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_queue_work(fs_info->submit_workers,
 419                                          &device->work);
 420                         goto done;
 421                 }
 422                 /* unplug every 64 requests just for good measure */
 423                 if (batch_run % 64 == 0) {
 424                         blk_finish_plug(&plug);
 425                         blk_start_plug(&plug);
 426                         sync_pending = 0;
 427                 }
 428         }
 429
 430         cond_resched();
 431         if (again)
 432                 goto loop;
 433
 434         spin_lock(&device->io_lock);
 435         if (device->pending_bios.head || device->pending_sync_bios.head)
 436                 goto loop_lock;
 437         spin_unlock(&device->io_lock);
 438
 439 done:
 440         blk_finish_plug(&plug);
 441 }
 442
 443 static void pending_bios_fn(struct btrfs_work *work)
 444 {
 445         struct btrfs_device *device;
 446
 447         device = container_of(work, struct btrfs_device, work);
 448         run_scheduled_bios(device);
 449 }
 450
 451 /*
 452  * Add new device to list of registered devices
 453  *
 454  * Returns:
 455  * 1   - first time device is seen
 456  * 0   - device already known
 457  * < 0 - error
 458  */
 459 static noinline int device_list_add(const char *path,
 460                            struct btrfs_super_block *disk_super,
 461                            u64 devid, struct btrfs_fs_devices **fs_devices_ret)
 462 {
 463         struct btrfs_device *device;
 464         struct btrfs_fs_devices *fs_devices;
 465         struct rcu_string *name;
 466         int ret = 0;
 467         u64 found_transid = btrfs_super_generation(disk_super);
 468
 469         fs_devices = find_fsid(disk_super->fsid);
 470         if (!fs_devices) {
 471                 fs_devices = alloc_fs_devices(disk_super->fsid);
 472                 if (IS_ERR(fs_devices))
 473                         return PTR_ERR(fs_devices);
 474
 475                 list_add(&fs_devices->list, &fs_uuids);
 476                 fs_devices->latest_devid = devid;
 477                 fs_devices->latest_trans = found_transid;
 478
 479                 device = NULL;
 480         } else {
 481                 device = __find_device(&fs_devices->devices, devid,
 482                                        disk_super->dev_item.uuid);
 483         }
 484         if (!device) {
 485                 if (fs_devices->opened)
 486                         return -EBUSY;
 487
 488                 device = btrfs_alloc_device(NULL, &devid,
 489                                             disk_super->dev_item.uuid);
 490                 if (IS_ERR(device)) {
 491                         /* we can safely leave the fs_devices entry around */
 492                         return PTR_ERR(device);
 493                 }
 494
 495                 name = rcu_string_strdup(path, GFP_NOFS);
 496                 if (!name) {
 497                         kfree(device);
 498                         return -ENOMEM;
 499                 }
 500                 rcu_assign_pointer(device->name, name);
 501
 502                 mutex_lock(&fs_devices->device_list_mutex);
 503                 list_add_rcu(&device->dev_list, &fs_devices->devices);
 504                 fs_devices->num_devices++;
 505                 mutex_unlock(&fs_devices->device_list_mutex);
 506
 507                 ret = 1;
 508                 device->fs_devices = fs_devices;
 509         } else if (!device->name || strcmp(device->name->str, path)) {
 510                 name = rcu_string_strdup(path, GFP_NOFS);
 511                 if (!name)
 512                         return -ENOMEM;
 513                 rcu_string_free(device->name);
 514                 rcu_assign_pointer(device->name, name);
 515                 if (device->missing) {
 516                         fs_devices->missing_devices--;
 517                         device->missing = 0;
 518                 }
 519         }
 520
 521         if (found_transid > fs_devices->latest_trans) {
 522                 fs_devices->latest_devid = devid;
 523                 fs_devices->latest_trans = found_transid;
 524         }
 525         *fs_devices_ret = fs_devices;
 526
 527         return ret;
 528 }
 529
 530 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
 531 {
 532         struct btrfs_fs_devices *fs_devices;
 533         struct btrfs_device *device;
 534         struct btrfs_device *orig_dev;
 535
 536         fs_devices = alloc_fs_devices(orig->fsid);
 537         if (IS_ERR(fs_devices))
 538                 return fs_devices;
 539
 540         fs_devices->latest_devid = orig->latest_devid;
 541         fs_devices->latest_trans = orig->latest_trans;
 542         fs_devices->total_devices = orig->total_devices;
 543
 544         /* We have held the volume lock, it is safe to get the devices. */
 545         list_for_each_entry(orig_dev, &orig->devices, dev_list) {
 546                 struct rcu_string *name;
 547
 548                 device = btrfs_alloc_device(NULL, &orig_dev->devid,
 549                                             orig_dev->uuid);
 550                 if (IS_ERR(device))
 551                         goto error;
 552
 553                 /*
 554                  * This is ok to do without rcu read locked because we hold the
 555                  * uuid mutex so nothing we touch in here is going to disappear.
 556                  */
 557                 name = rcu_string_strdup(orig_dev->name->str, GFP_NOFS);
 558                 if (!name) {
 559                         kfree(device);
 560                         goto error;
 561                 }
 562                 rcu_assign_pointer(device->name, name);
 563
 564                 list_add(&device->dev_list, &fs_devices->devices);
 565                 device->fs_devices = fs_devices;
 566                 fs_devices->num_devices++;
 567         }
 568         return fs_devices;
 569 error:
 570         free_fs_devices(fs_devices);
 571         return ERR_PTR(-ENOMEM);
 572 }
 573
 574 void btrfs_close_extra_devices(struct btrfs_fs_info *fs_info,
 575                                struct btrfs_fs_devices *fs_devices, int step)
 576 {
 577         struct btrfs_device *device, *next;
 578
 579         struct block_device *latest_bdev = NULL;
 580         u64 latest_devid = 0;
 581         u64 latest_transid = 0;
 582
 583         mutex_lock(&uuid_mutex);
 584 again:
 585         /* This is the initialized path, it is safe to release the devices. */
 586         list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
 587                 if (device->in_fs_metadata) {
 588                         if (!device->is_tgtdev_for_dev_replace &&
 589                             (!latest_transid ||
 590                              device->generation > latest_transid)) {
 591                                 latest_devid = device->devid;
 592                                 latest_transid = device->generation;
 593                                 latest_bdev = device->bdev;
 594                         }
 595                         continue;
 596                 }
 597
 598                 if (device->devid == BTRFS_DEV_REPLACE_DEVID) {
 599                         /*
 600                          * In the first step, keep the device which has
 601                          * the correct fsid and the devid that is used
 602                          * for the dev_replace procedure.
 603                          * In the second step, the dev_replace state is
 604                          * read from the device tree and it is known
 605                          * whether the procedure is really active or
 606                          * not, which means whether this device is
 607                          * used or whether it should be removed.
 608                          */
 609                         if (step == 0 || device->is_tgtdev_for_dev_replace) {
 610                                 continue;
 611                         }
 612                 }
 613                 if (device->bdev) {
 614                         blkdev_put(device->bdev, device->mode);
 615                         device->bdev = NULL;
 616                         fs_devices->open_devices--;
 617                 }
 618                 if (device->writeable) {
 619                         list_del_init(&device->dev_alloc_list);
 620                         device->writeable = 0;
 621                         if (!device->is_tgtdev_for_dev_replace)
 622                                 fs_devices->rw_devices--;
 623                 }
 624                 list_del_init(&device->dev_list);
 625                 fs_devices->num_devices--;
 626                 rcu_string_free(device->name);
 627                 kfree(device);
 628         }
 629
 630         if (fs_devices->seed) {
 631                 fs_devices = fs_devices->seed;
 632                 goto again;
 633         }
 634
 635         fs_devices->latest_bdev = latest_bdev;
 636         fs_devices->latest_devid = latest_devid;
 637         fs_devices->latest_trans = latest_transid;
 638
 639         mutex_unlock(&uuid_mutex);
 640 }
 641
 642 static void __free_device(struct work_struct *work)
 643 {
 644         struct btrfs_device *device;
 645
 646         device = container_of(work, struct btrfs_device, rcu_work);
 647
 648         if (device->bdev)
 649                 blkdev_put(device->bdev, device->mode);
 650
 651         rcu_string_free(device->name);
 652         kfree(device);
 653 }
 654
 655 static void free_device(struct rcu_head *head)
 656 {
 657         struct btrfs_device *device;
 658
 659         device = container_of(head, struct btrfs_device, rcu);
 660
 661         INIT_WORK(&device->rcu_work, __free_device);
 662         schedule_work(&device->rcu_work);
 663 }
 664
 665 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
 666 {
 667         struct btrfs_device *device;
 668
 669         if (--fs_devices->opened > 0)
 670                 return 0;
 671
 672         mutex_lock(&fs_devices->device_list_mutex);
 673         list_for_each_entry(device, &fs_devices->devices, dev_list) {
 674                 struct btrfs_device *new_device;
 675                 struct rcu_string *name;
 676
 677                 if (device->bdev)
 678                         fs_devices->open_devices--;
 679
 680                 if (device->writeable &&
 681                     device->devid != BTRFS_DEV_REPLACE_DEVID) {
 682                         list_del_init(&device->dev_alloc_list);
 683                         fs_devices->rw_devices--;
 684                 }
 685
 686                 if (device->can_discard)
 687                         fs_devices->num_can_discard--;
 688                 if (device->missing)
 689                         fs_devices->missing_devices--;
 690
 691                 new_device = btrfs_alloc_device(NULL, &device->devid,
 692                                                 device->uuid);
 693                 BUG_ON(IS_ERR(new_device)); /* -ENOMEM */
 694
 695                 /* Safe because we are under uuid_mutex */
 696                 if (device->name) {
 697                         name = rcu_string_strdup(device->name->str, GFP_NOFS);
 698                         BUG_ON(!name); /* -ENOMEM */
 699                         rcu_assign_pointer(new_device->name, name);
 700                 }
 701
 702                 list_replace_rcu(&device->dev_list, &new_device->dev_list);
 703                 new_device->fs_devices = device->fs_devices;
 704
 705                 call_rcu(&device->rcu, free_device);
 706         }
 707         mutex_unlock(&fs_devices->device_list_mutex);
 708
 709         WARN_ON(fs_devices->open_devices);
 710         WARN_ON(fs_devices->rw_devices);
 711         fs_devices->opened = 0;
 712         fs_devices->seeding = 0;
 713
 714         return 0;
 715 }
 716
 717 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
 718 {
 719         struct btrfs_fs_devices *seed_devices = NULL;
 720         int ret;
 721
 722         mutex_lock(&uuid_mutex);
 723         ret = __btrfs_close_devices(fs_devices);
 724         if (!fs_devices->opened) {
 725                 seed_devices = fs_devices->seed;
 726                 fs_devices->seed = NULL;
 727         }
 728         mutex_unlock(&uuid_mutex);
 729
 730         while (seed_devices) {
 731                 fs_devices = seed_devices;
 732                 seed_devices = fs_devices->seed;
 733                 __btrfs_close_devices(fs_devices);
 734                 free_fs_devices(fs_devices);
 735         }
 736         /*
 737          * Wait for rcu kworkers under __btrfs_close_devices
 738          * to finish all blkdev_puts so device is really
 739          * free when umount is done.
 740          */
 741         rcu_barrier();
 742         return ret;
 743 }
 744
 745 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
 746                                 fmode_t flags, void *holder)
 747 {
 748         struct request_queue *q;
 749         struct block_device *bdev;
 750         struct list_head *head = &fs_devices->devices;
 751         struct btrfs_device *device;
 752         struct block_device *latest_bdev = NULL;
 753         struct buffer_head *bh;
 754         struct btrfs_super_block *disk_super;
 755         u64 latest_devid = 0;
 756         u64 latest_transid = 0;
 757         u64 devid;
 758         int seeding = 1;
 759         int ret = 0;
 760
 761         flags |= FMODE_EXCL;
 762
 763         list_for_each_entry(device, head, dev_list) {
 764                 if (device->bdev)
 765                         continue;
 766                 if (!device->name)
 767                         continue;
 768
 769                 /* Just open everything we can; ignore failures here */
 770                 if (btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1,
 771                                             &bdev, &bh))
 772                         continue;
 773
 774                 disk_super = (struct btrfs_super_block *)bh->b_data;
 775                 devid = btrfs_stack_device_id(&disk_super->dev_item);
 776                 if (devid != device->devid)
 777                         goto error_brelse;
 778
 779                 if (memcmp(device->uuid, disk_super->dev_item.uuid,
 780                            BTRFS_UUID_SIZE))
 781                         goto error_brelse;
 782
 783                 device->generation = btrfs_super_generation(disk_super);
 784                 if (!latest_transid || device->generation > latest_transid) {
 785                         latest_devid = devid;
 786                         latest_transid = device->generation;
 787                         latest_bdev = bdev;
 788                 }
 789
 790                 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
 791                         device->writeable = 0;
 792                 } else {
 793                         device->writeable = !bdev_read_only(bdev);
 794                         seeding = 0;
 795                 }
 796
 797                 q = bdev_get_queue(bdev);
 798                 if (blk_queue_discard(q)) {
 799                         device->can_discard = 1;
 800                         fs_devices->num_can_discard++;
 801                 }
 802
 803                 device->bdev = bdev;
 804                 device->in_fs_metadata = 0;
 805                 device->mode = flags;
 806
 807                 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
 808                         fs_devices->rotating = 1;
 809
 810                 fs_devices->open_devices++;
 811                 if (device->writeable &&
 812                     device->devid != BTRFS_DEV_REPLACE_DEVID) {
 813                         fs_devices->rw_devices++;
 814                         list_add(&device->dev_alloc_list,
 815                                  &fs_devices->alloc_list);
 816                 }
 817                 brelse(bh);
 818                 continue;
 819
 820 error_brelse:
 821                 brelse(bh);
 822                 blkdev_put(bdev, flags);
 823                 continue;
 824         }
 825         if (fs_devices->open_devices == 0) {
 826                 ret = -EINVAL;
 827                 goto out;
 828         }
 829         fs_devices->seeding = seeding;
 830         fs_devices->opened = 1;
 831         fs_devices->latest_bdev = latest_bdev;
 832         fs_devices->latest_devid = latest_devid;
 833         fs_devices->latest_trans = latest_transid;
 834         fs_devices->total_rw_bytes = 0;
 835 out:
 836         return ret;
 837 }
 838
 839 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
 840                        fmode_t flags, void *holder)
 841 {
 842         int ret;
 843
 844         mutex_lock(&uuid_mutex);
 845         if (fs_devices->opened) {
 846                 fs_devices->opened++;
 847                 ret = 0;
 848         } else {
 849                 ret = __btrfs_open_devices(fs_devices, flags, holder);
 850         }
 851         mutex_unlock(&uuid_mutex);
 852         return ret;
 853 }
 854
 855 /*
 856  * Look for a btrfs signature on a device. This may be called out of the mount path
 857  * and we are not allowed to call set_blocksize during the scan. The superblock
 858  * is read via pagecache
 859  */
 860 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
 861                           struct btrfs_fs_devices **fs_devices_ret)
 862 {
 863         struct btrfs_super_block *disk_super;
 864         struct block_device *bdev;
 865         struct page *page;
 866         void *p;
 867         int ret = -EINVAL;
 868         u64 devid;
 869         u64 transid;
 870         u64 total_devices;
 871         u64 bytenr;
 872         pgoff_t index;
 873
 874         /*
 875          * we would like to check all the supers, but that would make
 876          * a btrfs mount succeed after a mkfs from a different FS.
 877          * So, we need to add a special mount option to scan for
 878          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
 879          */
 880         bytenr = btrfs_sb_offset(0);
 881         flags |= FMODE_EXCL;
 882         mutex_lock(&uuid_mutex);
 883
 884         bdev = blkdev_get_by_path(path, flags, holder);
 885
 886         if (IS_ERR(bdev)) {
 887                 ret = PTR_ERR(bdev);
 888                 goto error;
 889         }
 890
 891         /* make sure our super fits in the device */
 892         if (bytenr + PAGE_CACHE_SIZE >= i_size_read(bdev->bd_inode))
 893                 goto error_bdev_put;
 894
 895         /* make sure our super fits in the page */
 896         if (sizeof(*disk_super) > PAGE_CACHE_SIZE)
 897                 goto error_bdev_put;
 898
 899         /* make sure our super doesn't straddle pages on disk */
 900         index = bytenr >> PAGE_CACHE_SHIFT;
 901         if ((bytenr + sizeof(*disk_super) - 1) >> PAGE_CACHE_SHIFT != index)
 902                 goto error_bdev_put;
 903
 904         /* pull in the page with our super */
 905         page = read_cache_page_gfp(bdev->bd_inode->i_mapping,
 906                                    index, GFP_NOFS);
 907
 908         if (IS_ERR_OR_NULL(page))
 909                 goto error_bdev_put;
 910
 911         p = kmap(page);
 912
 913         /* align our pointer to the offset of the super block */
 914         disk_super = p + (bytenr & ~PAGE_CACHE_MASK);
 915
 916         if (btrfs_super_bytenr(disk_super) != bytenr ||
 917             btrfs_super_magic(disk_super) != BTRFS_MAGIC)
 918                 goto error_unmap;
 919
 920         devid = btrfs_stack_device_id(&disk_super->dev_item);
 921         transid = btrfs_super_generation(disk_super);
 922         total_devices = btrfs_super_num_devices(disk_super);
 923
 924         ret = device_list_add(path, disk_super, devid, fs_devices_ret);
 925         if (ret > 0) {
 926                 if (disk_super->label[0]) {
 927                         if (disk_super->label[BTRFS_LABEL_SIZE - 1])
 928                                 disk_super->label[BTRFS_LABEL_SIZE - 1] = '\0';
 929                         printk(KERN_INFO "BTRFS: device label %s ", disk_super->label);
 930                 } else {
 931                         printk(KERN_INFO "BTRFS: device fsid %pU ", disk_super->fsid);
 932                 }
 933
 934                 printk(KERN_CONT "devid %llu transid %llu %s\n", devid, transid, path);
 935                 ret = 0;
 936         }
 937         if (!ret && fs_devices_ret)
 938                 (*fs_devices_ret)->total_devices = total_devices;
 939
 940 error_unmap:
 941         kunmap(page);
 942         page_cache_release(page);
 943
 944 error_bdev_put:
 945         blkdev_put(bdev, flags);
 946 error:
 947         mutex_unlock(&uuid_mutex);
 948         return ret;
 949 }
 950
 951 /* helper to account the used device space in the range */
 952 int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
 953                                    u64 end, u64 *length)
 954 {
 955         struct btrfs_key key;
 956         struct btrfs_root *root = device->dev_root;
 957         struct btrfs_dev_extent *dev_extent;
 958         struct btrfs_path *path;
 959         u64 extent_end;
 960         int ret;
 961         int slot;
 962         struct extent_buffer *l;
 963
 964         *length = 0;
 965
 966         if (start >= device->total_bytes || device->is_tgtdev_for_dev_replace)
 967                 return 0;
 968
 969         path = btrfs_alloc_path();
 970         if (!path)
 971                 return -ENOMEM;
 972         path->reada = 2;
 973
 974         key.objectid = device->devid;
 975         key.offset = start;
 976         key.type = BTRFS_DEV_EXTENT_KEY;
 977
 978         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 979         if (ret < 0)
 980                 goto out;
 981         if (ret > 0) {
 982                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
 983                 if (ret < 0)
 984                         goto out;
 985         }
 986
 987         while (1) {
 988                 l = path->nodes[0];
 989                 slot = path->slots[0];
 990                 if (slot >= btrfs_header_nritems(l)) {
 991                         ret = btrfs_next_leaf(root, path);
 992                         if (ret == 0)
 993                                 continue;
 994                         if (ret < 0)
 995                                 goto out;
 996
 997                         break;
 998                 }
 999                 btrfs_item_key_to_cpu(l, &key, slot);
1000
1001                 if (key.objectid < device->devid)
1002                         goto next;
1003
1004                 if (key.objectid > device->devid)
1005                         break;
1006
1007                 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
1008                         goto next;
1009
1010                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1011                 extent_end = key.offset + btrfs_dev_extent_length(l,
1012                                                                   dev_extent);
1013                 if (key.offset <= start && extent_end > end) {
1014                         *length = end - start + 1;
1015                         break;
1016                 } else if (key.offset <= start && extent_end > start)
1017                         *length += extent_end - start;
1018                 else if (key.offset > start && extent_end <= end)
1019                         *length += extent_end - key.offset;
1020                 else if (key.offset > start && key.offset <= end) {
1021                         *length += end - key.offset + 1;
1022                         break;
1023                 } else if (key.offset > end)
1024                         break;
1025
1026 next:
1027                 path->slots[0]++;
1028         }
1029         ret = 0;
1030 out:
1031         btrfs_free_path(path);
1032         return ret;
1033 }
1034
1035 static int contains_pending_extent(struct btrfs_trans_handle *trans,
1036                                    struct btrfs_device *device,
1037                                    u64 *start, u64 len)
1038 {
1039         struct extent_map *em;
1040         int ret = 0;
1041
1042         list_for_each_entry(em, &trans->transaction->pending_chunks, list) {
1043                 struct map_lookup *map;
1044                 int i;
1045
1046                 map = (struct map_lookup *)em->bdev;
1047                 for (i = 0; i < map->num_stripes; i++) {
1048                         if (map->stripes[i].dev != device)
1049                                 continue;
1050                         if (map->stripes[i].physical >= *start + len ||
1051                             map->stripes[i].physical + em->orig_block_len <=
1052                             *start)
1053                                 continue;
1054                         *start = map->stripes[i].physical +
1055                                 em->orig_block_len;
1056                         ret = 1;
1057                 }
1058         }
1059
1060         return ret;
1061 }
1062
1063
1064 /*
1065  * find_free_dev_extent - find free space in the specified device
1066  * @device:     the device which we search the free space in
1067  * @num_bytes:  the size of the free space that we need
1068  * @start:      store the start of the free space.
1069  * @len:        the size of the free space. that we find, or the size of the max
1070  *              free space if we don't find suitable free space
1071  *
1072  * this uses a pretty simple search, the expectation is that it is
1073  * called very infrequently and that a given device has a small number
1074  * of extents
1075  *
1076  * @start is used to store the start of the free space if we find. But if we
1077  * don't find suitable free space, it will be used to store the start position
1078  * of the max free space.
1079  *
1080  * @len is used to store the size of the free space that we find.
1081  * But if we don't find suitable free space, it is used to store the size of
1082  * the max free space.
1083  */
1084 int find_free_dev_extent(struct btrfs_trans_handle *trans,
1085                          struct btrfs_device *device, u64 num_bytes,
1086                          u64 *start, u64 *len)
1087 {
1088         struct btrfs_key key;
1089         struct btrfs_root *root = device->dev_root;
1090         struct btrfs_dev_extent *dev_extent;
1091         struct btrfs_path *path;
1092         u64 hole_size;
1093         u64 max_hole_start;
1094         u64 max_hole_size;
1095         u64 extent_end;
1096         u64 search_start;
1097         u64 search_end = device->total_bytes;
1098         int ret;
1099         int slot;
1100         struct extent_buffer *l;
1101
1102         /* FIXME use last free of some kind */
1103
1104         /* we don't want to overwrite the superblock on the drive,
1105          * so we make sure to start at an offset of at least 1MB
1106          */
1107         search_start = max(root->fs_info->alloc_start, 1024ull * 1024);
1108
1109         path = btrfs_alloc_path();
1110         if (!path)
1111                 return -ENOMEM;
1112 again:
1113         max_hole_start = search_start;
1114         max_hole_size = 0;
1115         hole_size = 0;
1116
1117         if (search_start >= search_end || device->is_tgtdev_for_dev_replace) {
1118                 ret = -ENOSPC;
1119                 goto out;
1120         }
1121
1122         path->reada = 2;
1123         path->search_commit_root = 1;
1124         path->skip_locking = 1;
1125
1126         key.objectid = device->devid;
1127         key.offset = search_start;
1128         key.type = BTRFS_DEV_EXTENT_KEY;
1129
1130         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1131         if (ret < 0)
1132                 goto out;
1133         if (ret > 0) {
1134                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1135                 if (ret < 0)
1136                         goto out;
1137         }
1138
1139         while (1) {
1140                 l = path->nodes[0];
1141                 slot = path->slots[0];
1142                 if (slot >= btrfs_header_nritems(l)) {
1143                         ret = btrfs_next_leaf(root, path);
1144                         if (ret == 0)
1145                                 continue;
1146                         if (ret < 0)
1147                                 goto out;
1148
1149                         break;
1150                 }
1151                 btrfs_item_key_to_cpu(l, &key, slot);
1152
1153                 if (key.objectid < device->devid)
1154                         goto next;
1155
1156                 if (key.objectid > device->devid)
1157                         break;
1158
1159                 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
1160                         goto next;
1161
1162                 if (key.offset > search_start) {
1163                         hole_size = key.offset - search_start;
1164
1165                         /*
1166                          * Have to check before we set max_hole_start, otherwise
1167                          * we could end up sending back this offset anyway.
