * allow straddling of stripes
4692                  */
4693                 do_div(raid56_full_stripe_start, full_stripe_len);
4694                 raid56_full_stripe_start *= full_stripe_len;
4695         }
4696
4697         if (rw & REQ_DISCARD) {
4698                 /* we don't discard raid56 yet */
4699                 if (map->type &
4700                     (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)) {
4701                         ret = -EOPNOTSUPP;
4702                         goto out;
4703                 }
4704                 *length = min_t(u64, em->len - offset, *length);
4705         } else if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
4706                 u64 max_len;
4707                 /* For writes to RAID[56], allow a full stripeset across all disks.
4708                    For other RAID types and for RAID[56] reads, just allow a single
4709                    stripe (on a single disk). */
4710                 if (map->type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6) &&
4711                     (rw & REQ_WRITE)) {
4712                         max_len = stripe_len * nr_data_stripes(map) -
4713                                 (offset - raid56_full_stripe_start);
4714                 } else {
4715                         /* we limit the length of each bio to what fits in a stripe */
4716                         max_len = stripe_len - stripe_offset;
4717                 }
4718                 *length = min_t(u64, em->len - offset, max_len);
4719         } else {
4720                 *length = em->len - offset;
4721         }
4722
4723         /* This is for when we're called from btrfs_merge_bio_hook() and all
4724            it cares about is the length */
4725         if (!bbio_ret)
4726                 goto out;
4727
4728         btrfs_dev_replace_lock(dev_replace);
4729         dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
4730         if (!dev_replace_is_ongoing)
4731                 btrfs_dev_replace_unlock(dev_replace);
4732
4733         if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
4734             !(rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) &&
4735             dev_replace->tgtdev != NULL) {
4736                 /*
4737                  * in dev-replace case, for repair case (that's the only
4738                  * case where the mirror is selected explicitly when
4739                  * calling btrfs_map_block), blocks left of the left cursor
4740                  * can also be read from the target drive.
4741                  * For REQ_GET_READ_MIRRORS, the target drive is added as
4742                  * the last one to the array of stripes. For READ, it also
4743                  * needs to be supported using the same mirror number.
4744                  * If the requested block is not left of the left cursor,
4745                  * EIO is returned. This can happen because btrfs_num_copies()
4746                  * returns one more in the dev-replace case.
4747                  */
4748                 u64 tmp_length = *length;
4749                 struct btrfs_bio *tmp_bbio = NULL;
4750                 int tmp_num_stripes;
4751                 u64 srcdev_devid = dev_replace->srcdev->devid;
4752                 int index_srcdev = 0;
4753                 int found = 0;
4754                 u64 physical_of_found = 0;
4755
4756                 ret = __btrfs_map_block(fs_info, REQ_GET_READ_MIRRORS,
4757                              logical, &tmp_length, &tmp_bbio, 0, NULL);
4758                 if (ret) {
4759                         WARN_ON(tmp_bbio != NULL);
4760                         goto out;
4761                 }
4762
4763                 tmp_num_stripes = tmp_bbio->num_stripes;
4764                 if (mirror_num > tmp_num_stripes) {
4765                         /*
4766                          * REQ_GET_READ_MIRRORS does not contain this
4767                          * mirror, that means that the requested area
4768                          * is not left of the left cursor
4769                          */
4770                         ret = -EIO;
4771                         kfree(tmp_bbio);
4772                         goto out;
4773                 }
4774
4775                 /*
4776                  * process the rest of the function using the mirror_num
4777                  * of the source drive. Therefore look it up first.
4778                  * At the end, patch the device pointer to the one of the
4779                  * target drive.
4780                  */
4781                 for (i = 0; i < tmp_num_stripes; i++) {
4782                         if (tmp_bbio->stripes[i].dev->devid == srcdev_devid) {
4783                                 /*
4784                                  * In case of DUP, in order to keep it
4785                                  * simple, only add the mirror with the
4786                                  * lowest physical address
4787                                  */
4788                                 if (found &&
4789                                     physical_of_found <=
4790                                      tmp_bbio->stripes[i].physical)
4791                                         continue;
4792                                 index_srcdev = i;
4793                                 found = 1;
4794                                 physical_of_found =
4795                                         tmp_bbio->stripes[i].physical;
4796                         }
4797                 }
4798
4799                 if (found) {
4800                         mirror_num = index_srcdev + 1;
4801                         patch_the_first_stripe_for_dev_replace = 1;
4802                         physical_to_patch_in_first_stripe = physical_of_found;
4803                 } else {
4804                         WARN_ON(1);
4805                         ret = -EIO;
4806                         kfree(tmp_bbio);
4807                         goto out;
4808                 }
4809
4810                 kfree(tmp_bbio);
4811         } else if (mirror_num > map->num_stripes) {
4812                 mirror_num = 0;
4813         }
4814
4815         num_stripes = 1;
4816         stripe_index = 0;
4817         stripe_nr_orig = stripe_nr;
4818         stripe_nr_end = ALIGN(offset + *length, map->stripe_len);
4819         do_div(stripe_nr_end, map->stripe_len);
4820         stripe_end_offset = stripe_nr_end * map->stripe_len -
4821                             (offset + *length);
4822
4823         if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
4824                 if (rw & REQ_DISCARD)
4825                         num_stripes = min_t(u64, map->num_stripes,
4826                                             stripe_nr_end - stripe_nr_orig);
4827                 stripe_index = do_div(stripe_nr, map->num_stripes);
4828         } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
4829                 if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS))
4830                         num_stripes = map->num_stripes;
4831                 else if (mirror_num)
4832                         stripe_index = mirror_num - 1;
4833                 else {
4834                         stripe_index = find_live_mirror(fs_info, map, 0,
4835                                             map->num_stripes,
4836                                             current->pid % map->num_stripes,
4837                                             dev_replace_is_ongoing);
4838                         mirror_num = stripe_index + 1;
4839                 }
4840
4841         } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
4842                 if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) {
4843                         num_stripes = map->num_stripes;
4844                 } else if (mirror_num) {
4845                         stripe_index = mirror_num - 1;
4846                 } else {
4847                         mirror_num = 1;
4848                 }
4849
4850         } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
4851                 int factor = map->num_stripes / map->sub_stripes;
4852
4853                 stripe_index = do_div(stripe_nr, factor);
4854                 stripe_index *= map->sub_stripes;
4855
4856                 if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS))
4857                         num_stripes = map->sub_stripes;
4858                 else if (rw & REQ_DISCARD)
4859                         num_stripes = min_t(u64, map->sub_stripes *
4860                                             (stripe_nr_end - stripe_nr_orig),
4861                                             map->num_stripes);
4862                 else if (mirror_num)
4863                         stripe_index += mirror_num - 1;
4864                 else {
4865                         int old_stripe_index = stripe_index;
4866                         stripe_index = find_live_mirror(fs_info, map,
4867                                               stripe_index,
4868                                               map->sub_stripes, stripe_index +
4869                                               current->pid % map->sub_stripes,
4870                                               dev_replace_is_ongoing);
4871                         mirror_num = stripe_index - old_stripe_index + 1;
4872                 }
4873
4874         } else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
4875                                 BTRFS_BLOCK_GROUP_RAID6)) {
4876                 u64 tmp;
4877
4878                 if (bbio_ret && ((rw & REQ_WRITE) || mirror_num > 1)
4879                     && raid_map_ret) {
4880                         int i, rot;
4881
4882                         /* push stripe_nr back to the start of the full stripe */
4883                         stripe_nr = raid56_full_stripe_start;
4884                         do_div(stripe_nr, stripe_len);
4885
4886                         stripe_index = do_div(stripe_nr, nr_data_stripes(map));
4887
4888                         /* RAID[56] write or recovery. Return all stripes */
4889                         num_stripes = map->num_stripes;
4890                         max_errors = nr_parity_stripes(map);
4891
4892                         raid_map = kmalloc(sizeof(u64) * num_stripes,
4893                                            GFP_NOFS);
4894                         if (!raid_map) {
4895                                 ret = -ENOMEM;
4896                                 goto out;
4897                         }
4898
4899                         /* Work out the disk rotation on this stripe-set */
4900                         tmp = stripe_nr;
4901                         rot = do_div(tmp, num_stripes);
4902
4903                         /* Fill in the logical address of each stripe */
4904                         tmp = stripe_nr * nr_data_stripes(map);
4905                         for (i = 0; i < nr_data_stripes(map); i++)
4906                                 raid_map[(i+rot) % num_stripes] =
4907                                         em->start + (tmp + i) * map->stripe_len;
4908
4909                         raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE;
4910                         if (map->type & BTRFS_BLOCK_GROUP_RAID6)
4911                                 raid_map[(i+rot+1) % num_stripes] =
4912                                         RAID6_Q_STRIPE;
4913
4914                         *length = map->stripe_len;
4915                         stripe_index = 0;
4916                         stripe_offset = 0;
4917                 } else {
4918                         /*
4919                          * Mirror #0 or #1 means the original data block.
