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 #include "sysfs.h"
  44
  45 const struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = {
  46         [BTRFS_RAID_RAID10] = {
  47                 .sub_stripes    = 2,
  48                 .dev_stripes    = 1,
  49                 .devs_max       = 0,    /* 0 == as many as possible */
  50                 .devs_min       = 4,
  51                 .tolerated_failures = 1,
  52                 .devs_increment = 2,
  53                 .ncopies        = 2,
  54         },
  55         [BTRFS_RAID_RAID1] = {
  56                 .sub_stripes    = 1,
  57                 .dev_stripes    = 1,
  58                 .devs_max       = 2,
  59                 .devs_min       = 2,
  60                 .tolerated_failures = 1,
  61                 .devs_increment = 2,
  62                 .ncopies        = 2,
  63         },
  64         [BTRFS_RAID_DUP] = {
  65                 .sub_stripes    = 1,
  66                 .dev_stripes    = 2,
  67                 .devs_max       = 1,
  68                 .devs_min       = 1,
  69                 .tolerated_failures = 0,
  70                 .devs_increment = 1,
  71                 .ncopies        = 2,
  72         },
  73         [BTRFS_RAID_RAID0] = {
  74                 .sub_stripes    = 1,
  75                 .dev_stripes    = 1,
  76                 .devs_max       = 0,
  77                 .devs_min       = 2,
  78                 .tolerated_failures = 0,
  79                 .devs_increment = 1,
  80                 .ncopies        = 1,
  81         },
  82         [BTRFS_RAID_SINGLE] = {
  83                 .sub_stripes    = 1,
  84                 .dev_stripes    = 1,
  85                 .devs_max       = 1,
  86                 .devs_min       = 1,
  87                 .tolerated_failures = 0,
  88                 .devs_increment = 1,
  89                 .ncopies        = 1,
  90         },
  91         [BTRFS_RAID_RAID5] = {
  92                 .sub_stripes    = 1,
  93                 .dev_stripes    = 1,
  94                 .devs_max       = 0,
  95                 .devs_min       = 2,
  96                 .tolerated_failures = 1,
  97                 .devs_increment = 1,
  98                 .ncopies        = 2,
  99         },
 100         [BTRFS_RAID_RAID6] = {
 101                 .sub_stripes    = 1,
 102                 .dev_stripes    = 1,
 103                 .devs_max       = 0,
 104                 .devs_min       = 3,
 105                 .tolerated_failures = 2,
 106                 .devs_increment = 1,
 107                 .ncopies        = 3,
 108         },
 109 };
 110
 111 const u64 const btrfs_raid_group[BTRFS_NR_RAID_TYPES] = {
 112         [BTRFS_RAID_RAID10] = BTRFS_BLOCK_GROUP_RAID10,
 113         [BTRFS_RAID_RAID1]  = BTRFS_BLOCK_GROUP_RAID1,
 114         [BTRFS_RAID_DUP]    = BTRFS_BLOCK_GROUP_DUP,
 115         [BTRFS_RAID_RAID0]  = BTRFS_BLOCK_GROUP_RAID0,
 116         [BTRFS_RAID_SINGLE] = 0,
 117         [BTRFS_RAID_RAID5]  = BTRFS_BLOCK_GROUP_RAID5,
 118         [BTRFS_RAID_RAID6]  = BTRFS_BLOCK_GROUP_RAID6,
 119 };
 120
 121 static int init_first_rw_device(struct btrfs_trans_handle *trans,
 122                                 struct btrfs_root *root,
 123                                 struct btrfs_device *device);
 124 static int btrfs_relocate_sys_chunks(struct btrfs_root *root);
 125 static void __btrfs_reset_dev_stats(struct btrfs_device *dev);
 126 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev);
 127 static void btrfs_dev_stat_print_on_load(struct btrfs_device *device);
 128
 129 DEFINE_MUTEX(uuid_mutex);
 130 static LIST_HEAD(fs_uuids);
 131 struct list_head *btrfs_get_fs_uuids(void)
 132 {
 133         return &fs_uuids;
 134 }
 135
 136 static struct btrfs_fs_devices *__alloc_fs_devices(void)
 137 {
 138         struct btrfs_fs_devices *fs_devs;
 139
 140         fs_devs = kzalloc(sizeof(*fs_devs), GFP_NOFS);
 141         if (!fs_devs)
 142                 return ERR_PTR(-ENOMEM);
 143
 144         mutex_init(&fs_devs->device_list_mutex);
 145
 146         INIT_LIST_HEAD(&fs_devs->devices);
 147         INIT_LIST_HEAD(&fs_devs->resized_devices);
 148         INIT_LIST_HEAD(&fs_devs->alloc_list);
 149         INIT_LIST_HEAD(&fs_devs->list);
 150
 151         return fs_devs;
 152 }
 153
 154 /**
 155  * alloc_fs_devices - allocate struct btrfs_fs_devices
 156  * @fsid:       a pointer to UUID for this FS.  If NULL a new UUID is
 157  *              generated.
 158  *
 159  * Return: a pointer to a new &struct btrfs_fs_devices on success;
 160  * ERR_PTR() on error.  Returned struct is not linked onto any lists and
 161  * can be destroyed with kfree() right away.
 162  */
 163 static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid)
 164 {
 165         struct btrfs_fs_devices *fs_devs;
 166
 167         fs_devs = __alloc_fs_devices();
 168         if (IS_ERR(fs_devs))
 169                 return fs_devs;
 170
 171         if (fsid)
 172                 memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE);
 173         else
 174                 generate_random_uuid(fs_devs->fsid);
 175
 176         return fs_devs;
 177 }
 178
 179 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
 180 {
 181         struct btrfs_device *device;
 182         WARN_ON(fs_devices->opened);
 183         while (!list_empty(&fs_devices->devices)) {
 184                 device = list_entry(fs_devices->devices.next,
 185                                     struct btrfs_device, dev_list);
 186                 list_del(&device->dev_list);
 187                 rcu_string_free(device->name);
 188                 kfree(device);
 189         }
 190         kfree(fs_devices);
 191 }
 192
 193 static void btrfs_kobject_uevent(struct block_device *bdev,
 194                                  enum kobject_action action)
 195 {
 196         int ret;
 197
 198         ret = kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, action);
 199         if (ret)
 200                 pr_warn("BTRFS: Sending event '%d' to kobject: '%s' (%p): failed\n",
 201                         action,
 202                         kobject_name(&disk_to_dev(bdev->bd_disk)->kobj),
 203                         &disk_to_dev(bdev->bd_disk)->kobj);
 204 }
 205
 206 void btrfs_cleanup_fs_uuids(void)
 207 {
 208         struct btrfs_fs_devices *fs_devices;
 209
 210         while (!list_empty(&fs_uuids)) {
 211                 fs_devices = list_entry(fs_uuids.next,
 212                                         struct btrfs_fs_devices, list);
 213                 list_del(&fs_devices->list);
 214                 free_fs_devices(fs_devices);
 215         }
 216 }
 217
 218 static struct btrfs_device *__alloc_device(void)
 219 {
 220         struct btrfs_device *dev;
 221
 222         dev = kzalloc(sizeof(*dev), GFP_NOFS);
 223         if (!dev)
 224                 return ERR_PTR(-ENOMEM);
 225
 226         INIT_LIST_HEAD(&dev->dev_list);
 227         INIT_LIST_HEAD(&dev->dev_alloc_list);
 228         INIT_LIST_HEAD(&dev->resized_list);
 229
 230         spin_lock_init(&dev->io_lock);
 231
 232         spin_lock_init(&dev->reada_lock);
 233         atomic_set(&dev->reada_in_flight, 0);
 234         atomic_set(&dev->dev_stats_ccnt, 0);
 235         INIT_RADIX_TREE(&dev->reada_zones, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
 236         INIT_RADIX_TREE(&dev->reada_extents, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
 237
 238         return dev;
 239 }
 240
 241 static noinline struct btrfs_device *__find_device(struct list_head *head,
 242                                                    u64 devid, u8 *uuid)
 243 {
 244         struct btrfs_device *dev;
 245
 246         list_for_each_entry(dev, head, dev_list) {
 247                 if (dev->devid == devid &&
 248                     (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
 249                         return dev;
 250                 }
 251         }
 252         return NULL;
 253 }
 254
 255 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
 256 {
 257         struct btrfs_fs_devices *fs_devices;
 258
 259         list_for_each_entry(fs_devices, &fs_uuids, list) {
 260                 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
 261                         return fs_devices;
 262         }
 263         return NULL;
 264 }
 265
 266 static int
 267 btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder,
 268                       int flush, struct block_device **bdev,
 269                       struct buffer_head **bh)
 270 {
 271         int ret;
 272
 273         *bdev = blkdev_get_by_path(device_path, flags, holder);
 274
 275         if (IS_ERR(*bdev)) {
 276                 ret = PTR_ERR(*bdev);
 277                 goto error;
 278         }
 279
 280         if (flush)
 281                 filemap_write_and_wait((*bdev)->bd_inode->i_mapping);
 282         ret = set_blocksize(*bdev, 4096);
 283         if (ret) {
 284                 blkdev_put(*bdev, flags);
 285                 goto error;
 286         }
 287         invalidate_bdev(*bdev);
 288         *bh = btrfs_read_dev_super(*bdev);
 289         if (IS_ERR(*bh)) {
 290                 ret = PTR_ERR(*bh);
 291                 blkdev_put(*bdev, flags);
 292                 goto error;
 293         }
 294
 295         return 0;
 296
 297 error:
 298         *bdev = NULL;
 299         *bh = NULL;
 300         return ret;
 301 }
 302
 303 static void requeue_list(struct btrfs_pending_bios *pending_bios,
 304                         struct bio *head, struct bio *tail)
 305 {
 306
 307         struct bio *old_head;
 308
 309         old_head = pending_bios->head;
 310         pending_bios->head = head;
 311         if (pending_bios->tail)
 312                 tail->bi_next = old_head;
 313         else
 314                 pending_bios->tail = tail;
 315 }
 316
 317 /*
 318  * we try to collect pending bios for a device so we don't get a large
 319  * number of procs sending bios down to the same device.  This greatly
 320  * improves the schedulers ability to collect and merge the bios.
 321  *
 322  * But, it also turns into a long list of bios to process and that is sure
 323  * to eventually make the worker thread block.  The solution here is to
 324  * make some progress and then put this work struct back at the end of
 325  * the list if the block device is congested.  This way, multiple devices
 326  * can make progress from a single worker thread.
 327  */
 328 static noinline void run_scheduled_bios(struct btrfs_device *device)
 329 {
 330         struct bio *pending;
 331         struct backing_dev_info *bdi;
 332         struct btrfs_fs_info *fs_info;
 333         struct btrfs_pending_bios *pending_bios;
 334         struct bio *tail;
 335         struct bio *cur;
 336         int again = 0;
 337         unsigned long num_run;
 338         unsigned long batch_run = 0;
 339         unsigned long limit;
 340         unsigned long last_waited = 0;
 341         int force_reg = 0;
 342         int sync_pending = 0;
 343         struct blk_plug plug;
 344
 345         /*
 346          * this function runs all the bios we've collected for
 347          * a particular device.  We don't want to wander off to
 348          * another device without first sending all of these down.
 349          * So, setup a plug here and finish it off before we return
 350          */
 351         blk_start_plug(&plug);
 352
 353         bdi = blk_get_backing_dev_info(device->bdev);
 354         fs_info = device->dev_root->fs_info;
 355         limit = btrfs_async_submit_limit(fs_info);
 356         limit = limit * 2 / 3;
 357
 358 loop:
 359         spin_lock(&device->io_lock);
 360
 361 loop_lock:
 362         num_run = 0;
 363
 364         /* take all the bios off the list at once and process them
 365          * later on (without the lock held).  But, remember the
 366          * tail and other pointers so the bios can be properly reinserted
 367          * into the list if we hit congestion
 368          */
 369         if (!force_reg && device->pending_sync_bios.head) {
 370                 pending_bios = &device->pending_sync_bios;
 371                 force_reg = 1;
 372         } else {
 373                 pending_bios = &device->pending_bios;
 374                 force_reg = 0;
 375         }
 376
 377         pending = pending_bios->head;
 378         tail = pending_bios->tail;
 379         WARN_ON(pending && !tail);
 380
 381         /*
 382          * if pending was null this time around, no bios need processing
 383          * at all and we can stop.  Otherwise it'll loop back up again
 384          * and do an additional check so no bios are missed.