1168                          */
1169                         if (contains_pending_extent(trans, device,
1170                                                     &search_start,
1171                                                     hole_size))
1172                                 hole_size = 0;
1173
1174                         if (hole_size > max_hole_size) {
1175                                 max_hole_start = search_start;
1176                                 max_hole_size = hole_size;
1177                         }
1178
1179                         /*
1180                          * If this free space is greater than which we need,
1181                          * it must be the max free space that we have found
1182                          * until now, so max_hole_start must point to the start
1183                          * of this free space and the length of this free space
1184                          * is stored in max_hole_size. Thus, we return
1185                          * max_hole_start and max_hole_size and go back to the
1186                          * caller.
1187                          */
1188                         if (hole_size >= num_bytes) {
1189                                 ret = 0;
1190                                 goto out;
1191                         }
1192                 }
1193
1194                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1195                 extent_end = key.offset + btrfs_dev_extent_length(l,
1196                                                                   dev_extent);
1197                 if (extent_end > search_start)
1198                         search_start = extent_end;
1199 next:
1200                 path->slots[0]++;
1201                 cond_resched();
1202         }
1203
1204         /*
1205          * At this point, search_start should be the end of
1206          * allocated dev extents, and when shrinking the device,
1207          * search_end may be smaller than search_start.
1208          */
1209         if (search_end > search_start)
1210                 hole_size = search_end - search_start;
1211
1212         if (hole_size > max_hole_size) {
1213                 max_hole_start = search_start;
1214                 max_hole_size = hole_size;
1215         }
1216
1217         if (contains_pending_extent(trans, device, &search_start, hole_size)) {
1218                 btrfs_release_path(path);
1219                 goto again;
1220         }
1221
1222         /* See above. */
1223         if (hole_size < num_bytes)
1224                 ret = -ENOSPC;
1225         else
1226                 ret = 0;
1227
1228 out:
1229         btrfs_free_path(path);
1230         *start = max_hole_start;
1231         if (len)
1232                 *len = max_hole_size;
1233         return ret;
1234 }
1235
1236 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1237                           struct btrfs_device *device,
1238                           u64 start)
1239 {
1240         int ret;
1241         struct btrfs_path *path;
1242         struct btrfs_root *root = device->dev_root;
1243         struct btrfs_key key;
1244         struct btrfs_key found_key;
1245         struct extent_buffer *leaf = NULL;
1246         struct btrfs_dev_extent *extent = NULL;
1247
1248         path = btrfs_alloc_path();
1249         if (!path)
1250                 return -ENOMEM;
1251
1252         key.objectid = device->devid;
1253         key.offset = start;
1254         key.type = BTRFS_DEV_EXTENT_KEY;
1255 again:
1256         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1257         if (ret > 0) {
1258                 ret = btrfs_previous_item(root, path, key.objectid,
1259                                           BTRFS_DEV_EXTENT_KEY);
1260                 if (ret)
1261                         goto out;
1262                 leaf = path->nodes[0];
1263                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1264                 extent = btrfs_item_ptr(leaf, path->slots[0],
1265                                         struct btrfs_dev_extent);
1266                 BUG_ON(found_key.offset > start || found_key.offset +
1267                        btrfs_dev_extent_length(leaf, extent) < start);
1268                 key = found_key;
1269                 btrfs_release_path(path);
1270                 goto again;
1271         } else if (ret == 0) {
1272                 leaf = path->nodes[0];
1273                 extent = btrfs_item_ptr(leaf, path->slots[0],
1274                                         struct btrfs_dev_extent);
1275         } else {
1276                 btrfs_error(root->fs_info, ret, "Slot search failed");
1277                 goto out;
1278         }
1279
1280         if (device->bytes_used > 0) {
1281                 u64 len = btrfs_dev_extent_length(leaf, extent);
1282                 device->bytes_used -= len;
1283                 spin_lock(&root->fs_info->free_chunk_lock);
1284                 root->fs_info->free_chunk_space += len;
1285                 spin_unlock(&root->fs_info->free_chunk_lock);
1286         }
1287         ret = btrfs_del_item(trans, root, path);
1288         if (ret) {
1289                 btrfs_error(root->fs_info, ret,
1290                             "Failed to remove dev extent item");
1291         }
1292 out:
1293         btrfs_free_path(path);
1294         return ret;
1295 }
1296
1297 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1298                                   struct btrfs_device *device,
1299                                   u64 chunk_tree, u64 chunk_objectid,
1300                                   u64 chunk_offset, u64 start, u64 num_bytes)
1301 {
1302         int ret;
1303         struct btrfs_path *path;
1304         struct btrfs_root *root = device->dev_root;
1305         struct btrfs_dev_extent *extent;
1306         struct extent_buffer *leaf;
1307         struct btrfs_key key;
1308
1309         WARN_ON(!device->in_fs_metadata);
1310         WARN_ON(device->is_tgtdev_for_dev_replace);
1311         path = btrfs_alloc_path();
1312         if (!path)
1313                 return -ENOMEM;
1314
1315         key.objectid = device->devid;
1316         key.offset = start;
1317         key.type = BTRFS_DEV_EXTENT_KEY;
1318         ret = btrfs_insert_empty_item(trans, root, path, &key,
1319                                       sizeof(*extent));
1320         if (ret)
1321                 goto out;
1322
1323         leaf = path->nodes[0];
1324         extent = btrfs_item_ptr(leaf, path->slots[0],
1325                                 struct btrfs_dev_extent);
1326         btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
1327         btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
1328         btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1329
1330         write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
1331                     btrfs_dev_extent_chunk_tree_uuid(extent), BTRFS_UUID_SIZE);
1332
1333         btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1334         btrfs_mark_buffer_dirty(leaf);
1335 out:
1336         btrfs_free_path(path);
1337         return ret;
1338 }
1339
1340 static u64 find_next_chunk(struct btrfs_fs_info *fs_info)
1341 {
1342         struct extent_map_tree *em_tree;
1343         struct extent_map *em;
1344         struct rb_node *n;
1345         u64 ret = 0;
1346
1347         em_tree = &fs_info->mapping_tree.map_tree;
1348         read_lock(&em_tree->lock);
1349         n = rb_last(&em_tree->map);
1350         if (n) {
1351                 em = rb_entry(n, struct extent_map, rb_node);
1352                 ret = em->start + em->len;
1353         }
1354         read_unlock(&em_tree->lock);
1355
1356         return ret;
1357 }
1358
1359 static noinline int find_next_devid(struct btrfs_fs_info *fs_info,
1360                                     u64 *devid_ret)
1361 {
1362         int ret;
1363         struct btrfs_key key;
1364         struct btrfs_key found_key;
1365         struct btrfs_path *path;
1366
1367         path = btrfs_alloc_path();
1368         if (!path)
1369                 return -ENOMEM;
1370
1371         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1372         key.type = BTRFS_DEV_ITEM_KEY;
1373         key.offset = (u64)-1;
1374
1375         ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
1376         if (ret < 0)
1377                 goto error;
1378
1379         BUG_ON(ret == 0); /* Corruption */
1380
1381         ret = btrfs_previous_item(fs_info->chunk_root, path,
1382                                   BTRFS_DEV_ITEMS_OBJECTID,
1383                                   BTRFS_DEV_ITEM_KEY);
1384         if (ret) {
1385                 *devid_ret = 1;
1386         } else {
1387                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1388                                       path->slots[0]);
1389                 *devid_ret = found_key.offset + 1;
1390         }
1391         ret = 0;
1392 error:
1393         btrfs_free_path(path);
1394         return ret;
1395 }
1396
1397 /*
1398  * the device information is stored in the chunk root
1399  * the btrfs_device struct should be fully filled in
1400  */
1401 static int btrfs_add_device(struct btrfs_trans_handle *trans,
1402                             struct btrfs_root *root,
1403                             struct btrfs_device *device)
1404 {
1405         int ret;
1406         struct btrfs_path *path;
1407         struct btrfs_dev_item *dev_item;
1408         struct extent_buffer *leaf;
1409         struct btrfs_key key;
1410         unsigned long ptr;
1411
1412         root = root->fs_info->chunk_root;
1413
1414         path = btrfs_alloc_path();
1415         if (!path)
1416                 return -ENOMEM;
1417
1418         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1419         key.type = BTRFS_DEV_ITEM_KEY;
1420         key.offset = device->devid;
1421
1422         ret = btrfs_insert_empty_item(trans, root, path, &key,
1423                                       sizeof(*dev_item));
1424         if (ret)
1425                 goto out;
1426
1427         leaf = path->nodes[0];
1428         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1429
1430         btrfs_set_device_id(leaf, dev_item, device->devid);
1431         btrfs_set_device_generation(leaf, dev_item, 0);
1432         btrfs_set_device_type(leaf, dev_item, device->type);
1433         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1434         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1435         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1436         btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
1437         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1438         btrfs_set_device_group(leaf, dev_item, 0);
1439         btrfs_set_device_seek_speed(leaf, dev_item, 0);
1440         btrfs_set_device_bandwidth(leaf, dev_item, 0);
1441         btrfs_set_device_start_offset(leaf, dev_item, 0);
1442
1443         ptr = btrfs_device_uuid(dev_item);
1444         write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1445         ptr = btrfs_device_fsid(dev_item);
1446         write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1447         btrfs_mark_buffer_dirty(leaf);
1448
1449         ret = 0;
1450 out:
1451         btrfs_free_path(path);
1452         return ret;
1453 }
1454
1455 static int btrfs_rm_dev_item(struct btrfs_root *root,
1456                              struct btrfs_device *device)
1457 {
1458         int ret;
1459         struct btrfs_path *path;
1460         struct btrfs_key key;
1461         struct btrfs_trans_handle *trans;
1462
1463         root = root->fs_info->chunk_root;
1464
1465         path = btrfs_alloc_path();
1466         if (!path)
1467                 return -ENOMEM;
1468
1469         trans = btrfs_start_transaction(root, 0);
1470         if (IS_ERR(trans)) {
1471                 btrfs_free_path(path);
1472                 return PTR_ERR(trans);
1473         }
1474         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1475         key.type = BTRFS_DEV_ITEM_KEY;
1476         key.offset = device->devid;
1477         lock_chunks(root);
1478
1479         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1480         if (ret < 0)
1481                 goto out;
1482
1483         if (ret > 0) {
1484                 ret = -ENOENT;
1485                 goto out;
1486         }
1487
1488         ret = btrfs_del_item(trans, root, path);
1489         if (ret)
1490                 goto out;
1491 out:
1492         btrfs_free_path(path);
1493         unlock_chunks(root);
1494         btrfs_commit_transaction(trans, root);
1495         return ret;
1496 }
1497
1498 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
1499 {
1500         struct btrfs_device *device;
1501         struct btrfs_device *next_device;
1502         struct block_device *bdev;
1503         struct buffer_head *bh = NULL;
1504         struct btrfs_super_block *disk_super;
1505         struct btrfs_fs_devices *cur_devices;
1506         u64 all_avail;
1507         u64 devid;
1508         u64 num_devices;
1509         u8 *dev_uuid;
1510         unsigned seq;
1511         int ret = 0;
1512         bool clear_super = false;
1513
1514         mutex_lock(&uuid_mutex);
1515
1516         do {
1517                 seq = read_seqbegin(&root->fs_info->profiles_lock);
1518
1519                 all_avail = root->fs_info->avail_data_alloc_bits |
1520                             root->fs_info->avail_system_alloc_bits |
1521                             root->fs_info->avail_metadata_alloc_bits;
1522         } while (read_seqretry(&root->fs_info->profiles_lock, seq));
1523
1524         num_devices = root->fs_info->fs_devices->num_devices;
1525         btrfs_dev_replace_lock(&root->fs_info->dev_replace);
1526         if (btrfs_dev_replace_is_ongoing(&root->fs_info->dev_replace)) {
1527                 WARN_ON(num_devices < 1);
1528                 num_devices--;
1529         }
1530         btrfs_dev_replace_unlock(&root->fs_info->dev_replace);
1531
1532         if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) && num_devices <= 4) {
1533                 ret = BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET;
1534                 goto out;
1535         }
1536
1537         if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) && num_devices <= 2) {
1538                 ret = BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET;
1539                 goto out;
1540         }
1541
1542         if ((all_avail & BTRFS_BLOCK_GROUP_RAID5) &&
1543             root->fs_info->fs_devices->rw_devices <= 2) {
1544                 ret = BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET;
1545                 goto out;
1546         }
1547         if ((all_avail & BTRFS_BLOCK_GROUP_RAID6) &&
1548             root->fs_info->fs_devices->rw_devices <= 3) {
1549                 ret = BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET;
1550                 goto out;
1551         }
1552
1553         if (strcmp(device_path, "missing") == 0) {
1554                 struct list_head *devices;
1555                 struct btrfs_device *tmp;
1556
1557                 device = NULL;
1558                 devices = &root->fs_info->fs_devices->devices;
1559                 /*
1560                  * It is safe to read the devices since the volume_mutex
1561                  * is held.
1562                  */
1563                 list_for_each_entry(tmp, devices, dev_list) {
1564                         if (tmp->in_fs_metadata &&
1565                             !tmp->is_tgtdev_for_dev_replace &&
1566                             !tmp->bdev) {
1567                                 device = tmp;
1568                                 break;
1569                         }
1570                 }
1571                 bdev = NULL;
1572                 bh = NULL;
1573                 disk_super = NULL;
1574                 if (!device) {
1575                         ret = BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
1576                         goto out;
1577                 }
1578         } else {
1579                 ret = btrfs_get_bdev_and_sb(device_path,
1580                                             FMODE_WRITE | FMODE_EXCL,
1581                                             root->fs_info->bdev_holder, 0,
1582                                             &bdev, &bh);
1583                 if (ret)
1584                         goto out;
1585                 disk_super = (struct btrfs_super_block *)bh->b_data;
1586                 devid = btrfs_stack_device_id(&disk_super->dev_item);
1587                 dev_uuid = disk_super->dev_item.uuid;
1588                 device = btrfs_find_device(root->fs_info, devid, dev_uuid,
1589                                            disk_super->fsid);
1590                 if (!device) {
1591                         ret = -ENOENT;
1592                         goto error_brelse;
1593                 }
1594         }
1595
1596         if (device->is_tgtdev_for_dev_replace) {
1597                 ret = BTRFS_ERROR_DEV_TGT_REPLACE;
1598                 goto error_brelse;
1599         }
1600
1601         if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1602                 ret = BTRFS_ERROR_DEV_ONLY_WRITABLE;
1603                 goto error_brelse;
1604         }
1605
1606         if (device->writeable) {
1607                 lock_chunks(root);
1608                 list_del_init(&device->dev_alloc_list);
1609                 unlock_chunks(root);
1610                 root->fs_info->fs_devices->rw_devices--;
1611                 clear_super = true;
1612         }
1613
1614         mutex_unlock(&uuid_mutex);
1615         ret = btrfs_shrink_device(device, 0);
1616         mutex_lock(&uuid_mutex);
1617         if (ret)
1618                 goto error_undo;
1619
1620         /*
1621          * TODO: the superblock still includes this device in its num_devices
1622          * counter although write_all_supers() is not locked out. This
1623          * could give a filesystem state which requires a degraded mount.
1624          */
1625         ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1626         if (ret)
1627                 goto error_undo;
1628
1629         spin_lock(&root->fs_info->free_chunk_lock);
1630         root->fs_info->free_chunk_space = device->total_bytes -
1631                 device->bytes_used;
1632         spin_unlock(&root->fs_info->free_chunk_lock);
1633
1634         device->in_fs_metadata = 0;
1635         btrfs_scrub_cancel_dev(root->fs_info, device);
1636
1637         /*
1638          * the device list mutex makes sure that we don't change
1639          * the device list while someone else is writing out all
1640          * the device supers. Whoever is writing all supers, should
1641          * lock the device list mutex before getting the number of
1642          * devices in the super block (super_copy). Conversely,
1643          * whoever updates the number of devices in the super block
1644          * (super_copy) should hold the device list mutex.
1645          */
1646
1647         cur_devices = device->fs_devices;
1648         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1649         list_del_rcu(&device->dev_list);
1650
1651         device->fs_devices->num_devices--;
1652         device->fs_devices->total_devices--;
1653
1654         if (device->missing)
1655                 root->fs_info->fs_devices->missing_devices--;
1656
1657         next_device = list_entry(root->fs_info->fs_devices->devices.next,
1658                                  struct btrfs_device, dev_list);
1659         if (device->bdev == root->fs_info->sb->s_bdev)
1660                 root->fs_info->sb->s_bdev = next_device->bdev;
1661         if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1662                 root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1663
1664         if (device->bdev)
1665                 device->fs_devices->open_devices--;
1666
1667         call_rcu(&device->rcu, free_device);
1668
1669         num_devices = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
1670         btrfs_set_super_num_devices(root->fs_info->super_copy, num_devices);
1671         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1672
1673         if (cur_devices->open_devices == 0) {
1674                 struct btrfs_fs_devices *fs_devices;
1675                 fs_devices = root->fs_info->fs_devices;
1676                 while (fs_devices) {
1677                         if (fs_devices->seed == cur_devices)
1678                                 break;
1679                         fs_devices = fs_devices->seed;
1680                 }
1681                 fs_devices->seed = cur_devices->seed;
1682                 cur_devices->seed = NULL;
1683                 lock_chunks(root);
1684                 __btrfs_close_devices(cur_devices);
1685                 unlock_chunks(root);
1686                 free_fs_devices(cur_devices);
1687         }
1688
1689         root->fs_info->num_tolerated_disk_barrier_failures =
1690                 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
1691
1692         /*
1693          * at this point, the device is zero sized.  We want to
1694          * remove it from the devices list and zero out the old super
1695          */
1696         if (clear_super && disk_super) {
1697                 /* make sure this device isn't detected as part of
1698                  * the FS anymore
1699                  */
1700                 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1701                 set_buffer_dirty(bh);
1702                 sync_dirty_buffer(bh);
1703         }
1704
1705         ret = 0;
1706
1707         /* Notify udev that device has changed */
1708         if (bdev)
1709                 btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
1710
1711 error_brelse:
1712         brelse(bh);
1713         if (bdev)
1714                 blkdev_put(bdev, FMODE_READ | FMODE_EXCL);
1715 out:
1716         mutex_unlock(&uuid_mutex);
1717         return ret;
1718 error_undo:
1719         if (device->writeable) {
1720                 lock_chunks(root);
1721                 list_add(&device->dev_alloc_list,
1722                          &root->fs_info->fs_devices->alloc_list);
1723                 unlock_chunks(root);
1724                 root->fs_info->fs_devices->rw_devices++;
1725         }
1726         goto error_brelse;
1727 }
1728
1729 void btrfs_rm_dev_replace_srcdev(struct btrfs_fs_info *fs_info,
1730                                  struct btrfs_device *srcdev)
1731 {
1732         WARN_ON(!mutex_is_locked(&fs_info->fs_devices->device_list_mutex));
1733
1734         list_del_rcu(&srcdev->dev_list);
1735         list_del_rcu(&srcdev->dev_alloc_list);
1736         fs_info->fs_devices->num_devices--;
1737         if (srcdev->missing) {
1738                 fs_info->fs_devices->missing_devices--;
1739                 fs_info->fs_devices->rw_devices++;
1740         }
1741         if (srcdev->can_discard)
1742                 fs_info->fs_devices->num_can_discard--;
1743         if (srcdev->bdev) {
1744                 fs_info->fs_devices->open_devices--;
1745
1746                 /* zero out the old super */
1747                 btrfs_scratch_superblock(srcdev);
1748         }
1749
1750         call_rcu(&srcdev->rcu, free_device);
1751 }
1752
1753 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
1754                                       struct btrfs_device *tgtdev)
1755 {
1756         struct btrfs_device *next_device;
1757
1758         WARN_ON(!tgtdev);
1759         mutex_lock(&fs_info->fs_devices->device_list_mutex);
1760         if (tgtdev->bdev) {
1761                 btrfs_scratch_superblock(tgtdev);
1762                 fs_info->fs_devices->open_devices--;
1763         }
1764         fs_info->fs_devices->num_devices--;
1765         if (tgtdev->can_discard)
1766                 fs_info->fs_devices->num_can_discard++;
1767
1768         next_device = list_entry(fs_info->fs_devices->devices.next,
1769                                  struct btrfs_device, dev_list);
1770         if (tgtdev->bdev == fs_info->sb->s_bdev)
1771                 fs_info->sb->s_bdev = next_device->bdev;
1772         if (tgtdev->bdev == fs_info->fs_devices->latest_bdev)
1773                 fs_info->fs_devices->latest_bdev = next_device->bdev;
1774         list_del_rcu(&tgtdev->dev_list);
1775
1776         call_rcu(&tgtdev->rcu, free_device);
1777
1778         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
1779 }
1780
1781 static int btrfs_find_device_by_path(struct btrfs_root *root, char *device_path,
1782                                      struct btrfs_device **device)
1783 {
1784         int ret = 0;
1785         struct btrfs_super_block *disk_super;
1786         u64 devid;
1787         u8 *dev_uuid;
1788         struct block_device *bdev;
1789         struct buffer_head *bh;
1790
1791         *device = NULL;
1792         ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
1793                                     root->fs_info->bdev_holder, 0, &bdev, &bh);
1794         if (ret)
1795                 return ret;
1796         disk_super = (struct btrfs_super_block *)bh->b_data;
1797         devid = btrfs_stack_device_id(&disk_super->dev_item);
1798         dev_uuid = disk_super->dev_item.uuid;
1799         *device = btrfs_find_device(root->fs_info, devid, dev_uuid,
1800                                     disk_super->fsid);
1801         brelse(bh);
1802         if (!*device)
1803                 ret = -ENOENT;
1804         blkdev_put(bdev, FMODE_READ);
1805         return ret;
1806 }
1807
1808 int btrfs_find_device_missing_or_by_path(struct btrfs_root *root,
1809                                          char *device_path,
1810                                          struct btrfs_device **device)
1811 {
1812         *device = NULL;
1813         if (strcmp(device_path, "missing") == 0) {
1814                 struct list_head *devices;
1815                 struct btrfs_device *tmp;
1816
1817                 devices = &root->fs_info->fs_devices->devices;
1818                 /*
1819                  * It is safe to read the devices since the volume_mutex
1820                  * is held by the caller.
1821                  */
1822                 list_for_each_entry(tmp, devices, dev_list) {
1823                         if (tmp->in_fs_metadata && !tmp->bdev) {
1824                                 *device = tmp;
1825                                 break;
1826                         }
1827                 }
1828
1829                 if (!*device) {
1830                         btrfs_err(root->fs_info, "no missing device found");
1831                         return -ENOENT;
1832                 }
1833
1834                 return 0;
1835         } else {
1836                 return btrfs_find_device_by_path(root, device_path, device);
1837         }
1838 }
1839
1840 /*
1841  * does all the dirty work required for changing file system's UUID.
1842  */
1843 static int btrfs_prepare_sprout(struct btrfs_root *root)
1844 {
1845         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
1846         struct btrfs_fs_devices *old_devices;
1847         struct btrfs_fs_devices *seed_devices;
1848         struct btrfs_super_block *disk_super = root->fs_info->super_copy;
1849         struct btrfs_device *device;
1850         u64 super_flags;
1851
1852         BUG_ON(!mutex_is_locked(&uuid_mutex));
1853         if (!fs_devices->seeding)
1854                 return -EINVAL;
1855
1856         seed_devices = __alloc_fs_devices();
1857         if (IS_ERR(seed_devices))
1858                 return PTR_ERR(seed_devices);
1859
1860         old_devices = clone_fs_devices(fs_devices);
1861         if (IS_ERR(old_devices)) {
1862                 kfree(seed_devices);
1863                 return PTR_ERR(old_devices);
1864         }
1865
1866         list_add(&old_devices->list, &fs_uuids);
1867
1868         memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
1869         seed_devices->opened = 1;
1870         INIT_LIST_HEAD(&seed_devices->devices);
1871         INIT_LIST_HEAD(&seed_devices->alloc_list);
1872         mutex_init(&seed_devices->device_list_mutex);
1873
1874         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1875         list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
1876                               synchronize_rcu);
1877
1878         list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
1879         list_for_each_entry(device, &seed_devices->devices, dev_list) {
1880                 device->fs_devices = seed_devices;
1881         }
1882
1883         fs_devices->seeding = 0;
1884         fs_devices->num_devices = 0;
1885         fs_devices->open_devices = 0;
1886         fs_devices->total_devices = 0;
1887         fs_devices->seed = seed_devices;
1888
1889         generate_random_uuid(fs_devices->fsid);
1890         memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1891         memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1892         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1893
1894         super_flags = btrfs_super_flags(disk_super) &
1895                       ~BTRFS_SUPER_FLAG_SEEDING;
1896         btrfs_set_super_flags(disk_super, super_flags);
1897
1898         return 0;
1899 }
1900
1901 /*
1902  * strore the expected generation for seed devices in device items.
