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 "ctree.h"
  32 #include "extent_map.h"
  33 #include "disk-io.h"
  34 #include "transaction.h"
  35 #include "print-tree.h"
  36 #include "volumes.h"
  37 #include "raid56.h"
  38 #include "async-thread.h"
  39 #include "check-integrity.h"
  40 #include "rcu-string.h"
  41 #include "math.h"
  42 #include "dev-replace.h"
  43
  44 static int init_first_rw_device(struct btrfs_trans_handle *trans,
  45                                 struct btrfs_root *root,
  46                                 struct btrfs_device *device);
  47 static int btrfs_relocate_sys_chunks(struct btrfs_root *root);
  48 static void __btrfs_reset_dev_stats(struct btrfs_device *dev);
  49 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev);
  50 static void btrfs_dev_stat_print_on_load(struct btrfs_device *device);
  51
  52 static DEFINE_MUTEX(uuid_mutex);
  53 static LIST_HEAD(fs_uuids);
  54
  55 static void lock_chunks(struct btrfs_root *root)
  56 {
  57         mutex_lock(&root->fs_info->chunk_mutex);
  58 }
  59
  60 static void unlock_chunks(struct btrfs_root *root)
  61 {
  62         mutex_unlock(&root->fs_info->chunk_mutex);
  63 }
  64
  65 static struct btrfs_fs_devices *__alloc_fs_devices(void)
  66 {
  67         struct btrfs_fs_devices *fs_devs;
  68
  69         fs_devs = kzalloc(sizeof(*fs_devs), GFP_NOFS);
  70         if (!fs_devs)
  71                 return ERR_PTR(-ENOMEM);
  72
  73         mutex_init(&fs_devs->device_list_mutex);
  74
  75         INIT_LIST_HEAD(&fs_devs->devices);
  76         INIT_LIST_HEAD(&fs_devs->alloc_list);
  77         INIT_LIST_HEAD(&fs_devs->list);
  78
  79         return fs_devs;
  80 }
  81
  82 /**
  83  * alloc_fs_devices - allocate struct btrfs_fs_devices
  84  * @fsid:       a pointer to UUID for this FS.  If NULL a new UUID is
  85  *              generated.
  86  *
  87  * Return: a pointer to a new &struct btrfs_fs_devices on success;
  88  * ERR_PTR() on error.  Returned struct is not linked onto any lists and
  89  * can be destroyed with kfree() right away.
  90  */
  91 static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid)
  92 {
  93         struct btrfs_fs_devices *fs_devs;
  94
  95         fs_devs = __alloc_fs_devices();
  96         if (IS_ERR(fs_devs))
  97                 return fs_devs;
  98
  99         if (fsid)
 100                 memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE);
 101         else
 102                 generate_random_uuid(fs_devs->fsid);
 103
 104         return fs_devs;
 105 }
 106
 107 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
 108 {
 109         struct btrfs_device *device;
 110         WARN_ON(fs_devices->opened);
 111         while (!list_empty(&fs_devices->devices)) {
 112                 device = list_entry(fs_devices->devices.next,
 113                                     struct btrfs_device, dev_list);
 114                 list_del(&device->dev_list);
 115                 rcu_string_free(device->name);
 116                 kfree(device);
 117         }
 118         kfree(fs_devices);
 119 }
 120
 121 static void btrfs_kobject_uevent(struct block_device *bdev,
 122                                  enum kobject_action action)
 123 {
 124         int ret;
 125
 126         ret = kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, action);
 127         if (ret)
 128                 pr_warn("BTRFS: Sending event '%d' to kobject: '%s' (%p): failed\n",
 129                         action,
 130                         kobject_name(&disk_to_dev(bdev->bd_disk)->kobj),
 131                         &disk_to_dev(bdev->bd_disk)->kobj);
 132 }
 133
 134 void btrfs_cleanup_fs_uuids(void)
 135 {
 136         struct btrfs_fs_devices *fs_devices;
 137
 138         while (!list_empty(&fs_uuids)) {
 139                 fs_devices = list_entry(fs_uuids.next,
 140                                         struct btrfs_fs_devices, list);
 141                 list_del(&fs_devices->list);
 142                 free_fs_devices(fs_devices);
 143         }
 144 }
 145
 146 static struct btrfs_device *__alloc_device(void)
 147 {
 148         struct btrfs_device *dev;
 149
 150         dev = kzalloc(sizeof(*dev), GFP_NOFS);
 151         if (!dev)
 152                 return ERR_PTR(-ENOMEM);
 153
 154         INIT_LIST_HEAD(&dev->dev_list);
 155         INIT_LIST_HEAD(&dev->dev_alloc_list);
 156
 157         spin_lock_init(&dev->io_lock);
 158
 159         spin_lock_init(&dev->reada_lock);
 160         atomic_set(&dev->reada_in_flight, 0);
 161         INIT_RADIX_TREE(&dev->reada_zones, GFP_NOFS & ~__GFP_WAIT);
 162         INIT_RADIX_TREE(&dev->reada_extents, GFP_NOFS & ~__GFP_WAIT);
 163
 164         return dev;
 165 }
 166
 167 static noinline struct btrfs_device *__find_device(struct list_head *head,
 168                                                    u64 devid, u8 *uuid)
 169 {
 170         struct btrfs_device *dev;
 171
 172         list_for_each_entry(dev, head, dev_list) {
 173                 if (dev->devid == devid &&
 174                     (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
 175                         return dev;
 176                 }
 177         }
 178         return NULL;
 179 }
 180
 181 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
 182 {
 183         struct btrfs_fs_devices *fs_devices;
 184
 185         list_for_each_entry(fs_devices, &fs_uuids, list) {
 186                 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
 187                         return fs_devices;
 188         }
 189         return NULL;
 190 }
 191
 192 static int
 193 btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder,
 194                       int flush, struct block_device **bdev,
 195                       struct buffer_head **bh)
 196 {
 197         int ret;
 198
 199         *bdev = blkdev_get_by_path(device_path, flags, holder);
 200
 201         if (IS_ERR(*bdev)) {
 202                 ret = PTR_ERR(*bdev);
 203                 printk(KERN_INFO "BTRFS: open %s failed\n", device_path);
 204                 goto error;
 205         }
 206
 207         if (flush)
 208                 filemap_write_and_wait((*bdev)->bd_inode->i_mapping);
 209         ret = set_blocksize(*bdev, 4096);
 210         if (ret) {
 211                 blkdev_put(*bdev, flags);
 212                 goto error;
 213         }
 214         invalidate_bdev(*bdev);
 215         *bh = btrfs_read_dev_super(*bdev);
 216         if (!*bh) {
 217                 ret = -EINVAL;
 218                 blkdev_put(*bdev, flags);
 219                 goto error;
 220         }
 221
 222         return 0;
 223
 224 error:
 225         *bdev = NULL;
 226         *bh = NULL;
 227         return ret;
 228 }
 229
 230 static void requeue_list(struct btrfs_pending_bios *pending_bios,
 231                         struct bio *head, struct bio *tail)
 232 {
 233
 234         struct bio *old_head;
 235
 236         old_head = pending_bios->head;
 237         pending_bios->head = head;
 238         if (pending_bios->tail)
 239                 tail->bi_next = old_head;
 240         else
 241                 pending_bios->tail = tail;
 242 }
 243
 244 /*
 245  * we try to collect pending bios for a device so we don't get a large
 246  * number of procs sending bios down to the same device.  This greatly
 247  * improves the schedulers ability to collect and merge the bios.
 248  *
 249  * But, it also turns into a long list of bios to process and that is sure
 250  * to eventually make the worker thread block.  The solution here is to
 251  * make some progress and then put this work struct back at the end of
 252  * the list if the block device is congested.  This way, multiple devices
 253  * can make progress from a single worker thread.
 254  */
 255 static noinline void run_scheduled_bios(struct btrfs_device *device)
 256 {
 257         struct bio *pending;
 258         struct backing_dev_info *bdi;
 259         struct btrfs_fs_info *fs_info;
 260         struct btrfs_pending_bios *pending_bios;
 261         struct bio *tail;
 262         struct bio *cur;
 263         int again = 0;
 264         unsigned long num_run;
 265         unsigned long batch_run = 0;
 266         unsigned long limit;
 267         unsigned long last_waited = 0;
 268         int force_reg = 0;
 269         int sync_pending = 0;
 270         struct blk_plug plug;
 271
 272         /*
 273          * this function runs all the bios we've collected for
 274          * a particular device.  We don't want to wander off to
 275          * another device without first sending all of these down.
 276          * So, setup a plug here and finish it off before we return
 277          */
 278         blk_start_plug(&plug);
 279
 280         bdi = blk_get_backing_dev_info(device->bdev);
 281         fs_info = device->dev_root->fs_info;
 282         limit = btrfs_async_submit_limit(fs_info);
 283         limit = limit * 2 / 3;
 284
 285 loop:
 286         spin_lock(&device->io_lock);
 287
 288 loop_lock:
 289         num_run = 0;
 290
 291         /* take all the bios off the list at once and process them
 292          * later on (without the lock held).  But, remember the
 293          * tail and other pointers so the bios can be properly reinserted
 294          * into the list if we hit congestion
 295          */
 296         if (!force_reg && device->pending_sync_bios.head) {
 297                 pending_bios = &device->pending_sync_bios;
 298                 force_reg = 1;
 299         } else {
 300                 pending_bios = &device->pending_bios;
 301                 force_reg = 0;
 302         }
 303
 304         pending = pending_bios->head;
 305         tail = pending_bios->tail;
 306         WARN_ON(pending && !tail);
 307
 308         /*
 309          * if pending was null this time around, no bios need processing
 310          * at all and we can stop.  Otherwise it'll loop back up again
 311          * and do an additional check so no bios are missed.
 312          *
 313          * device->running_pending is used to synchronize with the
 314          * schedule_bio code.
 315          */
 316         if (device->pending_sync_bios.head == NULL &&
 317             device->pending_bios.head == NULL) {
 318                 again = 0;
 319                 device->running_pending = 0;
 320         } else {
 321                 again = 1;
 322                 device->running_pending = 1;
 323         }
 324
 325         pending_bios->head = NULL;
 326         pending_bios->tail = NULL;
 327
 328         spin_unlock(&device->io_lock);
 329
 330         while (pending) {
 331
 332                 rmb();
 333                 /* we want to work on both lists, but do more bios on the
 334                  * sync list than the regular list
 335                  */
 336                 if ((num_run > 32 &&
 337                     pending_bios != &device->pending_sync_bios &&
 338                     device->pending_sync_bios.head) ||
 339                    (num_run > 64 && pending_bios == &device->pending_sync_bios &&
 340                     device->pending_bios.head)) {
 341                         spin_lock(&device->io_lock);
 342                         requeue_list(pending_bios, pending, tail);
 343                         goto loop_lock;
 344                 }
 345
 346                 cur = pending;
 347                 pending = pending->bi_next;
 348                 cur->bi_next = NULL;
 349
 350                 if (atomic_dec_return(&fs_info->nr_async_bios) < limit &&
 351                     waitqueue_active(&fs_info->async_submit_wait))
 352                         wake_up(&fs_info->async_submit_wait);
 353
 354                 BUG_ON(atomic_read(&cur->bi_cnt) == 0);
 355
 356                 /*
 357                  * if we're doing the sync list, record that our
 358                  * plug has some sync requests on it
 359                  *
 360                  * If we're doing the regular list and there are
 361                  * sync requests sitting around, unplug before
 362                  * we add more
 363                  */
 364                 if (pending_bios == &device->pending_sync_bios) {
 365                         sync_pending = 1;
 366                 } else if (sync_pending) {
 367                         blk_finish_plug(&plug);
 368                         blk_start_plug(&plug);
 369                         sync_pending = 0;
 370                 }
 371
 372                 btrfsic_submit_bio(cur->bi_rw, cur);
 373                 num_run++;
 374                 batch_run++;
 375                 if (need_resched())
 376                         cond_resched();
 377
 378                 /*
 379                  * we made progress, there is more work to do and the bdi
 380                  * is now congested.  Back off and let other work structs
 381                  * run instead
 382                  */
 383                 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
 384                     fs_info->fs_devices->open_devices > 1) {
 385                         struct io_context *ioc;
 386
 387                         ioc = current->io_context;
 388
 389                         /*
 390                          * the main goal here is that we don't want to
 391                          * block if we're going to be able to submit
 392                          * more requests without blocking.
 393                          *
 394                          * This code does two great things, it pokes into
 395                          * the elevator code from a filesystem _and_
 396                          * it makes assumptions about how batching works.
 397                          */
 398                         if (ioc && ioc->nr_batch_requests > 0 &&
 399                             time_before(jiffies, ioc->last_waited + HZ/50UL) &&
 400                             (last_waited == 0 ||
 401                              ioc->last_waited == last_waited)) {
 402                                 /*
 403                                  * we want to go through our batch of
 404                                  * requests and stop.  So, we copy out
 405                                  * the ioc->last_waited time and test
 406                                  * against it before looping
 407                                  */
 408                                 last_waited = ioc->last_waited;
 409                                 if (need_resched())
 410                                         cond_resched();
 411                                 continue;
 412                         }
 413                         spin_lock(&device->io_lock);
 414                         requeue_list(pending_bios, pending, tail);
 415                         device->running_pending = 1;
 416
 417                         spin_unlock(&device->io_lock);
 418                         btrfs_requeue_work(&device->work);
 419                         goto done;
 420                 }
 421                 /* unplug every 64 requests just for good measure */
 422                 if (batch_run % 64 == 0) {
 423                         blk_finish_plug(&plug);
 424                         blk_start_plug(&plug);
 425                         sync_pending = 0;
 426                 }
 427         }
 428
 429         cond_resched();
 430         if (again)
 431                 goto loop;
 432
 433         spin_lock(&device->io_lock);
 434         if (device->pending_bios.head || device->pending_sync_bios.head)
 435                 goto loop_lock;
 436         spin_unlock(&device->io_lock);
 437
 438 done:
 439         blk_finish_plug(&plug);
 440 }
 441
 442 static void pending_bios_fn(struct btrfs_work *work)
 443 {
 444         struct btrfs_device *device;
 445
 446         device = container_of(work, struct btrfs_device, work);
 447         run_scheduled_bios(device);
 448 }
 449
 450 static noinline int device_list_add(const char *path,
 451                            struct btrfs_super_block *disk_super,
 452                            u64 devid, struct btrfs_fs_devices **fs_devices_ret)
 453 {
 454         struct btrfs_device *device;
 455         struct btrfs_fs_devices *fs_devices;
 456         struct rcu_string *name;
 457         u64 found_transid = btrfs_super_generation(disk_super);
 458
 459         fs_devices = find_fsid(disk_super->fsid);
 460         if (!fs_devices) {
 461                 fs_devices = alloc_fs_devices(disk_super->fsid);
 462                 if (IS_ERR(fs_devices))
 463                         return PTR_ERR(fs_devices);
 464
 465                 list_add(&fs_devices->list, &fs_uuids);
 466                 fs_devices->latest_devid = devid;
 467                 fs_devices->latest_trans = found_transid;
 468
 469                 device = NULL;
 470         } else {
 471                 device = __find_device(&fs_devices->devices, devid,
 472                                        disk_super->dev_item.uuid);
 473         }
 474         if (!device) {
 475                 if (fs_devices->opened)
 476                         return -EBUSY;
 477
 478                 device = btrfs_alloc_device(NULL, &devid,
 479                                             disk_super->dev_item.uuid);
 480                 if (IS_ERR(device)) {
 481                         /* we can safely leave the fs_devices entry around */
 482                         return PTR_ERR(device);
 483                 }
 484
 485                 name = rcu_string_strdup(path, GFP_NOFS);
 486                 if (!name) {
 487                         kfree(device);
 488                         return -ENOMEM;
 489                 }
 490                 rcu_assign_pointer(device->name, name);
 491
 492                 mutex_lock(&fs_devices->device_list_mutex);
 493                 list_add_rcu(&device->dev_list, &fs_devices->devices);
 494                 fs_devices->num_devices++;
 495                 mutex_unlock(&fs_devices->device_list_mutex);
 496
 497                 device->fs_devices = fs_devices;
 498         } else if (!device->name || strcmp(device->name->str, path)) {
 499                 name = rcu_string_strdup(path, GFP_NOFS);
 500                 if (!name)
 501                         return -ENOMEM;
 502                 rcu_string_free(device->name);
 503                 rcu_assign_pointer(device->name, name);
 504                 if (device->missing) {
 505                         fs_devices->missing_devices--;
 506                         device->missing = 0;
 507                 }
 508         }
 509
 510         if (found_transid > fs_devices->latest_trans) {
 511                 fs_devices->latest_devid = devid;
 512                 fs_devices->latest_trans = found_transid;
 513         }
 514         *fs_devices_ret = fs_devices;
 515         return 0;
 516 }
 517
 518 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
 519 {
 520         struct btrfs_fs_devices *fs_devices;
 521         struct btrfs_device *device;
 522         struct btrfs_device *orig_dev;
 523
 524         fs_devices = alloc_fs_devices(orig->fsid);
 525         if (IS_ERR(fs_devices))
 526                 return fs_devices;
 527
 528         fs_devices->latest_devid = orig->latest_devid;
 529         fs_devices->latest_trans = orig->latest_trans;
 530         fs_devices->total_devices = orig->total_devices;
 531
 532         /* We have held the volume lock, it is safe to get the devices. */
 533         list_for_each_entry(orig_dev, &orig->devices, dev_list) {
 534                 struct rcu_string *name;
 535
 536                 device = btrfs_alloc_device(NULL, &orig_dev->devid,
 537                                             orig_dev->uuid);
 538                 if (IS_ERR(device))
 539                         goto error;
 540
 541                 /*
 542                  * This is ok to do without rcu read locked because we hold the
 543                  * uuid mutex so nothing we touch in here is going to disappear.
 544                  */
 545                 name = rcu_string_strdup(orig_dev->name->str, GFP_NOFS);
 546                 if (!name) {
 547                         kfree(device);
 548                         goto error;
 549                 }
 550                 rcu_assign_pointer(device->name, name);
 551
 552                 list_add(&device->dev_list, &fs_devices->devices);
 553                 device->fs_devices = fs_devices;
 554                 fs_devices->num_devices++;
 555         }
 556         return fs_devices;
 557 error:
 558         free_fs_devices(fs_devices);
 559         return ERR_PTR(-ENOMEM);
 560 }
 561
 562 void btrfs_close_extra_devices(struct btrfs_fs_info *fs_info,
 563                                struct btrfs_fs_devices *fs_devices, int step)
 564 {
 565         struct btrfs_device *device, *next;
 566
 567         struct block_device *latest_bdev = NULL;
 568         u64 latest_devid = 0;
 569         u64 latest_transid = 0;
 570
 571         mutex_lock(&uuid_mutex);
 572 again:
 573         /* This is the initialized path, it is safe to release the devices. */
 574         list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
 575                 if (device->in_fs_metadata) {
 576                         if (!device->is_tgtdev_for_dev_replace &&
 577                             (!latest_transid ||
 578                              device->generation > latest_transid)) {
 579                                 latest_devid = device->devid;
 580                                 latest_transid = device->generation;
 581                                 latest_bdev = device->bdev;
 582                         }
 583                         continue;
 584                 }
 585
 586                 if (device->devid == BTRFS_DEV_REPLACE_DEVID) {
 587                         /*
 588                          * In the first step, keep the device which has
 589                          * the correct fsid and the devid that is used
 590                          * for the dev_replace procedure.
 591                          * In the second step, the dev_replace state is
 592                          * read from the device tree and it is known
 593                          * whether the procedure is really active or
 594                          * not, which means whether this device is
 595                          * used or whether it should be removed.
 596                          */
 597                         if (step == 0 || device->is_tgtdev_for_dev_replace) {
 598                                 continue;
 599                         }
 600                 }
 601                 if (device->bdev) {
 602                         blkdev_put(device->bdev, device->mode);
 603                         device->bdev = NULL;
 604                         fs_devices->open_devices--;
 605                 }
 606                 if (device->writeable) {
 607                         list_del_init(&device->dev_alloc_list);
 608                         device->writeable = 0;
 609                         if (!device->is_tgtdev_for_dev_replace)
 610                                 fs_devices->rw_devices--;
 611                 }
 612                 list_del_init(&device->dev_list);
 613                 fs_devices->num_devices--;
 614                 rcu_string_free(device->name);
 615                 kfree(device);
 616         }
 617
 618         if (fs_devices->seed) {
 619                 fs_devices = fs_devices->seed;
 620                 goto again;
 621         }
 622
 623         fs_devices->latest_bdev = latest_bdev;
 624         fs_devices->latest_devid = latest_devid;
 625         fs_devices->latest_trans = latest_transid;
 626
 627         mutex_unlock(&uuid_mutex);
 628 }
 629
 630 static void __free_device(struct work_struct *work)
 631 {
 632         struct btrfs_device *device;
 633
 634         device = container_of(work, struct btrfs_device, rcu_work);
 635
 636         if (device->bdev)
 637                 blkdev_put(device->bdev, device->mode);
 638
 639         rcu_string_free(device->name);
 640         kfree(device);
 641 }
 642
 643 static void free_device(struct rcu_head *head)
 644 {
 645         struct btrfs_device *device;
 646
 647         device = container_of(head, struct btrfs_device, rcu);
 648
 649         INIT_WORK(&device->rcu_work, __free_device);
 650         schedule_work(&device->rcu_work);
 651 }
 652
 653 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
 654 {
 655         struct btrfs_device *device;
 656
 657         if (--fs_devices->opened > 0)
 658                 return 0;
 659
 660         mutex_lock(&fs_devices->device_list_mutex);
 661         list_for_each_entry(device, &fs_devices->devices, dev_list) {
 662                 struct btrfs_device *new_device;
 663                 struct rcu_string *name;
 664
 665                 if (device->bdev)
 666                         fs_devices->open_devices--;
 667
 668                 if (device->writeable &&
 669                     device->devid != BTRFS_DEV_REPLACE_DEVID) {
 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 &&
 800                     device->devid != BTRFS_DEV_REPLACE_DEVID) {
 801                         fs_devices->rw_devices++;
 802                         list_add(&device->dev_alloc_list,
 803                                  &fs_devices->alloc_list);
 804                 }
 805                 brelse(bh);
 806                 continue;
 807
 808 error_brelse:
 809                 brelse(bh);
 810                 blkdev_put(bdev, flags);
 811                 continue;
 812         }
 813         if (fs_devices->open_devices == 0) {
 814                 ret = -EINVAL;
 815                 goto out;
 816         }
 817         fs_devices->seeding = seeding;
 818         fs_devices->opened = 1;
 819         fs_devices->latest_bdev = latest_bdev;
 820         fs_devices->latest_devid = latest_devid;
 821         fs_devices->latest_trans = latest_transid;
 822         fs_devices->total_rw_bytes = 0;
 823 out:
 824         return ret;
 825 }
 826
 827 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
 828                        fmode_t flags, void *holder)
 829 {
 830         int ret;
 831
 832         mutex_lock(&uuid_mutex);
 833         if (fs_devices->opened) {
 834                 fs_devices->opened++;
 835                 ret = 0;
 836         } else {
 837                 ret = __btrfs_open_devices(fs_devices, flags, holder);
 838         }
 839         mutex_unlock(&uuid_mutex);
 840         return ret;
 841 }
 842
 843 /*
 844  * Look for a btrfs signature on a device. This may be called out of the mount path
 845  * and we are not allowed to call set_blocksize during the scan. The superblock
 846  * is read via pagecache
 847  */
 848 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
 849                           struct btrfs_fs_devices **fs_devices_ret)
 850 {
 851         struct btrfs_super_block *disk_super;
 852         struct block_device *bdev;
 853         struct page *page;
 854         void *p;
 855         int ret = -EINVAL;
 856         u64 devid;
 857         u64 transid;
 858         u64 total_devices;
 859         u64 bytenr;
 860         pgoff_t index;
 861
 862         /*
 863          * we would like to check all the supers, but that would make
 864          * a btrfs mount succeed after a mkfs from a different FS.