4920                          * Mirror #2 is RAID5 parity block.
4921                          * Mirror #3 is RAID6 Q block.
4922                          */
4923                         stripe_index = do_div(stripe_nr, nr_data_stripes(map));
4924                         if (mirror_num > 1)
4925                                 stripe_index = nr_data_stripes(map) +
4926                                                 mirror_num - 2;
4927
4928                         /* We distribute the parity blocks across stripes */
4929                         tmp = stripe_nr + stripe_index;
4930                         stripe_index = do_div(tmp, map->num_stripes);
4931                 }
4932         } else {
4933                 /*
4934                  * after this do_div call, stripe_nr is the number of stripes
4935                  * on this device we have to walk to find the data, and
4936                  * stripe_index is the number of our device in the stripe array
4937                  */
4938                 stripe_index = do_div(stripe_nr, map->num_stripes);
4939                 mirror_num = stripe_index + 1;
4940         }
4941         BUG_ON(stripe_index >= map->num_stripes);
4942
4943         num_alloc_stripes = num_stripes;
4944         if (dev_replace_is_ongoing) {
4945                 if (rw & (REQ_WRITE | REQ_DISCARD))
4946                         num_alloc_stripes <<= 1;
4947                 if (rw & REQ_GET_READ_MIRRORS)
4948                         num_alloc_stripes++;
4949         }
4950         bbio = kzalloc(btrfs_bio_size(num_alloc_stripes), GFP_NOFS);
4951         if (!bbio) {
4952                 kfree(raid_map);
4953                 ret = -ENOMEM;
4954                 goto out;
4955         }
4956         atomic_set(&bbio->error, 0);
4957
4958         if (rw & REQ_DISCARD) {
4959                 int factor = 0;
4960                 int sub_stripes = 0;
4961                 u64 stripes_per_dev = 0;
4962                 u32 remaining_stripes = 0;
4963                 u32 last_stripe = 0;
4964
4965                 if (map->type &
4966                     (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID10)) {
4967                         if (map->type & BTRFS_BLOCK_GROUP_RAID0)
4968                                 sub_stripes = 1;
4969                         else
4970                                 sub_stripes = map->sub_stripes;
4971
4972                         factor = map->num_stripes / sub_stripes;
4973                         stripes_per_dev = div_u64_rem(stripe_nr_end -
4974                                                       stripe_nr_orig,
4975                                                       factor,
4976                                                       &remaining_stripes);
4977                         div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
4978                         last_stripe *= sub_stripes;
4979                 }
4980
4981                 for (i = 0; i < num_stripes; i++) {
4982                         bbio->stripes[i].physical =
4983                                 map->stripes[stripe_index].physical +
4984                                 stripe_offset + stripe_nr * map->stripe_len;
4985                         bbio->stripes[i].dev = map->stripes[stripe_index].dev;
4986
4987                         if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
4988                                          BTRFS_BLOCK_GROUP_RAID10)) {
4989                                 bbio->stripes[i].length = stripes_per_dev *
4990                                                           map->stripe_len;
4991
4992                                 if (i / sub_stripes < remaining_stripes)
4993                                         bbio->stripes[i].length +=
4994                                                 map->stripe_len;
4995
4996                                 /*
4997                                  * Special for the first stripe and
4998                                  * the last stripe:
4999                                  *
5000                                  * |-------|...|-------|
5001                                  *     |----------|
5002                                  *    off     end_off
5003                                  */
5004                                 if (i < sub_stripes)
5005                                         bbio->stripes[i].length -=
5006                                                 stripe_offset;
5007
5008                                 if (stripe_index >= last_stripe &&
5009                                     stripe_index <= (last_stripe +
5010                                                      sub_stripes - 1))
5011                                         bbio->stripes[i].length -=
5012                                                 stripe_end_offset;
5013
5014                                 if (i == sub_stripes - 1)
5015                                         stripe_offset = 0;
5016                         } else
5017                                 bbio->stripes[i].length = *length;
5018
5019                         stripe_index++;
5020                         if (stripe_index == map->num_stripes) {
5021                                 /* This could only happen for RAID0/10 */
5022                                 stripe_index = 0;
5023                                 stripe_nr++;
5024                         }
5025                 }
5026         } else {
5027                 for (i = 0; i < num_stripes; i++) {
5028                         bbio->stripes[i].physical =
5029                                 map->stripes[stripe_index].physical +
5030                                 stripe_offset +
5031                                 stripe_nr * map->stripe_len;
5032                         bbio->stripes[i].dev =
5033                                 map->stripes[stripe_index].dev;
5034                         stripe_index++;
5035                 }
5036         }
5037
5038         if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS)) {
5039                 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
5040                                  BTRFS_BLOCK_GROUP_RAID10 |
5041                                  BTRFS_BLOCK_GROUP_RAID5 |
5042                                  BTRFS_BLOCK_GROUP_DUP)) {
5043                         max_errors = 1;
5044                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID6) {
5045                         max_errors = 2;
5046                 }
5047         }
5048
5049         if (dev_replace_is_ongoing && (rw & (REQ_WRITE | REQ_DISCARD)) &&
5050             dev_replace->tgtdev != NULL) {
5051                 int index_where_to_add;
5052                 u64 srcdev_devid = dev_replace->srcdev->devid;
5053
5054                 /*
5055                  * duplicate the write operations while the dev replace
5056                  * procedure is running. Since the copying of the old disk
5057                  * to the new disk takes place at run time while the
5058                  * filesystem is mounted writable, the regular write
5059                  * operations to the old disk have to be duplicated to go
5060                  * to the new disk as well.