 385          *
 386          * device->running_pending is used to synchronize with the
 387          * schedule_bio code.
 388          */
 389         if (device->pending_sync_bios.head == NULL &&
 390             device->pending_bios.head == NULL) {
 391                 again = 0;
 392                 device->running_pending = 0;
 393         } else {
 394                 again = 1;
 395                 device->running_pending = 1;
 396         }
 397
 398         pending_bios->head = NULL;
 399         pending_bios->tail = NULL;
 400
 401         spin_unlock(&device->io_lock);
 402
 403         while (pending) {
 404
 405                 rmb();
 406                 /* we want to work on both lists, but do more bios on the
 407                  * sync list than the regular list
 408                  */
 409                 if ((num_run > 32 &&
 410                     pending_bios != &device->pending_sync_bios &&
 411                     device->pending_sync_bios.head) ||
 412                    (num_run > 64 && pending_bios == &device->pending_sync_bios &&
 413                     device->pending_bios.head)) {
 414                         spin_lock(&device->io_lock);
 415                         requeue_list(pending_bios, pending, tail);
 416                         goto loop_lock;
 417                 }
 418
 419                 cur = pending;
 420                 pending = pending->bi_next;
 421                 cur->bi_next = NULL;
 422
 423                 /*
 424                  * atomic_dec_return implies a barrier for waitqueue_active
 425                  */
 426                 if (atomic_dec_return(&fs_info->nr_async_bios) < limit &&
 427                     waitqueue_active(&fs_info->async_submit_wait))
 428                         wake_up(&fs_info->async_submit_wait);
 429
 430                 BUG_ON(atomic_read(&cur->__bi_cnt) == 0);
 431
 432                 /*
 433                  * if we're doing the sync list, record that our
 434                  * plug has some sync requests on it
 435                  *
 436                  * If we're doing the regular list and there are
 437                  * sync requests sitting around, unplug before
 438                  * we add more
 439                  */
 440                 if (pending_bios == &device->pending_sync_bios) {
 441                         sync_pending = 1;
 442                 } else if (sync_pending) {
 443                         blk_finish_plug(&plug);
 444                         blk_start_plug(&plug);
 445                         sync_pending = 0;
 446                 }
 447
 448                 btrfsic_submit_bio(cur->bi_rw, cur);
 449                 num_run++;
 450                 batch_run++;
 451
 452                 cond_resched();
 453
 454                 /*
 455                  * we made progress, there is more work to do and the bdi
 456                  * is now congested.  Back off and let other work structs
 457                  * run instead
 458                  */
 459                 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
 460                     fs_info->fs_devices->open_devices > 1) {
 461                         struct io_context *ioc;
 462
 463                         ioc = current->io_context;
 464
 465                         /*
 466                          * the main goal here is that we don't want to
 467                          * block if we're going to be able to submit
 468                          * more requests without blocking.
 469                          *
 470                          * This code does two great things, it pokes into
 471                          * the elevator code from a filesystem _and_
 472                          * it makes assumptions about how batching works.
 473                          */
 474                         if (ioc && ioc->nr_batch_requests > 0 &&
 475                             time_before(jiffies, ioc->last_waited + HZ/50UL) &&
 476                             (last_waited == 0 ||
 477                              ioc->last_waited == last_waited)) {
 478                                 /*
 479                                  * we want to go through our batch of
 480                                  * requests and stop.  So, we copy out
 481                                  * the ioc->last_waited time and test
 482                                  * against it before looping
 483                                  */
 484                                 last_waited = ioc->last_waited;
 485                                 cond_resched();
 486                                 continue;
 487                         }
 488                         spin_lock(&device->io_lock);
 489                         requeue_list(pending_bios, pending, tail);
 490                         device->running_pending = 1;
 491
 492                         spin_unlock(&device->io_lock);
 493                         btrfs_queue_work(fs_info->submit_workers,
 494                                          &device->work);
 495                         goto done;
 496                 }
 497                 /* unplug every 64 requests just for good measure */
 498                 if (batch_run % 64 == 0) {
 499                         blk_finish_plug(&plug);
 500                         blk_start_plug(&plug);
 501                         sync_pending = 0;
 502                 }
 503         }
 504
 505         cond_resched();
 506         if (again)
 507                 goto loop;
 508
 509         spin_lock(&device->io_lock);
 510         if (device->pending_bios.head || device->pending_sync_bios.head)
 511                 goto loop_lock;
 512         spin_unlock(&device->io_lock);
 513
 514 done:
 515         blk_finish_plug(&plug);
 516 }
 517
 518 static void pending_bios_fn(struct btrfs_work *work)
 519 {
 520         struct btrfs_device *device;
 521
 522         device = container_of(work, struct btrfs_device, work);
 523         run_scheduled_bios(device);
 524 }
 525
 526
 527 void btrfs_free_stale_device(struct btrfs_device *cur_dev)
 528 {
 529         struct btrfs_fs_devices *fs_devs;
 530         struct btrfs_device *dev;
 531
 532         if (!cur_dev->name)
 533                 return;
 534
 535         list_for_each_entry(fs_devs, &fs_uuids, list) {
 536                 int del = 1;
 537
 538                 if (fs_devs->opened)
 539                         continue;
 540                 if (fs_devs->seeding)
 541                         continue;
 542
 543                 list_for_each_entry(dev, &fs_devs->devices, dev_list) {
 544
 545                         if (dev == cur_dev)
 546                                 continue;
 547                         if (!dev->name)
 548                                 continue;
 549
 550                         /*
 551                          * Todo: This won't be enough. What if the same device
 552                          * comes back (with new uuid and) with its mapper path?
 553                          * But for now, this does help as mostly an admin will
 554                          * either use mapper or non mapper path throughout.
 555                          */
 556                         rcu_read_lock();
 557                         del = strcmp(rcu_str_deref(dev->name),
 558                                                 rcu_str_deref(cur_dev->name));
 559                         rcu_read_unlock();
 560                         if (!del)
 561                                 break;
 562                 }
 563
 564                 if (!del) {
 565                         /* delete the stale device */
 566                         if (fs_devs->num_devices == 1) {
 567                                 btrfs_sysfs_remove_fsid(fs_devs);
 568                                 list_del(&fs_devs->list);
 569                                 free_fs_devices(fs_devs);
 570                         } else {
 571                                 fs_devs->num_devices--;
 572                                 list_del(&dev->dev_list);
 573                                 rcu_string_free(dev->name);
 574                                 kfree(dev);
 575                         }
 576                         break;
 577                 }
 578         }
 579 }
 580
 581 /*
 582  * Add new device to list of registered devices
 583  *
 584  * Returns:
 585  * 1   - first time device is seen
 586  * 0   - device already known
 587  * < 0 - error
 588  */
 589 static noinline int device_list_add(const char *path,
 590                            struct btrfs_super_block *disk_super,
 591                            u64 devid, struct btrfs_fs_devices **fs_devices_ret)
 592 {
 593         struct btrfs_device *device;
 594         struct btrfs_fs_devices *fs_devices;
 595         struct rcu_string *name;
 596         int ret = 0;
 597         u64 found_transid = btrfs_super_generation(disk_super);
 598
 599         fs_devices = find_fsid(disk_super->fsid);
 600         if (!fs_devices) {
 601                 fs_devices = alloc_fs_devices(disk_super->fsid);
 602                 if (IS_ERR(fs_devices))
 603                         return PTR_ERR(fs_devices);
 604
 605                 list_add(&fs_devices->list, &fs_uuids);
 606
 607                 device = NULL;
 608         } else {
 609                 device = __find_device(&fs_devices->devices, devid,
 610                                        disk_super->dev_item.uuid);
 611         }
 612
 613         if (!device) {
 614                 if (fs_devices->opened)
 615                         return -EBUSY;
 616
 617                 device = btrfs_alloc_device(NULL, &devid,
 618                                             disk_super->dev_item.uuid);
 619                 if (IS_ERR(device)) {
 620                         /* we can safely leave the fs_devices entry around */
 621                         return PTR_ERR(device);
 622                 }
 623
 624                 name = rcu_string_strdup(path, GFP_NOFS);
 625                 if (!name) {
 626                         kfree(device);
 627                         return -ENOMEM;
 628                 }
 629                 rcu_assign_pointer(device->name, name);
 630
 631                 mutex_lock(&fs_devices->device_list_mutex);
 632                 list_add_rcu(&device->dev_list, &fs_devices->devices);
 633                 fs_devices->num_devices++;
 634                 mutex_unlock(&fs_devices->device_list_mutex);
 635
 636                 ret = 1;
 637                 device->fs_devices = fs_devices;
 638         } else if (!device->name || strcmp(device->name->str, path)) {
 639                 /*
 640                  * When FS is already mounted.
 641                  * 1. If you are here and if the device->name is NULL that
 642                  *    means this device was missing at time of FS mount.
 643                  * 2. If you are here and if the device->name is different
 644                  *    from 'path' that means either
 645                  *      a. The same device disappeared and reappeared with
 646                  *         different name. or
 647                  *      b. The missing-disk-which-was-replaced, has
 648                  *         reappeared now.
 649                  *
 650                  * We must allow 1 and 2a above. But 2b would be a spurious
 651                  * and unintentional.
 652                  *
 653                  * Further in case of 1 and 2a above, the disk at 'path'
 654                  * would have missed some transaction when it was away and
 655                  * in case of 2a the stale bdev has to be updated as well.
 656                  * 2b must not be allowed at all time.
 657                  */
 658
 659                 /*
 660                  * For now, we do allow update to btrfs_fs_device through the
 661                  * btrfs dev scan cli after FS has been mounted.  We're still
 662                  * tracking a problem where systems fail mount by subvolume id
 663                  * when we reject replacement on a mounted FS.
 664                  */
 665                 if (!fs_devices->opened && found_transid < device->generation) {
 666                         /*
 667                          * That is if the FS is _not_ mounted and if you
 668                          * are here, that means there is more than one
 669                          * disk with same uuid and devid.We keep the one
 670                          * with larger generation number or the last-in if
 671                          * generation are equal.
 672                          */
 673                         return -EEXIST;
 674                 }
 675
 676                 name = rcu_string_strdup(path, GFP_NOFS);
 677                 if (!name)
 678                         return -ENOMEM;
 679                 rcu_string_free(device->name);
 680                 rcu_assign_pointer(device->name, name);
 681                 if (device->missing) {
 682                         fs_devices->missing_devices--;
 683                         device->missing = 0;
 684                 }
 685         }
 686
 687         /*
 688          * Unmount does not free the btrfs_device struct but would zero
 689          * generation along with most of the other members. So just update
 690          * it back. We need it to pick the disk with largest generation
 691          * (as above).
 692          */
 693         if (!fs_devices->opened)
 694                 device->generation = found_transid;
 695
 696         /*
 697          * if there is new btrfs on an already registered device,
 698          * then remove the stale device entry.