 865          * So, we need to add a special mount option to scan for
 866          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
 867          */
 868         bytenr = btrfs_sb_offset(0);
 869         flags |= FMODE_EXCL;
 870         mutex_lock(&uuid_mutex);
 871
 872         bdev = blkdev_get_by_path(path, flags, holder);
 873
 874         if (IS_ERR(bdev)) {
 875                 ret = PTR_ERR(bdev);
 876                 goto error;
 877         }
 878
 879         /* make sure our super fits in the device */
 880         if (bytenr + PAGE_CACHE_SIZE >= i_size_read(bdev->bd_inode))
 881                 goto error_bdev_put;
 882
 883         /* make sure our super fits in the page */
 884         if (sizeof(*disk_super) > PAGE_CACHE_SIZE)
 885                 goto error_bdev_put;
 886
 887         /* make sure our super doesn't straddle pages on disk */
 888         index = bytenr >> PAGE_CACHE_SHIFT;
 889         if ((bytenr + sizeof(*disk_super) - 1) >> PAGE_CACHE_SHIFT != index)
 890                 goto error_bdev_put;
 891
 892         /* pull in the page with our super */
 893         page = read_cache_page_gfp(bdev->bd_inode->i_mapping,
 894                                    index, GFP_NOFS);
 895
 896         if (IS_ERR_OR_NULL(page))
 897                 goto error_bdev_put;
 898
 899         p = kmap(page);
 900
 901         /* align our pointer to the offset of the super block */
 902         disk_super = p + (bytenr & ~PAGE_CACHE_MASK);
 903
 904         if (btrfs_super_bytenr(disk_super) != bytenr ||
 905             btrfs_super_magic(disk_super) != BTRFS_MAGIC)
 906                 goto error_unmap;
 907
 908         devid = btrfs_stack_device_id(&disk_super->dev_item);
 909         transid = btrfs_super_generation(disk_super);
 910         total_devices = btrfs_super_num_devices(disk_super);
 911
 912         if (disk_super->label[0]) {
 913                 if (disk_super->label[BTRFS_LABEL_SIZE - 1])
 914                         disk_super->label[BTRFS_LABEL_SIZE - 1] = '\0';
 915                 printk(KERN_INFO "BTRFS: device label %s ", disk_super->label);
 916         } else {
 917                 printk(KERN_INFO "BTRFS: device fsid %pU ", disk_super->fsid);
 918         }
 919
 920         printk(KERN_CONT "devid %llu transid %llu %s\n", devid, transid, path);
 921
 922         ret = device_list_add(path, disk_super, devid, fs_devices_ret);
 923         if (!ret && fs_devices_ret)
 924                 (*fs_devices_ret)->total_devices = total_devices;
 925
 926 error_unmap:
 927         kunmap(page);
 928         page_cache_release(page);
 929
 930 error_bdev_put:
 931         blkdev_put(bdev, flags);
 932 error:
 933         mutex_unlock(&uuid_mutex);
 934         return ret;
 935 }
 936
 937 /* helper to account the used device space in the range */
 938 int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
 939                                    u64 end, u64 *length)
 940 {
 941         struct btrfs_key key;
 942         struct btrfs_root *root = device->dev_root;
 943         struct btrfs_dev_extent *dev_extent;
 944         struct btrfs_path *path;
 945         u64 extent_end;
 946         int ret;
 947         int slot;
 948         struct extent_buffer *l;
 949
 950         *length = 0;
 951
 952         if (start >= device->total_bytes || device->is_tgtdev_for_dev_replace)
 953                 return 0;
 954
 955         path = btrfs_alloc_path();
 956         if (!path)
 957                 return -ENOMEM;
 958         path->reada = 2;
 959
 960         key.objectid = device->devid;
 961         key.offset = start;
 962         key.type = BTRFS_DEV_EXTENT_KEY;
 963
 964         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 965         if (ret < 0)
 966                 goto out;
 967         if (ret > 0) {
 968                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
 969                 if (ret < 0)
 970                         goto out;
 971         }
 972
 973         while (1) {
 974                 l = path->nodes[0];
 975                 slot = path->slots[0];
 976                 if (slot >= btrfs_header_nritems(l)) {
 977                         ret = btrfs_next_leaf(root, path);
 978                         if (ret == 0)
 979                                 continue;
 980                         if (ret < 0)
 981                                 goto out;
 982
 983                         break;
 984                 }
 985                 btrfs_item_key_to_cpu(l, &key, slot);
 986
 987                 if (key.objectid < device->devid)
 988                         goto next;
 989
 990                 if (key.objectid > device->devid)
 991                         break;
 992
 993                 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
 994                         goto next;
 995
 996                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
 997                 extent_end = key.offset + btrfs_dev_extent_length(l,
 998                                                                   dev_extent);
 999                 if (key.offset <= start && extent_end > end) {
1000                         *length = end - start + 1;
1001                         break;
1002                 } else if (key.offset <= start && extent_end > start)
1003                         *length += extent_end - start;
1004                 else if (key.offset > start && extent_end <= end)
1005                         *length += extent_end - key.offset;
1006                 else if (key.offset > start && key.offset <= end) {
1007                         *length += end - key.offset + 1;
1008                         break;
1009                 } else if (key.offset > end)
1010                         break;
1011
1012 next:
1013                 path->slots[0]++;
1014         }
1015         ret = 0;
1016 out:
1017         btrfs_free_path(path);
1018         return ret;
1019 }
1020
1021 static int contains_pending_extent(struct btrfs_trans_handle *trans,
1022                                    struct btrfs_device *device,
1023                                    u64 *start, u64 len)
1024 {
1025         struct extent_map *em;
1026         int ret = 0;
1027
1028         list_for_each_entry(em, &trans->transaction->pending_chunks, list) {
1029                 struct map_lookup *map;
1030                 int i;
1031
1032                 map = (struct map_lookup *)em->bdev;
1033                 for (i = 0; i < map->num_stripes; i++) {
1034                         if (map->stripes[i].dev != device)
1035                                 continue;
1036                         if (map->stripes[i].physical >= *start + len ||
1037                             map->stripes[i].physical + em->orig_block_len <=
1038                             *start)
1039                                 continue;
1040                         *start = map->stripes[i].physical +
1041                                 em->orig_block_len;
1042                         ret = 1;
1043                 }
1044         }
1045
1046         return ret;
1047 }
1048
1049
1050 /*
1051  * find_free_dev_extent - find free space in the specified device
1052  * @device:     the device which we search the free space in
1053  * @num_bytes:  the size of the free space that we need
1054  * @start:      store the start of the free space.
1055  * @len:        the size of the free space. that we find, or the size of the max
1056  *              free space if we don't find suitable free space
1057  *
1058  * this uses a pretty simple search, the expectation is that it is
1059  * called very infrequently and that a given device has a small number
1060  * of extents
1061  *
1062  * @start is used to store the start of the free space if we find. But if we
1063  * don't find suitable free space, it will be used to store the start position
1064  * of the max free space.
1065  *
1066  * @len is used to store the size of the free space that we find.
1067  * But if we don't find suitable free space, it is used to store the size of
1068  * the max free space.
1069  */
1070 int find_free_dev_extent(struct btrfs_trans_handle *trans,
1071                          struct btrfs_device *device, u64 num_bytes,
1072                          u64 *start, u64 *len)
1073 {
1074         struct btrfs_key key;
1075         struct btrfs_root *root = device->dev_root;
1076         struct btrfs_dev_extent *dev_extent;
1077         struct btrfs_path *path;
1078         u64 hole_size;
1079         u64 max_hole_start;
1080         u64 max_hole_size;
1081         u64 extent_end;
1082         u64 search_start;
1083         u64 search_end = device->total_bytes;
1084         int ret;
1085         int slot;
1086         struct extent_buffer *l;
1087
1088         /* FIXME use last free of some kind */
1089
1090         /* we don't want to overwrite the superblock on the drive,
1091          * so we make sure to start at an offset of at least 1MB
1092          */
1093         search_start = max(root->fs_info->alloc_start, 1024ull * 1024);
1094
1095         path = btrfs_alloc_path();
1096         if (!path)
1097                 return -ENOMEM;
1098 again:
1099         max_hole_start = search_start;
1100         max_hole_size = 0;
1101         hole_size = 0;
1102
1103         if (search_start >= search_end || device->is_tgtdev_for_dev_replace) {
1104                 ret = -ENOSPC;
1105                 goto out;
1106         }
1107
1108         path->reada = 2;
1109         path->search_commit_root = 1;
1110         path->skip_locking = 1;
1111
1112         key.objectid = device->devid;
1113         key.offset = search_start;
1114         key.type = BTRFS_DEV_EXTENT_KEY;
1115
1116         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1117         if (ret < 0)
1118                 goto out;
1119         if (ret > 0) {
1120                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1121                 if (ret < 0)
1122                         goto out;
1123         }
1124
1125         while (1) {
1126                 l = path->nodes[0];
1127                 slot = path->slots[0];
1128                 if (slot >= btrfs_header_nritems(l)) {
1129                         ret = btrfs_next_leaf(root, path);
1130                         if (ret == 0)
1131                                 continue;
1132                         if (ret < 0)
1133                                 goto out;
1134
1135                         break;
1136                 }
1137                 btrfs_item_key_to_cpu(l, &key, slot);
1138
1139                 if (key.objectid < device->devid)
1140                         goto next;
1141
1142                 if (key.objectid > device->devid)
1143                         break;
1144
1145                 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
1146                         goto next;
1147
1148                 if (key.offset > search_start) {
1149                         hole_size = key.offset - search_start;
1150
1151                         /*
1152                          * Have to check before we set max_hole_start, otherwise
1153                          * we could end up sending back this offset anyway.
1154                          */
1155                         if (contains_pending_extent(trans, device,
1156                                                     &search_start,
1157                                                     hole_size))
1158                                 hole_size = 0;
1159
1160                         if (hole_size > max_hole_size) {
1161                                 max_hole_start = search_start;
1162                                 max_hole_size = hole_size;
1163                         }
1164
1165                         /*
1166                          * If this free space is greater than which we need,
1167                          * it must be the max free space that we have found
1168                          * until now, so max_hole_start must point to the start
1169                          * of this free space and the length of this free space
1170                          * is stored in max_hole_size. Thus, we return
1171                          * max_hole_start and max_hole_size and go back to the
1172                          * caller.
1173                          */
1174                         if (hole_size >= num_bytes) {
1175                                 ret = 0;
1176                                 goto out;
1177                         }
1178                 }
1179
1180                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1181                 extent_end = key.offset + btrfs_dev_extent_length(l,
1182                                                                   dev_extent);
1183                 if (extent_end > search_start)
1184                         search_start = extent_end;
1185 next:
1186                 path->slots[0]++;
1187                 cond_resched();
1188         }
1189
1190         /*
1191          * At this point, search_start should be the end of
1192          * allocated dev extents, and when shrinking the device,
1193          * search_end may be smaller than search_start.
1194          */
1195         if (search_end > search_start)
1196                 hole_size = search_end - search_start;
1197
1198         if (hole_size > max_hole_size) {
1199                 max_hole_start = search_start;
1200                 max_hole_size = hole_size;
1201         }
1202
1203         if (contains_pending_extent(trans, device, &search_start, hole_size)) {
1204                 btrfs_release_path(path);
1205                 goto again;
1206         }
1207
1208         /* See above. */
1209         if (hole_size < num_bytes)
1210                 ret = -ENOSPC;
1211         else
1212                 ret = 0;
1213
1214 out:
1215         btrfs_free_path(path);
1216         *start = max_hole_start;
1217         if (len)
1218                 *len = max_hole_size;
1219         return ret;
1220 }
1221
1222 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1223                           struct btrfs_device *device,
1224                           u64 start)
1225 {
1226         int ret;
1227         struct btrfs_path *path;
1228         struct btrfs_root *root = device->dev_root;
1229         struct btrfs_key key;
1230         struct btrfs_key found_key;
1231         struct extent_buffer *leaf = NULL;
1232         struct btrfs_dev_extent *extent = NULL;
1233
1234         path = btrfs_alloc_path();
1235         if (!path)
1236                 return -ENOMEM;
1237
1238         key.objectid = device->devid;
1239         key.offset = start;
1240         key.type = BTRFS_DEV_EXTENT_KEY;
1241 again:
1242         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1243         if (ret > 0) {
1244                 ret = btrfs_previous_item(root, path, key.objectid,
1245                                           BTRFS_DEV_EXTENT_KEY);
1246                 if (ret)
1247                         goto out;
1248                 leaf = path->nodes[0];
1249                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1250                 extent = btrfs_item_ptr(leaf, path->slots[0],
1251                                         struct btrfs_dev_extent);
1252                 BUG_ON(found_key.offset > start || found_key.offset +
1253                        btrfs_dev_extent_length(leaf, extent) < start);
1254                 key = found_key;
1255                 btrfs_release_path(path);
1256                 goto again;
1257         } else if (ret == 0) {
1258                 leaf = path->nodes[0];
1259                 extent = btrfs_item_ptr(leaf, path->slots[0],
1260                                         struct btrfs_dev_extent);
1261         } else {
1262                 btrfs_error(root->fs_info, ret, "Slot search failed");
1263                 goto out;
1264         }
1265
1266         if (device->bytes_used > 0) {
1267                 u64 len = btrfs_dev_extent_length(leaf, extent);
1268                 device->bytes_used -= len;
1269                 spin_lock(&root->fs_info->free_chunk_lock);
1270                 root->fs_info->free_chunk_space += len;
1271                 spin_unlock(&root->fs_info->free_chunk_lock);
1272         }
1273         ret = btrfs_del_item(trans, root, path);
1274         if (ret) {
1275                 btrfs_error(root->fs_info, ret,
1276                             "Failed to remove dev extent item");
1277         }
1278 out:
1279         btrfs_free_path(path);
1280         return ret;
1281 }
1282
1283 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1284                                   struct btrfs_device *device,
1285                                   u64 chunk_tree, u64 chunk_objectid,
1286                                   u64 chunk_offset, u64 start, u64 num_bytes)
1287 {
1288         int ret;
1289         struct btrfs_path *path;
1290         struct btrfs_root *root = device->dev_root;
1291         struct btrfs_dev_extent *extent;
1292         struct extent_buffer *leaf;
1293         struct btrfs_key key;
1294
1295         WARN_ON(!device->in_fs_metadata);
1296         WARN_ON(device->is_tgtdev_for_dev_replace);
1297         path = btrfs_alloc_path();
1298         if (!path)
1299                 return -ENOMEM;
1300
1301         key.objectid = device->devid;
1302         key.offset = start;
1303         key.type = BTRFS_DEV_EXTENT_KEY;
1304         ret = btrfs_insert_empty_item(trans, root, path, &key,
1305                                       sizeof(*extent));
1306         if (ret)
1307                 goto out;
1308
1309         leaf = path->nodes[0];
1310         extent = btrfs_item_ptr(leaf, path->slots[0],
1311                                 struct btrfs_dev_extent);
1312         btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
1313         btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
1314         btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1315
1316         write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
1317                     btrfs_dev_extent_chunk_tree_uuid(extent), BTRFS_UUID_SIZE);
1318
1319         btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1320         btrfs_mark_buffer_dirty(leaf);
1321 out:
1322         btrfs_free_path(path);
1323         return ret;
1324 }
1325
1326 static u64 find_next_chunk(struct btrfs_fs_info *fs_info)
1327 {
1328         struct extent_map_tree *em_tree;
1329         struct extent_map *em;
1330         struct rb_node *n;
1331         u64 ret = 0;
1332
1333         em_tree = &fs_info->mapping_tree.map_tree;
1334         read_lock(&em_tree->lock);
1335         n = rb_last(&em_tree->map);
1336         if (n) {
1337                 em = rb_entry(n, struct extent_map, rb_node);
1338                 ret = em->start + em->len;
1339         }
1340         read_unlock(&em_tree->lock);
1341
1342         return ret;
1343 }
1344
1345 static noinline int find_next_devid(struct btrfs_fs_info *fs_info,
1346                                     u64 *devid_ret)
1347 {
1348         int ret;
1349         struct btrfs_key key;
1350         struct btrfs_key found_key;
1351         struct btrfs_path *path;
1352
1353         path = btrfs_alloc_path();
1354         if (!path)
1355                 return -ENOMEM;
1356
1357         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1358         key.type = BTRFS_DEV_ITEM_KEY;
1359         key.offset = (u64)-1;
1360
1361         ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
1362         if (ret < 0)
1363                 goto error;
1364
1365         BUG_ON(ret == 0); /* Corruption */
1366
1367         ret = btrfs_previous_item(fs_info->chunk_root, path,
1368                                   BTRFS_DEV_ITEMS_OBJECTID,
1369                                   BTRFS_DEV_ITEM_KEY);
1370         if (ret) {
1371                 *devid_ret = 1;
1372         } else {
1373                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1374                                       path->slots[0]);
1375                 *devid_ret = found_key.offset + 1;
1376         }
1377         ret = 0;
1378 error:
1379         btrfs_free_path(path);
1380         return ret;
1381 }
1382
1383 /*
1384  * the device information is stored in the chunk root
1385  * the btrfs_device struct should be fully filled in
1386  */
1387 static int btrfs_add_device(struct btrfs_trans_handle *trans,
1388                             struct btrfs_root *root,
1389                             struct btrfs_device *device)
1390 {
1391         int ret;
1392         struct btrfs_path *path;
1393         struct btrfs_dev_item *dev_item;
1394         struct extent_buffer *leaf;
1395         struct btrfs_key key;
1396         unsigned long ptr;
1397
1398         root = root->fs_info->chunk_root;
1399
1400         path = btrfs_alloc_path();
1401         if (!path)
1402                 return -ENOMEM;
1403
1404         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1405         key.type = BTRFS_DEV_ITEM_KEY;
1406         key.offset = device->devid;
1407
1408         ret = btrfs_insert_empty_item(trans, root, path, &key,
1409                                       sizeof(*dev_item));
1410         if (ret)
1411                 goto out;
1412
1413         leaf = path->nodes[0];
1414         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1415
1416         btrfs_set_device_id(leaf, dev_item, device->devid);
1417         btrfs_set_device_generation(leaf, dev_item, 0);
1418         btrfs_set_device_type(leaf, dev_item, device->type);
1419         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1420         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1421         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1422         btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
1423         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1424         btrfs_set_device_group(leaf, dev_item, 0);
1425         btrfs_set_device_seek_speed(leaf, dev_item, 0);
1426         btrfs_set_device_bandwidth(leaf, dev_item, 0);
1427         btrfs_set_device_start_offset(leaf, dev_item, 0);
1428
1429         ptr = btrfs_device_uuid(dev_item);
1430         write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1431         ptr = btrfs_device_fsid(dev_item);
1432         write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1433         btrfs_mark_buffer_dirty(leaf);
1434
1435         ret = 0;
1436 out:
1437         btrfs_free_path(path);
1438         return ret;
1439 }
1440
1441 static int btrfs_rm_dev_item(struct btrfs_root *root,
1442                              struct btrfs_device *device)
1443 {
1444         int ret;
1445         struct btrfs_path *path;
1446         struct btrfs_key key;
1447         struct btrfs_trans_handle *trans;
1448
1449         root = root->fs_info->chunk_root;
1450
1451         path = btrfs_alloc_path();
1452         if (!path)
1453                 return -ENOMEM;
1454
1455         trans = btrfs_start_transaction(root, 0);
1456         if (IS_ERR(trans)) {
1457                 btrfs_free_path(path);
1458                 return PTR_ERR(trans);
1459         }
1460         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1461         key.type = BTRFS_DEV_ITEM_KEY;
1462         key.offset = device->devid;
1463         lock_chunks(root);
1464
1465         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1466         if (ret < 0)
1467                 goto out;
1468
1469         if (ret > 0) {
1470                 ret = -ENOENT;
1471                 goto out;
1472         }
1473
1474         ret = btrfs_del_item(trans, root, path);
1475         if (ret)
1476                 goto out;
1477 out:
1478         btrfs_free_path(path);
1479         unlock_chunks(root);
1480         btrfs_commit_transaction(trans, root);
1481         return ret;
1482 }
1483
1484 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
1485 {
1486         struct btrfs_device *device;
1487         struct btrfs_device *next_device;
1488         struct block_device *bdev;
1489         struct buffer_head *bh = NULL;
1490         struct btrfs_super_block *disk_super;
1491         struct btrfs_fs_devices *cur_devices;
1492         u64 all_avail;
1493         u64 devid;
1494         u64 num_devices;
1495         u8 *dev_uuid;
1496         unsigned seq;
1497         int ret = 0;
1498         bool clear_super = false;
1499
1500         mutex_lock(&uuid_mutex);
1501
1502         do {
1503                 seq = read_seqbegin(&root->fs_info->profiles_lock);
1504
1505                 all_avail = root->fs_info->avail_data_alloc_bits |
1506                             root->fs_info->avail_system_alloc_bits |
1507                             root->fs_info->avail_metadata_alloc_bits;
1508         } while (read_seqretry(&root->fs_info->profiles_lock, seq));
1509
1510         num_devices = root->fs_info->fs_devices->num_devices;
1511         btrfs_dev_replace_lock(&root->fs_info->dev_replace);
1512         if (btrfs_dev_replace_is_ongoing(&root->fs_info->dev_replace)) {
1513                 WARN_ON(num_devices < 1);
1514                 num_devices--;
1515         }
1516         btrfs_dev_replace_unlock(&root->fs_info->dev_replace);
1517
1518         if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) && num_devices <= 4) {
1519                 ret = BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET;
1520                 goto out;
1521         }
1522
1523         if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) && num_devices <= 2) {
1524                 ret = BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET;
1525                 goto out;
1526         }
1527
1528         if ((all_avail & BTRFS_BLOCK_GROUP_RAID5) &&
1529             root->fs_info->fs_devices->rw_devices <= 2) {
1530                 ret = BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET;
1531                 goto out;
1532         }
1533         if ((all_avail & BTRFS_BLOCK_GROUP_RAID6) &&
1534             root->fs_info->fs_devices->rw_devices <= 3) {
1535                 ret = BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET;
1536                 goto out;
1537         }
1538
1539         if (strcmp(device_path, "missing") == 0) {
1540                 struct list_head *devices;
1541                 struct btrfs_device *tmp;
1542
1543                 device = NULL;
1544                 devices = &root->fs_info->fs_devices->devices;
1545                 /*
1546                  * It is safe to read the devices since the volume_mutex
1547                  * is held.
1548                  */
1549                 list_for_each_entry(tmp, devices, dev_list) {
1550                         if (tmp->in_fs_metadata &&
1551                             !tmp->is_tgtdev_for_dev_replace &&
1552                             !tmp->bdev) {
1553                                 device = tmp;
1554                                 break;
1555                         }
1556                 }
1557                 bdev = NULL;
1558                 bh = NULL;
1559                 disk_super = NULL;
1560                 if (!device) {
1561                         ret = BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
1562                         goto out;
1563                 }
1564         } else {
1565                 ret = btrfs_get_bdev_and_sb(device_path,
1566                                             FMODE_WRITE | FMODE_EXCL,
1567                                             root->fs_info->bdev_holder, 0,
1568                                             &bdev, &bh);
1569                 if (ret)
1570                         goto out;
1571                 disk_super = (struct btrfs_super_block *)bh->b_data;
1572                 devid = btrfs_stack_device_id(&disk_super->dev_item);
1573                 dev_uuid = disk_super->dev_item.uuid;
1574                 device = btrfs_find_device(root->fs_info, devid, dev_uuid,
1575                                            disk_super->fsid);
1576                 if (!device) {
1577                         ret = -ENOENT;
1578                         goto error_brelse;
1579                 }
1580         }
1581
1582         if (device->is_tgtdev_for_dev_replace) {
1583                 ret = BTRFS_ERROR_DEV_TGT_REPLACE;
1584                 goto error_brelse;
1585         }
1586
1587         if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1588                 ret = BTRFS_ERROR_DEV_ONLY_WRITABLE;
1589                 goto error_brelse;
1590         }
1591
1592         if (device->writeable) {
1593                 lock_chunks(root);
1594                 list_del_init(&device->dev_alloc_list);
1595                 unlock_chunks(root);
1596                 root->fs_info->fs_devices->rw_devices--;
1597                 clear_super = true;
1598         }
1599
1600         mutex_unlock(&uuid_mutex);
1601         ret = btrfs_shrink_device(device, 0);
1602         mutex_lock(&uuid_mutex);
1603         if (ret)
1604                 goto error_undo;
1605
1606         /*
1607          * TODO: the superblock still includes this device in its num_devices
1608          * counter although write_all_supers() is not locked out. This
1609          * could give a filesystem state which requires a degraded mount.
1610          */
1611         ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1612         if (ret)
1613                 goto error_undo;
1614
1615         spin_lock(&root->fs_info->free_chunk_lock);
1616         root->fs_info->free_chunk_space = device->total_bytes -
1617                 device->bytes_used;
1618         spin_unlock(&root->fs_info->free_chunk_lock);
1619
1620         device->in_fs_metadata = 0;
1621         btrfs_scrub_cancel_dev(root->fs_info, device);
1622
1623         /*
1624          * the device list mutex makes sure that we don't change
1625          * the device list while someone else is writing out all
1626          * the device supers. Whoever is writing all supers, should
1627          * lock the device list mutex before getting the number of
1628          * devices in the super block (super_copy). Conversely,
1629          * whoever updates the number of devices in the super block
1630          * (super_copy) should hold the device list mutex.