5061                  * Note that device->missing is handled by the caller, and
5062                  * that the write to the old disk is already set up in the
5063                  * stripes array.
5064                  */
5065                 index_where_to_add = num_stripes;
5066                 for (i = 0; i < num_stripes; i++) {
5067                         if (bbio->stripes[i].dev->devid == srcdev_devid) {
5068                                 /* write to new disk, too */
5069                                 struct btrfs_bio_stripe *new =
5070                                         bbio->stripes + index_where_to_add;
5071                                 struct btrfs_bio_stripe *old =
5072                                         bbio->stripes + i;
5073
5074                                 new->physical = old->physical;
5075                                 new->length = old->length;
5076                                 new->dev = dev_replace->tgtdev;
5077                                 index_where_to_add++;
5078                                 max_errors++;
5079                         }
5080                 }
5081                 num_stripes = index_where_to_add;
5082         } else if (dev_replace_is_ongoing && (rw & REQ_GET_READ_MIRRORS) &&
5083                    dev_replace->tgtdev != NULL) {
5084                 u64 srcdev_devid = dev_replace->srcdev->devid;
5085                 int index_srcdev = 0;
5086                 int found = 0;
5087                 u64 physical_of_found = 0;
5088
5089                 /*
5090                  * During the dev-replace procedure, the target drive can
5091                  * also be used to read data in case it is needed to repair
5092                  * a corrupt block elsewhere. This is possible if the
5093                  * requested area is left of the left cursor. In this area,
5094                  * the target drive is a full copy of the source drive.
5095                  */
5096                 for (i = 0; i < num_stripes; i++) {
5097                         if (bbio->stripes[i].dev->devid == srcdev_devid) {
5098                                 /*
5099                                  * In case of DUP, in order to keep it
5100                                  * simple, only add the mirror with the
5101                                  * lowest physical address
5102                                  */
5103                                 if (found &&
5104                                     physical_of_found <=
5105                                      bbio->stripes[i].physical)
5106                                         continue;
5107                                 index_srcdev = i;
5108                                 found = 1;
5109                                 physical_of_found = bbio->stripes[i].physical;
5110                         }
5111                 }
5112                 if (found) {
5113                         u64 length = map->stripe_len;
5114
5115                         if (physical_of_found + length <=
5116                             dev_replace->cursor_left) {
5117                                 struct btrfs_bio_stripe *tgtdev_stripe =
5118                                         bbio->stripes + num_stripes;
5119
5120                                 tgtdev_stripe->physical = physical_of_found;
5121                                 tgtdev_stripe->length =
5122                                         bbio->stripes[index_srcdev].length;
5123                                 tgtdev_stripe->dev = dev_replace->tgtdev;
5124
5125                                 num_stripes++;
5126                         }
5127                 }
5128         }
5129
5130         *bbio_ret = bbio;
5131         bbio->num_stripes = num_stripes;
5132         bbio->max_errors = max_errors;
5133         bbio->mirror_num = mirror_num;
5134
5135         /*
5136          * this is the case that REQ_READ && dev_replace_is_ongoing &&
5137          * mirror_num == num_stripes + 1 && dev_replace target drive is
5138          * available as a mirror
5139          */
5140         if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) {
5141                 WARN_ON(num_stripes > 1);
5142                 bbio->stripes[0].dev = dev_replace->tgtdev;
5143                 bbio->stripes[0].physical = physical_to_patch_in_first_stripe;
5144                 bbio->mirror_num = map->num_stripes + 1;
5145         }
5146         if (raid_map) {
5147                 sort_parity_stripes(bbio, raid_map);
5148                 *raid_map_ret = raid_map;
5149         }
5150 out:
5151         if (dev_replace_is_ongoing)
5152                 btrfs_dev_replace_unlock(dev_replace);
5153         free_extent_map(em);
5154         return ret;
5155 }
5156
5157 int btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
5158                       u64 logical, u64 *length,
5159                       struct btrfs_bio **bbio_ret, int mirror_num)
5160 {
5161         return __btrfs_map_block(fs_info, rw, logical, length, bbio_ret,
5162                                  mirror_num, NULL);
5163 }
5164
5165 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
5166                      u64 chunk_start, u64 physical, u64 devid,
5167                      u64 **logical, int *naddrs, int *stripe_len)
5168 {
5169         struct extent_map_tree *em_tree = &map_tree->map_tree;
5170         struct extent_map *em;
5171         struct map_lookup *map;
5172         u64 *buf;
5173         u64 bytenr;
5174         u64 length;
5175         u64 stripe_nr;
5176         u64 rmap_len;
5177         int i, j, nr = 0;
5178
5179         read_lock(&em_tree->lock);
5180         em = lookup_extent_mapping(em_tree, chunk_start, 1);
5181         read_unlock(&em_tree->lock);
5182
5183         if (!em) {
5184                 printk(KERN_ERR "btrfs: couldn't find em for chunk %Lu\n",
5185                        chunk_start);
5186                 return -EIO;
5187         }
5188
5189         if (em->start != chunk_start) {
5190                 printk(KERN_ERR "btrfs: bad chunk start, em=%Lu, wanted=%Lu\n",
5191                        em->start, chunk_start);
5192                 free_extent_map(em);
5193                 return -EIO;
5194         }
5195         map = (struct map_lookup *)em->bdev;
5196
5197         length = em->len;
5198         rmap_len = map->stripe_len;
5199
5200         if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5201                 do_div(length, map->num_stripes / map->sub_stripes);
5202         else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5203                 do_div(length, map->num_stripes);
5204         else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
5205                               BTRFS_BLOCK_GROUP_RAID6)) {
5206                 do_div(length, nr_data_stripes(map));
5207                 rmap_len = map->stripe_len * nr_data_stripes(map);
5208         }
5209
5210         buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
5211         BUG_ON(!buf); /* -ENOMEM */
5212
5213         for (i = 0; i < map->num_stripes; i++) {
5214                 if (devid && map->stripes[i].dev->devid != devid)
5215                         continue;
5216                 if (map->stripes[i].physical > physical ||
5217                     map->stripes[i].physical + length <= physical)
5218                         continue;
5219
5220                 stripe_nr = physical - map->stripes[i].physical;
5221                 do_div(stripe_nr, map->stripe_len);
5222
5223                 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5224                         stripe_nr = stripe_nr * map->num_stripes + i;
5225                         do_div(stripe_nr, map->sub_stripes);
5226                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5227                         stripe_nr = stripe_nr * map->num_stripes + i;
5228                 } /* else if RAID[56], multiply by nr_data_stripes().