1903  */
1904 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
1905                                struct btrfs_root *root)
1906 {
1907         struct btrfs_path *path;
1908         struct extent_buffer *leaf;
1909         struct btrfs_dev_item *dev_item;
1910         struct btrfs_device *device;
1911         struct btrfs_key key;
1912         u8 fs_uuid[BTRFS_UUID_SIZE];
1913         u8 dev_uuid[BTRFS_UUID_SIZE];
1914         u64 devid;
1915         int ret;
1916
1917         path = btrfs_alloc_path();
1918         if (!path)
1919                 return -ENOMEM;
1920
1921         root = root->fs_info->chunk_root;
1922         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1923         key.offset = 0;
1924         key.type = BTRFS_DEV_ITEM_KEY;
1925
1926         while (1) {
1927                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1928                 if (ret < 0)
1929                         goto error;
1930
1931                 leaf = path->nodes[0];
1932 next_slot:
1933                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1934                         ret = btrfs_next_leaf(root, path);
1935                         if (ret > 0)
1936                                 break;
1937                         if (ret < 0)
1938                                 goto error;
1939                         leaf = path->nodes[0];
1940                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1941                         btrfs_release_path(path);
1942                         continue;
1943                 }
1944
1945                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1946                 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
1947                     key.type != BTRFS_DEV_ITEM_KEY)
1948                         break;
1949
1950                 dev_item = btrfs_item_ptr(leaf, path->slots[0],
1951                                           struct btrfs_dev_item);
1952                 devid = btrfs_device_id(leaf, dev_item);
1953                 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
1954                                    BTRFS_UUID_SIZE);
1955                 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
1956                                    BTRFS_UUID_SIZE);
1957                 device = btrfs_find_device(root->fs_info, devid, dev_uuid,
1958                                            fs_uuid);
1959                 BUG_ON(!device); /* Logic error */
1960
1961                 if (device->fs_devices->seeding) {
1962                         btrfs_set_device_generation(leaf, dev_item,
1963                                                     device->generation);
1964                         btrfs_mark_buffer_dirty(leaf);
1965                 }
1966
1967                 path->slots[0]++;
1968                 goto next_slot;
1969         }
1970         ret = 0;
1971 error:
1972         btrfs_free_path(path);
1973         return ret;
1974 }
1975
1976 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
1977 {
1978         struct request_queue *q;
1979         struct btrfs_trans_handle *trans;
1980         struct btrfs_device *device;
1981         struct block_device *bdev;
1982         struct list_head *devices;
1983         struct super_block *sb = root->fs_info->sb;
1984         struct rcu_string *name;
1985         u64 total_bytes;
1986         int seeding_dev = 0;
1987         int ret = 0;
1988
1989         if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
1990                 return -EROFS;
1991
1992         bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
1993                                   root->fs_info->bdev_holder);
1994         if (IS_ERR(bdev))
1995                 return PTR_ERR(bdev);
1996
1997         if (root->fs_info->fs_devices->seeding) {
1998                 seeding_dev = 1;
1999                 down_write(&sb->s_umount);
2000                 mutex_lock(&uuid_mutex);
2001         }
2002
2003         filemap_write_and_wait(bdev->bd_inode->i_mapping);
2004
2005         devices = &root->fs_info->fs_devices->devices;
2006
2007         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2008         list_for_each_entry(device, devices, dev_list) {
2009                 if (device->bdev == bdev) {
2010                         ret = -EEXIST;
2011                         mutex_unlock(
2012                                 &root->fs_info->fs_devices->device_list_mutex);
2013                         goto error;
2014                 }
2015         }
2016         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2017
2018         device = btrfs_alloc_device(root->fs_info, NULL, NULL);
2019         if (IS_ERR(device)) {
2020                 /* we can safely leave the fs_devices entry around */
2021                 ret = PTR_ERR(device);
2022                 goto error;
2023         }
2024
2025         name = rcu_string_strdup(device_path, GFP_NOFS);
2026         if (!name) {
2027                 kfree(device);
2028                 ret = -ENOMEM;
2029                 goto error;
2030         }
2031         rcu_assign_pointer(device->name, name);
2032
2033         trans = btrfs_start_transaction(root, 0);
2034         if (IS_ERR(trans)) {
2035                 rcu_string_free(device->name);
2036                 kfree(device);
2037                 ret = PTR_ERR(trans);
2038                 goto error;
2039         }
2040
2041         lock_chunks(root);
2042
2043         q = bdev_get_queue(bdev);
2044         if (blk_queue_discard(q))
2045                 device->can_discard = 1;
2046         device->writeable = 1;
2047         device->generation = trans->transid;
2048         device->io_width = root->sectorsize;
2049         device->io_align = root->sectorsize;
2050         device->sector_size = root->sectorsize;
2051         device->total_bytes = i_size_read(bdev->bd_inode);
2052         device->disk_total_bytes = device->total_bytes;
2053         device->dev_root = root->fs_info->dev_root;
2054         device->bdev = bdev;
2055         device->in_fs_metadata = 1;
2056         device->is_tgtdev_for_dev_replace = 0;
2057         device->mode = FMODE_EXCL;
2058         device->dev_stats_valid = 1;
2059         set_blocksize(device->bdev, 4096);
2060
2061         if (seeding_dev) {
2062                 sb->s_flags &= ~MS_RDONLY;
2063                 ret = btrfs_prepare_sprout(root);
2064                 BUG_ON(ret); /* -ENOMEM */
2065         }
2066
2067         device->fs_devices = root->fs_info->fs_devices;
2068
2069         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2070         list_add_rcu(&device->dev_list, &root->fs_info->fs_devices->devices);
2071         list_add(&device->dev_alloc_list,
2072                  &root->fs_info->fs_devices->alloc_list);
2073         root->fs_info->fs_devices->num_devices++;
2074         root->fs_info->fs_devices->open_devices++;
2075         root->fs_info->fs_devices->rw_devices++;
2076         root->fs_info->fs_devices->total_devices++;
2077         if (device->can_discard)
2078                 root->fs_info->fs_devices->num_can_discard++;
2079         root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
2080
2081         spin_lock(&root->fs_info->free_chunk_lock);
2082         root->fs_info->free_chunk_space += device->total_bytes;
2083         spin_unlock(&root->fs_info->free_chunk_lock);
2084
2085         if (!blk_queue_nonrot(bdev_get_queue(bdev)))
2086                 root->fs_info->fs_devices->rotating = 1;
2087
2088         total_bytes = btrfs_super_total_bytes(root->fs_info->super_copy);
2089         btrfs_set_super_total_bytes(root->fs_info->super_copy,
2090                                     total_bytes + device->total_bytes);
2091
2092         total_bytes = btrfs_super_num_devices(root->fs_info->super_copy);
2093         btrfs_set_super_num_devices(root->fs_info->super_copy,
2094                                     total_bytes + 1);
2095         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2096
2097         if (seeding_dev) {
2098                 ret = init_first_rw_device(trans, root, device);
2099                 if (ret) {
2100                         btrfs_abort_transaction(trans, root, ret);
2101                         goto error_trans;
2102                 }
2103                 ret = btrfs_finish_sprout(trans, root);
2104                 if (ret) {
2105                         btrfs_abort_transaction(trans, root, ret);
2106                         goto error_trans;
2107                 }
2108         } else {
2109                 ret = btrfs_add_device(trans, root, device);
2110                 if (ret) {
2111                         btrfs_abort_transaction(trans, root, ret);
2112                         goto error_trans;
2113                 }
2114         }
2115
2116         /*
2117          * we've got more storage, clear any full flags on the space
2118          * infos
2119          */
2120         btrfs_clear_space_info_full(root->fs_info);
2121
2122         unlock_chunks(root);
2123         root->fs_info->num_tolerated_disk_barrier_failures =
2124                 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
2125         ret = btrfs_commit_transaction(trans, root);
2126
2127         if (seeding_dev) {
2128                 mutex_unlock(&uuid_mutex);
2129                 up_write(&sb->s_umount);
2130
2131                 if (ret) /* transaction commit */
2132                         return ret;
2133
2134                 ret = btrfs_relocate_sys_chunks(root);
2135                 if (ret < 0)
2136                         btrfs_error(root->fs_info, ret,
2137                                     "Failed to relocate sys chunks after "
2138                                     "device initialization. This can be fixed "
2139                                     "using the \"btrfs balance\" command.");
2140                 trans = btrfs_attach_transaction(root);
2141                 if (IS_ERR(trans)) {
2142                         if (PTR_ERR(trans) == -ENOENT)
2143                                 return 0;
2144                         return PTR_ERR(trans);
2145                 }
2146                 ret = btrfs_commit_transaction(trans, root);
2147         }
2148
2149         return ret;
2150
2151 error_trans:
2152         unlock_chunks(root);
2153         btrfs_end_transaction(trans, root);
2154         rcu_string_free(device->name);
2155         kfree(device);
2156 error:
2157         blkdev_put(bdev, FMODE_EXCL);
2158         if (seeding_dev) {
2159                 mutex_unlock(&uuid_mutex);
2160                 up_write(&sb->s_umount);
2161         }
2162         return ret;
2163 }
2164
2165 int btrfs_init_dev_replace_tgtdev(struct btrfs_root *root, char *device_path,
2166                                   struct btrfs_device **device_out)
2167 {
2168         struct request_queue *q;
2169         struct btrfs_device *device;
2170         struct block_device *bdev;
2171         struct btrfs_fs_info *fs_info = root->fs_info;
2172         struct list_head *devices;
2173         struct rcu_string *name;
2174         u64 devid = BTRFS_DEV_REPLACE_DEVID;
2175         int ret = 0;
2176
2177         *device_out = NULL;
2178         if (fs_info->fs_devices->seeding)
2179                 return -EINVAL;
2180
2181         bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2182                                   fs_info->bdev_holder);
2183         if (IS_ERR(bdev))
2184                 return PTR_ERR(bdev);
2185
2186         filemap_write_and_wait(bdev->bd_inode->i_mapping);
2187
2188         devices = &fs_info->fs_devices->devices;
2189         list_for_each_entry(device, devices, dev_list) {
2190                 if (device->bdev == bdev) {
2191                         ret = -EEXIST;
2192                         goto error;
2193                 }
2194         }
2195
2196         device = btrfs_alloc_device(NULL, &devid, NULL);
2197         if (IS_ERR(device)) {
2198                 ret = PTR_ERR(device);
2199                 goto error;
2200         }
2201
2202         name = rcu_string_strdup(device_path, GFP_NOFS);
2203         if (!name) {
2204                 kfree(device);
2205                 ret = -ENOMEM;
2206                 goto error;
2207         }
2208         rcu_assign_pointer(device->name, name);
2209
2210         q = bdev_get_queue(bdev);
2211         if (blk_queue_discard(q))
2212                 device->can_discard = 1;
2213         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2214         device->writeable = 1;
2215         device->generation = 0;
2216         device->io_width = root->sectorsize;
2217         device->io_align = root->sectorsize;
2218         device->sector_size = root->sectorsize;
2219         device->total_bytes = i_size_read(bdev->bd_inode);
2220         device->disk_total_bytes = device->total_bytes;
2221         device->dev_root = fs_info->dev_root;
2222         device->bdev = bdev;
2223         device->in_fs_metadata = 1;
2224         device->is_tgtdev_for_dev_replace = 1;
2225         device->mode = FMODE_EXCL;
2226         device->dev_stats_valid = 1;
2227         set_blocksize(device->bdev, 4096);
2228         device->fs_devices = fs_info->fs_devices;
2229         list_add(&device->dev_list, &fs_info->fs_devices->devices);
2230         fs_info->fs_devices->num_devices++;
2231         fs_info->fs_devices->open_devices++;
2232         if (device->can_discard)
2233                 fs_info->fs_devices->num_can_discard++;
2234         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2235
2236         *device_out = device;
2237         return ret;
2238
2239 error:
2240         blkdev_put(bdev, FMODE_EXCL);
2241         return ret;
2242 }
2243
2244 void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info *fs_info,
2245                                               struct btrfs_device *tgtdev)
2246 {
2247         WARN_ON(fs_info->fs_devices->rw_devices == 0);
2248         tgtdev->io_width = fs_info->dev_root->sectorsize;
2249         tgtdev->io_align = fs_info->dev_root->sectorsize;
2250         tgtdev->sector_size = fs_info->dev_root->sectorsize;
2251         tgtdev->dev_root = fs_info->dev_root;
2252         tgtdev->in_fs_metadata = 1;
2253 }
2254
2255 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2256                                         struct btrfs_device *device)
2257 {
2258         int ret;
2259         struct btrfs_path *path;
2260         struct btrfs_root *root;
2261         struct btrfs_dev_item *dev_item;
2262         struct extent_buffer *leaf;
2263         struct btrfs_key key;
2264
2265         root = device->dev_root->fs_info->chunk_root;
2266
2267         path = btrfs_alloc_path();
2268         if (!path)
2269                 return -ENOMEM;
2270
2271         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2272         key.type = BTRFS_DEV_ITEM_KEY;
2273         key.offset = device->devid;
2274
2275         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2276         if (ret < 0)
2277                 goto out;
2278
2279         if (ret > 0) {
2280                 ret = -ENOENT;
2281                 goto out;
2282         }
2283
2284         leaf = path->nodes[0];
2285         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2286
2287         btrfs_set_device_id(leaf, dev_item, device->devid);
2288         btrfs_set_device_type(leaf, dev_item, device->type);
2289         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2290         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2291         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2292         btrfs_set_device_total_bytes(leaf, dev_item, device->disk_total_bytes);
2293         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
2294         btrfs_mark_buffer_dirty(leaf);
2295
2296 out:
2297         btrfs_free_path(path);
2298         return ret;
2299 }
2300
2301 static int __btrfs_grow_device(struct btrfs_trans_handle *trans,
2302                       struct btrfs_device *device, u64 new_size)
2303 {
2304         struct btrfs_super_block *super_copy =
2305                 device->dev_root->fs_info->super_copy;
2306         u64 old_total = btrfs_super_total_bytes(super_copy);
2307         u64 diff = new_size - device->total_bytes;
2308
2309         if (!device->writeable)
2310                 return -EACCES;
2311         if (new_size <= device->total_bytes ||
2312             device->is_tgtdev_for_dev_replace)
2313                 return -EINVAL;
2314
2315         btrfs_set_super_total_bytes(super_copy, old_total + diff);
2316         device->fs_devices->total_rw_bytes += diff;
2317
2318         device->total_bytes = new_size;
2319         device->disk_total_bytes = new_size;
2320         btrfs_clear_space_info_full(device->dev_root->fs_info);
2321
2322         return btrfs_update_device(trans, device);
2323 }
2324
2325 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2326                       struct btrfs_device *device, u64 new_size)
2327 {
2328         int ret;
2329         lock_chunks(device->dev_root);
2330         ret = __btrfs_grow_device(trans, device, new_size);
2331         unlock_chunks(device->dev_root);
2332         return ret;
2333 }
2334
2335 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
2336                             struct btrfs_root *root,
2337                             u64 chunk_tree, u64 chunk_objectid,
2338                             u64 chunk_offset)
2339 {
2340         int ret;
2341         struct btrfs_path *path;
2342         struct btrfs_key key;
2343
2344         root = root->fs_info->chunk_root;
2345         path = btrfs_alloc_path();
2346         if (!path)
2347                 return -ENOMEM;
2348
2349         key.objectid = chunk_objectid;
2350         key.offset = chunk_offset;
2351         key.type = BTRFS_CHUNK_ITEM_KEY;
2352
2353         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2354         if (ret < 0)
2355                 goto out;
2356         else if (ret > 0) { /* Logic error or corruption */
2357                 btrfs_error(root->fs_info, -ENOENT,
2358                             "Failed lookup while freeing chunk.");
2359                 ret = -ENOENT;
2360                 goto out;
2361         }
2362
2363         ret = btrfs_del_item(trans, root, path);
2364         if (ret < 0)
2365                 btrfs_error(root->fs_info, ret,
2366                             "Failed to delete chunk item.");
2367 out:
2368         btrfs_free_path(path);
2369         return ret;
2370 }
2371
2372 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
2373                         chunk_offset)
2374 {
2375         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
2376         struct btrfs_disk_key *disk_key;
2377         struct btrfs_chunk *chunk;
2378         u8 *ptr;
2379         int ret = 0;
2380         u32 num_stripes;
2381         u32 array_size;
2382         u32 len = 0;
2383         u32 cur;
2384         struct btrfs_key key;
2385
2386         array_size = btrfs_super_sys_array_size(super_copy);
2387
2388         ptr = super_copy->sys_chunk_array;
2389         cur = 0;
2390
2391         while (cur < array_size) {
2392                 disk_key = (struct btrfs_disk_key *)ptr;
2393                 btrfs_disk_key_to_cpu(&key, disk_key);
2394
2395                 len = sizeof(*disk_key);
2396
2397                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2398                         chunk = (struct btrfs_chunk *)(ptr + len);
2399                         num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2400                         len += btrfs_chunk_item_size(num_stripes);
2401                 } else {
2402                         ret = -EIO;
2403                         break;
2404                 }
2405                 if (key.objectid == chunk_objectid &&
2406                     key.offset == chunk_offset) {
2407                         memmove(ptr, ptr + len, array_size - (cur + len));
2408                         array_size -= len;
2409                         btrfs_set_super_sys_array_size(super_copy, array_size);
2410                 } else {
2411                         ptr += len;
2412                         cur += len;
2413                 }
2414         }
2415         return ret;
2416 }
2417
2418 static int btrfs_relocate_chunk(struct btrfs_root *root,
2419                          u64 chunk_tree, u64 chunk_objectid,
2420                          u64 chunk_offset)
2421 {
2422         struct extent_map_tree *em_tree;
2423         struct btrfs_root *extent_root;
2424         struct btrfs_trans_handle *trans;
2425         struct extent_map *em;
2426         struct map_lookup *map;
2427         int ret;
2428         int i;
2429
2430         root = root->fs_info->chunk_root;
2431         extent_root = root->fs_info->extent_root;
2432         em_tree = &root->fs_info->mapping_tree.map_tree;
2433
2434         ret = btrfs_can_relocate(extent_root, chunk_offset);
2435         if (ret)
2436                 return -ENOSPC;
2437
2438         /* step one, relocate all the extents inside this chunk */
2439         ret = btrfs_relocate_block_group(extent_root, chunk_offset);
2440         if (ret)
2441                 return ret;
2442
2443         trans = btrfs_start_transaction(root, 0);
2444         if (IS_ERR(trans)) {
2445                 ret = PTR_ERR(trans);
2446                 btrfs_std_error(root->fs_info, ret);
2447                 return ret;
2448         }
2449
2450         lock_chunks(root);
2451
2452         /*
2453          * step two, delete the device extents and the
2454          * chunk tree entries
2455          */
2456         read_lock(&em_tree->lock);
2457         em = lookup_extent_mapping(em_tree, chunk_offset, 1);
2458         read_unlock(&em_tree->lock);
2459
2460         BUG_ON(!em || em->start > chunk_offset ||
2461                em->start + em->len < chunk_offset);
2462         map = (struct map_lookup *)em->bdev;
2463
2464         for (i = 0; i < map->num_stripes; i++) {
2465                 ret = btrfs_free_dev_extent(trans, map->stripes[i].dev,
2466                                             map->stripes[i].physical);
2467                 BUG_ON(ret);
2468
2469                 if (map->stripes[i].dev) {
2470                         ret = btrfs_update_device(trans, map->stripes[i].dev);
2471                         BUG_ON(ret);
2472                 }
2473         }
2474         ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
2475                                chunk_offset);
2476
2477         BUG_ON(ret);
2478
2479         trace_btrfs_chunk_free(root, map, chunk_offset, em->len);
2480
2481         if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2482                 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
2483                 BUG_ON(ret);
2484         }
2485
2486         ret = btrfs_remove_block_group(trans, extent_root, chunk_offset);
2487         BUG_ON(ret);
2488
2489         write_lock(&em_tree->lock);
2490         remove_extent_mapping(em_tree, em);
2491         write_unlock(&em_tree->lock);
2492
2493         kfree(map);
2494         em->bdev = NULL;
2495
2496         /* once for the tree */
2497         free_extent_map(em);
2498         /* once for us */
2499         free_extent_map(em);
2500
2501         unlock_chunks(root);
2502         btrfs_end_transaction(trans, root);
2503         return 0;
2504 }
2505
2506 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
2507 {
2508         struct btrfs_root *chunk_root = root->fs_info->chunk_root;
2509         struct btrfs_path *path;
2510         struct extent_buffer *leaf;
2511         struct btrfs_chunk *chunk;
2512         struct btrfs_key key;
2513         struct btrfs_key found_key;
2514         u64 chunk_tree = chunk_root->root_key.objectid;
2515         u64 chunk_type;
2516         bool retried = false;
2517         int failed = 0;
2518         int ret;
2519
2520         path = btrfs_alloc_path();
2521         if (!path)
2522                 return -ENOMEM;
2523
2524 again:
2525         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2526         key.offset = (u64)-1;
2527         key.type = BTRFS_CHUNK_ITEM_KEY;
2528
2529         while (1) {
2530                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2531                 if (ret < 0)
2532                         goto error;
2533                 BUG_ON(ret == 0); /* Corruption */
2534
2535                 ret = btrfs_previous_item(chunk_root, path, key.objectid,
2536                                           key.type);
2537                 if (ret < 0)
2538                         goto error;
2539                 if (ret > 0)
2540                         break;
2541
2542                 leaf = path->nodes[0];
2543                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2544
2545                 chunk = btrfs_item_ptr(leaf, path->slots[0],
2546                                        struct btrfs_chunk);
2547                 chunk_type = btrfs_chunk_type(leaf, chunk);
2548                 btrfs_release_path(path);
2549
2550                 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
2551                         ret = btrfs_relocate_chunk(chunk_root, chunk_tree,
2552                                                    found_key.objectid,
2553                                                    found_key.offset);
2554                         if (ret == -ENOSPC)
2555                                 failed++;
2556                         else if (ret)
2557                                 BUG();
2558                 }
2559
2560                 if (found_key.offset == 0)
2561                         break;
2562                 key.offset = found_key.offset - 1;
2563         }
2564         ret = 0;
2565         if (failed && !retried) {
2566                 failed = 0;
2567                 retried = true;
2568                 goto again;
2569         } else if (WARN_ON(failed && retried)) {
2570                 ret = -ENOSPC;
2571         }
2572 error:
2573         btrfs_free_path(path);
2574         return ret;
2575 }
2576
2577 static int insert_balance_item(struct btrfs_root *root,
2578                                struct btrfs_balance_control *bctl)
2579 {
2580         struct btrfs_trans_handle *trans;
2581         struct btrfs_balance_item *item;
2582         struct btrfs_disk_balance_args disk_bargs;
2583         struct btrfs_path *path;
2584         struct extent_buffer *leaf;
2585         struct btrfs_key key;
2586         int ret, err;
2587
2588         path = btrfs_alloc_path();
2589         if (!path)
2590                 return -ENOMEM;
2591
2592         trans = btrfs_start_transaction(root, 0);
2593         if (IS_ERR(trans)) {
2594                 btrfs_free_path(path);
2595                 return PTR_ERR(trans);
2596         }
2597
2598         key.objectid = BTRFS_BALANCE_OBJECTID;
2599         key.type = BTRFS_BALANCE_ITEM_KEY;
2600         key.offset = 0;
2601
2602         ret = btrfs_insert_empty_item(trans, root, path, &key,
2603                                       sizeof(*item));
2604         if (ret)
2605                 goto out;
2606
2607         leaf = path->nodes[0];
2608         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
2609
2610         memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
2611
2612         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
2613         btrfs_set_balance_data(leaf, item, &disk_bargs);
2614         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
2615         btrfs_set_balance_meta(leaf, item, &disk_bargs);
2616         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
2617         btrfs_set_balance_sys(leaf, item, &disk_bargs);
2618
2619         btrfs_set_balance_flags(leaf, item, bctl->flags);
2620
2621         btrfs_mark_buffer_dirty(leaf);
2622 out:
2623         btrfs_free_path(path);
2624         err = btrfs_commit_transaction(trans, root);
2625         if (err && !ret)
2626                 ret = err;
2627         return ret;
2628 }
2629
2630 static int del_balance_item(struct btrfs_root *root)
2631 {
2632         struct btrfs_trans_handle *trans;
2633         struct btrfs_path *path;
2634         struct btrfs_key key;
2635         int ret, err;
2636
2637         path = btrfs_alloc_path();
2638         if (!path)
2639                 return -ENOMEM;
2640
2641         trans = btrfs_start_transaction(root, 0);
2642         if (IS_ERR(trans)) {
2643                 btrfs_free_path(path);
2644                 return PTR_ERR(trans);
2645         }
2646
2647         key.objectid = BTRFS_BALANCE_OBJECTID;
2648         key.type = BTRFS_BALANCE_ITEM_KEY;
2649         key.offset = 0;
2650
2651         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2652         if (ret < 0)
2653                 goto out;
2654         if (ret > 0) {
2655                 ret = -ENOENT;
2656                 goto out;
2657         }
2658
2659         ret = btrfs_del_item(trans, root, path);
2660 out:
2661         btrfs_free_path(path);
2662         err = btrfs_commit_transaction(trans, root);
2663         if (err && !ret)
2664                 ret = err;
2665         return ret;
2666 }
2667
2668 /*
2669  * This is a heuristic used to reduce the number of chunks balanced on
2670  * resume after balance was interrupted.
2671  */
2672 static void update_balance_args(struct btrfs_balance_control *bctl)
2673 {
2674         /*
2675          * Turn on soft mode for chunk types that were being converted.