1631          */
1632
1633         cur_devices = device->fs_devices;
1634         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1635         list_del_rcu(&device->dev_list);
1636
1637         device->fs_devices->num_devices--;
1638         device->fs_devices->total_devices--;
1639
1640         if (device->missing)
1641                 root->fs_info->fs_devices->missing_devices--;
1642
1643         next_device = list_entry(root->fs_info->fs_devices->devices.next,
1644                                  struct btrfs_device, dev_list);
1645         if (device->bdev == root->fs_info->sb->s_bdev)
1646                 root->fs_info->sb->s_bdev = next_device->bdev;
1647         if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1648                 root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1649
1650         if (device->bdev)
1651                 device->fs_devices->open_devices--;
1652
1653         call_rcu(&device->rcu, free_device);
1654
1655         num_devices = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
1656         btrfs_set_super_num_devices(root->fs_info->super_copy, num_devices);
1657         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1658
1659         if (cur_devices->open_devices == 0) {
1660                 struct btrfs_fs_devices *fs_devices;
1661                 fs_devices = root->fs_info->fs_devices;
1662                 while (fs_devices) {
1663                         if (fs_devices->seed == cur_devices)
1664                                 break;
1665                         fs_devices = fs_devices->seed;
1666                 }
1667                 fs_devices->seed = cur_devices->seed;
1668                 cur_devices->seed = NULL;
1669                 lock_chunks(root);
1670                 __btrfs_close_devices(cur_devices);
1671                 unlock_chunks(root);
1672                 free_fs_devices(cur_devices);
1673         }
1674
1675         root->fs_info->num_tolerated_disk_barrier_failures =
1676                 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
1677
1678         /*
1679          * at this point, the device is zero sized.  We want to
1680          * remove it from the devices list and zero out the old super
1681          */
1682         if (clear_super && disk_super) {
1683                 /* make sure this device isn't detected as part of
1684                  * the FS anymore
1685                  */
1686                 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1687                 set_buffer_dirty(bh);
1688                 sync_dirty_buffer(bh);
1689         }
1690
1691         ret = 0;
1692
1693         /* Notify udev that device has changed */
1694         if (bdev)
1695                 btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
1696
1697 error_brelse:
1698         brelse(bh);
1699         if (bdev)
1700                 blkdev_put(bdev, FMODE_READ | FMODE_EXCL);
1701 out:
1702         mutex_unlock(&uuid_mutex);
1703         return ret;
1704 error_undo:
1705         if (device->writeable) {
1706                 lock_chunks(root);
1707                 list_add(&device->dev_alloc_list,
1708                          &root->fs_info->fs_devices->alloc_list);
1709                 unlock_chunks(root);
1710                 root->fs_info->fs_devices->rw_devices++;
1711         }
1712         goto error_brelse;
1713 }
1714
1715 void btrfs_rm_dev_replace_srcdev(struct btrfs_fs_info *fs_info,
1716                                  struct btrfs_device *srcdev)
1717 {
1718         WARN_ON(!mutex_is_locked(&fs_info->fs_devices->device_list_mutex));
1719
1720         list_del_rcu(&srcdev->dev_list);
1721         list_del_rcu(&srcdev->dev_alloc_list);
1722         fs_info->fs_devices->num_devices--;
1723         if (srcdev->missing) {
1724                 fs_info->fs_devices->missing_devices--;
1725                 fs_info->fs_devices->rw_devices++;
1726         }
1727         if (srcdev->can_discard)
1728                 fs_info->fs_devices->num_can_discard--;
1729         if (srcdev->bdev) {
1730                 fs_info->fs_devices->open_devices--;
1731
1732                 /* zero out the old super */
1733                 btrfs_scratch_superblock(srcdev);
1734         }
1735
1736         call_rcu(&srcdev->rcu, free_device);
1737 }
1738
1739 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
1740                                       struct btrfs_device *tgtdev)
1741 {
1742         struct btrfs_device *next_device;
1743
1744         WARN_ON(!tgtdev);
1745         mutex_lock(&fs_info->fs_devices->device_list_mutex);
1746         if (tgtdev->bdev) {
1747                 btrfs_scratch_superblock(tgtdev);
1748                 fs_info->fs_devices->open_devices--;
1749         }
1750         fs_info->fs_devices->num_devices--;
1751         if (tgtdev->can_discard)
1752                 fs_info->fs_devices->num_can_discard++;
1753
1754         next_device = list_entry(fs_info->fs_devices->devices.next,
1755                                  struct btrfs_device, dev_list);
1756         if (tgtdev->bdev == fs_info->sb->s_bdev)
1757                 fs_info->sb->s_bdev = next_device->bdev;
1758         if (tgtdev->bdev == fs_info->fs_devices->latest_bdev)
1759                 fs_info->fs_devices->latest_bdev = next_device->bdev;
1760         list_del_rcu(&tgtdev->dev_list);
1761
1762         call_rcu(&tgtdev->rcu, free_device);
1763
1764         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
1765 }
1766
1767 static int btrfs_find_device_by_path(struct btrfs_root *root, char *device_path,
1768                                      struct btrfs_device **device)
1769 {
1770         int ret = 0;
1771         struct btrfs_super_block *disk_super;
1772         u64 devid;
1773         u8 *dev_uuid;
1774         struct block_device *bdev;
1775         struct buffer_head *bh;
1776
1777         *device = NULL;
1778         ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
1779                                     root->fs_info->bdev_holder, 0, &bdev, &bh);
1780         if (ret)
1781                 return ret;
1782         disk_super = (struct btrfs_super_block *)bh->b_data;
1783         devid = btrfs_stack_device_id(&disk_super->dev_item);
1784         dev_uuid = disk_super->dev_item.uuid;
1785         *device = btrfs_find_device(root->fs_info, devid, dev_uuid,
1786                                     disk_super->fsid);
1787         brelse(bh);
1788         if (!*device)
1789                 ret = -ENOENT;
1790         blkdev_put(bdev, FMODE_READ);
1791         return ret;
1792 }
1793
1794 int btrfs_find_device_missing_or_by_path(struct btrfs_root *root,
1795                                          char *device_path,
1796                                          struct btrfs_device **device)
1797 {
1798         *device = NULL;
1799         if (strcmp(device_path, "missing") == 0) {
1800                 struct list_head *devices;
1801                 struct btrfs_device *tmp;
1802
1803                 devices = &root->fs_info->fs_devices->devices;
1804                 /*
1805                  * It is safe to read the devices since the volume_mutex
1806                  * is held by the caller.
1807                  */
1808                 list_for_each_entry(tmp, devices, dev_list) {
1809                         if (tmp->in_fs_metadata && !tmp->bdev) {
1810                                 *device = tmp;
1811                                 break;
1812                         }
1813                 }
1814
1815                 if (!*device) {
1816                         btrfs_err(root->fs_info, "no missing device found");
1817                         return -ENOENT;
1818                 }
1819
1820                 return 0;
1821         } else {
1822                 return btrfs_find_device_by_path(root, device_path, device);
1823         }
1824 }
1825
1826 /*
1827  * does all the dirty work required for changing file system's UUID.
1828  */
1829 static int btrfs_prepare_sprout(struct btrfs_root *root)
1830 {
1831         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
1832         struct btrfs_fs_devices *old_devices;
1833         struct btrfs_fs_devices *seed_devices;
1834         struct btrfs_super_block *disk_super = root->fs_info->super_copy;
1835         struct btrfs_device *device;
1836         u64 super_flags;
1837
1838         BUG_ON(!mutex_is_locked(&uuid_mutex));
1839         if (!fs_devices->seeding)
1840                 return -EINVAL;
1841
1842         seed_devices = __alloc_fs_devices();
1843         if (IS_ERR(seed_devices))
1844                 return PTR_ERR(seed_devices);
1845
1846         old_devices = clone_fs_devices(fs_devices);
1847         if (IS_ERR(old_devices)) {
1848                 kfree(seed_devices);
1849                 return PTR_ERR(old_devices);
1850         }
1851
1852         list_add(&old_devices->list, &fs_uuids);
1853
1854         memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
1855         seed_devices->opened = 1;
1856         INIT_LIST_HEAD(&seed_devices->devices);
1857         INIT_LIST_HEAD(&seed_devices->alloc_list);
1858         mutex_init(&seed_devices->device_list_mutex);
1859
1860         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1861         list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
1862                               synchronize_rcu);
1863
1864         list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
1865         list_for_each_entry(device, &seed_devices->devices, dev_list) {
1866                 device->fs_devices = seed_devices;
1867         }
1868
1869         fs_devices->seeding = 0;
1870         fs_devices->num_devices = 0;
1871         fs_devices->open_devices = 0;
1872         fs_devices->total_devices = 0;
1873         fs_devices->seed = seed_devices;
1874
1875         generate_random_uuid(fs_devices->fsid);
1876         memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1877         memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1878         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1879
1880         super_flags = btrfs_super_flags(disk_super) &
1881                       ~BTRFS_SUPER_FLAG_SEEDING;
1882         btrfs_set_super_flags(disk_super, super_flags);
1883
1884         return 0;
1885 }
1886
1887 /*
1888  * strore the expected generation for seed devices in device items.
1889  */
1890 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
1891                                struct btrfs_root *root)
1892 {
1893         struct btrfs_path *path;
1894         struct extent_buffer *leaf;
1895         struct btrfs_dev_item *dev_item;
1896         struct btrfs_device *device;
1897         struct btrfs_key key;
1898         u8 fs_uuid[BTRFS_UUID_SIZE];
1899         u8 dev_uuid[BTRFS_UUID_SIZE];
1900         u64 devid;
1901         int ret;
1902
1903         path = btrfs_alloc_path();
1904         if (!path)
1905                 return -ENOMEM;
1906
1907         root = root->fs_info->chunk_root;
1908         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1909         key.offset = 0;
1910         key.type = BTRFS_DEV_ITEM_KEY;
1911
1912         while (1) {
1913                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1914                 if (ret < 0)
1915                         goto error;
1916
1917                 leaf = path->nodes[0];
1918 next_slot:
1919                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1920                         ret = btrfs_next_leaf(root, path);
1921                         if (ret > 0)
1922                                 break;
1923                         if (ret < 0)
1924                                 goto error;
1925                         leaf = path->nodes[0];
1926                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1927                         btrfs_release_path(path);
1928                         continue;
1929                 }
1930
1931                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1932                 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
1933                     key.type != BTRFS_DEV_ITEM_KEY)
1934                         break;
1935
1936                 dev_item = btrfs_item_ptr(leaf, path->slots[0],
1937                                           struct btrfs_dev_item);
1938                 devid = btrfs_device_id(leaf, dev_item);
1939                 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
1940                                    BTRFS_UUID_SIZE);
1941                 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
1942                                    BTRFS_UUID_SIZE);
1943                 device = btrfs_find_device(root->fs_info, devid, dev_uuid,
1944                                            fs_uuid);
1945                 BUG_ON(!device); /* Logic error */
1946
1947                 if (device->fs_devices->seeding) {
1948                         btrfs_set_device_generation(leaf, dev_item,
1949                                                     device->generation);
1950                         btrfs_mark_buffer_dirty(leaf);
1951                 }
1952
1953                 path->slots[0]++;
1954                 goto next_slot;
1955         }
1956         ret = 0;
1957 error:
1958         btrfs_free_path(path);
1959         return ret;
1960 }
1961
1962 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
1963 {
1964         struct request_queue *q;
1965         struct btrfs_trans_handle *trans;
1966         struct btrfs_device *device;
1967         struct block_device *bdev;
1968         struct list_head *devices;
1969         struct super_block *sb = root->fs_info->sb;
1970         struct rcu_string *name;
1971         u64 total_bytes;
1972         int seeding_dev = 0;
1973         int ret = 0;
1974
1975         if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
1976                 return -EROFS;
1977
1978         bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
1979                                   root->fs_info->bdev_holder);
1980         if (IS_ERR(bdev))
1981                 return PTR_ERR(bdev);
1982
1983         if (root->fs_info->fs_devices->seeding) {
1984                 seeding_dev = 1;
1985                 down_write(&sb->s_umount);
1986                 mutex_lock(&uuid_mutex);
1987         }
1988
1989         filemap_write_and_wait(bdev->bd_inode->i_mapping);
1990
1991         devices = &root->fs_info->fs_devices->devices;
1992
1993         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1994         list_for_each_entry(device, devices, dev_list) {
1995                 if (device->bdev == bdev) {
1996                         ret = -EEXIST;
1997                         mutex_unlock(
1998                                 &root->fs_info->fs_devices->device_list_mutex);
1999                         goto error;
2000                 }
2001         }
2002         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2003
2004         device = btrfs_alloc_device(root->fs_info, NULL, NULL);
2005         if (IS_ERR(device)) {
2006                 /* we can safely leave the fs_devices entry around */
2007                 ret = PTR_ERR(device);
2008                 goto error;
2009         }
2010
2011         name = rcu_string_strdup(device_path, GFP_NOFS);
2012         if (!name) {
2013                 kfree(device);
2014                 ret = -ENOMEM;
2015                 goto error;
2016         }
2017         rcu_assign_pointer(device->name, name);
2018
2019         trans = btrfs_start_transaction(root, 0);
2020         if (IS_ERR(trans)) {
2021                 rcu_string_free(device->name);
2022                 kfree(device);
2023                 ret = PTR_ERR(trans);
2024                 goto error;
2025         }
2026
2027         lock_chunks(root);
2028
2029         q = bdev_get_queue(bdev);
2030         if (blk_queue_discard(q))
2031                 device->can_discard = 1;
2032         device->writeable = 1;
2033         device->generation = trans->transid;
2034         device->io_width = root->sectorsize;
2035         device->io_align = root->sectorsize;
2036         device->sector_size = root->sectorsize;
2037         device->total_bytes = i_size_read(bdev->bd_inode);
2038         device->disk_total_bytes = device->total_bytes;
2039         device->dev_root = root->fs_info->dev_root;
2040         device->bdev = bdev;
2041         device->in_fs_metadata = 1;
2042         device->is_tgtdev_for_dev_replace = 0;
2043         device->mode = FMODE_EXCL;
2044         device->dev_stats_valid = 1;
2045         set_blocksize(device->bdev, 4096);
2046
2047         if (seeding_dev) {
2048                 sb->s_flags &= ~MS_RDONLY;
2049                 ret = btrfs_prepare_sprout(root);
2050                 BUG_ON(ret); /* -ENOMEM */
2051         }
2052
2053         device->fs_devices = root->fs_info->fs_devices;
2054
2055         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2056         list_add_rcu(&device->dev_list, &root->fs_info->fs_devices->devices);
2057         list_add(&device->dev_alloc_list,
2058                  &root->fs_info->fs_devices->alloc_list);
2059         root->fs_info->fs_devices->num_devices++;
2060         root->fs_info->fs_devices->open_devices++;
2061         root->fs_info->fs_devices->rw_devices++;
2062         root->fs_info->fs_devices->total_devices++;
2063         if (device->can_discard)
2064                 root->fs_info->fs_devices->num_can_discard++;
2065         root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
2066
2067         spin_lock(&root->fs_info->free_chunk_lock);
2068         root->fs_info->free_chunk_space += device->total_bytes;
2069         spin_unlock(&root->fs_info->free_chunk_lock);
2070
2071         if (!blk_queue_nonrot(bdev_get_queue(bdev)))
2072                 root->fs_info->fs_devices->rotating = 1;
2073
2074         total_bytes = btrfs_super_total_bytes(root->fs_info->super_copy);
2075         btrfs_set_super_total_bytes(root->fs_info->super_copy,
2076                                     total_bytes + device->total_bytes);
2077
2078         total_bytes = btrfs_super_num_devices(root->fs_info->super_copy);
2079         btrfs_set_super_num_devices(root->fs_info->super_copy,
2080                                     total_bytes + 1);
2081         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2082
2083         if (seeding_dev) {
2084                 ret = init_first_rw_device(trans, root, device);
2085                 if (ret) {
2086                         btrfs_abort_transaction(trans, root, ret);
2087                         goto error_trans;
2088                 }
2089                 ret = btrfs_finish_sprout(trans, root);
2090                 if (ret) {
2091                         btrfs_abort_transaction(trans, root, ret);
2092                         goto error_trans;
2093                 }
2094         } else {
2095                 ret = btrfs_add_device(trans, root, device);
2096                 if (ret) {
2097                         btrfs_abort_transaction(trans, root, ret);
2098                         goto error_trans;
2099                 }
2100         }
2101
2102         /*
2103          * we've got more storage, clear any full flags on the space
2104          * infos
2105          */
2106         btrfs_clear_space_info_full(root->fs_info);
2107
2108         unlock_chunks(root);
2109         root->fs_info->num_tolerated_disk_barrier_failures =
2110                 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
2111         ret = btrfs_commit_transaction(trans, root);
2112
2113         if (seeding_dev) {
2114                 mutex_unlock(&uuid_mutex);
2115                 up_write(&sb->s_umount);
2116
2117                 if (ret) /* transaction commit */
2118                         return ret;
2119
2120                 ret = btrfs_relocate_sys_chunks(root);
2121                 if (ret < 0)
2122                         btrfs_error(root->fs_info, ret,
2123                                     "Failed to relocate sys chunks after "
2124                                     "device initialization. This can be fixed "
2125                                     "using the \"btrfs balance\" command.");
2126                 trans = btrfs_attach_transaction(root);
2127                 if (IS_ERR(trans)) {
2128                         if (PTR_ERR(trans) == -ENOENT)
2129                                 return 0;
2130                         return PTR_ERR(trans);
2131                 }
2132                 ret = btrfs_commit_transaction(trans, root);
2133         }
2134
2135         return ret;
2136
2137 error_trans:
2138         unlock_chunks(root);
2139         btrfs_end_transaction(trans, root);
2140         rcu_string_free(device->name);
2141         kfree(device);
2142 error:
2143         blkdev_put(bdev, FMODE_EXCL);
2144         if (seeding_dev) {
2145                 mutex_unlock(&uuid_mutex);
2146                 up_write(&sb->s_umount);
2147         }
2148         return ret;
2149 }
2150
2151 int btrfs_init_dev_replace_tgtdev(struct btrfs_root *root, char *device_path,
2152                                   struct btrfs_device **device_out)
2153 {
2154         struct request_queue *q;
2155         struct btrfs_device *device;
2156         struct block_device *bdev;
2157         struct btrfs_fs_info *fs_info = root->fs_info;
2158         struct list_head *devices;
2159         struct rcu_string *name;
2160         u64 devid = BTRFS_DEV_REPLACE_DEVID;
2161         int ret = 0;
2162
2163         *device_out = NULL;
2164         if (fs_info->fs_devices->seeding)
2165                 return -EINVAL;
2166
2167         bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2168                                   fs_info->bdev_holder);
2169         if (IS_ERR(bdev))
2170                 return PTR_ERR(bdev);
2171
2172         filemap_write_and_wait(bdev->bd_inode->i_mapping);
2173
2174         devices = &fs_info->fs_devices->devices;
2175         list_for_each_entry(device, devices, dev_list) {
2176                 if (device->bdev == bdev) {
2177                         ret = -EEXIST;
2178                         goto error;
2179                 }
2180         }
2181
2182         device = btrfs_alloc_device(NULL, &devid, NULL);
2183         if (IS_ERR(device)) {
2184                 ret = PTR_ERR(device);
2185                 goto error;
2186         }
2187
2188         name = rcu_string_strdup(device_path, GFP_NOFS);
2189         if (!name) {
2190                 kfree(device);
2191                 ret = -ENOMEM;
2192                 goto error;
2193         }
2194         rcu_assign_pointer(device->name, name);
2195
2196         q = bdev_get_queue(bdev);
2197         if (blk_queue_discard(q))
2198                 device->can_discard = 1;
2199         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2200         device->writeable = 1;
2201         device->generation = 0;
2202         device->io_width = root->sectorsize;
2203         device->io_align = root->sectorsize;
2204         device->sector_size = root->sectorsize;
2205         device->total_bytes = i_size_read(bdev->bd_inode);
2206         device->disk_total_bytes = device->total_bytes;
2207         device->dev_root = fs_info->dev_root;
2208         device->bdev = bdev;
2209         device->in_fs_metadata = 1;
2210         device->is_tgtdev_for_dev_replace = 1;
2211         device->mode = FMODE_EXCL;
2212         device->dev_stats_valid = 1;
2213         set_blocksize(device->bdev, 4096);
2214         device->fs_devices = fs_info->fs_devices;
2215         list_add(&device->dev_list, &fs_info->fs_devices->devices);
2216         fs_info->fs_devices->num_devices++;
2217         fs_info->fs_devices->open_devices++;
2218         if (device->can_discard)
2219                 fs_info->fs_devices->num_can_discard++;
2220         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2221
2222         *device_out = device;
2223         return ret;
2224
2225 error:
2226         blkdev_put(bdev, FMODE_EXCL);
2227         return ret;
2228 }
2229
2230 void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info *fs_info,
2231                                               struct btrfs_device *tgtdev)
2232 {
2233         WARN_ON(fs_info->fs_devices->rw_devices == 0);
2234         tgtdev->io_width = fs_info->dev_root->sectorsize;
2235         tgtdev->io_align = fs_info->dev_root->sectorsize;
2236         tgtdev->sector_size = fs_info->dev_root->sectorsize;
2237         tgtdev->dev_root = fs_info->dev_root;
2238         tgtdev->in_fs_metadata = 1;
2239 }
2240
2241 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2242                                         struct btrfs_device *device)
2243 {
2244         int ret;
2245         struct btrfs_path *path;
2246         struct btrfs_root *root;
2247         struct btrfs_dev_item *dev_item;
2248         struct extent_buffer *leaf;
2249         struct btrfs_key key;
2250
2251         root = device->dev_root->fs_info->chunk_root;
2252
2253         path = btrfs_alloc_path();
2254         if (!path)
2255                 return -ENOMEM;
2256
2257         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2258         key.type = BTRFS_DEV_ITEM_KEY;
2259         key.offset = device->devid;
2260
2261         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2262         if (ret < 0)
2263                 goto out;
2264
2265         if (ret > 0) {
2266                 ret = -ENOENT;
2267                 goto out;
2268         }
2269
2270         leaf = path->nodes[0];
2271         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2272
2273         btrfs_set_device_id(leaf, dev_item, device->devid);
2274         btrfs_set_device_type(leaf, dev_item, device->type);
2275         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2276         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2277         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2278         btrfs_set_device_total_bytes(leaf, dev_item, device->disk_total_bytes);
2279         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
2280         btrfs_mark_buffer_dirty(leaf);
2281
2282 out:
2283         btrfs_free_path(path);
2284         return ret;
2285 }
2286
2287 static int __btrfs_grow_device(struct btrfs_trans_handle *trans,
2288                       struct btrfs_device *device, u64 new_size)
2289 {
2290         struct btrfs_super_block *super_copy =
2291                 device->dev_root->fs_info->super_copy;
2292         u64 old_total = btrfs_super_total_bytes(super_copy);
2293         u64 diff = new_size - device->total_bytes;
2294
2295         if (!device->writeable)
2296                 return -EACCES;
2297         if (new_size <= device->total_bytes ||
2298             device->is_tgtdev_for_dev_replace)
2299                 return -EINVAL;
2300
2301         btrfs_set_super_total_bytes(super_copy, old_total + diff);
2302         device->fs_devices->total_rw_bytes += diff;
2303
2304         device->total_bytes = new_size;
2305         device->disk_total_bytes = new_size;
2306         btrfs_clear_space_info_full(device->dev_root->fs_info);
2307
2308         return btrfs_update_device(trans, device);
2309 }
2310
2311 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2312                       struct btrfs_device *device, u64 new_size)
2313 {
2314         int ret;
2315         lock_chunks(device->dev_root);
2316         ret = __btrfs_grow_device(trans, device, new_size);
2317         unlock_chunks(device->dev_root);
2318         return ret;
2319 }
2320
2321 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
2322                             struct btrfs_root *root,
2323                             u64 chunk_tree, u64 chunk_objectid,
2324                             u64 chunk_offset)
2325 {
2326         int ret;
2327         struct btrfs_path *path;
2328         struct btrfs_key key;
2329
2330         root = root->fs_info->chunk_root;
2331         path = btrfs_alloc_path();
2332         if (!path)
2333                 return -ENOMEM;
2334
2335         key.objectid = chunk_objectid;
2336         key.offset = chunk_offset;
2337         key.type = BTRFS_CHUNK_ITEM_KEY;
2338
2339         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2340         if (ret < 0)
2341                 goto out;
2342         else if (ret > 0) { /* Logic error or corruption */
2343                 btrfs_error(root->fs_info, -ENOENT,
2344                             "Failed lookup while freeing chunk.");
2345                 ret = -ENOENT;
2346                 goto out;
2347         }
2348
2349         ret = btrfs_del_item(trans, root, path);
2350         if (ret < 0)
2351                 btrfs_error(root->fs_info, ret,
2352                             "Failed to delete chunk item.");
2353 out:
2354         btrfs_free_path(path);
2355         return ret;
2356 }
2357
2358 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
2359                         chunk_offset)
2360 {
2361         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
2362         struct btrfs_disk_key *disk_key;
2363         struct btrfs_chunk *chunk;
2364         u8 *ptr;
2365         int ret = 0;
2366         u32 num_stripes;
2367         u32 array_size;
2368         u32 len = 0;
2369         u32 cur;
2370         struct btrfs_key key;
2371
2372         array_size = btrfs_super_sys_array_size(super_copy);
2373
2374         ptr = super_copy->sys_chunk_array;
2375         cur = 0;
2376
2377         while (cur < array_size) {
2378                 disk_key = (struct btrfs_disk_key *)ptr;
2379                 btrfs_disk_key_to_cpu(&key, disk_key);
2380
2381                 len = sizeof(*disk_key);
2382
2383                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2384                         chunk = (struct btrfs_chunk *)(ptr + len);
2385                         num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2386                         len += btrfs_chunk_item_size(num_stripes);
2387                 } else {
2388                         ret = -EIO;
2389                         break;
2390                 }
2391                 if (key.objectid == chunk_objectid &&
2392                     key.offset == chunk_offset) {
2393                         memmove(ptr, ptr + len, array_size - (cur + len));
2394                         array_size -= len;
2395                         btrfs_set_super_sys_array_size(super_copy, array_size);
2396                 } else {
2397                         ptr += len;
2398                         cur += len;
2399                 }
2400         }
2401         return ret;
2402 }
2403
2404 static int btrfs_relocate_chunk(struct btrfs_root *root,
2405                          u64 chunk_tree, u64 chunk_objectid,
2406                          u64 chunk_offset)
2407 {
2408         struct extent_map_tree *em_tree;
2409         struct btrfs_root *extent_root;
2410         struct btrfs_trans_handle *trans;
2411         struct extent_map *em;
2412         struct map_lookup *map;
2413         int ret;
2414         int i;
2415
2416         root = root->fs_info->chunk_root;
2417         extent_root = root->fs_info->extent_root;
2418         em_tree = &root->fs_info->mapping_tree.map_tree;
2419
2420         ret = btrfs_can_relocate(extent_root, chunk_offset);
2421         if (ret)
2422                 return -ENOSPC;
2423
2424         /* step one, relocate all the extents inside this chunk */
2425         ret = btrfs_relocate_block_group(extent_root, chunk_offset);
2426         if (ret)
2427                 return ret;
2428
2429         trans = btrfs_start_transaction(root, 0);
2430         if (IS_ERR(trans)) {
2431                 ret = PTR_ERR(trans);
2432                 btrfs_std_error(root->fs_info, ret);
2433                 return ret;
2434         }
2435
2436         lock_chunks(root);
2437
2438         /*
2439          * step two, delete the device extents and the
2440          * chunk tree entries
2441          */
2442         read_lock(&em_tree->lock);
2443         em = lookup_extent_mapping(em_tree, chunk_offset, 1);
2444         read_unlock(&em_tree->lock);
2445
2446         BUG_ON(!em || em->start > chunk_offset ||
2447                em->start + em->len < chunk_offset);
2448         map = (struct map_lookup *)em->bdev;
2449
2450         for (i = 0; i < map->num_stripes; i++) {
2451                 ret = btrfs_free_dev_extent(trans, map->stripes[i].dev,
2452                                             map->stripes[i].physical);
2453                 BUG_ON(ret);
2454
2455                 if (map->stripes[i].dev) {
2456                         ret = btrfs_update_device(trans, map->stripes[i].dev);
2457                         BUG_ON(ret);
2458                 }
2459         }
2460         ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
2461                                chunk_offset);
2462
2463         BUG_ON(ret);
2464
2465         trace_btrfs_chunk_free(root, map, chunk_offset, em->len);
2466
2467         if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2468                 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
2469                 BUG_ON(ret);
2470         }
2471
2472         ret = btrfs_remove_block_group(trans, extent_root, chunk_offset);
2473         BUG_ON(ret);
2474
2475         write_lock(&em_tree->lock);
2476         remove_extent_mapping(em_tree, em);
2477         write_unlock(&em_tree->lock);
2478
2479         kfree(map);
2480         em->bdev = NULL;
2481
2482         /* once for the tree */
2483         free_extent_map(em);
2484         /* once for us */
2485         free_extent_map(em);
2486
2487         unlock_chunks(root);
2488         btrfs_end_transaction(trans, root);
2489         return 0;
2490 }
2491
2492 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
2493 {
2494         struct btrfs_root *chunk_root = root->fs_info->chunk_root;
2495         struct btrfs_path *path;
2496         struct extent_buffer *leaf;
2497         struct btrfs_chunk *chunk;
2498         struct btrfs_key key;
2499         struct btrfs_key found_key;
2500         u64 chunk_tree = chunk_root->root_key.objectid;
2501         u64 chunk_type;
2502         bool retried = false;
2503         int failed = 0;
2504         int ret;
2505
2506         path = btrfs_alloc_path();
2507         if (!path)
2508                 return -ENOMEM;
2509
2510 again:
2511         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2512         key.offset = (u64)-1;
2513         key.type = BTRFS_CHUNK_ITEM_KEY;
2514
2515         while (1) {
2516                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2517                 if (ret < 0)
2518                         goto error;
2519                 BUG_ON(ret == 0); /* Corruption */
2520
2521                 ret = btrfs_previous_item(chunk_root, path, key.objectid,
2522                                           key.type);
2523                 if (ret < 0)
2524                         goto error;
2525                 if (ret > 0)
2526                         break;
2527
2528                 leaf = path->nodes[0];
2529                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2530
2531                 chunk = btrfs_item_ptr(leaf, path->slots[0],
2532                                        struct btrfs_chunk);
2533                 chunk_type = btrfs_chunk_type(leaf, chunk);
2534                 btrfs_release_path(path);
2535
2536                 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
2537                         ret = btrfs_relocate_chunk(chunk_root, chunk_tree,
2538                                                    found_key.objectid,
2539                                                    found_key.offset);
2540                         if (ret == -ENOSPC)
2541                                 failed++;
2542                         else if (ret)
2543                                 BUG();
2544                 }
2545
2546                 if (found_key.offset == 0)
2547                         break;
2548                 key.offset = found_key.offset - 1;
2549         }
2550         ret = 0;
2551         if (failed && !retried) {
2552                 failed = 0;
2553                 retried = true;
2554                 goto again;
2555         } else if (WARN_ON(failed && retried)) {
2556                 ret = -ENOSPC;
2557         }
2558 error:
2559         btrfs_free_path(path);
2560         return ret;
2561 }
2562
2563 static int insert_balance_item(struct btrfs_root *root,
2564                                struct btrfs_balance_control *bctl)
2565 {
2566         struct btrfs_trans_handle *trans;
2567         struct btrfs_balance_item *item;
2568         struct btrfs_disk_balance_args disk_bargs;
2569         struct btrfs_path *path;
2570         struct extent_buffer *leaf;
2571         struct btrfs_key key;
2572         int ret, err;
2573
2574         path = btrfs_alloc_path();
2575         if (!path)
2576                 return -ENOMEM;
2577
2578         trans = btrfs_start_transaction(root, 0);
2579         if (IS_ERR(trans)) {
2580                 btrfs_free_path(path);
2581                 return PTR_ERR(trans);
2582         }
2583
2584         key.objectid = BTRFS_BALANCE_OBJECTID;
2585         key.type = BTRFS_BALANCE_ITEM_KEY;
2586         key.offset = 0;
2587
2588         ret = btrfs_insert_empty_item(trans, root, path, &key,
2589                                       sizeof(*item));
2590         if (ret)
2591                 goto out;
2592
2593         leaf = path->nodes[0];
2594         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
2595
2596         memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
2597
2598         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
2599         btrfs_set_balance_data(leaf, item, &disk_bargs);
2600         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
2601         btrfs_set_balance_meta(leaf, item, &disk_bargs);
2602         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
2603         btrfs_set_balance_sys(leaf, item, &disk_bargs);
2604
2605         btrfs_set_balance_flags(leaf, item, bctl->flags);
2606
2607         btrfs_mark_buffer_dirty(leaf);
2608 out:
2609         btrfs_free_path(path);
2610         err = btrfs_commit_transaction(trans, root);
2611         if (err && !ret)
2612                 ret = err;
2613         return ret;
2614 }
2615
2616 static int del_balance_item(struct btrfs_root *root)
2617 {
2618         struct btrfs_trans_handle *trans;
2619         struct btrfs_path *path;
2620         struct btrfs_key key;
2621         int ret, err;
2622
2623         path = btrfs_alloc_path();
2624         if (!path)
2625                 return -ENOMEM;
2626
2627         trans = btrfs_start_transaction(root, 0);
2628         if (IS_ERR(trans)) {
2629                 btrfs_free_path(path);
2630                 return PTR_ERR(trans);
2631         }
2632
2633         key.objectid = BTRFS_BALANCE_OBJECTID;
2634         key.type = BTRFS_BALANCE_ITEM_KEY;
2635         key.offset = 0;
2636
2637         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2638         if (ret < 0)
2639                 goto out;
2640         if (ret > 0) {
2641                 ret = -ENOENT;
2642                 goto out;
2643         }
2644
2645         ret = btrfs_del_item(trans, root, path);
2646 out:
2647         btrfs_free_path(path);
2648         err = btrfs_commit_transaction(trans, root);
2649         if (err && !ret)
2650                 ret = err;
2651         return ret;
2652 }
2653
2654 /*
2655  * This is a heuristic used to reduce the number of chunks balanced on
2656  * resume after balance was interrupted.