5229                    * Alternatively, just use rmap_len below instead of
5230                    * map->stripe_len */
5231
5232                 bytenr = chunk_start + stripe_nr * rmap_len;
5233                 WARN_ON(nr >= map->num_stripes);
5234                 for (j = 0; j < nr; j++) {
5235                         if (buf[j] == bytenr)
5236                                 break;
5237                 }
5238                 if (j == nr) {
5239                         WARN_ON(nr >= map->num_stripes);
5240                         buf[nr++] = bytenr;
5241                 }
5242         }
5243
5244         *logical = buf;
5245         *naddrs = nr;
5246         *stripe_len = rmap_len;
5247
5248         free_extent_map(em);
5249         return 0;
5250 }
5251
5252 static void btrfs_end_bio(struct bio *bio, int err)
5253 {
5254         struct btrfs_bio *bbio = bio->bi_private;
5255         int is_orig_bio = 0;
5256
5257         if (err) {
5258                 atomic_inc(&bbio->error);
5259                 if (err == -EIO || err == -EREMOTEIO) {
5260                         unsigned int stripe_index =
5261                                 btrfs_io_bio(bio)->stripe_index;
5262                         struct btrfs_device *dev;
5263
5264                         BUG_ON(stripe_index >= bbio->num_stripes);
5265                         dev = bbio->stripes[stripe_index].dev;
5266                         if (dev->bdev) {
5267                                 if (bio->bi_rw & WRITE)
5268                                         btrfs_dev_stat_inc(dev,
5269                                                 BTRFS_DEV_STAT_WRITE_ERRS);
5270                                 else
5271                                         btrfs_dev_stat_inc(dev,
5272                                                 BTRFS_DEV_STAT_READ_ERRS);
5273                                 if ((bio->bi_rw & WRITE_FLUSH) == WRITE_FLUSH)
5274                                         btrfs_dev_stat_inc(dev,
5275                                                 BTRFS_DEV_STAT_FLUSH_ERRS);
5276                                 btrfs_dev_stat_print_on_error(dev);
5277                         }
5278                 }
5279         }
5280
5281         if (bio == bbio->orig_bio)
5282                 is_orig_bio = 1;
5283
5284         if (atomic_dec_and_test(&bbio->stripes_pending)) {
5285                 if (!is_orig_bio) {
5286                         bio_put(bio);
5287                         bio = bbio->orig_bio;
5288                 }
5289                 bio->bi_private = bbio->private;
5290                 bio->bi_end_io = bbio->end_io;
5291                 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
5292                 /* only send an error to the higher layers if it is
5293                  * beyond the tolerance of the btrfs bio
5294                  */
5295                 if (atomic_read(&bbio->error) > bbio->max_errors) {
5296                         err = -EIO;
5297                 } else {
5298                         /*
5299                          * this bio is actually up to date, we didn't
5300                          * go over the max number of errors
5301                          */
5302                         set_bit(BIO_UPTODATE, &bio->bi_flags);
5303                         err = 0;
5304                 }
5305                 kfree(bbio);
5306
5307                 bio_endio(bio, err);
5308         } else if (!is_orig_bio) {
5309                 bio_put(bio);
5310         }
5311 }
5312
5313 struct async_sched {
5314         struct bio *bio;
5315         int rw;
5316         struct btrfs_fs_info *info;
5317         struct btrfs_work work;
5318 };
5319
5320 /*
5321  * see run_scheduled_bios for a description of why bios are collected for
5322  * async submit.
5323  *
5324  * This will add one bio to the pending list for a device and make sure
5325  * the work struct is scheduled.
5326  */
5327 static noinline void btrfs_schedule_bio(struct btrfs_root *root,
5328                                         struct btrfs_device *device,
5329                                         int rw, struct bio *bio)
5330 {
5331         int should_queue = 1;
5332         struct btrfs_pending_bios *pending_bios;
5333
5334         if (device->missing || !device->bdev) {
5335                 bio_endio(bio, -EIO);
5336                 return;
5337         }
5338
5339         /* don't bother with additional async steps for reads, right now */
5340         if (!(rw & REQ_WRITE)) {
5341                 bio_get(bio);
5342                 btrfsic_submit_bio(rw, bio);
5343                 bio_put(bio);
5344                 return;
5345         }
5346
5347         /*
5348          * nr_async_bios allows us to reliably return congestion to the
5349          * higher layers.  Otherwise, the async bio makes it appear we have
5350          * made progress against dirty pages when we've really just put it
5351          * on a queue for later
5352          */
5353         atomic_inc(&root->fs_info->nr_async_bios);
5354         WARN_ON(bio->bi_next);
5355         bio->bi_next = NULL;
5356         bio->bi_rw |= rw;
5357
5358         spin_lock(&device->io_lock);
5359         if (bio->bi_rw & REQ_SYNC)
5360                 pending_bios = &device->pending_sync_bios;
5361         else
5362                 pending_bios = &device->pending_bios;
5363
5364         if (pending_bios->tail)
5365                 pending_bios->tail->bi_next = bio;
5366
5367         pending_bios->tail = bio;
5368         if (!pending_bios->head)
5369                 pending_bios->head = bio;
5370         if (device->running_pending)
5371                 should_queue = 0;
5372
5373         spin_unlock(&device->io_lock);
5374
5375         if (should_queue)
5376                 btrfs_queue_worker(&root->fs_info->submit_workers,
5377                                    &device->work);
5378 }
5379
5380 static int bio_size_ok(struct block_device *bdev, struct bio *bio,
5381                        sector_t sector)
5382 {
5383         struct bio_vec *prev;
5384         struct request_queue *q = bdev_get_queue(bdev);
5385         unsigned short max_sectors = queue_max_sectors(q);
5386         struct bvec_merge_data bvm = {
5387                 .bi_bdev = bdev,
5388                 .bi_sector = sector,
5389                 .bi_rw = bio->bi_rw,
5390         };
5391
5392         if (bio->bi_vcnt == 0) {
5393                 WARN_ON(1);
5394                 return 1;
5395         }
5396
5397         prev = &bio->bi_io_vec[bio->bi_vcnt - 1];
5398         if (bio_sectors(bio) > max_sectors)
5399                 return 0;
5400
5401         if (!q->merge_bvec_fn)
5402                 return 1;
5403
5404         bvm.bi_size = bio->bi_size - prev->bv_len;
5405         if (q->merge_bvec_fn(q, &bvm, prev) < prev->bv_len)
5406                 return 0;
5407         return 1;
5408 }
5409
5410 static void submit_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio,
5411                               struct bio *bio, u64 physical, int dev_nr,
5412                               int rw, int async)
5413 {
5414         struct btrfs_device *dev = bbio->stripes[dev_nr].dev;
5415
5416         bio->bi_private = bbio;
5417         btrfs_io_bio(bio)->stripe_index = dev_nr;
5418         bio->bi_end_io = btrfs_end_bio;
5419         bio->bi_sector = physical >> 9;
5420 #ifdef DEBUG
5421         {
5422                 struct rcu_string *name;
5423
5424                 rcu_read_lock();
5425                 name = rcu_dereference(dev->name);
5426                 pr_debug("btrfs_map_bio: rw %d, sector=%llu, dev=%lu "
5427                          "(%s id %llu), size=%u\n", rw,
5428                          (u64)bio->bi_sector, (u_long)dev->bdev->bd_dev,