2676          */
2677         if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
2678                 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
2679         if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
2680                 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
2681         if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
2682                 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
2683
2684         /*
2685          * Turn on usage filter if is not already used.  The idea is
2686          * that chunks that we have already balanced should be
2687          * reasonably full.  Don't do it for chunks that are being
2688          * converted - that will keep us from relocating unconverted
2689          * (albeit full) chunks.
2690          */
2691         if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2692             !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2693                 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
2694                 bctl->data.usage = 90;
2695         }
2696         if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2697             !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2698                 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
2699                 bctl->sys.usage = 90;
2700         }
2701         if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2702             !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2703                 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
2704                 bctl->meta.usage = 90;
2705         }
2706 }
2707
2708 /*
2709  * Should be called with both balance and volume mutexes held to
2710  * serialize other volume operations (add_dev/rm_dev/resize) with
2711  * restriper.  Same goes for unset_balance_control.
2712  */
2713 static void set_balance_control(struct btrfs_balance_control *bctl)
2714 {
2715         struct btrfs_fs_info *fs_info = bctl->fs_info;
2716
2717         BUG_ON(fs_info->balance_ctl);
2718
2719         spin_lock(&fs_info->balance_lock);
2720         fs_info->balance_ctl = bctl;
2721         spin_unlock(&fs_info->balance_lock);
2722 }
2723
2724 static void unset_balance_control(struct btrfs_fs_info *fs_info)
2725 {
2726         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
2727
2728         BUG_ON(!fs_info->balance_ctl);
2729
2730         spin_lock(&fs_info->balance_lock);
2731         fs_info->balance_ctl = NULL;
2732         spin_unlock(&fs_info->balance_lock);
2733
2734         kfree(bctl);
2735 }
2736
2737 /*
2738  * Balance filters.  Return 1 if chunk should be filtered out
2739  * (should not be balanced).
2740  */
2741 static int chunk_profiles_filter(u64 chunk_type,
2742                                  struct btrfs_balance_args *bargs)
2743 {
2744         chunk_type = chunk_to_extended(chunk_type) &
2745                                 BTRFS_EXTENDED_PROFILE_MASK;
2746
2747         if (bargs->profiles & chunk_type)
2748                 return 0;
2749
2750         return 1;
2751 }
2752
2753 static int chunk_usage_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
2754                               struct btrfs_balance_args *bargs)
2755 {
2756         struct btrfs_block_group_cache *cache;
2757         u64 chunk_used, user_thresh;
2758         int ret = 1;
2759
2760         cache = btrfs_lookup_block_group(fs_info, chunk_offset);
2761         chunk_used = btrfs_block_group_used(&cache->item);
2762
2763         if (bargs->usage == 0)
2764                 user_thresh = 1;
2765         else if (bargs->usage > 100)
2766                 user_thresh = cache->key.offset;
2767         else
2768                 user_thresh = div_factor_fine(cache->key.offset,
2769                                               bargs->usage);
2770
2771         if (chunk_used < user_thresh)
2772                 ret = 0;
2773
2774         btrfs_put_block_group(cache);
2775         return ret;
2776 }
2777
2778 static int chunk_devid_filter(struct extent_buffer *leaf,
2779                               struct btrfs_chunk *chunk,
2780                               struct btrfs_balance_args *bargs)
2781 {
2782         struct btrfs_stripe *stripe;
2783         int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2784         int i;
2785
2786         for (i = 0; i < num_stripes; i++) {
2787                 stripe = btrfs_stripe_nr(chunk, i);
2788                 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
2789                         return 0;
2790         }
2791
2792         return 1;
2793 }
2794
2795 /* [pstart, pend) */
2796 static int chunk_drange_filter(struct extent_buffer *leaf,
2797                                struct btrfs_chunk *chunk,
2798                                u64 chunk_offset,
2799                                struct btrfs_balance_args *bargs)
2800 {
2801         struct btrfs_stripe *stripe;
2802         int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2803         u64 stripe_offset;
2804         u64 stripe_length;
2805         int factor;
2806         int i;
2807
2808         if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
2809                 return 0;
2810
2811         if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP |
2812              BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) {
2813                 factor = num_stripes / 2;
2814         } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) {
2815                 factor = num_stripes - 1;
2816         } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) {
2817                 factor = num_stripes - 2;
2818         } else {
2819                 factor = num_stripes;
2820         }
2821
2822         for (i = 0; i < num_stripes; i++) {
2823                 stripe = btrfs_stripe_nr(chunk, i);
2824                 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
2825                         continue;
2826
2827                 stripe_offset = btrfs_stripe_offset(leaf, stripe);
2828                 stripe_length = btrfs_chunk_length(leaf, chunk);
2829                 do_div(stripe_length, factor);
2830
2831                 if (stripe_offset < bargs->pend &&
2832                     stripe_offset + stripe_length > bargs->pstart)
2833                         return 0;
2834         }
2835
2836         return 1;
2837 }
2838
2839 /* [vstart, vend) */
2840 static int chunk_vrange_filter(struct extent_buffer *leaf,
2841                                struct btrfs_chunk *chunk,
2842                                u64 chunk_offset,
2843                                struct btrfs_balance_args *bargs)
2844 {
2845         if (chunk_offset < bargs->vend &&
2846             chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
2847                 /* at least part of the chunk is inside this vrange */
2848                 return 0;
2849
2850         return 1;
2851 }
2852
2853 static int chunk_soft_convert_filter(u64 chunk_type,
2854                                      struct btrfs_balance_args *bargs)
2855 {
2856         if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
2857                 return 0;
2858
2859         chunk_type = chunk_to_extended(chunk_type) &
2860                                 BTRFS_EXTENDED_PROFILE_MASK;
2861
2862         if (bargs->target == chunk_type)
2863                 return 1;
2864
2865         return 0;
2866 }
2867
2868 static int should_balance_chunk(struct btrfs_root *root,
2869                                 struct extent_buffer *leaf,
2870                                 struct btrfs_chunk *chunk, u64 chunk_offset)
2871 {
2872         struct btrfs_balance_control *bctl = root->fs_info->balance_ctl;
2873         struct btrfs_balance_args *bargs = NULL;
2874         u64 chunk_type = btrfs_chunk_type(leaf, chunk);
2875
2876         /* type filter */
2877         if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
2878               (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
2879                 return 0;
2880         }
2881
2882         if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
2883                 bargs = &bctl->data;
2884         else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
2885                 bargs = &bctl->sys;
2886         else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
2887                 bargs = &bctl->meta;
2888
2889         /* profiles filter */
2890         if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
2891             chunk_profiles_filter(chunk_type, bargs)) {
2892                 return 0;
2893         }
2894
2895         /* usage filter */
2896         if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
2897             chunk_usage_filter(bctl->fs_info, chunk_offset, bargs)) {
2898                 return 0;
2899         }
2900
2901         /* devid filter */
2902         if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
2903             chunk_devid_filter(leaf, chunk, bargs)) {
2904                 return 0;
2905         }
2906
2907         /* drange filter, makes sense only with devid filter */
2908         if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
2909             chunk_drange_filter(leaf, chunk, chunk_offset, bargs)) {
2910                 return 0;
2911         }
2912
2913         /* vrange filter */
2914         if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
2915             chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
2916                 return 0;
2917         }
2918
2919         /* soft profile changing mode */
2920         if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
2921             chunk_soft_convert_filter(chunk_type, bargs)) {
2922                 return 0;
2923         }
2924
2925         return 1;
2926 }
2927
2928 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
2929 {
2930         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
2931         struct btrfs_root *chunk_root = fs_info->chunk_root;
2932         struct btrfs_root *dev_root = fs_info->dev_root;
2933         struct list_head *devices;
2934         struct btrfs_device *device;
2935         u64 old_size;
2936         u64 size_to_free;
2937         struct btrfs_chunk *chunk;
2938         struct btrfs_path *path;
2939         struct btrfs_key key;
2940         struct btrfs_key found_key;
2941         struct btrfs_trans_handle *trans;
2942         struct extent_buffer *leaf;
2943         int slot;
2944         int ret;
2945         int enospc_errors = 0;
2946         bool counting = true;
2947
2948         /* step one make some room on all the devices */
2949         devices = &fs_info->fs_devices->devices;
2950         list_for_each_entry(device, devices, dev_list) {
2951                 old_size = device->total_bytes;
2952                 size_to_free = div_factor(old_size, 1);
2953                 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
2954                 if (!device->writeable ||
2955                     device->total_bytes - device->bytes_used > size_to_free ||
2956                     device->is_tgtdev_for_dev_replace)
2957                         continue;
2958
2959                 ret = btrfs_shrink_device(device, old_size - size_to_free);
2960                 if (ret == -ENOSPC)
2961                         break;
2962                 BUG_ON(ret);
2963
2964                 trans = btrfs_start_transaction(dev_root, 0);
2965                 BUG_ON(IS_ERR(trans));
2966
2967                 ret = btrfs_grow_device(trans, device, old_size);
2968                 BUG_ON(ret);
2969
2970                 btrfs_end_transaction(trans, dev_root);
2971         }
2972
2973         /* step two, relocate all the chunks */
2974         path = btrfs_alloc_path();
2975         if (!path) {
2976                 ret = -ENOMEM;
2977                 goto error;
2978         }
2979
2980         /* zero out stat counters */
2981         spin_lock(&fs_info->balance_lock);
2982         memset(&bctl->stat, 0, sizeof(bctl->stat));
2983         spin_unlock(&fs_info->balance_lock);
2984 again:
2985         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2986         key.offset = (u64)-1;
2987         key.type = BTRFS_CHUNK_ITEM_KEY;
2988
2989         while (1) {
2990                 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
2991                     atomic_read(&fs_info->balance_cancel_req)) {
2992                         ret = -ECANCELED;
2993                         goto error;
2994                 }
2995
2996                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2997                 if (ret < 0)
2998                         goto error;
2999
3000                 /*
3001                  * this shouldn't happen, it means the last relocate
3002                  * failed
3003                  */
3004                 if (ret == 0)
3005                         BUG(); /* FIXME break ? */
3006
3007                 ret = btrfs_previous_item(chunk_root, path, 0,
3008                                           BTRFS_CHUNK_ITEM_KEY);
3009                 if (ret) {
3010                         ret = 0;
3011                         break;
3012                 }
3013
3014                 leaf = path->nodes[0];
3015                 slot = path->slots[0];
3016                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3017
3018                 if (found_key.objectid != key.objectid)
3019                         break;
3020
3021                 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3022
3023                 if (!counting) {
3024                         spin_lock(&fs_info->balance_lock);
3025                         bctl->stat.considered++;
3026                         spin_unlock(&fs_info->balance_lock);
3027                 }
3028
3029                 ret = should_balance_chunk(chunk_root, leaf, chunk,
3030                                            found_key.offset);
3031                 btrfs_release_path(path);
3032                 if (!ret)
3033                         goto loop;
3034
3035                 if (counting) {
3036                         spin_lock(&fs_info->balance_lock);
3037                         bctl->stat.expected++;
3038                         spin_unlock(&fs_info->balance_lock);
3039                         goto loop;
3040                 }
3041
3042                 ret = btrfs_relocate_chunk(chunk_root,
3043                                            chunk_root->root_key.objectid,
3044                                            found_key.objectid,
3045                                            found_key.offset);
3046                 if (ret && ret != -ENOSPC)
3047                         goto error;
3048                 if (ret == -ENOSPC) {
3049                         enospc_errors++;
3050                 } else {
3051                         spin_lock(&fs_info->balance_lock);
3052                         bctl->stat.completed++;
3053                         spin_unlock(&fs_info->balance_lock);
3054                 }
3055 loop:
3056                 if (found_key.offset == 0)
3057                         break;
3058                 key.offset = found_key.offset - 1;
3059         }
3060
3061         if (counting) {
3062                 btrfs_release_path(path);
3063                 counting = false;
3064                 goto again;
3065         }
3066 error:
3067         btrfs_free_path(path);
3068         if (enospc_errors) {
3069                 btrfs_info(fs_info, "%d enospc errors during balance",
3070                        enospc_errors);
3071                 if (!ret)
3072                         ret = -ENOSPC;
3073         }
3074
3075         return ret;
3076 }
3077
3078 /**
3079  * alloc_profile_is_valid - see if a given profile is valid and reduced
3080  * @flags: profile to validate
3081  * @extended: if true @flags is treated as an extended profile
3082  */
3083 static int alloc_profile_is_valid(u64 flags, int extended)
3084 {
3085         u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
3086                                BTRFS_BLOCK_GROUP_PROFILE_MASK);
3087
3088         flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
3089
3090         /* 1) check that all other bits are zeroed */
3091         if (flags & ~mask)
3092                 return 0;
3093
3094         /* 2) see if profile is reduced */
3095         if (flags == 0)
3096                 return !extended; /* "0" is valid for usual profiles */
3097
3098         /* true if exactly one bit set */
3099         return (flags & (flags - 1)) == 0;
3100 }
3101
3102 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
3103 {
3104         /* cancel requested || normal exit path */
3105         return atomic_read(&fs_info->balance_cancel_req) ||
3106                 (atomic_read(&fs_info->balance_pause_req) == 0 &&
3107                  atomic_read(&fs_info->balance_cancel_req) == 0);
3108 }
3109
3110 static void __cancel_balance(struct btrfs_fs_info *fs_info)
3111 {
3112         int ret;
3113
3114         unset_balance_control(fs_info);
3115         ret = del_balance_item(fs_info->tree_root);
3116         if (ret)
3117                 btrfs_std_error(fs_info, ret);
3118
3119         atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3120 }
3121
3122 /*
3123  * Should be called with both balance and volume mutexes held
3124  */
3125 int btrfs_balance(struct btrfs_balance_control *bctl,
3126                   struct btrfs_ioctl_balance_args *bargs)
3127 {
3128         struct btrfs_fs_info *fs_info = bctl->fs_info;
3129         u64 allowed;
3130         int mixed = 0;
3131         int ret;
3132         u64 num_devices;
3133         unsigned seq;
3134
3135         if (btrfs_fs_closing(fs_info) ||
3136             atomic_read(&fs_info->balance_pause_req) ||
3137             atomic_read(&fs_info->balance_cancel_req)) {
3138                 ret = -EINVAL;
3139                 goto out;
3140         }
3141
3142         allowed = btrfs_super_incompat_flags(fs_info->super_copy);
3143         if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
3144                 mixed = 1;
3145
3146         /*
3147          * In case of mixed groups both data and meta should be picked,
3148          * and identical options should be given for both of them.
3149          */
3150         allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
3151         if (mixed && (bctl->flags & allowed)) {
3152                 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
3153                     !(bctl->flags & BTRFS_BALANCE_METADATA) ||
3154                     memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
3155                         btrfs_err(fs_info, "with mixed groups data and "
3156                                    "metadata balance options must be the same");
3157                         ret = -EINVAL;
3158                         goto out;
3159                 }
3160         }
3161
3162         num_devices = fs_info->fs_devices->num_devices;
3163         btrfs_dev_replace_lock(&fs_info->dev_replace);
3164         if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
3165                 BUG_ON(num_devices < 1);
3166                 num_devices--;
3167         }
3168         btrfs_dev_replace_unlock(&fs_info->dev_replace);
3169         allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE;
3170         if (num_devices == 1)
3171                 allowed |= BTRFS_BLOCK_GROUP_DUP;
3172         else if (num_devices > 1)
3173                 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1);
3174         if (num_devices > 2)
3175                 allowed |= BTRFS_BLOCK_GROUP_RAID5;
3176         if (num_devices > 3)
3177                 allowed |= (BTRFS_BLOCK_GROUP_RAID10 |
3178                             BTRFS_BLOCK_GROUP_RAID6);
3179         if ((bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3180             (!alloc_profile_is_valid(bctl->data.target, 1) ||
3181              (bctl->data.target & ~allowed))) {
3182                 btrfs_err(fs_info, "unable to start balance with target "
3183                            "data profile %llu",
3184                        bctl->data.target);
3185                 ret = -EINVAL;
3186                 goto out;
3187         }
3188         if ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3189             (!alloc_profile_is_valid(bctl->meta.target, 1) ||
3190              (bctl->meta.target & ~allowed))) {
3191                 btrfs_err(fs_info,
3192                            "unable to start balance with target metadata profile %llu",
3193                        bctl->meta.target);
3194                 ret = -EINVAL;
3195                 goto out;
3196         }
3197         if ((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3198             (!alloc_profile_is_valid(bctl->sys.target, 1) ||
3199              (bctl->sys.target & ~allowed))) {
3200                 btrfs_err(fs_info,
3201                            "unable to start balance with target system profile %llu",
3202                        bctl->sys.target);
3203                 ret = -EINVAL;
3204                 goto out;
3205         }
3206
3207         /* allow dup'ed data chunks only in mixed mode */
3208         if (!mixed && (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3209             (bctl->data.target & BTRFS_BLOCK_GROUP_DUP)) {
3210                 btrfs_err(fs_info, "dup for data is not allowed");
3211                 ret = -EINVAL;
3212                 goto out;
3213         }
3214
3215         /* allow to reduce meta or sys integrity only if force set */
3216         allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3217                         BTRFS_BLOCK_GROUP_RAID10 |
3218                         BTRFS_BLOCK_GROUP_RAID5 |
3219                         BTRFS_BLOCK_GROUP_RAID6;
3220         do {
3221                 seq = read_seqbegin(&fs_info->profiles_lock);
3222
3223                 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3224                      (fs_info->avail_system_alloc_bits & allowed) &&
3225                      !(bctl->sys.target & allowed)) ||
3226                     ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3227                      (fs_info->avail_metadata_alloc_bits & allowed) &&
3228                      !(bctl->meta.target & allowed))) {
3229                         if (bctl->flags & BTRFS_BALANCE_FORCE) {
3230                                 btrfs_info(fs_info, "force reducing metadata integrity");
3231                         } else {
3232                                 btrfs_err(fs_info, "balance will reduce metadata "
3233                                            "integrity, use force if you want this");
3234                                 ret = -EINVAL;
3235                                 goto out;
3236                         }
3237                 }
3238         } while (read_seqretry(&fs_info->profiles_lock, seq));
3239
3240         if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3241                 int num_tolerated_disk_barrier_failures;
3242                 u64 target = bctl->sys.target;
3243
3244                 num_tolerated_disk_barrier_failures =
3245                         btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3246                 if (num_tolerated_disk_barrier_failures > 0 &&
3247                     (target &
3248                      (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID0 |
3249                       BTRFS_AVAIL_ALLOC_BIT_SINGLE)))
3250                         num_tolerated_disk_barrier_failures = 0;
3251                 else if (num_tolerated_disk_barrier_failures > 1 &&
3252                          (target &
3253                           (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)))
3254                         num_tolerated_disk_barrier_failures = 1;
3255
3256                 fs_info->num_tolerated_disk_barrier_failures =
3257                         num_tolerated_disk_barrier_failures;
3258         }
3259
3260         ret = insert_balance_item(fs_info->tree_root, bctl);
3261         if (ret && ret != -EEXIST)
3262                 goto out;
3263
3264         if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
3265                 BUG_ON(ret == -EEXIST);
3266                 set_balance_control(bctl);
3267         } else {
3268                 BUG_ON(ret != -EEXIST);
3269                 spin_lock(&fs_info->balance_lock);
3270                 update_balance_args(bctl);
3271                 spin_unlock(&fs_info->balance_lock);
3272         }
3273
3274         atomic_inc(&fs_info->balance_running);
3275         mutex_unlock(&fs_info->balance_mutex);
3276
3277         ret = __btrfs_balance(fs_info);
3278
3279         mutex_lock(&fs_info->balance_mutex);
3280         atomic_dec(&fs_info->balance_running);
3281
3282         if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3283                 fs_info->num_tolerated_disk_barrier_failures =
3284                         btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3285         }
3286
3287         if (bargs) {
3288                 memset(bargs, 0, sizeof(*bargs));
3289                 update_ioctl_balance_args(fs_info, 0, bargs);
3290         }
3291
3292         if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
3293             balance_need_close(fs_info)) {
3294                 __cancel_balance(fs_info);
3295         }
3296
3297         wake_up(&fs_info->balance_wait_q);
3298
3299         return ret;
3300 out:
3301         if (bctl->flags & BTRFS_BALANCE_RESUME)
3302                 __cancel_balance(fs_info);
3303         else {
3304                 kfree(bctl);
3305                 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3306         }
3307         return ret;
3308 }
3309
3310 static int balance_kthread(void *data)
3311 {
3312         struct btrfs_fs_info *fs_info = data;
3313         int ret = 0;
3314
3315         mutex_lock(&fs_info->volume_mutex);
3316         mutex_lock(&fs_info->balance_mutex);
3317
3318         if (fs_info->balance_ctl) {
3319                 btrfs_info(fs_info, "continuing balance");
3320                 ret = btrfs_balance(fs_info->balance_ctl, NULL);
3321         }
3322
3323         mutex_unlock(&fs_info->balance_mutex);
3324         mutex_unlock(&fs_info->volume_mutex);
3325
3326         return ret;
3327 }
3328
3329 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
3330 {
3331         struct task_struct *tsk;
3332
3333         spin_lock(&fs_info->balance_lock);
3334         if (!fs_info->balance_ctl) {
3335                 spin_unlock(&fs_info->balance_lock);
3336                 return 0;
3337         }
3338         spin_unlock(&fs_info->balance_lock);
3339
3340         if (btrfs_test_opt(fs_info->tree_root, SKIP_BALANCE)) {
3341                 btrfs_info(fs_info, "force skipping balance");
3342                 return 0;
3343         }
3344
3345         tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
3346         return PTR_ERR_OR_ZERO(tsk);
3347 }
3348
3349 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
3350 {
3351         struct btrfs_balance_control *bctl;
3352         struct btrfs_balance_item *item;
3353         struct btrfs_disk_balance_args disk_bargs;
3354         struct btrfs_path *path;
3355         struct extent_buffer *leaf;
3356         struct btrfs_key key;
3357         int ret;
3358
3359         path = btrfs_alloc_path();
3360         if (!path)
3361                 return -ENOMEM;
3362
3363         key.objectid = BTRFS_BALANCE_OBJECTID;
3364         key.type = BTRFS_BALANCE_ITEM_KEY;
3365         key.offset = 0;
3366
3367         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
3368         if (ret < 0)
3369                 goto out;
3370         if (ret > 0) { /* ret = -ENOENT; */
3371                 ret = 0;
3372                 goto out;
3373         }
3374
3375         bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
3376         if (!bctl) {
3377                 ret = -ENOMEM;
3378                 goto out;
3379         }
3380
3381         leaf = path->nodes[0];
3382         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3383
3384         bctl->fs_info = fs_info;
3385         bctl->flags = btrfs_balance_flags(leaf, item);
3386         bctl->flags |= BTRFS_BALANCE_RESUME;
3387
3388         btrfs_balance_data(leaf, item, &disk_bargs);
3389         btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
3390         btrfs_balance_meta(leaf, item, &disk_bargs);
3391         btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
3392         btrfs_balance_sys(leaf, item, &disk_bargs);
3393         btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
3394
3395         WARN_ON(atomic_xchg(&fs_info->mutually_exclusive_operation_running, 1));
3396
3397         mutex_lock(&fs_info->volume_mutex);
3398         mutex_lock(&fs_info->balance_mutex);
3399
3400         set_balance_control(bctl);
3401
3402         mutex_unlock(&fs_info->balance_mutex);
3403         mutex_unlock(&fs_info->volume_mutex);
3404 out:
3405         btrfs_free_path(path);
3406         return ret;
3407 }
3408
3409 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
3410 {
3411         int ret = 0;
3412
3413         mutex_lock(&fs_info->balance_mutex);
3414         if (!fs_info->balance_ctl) {
3415                 mutex_unlock(&fs_info->balance_mutex);
3416                 return -ENOTCONN;
3417         }
3418
3419         if (atomic_read(&fs_info->balance_running)) {
3420                 atomic_inc(&fs_info->balance_pause_req);
3421                 mutex_unlock(&fs_info->balance_mutex);
3422
3423                 wait_event(fs_info->balance_wait_q,
3424                            atomic_read(&fs_info->balance_running) == 0);
3425
3426                 mutex_lock(&fs_info->balance_mutex);
3427                 /* we are good with balance_ctl ripped off from under us */
3428                 BUG_ON(atomic_read(&fs_info->balance_running));
3429                 atomic_dec(&fs_info->balance_pause_req);
3430         } else {
3431                 ret = -ENOTCONN;
3432         }
3433
3434         mutex_unlock(&fs_info->balance_mutex);
3435         return ret;
3436 }
3437
3438 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
3439 {
3440         if (fs_info->sb->s_flags & MS_RDONLY)
3441                 return -EROFS;
3442
3443         mutex_lock(&fs_info->balance_mutex);
3444         if (!fs_info->balance_ctl) {
3445                 mutex_unlock(&fs_info->balance_mutex);
3446                 return -ENOTCONN;
3447         }
3448
3449         atomic_inc(&fs_info->balance_cancel_req);
3450         /*
3451          * if we are running just wait and return, balance item is
3452          * deleted in btrfs_balance in this case
3453          */
3454         if (atomic_read(&fs_info->balance_running)) {
3455                 mutex_unlock(&fs_info->balance_mutex);
3456                 wait_event(fs_info->balance_wait_q,
3457                            atomic_read(&fs_info->balance_running) == 0);
3458                 mutex_lock(&fs_info->balance_mutex);
3459         } else {
3460                 /* __cancel_balance needs volume_mutex */
3461                 mutex_unlock(&fs_info->balance_mutex);
3462                 mutex_lock(&fs_info->volume_mutex);
3463                 mutex_lock(&fs_info->balance_mutex);
3464
3465                 if (fs_info->balance_ctl)
3466                         __cancel_balance(fs_info);
3467
3468                 mutex_unlock(&fs_info->volume_mutex);
3469         }
3470
3471         BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running));
3472         atomic_dec(&fs_info->balance_cancel_req);
3473         mutex_unlock(&fs_info->balance_mutex);
3474         return 0;
3475 }
3476
3477 static int btrfs_uuid_scan_kthread(void *data)
3478 {
3479         struct btrfs_fs_info *fs_info = data;
3480         struct btrfs_root *root = fs_info->tree_root;
3481         struct btrfs_key key;
3482         struct btrfs_key max_key;
3483         struct btrfs_path *path = NULL;
3484         int ret = 0;
3485         struct extent_buffer *eb;
3486         int slot;
3487         struct btrfs_root_item root_item;
3488         u32 item_size;
3489         struct btrfs_trans_handle *trans = NULL;
3490
3491         path = btrfs_alloc_path();
3492         if (!path) {
3493                 ret = -ENOMEM;
3494                 goto out;
3495         }
3496
3497         key.objectid = 0;
3498         key.type = BTRFS_ROOT_ITEM_KEY;
3499         key.offset = 0;
3500
3501         max_key.objectid = (u64)-1;
3502         max_key.type = BTRFS_ROOT_ITEM_KEY;
3503         max_key.offset = (u64)-1;
3504
3505         path->keep_locks = 1;
3506
3507         while (1) {
3508                 ret = btrfs_search_forward(root, &key, path, 0);
3509                 if (ret) {
3510                         if (ret > 0)
3511                                 ret = 0;
3512                         break;
3513                 }
3514
3515                 if (key.type != BTRFS_ROOT_ITEM_KEY ||
3516                     (key.objectid < BTRFS_FIRST_FREE_OBJECTID &&
3517                      key.objectid != BTRFS_FS_TREE_OBJECTID) ||
3518                     key.objectid > BTRFS_LAST_FREE_OBJECTID)
3519                         goto skip;
3520
3521                 eb = path->nodes[0];
3522                 slot = path->slots[0];
3523                 item_size = btrfs_item_size_nr(eb, slot);
3524                 if (item_size < sizeof(root_item))
3525                         goto skip;
3526
3527                 read_extent_buffer(eb, &root_item,
3528                                    btrfs_item_ptr_offset(eb, slot),
3529                                    (int)sizeof(root_item));
3530                 if (btrfs_root_refs(&root_item) == 0)
3531                         goto skip;
3532
3533                 if (!btrfs_is_empty_uuid(root_item.uuid) ||
3534                     !btrfs_is_empty_uuid(root_item.received_uuid)) {
3535                         if (trans)
3536                                 goto update_tree;
3537
3538                         btrfs_release_path(path);
3539                         /*
3540                          * 1 - subvol uuid item
3541                          * 1 - received_subvol uuid item
3542                          */
3543                         trans = btrfs_start_transaction(fs_info->uuid_root, 2);
3544                         if (IS_ERR(trans)) {
3545                                 ret = PTR_ERR(trans);
3546                                 break;
3547                         }
3548                         continue;
3549                 } else {
3550                         goto skip;
3551                 }
3552 update_tree:
3553                 if (!btrfs_is_empty_uuid(root_item.uuid)) {
3554                         ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
3555                                                   root_item.uuid,
3556                                                   BTRFS_UUID_KEY_SUBVOL,
3557                                                   key.objectid);
3558                         if (ret < 0) {
3559                                 btrfs_warn(fs_info, "uuid_tree_add failed %d",
3560                                         ret);
3561                                 break;
3562                         }
3563                 }
3564
3565                 if (!btrfs_is_empty_uuid(root_item.received_uuid)) {
3566                         ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
3567                                                   root_item.received_uuid,
3568                                                  BTRFS_UUID_KEY_RECEIVED_SUBVOL,
3569                                                   key.objectid);
3570                         if (ret < 0) {
3571                                 btrfs_warn(fs_info, "uuid_tree_add failed %d",
3572                                         ret);
3573                                 break;
3574                         }
3575                 }
3576
3577 skip:
3578                 if (trans) {
3579                         ret = btrfs_end_transaction(trans, fs_info->uuid_root);
3580                         trans = NULL;
3581                         if (ret)
3582                                 break;
3583                 }
3584
3585                 btrfs_release_path(path);
3586                 if (key.offset < (u64)-1) {
3587                         key.offset++;
3588                 } else if (key.type < BTRFS_ROOT_ITEM_KEY) {
3589                         key.offset = 0;
3590                         key.type = BTRFS_ROOT_ITEM_KEY;
3591                 } else if (key.objectid < (u64)-1) {
3592                         key.offset = 0;
3593                         key.type = BTRFS_ROOT_ITEM_KEY;
3594                         key.objectid++;
3595                 } else {
3596                         break;
3597                 }
3598                 cond_resched();
3599         }
3600
3601 out:
3602         btrfs_free_path(path);
3603         if (trans && !IS_ERR(trans))
3604                 btrfs_end_transaction(trans, fs_info->uuid_root);
3605         if (ret)
3606                 btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret);
3607         else
3608                 fs_info->update_uuid_tree_gen = 1;
3609         up(&fs_info->uuid_tree_rescan_sem);
3610         return 0;
3611 }
3612
3613 /*
3614  * Callback for btrfs_uuid_tree_iterate().