2657  */
2658 static void update_balance_args(struct btrfs_balance_control *bctl)
2659 {
2660         /*
2661          * Turn on soft mode for chunk types that were being converted.
2662          */
2663         if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
2664                 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
2665         if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
2666                 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
2667         if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
2668                 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
2669
2670         /*
2671          * Turn on usage filter if is not already used.  The idea is
2672          * that chunks that we have already balanced should be
2673          * reasonably full.  Don't do it for chunks that are being
2674          * converted - that will keep us from relocating unconverted
2675          * (albeit full) chunks.
2676          */
2677         if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2678             !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2679                 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
2680                 bctl->data.usage = 90;
2681         }
2682         if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2683             !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2684                 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
2685                 bctl->sys.usage = 90;
2686         }
2687         if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2688             !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2689                 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
2690                 bctl->meta.usage = 90;
2691         }
2692 }
2693
2694 /*
2695  * Should be called with both balance and volume mutexes held to
2696  * serialize other volume operations (add_dev/rm_dev/resize) with
2697  * restriper.  Same goes for unset_balance_control.
2698  */
2699 static void set_balance_control(struct btrfs_balance_control *bctl)
2700 {
2701         struct btrfs_fs_info *fs_info = bctl->fs_info;
2702
2703         BUG_ON(fs_info->balance_ctl);
2704
2705         spin_lock(&fs_info->balance_lock);
2706         fs_info->balance_ctl = bctl;
2707         spin_unlock(&fs_info->balance_lock);
2708 }
2709
2710 static void unset_balance_control(struct btrfs_fs_info *fs_info)
2711 {
2712         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
2713
2714         BUG_ON(!fs_info->balance_ctl);
2715
2716         spin_lock(&fs_info->balance_lock);
2717         fs_info->balance_ctl = NULL;
2718         spin_unlock(&fs_info->balance_lock);
2719
2720         kfree(bctl);
2721 }
2722
2723 /*
2724  * Balance filters.  Return 1 if chunk should be filtered out
2725  * (should not be balanced).
2726  */
2727 static int chunk_profiles_filter(u64 chunk_type,
2728                                  struct btrfs_balance_args *bargs)
2729 {
2730         chunk_type = chunk_to_extended(chunk_type) &
2731                                 BTRFS_EXTENDED_PROFILE_MASK;
2732
2733         if (bargs->profiles & chunk_type)
2734                 return 0;
2735
2736         return 1;
2737 }
2738
2739 static int chunk_usage_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
2740                               struct btrfs_balance_args *bargs)
2741 {
2742         struct btrfs_block_group_cache *cache;
2743         u64 chunk_used, user_thresh;
2744         int ret = 1;
2745
2746         cache = btrfs_lookup_block_group(fs_info, chunk_offset);
2747         chunk_used = btrfs_block_group_used(&cache->item);
2748
2749         if (bargs->usage == 0)
2750                 user_thresh = 1;
2751         else if (bargs->usage > 100)
2752                 user_thresh = cache->key.offset;
2753         else
2754                 user_thresh = div_factor_fine(cache->key.offset,
2755                                               bargs->usage);
2756
2757         if (chunk_used < user_thresh)
2758                 ret = 0;
2759
2760         btrfs_put_block_group(cache);
2761         return ret;
2762 }
2763
2764 static int chunk_devid_filter(struct extent_buffer *leaf,
2765                               struct btrfs_chunk *chunk,
2766                               struct btrfs_balance_args *bargs)
2767 {
2768         struct btrfs_stripe *stripe;
2769         int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2770         int i;
2771
2772         for (i = 0; i < num_stripes; i++) {
2773                 stripe = btrfs_stripe_nr(chunk, i);
2774                 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
2775                         return 0;
2776         }
2777
2778         return 1;
2779 }
2780
2781 /* [pstart, pend) */
2782 static int chunk_drange_filter(struct extent_buffer *leaf,
2783                                struct btrfs_chunk *chunk,
2784                                u64 chunk_offset,
2785                                struct btrfs_balance_args *bargs)
2786 {
2787         struct btrfs_stripe *stripe;
2788         int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2789         u64 stripe_offset;
2790         u64 stripe_length;
2791         int factor;
2792         int i;
2793
2794         if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
2795                 return 0;
2796
2797         if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP |
2798              BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) {
2799                 factor = num_stripes / 2;
2800         } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) {
2801                 factor = num_stripes - 1;
2802         } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) {
2803                 factor = num_stripes - 2;
2804         } else {
2805                 factor = num_stripes;
2806         }
2807
2808         for (i = 0; i < num_stripes; i++) {
2809                 stripe = btrfs_stripe_nr(chunk, i);
2810                 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
2811                         continue;
2812
2813                 stripe_offset = btrfs_stripe_offset(leaf, stripe);
2814                 stripe_length = btrfs_chunk_length(leaf, chunk);
2815                 do_div(stripe_length, factor);
2816
2817                 if (stripe_offset < bargs->pend &&
2818                     stripe_offset + stripe_length > bargs->pstart)
2819                         return 0;
2820         }
2821
2822         return 1;
2823 }
2824
2825 /* [vstart, vend) */
2826 static int chunk_vrange_filter(struct extent_buffer *leaf,
2827                                struct btrfs_chunk *chunk,
2828                                u64 chunk_offset,
2829                                struct btrfs_balance_args *bargs)
2830 {
2831         if (chunk_offset < bargs->vend &&
2832             chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
2833                 /* at least part of the chunk is inside this vrange */
2834                 return 0;
2835
2836         return 1;
2837 }
2838
2839 static int chunk_soft_convert_filter(u64 chunk_type,
2840                                      struct btrfs_balance_args *bargs)
2841 {
2842         if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
2843                 return 0;
2844
2845         chunk_type = chunk_to_extended(chunk_type) &
2846                                 BTRFS_EXTENDED_PROFILE_MASK;
2847
2848         if (bargs->target == chunk_type)
2849                 return 1;
2850
2851         return 0;
2852 }
2853
2854 static int should_balance_chunk(struct btrfs_root *root,
2855                                 struct extent_buffer *leaf,
2856                                 struct btrfs_chunk *chunk, u64 chunk_offset)
2857 {
2858         struct btrfs_balance_control *bctl = root->fs_info->balance_ctl;
2859         struct btrfs_balance_args *bargs = NULL;
2860         u64 chunk_type = btrfs_chunk_type(leaf, chunk);
2861
2862         /* type filter */
2863         if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
2864               (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
2865                 return 0;
2866         }
2867
2868         if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
2869                 bargs = &bctl->data;
2870         else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
2871                 bargs = &bctl->sys;
2872         else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
2873                 bargs = &bctl->meta;
2874
2875         /* profiles filter */
2876         if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
2877             chunk_profiles_filter(chunk_type, bargs)) {
2878                 return 0;
2879         }
2880
2881         /* usage filter */
2882         if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
2883             chunk_usage_filter(bctl->fs_info, chunk_offset, bargs)) {
2884                 return 0;
2885         }
2886
2887         /* devid filter */
2888         if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
2889             chunk_devid_filter(leaf, chunk, bargs)) {
2890                 return 0;
2891         }
2892
2893         /* drange filter, makes sense only with devid filter */
2894         if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
2895             chunk_drange_filter(leaf, chunk, chunk_offset, bargs)) {
2896                 return 0;
2897         }
2898
2899         /* vrange filter */
2900         if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
2901             chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
2902                 return 0;
2903         }
2904
2905         /* soft profile changing mode */
2906         if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
2907             chunk_soft_convert_filter(chunk_type, bargs)) {
2908                 return 0;
2909         }
2910
2911         return 1;
2912 }
2913
2914 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
2915 {
2916         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
2917         struct btrfs_root *chunk_root = fs_info->chunk_root;
2918         struct btrfs_root *dev_root = fs_info->dev_root;
2919         struct list_head *devices;
2920         struct btrfs_device *device;
2921         u64 old_size;
2922         u64 size_to_free;
2923         struct btrfs_chunk *chunk;
2924         struct btrfs_path *path;
2925         struct btrfs_key key;
2926         struct btrfs_key found_key;
2927         struct btrfs_trans_handle *trans;
2928         struct extent_buffer *leaf;
2929         int slot;
2930         int ret;
2931         int enospc_errors = 0;
2932         bool counting = true;
2933
2934         /* step one make some room on all the devices */
2935         devices = &fs_info->fs_devices->devices;
2936         list_for_each_entry(device, devices, dev_list) {
2937                 old_size = device->total_bytes;
2938                 size_to_free = div_factor(old_size, 1);
2939                 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
2940                 if (!device->writeable ||
2941                     device->total_bytes - device->bytes_used > size_to_free ||
2942                     device->is_tgtdev_for_dev_replace)
2943                         continue;
2944
2945                 ret = btrfs_shrink_device(device, old_size - size_to_free);
2946                 if (ret == -ENOSPC)
2947                         break;
2948                 BUG_ON(ret);
2949
2950                 trans = btrfs_start_transaction(dev_root, 0);
2951                 BUG_ON(IS_ERR(trans));
2952
2953                 ret = btrfs_grow_device(trans, device, old_size);
2954                 BUG_ON(ret);
2955
2956                 btrfs_end_transaction(trans, dev_root);
2957         }
2958
2959         /* step two, relocate all the chunks */
2960         path = btrfs_alloc_path();
2961         if (!path) {
2962                 ret = -ENOMEM;
2963                 goto error;
2964         }
2965
2966         /* zero out stat counters */
2967         spin_lock(&fs_info->balance_lock);
2968         memset(&bctl->stat, 0, sizeof(bctl->stat));
2969         spin_unlock(&fs_info->balance_lock);
2970 again:
2971         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2972         key.offset = (u64)-1;
2973         key.type = BTRFS_CHUNK_ITEM_KEY;
2974
2975         while (1) {
2976                 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
2977                     atomic_read(&fs_info->balance_cancel_req)) {
2978                         ret = -ECANCELED;
2979                         goto error;
2980                 }
2981
2982                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2983                 if (ret < 0)
2984                         goto error;
2985
2986                 /*
2987                  * this shouldn't happen, it means the last relocate
2988                  * failed
2989                  */
2990                 if (ret == 0)
2991                         BUG(); /* FIXME break ? */
2992
2993                 ret = btrfs_previous_item(chunk_root, path, 0,
2994                                           BTRFS_CHUNK_ITEM_KEY);
2995                 if (ret) {
2996                         ret = 0;
2997                         break;
2998                 }
2999
3000                 leaf = path->nodes[0];
3001                 slot = path->slots[0];
3002                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3003
3004                 if (found_key.objectid != key.objectid)
3005                         break;
3006
3007                 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3008
3009                 if (!counting) {
3010                         spin_lock(&fs_info->balance_lock);
3011                         bctl->stat.considered++;
3012                         spin_unlock(&fs_info->balance_lock);
3013                 }
3014
3015                 ret = should_balance_chunk(chunk_root, leaf, chunk,
3016                                            found_key.offset);
3017                 btrfs_release_path(path);
3018                 if (!ret)
3019                         goto loop;
3020
3021                 if (counting) {
3022                         spin_lock(&fs_info->balance_lock);
3023                         bctl->stat.expected++;
3024                         spin_unlock(&fs_info->balance_lock);
3025                         goto loop;
3026                 }
3027
3028                 ret = btrfs_relocate_chunk(chunk_root,
3029                                            chunk_root->root_key.objectid,
3030                                            found_key.objectid,
3031                                            found_key.offset);
3032                 if (ret && ret != -ENOSPC)
3033                         goto error;
3034                 if (ret == -ENOSPC) {
3035                         enospc_errors++;
3036                 } else {
3037                         spin_lock(&fs_info->balance_lock);
3038                         bctl->stat.completed++;
3039                         spin_unlock(&fs_info->balance_lock);
3040                 }
3041 loop:
3042                 if (found_key.offset == 0)
3043                         break;
3044                 key.offset = found_key.offset - 1;
3045         }
3046
3047         if (counting) {
3048                 btrfs_release_path(path);
3049                 counting = false;
3050                 goto again;
3051         }
3052 error:
3053         btrfs_free_path(path);
3054         if (enospc_errors) {
3055                 btrfs_info(fs_info, "%d enospc errors during balance",
3056                        enospc_errors);
3057                 if (!ret)
3058                         ret = -ENOSPC;
3059         }
3060
3061         return ret;
3062 }
3063
3064 /**
3065  * alloc_profile_is_valid - see if a given profile is valid and reduced
3066  * @flags: profile to validate
3067  * @extended: if true @flags is treated as an extended profile
3068  */
3069 static int alloc_profile_is_valid(u64 flags, int extended)
3070 {
3071         u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
3072                                BTRFS_BLOCK_GROUP_PROFILE_MASK);
3073
3074         flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
3075
3076         /* 1) check that all other bits are zeroed */
3077         if (flags & ~mask)
3078                 return 0;
3079
3080         /* 2) see if profile is reduced */
3081         if (flags == 0)
3082                 return !extended; /* "0" is valid for usual profiles */
3083
3084         /* true if exactly one bit set */
3085         return (flags & (flags - 1)) == 0;
3086 }
3087
3088 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
3089 {
3090         /* cancel requested || normal exit path */
3091         return atomic_read(&fs_info->balance_cancel_req) ||
3092                 (atomic_read(&fs_info->balance_pause_req) == 0 &&
3093                  atomic_read(&fs_info->balance_cancel_req) == 0);
3094 }
3095
3096 static void __cancel_balance(struct btrfs_fs_info *fs_info)
3097 {
3098         int ret;
3099
3100         unset_balance_control(fs_info);
3101         ret = del_balance_item(fs_info->tree_root);
3102         if (ret)
3103                 btrfs_std_error(fs_info, ret);
3104
3105         atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3106 }
3107
3108 /*
3109  * Should be called with both balance and volume mutexes held
3110  */
3111 int btrfs_balance(struct btrfs_balance_control *bctl,
3112                   struct btrfs_ioctl_balance_args *bargs)
3113 {
3114         struct btrfs_fs_info *fs_info = bctl->fs_info;
3115         u64 allowed;
3116         int mixed = 0;
3117         int ret;
3118         u64 num_devices;
3119         unsigned seq;
3120
3121         if (btrfs_fs_closing(fs_info) ||
3122             atomic_read(&fs_info->balance_pause_req) ||
3123             atomic_read(&fs_info->balance_cancel_req)) {
3124                 ret = -EINVAL;
3125                 goto out;
3126         }
3127
3128         allowed = btrfs_super_incompat_flags(fs_info->super_copy);
3129         if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
3130                 mixed = 1;
3131
3132         /*
3133          * In case of mixed groups both data and meta should be picked,
3134          * and identical options should be given for both of them.