 699          */
 700         btrfs_free_stale_device(device);
 701
 702         *fs_devices_ret = fs_devices;
 703
 704         return ret;
 705 }
 706
 707 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
 708 {
 709         struct btrfs_fs_devices *fs_devices;
 710         struct btrfs_device *device;
 711         struct btrfs_device *orig_dev;
 712
 713         fs_devices = alloc_fs_devices(orig->fsid);
 714         if (IS_ERR(fs_devices))
 715                 return fs_devices;
 716
 717         mutex_lock(&orig->device_list_mutex);
 718         fs_devices->total_devices = orig->total_devices;
 719
 720         /* We have held the volume lock, it is safe to get the devices. */
 721         list_for_each_entry(orig_dev, &orig->devices, dev_list) {
 722                 struct rcu_string *name;
 723
 724                 device = btrfs_alloc_device(NULL, &orig_dev->devid,
 725                                             orig_dev->uuid);
 726                 if (IS_ERR(device))
 727                         goto error;
 728
 729                 /*
 730                  * This is ok to do without rcu read locked because we hold the
 731                  * uuid mutex so nothing we touch in here is going to disappear.
 732                  */
 733                 if (orig_dev->name) {
 734                         name = rcu_string_strdup(orig_dev->name->str, GFP_NOFS);
 735                         if (!name) {
 736                                 kfree(device);
 737                                 goto error;
 738                         }
 739                         rcu_assign_pointer(device->name, name);
 740                 }
 741
 742                 list_add(&device->dev_list, &fs_devices->devices);
 743                 device->fs_devices = fs_devices;
 744                 fs_devices->num_devices++;
 745         }
 746         mutex_unlock(&orig->device_list_mutex);
 747         return fs_devices;
 748 error:
 749         mutex_unlock(&orig->device_list_mutex);
 750         free_fs_devices(fs_devices);
 751         return ERR_PTR(-ENOMEM);
 752 }
 753
 754 void btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices, int step)
 755 {
 756         struct btrfs_device *device, *next;
 757         struct btrfs_device *latest_dev = NULL;
 758
 759         mutex_lock(&uuid_mutex);
 760 again:
 761         /* This is the initialized path, it is safe to release the devices. */
 762         list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
 763                 if (device->in_fs_metadata) {
 764                         if (!device->is_tgtdev_for_dev_replace &&
 765                             (!latest_dev ||
 766                              device->generation > latest_dev->generation)) {
 767                                 latest_dev = device;
 768                         }
 769                         continue;
 770                 }
 771
 772                 if (device->devid == BTRFS_DEV_REPLACE_DEVID) {
 773                         /*
 774                          * In the first step, keep the device which has
 775                          * the correct fsid and the devid that is used
 776                          * for the dev_replace procedure.
 777                          * In the second step, the dev_replace state is
 778                          * read from the device tree and it is known
 779                          * whether the procedure is really active or
 780                          * not, which means whether this device is
 781                          * used or whether it should be removed.
 782                          */
 783                         if (step == 0 || device->is_tgtdev_for_dev_replace) {
 784                                 continue;
 785                         }
 786                 }
 787                 if (device->bdev) {
 788                         blkdev_put(device->bdev, device->mode);
 789                         device->bdev = NULL;
 790                         fs_devices->open_devices--;
 791                 }
 792                 if (device->writeable) {
 793                         list_del_init(&device->dev_alloc_list);
 794                         device->writeable = 0;
 795                         if (!device->is_tgtdev_for_dev_replace)
 796                                 fs_devices->rw_devices--;
 797                 }
 798                 list_del_init(&device->dev_list);
 799                 fs_devices->num_devices--;
 800                 rcu_string_free(device->name);
 801                 kfree(device);
 802         }
 803
 804         if (fs_devices->seed) {
 805                 fs_devices = fs_devices->seed;
 806                 goto again;
 807         }
 808
 809         fs_devices->latest_bdev = latest_dev->bdev;
 810
 811         mutex_unlock(&uuid_mutex);
 812 }
 813
 814 static void __free_device(struct work_struct *work)
 815 {
 816         struct btrfs_device *device;
 817
 818         device = container_of(work, struct btrfs_device, rcu_work);
 819
 820         if (device->bdev)
 821                 blkdev_put(device->bdev, device->mode);
 822
 823         rcu_string_free(device->name);
 824         kfree(device);
 825 }
 826
 827 static void free_device(struct rcu_head *head)
 828 {
 829         struct btrfs_device *device;
 830
 831         device = container_of(head, struct btrfs_device, rcu);
 832
 833         INIT_WORK(&device->rcu_work, __free_device);
 834         schedule_work(&device->rcu_work);
 835 }
 836
 837 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
 838 {
 839         struct btrfs_device *device, *tmp;
 840
 841         if (--fs_devices->opened > 0)
 842                 return 0;
 843
 844         mutex_lock(&fs_devices->device_list_mutex);
 845         list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list) {
 846                 btrfs_close_one_device(device);
 847         }
 848         mutex_unlock(&fs_devices->device_list_mutex);
 849
 850         WARN_ON(fs_devices->open_devices);
 851         WARN_ON(fs_devices->rw_devices);
 852         fs_devices->opened = 0;
 853         fs_devices->seeding = 0;
 854
 855         return 0;
 856 }
 857
 858 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
 859 {
 860         struct btrfs_fs_devices *seed_devices = NULL;
 861         int ret;
 862
 863         mutex_lock(&uuid_mutex);
 864         ret = __btrfs_close_devices(fs_devices);
 865         if (!fs_devices->opened) {
 866                 seed_devices = fs_devices->seed;
 867                 fs_devices->seed = NULL;
 868         }
 869         mutex_unlock(&uuid_mutex);
 870
 871         while (seed_devices) {
 872                 fs_devices = seed_devices;
 873                 seed_devices = fs_devices->seed;
 874                 __btrfs_close_devices(fs_devices);
 875                 free_fs_devices(fs_devices);
 876         }
 877         /*
 878          * Wait for rcu kworkers under __btrfs_close_devices
 879          * to finish all blkdev_puts so device is really
 880          * free when umount is done.
 881          */
 882         rcu_barrier();
 883         return ret;
 884 }
 885
 886 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
 887                                 fmode_t flags, void *holder)
 888 {
 889         struct request_queue *q;
 890         struct block_device *bdev;
 891         struct list_head *head = &fs_devices->devices;
 892         struct btrfs_device *device;
 893         struct btrfs_device *latest_dev = NULL;
 894         struct buffer_head *bh;
 895         struct btrfs_super_block *disk_super;
 896         u64 devid;
 897         int seeding = 1;
 898         int ret = 0;
 899
 900         flags |= FMODE_EXCL;
 901
 902         list_for_each_entry(device, head, dev_list) {
 903                 if (device->bdev)
 904                         continue;
 905                 if (!device->name)
 906                         continue;
 907
 908                 /* Just open everything we can; ignore failures here */
 909                 if (btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1,
 910                                             &bdev, &bh))
 911                         continue;
 912
 913                 disk_super = (struct btrfs_super_block *)bh->b_data;
 914                 devid = btrfs_stack_device_id(&disk_super->dev_item);
 915                 if (devid != device->devid)
 916                         goto error_brelse;
 917
 918                 if (memcmp(device->uuid, disk_super->dev_item.uuid,
 919                            BTRFS_UUID_SIZE))
 920                         goto error_brelse;
 921
 922                 device->generation = btrfs_super_generation(disk_super);
 923                 if (!latest_dev ||
 924                     device->generation > latest_dev->generation)
 925                         latest_dev = device;
 926
 927                 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
 928                         device->writeable = 0;
 929                 } else {
 930                         device->writeable = !bdev_read_only(bdev);
 931                         seeding = 0;
 932                 }
 933
 934                 q = bdev_get_queue(bdev);
 935                 if (blk_queue_discard(q))
 936                         device->can_discard = 1;
 937
 938                 device->bdev = bdev;
 939                 device->in_fs_metadata = 0;
 940                 device->mode = flags;
 941
 942                 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
 943                         fs_devices->rotating = 1;
 944
 945                 fs_devices->open_devices++;
 946                 if (device->writeable &&
 947                     device->devid != BTRFS_DEV_REPLACE_DEVID) {
 948                         fs_devices->rw_devices++;
 949                         list_add(&device->dev_alloc_list,
 950                                  &fs_devices->alloc_list);
 951                 }
 952                 brelse(bh);
 953                 continue;
 954
 955 error_brelse:
 956                 brelse(bh);
 957                 blkdev_put(bdev, flags);
 958                 continue;
 959         }
 960         if (fs_devices->open_devices == 0) {
 961                 ret = -EINVAL;
 962                 goto out;
 963         }
 964         fs_devices->seeding = seeding;
 965         fs_devices->opened = 1;
 966         fs_devices->latest_bdev = latest_dev->bdev;
 967         fs_devices->total_rw_bytes = 0;
 968 out:
 969         return ret;
 970 }
 971
 972 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
 973                        fmode_t flags, void *holder)
 974 {
 975         int ret;
 976
 977         mutex_lock(&uuid_mutex);
 978         if (fs_devices->opened) {
 979                 fs_devices->opened++;
 980                 ret = 0;
 981         } else {
 982                 ret = __btrfs_open_devices(fs_devices, flags, holder);
 983         }
 984         mutex_unlock(&uuid_mutex);
 985         return ret;
 986 }
 987
 988 /*
 989  * Look for a btrfs signature on a device. This may be called out of the mount path
 990  * and we are not allowed to call set_blocksize during the scan. The superblock
 991  * is read via pagecache
 992  */
 993 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
 994                           struct btrfs_fs_devices **fs_devices_ret)
 995 {
 996         struct btrfs_super_block *disk_super;
 997         struct block_device *bdev;
 998         struct page *page;
 999         void *p;
1000         int ret = -EINVAL;
1001         u64 devid;
1002         u64 transid;
1003         u64 total_devices;
1004         u64 bytenr;
1005         pgoff_t index;
1006
1007         /*
1008          * we would like to check all the supers, but that would make
1009          * a btrfs mount succeed after a mkfs from a different FS.