5429                          name->str, dev->devid, bio->bi_size);
5430                 rcu_read_unlock();
5431         }
5432 #endif
5433         bio->bi_bdev = dev->bdev;
5434         if (async)
5435                 btrfs_schedule_bio(root, dev, rw, bio);
5436         else
5437                 btrfsic_submit_bio(rw, bio);
5438 }
5439
5440 static int breakup_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio,
5441                               struct bio *first_bio, struct btrfs_device *dev,
5442                               int dev_nr, int rw, int async)
5443 {
5444         struct bio_vec *bvec = first_bio->bi_io_vec;
5445         struct bio *bio;
5446         int nr_vecs = bio_get_nr_vecs(dev->bdev);
5447         u64 physical = bbio->stripes[dev_nr].physical;
5448
5449 again:
5450         bio = btrfs_bio_alloc(dev->bdev, physical >> 9, nr_vecs, GFP_NOFS);
5451         if (!bio)
5452                 return -ENOMEM;
5453
5454         while (bvec <= (first_bio->bi_io_vec + first_bio->bi_vcnt - 1)) {
5455                 if (bio_add_page(bio, bvec->bv_page, bvec->bv_len,
5456                                  bvec->bv_offset) < bvec->bv_len) {
5457                         u64 len = bio->bi_size;
5458
5459                         atomic_inc(&bbio->stripes_pending);
5460                         submit_stripe_bio(root, bbio, bio, physical, dev_nr,
5461                                           rw, async);
5462                         physical += len;
5463                         goto again;
5464                 }
5465                 bvec++;
5466         }
5467
5468         submit_stripe_bio(root, bbio, bio, physical, dev_nr, rw, async);
5469         return 0;
5470 }
5471
5472 static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical)
5473 {
5474         atomic_inc(&bbio->error);
5475         if (atomic_dec_and_test(&bbio->stripes_pending)) {
5476                 bio->bi_private = bbio->private;
5477                 bio->bi_end_io = bbio->end_io;
5478                 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
5479                 bio->bi_sector = logical >> 9;
5480                 kfree(bbio);
5481                 bio_endio(bio, -EIO);
5482         }
5483 }
5484
5485 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
5486                   int mirror_num, int async_submit)
5487 {
5488         struct btrfs_device *dev;
5489         struct bio *first_bio = bio;
5490         u64 logical = (u64)bio->bi_sector << 9;
5491         u64 length = 0;
5492         u64 map_length;
5493         u64 *raid_map = NULL;
5494         int ret;
5495         int dev_nr = 0;
5496         int total_devs = 1;
5497         struct btrfs_bio *bbio = NULL;
5498
5499         length = bio->bi_size;
5500         map_length = length;
5501
5502         ret = __btrfs_map_block(root->fs_info, rw, logical, &map_length, &bbio,
5503                               mirror_num, &raid_map);
5504         if (ret) /* -ENOMEM */
5505                 return ret;
5506
5507         total_devs = bbio->num_stripes;
5508         bbio->orig_bio = first_bio;
5509         bbio->private = first_bio->bi_private;
5510         bbio->end_io = first_bio->bi_end_io;
5511         atomic_set(&bbio->stripes_pending, bbio->num_stripes);
5512
5513         if (raid_map) {
5514                 /* In this case, map_length has been set to the length of
5515                    a single stripe; not the whole write */
5516                 if (rw & WRITE) {
5517                         return raid56_parity_write(root, bio, bbio,
5518                                                    raid_map, map_length);
5519                 } else {
5520                         return raid56_parity_recover(root, bio, bbio,
5521                                                      raid_map, map_length,
5522                                                      mirror_num);
5523                 }
5524         }
5525
5526         if (map_length < length) {
5527                 btrfs_crit(root->fs_info, "mapping failed logical %llu bio len %llu len %llu",
5528                         logical, length, map_length);
5529                 BUG();
5530         }
5531
5532         while (dev_nr < total_devs) {
5533                 dev = bbio->stripes[dev_nr].dev;
5534                 if (!dev || !dev->bdev || (rw & WRITE && !dev->writeable)) {
5535                         bbio_error(bbio, first_bio, logical);
5536                         dev_nr++;
5537                         continue;
5538                 }
5539
5540                 /*
5541                  * Check and see if we're ok with this bio based on it's size
5542                  * and offset with the given device.
5543                  */
5544                 if (!bio_size_ok(dev->bdev, first_bio,
5545                                  bbio->stripes[dev_nr].physical >> 9)) {
5546                         ret = breakup_stripe_bio(root, bbio, first_bio, dev,
5547                                                  dev_nr, rw, async_submit);
5548                         BUG_ON(ret);
5549                         dev_nr++;
5550                         continue;
5551                 }
5552
5553                 if (dev_nr < total_devs - 1) {
5554                         bio = btrfs_bio_clone(first_bio, GFP_NOFS);
5555                         BUG_ON(!bio); /* -ENOMEM */
5556                 } else {
5557                         bio = first_bio;
5558                 }
5559
5560                 submit_stripe_bio(root, bbio, bio,
5561                                   bbio->stripes[dev_nr].physical, dev_nr, rw,
5562                                   async_submit);
5563                 dev_nr++;
5564         }
5565         return 0;
5566 }
5567
5568 struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid,
5569                                        u8 *uuid, u8 *fsid)
5570 {
5571         struct btrfs_device *device;
5572         struct btrfs_fs_devices *cur_devices;
5573
5574         cur_devices = fs_info->fs_devices;
5575         while (cur_devices) {
5576                 if (!fsid ||
5577                     !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
5578                         device = __find_device(&cur_devices->devices,
5579                                                devid, uuid);
5580                         if (device)
5581                                 return device;
5582                 }
5583                 cur_devices = cur_devices->seed;
5584         }
5585         return NULL;
5586 }
5587
5588 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
5589                                             u64 devid, u8 *dev_uuid)
5590 {
5591         struct btrfs_device *device;
5592         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
5593
5594         device = btrfs_alloc_device(NULL, &devid, dev_uuid);
5595         if (IS_ERR(device))
5596                 return NULL;
5597
5598         list_add(&device->dev_list, &fs_devices->devices);
5599         device->fs_devices = fs_devices;
5600         fs_devices->num_devices++;
5601
5602         device->missing = 1;
5603         fs_devices->missing_devices++;
5604
5605         return device;
5606 }
5607
5608 /**
5609  * btrfs_alloc_device - allocate struct btrfs_device
5610  * @fs_info:    used only for generating a new devid, can be NULL if
5611  *              devid is provided (i.e. @devid != NULL).
5612  * @devid:      a pointer to devid for this device.  If NULL a new devid
5613  *              is generated.
5614  * @uuid:       a pointer to UUID for this device.  If NULL a new UUID
5615  *              is generated.
5616  *
5617  * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
5618  * on error.  Returned struct is not linked onto any lists and can be
5619  * destroyed with kfree() right away.