3135          */
3136         allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
3137         if (mixed && (bctl->flags & allowed)) {
3138                 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
3139                     !(bctl->flags & BTRFS_BALANCE_METADATA) ||
3140                     memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
3141                         btrfs_err(fs_info, "with mixed groups data and "
3142                                    "metadata balance options must be the same");
3143                         ret = -EINVAL;
3144                         goto out;
3145                 }
3146         }
3147
3148         num_devices = fs_info->fs_devices->num_devices;
3149         btrfs_dev_replace_lock(&fs_info->dev_replace);
3150         if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
3151                 BUG_ON(num_devices < 1);
3152                 num_devices--;
3153         }
3154         btrfs_dev_replace_unlock(&fs_info->dev_replace);
3155         allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE;
3156         if (num_devices == 1)
3157                 allowed |= BTRFS_BLOCK_GROUP_DUP;
3158         else if (num_devices > 1)
3159                 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1);
3160         if (num_devices > 2)
3161                 allowed |= BTRFS_BLOCK_GROUP_RAID5;
3162         if (num_devices > 3)
3163                 allowed |= (BTRFS_BLOCK_GROUP_RAID10 |
3164                             BTRFS_BLOCK_GROUP_RAID6);
3165         if ((bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3166             (!alloc_profile_is_valid(bctl->data.target, 1) ||
3167              (bctl->data.target & ~allowed))) {
3168                 btrfs_err(fs_info, "unable to start balance with target "
3169                            "data profile %llu",
3170                        bctl->data.target);
3171                 ret = -EINVAL;
3172                 goto out;
3173         }
3174         if ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3175             (!alloc_profile_is_valid(bctl->meta.target, 1) ||
3176              (bctl->meta.target & ~allowed))) {
3177                 btrfs_err(fs_info,
3178                            "unable to start balance with target metadata profile %llu",
3179                        bctl->meta.target);
3180                 ret = -EINVAL;
3181                 goto out;
3182         }
3183         if ((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3184             (!alloc_profile_is_valid(bctl->sys.target, 1) ||
3185              (bctl->sys.target & ~allowed))) {
3186                 btrfs_err(fs_info,
3187                            "unable to start balance with target system profile %llu",
3188                        bctl->sys.target);
3189                 ret = -EINVAL;
3190                 goto out;
3191         }
3192
3193         /* allow dup'ed data chunks only in mixed mode */
3194         if (!mixed && (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3195             (bctl->data.target & BTRFS_BLOCK_GROUP_DUP)) {
3196                 btrfs_err(fs_info, "dup for data is not allowed");
3197                 ret = -EINVAL;
3198                 goto out;
3199         }
3200
3201         /* allow to reduce meta or sys integrity only if force set */
3202         allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3203                         BTRFS_BLOCK_GROUP_RAID10 |
3204                         BTRFS_BLOCK_GROUP_RAID5 |
3205                         BTRFS_BLOCK_GROUP_RAID6;
3206         do {
3207                 seq = read_seqbegin(&fs_info->profiles_lock);
3208
3209                 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3210                      (fs_info->avail_system_alloc_bits & allowed) &&
3211                      !(bctl->sys.target & allowed)) ||
3212                     ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3213                      (fs_info->avail_metadata_alloc_bits & allowed) &&
3214                      !(bctl->meta.target & allowed))) {
3215                         if (bctl->flags & BTRFS_BALANCE_FORCE) {
3216                                 btrfs_info(fs_info, "force reducing metadata integrity");
3217                         } else {
3218                                 btrfs_err(fs_info, "balance will reduce metadata "
3219                                            "integrity, use force if you want this");
3220                                 ret = -EINVAL;
3221                                 goto out;
3222                         }
3223                 }
3224         } while (read_seqretry(&fs_info->profiles_lock, seq));
3225
3226         if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3227                 int num_tolerated_disk_barrier_failures;
3228                 u64 target = bctl->sys.target;
3229
3230                 num_tolerated_disk_barrier_failures =
3231                         btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3232                 if (num_tolerated_disk_barrier_failures > 0 &&
3233                     (target &
3234                      (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID0 |
3235                       BTRFS_AVAIL_ALLOC_BIT_SINGLE)))
3236                         num_tolerated_disk_barrier_failures = 0;
3237                 else if (num_tolerated_disk_barrier_failures > 1 &&
3238                          (target &
3239                           (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)))
3240                         num_tolerated_disk_barrier_failures = 1;
3241
3242                 fs_info->num_tolerated_disk_barrier_failures =
3243                         num_tolerated_disk_barrier_failures;
3244         }
3245
3246         ret = insert_balance_item(fs_info->tree_root, bctl);
3247         if (ret && ret != -EEXIST)
3248                 goto out;
3249
3250         if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
3251                 BUG_ON(ret == -EEXIST);
3252                 set_balance_control(bctl);
3253         } else {
3254                 BUG_ON(ret != -EEXIST);
3255                 spin_lock(&fs_info->balance_lock);
3256                 update_balance_args(bctl);
3257                 spin_unlock(&fs_info->balance_lock);
3258         }
3259
3260         atomic_inc(&fs_info->balance_running);
3261         mutex_unlock(&fs_info->balance_mutex);
3262
3263         ret = __btrfs_balance(fs_info);
3264
3265         mutex_lock(&fs_info->balance_mutex);
3266         atomic_dec(&fs_info->balance_running);
3267
3268         if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3269                 fs_info->num_tolerated_disk_barrier_failures =
3270                         btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3271         }
3272
3273         if (bargs) {
3274                 memset(bargs, 0, sizeof(*bargs));
3275                 update_ioctl_balance_args(fs_info, 0, bargs);
3276         }
3277
3278         if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
3279             balance_need_close(fs_info)) {
3280                 __cancel_balance(fs_info);
3281         }
3282
3283         wake_up(&fs_info->balance_wait_q);
3284
3285         return ret;
3286 out:
3287         if (bctl->flags & BTRFS_BALANCE_RESUME)
3288                 __cancel_balance(fs_info);
3289         else {
3290                 kfree(bctl);
3291                 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3292         }
3293         return ret;
3294 }
3295
3296 static int balance_kthread(void *data)
3297 {
3298         struct btrfs_fs_info *fs_info = data;
3299         int ret = 0;
3300
3301         mutex_lock(&fs_info->volume_mutex);
3302         mutex_lock(&fs_info->balance_mutex);
3303
3304         if (fs_info->balance_ctl) {
3305                 btrfs_info(fs_info, "continuing balance");
3306                 ret = btrfs_balance(fs_info->balance_ctl, NULL);
3307         }
3308
3309         mutex_unlock(&fs_info->balance_mutex);
3310         mutex_unlock(&fs_info->volume_mutex);
3311
3312         return ret;
3313 }
3314
3315 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
3316 {
3317         struct task_struct *tsk;
3318
3319         spin_lock(&fs_info->balance_lock);
3320         if (!fs_info->balance_ctl) {
3321                 spin_unlock(&fs_info->balance_lock);
3322                 return 0;
3323         }
3324         spin_unlock(&fs_info->balance_lock);
3325
3326         if (btrfs_test_opt(fs_info->tree_root, SKIP_BALANCE)) {
3327                 btrfs_info(fs_info, "force skipping balance");
3328                 return 0;
3329         }
3330
3331         tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
3332         return PTR_ERR_OR_ZERO(tsk);
3333 }
3334
3335 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
3336 {
3337         struct btrfs_balance_control *bctl;
3338         struct btrfs_balance_item *item;
3339         struct btrfs_disk_balance_args disk_bargs;
3340         struct btrfs_path *path;
3341         struct extent_buffer *leaf;
3342         struct btrfs_key key;
3343         int ret;
3344
3345         path = btrfs_alloc_path();
3346         if (!path)
3347                 return -ENOMEM;
3348
3349         key.objectid = BTRFS_BALANCE_OBJECTID;
3350         key.type = BTRFS_BALANCE_ITEM_KEY;
3351         key.offset = 0;
3352
3353         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
3354         if (ret < 0)
3355                 goto out;
3356         if (ret > 0) { /* ret = -ENOENT; */
3357                 ret = 0;
3358                 goto out;
3359         }
3360
3361         bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
3362         if (!bctl) {
3363                 ret = -ENOMEM;
3364                 goto out;
3365         }
3366
3367         leaf = path->nodes[0];
3368         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3369
3370         bctl->fs_info = fs_info;
3371         bctl->flags = btrfs_balance_flags(leaf, item);
3372         bctl->flags |= BTRFS_BALANCE_RESUME;
3373
3374         btrfs_balance_data(leaf, item, &disk_bargs);
3375         btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
3376         btrfs_balance_meta(leaf, item, &disk_bargs);
3377         btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
3378         btrfs_balance_sys(leaf, item, &disk_bargs);
3379         btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
3380
3381         WARN_ON(atomic_xchg(&fs_info->mutually_exclusive_operation_running, 1));
3382
3383         mutex_lock(&fs_info->volume_mutex);
3384         mutex_lock(&fs_info->balance_mutex);
3385
3386         set_balance_control(bctl);
3387
3388         mutex_unlock(&fs_info->balance_mutex);
3389         mutex_unlock(&fs_info->volume_mutex);
3390 out:
3391         btrfs_free_path(path);
3392         return ret;
3393 }
3394
3395 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
3396 {
3397         int ret = 0;
3398
3399         mutex_lock(&fs_info->balance_mutex);
3400         if (!fs_info->balance_ctl) {
3401                 mutex_unlock(&fs_info->balance_mutex);
3402                 return -ENOTCONN;
3403         }
3404
3405         if (atomic_read(&fs_info->balance_running)) {
3406                 atomic_inc(&fs_info->balance_pause_req);
3407                 mutex_unlock(&fs_info->balance_mutex);
3408
3409                 wait_event(fs_info->balance_wait_q,
3410                            atomic_read(&fs_info->balance_running) == 0);
3411
3412                 mutex_lock(&fs_info->balance_mutex);
3413                 /* we are good with balance_ctl ripped off from under us */
3414                 BUG_ON(atomic_read(&fs_info->balance_running));
3415                 atomic_dec(&fs_info->balance_pause_req);
3416         } else {
3417                 ret = -ENOTCONN;
3418         }
3419
3420         mutex_unlock(&fs_info->balance_mutex);
3421         return ret;
3422 }
3423
3424 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
3425 {
3426         if (fs_info->sb->s_flags & MS_RDONLY)
3427                 return -EROFS;
3428
3429         mutex_lock(&fs_info->balance_mutex);
3430         if (!fs_info->balance_ctl) {
3431                 mutex_unlock(&fs_info->balance_mutex);
3432                 return -ENOTCONN;
3433         }
3434
3435         atomic_inc(&fs_info->balance_cancel_req);
3436         /*
3437          * if we are running just wait and return, balance item is
3438          * deleted in btrfs_balance in this case
3439          */
3440         if (atomic_read(&fs_info->balance_running)) {
3441                 mutex_unlock(&fs_info->balance_mutex);
3442                 wait_event(fs_info->balance_wait_q,
3443                            atomic_read(&fs_info->balance_running) == 0);
3444                 mutex_lock(&fs_info->balance_mutex);
3445         } else {
3446                 /* __cancel_balance needs volume_mutex */
3447                 mutex_unlock(&fs_info->balance_mutex);
3448                 mutex_lock(&fs_info->volume_mutex);
3449                 mutex_lock(&fs_info->balance_mutex);
3450
3451                 if (fs_info->balance_ctl)
3452                         __cancel_balance(fs_info);
3453
3454                 mutex_unlock(&fs_info->volume_mutex);
3455         }
3456
3457         BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running));
3458         atomic_dec(&fs_info->balance_cancel_req);
3459         mutex_unlock(&fs_info->balance_mutex);
3460         return 0;
3461 }
3462
3463 static int btrfs_uuid_scan_kthread(void *data)
3464 {
3465         struct btrfs_fs_info *fs_info = data;
3466         struct btrfs_root *root = fs_info->tree_root;
3467         struct btrfs_key key;
3468         struct btrfs_key max_key;
3469         struct btrfs_path *path = NULL;
3470         int ret = 0;
3471         struct extent_buffer *eb;
3472         int slot;
3473         struct btrfs_root_item root_item;
3474         u32 item_size;
3475         struct btrfs_trans_handle *trans = NULL;
3476
3477         path = btrfs_alloc_path();
3478         if (!path) {
3479                 ret = -ENOMEM;
3480                 goto out;
3481         }
3482
3483         key.objectid = 0;
3484         key.type = BTRFS_ROOT_ITEM_KEY;
3485         key.offset = 0;
3486
3487         max_key.objectid = (u64)-1;
3488         max_key.type = BTRFS_ROOT_ITEM_KEY;
3489         max_key.offset = (u64)-1;
3490
3491         path->keep_locks = 1;
3492
3493         while (1) {
3494                 ret = btrfs_search_forward(root, &key, path, 0);
3495                 if (ret) {
3496                         if (ret > 0)
3497                                 ret = 0;
3498                         break;
3499                 }
3500
3501                 if (key.type != BTRFS_ROOT_ITEM_KEY ||
3502                     (key.objectid < BTRFS_FIRST_FREE_OBJECTID &&
3503                      key.objectid != BTRFS_FS_TREE_OBJECTID) ||
3504                     key.objectid > BTRFS_LAST_FREE_OBJECTID)
3505                         goto skip;
3506
3507                 eb = path->nodes[0];
3508                 slot = path->slots[0];
3509                 item_size = btrfs_item_size_nr(eb, slot);
3510                 if (item_size < sizeof(root_item))
3511                         goto skip;
3512
3513                 read_extent_buffer(eb, &root_item,
3514                                    btrfs_item_ptr_offset(eb, slot),
3515                                    (int)sizeof(root_item));
3516                 if (btrfs_root_refs(&root_item) == 0)
3517                         goto skip;
3518
3519                 if (!btrfs_is_empty_uuid(root_item.uuid) ||
3520                     !btrfs_is_empty_uuid(root_item.received_uuid)) {
3521                         if (trans)
3522                                 goto update_tree;
3523
3524                         btrfs_release_path(path);
3525                         /*
3526                          * 1 - subvol uuid item
3527                          * 1 - received_subvol uuid item
3528                          */
3529                         trans = btrfs_start_transaction(fs_info->uuid_root, 2);
3530                         if (IS_ERR(trans)) {
3531                                 ret = PTR_ERR(trans);
3532                                 break;
3533                         }
3534                         continue;
3535                 } else {
3536                         goto skip;
3537                 }
3538 update_tree:
3539                 if (!btrfs_is_empty_uuid(root_item.uuid)) {
3540                         ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
3541                                                   root_item.uuid,
3542                                                   BTRFS_UUID_KEY_SUBVOL,
3543                                                   key.objectid);
3544                         if (ret < 0) {
3545                                 btrfs_warn(fs_info, "uuid_tree_add failed %d",
3546                                         ret);
3547                                 break;
3548                         }
3549                 }
3550
3551                 if (!btrfs_is_empty_uuid(root_item.received_uuid)) {
3552                         ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
3553                                                   root_item.received_uuid,
3554                                                  BTRFS_UUID_KEY_RECEIVED_SUBVOL,
3555                                                   key.objectid);
3556                         if (ret < 0) {
3557                                 btrfs_warn(fs_info, "uuid_tree_add failed %d",
3558                                         ret);
3559                                 break;
3560                         }
3561                 }
3562
3563 skip:
3564                 if (trans) {
3565                         ret = btrfs_end_transaction(trans, fs_info->uuid_root);
3566                         trans = NULL;
3567                         if (ret)
3568                                 break;
3569                 }
3570
3571                 btrfs_release_path(path);
3572                 if (key.offset < (u64)-1) {
3573                         key.offset++;
3574                 } else if (key.type < BTRFS_ROOT_ITEM_KEY) {
3575                         key.offset = 0;
3576                         key.type = BTRFS_ROOT_ITEM_KEY;
3577                 } else if (key.objectid < (u64)-1) {
3578                         key.offset = 0;
3579                         key.type = BTRFS_ROOT_ITEM_KEY;
3580                         key.objectid++;
3581                 } else {
3582                         break;
3583                 }
3584                 cond_resched();
3585         }
3586
3587 out:
3588         btrfs_free_path(path);
3589         if (trans && !IS_ERR(trans))
3590                 btrfs_end_transaction(trans, fs_info->uuid_root);
3591         if (ret)
3592                 btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret);
3593         else
3594                 fs_info->update_uuid_tree_gen = 1;
3595         up(&fs_info->uuid_tree_rescan_sem);
3596         return 0;
3597 }
3598
3599 /*
3600  * Callback for btrfs_uuid_tree_iterate().
3601  * returns:
3602  * 0    check succeeded, the entry is not outdated.
3603  * < 0  if an error occured.
3604  * > 0  if the check failed, which means the caller shall remove the entry.
3605  */
3606 static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info *fs_info,
3607                                        u8 *uuid, u8 type, u64 subid)
3608 {
3609         struct btrfs_key key;
3610         int ret = 0;
3611         struct btrfs_root *subvol_root;
3612
3613         if (type != BTRFS_UUID_KEY_SUBVOL &&
3614             type != BTRFS_UUID_KEY_RECEIVED_SUBVOL)
3615                 goto out;
3616
3617         key.objectid = subid;
3618         key.type = BTRFS_ROOT_ITEM_KEY;
3619         key.offset = (u64)-1;
3620         subvol_root = btrfs_read_fs_root_no_name(fs_info, &key);
3621         if (IS_ERR(subvol_root)) {
3622                 ret = PTR_ERR(subvol_root);
3623                 if (ret == -ENOENT)
3624                         ret = 1;
3625                 goto out;
3626         }
3627
3628         switch (type) {
3629         case BTRFS_UUID_KEY_SUBVOL:
3630                 if (memcmp(uuid, subvol_root->root_item.uuid, BTRFS_UUID_SIZE))
3631                         ret = 1;
3632                 break;
3633         case BTRFS_UUID_KEY_RECEIVED_SUBVOL:
3634                 if (memcmp(uuid, subvol_root->root_item.received_uuid,
3635                            BTRFS_UUID_SIZE))
3636                         ret = 1;
3637                 break;
3638         }
3639
3640 out:
3641         return ret;
3642 }
3643
3644 static int btrfs_uuid_rescan_kthread(void *data)
3645 {
3646         struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data;
3647         int ret;
3648
3649         /*
3650          * 1st step is to iterate through the existing UUID tree and
3651          * to delete all entries that contain outdated data.
3652          * 2nd step is to add all missing entries to the UUID tree.
3653          */
3654         ret = btrfs_uuid_tree_iterate(fs_info, btrfs_check_uuid_tree_entry);
3655         if (ret < 0) {
3656                 btrfs_warn(fs_info, "iterating uuid_tree failed %d", ret);
3657                 up(&fs_info->uuid_tree_rescan_sem);
3658                 return ret;
3659         }
3660         return btrfs_uuid_scan_kthread(data);
3661 }
3662
3663 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info)
3664 {
3665         struct btrfs_trans_handle *trans;
3666         struct btrfs_root *tree_root = fs_info->tree_root;
3667         struct btrfs_root *uuid_root;
3668         struct task_struct *task;
3669         int ret;
3670
3671         /*
3672          * 1 - root node
3673          * 1 - root item
3674          */
3675         trans = btrfs_start_transaction(tree_root, 2);
3676         if (IS_ERR(trans))
3677                 return PTR_ERR(trans);
3678
3679         uuid_root = btrfs_create_tree(trans, fs_info,
3680                                       BTRFS_UUID_TREE_OBJECTID);
3681         if (IS_ERR(uuid_root)) {
3682                 btrfs_abort_transaction(trans, tree_root,
3683                                         PTR_ERR(uuid_root));
3684                 return PTR_ERR(uuid_root);
3685         }
3686
3687         fs_info->uuid_root = uuid_root;
3688
3689         ret = btrfs_commit_transaction(trans, tree_root);
3690         if (ret)
3691                 return ret;
3692
3693         down(&fs_info->uuid_tree_rescan_sem);
3694         task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid");
3695         if (IS_ERR(task)) {
3696                 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
3697                 btrfs_warn(fs_info, "failed to start uuid_scan task");
3698                 up(&fs_info->uuid_tree_rescan_sem);
3699                 return PTR_ERR(task);
3700         }
3701
3702         return 0;
3703 }
3704
3705 int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
3706 {
3707         struct task_struct *task;
3708
3709         down(&fs_info->uuid_tree_rescan_sem);
3710         task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
3711         if (IS_ERR(task)) {
3712                 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
3713                 btrfs_warn(fs_info, "failed to start uuid_rescan task");
3714                 up(&fs_info->uuid_tree_rescan_sem);
3715                 return PTR_ERR(task);
3716         }
3717
3718         return 0;
3719 }
3720
3721 /*
3722  * shrinking a device means finding all of the device extents past
3723  * the new size, and then following the back refs to the chunks.
3724  * The chunk relocation code actually frees the device extent
3725  */
3726 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
3727 {
3728         struct btrfs_trans_handle *trans;
3729         struct btrfs_root *root = device->dev_root;
3730         struct btrfs_dev_extent *dev_extent = NULL;
3731         struct btrfs_path *path;
3732         u64 length;
3733         u64 chunk_tree;
3734         u64 chunk_objectid;
3735         u64 chunk_offset;
3736         int ret;
3737         int slot;
3738         int failed = 0;
3739         bool retried = false;
3740         struct extent_buffer *l;
3741         struct btrfs_key key;
3742         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
3743         u64 old_total = btrfs_super_total_bytes(super_copy);
3744         u64 old_size = device->total_bytes;
3745         u64 diff = device->total_bytes - new_size;
3746
3747         if (device->is_tgtdev_for_dev_replace)
3748                 return -EINVAL;
3749
3750         path = btrfs_alloc_path();
3751         if (!path)
3752                 return -ENOMEM;
3753
3754         path->reada = 2;
3755
3756         lock_chunks(root);
3757
3758         device->total_bytes = new_size;
3759         if (device->writeable) {
3760                 device->fs_devices->total_rw_bytes -= diff;
3761                 spin_lock(&root->fs_info->free_chunk_lock);
3762                 root->fs_info->free_chunk_space -= diff;
3763                 spin_unlock(&root->fs_info->free_chunk_lock);
3764         }
3765         unlock_chunks(root);
3766
3767 again:
3768         key.objectid = device->devid;
3769         key.offset = (u64)-1;
3770         key.type = BTRFS_DEV_EXTENT_KEY;
3771
3772         do {
3773                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3774                 if (ret < 0)
3775                         goto done;
3776
3777                 ret = btrfs_previous_item(root, path, 0, key.type);
3778                 if (ret < 0)
3779                         goto done;
3780                 if (ret) {
3781                         ret = 0;
3782                         btrfs_release_path(path);
3783                         break;
3784                 }
3785
3786                 l = path->nodes[0];
3787                 slot = path->slots[0];
3788                 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
3789
3790                 if (key.objectid != device->devid) {
3791                         btrfs_release_path(path);
3792                         break;
3793                 }
3794
3795                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
3796                 length = btrfs_dev_extent_length(l, dev_extent);
3797
3798                 if (key.offset + length <= new_size) {
3799                         btrfs_release_path(path);
3800                         break;
3801                 }
3802
3803                 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
3804                 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
3805                 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
3806                 btrfs_release_path(path);
3807
3808                 ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
3809                                            chunk_offset);
3810                 if (ret && ret != -ENOSPC)
3811                         goto done;
3812                 if (ret == -ENOSPC)
3813                         failed++;
3814         } while (key.offset-- > 0);
3815
3816         if (failed && !retried) {
3817                 failed = 0;
3818                 retried = true;
3819                 goto again;
3820         } else if (failed && retried) {
3821                 ret = -ENOSPC;
3822                 lock_chunks(root);
3823
3824                 device->total_bytes = old_size;
3825                 if (device->writeable)
3826                         device->fs_devices->total_rw_bytes += diff;
3827                 spin_lock(&root->fs_info->free_chunk_lock);
3828                 root->fs_info->free_chunk_space += diff;
3829                 spin_unlock(&root->fs_info->free_chunk_lock);
3830                 unlock_chunks(root);
3831                 goto done;
3832         }
3833
3834         /* Shrinking succeeded, else we would be at "done". */
3835         trans = btrfs_start_transaction(root, 0);
3836         if (IS_ERR(trans)) {
3837                 ret = PTR_ERR(trans);
3838                 goto done;
3839         }
3840
3841         lock_chunks(root);
3842
3843         device->disk_total_bytes = new_size;
3844         /* Now btrfs_update_device() will change the on-disk size. */
3845         ret = btrfs_update_device(trans, device);
3846         if (ret) {
3847                 unlock_chunks(root);
3848                 btrfs_end_transaction(trans, root);
3849                 goto done;
3850         }
3851         WARN_ON(diff > old_total);
3852         btrfs_set_super_total_bytes(super_copy, old_total - diff);
3853         unlock_chunks(root);
3854         btrfs_end_transaction(trans, root);
3855 done:
3856         btrfs_free_path(path);
3857         return ret;
3858 }
3859
3860 static int btrfs_add_system_chunk(struct btrfs_root *root,
3861                            struct btrfs_key *key,
3862                            struct btrfs_chunk *chunk, int item_size)
3863 {
3864         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
3865         struct btrfs_disk_key disk_key;
3866         u32 array_size;
3867         u8 *ptr;
3868
3869         array_size = btrfs_super_sys_array_size(super_copy);
3870         if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
3871                 return -EFBIG;
3872
3873         ptr = super_copy->sys_chunk_array + array_size;
3874         btrfs_cpu_key_to_disk(&disk_key, key);
3875         memcpy(ptr, &disk_key, sizeof(disk_key));
3876         ptr += sizeof(disk_key);
3877         memcpy(ptr, chunk, item_size);
3878         item_size += sizeof(disk_key);
3879         btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
3880         return 0;
3881 }
3882
3883 /*
3884  * sort the devices in descending order by max_avail, total_avail
3885  */
3886 static int btrfs_cmp_device_info(const void *a, const void *b)
3887 {
3888         const struct btrfs_device_info *di_a = a;
3889         const struct btrfs_device_info *di_b = b;
3890
3891         if (di_a->max_avail > di_b->max_avail)
3892                 return -1;
3893         if (di_a->max_avail < di_b->max_avail)
3894                 return 1;
3895         if (di_a->total_avail > di_b->total_avail)
3896                 return -1;
3897         if (di_a->total_avail < di_b->total_avail)
3898                 return 1;
3899         return 0;
3900 }
3901
3902 static struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = {
3903         [BTRFS_RAID_RAID10] = {
3904                 .sub_stripes    = 2,
3905                 .dev_stripes    = 1,
3906                 .devs_max       = 0,    /* 0 == as many as possible */
3907                 .devs_min       = 4,
3908                 .devs_increment = 2,
3909                 .ncopies        = 2,
3910         },
3911         [BTRFS_RAID_RAID1] = {
3912                 .sub_stripes    = 1,
3913                 .dev_stripes    = 1,
3914                 .devs_max       = 2,
3915                 .devs_min       = 2,
3916                 .devs_increment = 2,
3917                 .ncopies        = 2,
3918         },
3919         [BTRFS_RAID_DUP] = {
3920                 .sub_stripes    = 1,
3921                 .dev_stripes    = 2,
3922                 .devs_max       = 1,
3923                 .devs_min       = 1,
3924                 .devs_increment = 1,
3925                 .ncopies        = 2,
3926         },
3927         [BTRFS_RAID_RAID0] = {
3928                 .sub_stripes    = 1,
3929                 .dev_stripes    = 1,
3930                 .devs_max       = 0,
3931                 .devs_min       = 2,
3932                 .devs_increment = 1,
3933                 .ncopies        = 1,
3934         },
3935         [BTRFS_RAID_SINGLE] = {
3936                 .sub_stripes    = 1,
3937                 .dev_stripes    = 1,
3938                 .devs_max       = 1,
3939                 .devs_min       = 1,
3940                 .devs_increment = 1,
3941                 .ncopies        = 1,
3942         },
3943         [BTRFS_RAID_RAID5] = {
3944                 .sub_stripes    = 1,
3945                 .dev_stripes    = 1,
3946                 .devs_max       = 0,
3947                 .devs_min       = 2,
3948                 .devs_increment = 1,
3949                 .ncopies        = 2,
3950         },
3951         [BTRFS_RAID_RAID6] = {
3952                 .sub_stripes    = 1,
3953                 .dev_stripes    = 1,
3954                 .devs_max       = 0,
3955                 .devs_min       = 3,
3956                 .devs_increment = 1,
3957                 .ncopies        = 3,
3958         },
3959 };
3960
3961 static u32 find_raid56_stripe_len(u32 data_devices, u32 dev_stripe_target)
3962 {
3963         /* TODO allow them to set a preferred stripe size */
3964         return 64 * 1024;
3965 }
3966
3967 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
3968 {
3969         if (!(type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)))
3970                 return;
3971
3972         btrfs_set_fs_incompat(info, RAID56);
3973 }
3974
3975 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
3976                                struct btrfs_root *extent_root, u64 start,
3977                                u64 type)
3978 {
3979         struct btrfs_fs_info *info = extent_root->fs_info;
3980         struct btrfs_fs_devices *fs_devices = info->fs_devices;
3981         struct list_head *cur;
3982         struct map_lookup *map = NULL;
3983         struct extent_map_tree *em_tree;
3984         struct extent_map *em;
3985         struct btrfs_device_info *devices_info = NULL;
3986         u64 total_avail;
3987         int num_stripes;        /* total number of stripes to allocate */
3988         int data_stripes;       /* number of stripes that count for
3989                                    block group size */
3990         int sub_stripes;        /* sub_stripes info for map */
3991         int dev_stripes;        /* stripes per dev */
3992         int devs_max;           /* max devs to use */
3993         int devs_min;           /* min devs needed */
3994         int devs_increment;     /* ndevs has to be a multiple of this */
3995         int ncopies;            /* how many copies to data has */
3996         int ret;
3997         u64 max_stripe_size;
3998         u64 max_chunk_size;
3999         u64 stripe_size;
4000         u64 num_bytes;
4001         u64 raid_stripe_len = BTRFS_STRIPE_LEN;
4002         int ndevs;
4003         int i;
4004         int j;
4005         int index;
4006
4007         BUG_ON(!alloc_profile_is_valid(type, 0));
4008
4009         if (list_empty(&fs_devices->alloc_list))
4010                 return -ENOSPC;
4011
4012         index = __get_raid_index(type);
4013
4014         sub_stripes = btrfs_raid_array[index].sub_stripes;
4015         dev_stripes = btrfs_raid_array[index].dev_stripes;
4016         devs_max = btrfs_raid_array[index].devs_max;
4017         devs_min = btrfs_raid_array[index].devs_min;
4018         devs_increment = btrfs_raid_array[index].devs_increment;
4019         ncopies = btrfs_raid_array[index].ncopies;
4020
4021         if (type & BTRFS_BLOCK_GROUP_DATA) {
4022                 max_stripe_size = 1024 * 1024 * 1024;
4023                 max_chunk_size = 10 * max_stripe_size;
4024         } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
4025                 /* for larger filesystems, use larger metadata chunks */
4026                 if (fs_devices->total_rw_bytes > 50ULL * 1024 * 1024 * 1024)
4027                         max_stripe_size = 1024 * 1024 * 1024;
4028                 else
4029                         max_stripe_size = 256 * 1024 * 1024;
4030                 max_chunk_size = max_stripe_size;
4031         } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
4032                 max_stripe_size = 32 * 1024 * 1024;
4033                 max_chunk_size = 2 * max_stripe_size;
4034         } else {
4035                 btrfs_err(info, "invalid chunk type 0x%llx requested\n",
4036                        type);
4037                 BUG_ON(1);
4038         }
4039
4040         /* we don't want a chunk larger than 10% of writeable space */
4041         max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
4042                              max_chunk_size);
4043
4044         devices_info = kzalloc(sizeof(*devices_info) * fs_devices->rw_devices,
4045                                GFP_NOFS);
4046         if (!devices_info)
4047                 return -ENOMEM;
4048
4049         cur = fs_devices->alloc_list.next;
4050
4051         /*
4052          * in the first pass through the devices list, we gather information
4053          * about the available holes on each device.