1010          * So, we need to add a special mount option to scan for
1011          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
1012          */
1013         bytenr = btrfs_sb_offset(0);
1014         flags |= FMODE_EXCL;
1015         mutex_lock(&uuid_mutex);
1016
1017         bdev = blkdev_get_by_path(path, flags, holder);
1018
1019         if (IS_ERR(bdev)) {
1020                 ret = PTR_ERR(bdev);
1021                 goto error;
1022         }
1023
1024         /* make sure our super fits in the device */
1025         if (bytenr + PAGE_CACHE_SIZE >= i_size_read(bdev->bd_inode))
1026                 goto error_bdev_put;
1027
1028         /* make sure our super fits in the page */
1029         if (sizeof(*disk_super) > PAGE_CACHE_SIZE)
1030                 goto error_bdev_put;
1031
1032         /* make sure our super doesn't straddle pages on disk */
1033         index = bytenr >> PAGE_CACHE_SHIFT;
1034         if ((bytenr + sizeof(*disk_super) - 1) >> PAGE_CACHE_SHIFT != index)
1035                 goto error_bdev_put;
1036
1037         /* pull in the page with our super */
1038         page = read_cache_page_gfp(bdev->bd_inode->i_mapping,
1039                                    index, GFP_NOFS);
1040
1041         if (IS_ERR_OR_NULL(page))
1042                 goto error_bdev_put;
1043
1044         p = kmap(page);
1045
1046         /* align our pointer to the offset of the super block */
1047         disk_super = p + (bytenr & ~PAGE_CACHE_MASK);
1048
1049         if (btrfs_super_bytenr(disk_super) != bytenr ||
1050             btrfs_super_magic(disk_super) != BTRFS_MAGIC)
1051                 goto error_unmap;
1052
1053         devid = btrfs_stack_device_id(&disk_super->dev_item);
1054         transid = btrfs_super_generation(disk_super);
1055         total_devices = btrfs_super_num_devices(disk_super);
1056
1057         ret = device_list_add(path, disk_super, devid, fs_devices_ret);
1058         if (ret > 0) {
1059                 if (disk_super->label[0]) {
1060                         if (disk_super->label[BTRFS_LABEL_SIZE - 1])
1061                                 disk_super->label[BTRFS_LABEL_SIZE - 1] = '\0';
1062                         printk(KERN_INFO "BTRFS: device label %s ", disk_super->label);
1063                 } else {
1064                         printk(KERN_INFO "BTRFS: device fsid %pU ", disk_super->fsid);
1065                 }
1066
1067                 printk(KERN_CONT "devid %llu transid %llu %s\n", devid, transid, path);
1068                 ret = 0;
1069         }
1070         if (!ret && fs_devices_ret)
1071                 (*fs_devices_ret)->total_devices = total_devices;
1072
1073 error_unmap:
1074         kunmap(page);
1075         page_cache_release(page);
1076
1077 error_bdev_put:
1078         blkdev_put(bdev, flags);
1079 error:
1080         mutex_unlock(&uuid_mutex);
1081         return ret;
1082 }
1083
1084 /* helper to account the used device space in the range */
1085 int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
1086                                    u64 end, u64 *length)
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 extent_end;
1093         int ret;
1094         int slot;
1095         struct extent_buffer *l;
1096
1097         *length = 0;
1098
1099         if (start >= device->total_bytes || device->is_tgtdev_for_dev_replace)
1100                 return 0;
1101
1102         path = btrfs_alloc_path();
1103         if (!path)
1104                 return -ENOMEM;
1105         path->reada = 2;
1106
1107         key.objectid = device->devid;
1108         key.offset = start;
1109         key.type = BTRFS_DEV_EXTENT_KEY;
1110
1111         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1112         if (ret < 0)
1113                 goto out;
1114         if (ret > 0) {
1115                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1116                 if (ret < 0)
1117                         goto out;
1118         }
1119
1120         while (1) {
1121                 l = path->nodes[0];
1122                 slot = path->slots[0];
1123                 if (slot >= btrfs_header_nritems(l)) {
1124                         ret = btrfs_next_leaf(root, path);
1125                         if (ret == 0)
1126                                 continue;
1127                         if (ret < 0)
1128                                 goto out;
1129
1130                         break;
1131                 }
1132                 btrfs_item_key_to_cpu(l, &key, slot);
1133
1134                 if (key.objectid < device->devid)
1135                         goto next;
1136
1137                 if (key.objectid > device->devid)
1138                         break;
1139
1140                 if (key.type != BTRFS_DEV_EXTENT_KEY)
1141                         goto next;
1142
1143                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1144                 extent_end = key.offset + btrfs_dev_extent_length(l,
1145                                                                   dev_extent);
1146                 if (key.offset <= start && extent_end > end) {
1147                         *length = end - start + 1;
1148                         break;
1149                 } else if (key.offset <= start && extent_end > start)
1150                         *length += extent_end - start;
1151                 else if (key.offset > start && extent_end <= end)
1152                         *length += extent_end - key.offset;
1153                 else if (key.offset > start && key.offset <= end) {
1154                         *length += end - key.offset + 1;
1155                         break;
1156                 } else if (key.offset > end)
1157                         break;
1158
1159 next:
1160                 path->slots[0]++;
1161         }
1162         ret = 0;
1163 out:
1164         btrfs_free_path(path);
1165         return ret;
1166 }
1167
1168 static int contains_pending_extent(struct btrfs_transaction *transaction,
1169                                    struct btrfs_device *device,
1170                                    u64 *start, u64 len)
1171 {
1172         struct btrfs_fs_info *fs_info = device->dev_root->fs_info;
1173         struct extent_map *em;
1174         struct list_head *search_list = &fs_info->pinned_chunks;
1175         int ret = 0;
1176         u64 physical_start = *start;
1177
1178         if (transaction)
1179                 search_list = &transaction->pending_chunks;
1180 again:
1181         list_for_each_entry(em, search_list, list) {
1182                 struct map_lookup *map;
1183                 int i;
1184
1185                 map = (struct map_lookup *)em->bdev;
1186                 for (i = 0; i < map->num_stripes; i++) {
1187                         u64 end;
1188
1189                         if (map->stripes[i].dev != device)
1190                                 continue;
1191                         if (map->stripes[i].physical >= physical_start + len ||
1192                             map->stripes[i].physical + em->orig_block_len <=
1193                             physical_start)
1194                                 continue;
1195                         /*
1196                          * Make sure that while processing the pinned list we do
1197                          * not override our *start with a lower value, because
1198                          * we can have pinned chunks that fall within this
1199                          * device hole and that have lower physical addresses
1200                          * than the pending chunks we processed before. If we
1201                          * do not take this special care we can end up getting
1202                          * 2 pending chunks that start at the same physical
1203                          * device offsets because the end offset of a pinned
1204                          * chunk can be equal to the start offset of some
1205                          * pending chunk.
1206                          */
1207                         end = map->stripes[i].physical + em->orig_block_len;
1208                         if (end > *start) {
1209                                 *start = end;
1210                                 ret = 1;
1211                         }
1212                 }
1213         }
1214         if (search_list != &fs_info->pinned_chunks) {
1215                 search_list = &fs_info->pinned_chunks;
1216                 goto again;
1217         }
1218
1219         return ret;
1220 }
1221
1222
1223 /*
1224  * find_free_dev_extent_start - find free space in the specified device
1225  * @device:       the device which we search the free space in
1226  * @num_bytes:    the size of the free space that we need
1227  * @search_start: the position from which to begin the search
1228  * @start:        store the start of the free space.
1229  * @len:          the size of the free space. that we find, or the size
1230  *                of the max free space if we don't find suitable free space
1231  *
1232  * this uses a pretty simple search, the expectation is that it is
1233  * called very infrequently and that a given device has a small number
1234  * of extents
1235  *
1236  * @start is used to store the start of the free space if we find. But if we
1237  * don't find suitable free space, it will be used to store the start position
1238  * of the max free space.
1239  *
1240  * @len is used to store the size of the free space that we find.
1241  * But if we don't find suitable free space, it is used to store the size of
1242  * the max free space.
1243  */
1244 int find_free_dev_extent_start(struct btrfs_transaction *transaction,
1245                                struct btrfs_device *device, u64 num_bytes,
1246                                u64 search_start, u64 *start, u64 *len)
1247 {
1248         struct btrfs_key key;
1249         struct btrfs_root *root = device->dev_root;
1250         struct btrfs_dev_extent *dev_extent;
1251         struct btrfs_path *path;
1252         u64 hole_size;
1253         u64 max_hole_start;
1254         u64 max_hole_size;
1255         u64 extent_end;
1256         u64 search_end = device->total_bytes;
1257         int ret;
1258         int slot;
1259         struct extent_buffer *l;
1260
1261         path = btrfs_alloc_path();
1262         if (!path)
1263                 return -ENOMEM;
1264
1265         max_hole_start = search_start;
1266         max_hole_size = 0;
1267
1268 again:
1269         if (search_start >= search_end || device->is_tgtdev_for_dev_replace) {
1270                 ret = -ENOSPC;
1271                 goto out;
1272         }
1273
1274         path->reada = 2;
1275         path->search_commit_root = 1;
1276         path->skip_locking = 1;
1277
1278         key.objectid = device->devid;
1279         key.offset = search_start;
1280         key.type = BTRFS_DEV_EXTENT_KEY;
1281
1282         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1283         if (ret < 0)
1284                 goto out;
1285         if (ret > 0) {
1286                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1287                 if (ret < 0)
1288                         goto out;
1289         }
1290
1291         while (1) {
1292                 l = path->nodes[0];
1293                 slot = path->slots[0];
1294                 if (slot >= btrfs_header_nritems(l)) {
1295                         ret = btrfs_next_leaf(root, path);
1296                         if (ret == 0)
1297                                 continue;
1298                         if (ret < 0)
1299                                 goto out;
1300
1301                         break;
1302                 }
1303                 btrfs_item_key_to_cpu(l, &key, slot);
1304
1305                 if (key.objectid < device->devid)
1306                         goto next;
1307
1308                 if (key.objectid > device->devid)
1309                         break;
1310
1311                 if (key.type != BTRFS_DEV_EXTENT_KEY)
1312                         goto next;
1313
1314                 if (key.offset > search_start) {
1315                         hole_size = key.offset - search_start;
1316
1317                         /*
1318                          * Have to check before we set max_hole_start, otherwise
1319                          * we could end up sending back this offset anyway.
1320                          */
1321                         if (contains_pending_extent(transaction, device,
1322                                                     &search_start,
1323                                                     hole_size)) {
1324                                 if (key.offset >= search_start) {
1325                                         hole_size = key.offset - search_start;
1326                                 } else {
1327                                         WARN_ON_ONCE(1);
1328                                         hole_size = 0;
1329                                 }
1330                         }
1331
1332                         if (hole_size > max_hole_size) {
1333                                 max_hole_start = search_start;
1334                                 max_hole_size = hole_size;
1335                         }
1336
1337                         /*
1338                          * If this free space is greater than which we need,
1339                          * it must be the max free space that we have found
1340                          * until now, so max_hole_start must point to the start
1341                          * of this free space and the length of this free space
1342                          * is stored in max_hole_size. Thus, we return
1343                          * max_hole_start and max_hole_size and go back to the
1344                          * caller.
1345                          */
1346                         if (hole_size >= num_bytes) {
1347                                 ret = 0;
1348                                 goto out;
1349                         }
1350                 }
1351
1352                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1353                 extent_end = key.offset + btrfs_dev_extent_length(l,
1354                                                                   dev_extent);
1355                 if (extent_end > search_start)
1356                         search_start = extent_end;
1357 next:
1358                 path->slots[0]++;
1359                 cond_resched();
1360         }
1361
1362         /*
1363          * At this point, search_start should be the end of
1364          * allocated dev extents, and when shrinking the device,
1365          * search_end may be smaller than search_start.