3615  * returns:
3616  * 0    check succeeded, the entry is not outdated.
3617  * < 0  if an error occured.
3618  * > 0  if the check failed, which means the caller shall remove the entry.
3619  */
3620 static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info *fs_info,
3621                                        u8 *uuid, u8 type, u64 subid)
3622 {
3623         struct btrfs_key key;
3624         int ret = 0;
3625         struct btrfs_root *subvol_root;
3626
3627         if (type != BTRFS_UUID_KEY_SUBVOL &&
3628             type != BTRFS_UUID_KEY_RECEIVED_SUBVOL)
3629                 goto out;
3630
3631         key.objectid = subid;
3632         key.type = BTRFS_ROOT_ITEM_KEY;
3633         key.offset = (u64)-1;
3634         subvol_root = btrfs_read_fs_root_no_name(fs_info, &key);
3635         if (IS_ERR(subvol_root)) {
3636                 ret = PTR_ERR(subvol_root);
3637                 if (ret == -ENOENT)
3638                         ret = 1;
3639                 goto out;
3640         }
3641
3642         switch (type) {
3643         case BTRFS_UUID_KEY_SUBVOL:
3644                 if (memcmp(uuid, subvol_root->root_item.uuid, BTRFS_UUID_SIZE))
3645                         ret = 1;
3646                 break;
3647         case BTRFS_UUID_KEY_RECEIVED_SUBVOL:
3648                 if (memcmp(uuid, subvol_root->root_item.received_uuid,
3649                            BTRFS_UUID_SIZE))
3650                         ret = 1;
3651                 break;
3652         }
3653
3654 out:
3655         return ret;
3656 }
3657
3658 static int btrfs_uuid_rescan_kthread(void *data)
3659 {
3660         struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data;
3661         int ret;
3662
3663         /*
3664          * 1st step is to iterate through the existing UUID tree and
3665          * to delete all entries that contain outdated data.
3666          * 2nd step is to add all missing entries to the UUID tree.
3667          */
3668         ret = btrfs_uuid_tree_iterate(fs_info, btrfs_check_uuid_tree_entry);
3669         if (ret < 0) {
3670                 btrfs_warn(fs_info, "iterating uuid_tree failed %d", ret);
3671                 up(&fs_info->uuid_tree_rescan_sem);
3672                 return ret;
3673         }
3674         return btrfs_uuid_scan_kthread(data);
3675 }
3676
3677 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info)
3678 {
3679         struct btrfs_trans_handle *trans;
3680         struct btrfs_root *tree_root = fs_info->tree_root;
3681         struct btrfs_root *uuid_root;
3682         struct task_struct *task;
3683         int ret;
3684
3685         /*
3686          * 1 - root node
3687          * 1 - root item
3688          */
3689         trans = btrfs_start_transaction(tree_root, 2);
3690         if (IS_ERR(trans))
3691                 return PTR_ERR(trans);
3692
3693         uuid_root = btrfs_create_tree(trans, fs_info,
3694                                       BTRFS_UUID_TREE_OBJECTID);
3695         if (IS_ERR(uuid_root)) {
3696                 btrfs_abort_transaction(trans, tree_root,
3697                                         PTR_ERR(uuid_root));
3698                 return PTR_ERR(uuid_root);
3699         }
3700
3701         fs_info->uuid_root = uuid_root;
3702
3703         ret = btrfs_commit_transaction(trans, tree_root);
3704         if (ret)
3705                 return ret;
3706
3707         down(&fs_info->uuid_tree_rescan_sem);
3708         task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid");
3709         if (IS_ERR(task)) {
3710                 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
3711                 btrfs_warn(fs_info, "failed to start uuid_scan task");
3712                 up(&fs_info->uuid_tree_rescan_sem);
3713                 return PTR_ERR(task);
3714         }
3715
3716         return 0;
3717 }
3718
3719 int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
3720 {
3721         struct task_struct *task;
3722
3723         down(&fs_info->uuid_tree_rescan_sem);
3724         task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
3725         if (IS_ERR(task)) {
3726                 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
3727                 btrfs_warn(fs_info, "failed to start uuid_rescan task");
3728                 up(&fs_info->uuid_tree_rescan_sem);
3729                 return PTR_ERR(task);
3730         }
3731
3732         return 0;
3733 }
3734
3735 /*
3736  * shrinking a device means finding all of the device extents past
3737  * the new size, and then following the back refs to the chunks.
3738  * The chunk relocation code actually frees the device extent
3739  */
3740 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
3741 {
3742         struct btrfs_trans_handle *trans;
3743         struct btrfs_root *root = device->dev_root;
3744         struct btrfs_dev_extent *dev_extent = NULL;
3745         struct btrfs_path *path;
3746         u64 length;
3747         u64 chunk_tree;
3748         u64 chunk_objectid;
3749         u64 chunk_offset;
3750         int ret;
3751         int slot;
3752         int failed = 0;
3753         bool retried = false;
3754         struct extent_buffer *l;
3755         struct btrfs_key key;
3756         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
3757         u64 old_total = btrfs_super_total_bytes(super_copy);
3758         u64 old_size = device->total_bytes;
3759         u64 diff = device->total_bytes - new_size;
3760
3761         if (device->is_tgtdev_for_dev_replace)
3762                 return -EINVAL;
3763
3764         path = btrfs_alloc_path();
3765         if (!path)
3766                 return -ENOMEM;
3767
3768         path->reada = 2;
3769
3770         lock_chunks(root);
3771
3772         device->total_bytes = new_size;
3773         if (device->writeable) {
3774                 device->fs_devices->total_rw_bytes -= diff;
3775                 spin_lock(&root->fs_info->free_chunk_lock);
3776                 root->fs_info->free_chunk_space -= diff;
3777                 spin_unlock(&root->fs_info->free_chunk_lock);
3778         }
3779         unlock_chunks(root);
3780
3781 again:
3782         key.objectid = device->devid;
3783         key.offset = (u64)-1;
3784         key.type = BTRFS_DEV_EXTENT_KEY;
3785
3786         do {
3787                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3788                 if (ret < 0)
3789                         goto done;
3790
3791                 ret = btrfs_previous_item(root, path, 0, key.type);
3792                 if (ret < 0)
3793                         goto done;
3794                 if (ret) {
3795                         ret = 0;
3796                         btrfs_release_path(path);
3797                         break;
3798                 }
3799
3800                 l = path->nodes[0];
3801                 slot = path->slots[0];
3802                 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
3803
3804                 if (key.objectid != device->devid) {
3805                         btrfs_release_path(path);
3806                         break;
3807                 }
3808
3809                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
3810                 length = btrfs_dev_extent_length(l, dev_extent);
3811
3812                 if (key.offset + length <= new_size) {
3813                         btrfs_release_path(path);
3814                         break;
3815                 }
3816
3817                 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
3818                 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
3819                 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
3820                 btrfs_release_path(path);
3821
3822                 ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
3823                                            chunk_offset);
3824                 if (ret && ret != -ENOSPC)
3825                         goto done;
3826                 if (ret == -ENOSPC)
3827                         failed++;
3828         } while (key.offset-- > 0);
3829
3830         if (failed && !retried) {
3831                 failed = 0;
3832                 retried = true;
3833                 goto again;
3834         } else if (failed && retried) {
3835                 ret = -ENOSPC;
3836                 lock_chunks(root);
3837
3838                 device->total_bytes = old_size;
3839                 if (device->writeable)
3840                         device->fs_devices->total_rw_bytes += diff;
3841                 spin_lock(&root->fs_info->free_chunk_lock);
3842                 root->fs_info->free_chunk_space += diff;
3843                 spin_unlock(&root->fs_info->free_chunk_lock);
3844                 unlock_chunks(root);
3845                 goto done;
3846         }
3847
3848         /* Shrinking succeeded, else we would be at "done". */
3849         trans = btrfs_start_transaction(root, 0);
3850         if (IS_ERR(trans)) {
3851                 ret = PTR_ERR(trans);
3852                 goto done;
3853         }
3854
3855         lock_chunks(root);
3856
3857         device->disk_total_bytes = new_size;
3858         /* Now btrfs_update_device() will change the on-disk size. */
3859         ret = btrfs_update_device(trans, device);
3860         if (ret) {
3861                 unlock_chunks(root);
3862                 btrfs_end_transaction(trans, root);
3863                 goto done;
3864         }
3865         WARN_ON(diff > old_total);
3866         btrfs_set_super_total_bytes(super_copy, old_total - diff);
3867         unlock_chunks(root);
3868         btrfs_end_transaction(trans, root);
3869 done:
3870         btrfs_free_path(path);
3871         return ret;
3872 }
3873
3874 static int btrfs_add_system_chunk(struct btrfs_root *root,
3875                            struct btrfs_key *key,
3876                            struct btrfs_chunk *chunk, int item_size)
3877 {
3878         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
3879         struct btrfs_disk_key disk_key;
3880         u32 array_size;
3881         u8 *ptr;
3882
3883         array_size = btrfs_super_sys_array_size(super_copy);
3884         if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
3885                 return -EFBIG;
3886
3887         ptr = super_copy->sys_chunk_array + array_size;
3888         btrfs_cpu_key_to_disk(&disk_key, key);
3889         memcpy(ptr, &disk_key, sizeof(disk_key));
3890         ptr += sizeof(disk_key);
3891         memcpy(ptr, chunk, item_size);
3892         item_size += sizeof(disk_key);
3893         btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
3894         return 0;
3895 }
3896
3897 /*
3898  * sort the devices in descending order by max_avail, total_avail
3899  */
3900 static int btrfs_cmp_device_info(const void *a, const void *b)
3901 {
3902         const struct btrfs_device_info *di_a = a;
3903         const struct btrfs_device_info *di_b = b;
3904
3905         if (di_a->max_avail > di_b->max_avail)
3906                 return -1;
3907         if (di_a->max_avail < di_b->max_avail)
3908                 return 1;
3909         if (di_a->total_avail > di_b->total_avail)
3910                 return -1;
3911         if (di_a->total_avail < di_b->total_avail)
3912                 return 1;
3913         return 0;
3914 }
3915
3916 static struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = {
3917         [BTRFS_RAID_RAID10] = {
3918                 .sub_stripes    = 2,
3919                 .dev_stripes    = 1,
3920                 .devs_max       = 0,    /* 0 == as many as possible */
3921                 .devs_min       = 4,
3922                 .devs_increment = 2,
3923                 .ncopies        = 2,
3924         },
3925         [BTRFS_RAID_RAID1] = {
3926                 .sub_stripes    = 1,
3927                 .dev_stripes    = 1,
3928                 .devs_max       = 2,
3929                 .devs_min       = 2,
3930                 .devs_increment = 2,
3931                 .ncopies        = 2,
3932         },
3933         [BTRFS_RAID_DUP] = {
3934                 .sub_stripes    = 1,
3935                 .dev_stripes    = 2,
3936                 .devs_max       = 1,
3937                 .devs_min       = 1,
3938                 .devs_increment = 1,
3939                 .ncopies        = 2,
3940         },
3941         [BTRFS_RAID_RAID0] = {
3942                 .sub_stripes    = 1,
3943                 .dev_stripes    = 1,
3944                 .devs_max       = 0,
3945                 .devs_min       = 2,
3946                 .devs_increment = 1,
3947                 .ncopies        = 1,
3948         },
3949         [BTRFS_RAID_SINGLE] = {
3950                 .sub_stripes    = 1,
3951                 .dev_stripes    = 1,
3952                 .devs_max       = 1,
3953                 .devs_min       = 1,
3954                 .devs_increment = 1,
3955                 .ncopies        = 1,
3956         },
3957         [BTRFS_RAID_RAID5] = {
3958                 .sub_stripes    = 1,
3959                 .dev_stripes    = 1,
3960                 .devs_max       = 0,
3961                 .devs_min       = 2,
3962                 .devs_increment = 1,
3963                 .ncopies        = 2,
3964         },
3965         [BTRFS_RAID_RAID6] = {
3966                 .sub_stripes    = 1,
3967                 .dev_stripes    = 1,
3968                 .devs_max       = 0,
3969                 .devs_min       = 3,
3970                 .devs_increment = 1,
3971                 .ncopies        = 3,
3972         },
3973 };
3974
3975 static u32 find_raid56_stripe_len(u32 data_devices, u32 dev_stripe_target)
3976 {
3977         /* TODO allow them to set a preferred stripe size */
3978         return 64 * 1024;
3979 }
3980
3981 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
3982 {
3983         if (!(type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)))
3984                 return;
3985
3986         btrfs_set_fs_incompat(info, RAID56);
3987 }
3988
3989 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
3990                                struct btrfs_root *extent_root, u64 start,
3991                                u64 type)
3992 {
3993         struct btrfs_fs_info *info = extent_root->fs_info;
3994         struct btrfs_fs_devices *fs_devices = info->fs_devices;
3995         struct list_head *cur;
3996         struct map_lookup *map = NULL;
3997         struct extent_map_tree *em_tree;
3998         struct extent_map *em;
3999         struct btrfs_device_info *devices_info = NULL;
4000         u64 total_avail;
4001         int num_stripes;        /* total number of stripes to allocate */
4002         int data_stripes;       /* number of stripes that count for
4003                                    block group size */
4004         int sub_stripes;        /* sub_stripes info for map */
4005         int dev_stripes;        /* stripes per dev */
4006         int devs_max;           /* max devs to use */
4007         int devs_min;           /* min devs needed */
4008         int devs_increment;     /* ndevs has to be a multiple of this */
4009         int ncopies;            /* how many copies to data has */
4010         int ret;
4011         u64 max_stripe_size;
4012         u64 max_chunk_size;
4013         u64 stripe_size;
4014         u64 num_bytes;
4015         u64 raid_stripe_len = BTRFS_STRIPE_LEN;
4016         int ndevs;
4017         int i;
4018         int j;
4019         int index;
4020
4021         BUG_ON(!alloc_profile_is_valid(type, 0));
4022
4023         if (list_empty(&fs_devices->alloc_list))
4024                 return -ENOSPC;
4025
4026         index = __get_raid_index(type);
4027
4028         sub_stripes = btrfs_raid_array[index].sub_stripes;
4029         dev_stripes = btrfs_raid_array[index].dev_stripes;
4030         devs_max = btrfs_raid_array[index].devs_max;
4031         devs_min = btrfs_raid_array[index].devs_min;
4032         devs_increment = btrfs_raid_array[index].devs_increment;
4033         ncopies = btrfs_raid_array[index].ncopies;
4034
4035         if (type & BTRFS_BLOCK_GROUP_DATA) {
4036                 max_stripe_size = 1024 * 1024 * 1024;
4037                 max_chunk_size = 10 * max_stripe_size;
4038         } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
4039                 /* for larger filesystems, use larger metadata chunks */
4040                 if (fs_devices->total_rw_bytes > 50ULL * 1024 * 1024 * 1024)
4041                         max_stripe_size = 1024 * 1024 * 1024;
4042                 else
4043                         max_stripe_size = 256 * 1024 * 1024;
4044                 max_chunk_size = max_stripe_size;
4045         } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
4046                 max_stripe_size = 32 * 1024 * 1024;
4047                 max_chunk_size = 2 * max_stripe_size;
4048         } else {
4049                 btrfs_err(info, "invalid chunk type 0x%llx requested\n",
4050                        type);
4051                 BUG_ON(1);
4052         }
4053
4054         /* we don't want a chunk larger than 10% of writeable space */
4055         max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
4056                              max_chunk_size);
4057
4058         devices_info = kzalloc(sizeof(*devices_info) * fs_devices->rw_devices,
4059                                GFP_NOFS);
4060         if (!devices_info)
4061                 return -ENOMEM;
4062
4063         cur = fs_devices->alloc_list.next;
4064
4065         /*
4066          * in the first pass through the devices list, we gather information
4067          * about the available holes on each device.
4068          */
4069         ndevs = 0;
4070         while (cur != &fs_devices->alloc_list) {
4071                 struct btrfs_device *device;
4072                 u64 max_avail;
4073                 u64 dev_offset;
4074
4075                 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
4076
4077                 cur = cur->next;
4078
4079                 if (!device->writeable) {
4080                         WARN(1, KERN_ERR
4081                                "BTRFS: read-only device in alloc_list\n");
4082                         continue;
4083                 }
4084
4085                 if (!device->in_fs_metadata ||
4086                     device->is_tgtdev_for_dev_replace)
4087                         continue;
4088
4089                 if (device->total_bytes > device->bytes_used)
4090                         total_avail = device->total_bytes - device->bytes_used;
4091                 else
4092                         total_avail = 0;
4093
4094                 /* If there is no space on this device, skip it. */
4095                 if (total_avail == 0)
4096                         continue;
4097
4098                 ret = find_free_dev_extent(trans, device,
4099                                            max_stripe_size * dev_stripes,
4100                                            &dev_offset, &max_avail);
4101                 if (ret && ret != -ENOSPC)
4102                         goto error;
4103
4104                 if (ret == 0)
4105                         max_avail = max_stripe_size * dev_stripes;
4106
4107                 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes)
4108                         continue;
4109
4110                 if (ndevs == fs_devices->rw_devices) {
4111                         WARN(1, "%s: found more than %llu devices\n",
4112                              __func__, fs_devices->rw_devices);
4113                         break;
4114                 }
4115                 devices_info[ndevs].dev_offset = dev_offset;
4116                 devices_info[ndevs].max_avail = max_avail;
4117                 devices_info[ndevs].total_avail = total_avail;
4118                 devices_info[ndevs].dev = device;
4119                 ++ndevs;
4120         }
4121
4122         /*
4123          * now sort the devices by hole size / available space
4124          */
4125         sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
4126              btrfs_cmp_device_info, NULL);
4127
4128         /* round down to number of usable stripes */
4129         ndevs -= ndevs % devs_increment;
4130
4131         if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) {
4132                 ret = -ENOSPC;
4133                 goto error;
4134         }
4135
4136         if (devs_max && ndevs > devs_max)
4137                 ndevs = devs_max;
4138         /*
4139          * the primary goal is to maximize the number of stripes, so use as many
4140          * devices as possible, even if the stripes are not maximum sized.