5620  */
5621 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
5622                                         const u64 *devid,
5623                                         const u8 *uuid)
5624 {
5625         struct btrfs_device *dev;
5626         u64 tmp;
5627
5628         if (!devid && !fs_info) {
5629                 WARN_ON(1);
5630                 return ERR_PTR(-EINVAL);
5631         }
5632
5633         dev = __alloc_device();
5634         if (IS_ERR(dev))
5635                 return dev;
5636
5637         if (devid)
5638                 tmp = *devid;
5639         else {
5640                 int ret;
5641
5642                 ret = find_next_devid(fs_info, &tmp);
5643                 if (ret) {
5644                         kfree(dev);
5645                         return ERR_PTR(ret);
5646                 }
5647         }
5648         dev->devid = tmp;
5649
5650         if (uuid)
5651                 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE);
5652         else
5653                 generate_random_uuid(dev->uuid);
5654
5655         dev->work.func = pending_bios_fn;
5656
5657         return dev;
5658 }
5659
5660 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
5661                           struct extent_buffer *leaf,
5662                           struct btrfs_chunk *chunk)
5663 {
5664         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
5665         struct map_lookup *map;
5666         struct extent_map *em;
5667         u64 logical;
5668         u64 length;
5669         u64 devid;
5670         u8 uuid[BTRFS_UUID_SIZE];
5671         int num_stripes;
5672         int ret;
5673         int i;
5674
5675         logical = key->offset;
5676         length = btrfs_chunk_length(leaf, chunk);
5677
5678         read_lock(&map_tree->map_tree.lock);
5679         em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
5680         read_unlock(&map_tree->map_tree.lock);
5681
5682         /* already mapped? */
5683         if (em && em->start <= logical && em->start + em->len > logical) {
5684                 free_extent_map(em);
5685                 return 0;
5686         } else if (em) {
5687                 free_extent_map(em);
5688         }
5689
5690         em = alloc_extent_map();
5691         if (!em)
5692                 return -ENOMEM;
5693         num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
5694         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
5695         if (!map) {
5696                 free_extent_map(em);
5697                 return -ENOMEM;
5698         }
5699
5700         em->bdev = (struct block_device *)map;
5701         em->start = logical;
5702         em->len = length;
5703         em->orig_start = 0;
5704         em->block_start = 0;
5705         em->block_len = em->len;
5706
5707         map->num_stripes = num_stripes;
5708         map->io_width = btrfs_chunk_io_width(leaf, chunk);
5709         map->io_align = btrfs_chunk_io_align(leaf, chunk);
5710         map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
5711         map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
5712         map->type = btrfs_chunk_type(leaf, chunk);
5713         map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
5714         for (i = 0; i < num_stripes; i++) {
5715                 map->stripes[i].physical =
5716                         btrfs_stripe_offset_nr(leaf, chunk, i);
5717                 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
5718                 read_extent_buffer(leaf, uuid, (unsigned long)
5719                                    btrfs_stripe_dev_uuid_nr(chunk, i),
5720                                    BTRFS_UUID_SIZE);
5721                 map->stripes[i].dev = btrfs_find_device(root->fs_info, devid,
5722                                                         uuid, NULL);
5723                 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
5724                         kfree(map);
5725                         free_extent_map(em);
5726                         return -EIO;
5727                 }
5728                 if (!map->stripes[i].dev) {
5729                         map->stripes[i].dev =
5730                                 add_missing_dev(root, devid, uuid);
5731                         if (!map->stripes[i].dev) {
5732                                 kfree(map);
5733                                 free_extent_map(em);
5734                                 return -EIO;
5735                         }
5736                 }
5737                 map->stripes[i].dev->in_fs_metadata = 1;
5738         }
5739
5740         write_lock(&map_tree->map_tree.lock);
5741         ret = add_extent_mapping(&map_tree->map_tree, em, 0);
5742         write_unlock(&map_tree->map_tree.lock);
5743         BUG_ON(ret); /* Tree corruption */
5744         free_extent_map(em);
5745
5746         return 0;
5747 }
5748
5749 static void fill_device_from_item(struct extent_buffer *leaf,
5750                                  struct btrfs_dev_item *dev_item,
5751                                  struct btrfs_device *device)
5752 {
5753         unsigned long ptr;
5754
5755         device->devid = btrfs_device_id(leaf, dev_item);
5756         device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
5757         device->total_bytes = device->disk_total_bytes;
5758         device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
5759         device->type = btrfs_device_type(leaf, dev_item);
5760         device->io_align = btrfs_device_io_align(leaf, dev_item);
5761         device->io_width = btrfs_device_io_width(leaf, dev_item);
5762         device->sector_size = btrfs_device_sector_size(leaf, dev_item);
5763         WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
5764         device->is_tgtdev_for_dev_replace = 0;
5765
5766         ptr = btrfs_device_uuid(dev_item);
5767         read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
5768 }
5769
5770 static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
5771 {
5772         struct btrfs_fs_devices *fs_devices;
5773         int ret;
5774
5775         BUG_ON(!mutex_is_locked(&uuid_mutex));
5776
5777         fs_devices = root->fs_info->fs_devices->seed;
5778         while (fs_devices) {
5779                 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
5780                         ret = 0;
5781                         goto out;
5782                 }
5783                 fs_devices = fs_devices->seed;
5784         }
5785
5786         fs_devices = find_fsid(fsid);
5787         if (!fs_devices) {
5788                 ret = -ENOENT;
5789                 goto out;
5790         }
5791
5792         fs_devices = clone_fs_devices(fs_devices);
5793         if (IS_ERR(fs_devices)) {
5794                 ret = PTR_ERR(fs_devices);
5795                 goto out;
5796         }
5797
5798         ret = __btrfs_open_devices(fs_devices, FMODE_READ,
5799                                    root->fs_info->bdev_holder);
5800         if (ret) {
5801                 free_fs_devices(fs_devices);
5802                 goto out;
5803         }
5804
5805         if (!fs_devices->seeding) {
5806                 __btrfs_close_devices(fs_devices);
5807                 free_fs_devices(fs_devices);
5808                 ret = -EINVAL;
5809                 goto out;
5810         }
5811
5812         fs_devices->seed = root->fs_info->fs_devices->seed;
5813         root->fs_info->fs_devices->seed = fs_devices;
5814 out:
5815         return ret;
5816 }
5817
5818 static int read_one_dev(struct btrfs_root *root,
5819                         struct extent_buffer *leaf,
5820                         struct btrfs_dev_item *dev_item)
5821 {
5822         struct btrfs_device *device;
5823         u64 devid;
5824         int ret;
5825         u8 fs_uuid[BTRFS_UUID_SIZE];
5826         u8 dev_uuid[BTRFS_UUID_SIZE];
5827
5828         devid = btrfs_device_id(leaf, dev_item);
5829         read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
5830                            BTRFS_UUID_SIZE);
5831         read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
5832                            BTRFS_UUID_SIZE);
5833
5834         if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
5835                 ret = open_seed_devices(root, fs_uuid);
5836                 if (ret && !btrfs_test_opt(root, DEGRADED))
5837                         return ret;
5838         }
5839
5840         device = btrfs_find_device(root->fs_info, devid, dev_uuid, fs_uuid);
5841         if (!device || !device->bdev) {
5842                 if (!btrfs_test_opt(root, DEGRADED))
5843                         return -EIO;
5844
5845                 if (!device) {
5846                         btrfs_warn(root->fs_info, "devid %llu missing", devid);
5847                         device = add_missing_dev(root, devid, dev_uuid);
5848                         if (!device)
5849                                 return -ENOMEM;
5850                 } else if (!device->missing) {
5851                         /*
5852                          * this happens when a device that was properly setup
5853                          * in the device info lists suddenly goes bad.