4054          */
4055         ndevs = 0;
4056         while (cur != &fs_devices->alloc_list) {
4057                 struct btrfs_device *device;
4058                 u64 max_avail;
4059                 u64 dev_offset;
4060
4061                 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
4062
4063                 cur = cur->next;
4064
4065                 if (!device->writeable) {
4066                         WARN(1, KERN_ERR
4067                                "BTRFS: read-only device in alloc_list\n");
4068                         continue;
4069                 }
4070
4071                 if (!device->in_fs_metadata ||
4072                     device->is_tgtdev_for_dev_replace)
4073                         continue;
4074
4075                 if (device->total_bytes > device->bytes_used)
4076                         total_avail = device->total_bytes - device->bytes_used;
4077                 else
4078                         total_avail = 0;
4079
4080                 /* If there is no space on this device, skip it. */
4081                 if (total_avail == 0)
4082                         continue;
4083
4084                 ret = find_free_dev_extent(trans, device,
4085                                            max_stripe_size * dev_stripes,
4086                                            &dev_offset, &max_avail);
4087                 if (ret && ret != -ENOSPC)
4088                         goto error;
4089
4090                 if (ret == 0)
4091                         max_avail = max_stripe_size * dev_stripes;
4092
4093                 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes)
4094                         continue;
4095
4096                 if (ndevs == fs_devices->rw_devices) {
4097                         WARN(1, "%s: found more than %llu devices\n",
4098                              __func__, fs_devices->rw_devices);
4099                         break;
4100                 }
4101                 devices_info[ndevs].dev_offset = dev_offset;
4102                 devices_info[ndevs].max_avail = max_avail;
4103                 devices_info[ndevs].total_avail = total_avail;
4104                 devices_info[ndevs].dev = device;
4105                 ++ndevs;
4106         }
4107
4108         /*
4109          * now sort the devices by hole size / available space
4110          */
4111         sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
4112              btrfs_cmp_device_info, NULL);
4113
4114         /* round down to number of usable stripes */
4115         ndevs -= ndevs % devs_increment;
4116
4117         if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) {
4118                 ret = -ENOSPC;
4119                 goto error;
4120         }
4121
4122         if (devs_max && ndevs > devs_max)
4123                 ndevs = devs_max;
4124         /*
4125          * the primary goal is to maximize the number of stripes, so use as many
4126          * devices as possible, even if the stripes are not maximum sized.
4127          */
4128         stripe_size = devices_info[ndevs-1].max_avail;
4129         num_stripes = ndevs * dev_stripes;
4130
4131         /*
4132          * this will have to be fixed for RAID1 and RAID10 over
4133          * more drives
4134          */
4135         data_stripes = num_stripes / ncopies;
4136
4137         if (type & BTRFS_BLOCK_GROUP_RAID5) {
4138                 raid_stripe_len = find_raid56_stripe_len(ndevs - 1,
4139                                  btrfs_super_stripesize(info->super_copy));
4140                 data_stripes = num_stripes - 1;
4141         }
4142         if (type & BTRFS_BLOCK_GROUP_RAID6) {
4143                 raid_stripe_len = find_raid56_stripe_len(ndevs - 2,
4144                                  btrfs_super_stripesize(info->super_copy));
4145                 data_stripes = num_stripes - 2;
4146         }
4147
4148         /*
4149          * Use the number of data stripes to figure out how big this chunk
4150          * is really going to be in terms of logical address space,
4151          * and compare that answer with the max chunk size
4152          */
4153         if (stripe_size * data_stripes > max_chunk_size) {
4154                 u64 mask = (1ULL << 24) - 1;
4155                 stripe_size = max_chunk_size;
4156                 do_div(stripe_size, data_stripes);
4157
4158                 /* bump the answer up to a 16MB boundary */
4159                 stripe_size = (stripe_size + mask) & ~mask;
4160
4161                 /* but don't go higher than the limits we found
4162                  * while searching for free extents
4163                  */
4164                 if (stripe_size > devices_info[ndevs-1].max_avail)
4165                         stripe_size = devices_info[ndevs-1].max_avail;
4166         }
4167
4168         do_div(stripe_size, dev_stripes);
4169
4170         /* align to BTRFS_STRIPE_LEN */
4171         do_div(stripe_size, raid_stripe_len);
4172         stripe_size *= raid_stripe_len;
4173
4174         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
4175         if (!map) {
4176                 ret = -ENOMEM;
4177                 goto error;
4178         }
4179         map->num_stripes = num_stripes;
4180
4181         for (i = 0; i < ndevs; ++i) {
4182                 for (j = 0; j < dev_stripes; ++j) {
4183                         int s = i * dev_stripes + j;
4184                         map->stripes[s].dev = devices_info[i].dev;
4185                         map->stripes[s].physical = devices_info[i].dev_offset +
4186                                                    j * stripe_size;
4187                 }
4188         }
4189         map->sector_size = extent_root->sectorsize;
4190         map->stripe_len = raid_stripe_len;
4191         map->io_align = raid_stripe_len;
4192         map->io_width = raid_stripe_len;
4193         map->type = type;
4194         map->sub_stripes = sub_stripes;
4195
4196         num_bytes = stripe_size * data_stripes;
4197
4198         trace_btrfs_chunk_alloc(info->chunk_root, map, start, num_bytes);
4199
4200         em = alloc_extent_map();
4201         if (!em) {
4202                 ret = -ENOMEM;
4203                 goto error;
4204         }
4205         em->bdev = (struct block_device *)map;
4206         em->start = start;
4207         em->len = num_bytes;
4208         em->block_start = 0;
4209         em->block_len = em->len;
4210         em->orig_block_len = stripe_size;
4211
4212         em_tree = &extent_root->fs_info->mapping_tree.map_tree;
4213         write_lock(&em_tree->lock);
4214         ret = add_extent_mapping(em_tree, em, 0);
4215         if (!ret) {
4216                 list_add_tail(&em->list, &trans->transaction->pending_chunks);
4217                 atomic_inc(&em->refs);
4218         }
4219         write_unlock(&em_tree->lock);
4220         if (ret) {
4221                 free_extent_map(em);
4222                 goto error;
4223         }
4224
4225         ret = btrfs_make_block_group(trans, extent_root, 0, type,
4226                                      BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4227                                      start, num_bytes);
4228         if (ret)
4229                 goto error_del_extent;
4230
4231         free_extent_map(em);
4232         check_raid56_incompat_flag(extent_root->fs_info, type);
4233
4234         kfree(devices_info);
4235         return 0;
4236
4237 error_del_extent:
4238         write_lock(&em_tree->lock);
4239         remove_extent_mapping(em_tree, em);
4240         write_unlock(&em_tree->lock);
4241
4242         /* One for our allocation */
4243         free_extent_map(em);
4244         /* One for the tree reference */
4245         free_extent_map(em);
4246 error:
4247         kfree(map);
4248         kfree(devices_info);
4249         return ret;
4250 }
4251
4252 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans,
4253                                 struct btrfs_root *extent_root,
4254                                 u64 chunk_offset, u64 chunk_size)
4255 {
4256         struct btrfs_key key;
4257         struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
4258         struct btrfs_device *device;
4259         struct btrfs_chunk *chunk;
4260         struct btrfs_stripe *stripe;
4261         struct extent_map_tree *em_tree;
4262         struct extent_map *em;
4263         struct map_lookup *map;
4264         size_t item_size;
4265         u64 dev_offset;
4266         u64 stripe_size;
4267         int i = 0;
4268         int ret;
4269
4270         em_tree = &extent_root->fs_info->mapping_tree.map_tree;
4271         read_lock(&em_tree->lock);
4272         em = lookup_extent_mapping(em_tree, chunk_offset, chunk_size);
4273         read_unlock(&em_tree->lock);
4274
4275         if (!em) {
4276                 btrfs_crit(extent_root->fs_info, "unable to find logical "
4277                            "%Lu len %Lu", chunk_offset, chunk_size);
4278                 return -EINVAL;
4279         }
4280
4281         if (em->start != chunk_offset || em->len != chunk_size) {
4282                 btrfs_crit(extent_root->fs_info, "found a bad mapping, wanted"
4283                           " %Lu-%Lu, found %Lu-%Lu\n", chunk_offset,
4284                           chunk_size, em->start, em->len);
4285                 free_extent_map(em);
4286                 return -EINVAL;
4287         }
4288
4289         map = (struct map_lookup *)em->bdev;
4290         item_size = btrfs_chunk_item_size(map->num_stripes);
4291         stripe_size = em->orig_block_len;
4292
4293         chunk = kzalloc(item_size, GFP_NOFS);
4294         if (!chunk) {
4295                 ret = -ENOMEM;
4296                 goto out;
4297         }
4298
4299         for (i = 0; i < map->num_stripes; i++) {
4300                 device = map->stripes[i].dev;
4301                 dev_offset = map->stripes[i].physical;
4302
4303                 device->bytes_used += stripe_size;
4304                 ret = btrfs_update_device(trans, device);
4305                 if (ret)
4306                         goto out;
4307                 ret = btrfs_alloc_dev_extent(trans, device,
4308                                              chunk_root->root_key.objectid,
4309                                              BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4310                                              chunk_offset, dev_offset,
4311                                              stripe_size);
4312                 if (ret)
4313                         goto out;
4314         }
4315
4316         spin_lock(&extent_root->fs_info->free_chunk_lock);
4317         extent_root->fs_info->free_chunk_space -= (stripe_size *
4318                                                    map->num_stripes);
4319         spin_unlock(&extent_root->fs_info->free_chunk_lock);
4320
4321         stripe = &chunk->stripe;
4322         for (i = 0; i < map->num_stripes; i++) {
4323                 device = map->stripes[i].dev;
4324                 dev_offset = map->stripes[i].physical;
4325
4326                 btrfs_set_stack_stripe_devid(stripe, device->devid);
4327                 btrfs_set_stack_stripe_offset(stripe, dev_offset);
4328                 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
4329                 stripe++;
4330         }
4331
4332         btrfs_set_stack_chunk_length(chunk, chunk_size);
4333         btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
4334         btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
4335         btrfs_set_stack_chunk_type(chunk, map->type);
4336         btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
4337         btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
4338         btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
4339         btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
4340         btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
4341
4342         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
4343         key.type = BTRFS_CHUNK_ITEM_KEY;
4344         key.offset = chunk_offset;
4345
4346         ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
4347         if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
4348                 /*
4349                  * TODO: Cleanup of inserted chunk root in case of
4350                  * failure.
4351                  */
4352                 ret = btrfs_add_system_chunk(chunk_root, &key, chunk,
4353                                              item_size);
4354         }
4355
4356 out:
4357         kfree(chunk);
4358         free_extent_map(em);
4359         return ret;
4360 }
4361
4362 /*
4363  * Chunk allocation falls into two parts. The first part does works
4364  * that make the new allocated chunk useable, but not do any operation
4365  * that modifies the chunk tree. The second part does the works that
4366  * require modifying the chunk tree. This division is important for the
4367  * bootstrap process of adding storage to a seed btrfs.
4368  */
4369 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4370                       struct btrfs_root *extent_root, u64 type)
4371 {
4372         u64 chunk_offset;
4373
4374         chunk_offset = find_next_chunk(extent_root->fs_info);
4375         return __btrfs_alloc_chunk(trans, extent_root, chunk_offset, type);
4376 }
4377
4378 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
4379                                          struct btrfs_root *root,
4380                                          struct btrfs_device *device)
4381 {
4382         u64 chunk_offset;
4383         u64 sys_chunk_offset;
4384         u64 alloc_profile;
4385         struct btrfs_fs_info *fs_info = root->fs_info;
4386         struct btrfs_root *extent_root = fs_info->extent_root;
4387         int ret;
4388
4389         chunk_offset = find_next_chunk(fs_info);
4390         alloc_profile = btrfs_get_alloc_profile(extent_root, 0);
4391         ret = __btrfs_alloc_chunk(trans, extent_root, chunk_offset,
4392                                   alloc_profile);
4393         if (ret)
4394                 return ret;
4395
4396         sys_chunk_offset = find_next_chunk(root->fs_info);
4397         alloc_profile = btrfs_get_alloc_profile(fs_info->chunk_root, 0);
4398         ret = __btrfs_alloc_chunk(trans, extent_root, sys_chunk_offset,
4399                                   alloc_profile);
4400         if (ret) {
4401                 btrfs_abort_transaction(trans, root, ret);
4402                 goto out;
4403         }
4404
4405         ret = btrfs_add_device(trans, fs_info->chunk_root, device);
4406         if (ret)
4407                 btrfs_abort_transaction(trans, root, ret);
4408 out:
4409         return ret;
4410 }
4411
4412 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
4413 {
4414         struct extent_map *em;
4415         struct map_lookup *map;
4416         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
4417         int readonly = 0;
4418         int i;
4419
4420         read_lock(&map_tree->map_tree.lock);
4421         em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
4422         read_unlock(&map_tree->map_tree.lock);
4423         if (!em)
4424                 return 1;
4425
4426         if (btrfs_test_opt(root, DEGRADED)) {
4427                 free_extent_map(em);
4428                 return 0;
4429         }
4430
4431         map = (struct map_lookup *)em->bdev;
4432         for (i = 0; i < map->num_stripes; i++) {
4433                 if (!map->stripes[i].dev->writeable) {
4434                         readonly = 1;
4435                         break;
4436                 }
4437         }
4438         free_extent_map(em);
4439         return readonly;
4440 }
4441
4442 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
4443 {
4444         extent_map_tree_init(&tree->map_tree);
4445 }
4446
4447 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
4448 {
4449         struct extent_map *em;
4450
4451         while (1) {
4452                 write_lock(&tree->map_tree.lock);
4453                 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
4454                 if (em)
4455                         remove_extent_mapping(&tree->map_tree, em);
4456                 write_unlock(&tree->map_tree.lock);
4457                 if (!em)
4458                         break;
4459                 kfree(em->bdev);
4460                 /* once for us */
4461                 free_extent_map(em);
4462                 /* once for the tree */
4463                 free_extent_map(em);
4464         }
4465 }
4466
4467 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
4468 {
4469         struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
4470         struct extent_map *em;
4471         struct map_lookup *map;
4472         struct extent_map_tree *em_tree = &map_tree->map_tree;
4473         int ret;
4474
4475         read_lock(&em_tree->lock);
4476         em = lookup_extent_mapping(em_tree, logical, len);
4477         read_unlock(&em_tree->lock);
4478
4479         /*
4480          * We could return errors for these cases, but that could get ugly and
4481          * we'd probably do the same thing which is just not do anything else
4482          * and exit, so return 1 so the callers don't try to use other copies.
4483          */
4484         if (!em) {
4485                 btrfs_crit(fs_info, "No mapping for %Lu-%Lu\n", logical,
4486                             logical+len);
4487                 return 1;
4488         }
4489
4490         if (em->start > logical || em->start + em->len < logical) {
4491                 btrfs_crit(fs_info, "Invalid mapping for %Lu-%Lu, got "
4492                             "%Lu-%Lu\n", logical, logical+len, em->start,
4493                             em->start + em->len);
4494                 free_extent_map(em);
4495                 return 1;
4496         }
4497
4498         map = (struct map_lookup *)em->bdev;
4499         if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
4500                 ret = map->num_stripes;
4501         else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
4502                 ret = map->sub_stripes;
4503         else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
4504                 ret = 2;
4505         else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
4506                 ret = 3;
4507         else
4508                 ret = 1;
4509         free_extent_map(em);
4510
4511         btrfs_dev_replace_lock(&fs_info->dev_replace);
4512         if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace))
4513                 ret++;
4514         btrfs_dev_replace_unlock(&fs_info->dev_replace);
4515
4516         return ret;
4517 }
4518
4519 unsigned long btrfs_full_stripe_len(struct btrfs_root *root,
4520                                     struct btrfs_mapping_tree *map_tree,
4521                                     u64 logical)
4522 {
4523         struct extent_map *em;
4524         struct map_lookup *map;
4525         struct extent_map_tree *em_tree = &map_tree->map_tree;
4526         unsigned long len = root->sectorsize;
4527
4528         read_lock(&em_tree->lock);
4529         em = lookup_extent_mapping(em_tree, logical, len);
4530         read_unlock(&em_tree->lock);
4531         BUG_ON(!em);
4532
4533         BUG_ON(em->start > logical || em->start + em->len < logical);
4534         map = (struct map_lookup *)em->bdev;
4535         if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
4536                          BTRFS_BLOCK_GROUP_RAID6)) {
4537                 len = map->stripe_len * nr_data_stripes(map);
4538         }
4539         free_extent_map(em);
4540         return len;
4541 }
4542
4543 int btrfs_is_parity_mirror(struct btrfs_mapping_tree *map_tree,
4544                            u64 logical, u64 len, int mirror_num)
4545 {
4546         struct extent_map *em;
4547         struct map_lookup *map;
4548         struct extent_map_tree *em_tree = &map_tree->map_tree;
4549         int ret = 0;
4550
4551         read_lock(&em_tree->lock);
4552         em = lookup_extent_mapping(em_tree, logical, len);
4553         read_unlock(&em_tree->lock);
4554         BUG_ON(!em);
4555
4556         BUG_ON(em->start > logical || em->start + em->len < logical);
4557         map = (struct map_lookup *)em->bdev;
4558         if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
4559                          BTRFS_BLOCK_GROUP_RAID6))
4560                 ret = 1;
4561         free_extent_map(em);
4562         return ret;
4563 }
4564
4565 static int find_live_mirror(struct btrfs_fs_info *fs_info,
4566                             struct map_lookup *map, int first, int num,
4567                             int optimal, int dev_replace_is_ongoing)
4568 {
4569         int i;
4570         int tolerance;
4571         struct btrfs_device *srcdev;
4572
4573         if (dev_replace_is_ongoing &&
4574             fs_info->dev_replace.cont_reading_from_srcdev_mode ==
4575              BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
4576                 srcdev = fs_info->dev_replace.srcdev;
4577         else
4578                 srcdev = NULL;
4579
4580         /*
4581          * try to avoid the drive that is the source drive for a
4582          * dev-replace procedure, only choose it if no other non-missing
4583          * mirror is available
4584          */
4585         for (tolerance = 0; tolerance < 2; tolerance++) {
4586                 if (map->stripes[optimal].dev->bdev &&
4587                     (tolerance || map->stripes[optimal].dev != srcdev))
4588                         return optimal;
4589                 for (i = first; i < first + num; i++) {
4590                         if (map->stripes[i].dev->bdev &&
4591                             (tolerance || map->stripes[i].dev != srcdev))
4592                                 return i;
4593                 }
4594         }
4595
4596         /* we couldn't find one that doesn't fail.  Just return something
4597          * and the io error handling code will clean up eventually
4598          */
4599         return optimal;
4600 }
4601
4602 static inline int parity_smaller(u64 a, u64 b)
4603 {
4604         return a > b;
4605 }
4606
4607 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
4608 static void sort_parity_stripes(struct btrfs_bio *bbio, u64 *raid_map)
4609 {
4610         struct btrfs_bio_stripe s;
4611         int i;
4612         u64 l;
4613         int again = 1;
4614
4615         while (again) {
4616                 again = 0;
4617                 for (i = 0; i < bbio->num_stripes - 1; i++) {
4618                         if (parity_smaller(raid_map[i], raid_map[i+1])) {
4619                                 s = bbio->stripes[i];
4620                                 l = raid_map[i];
4621                                 bbio->stripes[i] = bbio->stripes[i+1];
4622                                 raid_map[i] = raid_map[i+1];
4623                                 bbio->stripes[i+1] = s;
4624                                 raid_map[i+1] = l;
4625                                 again = 1;
4626                         }
4627                 }
4628         }
4629 }
4630
4631 static int __btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
4632                              u64 logical, u64 *length,
4633                              struct btrfs_bio **bbio_ret,
4634                              int mirror_num, u64 **raid_map_ret)
4635 {
4636         struct extent_map *em;
4637         struct map_lookup *map;
4638         struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
4639         struct extent_map_tree *em_tree = &map_tree->map_tree;
4640         u64 offset;
4641         u64 stripe_offset;
4642         u64 stripe_end_offset;
4643         u64 stripe_nr;
4644         u64 stripe_nr_orig;
4645         u64 stripe_nr_end;
4646         u64 stripe_len;
4647         u64 *raid_map = NULL;
4648         int stripe_index;
4649         int i;
4650         int ret = 0;
4651         int num_stripes;
4652         int max_errors = 0;
4653         struct btrfs_bio *bbio = NULL;
4654         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
4655         int dev_replace_is_ongoing = 0;
4656         int num_alloc_stripes;
4657         int patch_the_first_stripe_for_dev_replace = 0;
4658         u64 physical_to_patch_in_first_stripe = 0;
4659         u64 raid56_full_stripe_start = (u64)-1;
4660
4661         read_lock(&em_tree->lock);
4662         em = lookup_extent_mapping(em_tree, logical, *length);
4663         read_unlock(&em_tree->lock);
4664
4665         if (!em) {
4666                 btrfs_crit(fs_info, "unable to find logical %llu len %llu",
4667                         logical, *length);
4668                 return -EINVAL;
4669         }
4670
4671         if (em->start > logical || em->start + em->len < logical) {
4672                 btrfs_crit(fs_info, "found a bad mapping, wanted %Lu, "
4673                            "found %Lu-%Lu\n", logical, em->start,
4674                            em->start + em->len);
4675                 free_extent_map(em);
4676                 return -EINVAL;
4677         }
4678
4679         map = (struct map_lookup *)em->bdev;
4680         offset = logical - em->start;
4681
4682         stripe_len = map->stripe_len;
4683         stripe_nr = offset;
4684         /*
4685          * stripe_nr counts the total number of stripes we have to stride
4686          * to get to this block
4687          */
4688         do_div(stripe_nr, stripe_len);
4689
4690         stripe_offset = stripe_nr * stripe_len;
4691         BUG_ON(offset < stripe_offset);
4692
4693         /* stripe_offset is the offset of this block in its stripe*/
4694         stripe_offset = offset - stripe_offset;
4695
4696         /* if we're here for raid56, we need to know the stripe aligned start */
4697         if (map->type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)) {
4698                 unsigned long full_stripe_len = stripe_len * nr_data_stripes(map);
4699                 raid56_full_stripe_start = offset;
4700
4701                 /* allow a write of a full stripe, but make sure we don't
4702                  * allow straddling of stripes
4703                  */
4704                 do_div(raid56_full_stripe_start, full_stripe_len);
4705                 raid56_full_stripe_start *= full_stripe_len;
4706         }
4707
4708         if (rw & REQ_DISCARD) {
4709                 /* we don't discard raid56 yet */
4710                 if (map->type &
4711                     (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)) {
4712                         ret = -EOPNOTSUPP;
4713                         goto out;
4714                 }
4715                 *length = min_t(u64, em->len - offset, *length);
4716         } else if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
4717                 u64 max_len;
4718                 /* For writes to RAID[56], allow a full stripeset across all disks.
4719                    For other RAID types and for RAID[56] reads, just allow a single
4720                    stripe (on a single disk). */
4721                 if (map->type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6) &&
4722                     (rw & REQ_WRITE)) {
4723                         max_len = stripe_len * nr_data_stripes(map) -
4724                                 (offset - raid56_full_stripe_start);
4725                 } else {
4726                         /* we limit the length of each bio to what fits in a stripe */
4727                         max_len = stripe_len - stripe_offset;
4728                 }
4729                 *length = min_t(u64, em->len - offset, max_len);
4730         } else {
4731                 *length = em->len - offset;
4732         }
4733
4734         /* This is for when we're called from btrfs_merge_bio_hook() and all
4735            it cares about is the length */
4736         if (!bbio_ret)
4737                 goto out;
4738
4739         btrfs_dev_replace_lock(dev_replace);
4740         dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
4741         if (!dev_replace_is_ongoing)
4742                 btrfs_dev_replace_unlock(dev_replace);
4743
4744         if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
4745             !(rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) &&
4746             dev_replace->tgtdev != NULL) {
4747                 /*
4748                  * in dev-replace case, for repair case (that's the only
4749                  * case where the mirror is selected explicitly when
4750                  * calling btrfs_map_block), blocks left of the left cursor
4751                  * can also be read from the target drive.
4752                  * For REQ_GET_READ_MIRRORS, the target drive is added as
4753                  * the last one to the array of stripes. For READ, it also
4754                  * needs to be supported using the same mirror number.
4755                  * If the requested block is not left of the left cursor,
4756                  * EIO is returned. This can happen because btrfs_num_copies()
4757                  * returns one more in the dev-replace case.
4758                  */
4759                 u64 tmp_length = *length;
4760                 struct btrfs_bio *tmp_bbio = NULL;
4761                 int tmp_num_stripes;
4762                 u64 srcdev_devid = dev_replace->srcdev->devid;
4763                 int index_srcdev = 0;
4764                 int found = 0;
4765                 u64 physical_of_found = 0;
4766
4767                 ret = __btrfs_map_block(fs_info, REQ_GET_READ_MIRRORS,
4768                              logical, &tmp_length, &tmp_bbio, 0, NULL);
4769                 if (ret) {
4770                         WARN_ON(tmp_bbio != NULL);
4771                         goto out;
4772                 }
4773
4774                 tmp_num_stripes = tmp_bbio->num_stripes;
4775                 if (mirror_num > tmp_num_stripes) {
4776                         /*
4777                          * REQ_GET_READ_MIRRORS does not contain this
4778                          * mirror, that means that the requested area
4779                          * is not left of the left cursor
4780                          */
4781                         ret = -EIO;
4782                         kfree(tmp_bbio);
4783                         goto out;
4784                 }
4785
4786                 /*
4787                  * process the rest of the function using the mirror_num
4788                  * of the source drive. Therefore look it up first.
4789                  * At the end, patch the device pointer to the one of the
4790                  * target drive.