4141          */
4142         stripe_size = devices_info[ndevs-1].max_avail;
4143         num_stripes = ndevs * dev_stripes;
4144
4145         /*
4146          * this will have to be fixed for RAID1 and RAID10 over
4147          * more drives
4148          */
4149         data_stripes = num_stripes / ncopies;
4150
4151         if (type & BTRFS_BLOCK_GROUP_RAID5) {
4152                 raid_stripe_len = find_raid56_stripe_len(ndevs - 1,
4153                                  btrfs_super_stripesize(info->super_copy));
4154                 data_stripes = num_stripes - 1;
4155         }
4156         if (type & BTRFS_BLOCK_GROUP_RAID6) {
4157                 raid_stripe_len = find_raid56_stripe_len(ndevs - 2,
4158                                  btrfs_super_stripesize(info->super_copy));
4159                 data_stripes = num_stripes - 2;
4160         }
4161
4162         /*
4163          * Use the number of data stripes to figure out how big this chunk
4164          * is really going to be in terms of logical address space,
4165          * and compare that answer with the max chunk size
4166          */
4167         if (stripe_size * data_stripes > max_chunk_size) {
4168                 u64 mask = (1ULL << 24) - 1;
4169                 stripe_size = max_chunk_size;
4170                 do_div(stripe_size, data_stripes);
4171
4172                 /* bump the answer up to a 16MB boundary */
4173                 stripe_size = (stripe_size + mask) & ~mask;
4174
4175                 /* but don't go higher than the limits we found
4176                  * while searching for free extents
4177                  */
4178                 if (stripe_size > devices_info[ndevs-1].max_avail)
4179                         stripe_size = devices_info[ndevs-1].max_avail;
4180         }
4181
4182         do_div(stripe_size, dev_stripes);
4183
4184         /* align to BTRFS_STRIPE_LEN */
4185         do_div(stripe_size, raid_stripe_len);
4186         stripe_size *= raid_stripe_len;
4187
4188         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
4189         if (!map) {
4190                 ret = -ENOMEM;
4191                 goto error;
4192         }
4193         map->num_stripes = num_stripes;
4194
4195         for (i = 0; i < ndevs; ++i) {
4196                 for (j = 0; j < dev_stripes; ++j) {
4197                         int s = i * dev_stripes + j;
4198                         map->stripes[s].dev = devices_info[i].dev;
4199                         map->stripes[s].physical = devices_info[i].dev_offset +
4200                                                    j * stripe_size;
4201                 }
4202         }
4203         map->sector_size = extent_root->sectorsize;
4204         map->stripe_len = raid_stripe_len;
4205         map->io_align = raid_stripe_len;
4206         map->io_width = raid_stripe_len;
4207         map->type = type;
4208         map->sub_stripes = sub_stripes;
4209
4210         num_bytes = stripe_size * data_stripes;
4211
4212         trace_btrfs_chunk_alloc(info->chunk_root, map, start, num_bytes);
4213
4214         em = alloc_extent_map();
4215         if (!em) {
4216                 ret = -ENOMEM;
4217                 goto error;
4218         }
4219         em->bdev = (struct block_device *)map;
4220         em->start = start;
4221         em->len = num_bytes;
4222         em->block_start = 0;
4223         em->block_len = em->len;
4224         em->orig_block_len = stripe_size;
4225
4226         em_tree = &extent_root->fs_info->mapping_tree.map_tree;
4227         write_lock(&em_tree->lock);
4228         ret = add_extent_mapping(em_tree, em, 0);
4229         if (!ret) {
4230                 list_add_tail(&em->list, &trans->transaction->pending_chunks);
4231                 atomic_inc(&em->refs);
4232         }
4233         write_unlock(&em_tree->lock);
4234         if (ret) {
4235                 free_extent_map(em);
4236                 goto error;
4237         }
4238
4239         ret = btrfs_make_block_group(trans, extent_root, 0, type,
4240                                      BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4241                                      start, num_bytes);
4242         if (ret)
4243                 goto error_del_extent;
4244
4245         free_extent_map(em);
4246         check_raid56_incompat_flag(extent_root->fs_info, type);
4247
4248         kfree(devices_info);
4249         return 0;
4250
4251 error_del_extent:
4252         write_lock(&em_tree->lock);
4253         remove_extent_mapping(em_tree, em);
4254         write_unlock(&em_tree->lock);
4255
4256         /* One for our allocation */
4257         free_extent_map(em);
4258         /* One for the tree reference */
4259         free_extent_map(em);
4260 error:
4261         kfree(map);
4262         kfree(devices_info);
4263         return ret;
4264 }
4265
4266 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans,
4267                                 struct btrfs_root *extent_root,
4268                                 u64 chunk_offset, u64 chunk_size)
4269 {
4270         struct btrfs_key key;
4271         struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
4272         struct btrfs_device *device;
4273         struct btrfs_chunk *chunk;
4274         struct btrfs_stripe *stripe;
4275         struct extent_map_tree *em_tree;
4276         struct extent_map *em;
4277         struct map_lookup *map;
4278         size_t item_size;
4279         u64 dev_offset;
4280         u64 stripe_size;
4281         int i = 0;
4282         int ret;
4283
4284         em_tree = &extent_root->fs_info->mapping_tree.map_tree;
4285         read_lock(&em_tree->lock);
4286         em = lookup_extent_mapping(em_tree, chunk_offset, chunk_size);
4287         read_unlock(&em_tree->lock);
4288
4289         if (!em) {
4290                 btrfs_crit(extent_root->fs_info, "unable to find logical "
4291                            "%Lu len %Lu", chunk_offset, chunk_size);
4292                 return -EINVAL;
4293         }
4294
4295         if (em->start != chunk_offset || em->len != chunk_size) {
4296                 btrfs_crit(extent_root->fs_info, "found a bad mapping, wanted"
4297                           " %Lu-%Lu, found %Lu-%Lu\n", chunk_offset,
4298                           chunk_size, em->start, em->len);
4299                 free_extent_map(em);
4300                 return -EINVAL;
4301         }
4302
4303         map = (struct map_lookup *)em->bdev;
4304         item_size = btrfs_chunk_item_size(map->num_stripes);
4305         stripe_size = em->orig_block_len;
4306
4307         chunk = kzalloc(item_size, GFP_NOFS);
4308         if (!chunk) {
4309                 ret = -ENOMEM;
4310                 goto out;
4311         }
4312
4313         for (i = 0; i < map->num_stripes; i++) {
4314                 device = map->stripes[i].dev;
4315                 dev_offset = map->stripes[i].physical;
4316
4317                 device->bytes_used += stripe_size;
4318                 ret = btrfs_update_device(trans, device);
4319                 if (ret)
4320                         goto out;
4321                 ret = btrfs_alloc_dev_extent(trans, device,
4322                                              chunk_root->root_key.objectid,
4323                                              BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4324                                              chunk_offset, dev_offset,
4325                                              stripe_size);
4326                 if (ret)
4327                         goto out;
4328         }
4329
4330         spin_lock(&extent_root->fs_info->free_chunk_lock);
4331         extent_root->fs_info->free_chunk_space -= (stripe_size *
4332                                                    map->num_stripes);
4333         spin_unlock(&extent_root->fs_info->free_chunk_lock);
4334
4335         stripe = &chunk->stripe;
4336         for (i = 0; i < map->num_stripes; i++) {
4337                 device = map->stripes[i].dev;
4338                 dev_offset = map->stripes[i].physical;
4339
4340                 btrfs_set_stack_stripe_devid(stripe, device->devid);
4341                 btrfs_set_stack_stripe_offset(stripe, dev_offset);
4342                 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
4343                 stripe++;
4344         }
4345
4346         btrfs_set_stack_chunk_length(chunk, chunk_size);
4347         btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
4348         btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
4349         btrfs_set_stack_chunk_type(chunk, map->type);
4350         btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
4351         btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
4352         btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
4353         btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
4354         btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
4355
4356         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
4357         key.type = BTRFS_CHUNK_ITEM_KEY;
4358         key.offset = chunk_offset;
4359
4360         ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
4361         if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
4362                 /*
4363                  * TODO: Cleanup of inserted chunk root in case of
4364                  * failure.
4365                  */
4366                 ret = btrfs_add_system_chunk(chunk_root, &key, chunk,
4367                                              item_size);
4368         }
4369
4370 out:
4371         kfree(chunk);
4372         free_extent_map(em);
4373         return ret;
4374 }
4375
4376 /*
4377  * Chunk allocation falls into two parts. The first part does works
4378  * that make the new allocated chunk useable, but not do any operation
4379  * that modifies the chunk tree. The second part does the works that
4380  * require modifying the chunk tree. This division is important for the
4381  * bootstrap process of adding storage to a seed btrfs.
4382  */
4383 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4384                       struct btrfs_root *extent_root, u64 type)
4385 {
4386         u64 chunk_offset;
4387
4388         chunk_offset = find_next_chunk(extent_root->fs_info);
4389         return __btrfs_alloc_chunk(trans, extent_root, chunk_offset, type);
4390 }
4391
4392 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
4393                                          struct btrfs_root *root,
4394                                          struct btrfs_device *device)
4395 {
4396         u64 chunk_offset;
4397         u64 sys_chunk_offset;
4398         u64 alloc_profile;
4399         struct btrfs_fs_info *fs_info = root->fs_info;
4400         struct btrfs_root *extent_root = fs_info->extent_root;
4401         int ret;
4402
4403         chunk_offset = find_next_chunk(fs_info);
4404         alloc_profile = btrfs_get_alloc_profile(extent_root, 0);
4405         ret = __btrfs_alloc_chunk(trans, extent_root, chunk_offset,
4406                                   alloc_profile);
4407         if (ret)
4408                 return ret;
4409
4410         sys_chunk_offset = find_next_chunk(root->fs_info);
4411         alloc_profile = btrfs_get_alloc_profile(fs_info->chunk_root, 0);
4412         ret = __btrfs_alloc_chunk(trans, extent_root, sys_chunk_offset,
4413                                   alloc_profile);
4414         if (ret) {
4415                 btrfs_abort_transaction(trans, root, ret);
4416                 goto out;
4417         }
4418
4419         ret = btrfs_add_device(trans, fs_info->chunk_root, device);
4420         if (ret)
4421                 btrfs_abort_transaction(trans, root, ret);
4422 out:
4423         return ret;
4424 }
4425
4426 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
4427 {
4428         struct extent_map *em;
4429         struct map_lookup *map;
4430         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
4431         int readonly = 0;
4432         int i;
4433
4434         read_lock(&map_tree->map_tree.lock);
4435         em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
4436         read_unlock(&map_tree->map_tree.lock);
4437         if (!em)
4438                 return 1;
4439
4440         if (btrfs_test_opt(root, DEGRADED)) {
4441                 free_extent_map(em);
4442                 return 0;
4443         }
4444
4445         map = (struct map_lookup *)em->bdev;
4446         for (i = 0; i < map->num_stripes; i++) {
4447                 if (!map->stripes[i].dev->writeable) {
4448                         readonly = 1;
4449                         break;
4450                 }
4451         }
4452         free_extent_map(em);
4453         return readonly;
4454 }
4455
4456 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
4457 {
4458         extent_map_tree_init(&tree->map_tree);
4459 }
4460
4461 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
4462 {
4463         struct extent_map *em;
4464
4465         while (1) {
4466                 write_lock(&tree->map_tree.lock);
4467                 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
4468                 if (em)
4469                         remove_extent_mapping(&tree->map_tree, em);
4470                 write_unlock(&tree->map_tree.lock);
4471                 if (!em)
4472                         break;
4473                 kfree(em->bdev);
4474                 /* once for us */
4475                 free_extent_map(em);
4476                 /* once for the tree */
4477                 free_extent_map(em);
4478         }
4479 }
4480
4481 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
4482 {
4483         struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
4484         struct extent_map *em;
4485         struct map_lookup *map;
4486         struct extent_map_tree *em_tree = &map_tree->map_tree;
4487         int ret;
4488
4489         read_lock(&em_tree->lock);
4490         em = lookup_extent_mapping(em_tree, logical, len);
4491         read_unlock(&em_tree->lock);
4492
4493         /*
4494          * We could return errors for these cases, but that could get ugly and
4495          * we'd probably do the same thing which is just not do anything else
4496          * and exit, so return 1 so the callers don't try to use other copies.
4497          */
4498         if (!em) {
4499                 btrfs_crit(fs_info, "No mapping for %Lu-%Lu\n", logical,
4500                             logical+len);
4501                 return 1;
4502         }
4503
4504         if (em->start > logical || em->start + em->len < logical) {
4505                 btrfs_crit(fs_info, "Invalid mapping for %Lu-%Lu, got "
4506                             "%Lu-%Lu\n", logical, logical+len, em->start,
4507                             em->start + em->len);
4508                 free_extent_map(em);
4509                 return 1;
4510         }
4511
4512         map = (struct map_lookup *)em->bdev;
4513         if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
4514                 ret = map->num_stripes;
4515         else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
4516                 ret = map->sub_stripes;
4517         else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
4518                 ret = 2;
4519         else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
4520                 ret = 3;
4521         else
4522                 ret = 1;
4523         free_extent_map(em);
4524
4525         btrfs_dev_replace_lock(&fs_info->dev_replace);
4526         if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace))
4527                 ret++;
4528         btrfs_dev_replace_unlock(&fs_info->dev_replace);
4529
4530         return ret;
4531 }
4532
4533 unsigned long btrfs_full_stripe_len(struct btrfs_root *root,
4534                                     struct btrfs_mapping_tree *map_tree,
4535                                     u64 logical)
4536 {
4537         struct extent_map *em;
4538         struct map_lookup *map;
4539         struct extent_map_tree *em_tree = &map_tree->map_tree;
4540         unsigned long len = root->sectorsize;
4541
4542         read_lock(&em_tree->lock);
4543         em = lookup_extent_mapping(em_tree, logical, len);
4544         read_unlock(&em_tree->lock);
4545         BUG_ON(!em);
4546
4547         BUG_ON(em->start > logical || em->start + em->len < logical);
4548         map = (struct map_lookup *)em->bdev;
4549         if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
4550                          BTRFS_BLOCK_GROUP_RAID6)) {
4551                 len = map->stripe_len * nr_data_stripes(map);
4552         }
4553         free_extent_map(em);
4554         return len;
4555 }
4556
4557 int btrfs_is_parity_mirror(struct btrfs_mapping_tree *map_tree,
4558                            u64 logical, u64 len, int mirror_num)
4559 {
4560         struct extent_map *em;
4561         struct map_lookup *map;
4562         struct extent_map_tree *em_tree = &map_tree->map_tree;
4563         int ret = 0;
4564
4565         read_lock(&em_tree->lock);
4566         em = lookup_extent_mapping(em_tree, logical, len);
4567         read_unlock(&em_tree->lock);
4568         BUG_ON(!em);
4569
4570         BUG_ON(em->start > logical || em->start + em->len < logical);
4571         map = (struct map_lookup *)em->bdev;
4572         if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
4573                          BTRFS_BLOCK_GROUP_RAID6))
4574                 ret = 1;
4575         free_extent_map(em);
4576         return ret;
4577 }
4578
4579 static int find_live_mirror(struct btrfs_fs_info *fs_info,
4580                             struct map_lookup *map, int first, int num,
4581                             int optimal, int dev_replace_is_ongoing)
4582 {
4583         int i;
4584         int tolerance;
4585         struct btrfs_device *srcdev;
4586
4587         if (dev_replace_is_ongoing &&
4588             fs_info->dev_replace.cont_reading_from_srcdev_mode ==
4589              BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
4590                 srcdev = fs_info->dev_replace.srcdev;
4591         else
4592                 srcdev = NULL;
4593
4594         /*
4595          * try to avoid the drive that is the source drive for a
4596          * dev-replace procedure, only choose it if no other non-missing
4597          * mirror is available
4598          */
4599         for (tolerance = 0; tolerance < 2; tolerance++) {
4600                 if (map->stripes[optimal].dev->bdev &&
4601                     (tolerance || map->stripes[optimal].dev != srcdev))
4602                         return optimal;
4603                 for (i = first; i < first + num; i++) {
4604                         if (map->stripes[i].dev->bdev &&
4605                             (tolerance || map->stripes[i].dev != srcdev))
4606                                 return i;
4607                 }
4608         }
4609
4610         /* we couldn't find one that doesn't fail.  Just return something
4611          * and the io error handling code will clean up eventually
4612          */
4613         return optimal;
4614 }
4615
4616 static inline int parity_smaller(u64 a, u64 b)
4617 {
4618         return a > b;
4619 }
4620
4621 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
4622 static void sort_parity_stripes(struct btrfs_bio *bbio, u64 *raid_map)
4623 {
4624         struct btrfs_bio_stripe s;
4625         int i;
4626         u64 l;
4627         int again = 1;
4628
4629         while (again) {
4630                 again = 0;
4631                 for (i = 0; i < bbio->num_stripes - 1; i++) {
4632                         if (parity_smaller(raid_map[i], raid_map[i+1])) {
4633                                 s = bbio->stripes[i];
4634                                 l = raid_map[i];
4635                                 bbio->stripes[i] = bbio->stripes[i+1];
4636                                 raid_map[i] = raid_map[i+1];
4637                                 bbio->stripes[i+1] = s;
4638                                 raid_map[i+1] = l;
4639                                 again = 1;
4640                         }
4641                 }
4642         }
4643 }
4644
4645 static int __btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
4646                              u64 logical, u64 *length,
4647                              struct btrfs_bio **bbio_ret,
4648                              int mirror_num, u64 **raid_map_ret)
4649 {
4650         struct extent_map *em;
4651         struct map_lookup *map;
4652         struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
4653         struct extent_map_tree *em_tree = &map_tree->map_tree;
4654         u64 offset;
4655         u64 stripe_offset;
4656         u64 stripe_end_offset;
4657         u64 stripe_nr;
4658         u64 stripe_nr_orig;
4659         u64 stripe_nr_end;
4660         u64 stripe_len;
4661         u64 *raid_map = NULL;
4662         int stripe_index;
4663         int i;
4664         int ret = 0;
4665         int num_stripes;
4666         int max_errors = 0;
4667         struct btrfs_bio *bbio = NULL;
4668         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
4669         int dev_replace_is_ongoing = 0;
4670         int num_alloc_stripes;
4671         int patch_the_first_stripe_for_dev_replace = 0;
4672         u64 physical_to_patch_in_first_stripe = 0;
4673         u64 raid56_full_stripe_start = (u64)-1;
4674
4675         read_lock(&em_tree->lock);
4676         em = lookup_extent_mapping(em_tree, logical, *length);
4677         read_unlock(&em_tree->lock);
4678
4679         if (!em) {
4680                 btrfs_crit(fs_info, "unable to find logical %llu len %llu",
4681                         logical, *length);
4682                 return -EINVAL;
4683         }
4684
4685         if (em->start > logical || em->start + em->len < logical) {
4686                 btrfs_crit(fs_info, "found a bad mapping, wanted %Lu, "
4687                            "found %Lu-%Lu\n", logical, em->start,
4688                            em->start + em->len);
4689                 free_extent_map(em);
4690                 return -EINVAL;
4691         }
4692
4693         map = (struct map_lookup *)em->bdev;
4694         offset = logical - em->start;
4695
4696         stripe_len = map->stripe_len;
4697         stripe_nr = offset;
4698         /*
4699          * stripe_nr counts the total number of stripes we have to stride
4700          * to get to this block
4701          */
4702         do_div(stripe_nr, stripe_len);
4703
4704         stripe_offset = stripe_nr * stripe_len;
4705         BUG_ON(offset < stripe_offset);
4706
4707         /* stripe_offset is the offset of this block in its stripe*/
4708         stripe_offset = offset - stripe_offset;
4709
4710         /* if we're here for raid56, we need to know the stripe aligned start */
4711         if (map->type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)) {
4712                 unsigned long full_stripe_len = stripe_len * nr_data_stripes(map);
4713                 raid56_full_stripe_start = offset;
4714
4715                 /* allow a write of a full stripe, but make sure we don't
4716                  * allow straddling of stripes
4717                  */
4718                 do_div(raid56_full_stripe_start, full_stripe_len);
4719                 raid56_full_stripe_start *= full_stripe_len;
4720         }
4721
4722         if (rw & REQ_DISCARD) {
4723                 /* we don't discard raid56 yet */
4724                 if (map->type &
4725                     (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)) {
4726                         ret = -EOPNOTSUPP;
4727                         goto out;
4728                 }
4729                 *length = min_t(u64, em->len - offset, *length);
4730         } else if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
4731                 u64 max_len;
4732                 /* For writes to RAID[56], allow a full stripeset across all disks.
4733                    For other RAID types and for RAID[56] reads, just allow a single
4734                    stripe (on a single disk). */
4735                 if (map->type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6) &&
4736                     (rw & REQ_WRITE)) {
4737                         max_len = stripe_len * nr_data_stripes(map) -
4738                                 (offset - raid56_full_stripe_start);
4739                 } else {
4740                         /* we limit the length of each bio to what fits in a stripe */
4741                         max_len = stripe_len - stripe_offset;
4742                 }
4743                 *length = min_t(u64, em->len - offset, max_len);
4744         } else {
4745                 *length = em->len - offset;
4746         }
4747
4748         /* This is for when we're called from btrfs_merge_bio_hook() and all
4749            it cares about is the length */
4750         if (!bbio_ret)
4751                 goto out;
4752
4753         btrfs_dev_replace_lock(dev_replace);
4754         dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
4755         if (!dev_replace_is_ongoing)
4756                 btrfs_dev_replace_unlock(dev_replace);
4757
4758         if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
4759             !(rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) &&
4760             dev_replace->tgtdev != NULL) {
4761                 /*
4762                  * in dev-replace case, for repair case (that's the only
4763                  * case where the mirror is selected explicitly when
4764                  * calling btrfs_map_block), blocks left of the left cursor
4765                  * can also be read from the target drive.
4766                  * For REQ_GET_READ_MIRRORS, the target drive is added as
4767                  * the last one to the array of stripes. For READ, it also
4768                  * needs to be supported using the same mirror number.
4769                  * If the requested block is not left of the left cursor,
4770                  * EIO is returned. This can happen because btrfs_num_copies()
4771                  * returns one more in the dev-replace case.
4772                  */
4773                 u64 tmp_length = *length;
4774                 struct btrfs_bio *tmp_bbio = NULL;
4775                 int tmp_num_stripes;
4776                 u64 srcdev_devid = dev_replace->srcdev->devid;
4777                 int index_srcdev = 0;
4778                 int found = 0;
4779                 u64 physical_of_found = 0;
4780
4781                 ret = __btrfs_map_block(fs_info, REQ_GET_READ_MIRRORS,
4782                              logical, &tmp_length, &tmp_bbio, 0, NULL);
4783                 if (ret) {
4784                         WARN_ON(tmp_bbio != NULL);
4785                         goto out;
4786                 }
4787
4788                 tmp_num_stripes = tmp_bbio->num_stripes;
4789                 if (mirror_num > tmp_num_stripes) {
4790                         /*
4791                          * REQ_GET_READ_MIRRORS does not contain this
4792                          * mirror, that means that the requested area
4793                          * is not left of the left cursor
4794                          */
4795                         ret = -EIO;
4796                         kfree(tmp_bbio);
4797                         goto out;
4798                 }
4799
4800                 /*
4801                  * process the rest of the function using the mirror_num
4802                  * of the source drive. Therefore look it up first.
4803                  * At the end, patch the device pointer to the one of the
4804                  * target drive.
4805                  */
4806                 for (i = 0; i < tmp_num_stripes; i++) {
4807                         if (tmp_bbio->stripes[i].dev->devid == srcdev_devid) {
4808                                 /*
4809                                  * In case of DUP, in order to keep it
4810                                  * simple, only add the mirror with the
4811                                  * lowest physical address
4812                                  */
4813                                 if (found &&
4814                                     physical_of_found <=
4815                                      tmp_bbio->stripes[i].physical)
4816                                         continue;
4817                                 index_srcdev = i;
4818                                 found = 1;
4819                                 physical_of_found =
4820                                         tmp_bbio->stripes[i].physical;
4821                         }
4822                 }
4823
4824                 if (found) {
4825                         mirror_num = index_srcdev + 1;
4826                         patch_the_first_stripe_for_dev_replace = 1;
4827                         physical_to_patch_in_first_stripe = physical_of_found;
4828                 } else {
4829                         WARN_ON(1);
4830                         ret = -EIO;
4831                         kfree(tmp_bbio);
4832                         goto out;
4833                 }
4834
4835                 kfree(tmp_bbio);
4836         } else if (mirror_num > map->num_stripes) {
4837                 mirror_num = 0;
4838         }
4839
4840         num_stripes = 1;
4841         stripe_index = 0;
4842         stripe_nr_orig = stripe_nr;
4843         stripe_nr_end = ALIGN(offset + *length, map->stripe_len);
4844         do_div(stripe_nr_end, map->stripe_len);
4845         stripe_end_offset = stripe_nr_end * map->stripe_len -
4846                             (offset + *length);
4847
4848         if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
4849                 if (rw & REQ_DISCARD)
4850                         num_stripes = min_t(u64, map->num_stripes,
4851                                             stripe_nr_end - stripe_nr_orig);
4852                 stripe_index = do_div(stripe_nr, map->num_stripes);
4853         } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
4854                 if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS))
4855                         num_stripes = map->num_stripes;
4856                 else if (mirror_num)
4857                         stripe_index = mirror_num - 1;
4858                 else {
4859                         stripe_index = find_live_mirror(fs_info, map, 0,
4860                                             map->num_stripes,
4861                                             current->pid % map->num_stripes,
4862                                             dev_replace_is_ongoing);
4863                         mirror_num = stripe_index + 1;
4864                 }
4865
4866         } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
4867                 if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) {
4868                         num_stripes = map->num_stripes;
4869                 } else if (mirror_num) {
4870                         stripe_index = mirror_num - 1;
4871                 } else {
4872                         mirror_num = 1;
4873                 }
4874
4875         } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
4876                 int factor = map->num_stripes / map->sub_stripes;
4877
4878                 stripe_index = do_div(stripe_nr, factor);
4879                 stripe_index *= map->sub_stripes;
4880
4881                 if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS))
4882                         num_stripes = map->sub_stripes;
4883                 else if (rw & REQ_DISCARD)
4884                         num_stripes = min_t(u64, map->sub_stripes *
4885                                             (stripe_nr_end - stripe_nr_orig),
4886                                             map->num_stripes);
4887                 else if (mirror_num)
4888                         stripe_index += mirror_num - 1;
4889                 else {
4890                         int old_stripe_index = stripe_index;
4891                         stripe_index = find_live_mirror(fs_info, map,
4892                                               stripe_index,
4893                                               map->sub_stripes, stripe_index +
4894                                               current->pid % map->sub_stripes,
4895                                               dev_replace_is_ongoing);
4896                         mirror_num = stripe_index - old_stripe_index + 1;
4897                 }
4898
4899         } else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
4900                                 BTRFS_BLOCK_GROUP_RAID6)) {
4901                 u64 tmp;
4902
4903                 if (bbio_ret && ((rw & REQ_WRITE) || mirror_num > 1)
4904                     && raid_map_ret) {
4905                         int i, rot;
4906
4907                         /* push stripe_nr back to the start of the full stripe */
4908                         stripe_nr = raid56_full_stripe_start;
4909                         do_div(stripe_nr, stripe_len);
4910
4911                         stripe_index = do_div(stripe_nr, nr_data_stripes(map));
4912
4913                         /* RAID[56] write or recovery. Return all stripes */
4914                         num_stripes = map->num_stripes;
4915                         max_errors = nr_parity_stripes(map);
4916
4917                         raid_map = kmalloc_array(num_stripes, sizeof(u64),
4918                                            GFP_NOFS);
4919                         if (!raid_map) {
4920                                 ret = -ENOMEM;
4921                                 goto out;
4922                         }
4923
4924                         /* Work out the disk rotation on this stripe-set */
4925                         tmp = stripe_nr;
4926                         rot = do_div(tmp, num_stripes);
4927
4928                         /* Fill in the logical address of each stripe */
4929                         tmp = stripe_nr * nr_data_stripes(map);
4930                         for (i = 0; i < nr_data_stripes(map); i++)
4931                                 raid_map[(i+rot) % num_stripes] =
4932                                         em->start + (tmp + i) * map->stripe_len;
4933
4934                         raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE;
4935                         if (map->type & BTRFS_BLOCK_GROUP_RAID6)
4936                                 raid_map[(i+rot+1) % num_stripes] =
4937                                         RAID6_Q_STRIPE;
4938
4939                         *length = map->stripe_len;
4940                         stripe_index = 0;
4941                         stripe_offset = 0;
4942                 } else {
4943                         /*
4944                          * Mirror #0 or #1 means the original data block.