1366          */
1367         if (search_end > search_start) {
1368                 hole_size = search_end - search_start;
1369
1370                 if (contains_pending_extent(transaction, device, &search_start,
1371                                             hole_size)) {
1372                         btrfs_release_path(path);
1373                         goto again;
1374                 }
1375
1376                 if (hole_size > max_hole_size) {
1377                         max_hole_start = search_start;
1378                         max_hole_size = hole_size;
1379                 }
1380         }
1381
1382         /* See above. */
1383         if (max_hole_size < num_bytes)
1384                 ret = -ENOSPC;
1385         else
1386                 ret = 0;
1387
1388 out:
1389         btrfs_free_path(path);
1390         *start = max_hole_start;
1391         if (len)
1392                 *len = max_hole_size;
1393         return ret;
1394 }
1395
1396 int find_free_dev_extent(struct btrfs_trans_handle *trans,
1397                          struct btrfs_device *device, u64 num_bytes,
1398                          u64 *start, u64 *len)
1399 {
1400         struct btrfs_root *root = device->dev_root;
1401         u64 search_start;
1402
1403         /* FIXME use last free of some kind */
1404
1405         /*
1406          * we don't want to overwrite the superblock on the drive,
1407          * so we make sure to start at an offset of at least 1MB
1408          */
1409         search_start = max(root->fs_info->alloc_start, 1024ull * 1024);
1410         return find_free_dev_extent_start(trans->transaction, device,
1411                                           num_bytes, search_start, start, len);
1412 }
1413
1414 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1415                           struct btrfs_device *device,
1416                           u64 start, u64 *dev_extent_len)
1417 {
1418         int ret;
1419         struct btrfs_path *path;
1420         struct btrfs_root *root = device->dev_root;
1421         struct btrfs_key key;
1422         struct btrfs_key found_key;
1423         struct extent_buffer *leaf = NULL;
1424         struct btrfs_dev_extent *extent = NULL;
1425
1426         path = btrfs_alloc_path();
1427         if (!path)
1428                 return -ENOMEM;
1429
1430         key.objectid = device->devid;
1431         key.offset = start;
1432         key.type = BTRFS_DEV_EXTENT_KEY;
1433 again:
1434         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1435         if (ret > 0) {
1436                 ret = btrfs_previous_item(root, path, key.objectid,
1437                                           BTRFS_DEV_EXTENT_KEY);
1438                 if (ret)
1439                         goto out;
1440                 leaf = path->nodes[0];
1441                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1442                 extent = btrfs_item_ptr(leaf, path->slots[0],
1443                                         struct btrfs_dev_extent);
1444                 BUG_ON(found_key.offset > start || found_key.offset +
1445                        btrfs_dev_extent_length(leaf, extent) < start);
1446                 key = found_key;
1447                 btrfs_release_path(path);
1448                 goto again;
1449         } else if (ret == 0) {
1450                 leaf = path->nodes[0];
1451                 extent = btrfs_item_ptr(leaf, path->slots[0],
1452                                         struct btrfs_dev_extent);
1453         } else {
1454                 btrfs_std_error(root->fs_info, ret, "Slot search failed");
1455                 goto out;
1456         }
1457
1458         *dev_extent_len = btrfs_dev_extent_length(leaf, extent);
1459
1460         ret = btrfs_del_item(trans, root, path);
1461         if (ret) {
1462                 btrfs_std_error(root->fs_info, ret,
1463                             "Failed to remove dev extent item");
1464         } else {
1465                 set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags);
1466         }
1467 out:
1468         btrfs_free_path(path);
1469         return ret;
1470 }
1471
1472 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1473                                   struct btrfs_device *device,
1474                                   u64 chunk_tree, u64 chunk_objectid,
1475                                   u64 chunk_offset, u64 start, u64 num_bytes)
1476 {
1477         int ret;
1478         struct btrfs_path *path;
1479         struct btrfs_root *root = device->dev_root;
1480         struct btrfs_dev_extent *extent;
1481         struct extent_buffer *leaf;
1482         struct btrfs_key key;
1483
1484         WARN_ON(!device->in_fs_metadata);
1485         WARN_ON(device->is_tgtdev_for_dev_replace);
1486         path = btrfs_alloc_path();
1487         if (!path)
1488                 return -ENOMEM;
1489
1490         key.objectid = device->devid;
1491         key.offset = start;
1492         key.type = BTRFS_DEV_EXTENT_KEY;
1493         ret = btrfs_insert_empty_item(trans, root, path, &key,
1494                                       sizeof(*extent));
1495         if (ret)
1496                 goto out;
1497
1498         leaf = path->nodes[0];
1499         extent = btrfs_item_ptr(leaf, path->slots[0],
1500                                 struct btrfs_dev_extent);
1501         btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
1502         btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
1503         btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1504
1505         write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
1506                     btrfs_dev_extent_chunk_tree_uuid(extent), BTRFS_UUID_SIZE);
1507
1508         btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1509         btrfs_mark_buffer_dirty(leaf);
1510 out:
1511         btrfs_free_path(path);
1512         return ret;
1513 }
1514
1515 static u64 find_next_chunk(struct btrfs_fs_info *fs_info)
1516 {
1517         struct extent_map_tree *em_tree;
1518         struct extent_map *em;
1519         struct rb_node *n;
1520         u64 ret = 0;
1521
1522         em_tree = &fs_info->mapping_tree.map_tree;
1523         read_lock(&em_tree->lock);
1524         n = rb_last(&em_tree->map);
1525         if (n) {
1526                 em = rb_entry(n, struct extent_map, rb_node);
1527                 ret = em->start + em->len;
1528         }
1529         read_unlock(&em_tree->lock);
1530
1531         return ret;
1532 }
1533
1534 static noinline int find_next_devid(struct btrfs_fs_info *fs_info,
1535                                     u64 *devid_ret)
1536 {
1537         int ret;
1538         struct btrfs_key key;
1539         struct btrfs_key found_key;
1540         struct btrfs_path *path;
1541
1542         path = btrfs_alloc_path();
1543         if (!path)
1544                 return -ENOMEM;
1545
1546         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1547         key.type = BTRFS_DEV_ITEM_KEY;
1548         key.offset = (u64)-1;
1549
1550         ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
1551         if (ret < 0)
1552                 goto error;
1553
1554         BUG_ON(ret == 0); /* Corruption */
1555
1556         ret = btrfs_previous_item(fs_info->chunk_root, path,
1557                                   BTRFS_DEV_ITEMS_OBJECTID,
1558                                   BTRFS_DEV_ITEM_KEY);
1559         if (ret) {
1560                 *devid_ret = 1;
1561         } else {
1562                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1563                                       path->slots[0]);
1564                 *devid_ret = found_key.offset + 1;
1565         }
1566         ret = 0;
1567 error:
1568         btrfs_free_path(path);
1569         return ret;
1570 }
1571
1572 /*
1573  * the device information is stored in the chunk root
1574  * the btrfs_device struct should be fully filled in
1575  */
1576 static int btrfs_add_device(struct btrfs_trans_handle *trans,
1577                             struct btrfs_root *root,
1578                             struct btrfs_device *device)
1579 {
1580         int ret;
1581         struct btrfs_path *path;
1582         struct btrfs_dev_item *dev_item;
1583         struct extent_buffer *leaf;
1584         struct btrfs_key key;
1585         unsigned long ptr;
1586
1587         root = root->fs_info->chunk_root;
1588
1589         path = btrfs_alloc_path();
1590         if (!path)
1591                 return -ENOMEM;
1592
1593         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1594         key.type = BTRFS_DEV_ITEM_KEY;
1595         key.offset = device->devid;
1596
1597         ret = btrfs_insert_empty_item(trans, root, path, &key,
1598                                       sizeof(*dev_item));
1599         if (ret)
1600                 goto out;
1601
1602         leaf = path->nodes[0];
1603         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1604
1605         btrfs_set_device_id(leaf, dev_item, device->devid);
1606         btrfs_set_device_generation(leaf, dev_item, 0);
1607         btrfs_set_device_type(leaf, dev_item, device->type);
1608         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1609         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1610         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1611         btrfs_set_device_total_bytes(leaf, dev_item,
1612                                      btrfs_device_get_disk_total_bytes(device));
1613         btrfs_set_device_bytes_used(leaf, dev_item,
1614                                     btrfs_device_get_bytes_used(device));
1615         btrfs_set_device_group(leaf, dev_item, 0);
1616         btrfs_set_device_seek_speed(leaf, dev_item, 0);
1617         btrfs_set_device_bandwidth(leaf, dev_item, 0);
1618         btrfs_set_device_start_offset(leaf, dev_item, 0);
1619
1620         ptr = btrfs_device_uuid(dev_item);
1621         write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1622         ptr = btrfs_device_fsid(dev_item);
1623         write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1624         btrfs_mark_buffer_dirty(leaf);
1625
1626         ret = 0;
1627 out:
1628         btrfs_free_path(path);
1629         return ret;
1630 }
1631
1632 /*
1633  * Function to update ctime/mtime for a given device path.
1634  * Mainly used for ctime/mtime based probe like libblkid.
1635  */
1636 static void update_dev_time(char *path_name)
1637 {
1638         struct file *filp;
1639
1640         filp = filp_open(path_name, O_RDWR, 0);
1641         if (IS_ERR(filp))
1642                 return;
1643         file_update_time(filp);
1644         filp_close(filp, NULL);
1645         return;
1646 }
1647
1648 static int btrfs_rm_dev_item(struct btrfs_root *root,
1649                              struct btrfs_device *device)
1650 {
1651         int ret;
1652         struct btrfs_path *path;
1653         struct btrfs_key key;
1654         struct btrfs_trans_handle *trans;
1655
1656         root = root->fs_info->chunk_root;
1657
1658         path = btrfs_alloc_path();
1659         if (!path)
1660                 return -ENOMEM;
1661
1662         trans = btrfs_start_transaction(root, 0);
1663         if (IS_ERR(trans)) {
1664                 btrfs_free_path(path);
1665                 return PTR_ERR(trans);
1666         }
1667         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1668         key.type = BTRFS_DEV_ITEM_KEY;
1669         key.offset = device->devid;
1670
1671         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1672         if (ret < 0)
1673                 goto out;
1674
1675         if (ret > 0) {
1676                 ret = -ENOENT;
1677                 goto out;
1678         }
1679
1680         ret = btrfs_del_item(trans, root, path);
1681         if (ret)
1682                 goto out;
1683 out:
1684         btrfs_free_path(path);
1685         btrfs_commit_transaction(trans, root);
1686         return ret;
1687 }
1688
1689 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
1690 {
1691         struct btrfs_device *device;
1692         struct btrfs_device *next_device;
1693         struct block_device *bdev;
1694         struct buffer_head *bh = NULL;
1695         struct btrfs_super_block *disk_super;
1696         struct btrfs_fs_devices *cur_devices;
1697         u64 all_avail;
1698         u64 devid;
1699         u64 num_devices;
1700         u8 *dev_uuid;
1701         unsigned seq;
1702         int ret = 0;
1703         bool clear_super = false;
1704
1705         mutex_lock(&uuid_mutex);
1706
1707         do {
1708                 seq = read_seqbegin(&root->fs_info->profiles_lock);
1709
1710                 all_avail = root->fs_info->avail_data_alloc_bits |
1711                             root->fs_info->avail_system_alloc_bits |
1712                             root->fs_info->avail_metadata_alloc_bits;
1713         } while (read_seqretry(&root->fs_info->profiles_lock, seq));
1714
1715         num_devices = root->fs_info->fs_devices->num_devices;
1716         btrfs_dev_replace_lock(&root->fs_info->dev_replace);
1717         if (btrfs_dev_replace_is_ongoing(&root->fs_info->dev_replace)) {
1718                 WARN_ON(num_devices < 1);
1719                 num_devices--;
1720         }
1721         btrfs_dev_replace_unlock(&root->fs_info->dev_replace);
1722
1723         if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) && num_devices <= 4) {
1724                 ret = BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET;
1725                 goto out;
1726         }
1727
1728         if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) && num_devices <= 2) {
1729                 ret = BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET;
1730                 goto out;
1731         }
1732
1733         if ((all_avail & BTRFS_BLOCK_GROUP_RAID5) &&
1734             root->fs_info->fs_devices->rw_devices <= 2) {
1735                 ret = BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET;
1736                 goto out;
1737         }
1738         if ((all_avail & BTRFS_BLOCK_GROUP_RAID6) &&
1739             root->fs_info->fs_devices->rw_devices <= 3) {
1740                 ret = BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET;
1741                 goto out;
1742         }
1743
1744         if (strcmp(device_path, "missing") == 0) {
1745                 struct list_head *devices;
1746                 struct btrfs_device *tmp;
1747
1748                 device = NULL;
1749                 devices = &root->fs_info->fs_devices->devices;
1750                 /*
1751                  * It is safe to read the devices since the volume_mutex
1752                  * is held.
1753                  */
1754                 list_for_each_entry(tmp, devices, dev_list) {
1755                         if (tmp->in_fs_metadata &&
1756                             !tmp->is_tgtdev_for_dev_replace &&
1757                             !tmp->bdev) {
1758                                 device = tmp;
1759                                 break;
1760                         }
1761                 }
1762                 bdev = NULL;
1763                 bh = NULL;
1764                 disk_super = NULL;
1765                 if (!device) {
1766                         ret = BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
1767                         goto out;
1768                 }
1769         } else {
1770                 ret = btrfs_get_bdev_and_sb(device_path,
1771                                             FMODE_WRITE | FMODE_EXCL,
1772                                             root->fs_info->bdev_holder, 0,
1773                                             &bdev, &bh);
1774                 if (ret)
1775                         goto out;
1776                 disk_super = (struct btrfs_super_block *)bh->b_data;
1777                 devid = btrfs_stack_device_id(&disk_super->dev_item);
1778                 dev_uuid = disk_super->dev_item.uuid;
1779                 device = btrfs_find_device(root->fs_info, devid, dev_uuid,
1780                                            disk_super->fsid);
1781                 if (!device) {
1782                         ret = -ENOENT;
1783                         goto error_brelse;
1784                 }
1785         }
1786
1787         if (device->is_tgtdev_for_dev_replace) {
1788                 ret = BTRFS_ERROR_DEV_TGT_REPLACE;
1789                 goto error_brelse;
1790         }
1791
1792         if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1793                 ret = BTRFS_ERROR_DEV_ONLY_WRITABLE;
1794                 goto error_brelse;
1795         }
1796
1797         if (device->writeable) {
1798                 lock_chunks(root);
1799                 list_del_init(&device->dev_alloc_list);
1800                 device->fs_devices->rw_devices--;
1801                 unlock_chunks(root);
1802                 clear_super = true;
1803         }
1804
1805         mutex_unlock(&uuid_mutex);
1806         ret = btrfs_shrink_device(device, 0);
1807         mutex_lock(&uuid_mutex);
1808         if (ret)
1809                 goto error_undo;
1810
1811         /*
1812          * TODO: the superblock still includes this device in its num_devices
1813          * counter although write_all_supers() is not locked out. This
1814          * could give a filesystem state which requires a degraded mount.