5854                          * device->bdev is NULL, and so we have to set
5855                          * device->missing to one here
5856                          */
5857                         root->fs_info->fs_devices->missing_devices++;
5858                         device->missing = 1;
5859                 }
5860         }
5861
5862         if (device->fs_devices != root->fs_info->fs_devices) {
5863                 BUG_ON(device->writeable);
5864                 if (device->generation !=
5865                     btrfs_device_generation(leaf, dev_item))
5866                         return -EINVAL;
5867         }
5868
5869         fill_device_from_item(leaf, dev_item, device);
5870         device->in_fs_metadata = 1;
5871         if (device->writeable && !device->is_tgtdev_for_dev_replace) {
5872                 device->fs_devices->total_rw_bytes += device->total_bytes;
5873                 spin_lock(&root->fs_info->free_chunk_lock);
5874                 root->fs_info->free_chunk_space += device->total_bytes -
5875                         device->bytes_used;
5876                 spin_unlock(&root->fs_info->free_chunk_lock);
5877         }
5878         ret = 0;
5879         return ret;
5880 }
5881
5882 int btrfs_read_sys_array(struct btrfs_root *root)
5883 {
5884         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
5885         struct extent_buffer *sb;
5886         struct btrfs_disk_key *disk_key;
5887         struct btrfs_chunk *chunk;
5888         u8 *ptr;
5889         unsigned long sb_ptr;
5890         int ret = 0;
5891         u32 num_stripes;
5892         u32 array_size;
5893         u32 len = 0;
5894         u32 cur;
5895         struct btrfs_key key;
5896
5897         sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
5898                                           BTRFS_SUPER_INFO_SIZE);
5899         if (!sb)
5900                 return -ENOMEM;
5901         btrfs_set_buffer_uptodate(sb);
5902         btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
5903         /*
5904          * The sb extent buffer is artifical and just used to read the system array.
5905          * btrfs_set_buffer_uptodate() call does not properly mark all it's
5906          * pages up-to-date when the page is larger: extent does not cover the
5907          * whole page and consequently check_page_uptodate does not find all
5908          * the page's extents up-to-date (the hole beyond sb),
5909          * write_extent_buffer then triggers a WARN_ON.
5910          *
5911          * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
5912          * but sb spans only this function. Add an explicit SetPageUptodate call
5913          * to silence the warning eg. on PowerPC 64.
5914          */
5915         if (PAGE_CACHE_SIZE > BTRFS_SUPER_INFO_SIZE)
5916                 SetPageUptodate(sb->pages[0]);
5917
5918         write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
5919         array_size = btrfs_super_sys_array_size(super_copy);
5920
5921         ptr = super_copy->sys_chunk_array;
5922         sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
5923         cur = 0;
5924
5925         while (cur < array_size) {
5926                 disk_key = (struct btrfs_disk_key *)ptr;
5927                 btrfs_disk_key_to_cpu(&key, disk_key);
5928
5929                 len = sizeof(*disk_key); ptr += len;
5930                 sb_ptr += len;
5931                 cur += len;
5932
5933                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
5934                         chunk = (struct btrfs_chunk *)sb_ptr;
5935                         ret = read_one_chunk(root, &key, sb, chunk);
5936                         if (ret)
5937                                 break;
5938                         num_stripes = btrfs_chunk_num_stripes(sb, chunk);
5939                         len = btrfs_chunk_item_size(num_stripes);
5940                 } else {
5941                         ret = -EIO;
5942                         break;
5943                 }
5944                 ptr += len;
5945                 sb_ptr += len;
5946                 cur += len;
5947         }
5948         free_extent_buffer(sb);
5949         return ret;
5950 }
5951
5952 int btrfs_read_chunk_tree(struct btrfs_root *root)
5953 {
5954         struct btrfs_path *path;
5955         struct extent_buffer *leaf;
5956         struct btrfs_key key;
5957         struct btrfs_key found_key;
5958         int ret;
5959         int slot;
5960
5961         root = root->fs_info->chunk_root;
5962
5963         path = btrfs_alloc_path();
5964         if (!path)
5965                 return -ENOMEM;
5966
5967         mutex_lock(&uuid_mutex);
5968         lock_chunks(root);
5969
5970         /*
5971          * Read all device items, and then all the chunk items. All
5972          * device items are found before any chunk item (their object id
5973          * is smaller than the lowest possible object id for a chunk
5974          * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
5975          */
5976         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
5977         key.offset = 0;
5978         key.type = 0;
5979         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5980         if (ret < 0)
5981                 goto error;
5982         while (1) {
5983                 leaf = path->nodes[0];
5984                 slot = path->slots[0];
5985                 if (slot >= btrfs_header_nritems(leaf)) {
5986                         ret = btrfs_next_leaf(root, path);
5987                         if (ret == 0)
5988                                 continue;
5989                         if (ret < 0)
5990                                 goto error;
5991                         break;
5992                 }
5993                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5994                 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
5995                         struct btrfs_dev_item *dev_item;
5996                         dev_item = btrfs_item_ptr(leaf, slot,
5997                                                   struct btrfs_dev_item);
5998                         ret = read_one_dev(root, leaf, dev_item);
5999                         if (ret)
6000                                 goto error;
6001                 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
6002                         struct btrfs_chunk *chunk;
6003                         chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
6004                         ret = read_one_chunk(root, &found_key, leaf, chunk);
6005                         if (ret)
6006                                 goto error;
6007                 }
6008                 path->slots[0]++;
6009         }
6010         ret = 0;
6011 error:
6012         unlock_chunks(root);
6013         mutex_unlock(&uuid_mutex);
6014
6015         btrfs_free_path(path);
6016         return ret;
6017 }
6018
6019 void btrfs_init_devices_late(struct btrfs_fs_info *fs_info)
6020 {
6021         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6022         struct btrfs_device *device;
6023
6024         mutex_lock(&fs_devices->device_list_mutex);
6025         list_for_each_entry(device, &fs_devices->devices, dev_list)
6026                 device->dev_root = fs_info->dev_root;
6027         mutex_unlock(&fs_devices->device_list_mutex);
6028 }
6029
6030 static void __btrfs_reset_dev_stats(struct btrfs_device *dev)
6031 {
6032         int i;
6033
6034         for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6035                 btrfs_dev_stat_reset(dev, i);
6036 }
6037
6038 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
6039 {
6040         struct btrfs_key key;
6041         struct btrfs_key found_key;
6042         struct btrfs_root *dev_root = fs_info->dev_root;
6043         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6044         struct extent_buffer *eb;
6045         int slot;
6046         int ret = 0;
6047         struct btrfs_device *device;
6048         struct btrfs_path *path = NULL;
6049         int i;
6050
6051         path = btrfs_alloc_path();
6052         if (!path) {
6053                 ret = -ENOMEM;
6054                 goto out;
6055         }
6056
6057         mutex_lock(&fs_devices->device_list_mutex);
6058         list_for_each_entry(device, &fs_devices->devices, dev_list) {
6059                 int item_size;
6060                 struct btrfs_dev_stats_item *ptr;
6061
6062                 key.objectid = 0;
6063                 key.type = BTRFS_DEV_STATS_KEY;
6064                 key.offset = device->devid;
6065                 ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
6066                 if (ret) {
6067                         __btrfs_reset_dev_stats(device);
6068                         device->dev_stats_valid = 1;
6069                         btrfs_release_path(path);