4945                          * Mirror #2 is RAID5 parity block.
4946                          * Mirror #3 is RAID6 Q block.
4947                          */
4948                         stripe_index = do_div(stripe_nr, nr_data_stripes(map));
4949                         if (mirror_num > 1)
4950                                 stripe_index = nr_data_stripes(map) +
4951                                                 mirror_num - 2;
4952
4953                         /* We distribute the parity blocks across stripes */
4954                         tmp = stripe_nr + stripe_index;
4955                         stripe_index = do_div(tmp, map->num_stripes);
4956                 }
4957         } else {
4958                 /*
4959                  * after this do_div call, stripe_nr is the number of stripes
4960                  * on this device we have to walk to find the data, and
4961                  * stripe_index is the number of our device in the stripe array
4962                  */
4963                 stripe_index = do_div(stripe_nr, map->num_stripes);
4964                 mirror_num = stripe_index + 1;
4965         }
4966         BUG_ON(stripe_index >= map->num_stripes);
4967
4968         num_alloc_stripes = num_stripes;
4969         if (dev_replace_is_ongoing) {
4970                 if (rw & (REQ_WRITE | REQ_DISCARD))
4971                         num_alloc_stripes <<= 1;
4972                 if (rw & REQ_GET_READ_MIRRORS)
4973                         num_alloc_stripes++;
4974         }
4975         bbio = kzalloc(btrfs_bio_size(num_alloc_stripes), GFP_NOFS);
4976         if (!bbio) {
4977                 kfree(raid_map);
4978                 ret = -ENOMEM;
4979                 goto out;
4980         }
4981         atomic_set(&bbio->error, 0);
4982
4983         if (rw & REQ_DISCARD) {
4984                 int factor = 0;
4985                 int sub_stripes = 0;
4986                 u64 stripes_per_dev = 0;
4987                 u32 remaining_stripes = 0;
4988                 u32 last_stripe = 0;
4989
4990                 if (map->type &
4991                     (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID10)) {
4992                         if (map->type & BTRFS_BLOCK_GROUP_RAID0)
4993                                 sub_stripes = 1;
4994                         else
4995                                 sub_stripes = map->sub_stripes;
4996
4997                         factor = map->num_stripes / sub_stripes;
4998                         stripes_per_dev = div_u64_rem(stripe_nr_end -
4999                                                       stripe_nr_orig,
5000                                                       factor,
5001                                                       &remaining_stripes);
5002                         div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
5003                         last_stripe *= sub_stripes;
5004                 }
5005
5006                 for (i = 0; i < num_stripes; i++) {
5007                         bbio->stripes[i].physical =
5008                                 map->stripes[stripe_index].physical +
5009                                 stripe_offset + stripe_nr * map->stripe_len;
5010                         bbio->stripes[i].dev = map->stripes[stripe_index].dev;
5011
5012                         if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5013                                          BTRFS_BLOCK_GROUP_RAID10)) {
5014                                 bbio->stripes[i].length = stripes_per_dev *
5015                                                           map->stripe_len;
5016
5017                                 if (i / sub_stripes < remaining_stripes)
5018                                         bbio->stripes[i].length +=
5019                                                 map->stripe_len;
5020
5021                                 /*
5022                                  * Special for the first stripe and
5023                                  * the last stripe:
5024                                  *
5025                                  * |-------|...|-------|
5026                                  *     |----------|
5027                                  *    off     end_off
5028                                  */
5029                                 if (i < sub_stripes)
5030                                         bbio->stripes[i].length -=
5031                                                 stripe_offset;
5032
5033                                 if (stripe_index >= last_stripe &&
5034                                     stripe_index <= (last_stripe +
5035                                                      sub_stripes - 1))
5036                                         bbio->stripes[i].length -=
5037                                                 stripe_end_offset;
5038
5039                                 if (i == sub_stripes - 1)
5040                                         stripe_offset = 0;
5041                         } else
5042                                 bbio->stripes[i].length = *length;
5043
5044                         stripe_index++;
5045                         if (stripe_index == map->num_stripes) {
5046                                 /* This could only happen for RAID0/10 */
5047                                 stripe_index = 0;
5048                                 stripe_nr++;
5049                         }
5050                 }
5051         } else {
5052                 for (i = 0; i < num_stripes; i++) {
5053                         bbio->stripes[i].physical =
5054                                 map->stripes[stripe_index].physical +
5055                                 stripe_offset +
5056                                 stripe_nr * map->stripe_len;
5057                         bbio->stripes[i].dev =
5058                                 map->stripes[stripe_index].dev;
5059                         stripe_index++;
5060                 }
5061         }
5062
5063         if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS)) {
5064                 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
5065                                  BTRFS_BLOCK_GROUP_RAID10 |
5066                                  BTRFS_BLOCK_GROUP_RAID5 |
5067                                  BTRFS_BLOCK_GROUP_DUP)) {
5068                         max_errors = 1;
5069                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID6) {
5070                         max_errors = 2;
5071                 }
5072         }
5073
5074         if (dev_replace_is_ongoing && (rw & (REQ_WRITE | REQ_DISCARD)) &&
5075             dev_replace->tgtdev != NULL) {
5076                 int index_where_to_add;
5077                 u64 srcdev_devid = dev_replace->srcdev->devid;
5078
5079                 /*
5080                  * duplicate the write operations while the dev replace
5081                  * procedure is running. Since the copying of the old disk
5082                  * to the new disk takes place at run time while the
5083                  * filesystem is mounted writable, the regular write
5084                  * operations to the old disk have to be duplicated to go
5085                  * to the new disk as well.
5086                  * Note that device->missing is handled by the caller, and
5087                  * that the write to the old disk is already set up in the
5088                  * stripes array.
5089                  */
5090                 index_where_to_add = num_stripes;
5091                 for (i = 0; i < num_stripes; i++) {
5092                         if (bbio->stripes[i].dev->devid == srcdev_devid) {
5093                                 /* write to new disk, too */
5094                                 struct btrfs_bio_stripe *new =
5095                                         bbio->stripes + index_where_to_add;
5096                                 struct btrfs_bio_stripe *old =
5097                                         bbio->stripes + i;
5098
5099                                 new->physical = old->physical;
5100                                 new->length = old->length;
5101                                 new->dev = dev_replace->tgtdev;
5102                                 index_where_to_add++;
5103                                 max_errors++;
5104                         }
5105                 }
5106                 num_stripes = index_where_to_add;
5107         } else if (dev_replace_is_ongoing && (rw & REQ_GET_READ_MIRRORS) &&
5108                    dev_replace->tgtdev != NULL) {
5109                 u64 srcdev_devid = dev_replace->srcdev->devid;
5110                 int index_srcdev = 0;
5111                 int found = 0;
5112                 u64 physical_of_found = 0;
5113
5114                 /*
5115                  * During the dev-replace procedure, the target drive can
5116                  * also be used to read data in case it is needed to repair
5117                  * a corrupt block elsewhere. This is possible if the
5118                  * requested area is left of the left cursor. In this area,
5119                  * the target drive is a full copy of the source drive.
5120                  */
5121                 for (i = 0; i < num_stripes; i++) {
5122                         if (bbio->stripes[i].dev->devid == srcdev_devid) {
5123                                 /*
5124                                  * In case of DUP, in order to keep it
5125                                  * simple, only add the mirror with the
5126                                  * lowest physical address
5127                                  */
5128                                 if (found &&
5129                                     physical_of_found <=
5130                                      bbio->stripes[i].physical)
5131                                         continue;
5132                                 index_srcdev = i;
5133                                 found = 1;
5134                                 physical_of_found = bbio->stripes[i].physical;
5135                         }
5136                 }
5137                 if (found) {
5138                         u64 length = map->stripe_len;
5139
5140                         if (physical_of_found + length <=
5141                             dev_replace->cursor_left) {
5142                                 struct btrfs_bio_stripe *tgtdev_stripe =
5143                                         bbio->stripes + num_stripes;
5144
5145                                 tgtdev_stripe->physical = physical_of_found;
5146                                 tgtdev_stripe->length =
5147                                         bbio->stripes[index_srcdev].length;
5148                                 tgtdev_stripe->dev = dev_replace->tgtdev;
5149
5150                                 num_stripes++;
5151                         }
5152                 }
5153         }
5154
5155         *bbio_ret = bbio;
5156         bbio->num_stripes = num_stripes;
5157         bbio->max_errors = max_errors;
5158         bbio->mirror_num = mirror_num;
5159
5160         /*
5161          * this is the case that REQ_READ && dev_replace_is_ongoing &&
5162          * mirror_num == num_stripes + 1 && dev_replace target drive is
5163          * available as a mirror
5164          */
5165         if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) {
5166                 WARN_ON(num_stripes > 1);
5167                 bbio->stripes[0].dev = dev_replace->tgtdev;
5168                 bbio->stripes[0].physical = physical_to_patch_in_first_stripe;
5169                 bbio->mirror_num = map->num_stripes + 1;
5170         }
5171         if (raid_map) {
5172                 sort_parity_stripes(bbio, raid_map);
5173                 *raid_map_ret = raid_map;
5174         }
5175 out:
5176         if (dev_replace_is_ongoing)
5177                 btrfs_dev_replace_unlock(dev_replace);
5178         free_extent_map(em);
5179         return ret;
5180 }
5181
5182 int btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
5183                       u64 logical, u64 *length,
5184                       struct btrfs_bio **bbio_ret, int mirror_num)
5185 {
5186         return __btrfs_map_block(fs_info, rw, logical, length, bbio_ret,
5187                                  mirror_num, NULL);
5188 }
5189
5190 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
5191                      u64 chunk_start, u64 physical, u64 devid,
5192                      u64 **logical, int *naddrs, int *stripe_len)
5193 {
5194         struct extent_map_tree *em_tree = &map_tree->map_tree;
5195         struct extent_map *em;
5196         struct map_lookup *map;
5197         u64 *buf;
5198         u64 bytenr;
5199         u64 length;
5200         u64 stripe_nr;
5201         u64 rmap_len;
5202         int i, j, nr = 0;
5203
5204         read_lock(&em_tree->lock);
5205         em = lookup_extent_mapping(em_tree, chunk_start, 1);
5206         read_unlock(&em_tree->lock);
5207
5208         if (!em) {
5209                 printk(KERN_ERR "BTRFS: couldn't find em for chunk %Lu\n",
5210                        chunk_start);
5211                 return -EIO;
5212         }
5213
5214         if (em->start != chunk_start) {
5215                 printk(KERN_ERR "BTRFS: bad chunk start, em=%Lu, wanted=%Lu\n",
5216                        em->start, chunk_start);
5217                 free_extent_map(em);
5218                 return -EIO;
5219         }
5220         map = (struct map_lookup *)em->bdev;
5221
5222         length = em->len;
5223         rmap_len = map->stripe_len;
5224
5225         if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5226                 do_div(length, map->num_stripes / map->sub_stripes);
5227         else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5228                 do_div(length, map->num_stripes);
5229         else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
5230                               BTRFS_BLOCK_GROUP_RAID6)) {
5231                 do_div(length, nr_data_stripes(map));
5232                 rmap_len = map->stripe_len * nr_data_stripes(map);
5233         }
5234
5235         buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
5236         BUG_ON(!buf); /* -ENOMEM */
5237
5238         for (i = 0; i < map->num_stripes; i++) {
5239                 if (devid && map->stripes[i].dev->devid != devid)
5240                         continue;
5241                 if (map->stripes[i].physical > physical ||
5242                     map->stripes[i].physical + length <= physical)
5243                         continue;
5244
5245                 stripe_nr = physical - map->stripes[i].physical;
5246                 do_div(stripe_nr, map->stripe_len);
5247
5248                 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5249                         stripe_nr = stripe_nr * map->num_stripes + i;
5250                         do_div(stripe_nr, map->sub_stripes);
5251                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5252                         stripe_nr = stripe_nr * map->num_stripes + i;
5253                 } /* else if RAID[56], multiply by nr_data_stripes().
5254                    * Alternatively, just use rmap_len below instead of
5255                    * map->stripe_len */
5256
5257                 bytenr = chunk_start + stripe_nr * rmap_len;
5258                 WARN_ON(nr >= map->num_stripes);
5259                 for (j = 0; j < nr; j++) {
5260                         if (buf[j] == bytenr)
5261                                 break;
5262                 }
5263                 if (j == nr) {
5264                         WARN_ON(nr >= map->num_stripes);
5265                         buf[nr++] = bytenr;
5266                 }
5267         }
5268
5269         *logical = buf;
5270         *naddrs = nr;
5271         *stripe_len = rmap_len;
5272
5273         free_extent_map(em);
5274         return 0;
5275 }
5276
5277 static void btrfs_end_bio(struct bio *bio, int err)
5278 {
5279         struct btrfs_bio *bbio = bio->bi_private;
5280         struct btrfs_device *dev = bbio->stripes[0].dev;
5281         int is_orig_bio = 0;
5282
5283         if (err) {
5284                 atomic_inc(&bbio->error);
5285                 if (err == -EIO || err == -EREMOTEIO) {
5286                         unsigned int stripe_index =
5287                                 btrfs_io_bio(bio)->stripe_index;
5288
5289                         BUG_ON(stripe_index >= bbio->num_stripes);
5290                         dev = bbio->stripes[stripe_index].dev;
5291                         if (dev->bdev) {
5292                                 if (bio->bi_rw & WRITE)
5293                                         btrfs_dev_stat_inc(dev,
5294                                                 BTRFS_DEV_STAT_WRITE_ERRS);
5295                                 else
5296                                         btrfs_dev_stat_inc(dev,
5297                                                 BTRFS_DEV_STAT_READ_ERRS);
5298                                 if ((bio->bi_rw & WRITE_FLUSH) == WRITE_FLUSH)
5299                                         btrfs_dev_stat_inc(dev,
5300                                                 BTRFS_DEV_STAT_FLUSH_ERRS);
5301                                 btrfs_dev_stat_print_on_error(dev);
5302                         }
5303                 }
5304         }
5305
5306         if (bio == bbio->orig_bio)
5307                 is_orig_bio = 1;
5308
5309         btrfs_bio_counter_dec(bbio->fs_info);
5310
5311         if (atomic_dec_and_test(&bbio->stripes_pending)) {
5312                 if (!is_orig_bio) {
5313                         bio_put(bio);
5314                         bio = bbio->orig_bio;
5315                 }
5316
5317                 /*
5318                  * We have original bio now. So increment bi_remaining to
5319                  * account for it in endio
5320                  */
5321                 atomic_inc(&bio->bi_remaining);
5322
5323                 bio->bi_private = bbio->private;
5324                 bio->bi_end_io = bbio->end_io;
5325                 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
5326                 /* only send an error to the higher layers if it is
5327                  * beyond the tolerance of the btrfs bio
5328                  */
5329                 if (atomic_read(&bbio->error) > bbio->max_errors) {
5330                         err = -EIO;
5331                 } else {
5332                         /*
5333                          * this bio is actually up to date, we didn't
5334                          * go over the max number of errors
5335                          */
5336                         set_bit(BIO_UPTODATE, &bio->bi_flags);
5337                         err = 0;
5338                 }
5339                 kfree(bbio);
5340
5341                 bio_endio(bio, err);
5342         } else if (!is_orig_bio) {
5343                 bio_put(bio);
5344         }
5345 }
5346
5347 /*
5348  * see run_scheduled_bios for a description of why bios are collected for
5349  * async submit.
5350  *
5351  * This will add one bio to the pending list for a device and make sure
5352  * the work struct is scheduled.
5353  */
5354 static noinline void btrfs_schedule_bio(struct btrfs_root *root,
5355                                         struct btrfs_device *device,
5356                                         int rw, struct bio *bio)
5357 {
5358         int should_queue = 1;
5359         struct btrfs_pending_bios *pending_bios;
5360
5361         if (device->missing || !device->bdev) {
5362                 bio_endio(bio, -EIO);
5363                 return;
5364         }
5365
5366         /* don't bother with additional async steps for reads, right now */
5367         if (!(rw & REQ_WRITE)) {
5368                 bio_get(bio);
5369                 btrfsic_submit_bio(rw, bio);
5370                 bio_put(bio);
5371                 return;
5372         }
5373
5374         /*
5375          * nr_async_bios allows us to reliably return congestion to the
5376          * higher layers.  Otherwise, the async bio makes it appear we have
5377          * made progress against dirty pages when we've really just put it
5378          * on a queue for later
5379          */
5380         atomic_inc(&root->fs_info->nr_async_bios);
5381         WARN_ON(bio->bi_next);
5382         bio->bi_next = NULL;
5383         bio->bi_rw |= rw;
5384
5385         spin_lock(&device->io_lock);
5386         if (bio->bi_rw & REQ_SYNC)
5387                 pending_bios = &device->pending_sync_bios;
5388         else
5389                 pending_bios = &device->pending_bios;
5390
5391         if (pending_bios->tail)
5392                 pending_bios->tail->bi_next = bio;
5393
5394         pending_bios->tail = bio;
5395         if (!pending_bios->head)
5396                 pending_bios->head = bio;
5397         if (device->running_pending)
5398                 should_queue = 0;
5399
5400         spin_unlock(&device->io_lock);
5401
5402         if (should_queue)
5403                 btrfs_queue_work(root->fs_info->submit_workers,
5404                                  &device->work);
5405 }
5406
5407 static int bio_size_ok(struct block_device *bdev, struct bio *bio,
5408                        sector_t sector)
5409 {
5410         struct bio_vec *prev;
5411         struct request_queue *q = bdev_get_queue(bdev);
5412         unsigned int max_sectors = queue_max_sectors(q);
5413         struct bvec_merge_data bvm = {
5414                 .bi_bdev = bdev,
5415                 .bi_sector = sector,
5416                 .bi_rw = bio->bi_rw,
5417         };
5418
5419         if (WARN_ON(bio->bi_vcnt == 0))
5420                 return 1;
5421
5422         prev = &bio->bi_io_vec[bio->bi_vcnt - 1];
5423         if (bio_sectors(bio) > max_sectors)
5424                 return 0;
5425
5426         if (!q->merge_bvec_fn)
5427                 return 1;
5428
5429         bvm.bi_size = bio->bi_iter.bi_size - prev->bv_len;
5430         if (q->merge_bvec_fn(q, &bvm, prev) < prev->bv_len)
5431                 return 0;
5432         return 1;
5433 }
5434
5435 static void submit_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio,
5436                               struct bio *bio, u64 physical, int dev_nr,
5437                               int rw, int async)
5438 {
5439         struct btrfs_device *dev = bbio->stripes[dev_nr].dev;
5440
5441         bio->bi_private = bbio;
5442         btrfs_io_bio(bio)->stripe_index = dev_nr;
5443         bio->bi_end_io = btrfs_end_bio;
5444         bio->bi_iter.bi_sector = physical >> 9;
5445 #ifdef DEBUG
5446         {
5447                 struct rcu_string *name;
5448
5449                 rcu_read_lock();
5450                 name = rcu_dereference(dev->name);
5451                 pr_debug("btrfs_map_bio: rw %d, sector=%llu, dev=%lu "
5452                          "(%s id %llu), size=%u\n", rw,
5453                          (u64)bio->bi_sector, (u_long)dev->bdev->bd_dev,
5454                          name->str, dev->devid, bio->bi_size);
5455                 rcu_read_unlock();
5456         }
5457 #endif
5458         bio->bi_bdev = dev->bdev;
5459
5460         btrfs_bio_counter_inc_noblocked(root->fs_info);
5461
5462         if (async)
5463                 btrfs_schedule_bio(root, dev, rw, bio);
5464         else
5465                 btrfsic_submit_bio(rw, bio);
5466 }
5467
5468 static int breakup_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio,
5469                               struct bio *first_bio, struct btrfs_device *dev,
5470                               int dev_nr, int rw, int async)
5471 {
5472         struct bio_vec *bvec = first_bio->bi_io_vec;
5473         struct bio *bio;
5474         int nr_vecs = bio_get_nr_vecs(dev->bdev);
5475         u64 physical = bbio->stripes[dev_nr].physical;
5476
5477 again:
5478         bio = btrfs_bio_alloc(dev->bdev, physical >> 9, nr_vecs, GFP_NOFS);
5479         if (!bio)
5480                 return -ENOMEM;
5481
5482         while (bvec <= (first_bio->bi_io_vec + first_bio->bi_vcnt - 1)) {
5483                 if (bio_add_page(bio, bvec->bv_page, bvec->bv_len,
5484                                  bvec->bv_offset) < bvec->bv_len) {
5485                         u64 len = bio->bi_iter.bi_size;
5486
5487                         atomic_inc(&bbio->stripes_pending);
5488                         submit_stripe_bio(root, bbio, bio, physical, dev_nr,
5489                                           rw, async);
5490                         physical += len;
5491                         goto again;
5492                 }
5493                 bvec++;
5494         }
5495
5496         submit_stripe_bio(root, bbio, bio, physical, dev_nr, rw, async);
5497         return 0;
5498 }
5499
5500 static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical)
5501 {
5502         atomic_inc(&bbio->error);
5503         if (atomic_dec_and_test(&bbio->stripes_pending)) {
5504                 bio->bi_private = bbio->private;
5505                 bio->bi_end_io = bbio->end_io;
5506                 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
5507                 bio->bi_iter.bi_sector = logical >> 9;
5508                 kfree(bbio);
5509                 bio_endio(bio, -EIO);
5510         }
5511 }
5512
5513 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
5514                   int mirror_num, int async_submit)
5515 {
5516         struct btrfs_device *dev;
5517         struct bio *first_bio = bio;
5518         u64 logical = (u64)bio->bi_iter.bi_sector << 9;
5519         u64 length = 0;
5520         u64 map_length;
5521         u64 *raid_map = NULL;
5522         int ret;
5523         int dev_nr = 0;
5524         int total_devs = 1;
5525         struct btrfs_bio *bbio = NULL;
5526
5527         length = bio->bi_iter.bi_size;
5528         map_length = length;
5529
5530         btrfs_bio_counter_inc_blocked(root->fs_info);
5531         ret = __btrfs_map_block(root->fs_info, rw, logical, &map_length, &bbio,
5532                               mirror_num, &raid_map);
5533         if (ret) {
5534                 btrfs_bio_counter_dec(root->fs_info);
5535                 return ret;
5536         }
5537
5538         total_devs = bbio->num_stripes;
5539         bbio->orig_bio = first_bio;
5540         bbio->private = first_bio->bi_private;
5541         bbio->end_io = first_bio->bi_end_io;
5542         bbio->fs_info = root->fs_info;
5543         atomic_set(&bbio->stripes_pending, bbio->num_stripes);
5544
5545         if (raid_map) {
5546                 /* In this case, map_length has been set to the length of
5547                    a single stripe; not the whole write */
5548                 if (rw & WRITE) {
5549                         ret = raid56_parity_write(root, bio, bbio,
5550                                                   raid_map, map_length);
5551                 } else {
5552                         ret = raid56_parity_recover(root, bio, bbio,
5553                                                     raid_map, map_length,
5554                                                     mirror_num);
5555                 }
5556                 /*
5557                  * FIXME, replace dosen't support raid56 yet, please fix
5558                  * it in the future.