4791                  */
4792                 for (i = 0; i < tmp_num_stripes; i++) {
4793                         if (tmp_bbio->stripes[i].dev->devid == srcdev_devid) {
4794                                 /*
4795                                  * In case of DUP, in order to keep it
4796                                  * simple, only add the mirror with the
4797                                  * lowest physical address
4798                                  */
4799                                 if (found &&
4800                                     physical_of_found <=
4801                                      tmp_bbio->stripes[i].physical)
4802                                         continue;
4803                                 index_srcdev = i;
4804                                 found = 1;
4805                                 physical_of_found =
4806                                         tmp_bbio->stripes[i].physical;
4807                         }
4808                 }
4809
4810                 if (found) {
4811                         mirror_num = index_srcdev + 1;
4812                         patch_the_first_stripe_for_dev_replace = 1;
4813                         physical_to_patch_in_first_stripe = physical_of_found;
4814                 } else {
4815                         WARN_ON(1);
4816                         ret = -EIO;
4817                         kfree(tmp_bbio);
4818                         goto out;
4819                 }
4820
4821                 kfree(tmp_bbio);
4822         } else if (mirror_num > map->num_stripes) {
4823                 mirror_num = 0;
4824         }
4825
4826         num_stripes = 1;
4827         stripe_index = 0;
4828         stripe_nr_orig = stripe_nr;
4829         stripe_nr_end = ALIGN(offset + *length, map->stripe_len);
4830         do_div(stripe_nr_end, map->stripe_len);
4831         stripe_end_offset = stripe_nr_end * map->stripe_len -
4832                             (offset + *length);
4833
4834         if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
4835                 if (rw & REQ_DISCARD)
4836                         num_stripes = min_t(u64, map->num_stripes,
4837                                             stripe_nr_end - stripe_nr_orig);
4838                 stripe_index = do_div(stripe_nr, map->num_stripes);
4839         } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
4840                 if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS))
4841                         num_stripes = map->num_stripes;
4842                 else if (mirror_num)
4843                         stripe_index = mirror_num - 1;
4844                 else {
4845                         stripe_index = find_live_mirror(fs_info, map, 0,
4846                                             map->num_stripes,
4847                                             current->pid % map->num_stripes,
4848                                             dev_replace_is_ongoing);
4849                         mirror_num = stripe_index + 1;
4850                 }
4851
4852         } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
4853                 if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) {
4854                         num_stripes = map->num_stripes;
4855                 } else if (mirror_num) {
4856                         stripe_index = mirror_num - 1;
4857                 } else {
4858                         mirror_num = 1;
4859                 }
4860
4861         } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
4862                 int factor = map->num_stripes / map->sub_stripes;
4863
4864                 stripe_index = do_div(stripe_nr, factor);
4865                 stripe_index *= map->sub_stripes;
4866
4867                 if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS))
4868                         num_stripes = map->sub_stripes;
4869                 else if (rw & REQ_DISCARD)
4870                         num_stripes = min_t(u64, map->sub_stripes *
4871                                             (stripe_nr_end - stripe_nr_orig),
4872                                             map->num_stripes);
4873                 else if (mirror_num)
4874                         stripe_index += mirror_num - 1;
4875                 else {
4876                         int old_stripe_index = stripe_index;
4877                         stripe_index = find_live_mirror(fs_info, map,
4878                                               stripe_index,
4879                                               map->sub_stripes, stripe_index +
4880                                               current->pid % map->sub_stripes,
4881                                               dev_replace_is_ongoing);
4882                         mirror_num = stripe_index - old_stripe_index + 1;
4883                 }
4884
4885         } else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
4886                                 BTRFS_BLOCK_GROUP_RAID6)) {
4887                 u64 tmp;
4888
4889                 if (bbio_ret && ((rw & REQ_WRITE) || mirror_num > 1)
4890                     && raid_map_ret) {
4891                         int i, rot;
4892
4893                         /* push stripe_nr back to the start of the full stripe */
4894                         stripe_nr = raid56_full_stripe_start;
4895                         do_div(stripe_nr, stripe_len);
4896
4897                         stripe_index = do_div(stripe_nr, nr_data_stripes(map));
4898
4899                         /* RAID[56] write or recovery. Return all stripes */
4900                         num_stripes = map->num_stripes;
4901                         max_errors = nr_parity_stripes(map);
4902
4903                         raid_map = kmalloc_array(num_stripes, sizeof(u64),
4904                                            GFP_NOFS);
4905                         if (!raid_map) {
4906                                 ret = -ENOMEM;
4907                                 goto out;
4908                         }
4909
4910                         /* Work out the disk rotation on this stripe-set */
4911                         tmp = stripe_nr;
4912                         rot = do_div(tmp, num_stripes);
4913
4914                         /* Fill in the logical address of each stripe */
4915                         tmp = stripe_nr * nr_data_stripes(map);
4916                         for (i = 0; i < nr_data_stripes(map); i++)
4917                                 raid_map[(i+rot) % num_stripes] =
4918                                         em->start + (tmp + i) * map->stripe_len;
4919
4920                         raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE;
4921                         if (map->type & BTRFS_BLOCK_GROUP_RAID6)
4922                                 raid_map[(i+rot+1) % num_stripes] =
4923                                         RAID6_Q_STRIPE;
4924
4925                         *length = map->stripe_len;
4926                         stripe_index = 0;
4927                         stripe_offset = 0;
4928                 } else {
4929                         /*
4930                          * Mirror #0 or #1 means the original data block.
4931                          * Mirror #2 is RAID5 parity block.
4932                          * Mirror #3 is RAID6 Q block.
4933                          */
4934                         stripe_index = do_div(stripe_nr, nr_data_stripes(map));
4935                         if (mirror_num > 1)
4936                                 stripe_index = nr_data_stripes(map) +
4937                                                 mirror_num - 2;
4938
4939                         /* We distribute the parity blocks across stripes */
4940                         tmp = stripe_nr + stripe_index;
4941                         stripe_index = do_div(tmp, map->num_stripes);
4942                 }
4943         } else {
4944                 /*
4945                  * after this do_div call, stripe_nr is the number of stripes
4946                  * on this device we have to walk to find the data, and
4947                  * stripe_index is the number of our device in the stripe array
4948                  */
4949                 stripe_index = do_div(stripe_nr, map->num_stripes);
4950                 mirror_num = stripe_index + 1;
4951         }
4952         BUG_ON(stripe_index >= map->num_stripes);
4953
4954         num_alloc_stripes = num_stripes;
4955         if (dev_replace_is_ongoing) {
4956                 if (rw & (REQ_WRITE | REQ_DISCARD))
4957                         num_alloc_stripes <<= 1;
4958                 if (rw & REQ_GET_READ_MIRRORS)
4959                         num_alloc_stripes++;
4960         }
4961         bbio = kzalloc(btrfs_bio_size(num_alloc_stripes), GFP_NOFS);
4962         if (!bbio) {
4963                 kfree(raid_map);
4964                 ret = -ENOMEM;
4965                 goto out;
4966         }
4967         atomic_set(&bbio->error, 0);
4968
4969         if (rw & REQ_DISCARD) {
4970                 int factor = 0;
4971                 int sub_stripes = 0;
4972                 u64 stripes_per_dev = 0;
4973                 u32 remaining_stripes = 0;
4974                 u32 last_stripe = 0;
4975
4976                 if (map->type &
4977                     (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID10)) {
4978                         if (map->type & BTRFS_BLOCK_GROUP_RAID0)
4979                                 sub_stripes = 1;
4980                         else
4981                                 sub_stripes = map->sub_stripes;
4982
4983                         factor = map->num_stripes / sub_stripes;
4984                         stripes_per_dev = div_u64_rem(stripe_nr_end -
4985                                                       stripe_nr_orig,
4986                                                       factor,
4987                                                       &remaining_stripes);
4988                         div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
4989                         last_stripe *= sub_stripes;
4990                 }
4991
4992                 for (i = 0; i < num_stripes; i++) {
4993                         bbio->stripes[i].physical =
4994                                 map->stripes[stripe_index].physical +
4995                                 stripe_offset + stripe_nr * map->stripe_len;
4996                         bbio->stripes[i].dev = map->stripes[stripe_index].dev;
4997
4998                         if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
4999                                          BTRFS_BLOCK_GROUP_RAID10)) {
5000                                 bbio->stripes[i].length = stripes_per_dev *
5001                                                           map->stripe_len;
5002
5003                                 if (i / sub_stripes < remaining_stripes)
5004                                         bbio->stripes[i].length +=
5005                                                 map->stripe_len;
5006
5007                                 /*
5008                                  * Special for the first stripe and
5009                                  * the last stripe:
5010                                  *
5011                                  * |-------|...|-------|
5012                                  *     |----------|
5013                                  *    off     end_off
5014                                  */
5015                                 if (i < sub_stripes)
5016                                         bbio->stripes[i].length -=
5017                                                 stripe_offset;
5018
5019                                 if (stripe_index >= last_stripe &&
5020                                     stripe_index <= (last_stripe +
5021                                                      sub_stripes - 1))
5022                                         bbio->stripes[i].length -=
5023                                                 stripe_end_offset;
5024
5025                                 if (i == sub_stripes - 1)
5026                                         stripe_offset = 0;
5027                         } else
5028                                 bbio->stripes[i].length = *length;
5029
5030                         stripe_index++;
5031                         if (stripe_index == map->num_stripes) {
5032                                 /* This could only happen for RAID0/10 */
5033                                 stripe_index = 0;
5034                                 stripe_nr++;
5035                         }
5036                 }
5037         } else {
5038                 for (i = 0; i < num_stripes; i++) {
5039                         bbio->stripes[i].physical =
5040                                 map->stripes[stripe_index].physical +
5041                                 stripe_offset +
5042                                 stripe_nr * map->stripe_len;
5043                         bbio->stripes[i].dev =
5044                                 map->stripes[stripe_index].dev;
5045                         stripe_index++;
5046                 }
5047         }
5048
5049         if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS)) {
5050                 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
5051                                  BTRFS_BLOCK_GROUP_RAID10 |
5052                                  BTRFS_BLOCK_GROUP_RAID5 |
5053                                  BTRFS_BLOCK_GROUP_DUP)) {
5054                         max_errors = 1;
5055                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID6) {
5056                         max_errors = 2;
5057                 }
5058         }
5059
5060         if (dev_replace_is_ongoing && (rw & (REQ_WRITE | REQ_DISCARD)) &&
5061             dev_replace->tgtdev != NULL) {
5062                 int index_where_to_add;
5063                 u64 srcdev_devid = dev_replace->srcdev->devid;
5064
5065                 /*
5066                  * duplicate the write operations while the dev replace
5067                  * procedure is running. Since the copying of the old disk
5068                  * to the new disk takes place at run time while the
5069                  * filesystem is mounted writable, the regular write
5070                  * operations to the old disk have to be duplicated to go
5071                  * to the new disk as well.
5072                  * Note that device->missing is handled by the caller, and
5073                  * that the write to the old disk is already set up in the
5074                  * stripes array.
5075                  */
5076                 index_where_to_add = num_stripes;
5077                 for (i = 0; i < num_stripes; i++) {
5078                         if (bbio->stripes[i].dev->devid == srcdev_devid) {
5079                                 /* write to new disk, too */
5080                                 struct btrfs_bio_stripe *new =
5081                                         bbio->stripes + index_where_to_add;
5082                                 struct btrfs_bio_stripe *old =
5083                                         bbio->stripes + i;
5084
5085                                 new->physical = old->physical;
5086                                 new->length = old->length;
5087                                 new->dev = dev_replace->tgtdev;
5088                                 index_where_to_add++;
5089                                 max_errors++;
5090                         }
5091                 }
5092                 num_stripes = index_where_to_add;
5093         } else if (dev_replace_is_ongoing && (rw & REQ_GET_READ_MIRRORS) &&
5094                    dev_replace->tgtdev != NULL) {
5095                 u64 srcdev_devid = dev_replace->srcdev->devid;
5096                 int index_srcdev = 0;
5097                 int found = 0;
5098                 u64 physical_of_found = 0;
5099
5100                 /*
5101                  * During the dev-replace procedure, the target drive can
5102                  * also be used to read data in case it is needed to repair
5103                  * a corrupt block elsewhere. This is possible if the
5104                  * requested area is left of the left cursor. In this area,
5105                  * the target drive is a full copy of the source drive.
5106                  */
5107                 for (i = 0; i < num_stripes; i++) {
5108                         if (bbio->stripes[i].dev->devid == srcdev_devid) {
5109                                 /*
5110                                  * In case of DUP, in order to keep it
5111                                  * simple, only add the mirror with the
5112                                  * lowest physical address
5113                                  */
5114                                 if (found &&
5115                                     physical_of_found <=
5116                                      bbio->stripes[i].physical)
5117                                         continue;
5118                                 index_srcdev = i;
5119                                 found = 1;
5120                                 physical_of_found = bbio->stripes[i].physical;
5121                         }
5122                 }
5123                 if (found) {
5124                         u64 length = map->stripe_len;
5125
5126                         if (physical_of_found + length <=
5127                             dev_replace->cursor_left) {
5128                                 struct btrfs_bio_stripe *tgtdev_stripe =
5129                                         bbio->stripes + num_stripes;
5130
5131                                 tgtdev_stripe->physical = physical_of_found;
5132                                 tgtdev_stripe->length =
5133                                         bbio->stripes[index_srcdev].length;
5134                                 tgtdev_stripe->dev = dev_replace->tgtdev;
5135
5136                                 num_stripes++;
5137                         }
5138                 }
5139         }
5140
5141         *bbio_ret = bbio;
5142         bbio->num_stripes = num_stripes;
5143         bbio->max_errors = max_errors;
5144         bbio->mirror_num = mirror_num;
5145
5146         /*
5147          * this is the case that REQ_READ && dev_replace_is_ongoing &&
5148          * mirror_num == num_stripes + 1 && dev_replace target drive is
5149          * available as a mirror
5150          */
5151         if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) {
5152                 WARN_ON(num_stripes > 1);
5153                 bbio->stripes[0].dev = dev_replace->tgtdev;
5154                 bbio->stripes[0].physical = physical_to_patch_in_first_stripe;
5155                 bbio->mirror_num = map->num_stripes + 1;
5156         }
5157         if (raid_map) {
5158                 sort_parity_stripes(bbio, raid_map);
5159                 *raid_map_ret = raid_map;
5160         }
5161 out:
5162         if (dev_replace_is_ongoing)
5163                 btrfs_dev_replace_unlock(dev_replace);
5164         free_extent_map(em);
5165         return ret;
5166 }
5167
5168 int btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
5169                       u64 logical, u64 *length,
5170                       struct btrfs_bio **bbio_ret, int mirror_num)
5171 {
5172         return __btrfs_map_block(fs_info, rw, logical, length, bbio_ret,
5173                                  mirror_num, NULL);
5174 }
5175
5176 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
5177                      u64 chunk_start, u64 physical, u64 devid,
5178                      u64 **logical, int *naddrs, int *stripe_len)
5179 {
5180         struct extent_map_tree *em_tree = &map_tree->map_tree;
5181         struct extent_map *em;
5182         struct map_lookup *map;
5183         u64 *buf;
5184         u64 bytenr;
5185         u64 length;
5186         u64 stripe_nr;
5187         u64 rmap_len;
5188         int i, j, nr = 0;
5189
5190         read_lock(&em_tree->lock);
5191         em = lookup_extent_mapping(em_tree, chunk_start, 1);
5192         read_unlock(&em_tree->lock);
5193
5194         if (!em) {
5195                 printk(KERN_ERR "BTRFS: couldn't find em for chunk %Lu\n",
5196                        chunk_start);
5197                 return -EIO;
5198         }
5199
5200         if (em->start != chunk_start) {
5201                 printk(KERN_ERR "BTRFS: bad chunk start, em=%Lu, wanted=%Lu\n",
5202                        em->start, chunk_start);
5203                 free_extent_map(em);
5204                 return -EIO;
5205         }
5206         map = (struct map_lookup *)em->bdev;
5207
5208         length = em->len;
5209         rmap_len = map->stripe_len;
5210
5211         if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5212                 do_div(length, map->num_stripes / map->sub_stripes);
5213         else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5214                 do_div(length, map->num_stripes);
5215         else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
5216                               BTRFS_BLOCK_GROUP_RAID6)) {
5217                 do_div(length, nr_data_stripes(map));
5218                 rmap_len = map->stripe_len * nr_data_stripes(map);
5219         }
5220
5221         buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
5222         BUG_ON(!buf); /* -ENOMEM */
5223
5224         for (i = 0; i < map->num_stripes; i++) {
5225                 if (devid && map->stripes[i].dev->devid != devid)
5226                         continue;
5227                 if (map->stripes[i].physical > physical ||
5228                     map->stripes[i].physical + length <= physical)
5229                         continue;
5230
5231                 stripe_nr = physical - map->stripes[i].physical;
5232                 do_div(stripe_nr, map->stripe_len);
5233
5234                 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5235                         stripe_nr = stripe_nr * map->num_stripes + i;
5236                         do_div(stripe_nr, map->sub_stripes);
5237                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5238                         stripe_nr = stripe_nr * map->num_stripes + i;
5239                 } /* else if RAID[56], multiply by nr_data_stripes().
5240                    * Alternatively, just use rmap_len below instead of
5241                    * map->stripe_len */
5242
5243                 bytenr = chunk_start + stripe_nr * rmap_len;
5244                 WARN_ON(nr >= map->num_stripes);
5245                 for (j = 0; j < nr; j++) {
5246                         if (buf[j] == bytenr)
5247                                 break;
5248                 }
5249                 if (j == nr) {
5250                         WARN_ON(nr >= map->num_stripes);
5251                         buf[nr++] = bytenr;
5252                 }
5253         }
5254
5255         *logical = buf;
5256         *naddrs = nr;
5257         *stripe_len = rmap_len;
5258
5259         free_extent_map(em);
5260         return 0;
5261 }
5262
5263 static void btrfs_end_bio(struct bio *bio, int err)
5264 {
5265         struct btrfs_bio *bbio = bio->bi_private;
5266         int is_orig_bio = 0;
5267
5268         if (err) {
5269                 atomic_inc(&bbio->error);
5270                 if (err == -EIO || err == -EREMOTEIO) {
5271                         unsigned int stripe_index =
5272                                 btrfs_io_bio(bio)->stripe_index;
5273                         struct btrfs_device *dev;
5274
5275                         BUG_ON(stripe_index >= bbio->num_stripes);
5276                         dev = bbio->stripes[stripe_index].dev;
5277                         if (dev->bdev) {
5278                                 if (bio->bi_rw & WRITE)
5279                                         btrfs_dev_stat_inc(dev,
5280                                                 BTRFS_DEV_STAT_WRITE_ERRS);
5281                                 else
5282                                         btrfs_dev_stat_inc(dev,
5283                                                 BTRFS_DEV_STAT_READ_ERRS);
5284                                 if ((bio->bi_rw & WRITE_FLUSH) == WRITE_FLUSH)
5285                                         btrfs_dev_stat_inc(dev,
5286                                                 BTRFS_DEV_STAT_FLUSH_ERRS);
5287                                 btrfs_dev_stat_print_on_error(dev);
5288                         }
5289                 }
5290         }
5291
5292         if (bio == bbio->orig_bio)
5293                 is_orig_bio = 1;
5294
5295         if (atomic_dec_and_test(&bbio->stripes_pending)) {
5296                 if (!is_orig_bio) {
5297                         bio_put(bio);
5298                         bio = bbio->orig_bio;
5299                 }
5300
5301                 /*
5302                  * We have original bio now. So increment bi_remaining to
5303                  * account for it in endio
5304                  */
5305                 atomic_inc(&bio->bi_remaining);
5306
5307                 bio->bi_private = bbio->private;
5308                 bio->bi_end_io = bbio->end_io;
5309                 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
5310                 /* only send an error to the higher layers if it is
5311                  * beyond the tolerance of the btrfs bio
5312                  */
5313                 if (atomic_read(&bbio->error) > bbio->max_errors) {
5314                         err = -EIO;
5315                 } else {
5316                         /*
5317                          * this bio is actually up to date, we didn't
5318                          * go over the max number of errors
5319                          */
5320                         set_bit(BIO_UPTODATE, &bio->bi_flags);
5321                         err = 0;
5322                 }
5323                 kfree(bbio);
5324
5325                 bio_endio(bio, err);
5326         } else if (!is_orig_bio) {
5327                 bio_put(bio);
5328         }
5329 }
5330
5331 struct async_sched {
5332         struct bio *bio;
5333         int rw;
5334         struct btrfs_fs_info *info;
5335         struct btrfs_work work;
5336 };
5337
5338 /*
5339  * see run_scheduled_bios for a description of why bios are collected for
5340  * async submit.
5341  *
5342  * This will add one bio to the pending list for a device and make sure
5343  * the work struct is scheduled.
5344  */
5345 static noinline void btrfs_schedule_bio(struct btrfs_root *root,
5346                                         struct btrfs_device *device,
5347                                         int rw, struct bio *bio)
5348 {
5349         int should_queue = 1;
5350         struct btrfs_pending_bios *pending_bios;
5351
5352         if (device->missing || !device->bdev) {
5353                 bio_endio(bio, -EIO);
5354                 return;
5355         }
5356
5357         /* don't bother with additional async steps for reads, right now */
5358         if (!(rw & REQ_WRITE)) {
5359                 bio_get(bio);
5360                 btrfsic_submit_bio(rw, bio);
5361                 bio_put(bio);
5362                 return;
5363         }
5364
5365         /*
5366          * nr_async_bios allows us to reliably return congestion to the
5367          * higher layers.  Otherwise, the async bio makes it appear we have
5368          * made progress against dirty pages when we've really just put it
5369          * on a queue for later
5370          */
5371         atomic_inc(&root->fs_info->nr_async_bios);
5372         WARN_ON(bio->bi_next);
5373         bio->bi_next = NULL;
5374         bio->bi_rw |= rw;
5375
5376         spin_lock(&device->io_lock);
5377         if (bio->bi_rw & REQ_SYNC)
5378                 pending_bios = &device->pending_sync_bios;
5379         else
5380                 pending_bios = &device->pending_bios;
5381
5382         if (pending_bios->tail)
5383                 pending_bios->tail->bi_next = bio;
5384
5385         pending_bios->tail = bio;
5386         if (!pending_bios->head)
5387                 pending_bios->head = bio;
5388         if (device->running_pending)
5389                 should_queue = 0;
5390
5391         spin_unlock(&device->io_lock);
5392
5393         if (should_queue)
5394                 btrfs_queue_worker(&root->fs_info->submit_workers,
5395                                    &device->work);
5396 }
5397
5398 static int bio_size_ok(struct block_device *bdev, struct bio *bio,
5399                        sector_t sector)
5400 {
5401         struct bio_vec *prev;
5402         struct request_queue *q = bdev_get_queue(bdev);
5403         unsigned int max_sectors = queue_max_sectors(q);
5404         struct bvec_merge_data bvm = {
5405                 .bi_bdev = bdev,
5406                 .bi_sector = sector,
5407                 .bi_rw = bio->bi_rw,
5408         };
5409
5410         if (WARN_ON(bio->bi_vcnt == 0))
5411                 return 1;
5412
5413         prev = &bio->bi_io_vec[bio->bi_vcnt - 1];
5414         if (bio_sectors(bio) > max_sectors)
5415                 return 0;
5416
5417         if (!q->merge_bvec_fn)
5418                 return 1;
5419
5420         bvm.bi_size = bio->bi_iter.bi_size - prev->bv_len;
5421         if (q->merge_bvec_fn(q, &bvm, prev) < prev->bv_len)
5422                 return 0;
5423         return 1;
5424 }
5425
5426 static void submit_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio,
5427                               struct bio *bio, u64 physical, int dev_nr,
5428                               int rw, int async)
5429 {
5430         struct btrfs_device *dev = bbio->stripes[dev_nr].dev;
5431
5432         bio->bi_private = bbio;
5433         btrfs_io_bio(bio)->stripe_index = dev_nr;
5434         bio->bi_end_io = btrfs_end_bio;
5435         bio->bi_iter.bi_sector = physical >> 9;
5436 #ifdef DEBUG
5437         {
5438                 struct rcu_string *name;
5439
5440                 rcu_read_lock();
5441                 name = rcu_dereference(dev->name);
5442                 pr_debug("btrfs_map_bio: rw %d, sector=%llu, dev=%lu "
5443                          "(%s id %llu), size=%u\n", rw,
5444                          (u64)bio->bi_sector, (u_long)dev->bdev->bd_dev,
5445                          name->str, dev->devid, bio->bi_size);
5446                 rcu_read_unlock();
5447         }
5448 #endif
5449         bio->bi_bdev = dev->bdev;
5450         if (async)
5451                 btrfs_schedule_bio(root, dev, rw, bio);
5452         else
5453                 btrfsic_submit_bio(rw, bio);
5454 }
5455
5456 static int breakup_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio,
5457                               struct bio *first_bio, struct btrfs_device *dev,
5458                               int dev_nr, int rw, int async)
5459 {
5460         struct bio_vec *bvec = first_bio->bi_io_vec;
5461         struct bio *bio;
5462         int nr_vecs = bio_get_nr_vecs(dev->bdev);
5463         u64 physical = bbio->stripes[dev_nr].physical;
5464
5465 again:
5466         bio = btrfs_bio_alloc(dev->bdev, physical >> 9, nr_vecs, GFP_NOFS);
5467         if (!bio)
5468                 return -ENOMEM;
5469
5470         while (bvec <= (first_bio->bi_io_vec + first_bio->bi_vcnt - 1)) {
5471                 if (bio_add_page(bio, bvec->bv_page, bvec->bv_len,
5472                                  bvec->bv_offset) < bvec->bv_len) {
5473                         u64 len = bio->bi_iter.bi_size;
5474
5475                         atomic_inc(&bbio->stripes_pending);
5476                         submit_stripe_bio(root, bbio, bio, physical, dev_nr,
5477                                           rw, async);
5478                         physical += len;
5479                         goto again;
5480                 }
5481                 bvec++;
5482         }
5483
5484         submit_stripe_bio(root, bbio, bio, physical, dev_nr, rw, async);
5485         return 0;
5486 }
5487
5488 static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical)
5489 {
5490         atomic_inc(&bbio->error);
5491         if (atomic_dec_and_test(&bbio->stripes_pending)) {
5492                 bio->bi_private = bbio->private;
5493                 bio->bi_end_io = bbio->end_io;
5494                 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
5495                 bio->bi_iter.bi_sector = logical >> 9;
5496                 kfree(bbio);
5497                 bio_endio(bio, -EIO);
5498         }
5499 }
5500
5501 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
5502                   int mirror_num, int async_submit)
5503 {
5504         struct btrfs_device *dev;
5505         struct bio *first_bio = bio;
5506         u64 logical = (u64)bio->bi_iter.bi_sector << 9;
5507         u64 length = 0;
5508         u64 map_length;
5509         u64 *raid_map = NULL;
5510         int ret;
5511         int dev_nr = 0;
5512         int total_devs = 1;
5513         struct btrfs_bio *bbio = NULL;
5514
5515         length = bio->bi_iter.bi_size;
5516         map_length = length;
5517
5518         ret = __btrfs_map_block(root->fs_info, rw, logical, &map_length, &bbio,
5519                               mirror_num, &raid_map);
5520         if (ret) /* -ENOMEM */
5521                 return ret;
5522
5523         total_devs = bbio->num_stripes;
5524         bbio->orig_bio = first_bio;
5525         bbio->private = first_bio->bi_private;
5526         bbio->end_io = first_bio->bi_end_io;
5527         atomic_set(&bbio->stripes_pending, bbio->num_stripes);
5528
5529         if (raid_map) {
5530                 /* In this case, map_length has been set to the length of
5531                    a single stripe; not the whole write */
5532                 if (rw & WRITE) {
5533                         return raid56_parity_write(root, bio, bbio,
5534                                                    raid_map, map_length);
5535                 } else {
5536                         return raid56_parity_recover(root, bio, bbio,
5537                                                      raid_map, map_length,
5538                                                      mirror_num);
5539                 }
5540         }
5541
5542         if (map_length < length) {
5543                 btrfs_crit(root->fs_info, "mapping failed logical %llu bio len %llu len %llu",
5544                         logical, length, map_length);
5545                 BUG();
5546         }
5547
5548         while (dev_nr < total_devs) {
5549                 dev = bbio->stripes[dev_nr].dev;
5550                 if (!dev || !dev->bdev || (rw & WRITE && !dev->writeable)) {
5551                         bbio_error(bbio, first_bio, logical);
5552                         dev_nr++;
5553                         continue;
5554                 }
5555
5556                 /*
5557                  * Check and see if we're ok with this bio based on it's size
5558                  * and offset with the given device.