1815          */
1816         ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1817         if (ret)
1818                 goto error_undo;
1819
1820         device->in_fs_metadata = 0;
1821         btrfs_scrub_cancel_dev(root->fs_info, device);
1822
1823         /*
1824          * the device list mutex makes sure that we don't change
1825          * the device list while someone else is writing out all
1826          * the device supers. Whoever is writing all supers, should
1827          * lock the device list mutex before getting the number of
1828          * devices in the super block (super_copy). Conversely,
1829          * whoever updates the number of devices in the super block
1830          * (super_copy) should hold the device list mutex.
1831          */
1832
1833         cur_devices = device->fs_devices;
1834         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1835         list_del_rcu(&device->dev_list);
1836
1837         device->fs_devices->num_devices--;
1838         device->fs_devices->total_devices--;
1839
1840         if (device->missing)
1841                 device->fs_devices->missing_devices--;
1842
1843         next_device = list_entry(root->fs_info->fs_devices->devices.next,
1844                                  struct btrfs_device, dev_list);
1845         if (device->bdev == root->fs_info->sb->s_bdev)
1846                 root->fs_info->sb->s_bdev = next_device->bdev;
1847         if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1848                 root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1849
1850         if (device->bdev) {
1851                 device->fs_devices->open_devices--;
1852                 /* remove sysfs entry */
1853                 btrfs_sysfs_rm_device_link(root->fs_info->fs_devices, device);
1854         }
1855
1856         call_rcu(&device->rcu, free_device);
1857
1858         num_devices = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
1859         btrfs_set_super_num_devices(root->fs_info->super_copy, num_devices);
1860         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1861
1862         if (cur_devices->open_devices == 0) {
1863                 struct btrfs_fs_devices *fs_devices;
1864                 fs_devices = root->fs_info->fs_devices;
1865                 while (fs_devices) {
1866                         if (fs_devices->seed == cur_devices) {
1867                                 fs_devices->seed = cur_devices->seed;
1868                                 break;
1869                         }
1870                         fs_devices = fs_devices->seed;
1871                 }
1872                 cur_devices->seed = NULL;
1873                 __btrfs_close_devices(cur_devices);
1874                 free_fs_devices(cur_devices);
1875         }
1876
1877         root->fs_info->num_tolerated_disk_barrier_failures =
1878                 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
1879
1880         /*
1881          * at this point, the device is zero sized.  We want to
1882          * remove it from the devices list and zero out the old super
1883          */
1884         if (clear_super && disk_super) {
1885                 u64 bytenr;
1886                 int i;
1887
1888                 /* make sure this device isn't detected as part of
1889                  * the FS anymore
1890                  */
1891                 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1892                 set_buffer_dirty(bh);
1893                 sync_dirty_buffer(bh);
1894
1895                 /* clear the mirror copies of super block on the disk
1896                  * being removed, 0th copy is been taken care above and
1897                  * the below would take of the rest
1898                  */
1899                 for (i = 1; i < BTRFS_SUPER_MIRROR_MAX; i++) {
1900                         bytenr = btrfs_sb_offset(i);
1901                         if (bytenr + BTRFS_SUPER_INFO_SIZE >=
1902                                         i_size_read(bdev->bd_inode))
1903                                 break;
1904
1905                         brelse(bh);
1906                         bh = __bread(bdev, bytenr / 4096,
1907                                         BTRFS_SUPER_INFO_SIZE);
1908                         if (!bh)
1909                                 continue;
1910
1911                         disk_super = (struct btrfs_super_block *)bh->b_data;
1912
1913                         if (btrfs_super_bytenr(disk_super) != bytenr ||
1914                                 btrfs_super_magic(disk_super) != BTRFS_MAGIC) {
1915                                 continue;
1916                         }
1917                         memset(&disk_super->magic, 0,
1918                                                 sizeof(disk_super->magic));
1919                         set_buffer_dirty(bh);
1920                         sync_dirty_buffer(bh);
1921                 }
1922         }
1923
1924         ret = 0;
1925
1926         if (bdev) {
1927                 /* Notify udev that device has changed */
1928                 btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
1929
1930                 /* Update ctime/mtime for device path for libblkid */
1931                 update_dev_time(device_path);
1932         }
1933
1934 error_brelse:
1935         brelse(bh);
1936         if (bdev)
1937                 blkdev_put(bdev, FMODE_READ | FMODE_EXCL);
1938 out:
1939         mutex_unlock(&uuid_mutex);
1940         return ret;
1941 error_undo:
1942         if (device->writeable) {
1943                 lock_chunks(root);
1944                 list_add(&device->dev_alloc_list,
1945                          &root->fs_info->fs_devices->alloc_list);
1946                 device->fs_devices->rw_devices++;
1947                 unlock_chunks(root);
1948         }
1949         goto error_brelse;
1950 }
1951
1952 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_fs_info *fs_info,
1953                                         struct btrfs_device *srcdev)
1954 {
1955         struct btrfs_fs_devices *fs_devices;
1956
1957         WARN_ON(!mutex_is_locked(&fs_info->fs_devices->device_list_mutex));
1958
1959         /*
1960          * in case of fs with no seed, srcdev->fs_devices will point
1961          * to fs_devices of fs_info. However when the dev being replaced is
1962          * a seed dev it will point to the seed's local fs_devices. In short
1963          * srcdev will have its correct fs_devices in both the cases.
1964          */
1965         fs_devices = srcdev->fs_devices;
1966
1967         list_del_rcu(&srcdev->dev_list);
1968         list_del_rcu(&srcdev->dev_alloc_list);
1969         fs_devices->num_devices--;
1970         if (srcdev->missing)
1971                 fs_devices->missing_devices--;
1972
1973         if (srcdev->writeable) {
1974                 fs_devices->rw_devices--;
1975                 /* zero out the old super if it is writable */
1976                 btrfs_scratch_superblocks(srcdev->bdev, srcdev->name->str);
1977         }
1978
1979         if (srcdev->bdev)
1980                 fs_devices->open_devices--;
1981 }
1982
1983 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_fs_info *fs_info,
1984                                       struct btrfs_device *srcdev)
1985 {
1986         struct btrfs_fs_devices *fs_devices = srcdev->fs_devices;
1987
1988         call_rcu(&srcdev->rcu, free_device);
1989
1990         /*
1991          * unless fs_devices is seed fs, num_devices shouldn't go
1992          * zero
1993          */
1994         BUG_ON(!fs_devices->num_devices && !fs_devices->seeding);
1995
1996         /* if this is no devs we rather delete the fs_devices */
1997         if (!fs_devices->num_devices) {
1998                 struct btrfs_fs_devices *tmp_fs_devices;
1999
2000                 tmp_fs_devices = fs_info->fs_devices;
2001                 while (tmp_fs_devices) {
2002                         if (tmp_fs_devices->seed == fs_devices) {
2003                                 tmp_fs_devices->seed = fs_devices->seed;
2004                                 break;
2005                         }
2006                         tmp_fs_devices = tmp_fs_devices->seed;
2007                 }
2008                 fs_devices->seed = NULL;
2009                 __btrfs_close_devices(fs_devices);
2010                 free_fs_devices(fs_devices);
2011         }
2012 }
2013
2014 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
2015                                       struct btrfs_device *tgtdev)
2016 {
2017         struct btrfs_device *next_device;
2018
2019         mutex_lock(&uuid_mutex);
2020         WARN_ON(!tgtdev);
2021         mutex_lock(&fs_info->fs_devices->device_list_mutex);
2022
2023         btrfs_sysfs_rm_device_link(fs_info->fs_devices, tgtdev);
2024
2025         if (tgtdev->bdev) {
2026                 btrfs_scratch_superblocks(tgtdev->bdev, tgtdev->name->str);
2027                 fs_info->fs_devices->open_devices--;
2028         }
2029         fs_info->fs_devices->num_devices--;
2030
2031         next_device = list_entry(fs_info->fs_devices->devices.next,
2032                                  struct btrfs_device, dev_list);
2033         if (tgtdev->bdev == fs_info->sb->s_bdev)
2034                 fs_info->sb->s_bdev = next_device->bdev;
2035         if (tgtdev->bdev == fs_info->fs_devices->latest_bdev)
2036                 fs_info->fs_devices->latest_bdev = next_device->bdev;
2037         list_del_rcu(&tgtdev->dev_list);
2038
2039         call_rcu(&tgtdev->rcu, free_device);
2040
2041         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2042         mutex_unlock(&uuid_mutex);
2043 }
2044
2045 static int btrfs_find_device_by_path(struct btrfs_root *root, char *device_path,
2046                                      struct btrfs_device **device)
2047 {
2048         int ret = 0;
2049         struct btrfs_super_block *disk_super;
2050         u64 devid;
2051         u8 *dev_uuid;
2052         struct block_device *bdev;
2053         struct buffer_head *bh;
2054
2055         *device = NULL;
2056         ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
2057                                     root->fs_info->bdev_holder, 0, &bdev, &bh);
2058         if (ret)
2059                 return ret;
2060         disk_super = (struct btrfs_super_block *)bh->b_data;
2061         devid = btrfs_stack_device_id(&disk_super->dev_item);
2062         dev_uuid = disk_super->dev_item.uuid;
2063         *device = btrfs_find_device(root->fs_info, devid, dev_uuid,
2064                                     disk_super->fsid);
2065         brelse(bh);
2066         if (!*device)
2067                 ret = -ENOENT;
2068         blkdev_put(bdev, FMODE_READ);
2069         return ret;
2070 }
2071
2072 int btrfs_find_device_missing_or_by_path(struct btrfs_root *root,
2073                                          char *device_path,
2074                                          struct btrfs_device **device)
2075 {
2076         *device = NULL;
2077         if (strcmp(device_path, "missing") == 0) {
2078                 struct list_head *devices;
2079                 struct btrfs_device *tmp;
2080
2081                 devices = &root->fs_info->fs_devices->devices;
2082                 /*
2083                  * It is safe to read the devices since the volume_mutex
2084                  * is held by the caller.
2085                  */
2086                 list_for_each_entry(tmp, devices, dev_list) {
2087                         if (tmp->in_fs_metadata && !tmp->bdev) {
2088                                 *device = tmp;
2089                                 break;
2090                         }
2091                 }
2092
2093                 if (!*device)
2094                         return BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
2095
2096                 return 0;
2097         } else {
2098                 return btrfs_find_device_by_path(root, device_path, device);
2099         }
2100 }
2101
2102 /*
2103  * does all the dirty work required for changing file system's UUID.