6070                         continue;
6071                 }
6072                 slot = path->slots[0];
6073                 eb = path->nodes[0];
6074                 btrfs_item_key_to_cpu(eb, &found_key, slot);
6075                 item_size = btrfs_item_size_nr(eb, slot);
6076
6077                 ptr = btrfs_item_ptr(eb, slot,
6078                                      struct btrfs_dev_stats_item);
6079
6080                 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
6081                         if (item_size >= (1 + i) * sizeof(__le64))
6082                                 btrfs_dev_stat_set(device, i,
6083                                         btrfs_dev_stats_value(eb, ptr, i));
6084                         else
6085                                 btrfs_dev_stat_reset(device, i);
6086                 }
6087
6088                 device->dev_stats_valid = 1;
6089                 btrfs_dev_stat_print_on_load(device);
6090                 btrfs_release_path(path);
6091         }
6092         mutex_unlock(&fs_devices->device_list_mutex);
6093
6094 out:
6095         btrfs_free_path(path);
6096         return ret < 0 ? ret : 0;
6097 }
6098
6099 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
6100                                 struct btrfs_root *dev_root,
6101                                 struct btrfs_device *device)
6102 {
6103         struct btrfs_path *path;
6104         struct btrfs_key key;
6105         struct extent_buffer *eb;
6106         struct btrfs_dev_stats_item *ptr;
6107         int ret;
6108         int i;
6109
6110         key.objectid = 0;
6111         key.type = BTRFS_DEV_STATS_KEY;
6112         key.offset = device->devid;
6113
6114         path = btrfs_alloc_path();
6115         BUG_ON(!path);
6116         ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
6117         if (ret < 0) {
6118                 printk_in_rcu(KERN_WARNING "btrfs: error %d while searching for dev_stats item for device %s!\n",
6119                               ret, rcu_str_deref(device->name));
6120                 goto out;
6121         }
6122
6123         if (ret == 0 &&
6124             btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
6125                 /* need to delete old one and insert a new one */
6126                 ret = btrfs_del_item(trans, dev_root, path);
6127                 if (ret != 0) {
6128                         printk_in_rcu(KERN_WARNING "btrfs: delete too small dev_stats item for device %s failed %d!\n",
6129                                       rcu_str_deref(device->name), ret);
6130                         goto out;
6131                 }
6132                 ret = 1;
6133         }
6134
6135         if (ret == 1) {
6136                 /* need to insert a new item */
6137                 btrfs_release_path(path);
6138                 ret = btrfs_insert_empty_item(trans, dev_root, path,
6139                                               &key, sizeof(*ptr));
6140                 if (ret < 0) {
6141                         printk_in_rcu(KERN_WARNING "btrfs: insert dev_stats item for device %s failed %d!\n",
6142                                       rcu_str_deref(device->name), ret);
6143                         goto out;
6144                 }
6145         }
6146
6147         eb = path->nodes[0];
6148         ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
6149         for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6150                 btrfs_set_dev_stats_value(eb, ptr, i,
6151                                           btrfs_dev_stat_read(device, i));
6152         btrfs_mark_buffer_dirty(eb);
6153
6154 out:
6155         btrfs_free_path(path);
6156         return ret;
6157 }
6158
6159 /*
6160  * called from commit_transaction. Writes all changed device stats to disk.
6161  */
6162 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans,
6163                         struct btrfs_fs_info *fs_info)
6164 {
6165         struct btrfs_root *dev_root = fs_info->dev_root;
6166         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6167         struct btrfs_device *device;
6168         int ret = 0;
6169
6170         mutex_lock(&fs_devices->device_list_mutex);
6171         list_for_each_entry(device, &fs_devices->devices, dev_list) {
6172                 if (!device->dev_stats_valid || !device->dev_stats_dirty)
6173                         continue;
6174
6175                 ret = update_dev_stat_item(trans, dev_root, device);
6176                 if (!ret)
6177                         device->dev_stats_dirty = 0;
6178         }
6179         mutex_unlock(&fs_devices->device_list_mutex);
6180
6181         return ret;
6182 }
6183
6184 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
6185 {
6186         btrfs_dev_stat_inc(dev, index);
6187         btrfs_dev_stat_print_on_error(dev);
6188 }
6189
6190 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
6191 {
6192         if (!dev->dev_stats_valid)
6193                 return;
6194         printk_ratelimited_in_rcu(KERN_ERR
6195                            "btrfs: bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
6196                            rcu_str_deref(dev->name),
6197                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
6198                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
6199                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
6200                            btrfs_dev_stat_read(dev,
6201                                                BTRFS_DEV_STAT_CORRUPTION_ERRS),
6202                            btrfs_dev_stat_read(dev,
6203                                                BTRFS_DEV_STAT_GENERATION_ERRS));
6204 }
6205
6206 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
6207 {
6208         int i;
6209
6210         for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6211                 if (btrfs_dev_stat_read(dev, i) != 0)
6212                         break;
6213         if (i == BTRFS_DEV_STAT_VALUES_MAX)
6214                 return; /* all values == 0, suppress message */
6215
6216         printk_in_rcu(KERN_INFO "btrfs: bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
6217                rcu_str_deref(dev->name),
6218                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
6219                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
6220                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
6221                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
6222                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
6223 }
6224
6225 int btrfs_get_dev_stats(struct btrfs_root *root,
6226                         struct btrfs_ioctl_get_dev_stats *stats)
6227 {
6228         struct btrfs_device *dev;
6229         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
6230         int i;
6231
6232         mutex_lock(&fs_devices->device_list_mutex);
6233         dev = btrfs_find_device(root->fs_info, stats->devid, NULL, NULL);
6234         mutex_unlock(&fs_devices->device_list_mutex);
6235
6236         if (!dev) {
6237                 printk(KERN_WARNING
6238                        "btrfs: get dev_stats failed, device not found\n");
6239                 return -ENODEV;
6240         } else if (!dev->dev_stats_valid) {
6241                 printk(KERN_WARNING
6242                        "btrfs: get dev_stats failed, not yet valid\n");
6243                 return -ENODEV;
6244         } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
6245                 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
6246                         if (stats->nr_items > i)
6247                                 stats->values[i] =
6248                                         btrfs_dev_stat_read_and_reset(dev, i);
6249                         else
6250                                 btrfs_dev_stat_reset(dev, i);
6251                 }
6252         } else {
6253                 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6254                         if (stats->nr_items > i)
6255                                 stats->values[i] = btrfs_dev_stat_read(dev, i);
6256         }
6257         if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
6258                 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
6259         return 0;
6260 }
6261
6262 int btrfs_scratch_superblock(struct btrfs_device *device)
6263 {
6264         struct buffer_head *bh;
6265         struct btrfs_super_block *disk_super;
6266
6267         bh = btrfs_read_dev_super(device->bdev);
6268         if (!bh)
6269                 return -EINVAL;
6270         disk_super = (struct btrfs_super_block *)bh->b_data;
6271
6272         memset(&disk_super->magic, 0, sizeof(disk_super->magic));
6273         set_buffer_dirty(bh);
6274         sync_dirty_buffer(bh);
6275         brelse(bh);
6276
6277         return 0;
6278 }