5559                  */
5560                 if (!bio_size_ok(dev->bdev, first_bio,
5561                                  bbio->stripes[dev_nr].physical >> 9)) {
5562                         ret = breakup_stripe_bio(root, bbio, first_bio, dev,
5563                                                  dev_nr, rw, async_submit);
5564                         BUG_ON(ret);
5565                         dev_nr++;
5566                         continue;
5567                 }
5568
5569                 if (dev_nr < total_devs - 1) {
5570                         bio = btrfs_bio_clone(first_bio, GFP_NOFS);
5571                         BUG_ON(!bio); /* -ENOMEM */
5572                 } else {
5573                         bio = first_bio;
5574                 }
5575
5576                 submit_stripe_bio(root, bbio, bio,
5577                                   bbio->stripes[dev_nr].physical, dev_nr, rw,
5578                                   async_submit);
5579                 dev_nr++;
5580         }
5581         return 0;
5582 }
5583
5584 struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid,
5585                                        u8 *uuid, u8 *fsid)
5586 {
5587         struct btrfs_device *device;
5588         struct btrfs_fs_devices *cur_devices;
5589
5590         cur_devices = fs_info->fs_devices;
5591         while (cur_devices) {
5592                 if (!fsid ||
5593                     !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
5594                         device = __find_device(&cur_devices->devices,
5595                                                devid, uuid);
5596                         if (device)
5597                                 return device;
5598                 }
5599                 cur_devices = cur_devices->seed;
5600         }
5601         return NULL;
5602 }
5603
5604 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
5605                                             u64 devid, u8 *dev_uuid)
5606 {
5607         struct btrfs_device *device;
5608         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
5609
5610         device = btrfs_alloc_device(NULL, &devid, dev_uuid);
5611         if (IS_ERR(device))
5612                 return NULL;
5613
5614         list_add(&device->dev_list, &fs_devices->devices);
5615         device->fs_devices = fs_devices;
5616         fs_devices->num_devices++;
5617
5618         device->missing = 1;
5619         fs_devices->missing_devices++;
5620
5621         return device;
5622 }
5623
5624 /**
5625  * btrfs_alloc_device - allocate struct btrfs_device
5626  * @fs_info:    used only for generating a new devid, can be NULL if
5627  *              devid is provided (i.e. @devid != NULL).
5628  * @devid:      a pointer to devid for this device.  If NULL a new devid
5629  *              is generated.
5630  * @uuid:       a pointer to UUID for this device.  If NULL a new UUID
5631  *              is generated.
5632  *
5633  * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
5634  * on error.  Returned struct is not linked onto any lists and can be
5635  * destroyed with kfree() right away.
5636  */
5637 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
5638                                         const u64 *devid,
5639                                         const u8 *uuid)
5640 {
5641         struct btrfs_device *dev;
5642         u64 tmp;
5643
5644         if (WARN_ON(!devid && !fs_info))
5645                 return ERR_PTR(-EINVAL);
5646
5647         dev = __alloc_device();
5648         if (IS_ERR(dev))
5649                 return dev;
5650
5651         if (devid)
5652                 tmp = *devid;
5653         else {
5654                 int ret;
5655
5656                 ret = find_next_devid(fs_info, &tmp);
5657                 if (ret) {
5658                         kfree(dev);
5659                         return ERR_PTR(ret);
5660                 }
5661         }
5662         dev->devid = tmp;
5663
5664         if (uuid)
5665                 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE);
5666         else
5667                 generate_random_uuid(dev->uuid);
5668
5669         dev->work.func = pending_bios_fn;
5670
5671         return dev;
5672 }
5673
5674 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
5675                           struct extent_buffer *leaf,
5676                           struct btrfs_chunk *chunk)
5677 {
5678         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
5679         struct map_lookup *map;
5680         struct extent_map *em;
5681         u64 logical;
5682         u64 length;
5683         u64 devid;
5684         u8 uuid[BTRFS_UUID_SIZE];
5685         int num_stripes;
5686         int ret;
5687         int i;
5688
5689         logical = key->offset;
5690         length = btrfs_chunk_length(leaf, chunk);
5691
5692         read_lock(&map_tree->map_tree.lock);
5693         em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
5694         read_unlock(&map_tree->map_tree.lock);
5695
5696         /* already mapped? */
5697         if (em && em->start <= logical && em->start + em->len > logical) {
5698                 free_extent_map(em);
5699                 return 0;
5700         } else if (em) {
5701                 free_extent_map(em);
5702         }
5703
5704         em = alloc_extent_map();
5705         if (!em)
5706                 return -ENOMEM;
5707         num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
5708         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
5709         if (!map) {
5710                 free_extent_map(em);
5711                 return -ENOMEM;
5712         }
5713
5714         em->bdev = (struct block_device *)map;
5715         em->start = logical;
5716         em->len = length;
5717         em->orig_start = 0;
5718         em->block_start = 0;
5719         em->block_len = em->len;
5720
5721         map->num_stripes = num_stripes;
5722         map->io_width = btrfs_chunk_io_width(leaf, chunk);
5723         map->io_align = btrfs_chunk_io_align(leaf, chunk);
5724         map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
5725         map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
5726         map->type = btrfs_chunk_type(leaf, chunk);
5727         map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
5728         for (i = 0; i < num_stripes; i++) {
5729                 map->stripes[i].physical =
5730                         btrfs_stripe_offset_nr(leaf, chunk, i);
5731                 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
5732                 read_extent_buffer(leaf, uuid, (unsigned long)
5733                                    btrfs_stripe_dev_uuid_nr(chunk, i),
5734                                    BTRFS_UUID_SIZE);
5735                 map->stripes[i].dev = btrfs_find_device(root->fs_info, devid,
5736                                                         uuid, NULL);
5737                 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
5738                         kfree(map);
5739                         free_extent_map(em);
5740                         return -EIO;
5741                 }
5742                 if (!map->stripes[i].dev) {
5743                         map->stripes[i].dev =
5744                                 add_missing_dev(root, devid, uuid);
5745                         if (!map->stripes[i].dev) {
5746                                 kfree(map);
5747                                 free_extent_map(em);
5748                                 return -EIO;
5749                         }
5750                 }
5751                 map->stripes[i].dev->in_fs_metadata = 1;
5752         }
5753
5754         write_lock(&map_tree->map_tree.lock);
5755         ret = add_extent_mapping(&map_tree->map_tree, em, 0);
5756         write_unlock(&map_tree->map_tree.lock);
5757         BUG_ON(ret); /* Tree corruption */
5758         free_extent_map(em);
5759
5760         return 0;
5761 }
5762
5763 static void fill_device_from_item(struct extent_buffer *leaf,
5764                                  struct btrfs_dev_item *dev_item,
5765                                  struct btrfs_device *device)
5766 {
5767         unsigned long ptr;
5768
5769         device->devid = btrfs_device_id(leaf, dev_item);
5770         device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
5771         device->total_bytes = device->disk_total_bytes;
5772         device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
5773         device->type = btrfs_device_type(leaf, dev_item);
5774         device->io_align = btrfs_device_io_align(leaf, dev_item);
5775         device->io_width = btrfs_device_io_width(leaf, dev_item);
5776         device->sector_size = btrfs_device_sector_size(leaf, dev_item);
5777         WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
5778         device->is_tgtdev_for_dev_replace = 0;
5779
5780         ptr = btrfs_device_uuid(dev_item);
5781         read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
5782 }
5783
5784 static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
5785 {
5786         struct btrfs_fs_devices *fs_devices;
5787         int ret;
5788
5789         BUG_ON(!mutex_is_locked(&uuid_mutex));
5790
5791         fs_devices = root->fs_info->fs_devices->seed;
5792         while (fs_devices) {
5793                 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
5794                         ret = 0;
5795                         goto out;
5796                 }
5797                 fs_devices = fs_devices->seed;
5798         }
5799
5800         fs_devices = find_fsid(fsid);
5801         if (!fs_devices) {
5802                 ret = -ENOENT;
5803                 goto out;
5804         }
5805
5806         fs_devices = clone_fs_devices(fs_devices);
5807         if (IS_ERR(fs_devices)) {
5808                 ret = PTR_ERR(fs_devices);
5809                 goto out;
5810         }
5811
5812         ret = __btrfs_open_devices(fs_devices, FMODE_READ,
5813                                    root->fs_info->bdev_holder);
5814         if (ret) {
5815                 free_fs_devices(fs_devices);
5816                 goto out;
5817         }
5818
5819         if (!fs_devices->seeding) {
5820                 __btrfs_close_devices(fs_devices);
5821                 free_fs_devices(fs_devices);
5822                 ret = -EINVAL;
5823                 goto out;
5824         }
5825
5826         fs_devices->seed = root->fs_info->fs_devices->seed;
5827         root->fs_info->fs_devices->seed = fs_devices;
5828 out:
5829         return ret;
5830 }
5831
5832 static int read_one_dev(struct btrfs_root *root,
5833                         struct extent_buffer *leaf,
5834                         struct btrfs_dev_item *dev_item)
5835 {
5836         struct btrfs_device *device;
5837         u64 devid;
5838         int ret;
5839         u8 fs_uuid[BTRFS_UUID_SIZE];
5840         u8 dev_uuid[BTRFS_UUID_SIZE];
5841
5842         devid = btrfs_device_id(leaf, dev_item);
5843         read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
5844                            BTRFS_UUID_SIZE);
5845         read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
5846                            BTRFS_UUID_SIZE);
5847
5848         if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
5849                 ret = open_seed_devices(root, fs_uuid);
5850                 if (ret && !btrfs_test_opt(root, DEGRADED))
5851                         return ret;
5852         }
5853
5854         device = btrfs_find_device(root->fs_info, devid, dev_uuid, fs_uuid);
5855         if (!device || !device->bdev) {
5856                 if (!btrfs_test_opt(root, DEGRADED))
5857                         return -EIO;
5858
5859                 if (!device) {
5860                         btrfs_warn(root->fs_info, "devid %llu missing", devid);
5861                         device = add_missing_dev(root, devid, dev_uuid);
5862                         if (!device)
5863                                 return -ENOMEM;
5864                 } else if (!device->missing) {
5865                         /*
5866                          * this happens when a device that was properly setup
5867                          * in the device info lists suddenly goes bad.
5868                          * device->bdev is NULL, and so we have to set
5869                          * device->missing to one here
5870                          */
5871                         root->fs_info->fs_devices->missing_devices++;
5872                         device->missing = 1;
5873                 }
5874         }
5875
5876         if (device->fs_devices != root->fs_info->fs_devices) {
5877                 BUG_ON(device->writeable);
5878                 if (device->generation !=
5879                     btrfs_device_generation(leaf, dev_item))
5880                         return -EINVAL;
5881         }
5882
5883         fill_device_from_item(leaf, dev_item, device);
5884         device->in_fs_metadata = 1;
5885         if (device->writeable && !device->is_tgtdev_for_dev_replace) {
5886                 device->fs_devices->total_rw_bytes += device->total_bytes;
5887                 spin_lock(&root->fs_info->free_chunk_lock);
5888                 root->fs_info->free_chunk_space += device->total_bytes -
5889                         device->bytes_used;
5890                 spin_unlock(&root->fs_info->free_chunk_lock);
5891         }
5892         ret = 0;
5893         return ret;
5894 }
5895
5896 int btrfs_read_sys_array(struct btrfs_root *root)
5897 {
5898         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
5899         struct extent_buffer *sb;
5900         struct btrfs_disk_key *disk_key;
5901         struct btrfs_chunk *chunk;
5902         u8 *ptr;
5903         unsigned long sb_ptr;
5904         int ret = 0;
5905         u32 num_stripes;
5906         u32 array_size;
5907         u32 len = 0;
5908         u32 cur;
5909         struct btrfs_key key;
5910
5911         sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
5912                                           BTRFS_SUPER_INFO_SIZE);
5913         if (!sb)
5914                 return -ENOMEM;
5915         btrfs_set_buffer_uptodate(sb);
5916         btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
5917         /*
5918          * The sb extent buffer is artifical and just used to read the system array.
5919          * btrfs_set_buffer_uptodate() call does not properly mark all it's
5920          * pages up-to-date when the page is larger: extent does not cover the
5921          * whole page and consequently check_page_uptodate does not find all
5922          * the page's extents up-to-date (the hole beyond sb),
5923          * write_extent_buffer then triggers a WARN_ON.
5924          *
5925          * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
5926          * but sb spans only this function. Add an explicit SetPageUptodate call
5927          * to silence the warning eg. on PowerPC 64.
5928          */
5929         if (PAGE_CACHE_SIZE > BTRFS_SUPER_INFO_SIZE)
5930                 SetPageUptodate(sb->pages[0]);
5931
5932         write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
5933         array_size = btrfs_super_sys_array_size(super_copy);
5934
5935         ptr = super_copy->sys_chunk_array;
5936         sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
5937         cur = 0;
5938
5939         while (cur < array_size) {
5940                 disk_key = (struct btrfs_disk_key *)ptr;
5941                 btrfs_disk_key_to_cpu(&key, disk_key);
5942
5943                 len = sizeof(*disk_key); ptr += len;
5944                 sb_ptr += len;
5945                 cur += len;
5946
5947                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
5948                         chunk = (struct btrfs_chunk *)sb_ptr;
5949                         ret = read_one_chunk(root, &key, sb, chunk);
5950                         if (ret)
5951                                 break;
5952                         num_stripes = btrfs_chunk_num_stripes(sb, chunk);
5953                         len = btrfs_chunk_item_size(num_stripes);
5954                 } else {
5955                         ret = -EIO;
5956                         break;
5957                 }
5958                 ptr += len;
5959                 sb_ptr += len;
5960                 cur += len;
5961         }
5962         free_extent_buffer(sb);
5963         return ret;
5964 }
5965
5966 int btrfs_read_chunk_tree(struct btrfs_root *root)
5967 {
5968         struct btrfs_path *path;
5969         struct extent_buffer *leaf;
5970         struct btrfs_key key;
5971         struct btrfs_key found_key;
5972         int ret;
5973         int slot;
5974
5975         root = root->fs_info->chunk_root;
5976
5977         path = btrfs_alloc_path();
5978         if (!path)
5979                 return -ENOMEM;
5980
5981         mutex_lock(&uuid_mutex);
5982         lock_chunks(root);
5983
5984         /*
5985          * Read all device items, and then all the chunk items. All
5986          * device items are found before any chunk item (their object id
5987          * is smaller than the lowest possible object id for a chunk
5988          * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
5989          */
5990         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
5991         key.offset = 0;
5992         key.type = 0;
5993         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5994         if (ret < 0)
5995                 goto error;
5996         while (1) {
5997                 leaf = path->nodes[0];
5998                 slot = path->slots[0];
5999                 if (slot >= btrfs_header_nritems(leaf)) {
6000                         ret = btrfs_next_leaf(root, path);
6001                         if (ret == 0)
6002                                 continue;
6003                         if (ret < 0)
6004                                 goto error;
6005                         break;
6006                 }
6007                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6008                 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
6009                         struct btrfs_dev_item *dev_item;
6010                         dev_item = btrfs_item_ptr(leaf, slot,
6011                                                   struct btrfs_dev_item);
6012                         ret = read_one_dev(root, leaf, dev_item);
6013                         if (ret)
6014                                 goto error;
6015                 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
6016                         struct btrfs_chunk *chunk;
6017                         chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
6018                         ret = read_one_chunk(root, &found_key, leaf, chunk);
6019                         if (ret)
6020                                 goto error;
6021                 }
6022                 path->slots[0]++;
6023         }
6024         ret = 0;
6025 error:
6026         unlock_chunks(root);
6027         mutex_unlock(&uuid_mutex);
6028
6029         btrfs_free_path(path);
6030         return ret;
6031 }
6032
6033 void btrfs_init_devices_late(struct btrfs_fs_info *fs_info)
6034 {
6035         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6036         struct btrfs_device *device;
6037
6038         mutex_lock(&fs_devices->device_list_mutex);
6039         list_for_each_entry(device, &fs_devices->devices, dev_list)
6040                 device->dev_root = fs_info->dev_root;
6041         mutex_unlock(&fs_devices->device_list_mutex);
6042 }
6043
6044 static void __btrfs_reset_dev_stats(struct btrfs_device *dev)
6045 {
6046         int i;
6047
6048         for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6049                 btrfs_dev_stat_reset(dev, i);
6050 }
6051
6052 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
6053 {
6054         struct btrfs_key key;
6055         struct btrfs_key found_key;
6056         struct btrfs_root *dev_root = fs_info->dev_root;
6057         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6058         struct extent_buffer *eb;
6059         int slot;
6060         int ret = 0;
6061         struct btrfs_device *device;
6062         struct btrfs_path *path = NULL;
6063         int i;
6064
6065         path = btrfs_alloc_path();
6066         if (!path) {
6067                 ret = -ENOMEM;
6068                 goto out;
6069         }
6070
6071         mutex_lock(&fs_devices->device_list_mutex);
6072         list_for_each_entry(device, &fs_devices->devices, dev_list) {
6073                 int item_size;
6074                 struct btrfs_dev_stats_item *ptr;
6075
6076                 key.objectid = 0;
6077                 key.type = BTRFS_DEV_STATS_KEY;
6078                 key.offset = device->devid;
6079                 ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
6080                 if (ret) {
6081                         __btrfs_reset_dev_stats(device);
6082                         device->dev_stats_valid = 1;
6083                         btrfs_release_path(path);
6084                         continue;
6085                 }
6086                 slot = path->slots[0];
6087                 eb = path->nodes[0];
6088                 btrfs_item_key_to_cpu(eb, &found_key, slot);
6089                 item_size = btrfs_item_size_nr(eb, slot);
6090
6091                 ptr = btrfs_item_ptr(eb, slot,
6092                                      struct btrfs_dev_stats_item);
6093
6094                 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
6095                         if (item_size >= (1 + i) * sizeof(__le64))
6096                                 btrfs_dev_stat_set(device, i,
6097                                         btrfs_dev_stats_value(eb, ptr, i));
6098                         else
6099                                 btrfs_dev_stat_reset(device, i);
6100                 }
6101
6102                 device->dev_stats_valid = 1;
6103                 btrfs_dev_stat_print_on_load(device);
6104                 btrfs_release_path(path);
6105         }
6106         mutex_unlock(&fs_devices->device_list_mutex);
6107
6108 out:
6109         btrfs_free_path(path);
6110         return ret < 0 ? ret : 0;
6111 }
6112
6113 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
6114                                 struct btrfs_root *dev_root,
6115                                 struct btrfs_device *device)
6116 {
6117         struct btrfs_path *path;
6118         struct btrfs_key key;
6119         struct extent_buffer *eb;
6120         struct btrfs_dev_stats_item *ptr;
6121         int ret;
6122         int i;
6123
6124         key.objectid = 0;
6125         key.type = BTRFS_DEV_STATS_KEY;
6126         key.offset = device->devid;
6127
6128         path = btrfs_alloc_path();
6129         BUG_ON(!path);
6130         ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
6131         if (ret < 0) {
6132                 printk_in_rcu(KERN_WARNING "BTRFS: "
6133                         "error %d while searching for dev_stats item for device %s!\n",
6134                               ret, rcu_str_deref(device->name));
6135                 goto out;
6136         }
6137
6138         if (ret == 0 &&
6139             btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
6140                 /* need to delete old one and insert a new one */
6141                 ret = btrfs_del_item(trans, dev_root, path);
6142                 if (ret != 0) {
6143                         printk_in_rcu(KERN_WARNING "BTRFS: "
6144                                 "delete too small dev_stats item for device %s failed %d!\n",
6145                                       rcu_str_deref(device->name), ret);
6146                         goto out;
6147                 }
6148                 ret = 1;
6149         }
6150
6151         if (ret == 1) {
6152                 /* need to insert a new item */
6153                 btrfs_release_path(path);
6154                 ret = btrfs_insert_empty_item(trans, dev_root, path,
6155                                               &key, sizeof(*ptr));
6156                 if (ret < 0) {
6157                         printk_in_rcu(KERN_WARNING "BTRFS: "
6158                                           "insert dev_stats item for device %s failed %d!\n",
6159                                       rcu_str_deref(device->name), ret);
6160                         goto out;
6161                 }
6162         }
6163
6164         eb = path->nodes[0];
6165         ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
6166         for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6167                 btrfs_set_dev_stats_value(eb, ptr, i,
6168                                           btrfs_dev_stat_read(device, i));
6169         btrfs_mark_buffer_dirty(eb);
6170
6171 out:
6172         btrfs_free_path(path);
6173         return ret;
6174 }
6175
6176 /*
6177  * called from commit_transaction. Writes all changed device stats to disk.
6178  */
6179 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans,
6180                         struct btrfs_fs_info *fs_info)
6181 {
6182         struct btrfs_root *dev_root = fs_info->dev_root;
6183         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6184         struct btrfs_device *device;
6185         int ret = 0;
6186
6187         mutex_lock(&fs_devices->device_list_mutex);
6188         list_for_each_entry(device, &fs_devices->devices, dev_list) {
6189                 if (!device->dev_stats_valid || !device->dev_stats_dirty)
6190                         continue;
6191
6192                 ret = update_dev_stat_item(trans, dev_root, device);
6193                 if (!ret)
6194                         device->dev_stats_dirty = 0;
6195         }
6196         mutex_unlock(&fs_devices->device_list_mutex);
6197
6198         return ret;
6199 }
6200
6201 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
6202 {
6203         btrfs_dev_stat_inc(dev, index);
6204         btrfs_dev_stat_print_on_error(dev);
6205 }
6206
6207 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
6208 {
6209         if (!dev->dev_stats_valid)
6210                 return;
6211         printk_ratelimited_in_rcu(KERN_ERR "BTRFS: "
6212                            "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
6213                            rcu_str_deref(dev->name),
6214                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
6215                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
6216                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
6217                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
6218                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
6219 }
6220
6221 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
6222 {
6223         int i;
6224
6225         for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6226                 if (btrfs_dev_stat_read(dev, i) != 0)
6227                         break;
6228         if (i == BTRFS_DEV_STAT_VALUES_MAX)
6229                 return; /* all values == 0, suppress message */
6230
6231         printk_in_rcu(KERN_INFO "BTRFS: "
6232                    "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
6233                rcu_str_deref(dev->name),
6234                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
6235                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
6236                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
6237                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
6238                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
6239 }
6240
6241 int btrfs_get_dev_stats(struct btrfs_root *root,
6242                         struct btrfs_ioctl_get_dev_stats *stats)
6243 {
6244         struct btrfs_device *dev;
6245         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
6246         int i;
6247
6248         mutex_lock(&fs_devices->device_list_mutex);
6249         dev = btrfs_find_device(root->fs_info, stats->devid, NULL, NULL);
6250         mutex_unlock(&fs_devices->device_list_mutex);
6251
6252         if (!dev) {
6253                 btrfs_warn(root->fs_info, "get dev_stats failed, device not found");
6254                 return -ENODEV;
6255         } else if (!dev->dev_stats_valid) {
6256                 btrfs_warn(root->fs_info, "get dev_stats failed, not yet valid");
6257                 return -ENODEV;
6258         } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
6259                 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
6260                         if (stats->nr_items > i)
6261                                 stats->values[i] =
6262                                         btrfs_dev_stat_read_and_reset(dev, i);
6263                         else
6264                                 btrfs_dev_stat_reset(dev, i);
6265                 }
6266         } else {
6267                 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6268                         if (stats->nr_items > i)
6269                                 stats->values[i] = btrfs_dev_stat_read(dev, i);
6270         }
6271         if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
6272                 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
6273         return 0;
6274 }
6275
6276 int btrfs_scratch_superblock(struct btrfs_device *device)
6277 {
6278         struct buffer_head *bh;
6279         struct btrfs_super_block *disk_super;
6280
6281         bh = btrfs_read_dev_super(device->bdev);
6282         if (!bh)
6283                 return -EINVAL;
6284         disk_super = (struct btrfs_super_block *)bh->b_data;
6285
6286         memset(&disk_super->magic, 0, sizeof(disk_super->magic));
6287         set_buffer_dirty(bh);
6288         sync_dirty_buffer(bh);
6289         brelse(bh);
6290
6291         return 0;
6292 }