2104  */
2105 static int btrfs_prepare_sprout(struct btrfs_root *root)
2106 {
2107         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2108         struct btrfs_fs_devices *old_devices;
2109         struct btrfs_fs_devices *seed_devices;
2110         struct btrfs_super_block *disk_super = root->fs_info->super_copy;
2111         struct btrfs_device *device;
2112         u64 super_flags;
2113
2114         BUG_ON(!mutex_is_locked(&uuid_mutex));
2115         if (!fs_devices->seeding)
2116                 return -EINVAL;
2117
2118         seed_devices = __alloc_fs_devices();
2119         if (IS_ERR(seed_devices))
2120                 return PTR_ERR(seed_devices);
2121
2122         old_devices = clone_fs_devices(fs_devices);
2123         if (IS_ERR(old_devices)) {
2124                 kfree(seed_devices);
2125                 return PTR_ERR(old_devices);
2126         }
2127
2128         list_add(&old_devices->list, &fs_uuids);
2129
2130         memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
2131         seed_devices->opened = 1;
2132         INIT_LIST_HEAD(&seed_devices->devices);
2133         INIT_LIST_HEAD(&seed_devices->alloc_list);
2134         mutex_init(&seed_devices->device_list_mutex);
2135
2136         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2137         list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
2138                               synchronize_rcu);
2139         list_for_each_entry(device, &seed_devices->devices, dev_list)
2140                 device->fs_devices = seed_devices;
2141
2142         lock_chunks(root);
2143         list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
2144         unlock_chunks(root);
2145
2146         fs_devices->seeding = 0;
2147         fs_devices->num_devices = 0;
2148         fs_devices->open_devices = 0;
2149         fs_devices->missing_devices = 0;
2150         fs_devices->rotating = 0;
2151         fs_devices->seed = seed_devices;
2152
2153         generate_random_uuid(fs_devices->fsid);
2154         memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2155         memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2156         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2157
2158         super_flags = btrfs_super_flags(disk_super) &
2159                       ~BTRFS_SUPER_FLAG_SEEDING;
2160         btrfs_set_super_flags(disk_super, super_flags);
2161
2162         return 0;
2163 }
2164
2165 /*
2166  * strore the expected generation for seed devices in device items.
2167  */
2168 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
2169                                struct btrfs_root *root)
2170 {
2171         struct btrfs_path *path;
2172         struct extent_buffer *leaf;
2173         struct btrfs_dev_item *dev_item;
2174         struct btrfs_device *device;
2175         struct btrfs_key key;
2176         u8 fs_uuid[BTRFS_UUID_SIZE];
2177         u8 dev_uuid[BTRFS_UUID_SIZE];
2178         u64 devid;
2179         int ret;
2180
2181         path = btrfs_alloc_path();
2182         if (!path)
2183                 return -ENOMEM;
2184
2185         root = root->fs_info->chunk_root;
2186         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2187         key.offset = 0;
2188         key.type = BTRFS_DEV_ITEM_KEY;
2189
2190         while (1) {
2191                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2192                 if (ret < 0)
2193                         goto error;
2194
2195                 leaf = path->nodes[0];
2196 next_slot:
2197                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2198                         ret = btrfs_next_leaf(root, path);
2199                         if (ret > 0)
2200                                 break;
2201                         if (ret < 0)
2202                                 goto error;
2203                         leaf = path->nodes[0];
2204                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2205                         btrfs_release_path(path);
2206                         continue;
2207                 }
2208
2209                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2210                 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
2211                     key.type != BTRFS_DEV_ITEM_KEY)
2212                         break;
2213
2214                 dev_item = btrfs_item_ptr(leaf, path->slots[0],
2215                                           struct btrfs_dev_item);
2216                 devid = btrfs_device_id(leaf, dev_item);
2217                 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
2218                                    BTRFS_UUID_SIZE);
2219                 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
2220                                    BTRFS_UUID_SIZE);
2221                 device = btrfs_find_device(root->fs_info, devid, dev_uuid,
2222                                            fs_uuid);
2223                 BUG_ON(!device); /* Logic error */
2224
2225                 if (device->fs_devices->seeding) {
2226                         btrfs_set_device_generation(leaf, dev_item,
2227                                                     device->generation);
2228                         btrfs_mark_buffer_dirty(leaf);
2229                 }
2230
2231                 path->slots[0]++;
2232                 goto next_slot;
2233         }
2234         ret = 0;
2235 error:
2236         btrfs_free_path(path);
2237         return ret;
2238 }
2239
2240 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
2241 {
2242         struct request_queue *q;
2243         struct btrfs_trans_handle *trans;
2244         struct btrfs_device *device;
2245         struct block_device *bdev;
2246         struct list_head *devices;
2247         struct super_block *sb = root->fs_info->sb;
2248         struct rcu_string *name;
2249         u64 tmp;
2250         int seeding_dev = 0;
2251         int ret = 0;
2252
2253         if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
2254                 return -EROFS;
2255
2256         bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2257                                   root->fs_info->bdev_holder);
2258         if (IS_ERR(bdev))
2259                 return PTR_ERR(bdev);
2260
2261         if (root->fs_info->fs_devices->seeding) {
2262                 seeding_dev = 1;
2263                 down_write(&sb->s_umount);
2264                 mutex_lock(&uuid_mutex);
2265         }
2266
2267         filemap_write_and_wait(bdev->bd_inode->i_mapping);
2268
2269         devices = &root->fs_info->fs_devices->devices;
2270
2271         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2272         list_for_each_entry(device, devices, dev_list) {
2273                 if (device->bdev == bdev) {
2274                         ret = -EEXIST;
2275                         mutex_unlock(
2276                                 &root->fs_info->fs_devices->device_list_mutex);
2277                         goto error;
2278                 }
2279         }
2280         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2281
2282         device = btrfs_alloc_device(root->fs_info, NULL, NULL);
2283         if (IS_ERR(device)) {
2284                 /* we can safely leave the fs_devices entry around */
2285                 ret = PTR_ERR(device);
2286                 goto error;
2287         }
2288
2289         name = rcu_string_strdup(device_path, GFP_NOFS);
2290         if (!name) {
2291                 kfree(device);
2292                 ret = -ENOMEM;
2293                 goto error;
2294         }
2295         rcu_assign_pointer(device->name, name);
2296
2297         trans = btrfs_start_transaction(root, 0);
2298         if (IS_ERR(trans)) {
2299                 rcu_string_free(device->name);
2300                 kfree(device);
2301                 ret = PTR_ERR(trans);
2302                 goto error;
2303         }
2304
2305         q = bdev_get_queue(bdev);
2306         if (blk_queue_discard(q))
2307                 device->can_discard = 1;
2308         device->writeable = 1;
2309         device->generation = trans->transid;
2310         device->io_width = root->sectorsize;
2311         device->io_align = root->sectorsize;
2312         device->sector_size = root->sectorsize;
2313         device->total_bytes = i_size_read(bdev->bd_inode);
2314         device->disk_total_bytes = device->total_bytes;
2315         device->commit_total_bytes = device->total_bytes;
2316         device->dev_root = root->fs_info->dev_root;
2317         device->bdev = bdev;
2318         device->in_fs_metadata = 1;
2319         device->is_tgtdev_for_dev_replace = 0;
2320         device->mode = FMODE_EXCL;
2321         device->dev_stats_valid = 1;
2322         set_blocksize(device->bdev, 4096);
2323
2324         if (seeding_dev) {
2325                 sb->s_flags &= ~MS_RDONLY;
2326                 ret = btrfs_prepare_sprout(root);
2327                 BUG_ON(ret); /* -ENOMEM */
2328         }
2329
2330         device->fs_devices = root->fs_info->fs_devices;
2331
2332         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2333         lock_chunks(root);
2334         list_add_rcu(&device->dev_list, &root->fs_info->fs_devices->devices);
2335         list_add(&device->dev_alloc_list,
2336                  &root->fs_info->fs_devices->alloc_list);
2337         root->fs_info->fs_devices->num_devices++;
2338         root->fs_info->fs_devices->open_devices++;
2339         root->fs_info->fs_devices->rw_devices++;
2340         root->fs_info->fs_devices->total_devices++;
2341         root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
2342
2343         spin_lock(&root->fs_info->free_chunk_lock);
2344         root->fs_info->free_chunk_space += device->total_bytes;
2345         spin_unlock(&root->fs_info->free_chunk_lock);
2346
2347         if (!blk_queue_nonrot(bdev_get_queue(bdev)))
2348                 root->fs_info->fs_devices->rotating = 1;
2349
2350         tmp = btrfs_super_total_bytes(root->fs_info->super_copy);
2351         btrfs_set_super_total_bytes(root->fs_info->super_copy,
2352                                     tmp + device->total_bytes);
2353
2354         tmp = btrfs_super_num_devices(root->fs_info->super_copy);
2355         btrfs_set_super_num_devices(root->fs_info->super_copy,
2356                                     tmp + 1);
2357
2358         /* add sysfs device entry */
2359         btrfs_sysfs_add_device_link(root->fs_info->fs_devices, device);
2360
2361         /*
2362          * we've got more storage, clear any full flags on the space
2363          * infos
2364          */
2365         btrfs_clear_space_info_full(root->fs_info);
2366
2367         unlock_chunks(root);
2368         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2369
2370         if (seeding_dev) {
2371                 lock_chunks(root);
2372                 ret = init_first_rw_device(trans, root, device);
2373                 unlock_chunks(root);
2374                 if (ret) {
2375                         btrfs_abort_transaction(trans, root, ret);
2376                         goto error_trans;
2377                 }
2378         }
2379
2380         ret = btrfs_add_device(trans, root, device);
2381         if (ret) {
2382                 btrfs_abort_transaction(trans, root, ret);
2383                 goto error_trans;
2384         }
2385
2386         if (seeding_dev) {
2387                 char fsid_buf[BTRFS_UUID_UNPARSED_SIZE];
2388
2389                 ret = btrfs_finish_sprout(trans, root);
2390                 if (ret) {
2391                         btrfs_abort_transaction(trans, root, ret);
2392                         goto error_trans;
2393                 }
2394
2395                 /* Sprouting would change fsid of the mounted root,
2396                  * so rename the fsid on the sysfs
2397                  */
2398                 snprintf(fsid_buf, BTRFS_UUID_UNPARSED_SIZE, "%pU",
2399                                                 root->fs_info->fsid);
2400                 if (kobject_rename(&root->fs_info->fs_devices->fsid_kobj,
2401                                                                 fsid_buf))
2402                         btrfs_warn(root->fs_info,
2403                                 "sysfs: failed to create fsid for sprout");
2404         }
2405
2406         root->fs_info->num_tolerated_disk_barrier_failures =
2407                 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
2408         ret = btrfs_commit_transaction(trans, root);
2409
2410         if (seeding_dev) {
2411                 mutex_unlock(&uuid_mutex);
2412                 up_write(&sb->s_umount);
2413
2414                 if (ret) /* transaction commit */
2415                         return ret;
2416
2417                 ret = btrfs_relocate_sys_chunks(root);
2418                 if (ret < 0)
2419                         btrfs_std_error(root->fs_info, ret,
2420                                     "Failed to relocate sys chunks after "
2421                                     "device initialization. This can be fixed "
2422                                     "using the \"btrfs balance\" command.");
2423                 trans = btrfs_attach_transaction(root);
2424                 if (IS_ERR(trans)) {
2425                         if (PTR_ERR(trans) == -ENOENT)
2426                                 return 0;
2427                         return PTR_ERR(trans);
2428                 }
2429                 ret = btrfs_commit_transaction(trans, root);
2430         }
2431
2432         /* Update ctime/mtime for libblkid */
2433         update_dev_time(device_path);
2434         return ret;
2435
2436 error_trans:
2437         btrfs_end_transaction(trans, root);
2438         rcu_string_free(device->name);
2439         btrfs_sysfs_rm_device_link(root->fs_info->fs_devices, device);
2440         kfree(device);
2441 error:
2442         blkdev_put(bdev, FMODE_EXCL);