5559                  */
5560                 btrfs_bio_counter_dec(root->fs_info);
5561                 return ret;
5562         }
5563
5564         if (map_length < length) {
5565                 btrfs_crit(root->fs_info, "mapping failed logical %llu bio len %llu len %llu",
5566                         logical, length, map_length);
5567                 BUG();
5568         }
5569
5570         while (dev_nr < total_devs) {
5571                 dev = bbio->stripes[dev_nr].dev;
5572                 if (!dev || !dev->bdev || (rw & WRITE && !dev->writeable)) {
5573                         bbio_error(bbio, first_bio, logical);
5574                         dev_nr++;
5575                         continue;
5576                 }
5577
5578                 /*
5579                  * Check and see if we're ok with this bio based on it's size
5580                  * and offset with the given device.
5581                  */
5582                 if (!bio_size_ok(dev->bdev, first_bio,
5583                                  bbio->stripes[dev_nr].physical >> 9)) {
5584                         ret = breakup_stripe_bio(root, bbio, first_bio, dev,
5585                                                  dev_nr, rw, async_submit);
5586                         BUG_ON(ret);
5587                         dev_nr++;
5588                         continue;
5589                 }
5590
5591                 if (dev_nr < total_devs - 1) {
5592                         bio = btrfs_bio_clone(first_bio, GFP_NOFS);
5593                         BUG_ON(!bio); /* -ENOMEM */
5594                 } else {
5595                         bio = first_bio;
5596                 }
5597
5598                 submit_stripe_bio(root, bbio, bio,
5599                                   bbio->stripes[dev_nr].physical, dev_nr, rw,
5600                                   async_submit);
5601                 dev_nr++;
5602         }
5603         btrfs_bio_counter_dec(root->fs_info);
5604         return 0;
5605 }
5606
5607 struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid,
5608                                        u8 *uuid, u8 *fsid)
5609 {
5610         struct btrfs_device *device;
5611         struct btrfs_fs_devices *cur_devices;
5612
5613         cur_devices = fs_info->fs_devices;
5614         while (cur_devices) {
5615                 if (!fsid ||
5616                     !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
5617                         device = __find_device(&cur_devices->devices,
5618                                                devid, uuid);
5619                         if (device)
5620                                 return device;
5621                 }
5622                 cur_devices = cur_devices->seed;
5623         }
5624         return NULL;
5625 }
5626
5627 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
5628                                             u64 devid, u8 *dev_uuid)
5629 {
5630         struct btrfs_device *device;
5631         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
5632
5633         device = btrfs_alloc_device(NULL, &devid, dev_uuid);
5634         if (IS_ERR(device))
5635                 return NULL;
5636
5637         list_add(&device->dev_list, &fs_devices->devices);
5638         device->fs_devices = fs_devices;
5639         fs_devices->num_devices++;
5640
5641         device->missing = 1;
5642         fs_devices->missing_devices++;
5643
5644         return device;
5645 }
5646
5647 /**
5648  * btrfs_alloc_device - allocate struct btrfs_device
5649  * @fs_info:    used only for generating a new devid, can be NULL if
5650  *              devid is provided (i.e. @devid != NULL).
5651  * @devid:      a pointer to devid for this device.  If NULL a new devid
5652  *              is generated.
5653  * @uuid:       a pointer to UUID for this device.  If NULL a new UUID
5654  *              is generated.
5655  *
5656  * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
5657  * on error.  Returned struct is not linked onto any lists and can be
5658  * destroyed with kfree() right away.
5659  */
5660 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
5661                                         const u64 *devid,
5662                                         const u8 *uuid)
5663 {
5664         struct btrfs_device *dev;
5665         u64 tmp;
5666
5667         if (WARN_ON(!devid && !fs_info))
5668                 return ERR_PTR(-EINVAL);
5669
5670         dev = __alloc_device();
5671         if (IS_ERR(dev))
5672                 return dev;
5673
5674         if (devid)
5675                 tmp = *devid;
5676         else {
5677                 int ret;
5678
5679                 ret = find_next_devid(fs_info, &tmp);
5680                 if (ret) {
5681                         kfree(dev);
5682                         return ERR_PTR(ret);
5683                 }
5684         }
5685         dev->devid = tmp;
5686
5687         if (uuid)
5688                 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE);
5689         else
5690                 generate_random_uuid(dev->uuid);
5691
5692         btrfs_init_work(&dev->work, pending_bios_fn, NULL, NULL);
5693
5694         return dev;
5695 }
5696
5697 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
5698                           struct extent_buffer *leaf,
5699                           struct btrfs_chunk *chunk)
5700 {
5701         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
5702         struct map_lookup *map;
5703         struct extent_map *em;
5704         u64 logical;
5705         u64 length;
5706         u64 devid;
5707         u8 uuid[BTRFS_UUID_SIZE];
5708         int num_stripes;
5709         int ret;
5710         int i;
5711
5712         logical = key->offset;
5713         length = btrfs_chunk_length(leaf, chunk);
5714
5715         read_lock(&map_tree->map_tree.lock);
5716         em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
5717         read_unlock(&map_tree->map_tree.lock);
5718
5719         /* already mapped? */
5720         if (em && em->start <= logical && em->start + em->len > logical) {
5721                 free_extent_map(em);
5722                 return 0;
5723         } else if (em) {
5724                 free_extent_map(em);
5725         }
5726
5727         em = alloc_extent_map();
5728         if (!em)
5729                 return -ENOMEM;
5730         num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
5731         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
5732         if (!map) {
5733                 free_extent_map(em);
5734                 return -ENOMEM;
5735         }
5736
5737         em->bdev = (struct block_device *)map;
5738         em->start = logical;
5739         em->len = length;
5740         em->orig_start = 0;
5741         em->block_start = 0;
5742         em->block_len = em->len;
5743
5744         map->num_stripes = num_stripes;
5745         map->io_width = btrfs_chunk_io_width(leaf, chunk);
5746         map->io_align = btrfs_chunk_io_align(leaf, chunk);
5747         map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
5748         map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
5749         map->type = btrfs_chunk_type(leaf, chunk);
5750         map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
5751         for (i = 0; i < num_stripes; i++) {
5752                 map->stripes[i].physical =
5753                         btrfs_stripe_offset_nr(leaf, chunk, i);
5754                 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
5755                 read_extent_buffer(leaf, uuid, (unsigned long)
5756                                    btrfs_stripe_dev_uuid_nr(chunk, i),
5757                                    BTRFS_UUID_SIZE);
5758                 map->stripes[i].dev = btrfs_find_device(root->fs_info, devid,
5759                                                         uuid, NULL);
5760                 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
5761                         kfree(map);
5762                         free_extent_map(em);
5763                         return -EIO;
5764                 }
5765                 if (!map->stripes[i].dev) {
5766                         map->stripes[i].dev =
5767                                 add_missing_dev(root, devid, uuid);
5768                         if (!map->stripes[i].dev) {
5769                                 kfree(map);
5770                                 free_extent_map(em);
5771                                 return -EIO;
5772                         }
5773                 }
5774                 map->stripes[i].dev->in_fs_metadata = 1;
5775         }
5776
5777         write_lock(&map_tree->map_tree.lock);
5778         ret = add_extent_mapping(&map_tree->map_tree, em, 0);
5779         write_unlock(&map_tree->map_tree.lock);
5780         BUG_ON(ret); /* Tree corruption */
5781         free_extent_map(em);
5782
5783         return 0;
5784 }
5785
5786 static void fill_device_from_item(struct extent_buffer *leaf,
5787                                  struct btrfs_dev_item *dev_item,
5788                                  struct btrfs_device *device)
5789 {
5790         unsigned long ptr;
5791
5792         device->devid = btrfs_device_id(leaf, dev_item);
5793         device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
5794         device->total_bytes = device->disk_total_bytes;
5795         device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
5796         device->type = btrfs_device_type(leaf, dev_item);
5797         device->io_align = btrfs_device_io_align(leaf, dev_item);
5798         device->io_width = btrfs_device_io_width(leaf, dev_item);
5799         device->sector_size = btrfs_device_sector_size(leaf, dev_item);
5800         WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
5801         device->is_tgtdev_for_dev_replace = 0;
5802
5803         ptr = btrfs_device_uuid(dev_item);
5804         read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
5805 }
5806
5807 static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
5808 {
5809         struct btrfs_fs_devices *fs_devices;
5810         int ret;
5811
5812         BUG_ON(!mutex_is_locked(&uuid_mutex));
5813
5814         fs_devices = root->fs_info->fs_devices->seed;
5815         while (fs_devices) {
5816                 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
5817                         ret = 0;
5818                         goto out;
5819                 }
5820                 fs_devices = fs_devices->seed;
5821         }
5822
5823         fs_devices = find_fsid(fsid);
5824         if (!fs_devices) {
5825                 ret = -ENOENT;
5826                 goto out;
5827         }
5828
5829         fs_devices = clone_fs_devices(fs_devices);
5830         if (IS_ERR(fs_devices)) {
5831                 ret = PTR_ERR(fs_devices);
5832                 goto out;
5833         }
5834
5835         ret = __btrfs_open_devices(fs_devices, FMODE_READ,
5836                                    root->fs_info->bdev_holder);
5837         if (ret) {
5838                 free_fs_devices(fs_devices);
5839                 goto out;
5840         }
5841
5842         if (!fs_devices->seeding) {
5843                 __btrfs_close_devices(fs_devices);
5844                 free_fs_devices(fs_devices);
5845                 ret = -EINVAL;
5846                 goto out;
5847         }
5848
5849         fs_devices->seed = root->fs_info->fs_devices->seed;
5850         root->fs_info->fs_devices->seed = fs_devices;
5851 out:
5852         return ret;
5853 }
5854
5855 static int read_one_dev(struct btrfs_root *root,
5856                         struct extent_buffer *leaf,
5857                         struct btrfs_dev_item *dev_item)
5858 {
5859         struct btrfs_device *device;
5860         u64 devid;
5861         int ret;
5862         u8 fs_uuid[BTRFS_UUID_SIZE];
5863         u8 dev_uuid[BTRFS_UUID_SIZE];
5864
5865         devid = btrfs_device_id(leaf, dev_item);
5866         read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
5867                            BTRFS_UUID_SIZE);
5868         read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
5869                            BTRFS_UUID_SIZE);
5870
5871         if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
5872                 ret = open_seed_devices(root, fs_uuid);
5873                 if (ret && !btrfs_test_opt(root, DEGRADED))
5874                         return ret;
5875         }
5876
5877         device = btrfs_find_device(root->fs_info, devid, dev_uuid, fs_uuid);
5878         if (!device || !device->bdev) {
5879                 if (!btrfs_test_opt(root, DEGRADED))
5880                         return -EIO;
5881
5882                 if (!device) {
5883                         btrfs_warn(root->fs_info, "devid %llu missing", devid);
5884                         device = add_missing_dev(root, devid, dev_uuid);
5885                         if (!device)
5886                                 return -ENOMEM;
5887                 } else if (!device->missing) {
5888                         /*
5889                          * this happens when a device that was properly setup
5890                          * in the device info lists suddenly goes bad.
5891                          * device->bdev is NULL, and so we have to set
5892                          * device->missing to one here
5893                          */
5894                         root->fs_info->fs_devices->missing_devices++;
5895                         device->missing = 1;
5896                 }
5897         }
5898
5899         if (device->fs_devices != root->fs_info->fs_devices) {
5900                 BUG_ON(device->writeable);
5901                 if (device->generation !=
5902                     btrfs_device_generation(leaf, dev_item))
5903                         return -EINVAL;
5904         }
5905
5906         fill_device_from_item(leaf, dev_item, device);
5907         device->in_fs_metadata = 1;
5908         if (device->writeable && !device->is_tgtdev_for_dev_replace) {
5909                 device->fs_devices->total_rw_bytes += device->total_bytes;
5910                 spin_lock(&root->fs_info->free_chunk_lock);
5911                 root->fs_info->free_chunk_space += device->total_bytes -
5912                         device->bytes_used;
5913                 spin_unlock(&root->fs_info->free_chunk_lock);
5914         }
5915         ret = 0;
5916         return ret;
5917 }
5918
5919 int btrfs_read_sys_array(struct btrfs_root *root)
5920 {
5921         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
5922         struct extent_buffer *sb;
5923         struct btrfs_disk_key *disk_key;
5924         struct btrfs_chunk *chunk;
5925         u8 *ptr;
5926         unsigned long sb_ptr;
5927         int ret = 0;
5928         u32 num_stripes;
5929         u32 array_size;
5930         u32 len = 0;
5931         u32 cur;
5932         struct btrfs_key key;
5933
5934         sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
5935                                           BTRFS_SUPER_INFO_SIZE);
5936         if (!sb)
5937                 return -ENOMEM;
5938         btrfs_set_buffer_uptodate(sb);
5939         btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
5940         /*
5941          * The sb extent buffer is artifical and just used to read the system array.
5942          * btrfs_set_buffer_uptodate() call does not properly mark all it's
5943          * pages up-to-date when the page is larger: extent does not cover the
5944          * whole page and consequently check_page_uptodate does not find all
5945          * the page's extents up-to-date (the hole beyond sb),
5946          * write_extent_buffer then triggers a WARN_ON.
5947          *
5948          * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
5949          * but sb spans only this function. Add an explicit SetPageUptodate call
5950          * to silence the warning eg. on PowerPC 64.
5951          */
5952         if (PAGE_CACHE_SIZE > BTRFS_SUPER_INFO_SIZE)
5953                 SetPageUptodate(sb->pages[0]);
5954
5955         write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
5956         array_size = btrfs_super_sys_array_size(super_copy);
5957
5958         ptr = super_copy->sys_chunk_array;
5959         sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
5960         cur = 0;
5961
5962         while (cur < array_size) {
5963                 disk_key = (struct btrfs_disk_key *)ptr;
5964                 btrfs_disk_key_to_cpu(&key, disk_key);
5965
5966                 len = sizeof(*disk_key); ptr += len;
5967                 sb_ptr += len;
5968                 cur += len;
5969
5970                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
5971                         chunk = (struct btrfs_chunk *)sb_ptr;
5972                         ret = read_one_chunk(root, &key, sb, chunk);
5973                         if (ret)
5974                                 break;
5975                         num_stripes = btrfs_chunk_num_stripes(sb, chunk);
5976                         len = btrfs_chunk_item_size(num_stripes);
5977                 } else {
5978                         ret = -EIO;
5979                         break;
5980                 }
5981                 ptr += len;
5982                 sb_ptr += len;
5983                 cur += len;
5984         }
5985         free_extent_buffer(sb);
5986         return ret;
5987 }
5988
5989 int btrfs_read_chunk_tree(struct btrfs_root *root)
5990 {
5991         struct btrfs_path *path;
5992         struct extent_buffer *leaf;
5993         struct btrfs_key key;
5994         struct btrfs_key found_key;
5995         int ret;
5996         int slot;
5997
5998         root = root->fs_info->chunk_root;
5999
6000         path = btrfs_alloc_path();
6001         if (!path)
6002                 return -ENOMEM;
6003
6004         mutex_lock(&uuid_mutex);
6005         lock_chunks(root);
6006
6007         /*
6008          * Read all device items, and then all the chunk items. All
6009          * device items are found before any chunk item (their object id
6010          * is smaller than the lowest possible object id for a chunk
6011          * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
6012          */
6013         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
6014         key.offset = 0;
6015         key.type = 0;
6016         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6017         if (ret < 0)
6018                 goto error;
6019         while (1) {
6020                 leaf = path->nodes[0];
6021                 slot = path->slots[0];
6022                 if (slot >= btrfs_header_nritems(leaf)) {
6023                         ret = btrfs_next_leaf(root, path);
6024                         if (ret == 0)
6025                                 continue;
6026                         if (ret < 0)
6027                                 goto error;
6028                         break;
6029                 }
6030                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6031                 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
6032                         struct btrfs_dev_item *dev_item;
6033                         dev_item = btrfs_item_ptr(leaf, slot,
6034                                                   struct btrfs_dev_item);
6035                         ret = read_one_dev(root, leaf, dev_item);
6036                         if (ret)
6037                                 goto error;
6038                 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
6039                         struct btrfs_chunk *chunk;
6040                         chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
6041                         ret = read_one_chunk(root, &found_key, leaf, chunk);
6042                         if (ret)
6043                                 goto error;
6044                 }
6045                 path->slots[0]++;
6046         }
6047         ret = 0;
6048 error:
6049         unlock_chunks(root);
6050         mutex_unlock(&uuid_mutex);
6051
6052         btrfs_free_path(path);
6053         return ret;
6054 }
6055
6056 void btrfs_init_devices_late(struct btrfs_fs_info *fs_info)
6057 {
6058         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6059         struct btrfs_device *device;
6060
6061         mutex_lock(&fs_devices->device_list_mutex);
6062         list_for_each_entry(device, &fs_devices->devices, dev_list)
6063                 device->dev_root = fs_info->dev_root;
6064         mutex_unlock(&fs_devices->device_list_mutex);
6065 }
6066
6067 static void __btrfs_reset_dev_stats(struct btrfs_device *dev)
6068 {
6069         int i;
6070
6071         for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6072                 btrfs_dev_stat_reset(dev, i);
6073 }
6074
6075 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
6076 {
6077         struct btrfs_key key;
6078         struct btrfs_key found_key;
6079         struct btrfs_root *dev_root = fs_info->dev_root;
6080         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6081         struct extent_buffer *eb;
6082         int slot;
6083         int ret = 0;
6084         struct btrfs_device *device;
6085         struct btrfs_path *path = NULL;
6086         int i;
6087
6088         path = btrfs_alloc_path();
6089         if (!path) {
6090                 ret = -ENOMEM;
6091                 goto out;
6092         }
6093
6094         mutex_lock(&fs_devices->device_list_mutex);
6095         list_for_each_entry(device, &fs_devices->devices, dev_list) {
6096                 int item_size;
6097                 struct btrfs_dev_stats_item *ptr;
6098
6099                 key.objectid = 0;
6100                 key.type = BTRFS_DEV_STATS_KEY;
6101                 key.offset = device->devid;
6102                 ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
6103                 if (ret) {
6104                         __btrfs_reset_dev_stats(device);
6105                         device->dev_stats_valid = 1;
6106                         btrfs_release_path(path);
6107                         continue;
6108                 }
6109                 slot = path->slots[0];
6110                 eb = path->nodes[0];
6111                 btrfs_item_key_to_cpu(eb, &found_key, slot);
6112                 item_size = btrfs_item_size_nr(eb, slot);
6113
6114                 ptr = btrfs_item_ptr(eb, slot,
6115                                      struct btrfs_dev_stats_item);
6116
6117                 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
6118                         if (item_size >= (1 + i) * sizeof(__le64))
6119                                 btrfs_dev_stat_set(device, i,
6120                                         btrfs_dev_stats_value(eb, ptr, i));
6121                         else
6122                                 btrfs_dev_stat_reset(device, i);
6123                 }
6124
6125                 device->dev_stats_valid = 1;
6126                 btrfs_dev_stat_print_on_load(device);
6127                 btrfs_release_path(path);
6128         }
6129         mutex_unlock(&fs_devices->device_list_mutex);
6130
6131 out:
6132         btrfs_free_path(path);
6133         return ret < 0 ? ret : 0;
6134 }
6135
6136 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
6137                                 struct btrfs_root *dev_root,
6138                                 struct btrfs_device *device)
6139 {
6140         struct btrfs_path *path;
6141         struct btrfs_key key;
6142         struct extent_buffer *eb;
6143         struct btrfs_dev_stats_item *ptr;
6144         int ret;
6145         int i;
6146
6147         key.objectid = 0;
6148         key.type = BTRFS_DEV_STATS_KEY;
6149         key.offset = device->devid;
6150
6151         path = btrfs_alloc_path();
6152         BUG_ON(!path);
6153         ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
6154         if (ret < 0) {
6155                 printk_in_rcu(KERN_WARNING "BTRFS: "
6156                         "error %d while searching for dev_stats item for device %s!\n",
6157                               ret, rcu_str_deref(device->name));
6158                 goto out;
6159         }
6160
6161         if (ret == 0 &&
6162             btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
6163                 /* need to delete old one and insert a new one */
6164                 ret = btrfs_del_item(trans, dev_root, path);
6165                 if (ret != 0) {
6166                         printk_in_rcu(KERN_WARNING "BTRFS: "
6167                                 "delete too small dev_stats item for device %s failed %d!\n",
6168                                       rcu_str_deref(device->name), ret);
6169                         goto out;
6170                 }
6171                 ret = 1;
6172         }
6173
6174         if (ret == 1) {
6175                 /* need to insert a new item */
6176                 btrfs_release_path(path);
6177                 ret = btrfs_insert_empty_item(trans, dev_root, path,
6178                                               &key, sizeof(*ptr));
6179                 if (ret < 0) {
6180                         printk_in_rcu(KERN_WARNING "BTRFS: "
6181                                           "insert dev_stats item for device %s failed %d!\n",
6182                                       rcu_str_deref(device->name), ret);
6183                         goto out;
6184                 }
6185         }
6186
6187         eb = path->nodes[0];
6188         ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
6189         for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6190                 btrfs_set_dev_stats_value(eb, ptr, i,
6191                                           btrfs_dev_stat_read(device, i));
6192         btrfs_mark_buffer_dirty(eb);
6193
6194 out:
6195         btrfs_free_path(path);
6196         return ret;
6197 }
6198
6199 /*
6200  * called from commit_transaction. Writes all changed device stats to disk.
6201  */
6202 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans,
6203                         struct btrfs_fs_info *fs_info)
6204 {
6205         struct btrfs_root *dev_root = fs_info->dev_root;
6206         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6207         struct btrfs_device *device;
6208         int ret = 0;
6209
6210         mutex_lock(&fs_devices->device_list_mutex);
6211         list_for_each_entry(device, &fs_devices->devices, dev_list) {
6212                 if (!device->dev_stats_valid || !device->dev_stats_dirty)
6213                         continue;
6214
6215                 ret = update_dev_stat_item(trans, dev_root, device);
6216                 if (!ret)
6217                         device->dev_stats_dirty = 0;
6218         }
6219         mutex_unlock(&fs_devices->device_list_mutex);
6220
6221         return ret;
6222 }
6223
6224 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
6225 {
6226         btrfs_dev_stat_inc(dev, index);
6227         btrfs_dev_stat_print_on_error(dev);
6228 }
6229
6230 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
6231 {
6232         if (!dev->dev_stats_valid)
6233                 return;
6234         printk_ratelimited_in_rcu(KERN_ERR "BTRFS: "
6235                            "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
6236                            rcu_str_deref(dev->name),
6237                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
6238                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
6239                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
6240                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
6241                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
6242 }
6243
6244 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
6245 {
6246         int i;
6247
6248         for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6249                 if (btrfs_dev_stat_read(dev, i) != 0)
6250                         break;
6251         if (i == BTRFS_DEV_STAT_VALUES_MAX)
6252                 return; /* all values == 0, suppress message */
6253
6254         printk_in_rcu(KERN_INFO "BTRFS: "
6255                    "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
6256                rcu_str_deref(dev->name),
6257                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
6258                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
6259                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
6260                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
6261                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
6262 }
6263
6264 int btrfs_get_dev_stats(struct btrfs_root *root,
6265                         struct btrfs_ioctl_get_dev_stats *stats)
6266 {
6267         struct btrfs_device *dev;
6268         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
6269         int i;
6270
6271         mutex_lock(&fs_devices->device_list_mutex);
6272         dev = btrfs_find_device(root->fs_info, stats->devid, NULL, NULL);
6273         mutex_unlock(&fs_devices->device_list_mutex);
6274
6275         if (!dev) {
6276                 btrfs_warn(root->fs_info, "get dev_stats failed, device not found");
6277                 return -ENODEV;
6278         } else if (!dev->dev_stats_valid) {
6279                 btrfs_warn(root->fs_info, "get dev_stats failed, not yet valid");
6280                 return -ENODEV;
6281         } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
6282                 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
6283                         if (stats->nr_items > i)
6284                                 stats->values[i] =
6285                                         btrfs_dev_stat_read_and_reset(dev, i);
6286                         else
6287                                 btrfs_dev_stat_reset(dev, i);
6288                 }
6289         } else {
6290                 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6291                         if (stats->nr_items > i)
6292                                 stats->values[i] = btrfs_dev_stat_read(dev, i);
6293         }
6294         if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
6295                 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
6296         return 0;
6297 }
6298
6299 int btrfs_scratch_superblock(struct btrfs_device *device)
6300 {
6301         struct buffer_head *bh;
6302         struct btrfs_super_block *disk_super;
6303
6304         bh = btrfs_read_dev_super(device->bdev);
6305         if (!bh)
6306                 return -EINVAL;
6307         disk_super = (struct btrfs_super_block *)bh->b_data;
6308
6309         memset(&disk_super->magic, 0, sizeof(disk_super->magic));
6310         set_buffer_dirty(bh);
6311         sync_dirty_buffer(bh);
6312         brelse(bh);
6313
6314         return 0;
6315 }