2443         if (seeding_dev) {
2444                 mutex_unlock(&uuid_mutex);
2445                 up_write(&sb->s_umount);
2446         }
2447         return ret;
2448 }
2449
2450 int btrfs_init_dev_replace_tgtdev(struct btrfs_root *root, char *device_path,
2451                                   struct btrfs_device *srcdev,
2452                                   struct btrfs_device **device_out)
2453 {
2454         struct request_queue *q;
2455         struct btrfs_device *device;
2456         struct block_device *bdev;
2457         struct btrfs_fs_info *fs_info = root->fs_info;
2458         struct list_head *devices;
2459         struct rcu_string *name;
2460         u64 devid = BTRFS_DEV_REPLACE_DEVID;
2461         int ret = 0;
2462
2463         *device_out = NULL;
2464         if (fs_info->fs_devices->seeding) {
2465                 btrfs_err(fs_info, "the filesystem is a seed filesystem!");
2466                 return -EINVAL;
2467         }
2468
2469         bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2470                                   fs_info->bdev_holder);
2471         if (IS_ERR(bdev)) {
2472                 btrfs_err(fs_info, "target device %s is invalid!", device_path);
2473                 return PTR_ERR(bdev);
2474         }
2475
2476         filemap_write_and_wait(bdev->bd_inode->i_mapping);
2477
2478         devices = &fs_info->fs_devices->devices;
2479         list_for_each_entry(device, devices, dev_list) {
2480                 if (device->bdev == bdev) {
2481                         btrfs_err(fs_info, "target device is in the filesystem!");
2482                         ret = -EEXIST;
2483                         goto error;
2484                 }
2485         }
2486
2487
2488         if (i_size_read(bdev->bd_inode) <
2489             btrfs_device_get_total_bytes(srcdev)) {
2490                 btrfs_err(fs_info, "target device is smaller than source device!");
2491                 ret = -EINVAL;
2492                 goto error;
2493         }
2494
2495
2496         device = btrfs_alloc_device(NULL, &devid, NULL);
2497         if (IS_ERR(device)) {
2498                 ret = PTR_ERR(device);
2499                 goto error;
2500         }
2501
2502         name = rcu_string_strdup(device_path, GFP_NOFS);
2503         if (!name) {
2504                 kfree(device);
2505                 ret = -ENOMEM;
2506                 goto error;
2507         }
2508         rcu_assign_pointer(device->name, name);
2509
2510         q = bdev_get_queue(bdev);
2511         if (blk_queue_discard(q))
2512                 device->can_discard = 1;
2513         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2514         device->writeable = 1;
2515         device->generation = 0;
2516         device->io_width = root->sectorsize;
2517         device->io_align = root->sectorsize;
2518         device->sector_size = root->sectorsize;
2519         device->total_bytes = btrfs_device_get_total_bytes(srcdev);
2520         device->disk_total_bytes = btrfs_device_get_disk_total_bytes(srcdev);
2521         device->bytes_used = btrfs_device_get_bytes_used(srcdev);
2522         ASSERT(list_empty(&srcdev->resized_list));
2523         device->commit_total_bytes = srcdev->commit_total_bytes;
2524         device->commit_bytes_used = device->bytes_used;
2525         device->dev_root = fs_info->dev_root;
2526         device->bdev = bdev;
2527         device->in_fs_metadata = 1;
2528         device->is_tgtdev_for_dev_replace = 1;
2529         device->mode = FMODE_EXCL;
2530         device->dev_stats_valid = 1;
2531         set_blocksize(device->bdev, 4096);
2532         device->fs_devices = fs_info->fs_devices;
2533         list_add(&device->dev_list, &fs_info->fs_devices->devices);
2534         fs_info->fs_devices->num_devices++;
2535         fs_info->fs_devices->open_devices++;
2536         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2537
2538         *device_out = device;
2539         return ret;
2540
2541 error:
2542         blkdev_put(bdev, FMODE_EXCL);
2543         return ret;
2544 }
2545
2546 void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info *fs_info,
2547                                               struct btrfs_device *tgtdev)
2548 {
2549         WARN_ON(fs_info->fs_devices->rw_devices == 0);
2550         tgtdev->io_width = fs_info->dev_root->sectorsize;
2551         tgtdev->io_align = fs_info->dev_root->sectorsize;
2552         tgtdev->sector_size = fs_info->dev_root->sectorsize;
2553         tgtdev->dev_root = fs_info->dev_root;
2554         tgtdev->in_fs_metadata = 1;
2555 }
2556
2557 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2558                                         struct btrfs_device *device)
2559 {
2560         int ret;
2561         struct btrfs_path *path;
2562         struct btrfs_root *root;
2563         struct btrfs_dev_item *dev_item;
2564         struct extent_buffer *leaf;
2565         struct btrfs_key key;
2566
2567         root = device->dev_root->fs_info->chunk_root;
2568
2569         path = btrfs_alloc_path();
2570         if (!path)
2571                 return -ENOMEM;
2572
2573         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2574         key.type = BTRFS_DEV_ITEM_KEY;
2575         key.offset = device->devid;
2576
2577         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2578         if (ret < 0)
2579                 goto out;
2580
2581         if (ret > 0) {
2582                 ret = -ENOENT;
2583                 goto out;
2584         }
2585
2586         leaf = path->nodes[0];
2587         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2588
2589         btrfs_set_device_id(leaf, dev_item, device->devid);
2590         btrfs_set_device_type(leaf, dev_item, device->type);
2591         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2592         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2593         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2594         btrfs_set_device_total_bytes(leaf, dev_item,
2595                                      btrfs_device_get_disk_total_bytes(device));
2596         btrfs_set_device_bytes_used(leaf, dev_item,
2597                                     btrfs_device_get_bytes_used(device));
2598         btrfs_mark_buffer_dirty(leaf);
2599
2600 out:
2601         btrfs_free_path(path);
2602         return ret;
2603 }
2604
2605 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2606                       struct btrfs_device *device, u64 new_size)
2607 {
2608         struct btrfs_super_block *super_copy =
2609                 device->dev_root->fs_info->super_copy;
2610         struct btrfs_fs_devices *fs_devices;
2611         u64 old_total;
2612         u64 diff;
2613
2614         if (!device->writeable)
2615                 return -EACCES;
2616
2617         lock_chunks(device->dev_root);
2618         old_total = btrfs_super_total_bytes(super_copy);
2619         diff = new_size - device->total_bytes;
2620
2621         if (new_size <= device->total_bytes ||
2622             device->is_tgtdev_for_dev_replace) {
2623                 unlock_chunks(device->dev_root);
2624                 return -EINVAL;
2625         }
2626
2627         fs_devices = device->dev_root->fs_info->fs_devices;
2628
2629         btrfs_set_super_total_bytes(super_copy, old_total + diff);
2630         device->fs_devices->total_rw_bytes += diff;
2631
2632         btrfs_device_set_total_bytes(device, new_size);
2633         btrfs_device_set_disk_total_bytes(device, new_size);
2634         btrfs_clear_space_info_full(device->dev_root->fs_info);
2635         if (list_empty(&device->resized_list))
2636                 list_add_tail(&device->resized_list,
2637                               &fs_devices->resized_devices);
2638         unlock_chunks(device->dev_root);
2639
2640         return btrfs_update_device(trans, device);
2641 }
2642
2643 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
2644                             struct btrfs_root *root, u64 chunk_objectid,
2645                             u64 chunk_offset)
2646 {
2647         int ret;
2648         struct btrfs_path *path;
2649         struct btrfs_key key;
2650
2651         root = root->fs_info->chunk_root;
2652         path = btrfs_alloc_path();
2653         if (!path)
2654                 return -ENOMEM;
2655
2656         key.objectid = chunk_objectid;
2657         key.offset = chunk_offset;
2658         key.type = BTRFS_CHUNK_ITEM_KEY;
2659
2660         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2661         if (ret < 0)
2662                 goto out;
2663         else if (ret > 0) { /* Logic error or corruption */
2664                 btrfs_std_error(root->fs_info, -ENOENT,
2665                             "Failed lookup while freeing chunk.");
2666                 ret = -ENOENT;
2667                 goto out;
2668         }
2669
2670         ret = btrfs_del_item(trans, root, path);
2671         if (ret < 0)
2672                 btrfs_std_error(root->fs_info, ret,
2673                             "Failed to delete chunk item.");
2674 out:
2675         btrfs_free_path(path);
2676         return ret;
2677 }
2678
2679 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
2680                         chunk_offset)
2681 {
2682         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
2683         struct btrfs_disk_key *disk_key;
2684         struct btrfs_chunk *chunk;
2685         u8 *ptr;
2686         int ret = 0;
2687         u32 num_stripes;
2688         u32 array_size;
2689         u32 len = 0;
2690         u32 cur;
2691         struct btrfs_key key;
2692
2693         lock_chunks(root);
2694         array_size = btrfs_super_sys_array_size(super_copy);
2695
2696         ptr = super_copy->sys_chunk_array;
2697         cur = 0;
2698
2699         while (cur < array_size) {
2700                 disk_key = (struct btrfs_disk_key *)ptr;
2701                 btrfs_disk_key_to_cpu(&key, disk_key);
2702
2703                 len = sizeof(*disk_key);
2704
2705                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2706                         chunk = (struct btrfs_chunk *)(ptr + len);
2707                         num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2708                         len += btrfs_chunk_item_size(num_stripes);
2709                 } else {
2710                         ret = -EIO;
2711                         break;
2712                 }
2713                 if (key.objectid == chunk_objectid &&
2714                     key.offset == chunk_offset) {
2715                         memmove(ptr, ptr + len, array_size - (cur + len));
2716                         array_size -= len;
2717                         btrfs_set_super_sys_array_size(super_copy, array_size);
2718                 } else {
2719                         ptr += len;
2720                         cur += len;
2721                 }
2722         }
2723         unlock_chunks(root);
2724         return ret;
2725 }
2726
2727 int btrfs_remove_chunk(struct btrfs_trans_handle *trans,
2728                        struct btrfs_root *root, u64 chunk_offset)
2729 {
2730         struct extent_map_tree *em_tree;
2731         struct extent_map *em;
2732         struct btrfs_root *extent_root = root->fs_info->extent_root;
2733         struct map_lookup *map;
2734         u64 dev_extent_len = 0;
2735         u64 chunk_objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2736         int i, ret = 0;
2737
2738         /* Just in case */
2739         root = root->fs_info->chunk_root;
2740         em_tree = &root->fs_info->mapping_tree.map_tree;
2741
2742         read_lock(&em_tree->lock);
2743         em = lookup_extent_mapping(em_tree, chunk_offset, 1);
2744         read_unlock(&em_tree->lock);
2745
2746         if (!em || em->start > chunk_offset ||
2747             em->start + em->len < chunk_offset) {
2748                 /*
2749                  * This is a logic error, but we don't want to just rely on the
2750                  * user having built with ASSERT enabled, so if ASSERT doens't
2751                  * do anything we still error out.
2752                  */
2753                 ASSERT(0);
2754                 if (em)
2755                         free_extent_map(em);
2756                 return -EINVAL;
2757         }
2758         map = (struct map_lookup *)em->bdev;
2759         lock_chunks(root->fs_info->chunk_root);
2760         check_system_chunk(trans, extent_root, map->type);
2761         unlock_chunks(root->fs_info->chunk_root);
2762
2763         for (i = 0; i < map->num_stripes; i++) {
2764                 struct btrfs_device *device = map->stripes[i].dev;
2765                 ret = btrfs_free_dev_extent(trans, device,
2766                                             map->stripes[i].physical,
2767                                             &dev_extent_len);
2768                 if (ret) {
2769                         btrfs_abort_transaction(trans, root, ret);
2770                         goto out;
2771                 }
2772
2773                 if (device->bytes_used > 0) {
2774                         lock_chunks(root);
2775                         btrfs_device_set_bytes_used(device,
2776                                         device->bytes_used - dev_extent_len);
2777                         spin_lock(&root->fs_info->free_chunk_lock);
2778                         root->fs_info->free_chunk_space += dev_extent_len;
2779                         spin_unlock(&root->fs_info->free_chunk_lock);
2780                         btrfs_clear_space_info_full(root->fs_info);
2781                         unlock_chunks(root);
2782                 }
2783
2784                 if (map->stripes[i].dev) {
2785                         ret = btrfs_update_device(trans, map->stripes[i].dev);
2786                         if (ret) {
2787                                 btrfs_abort_transaction(trans, root, ret);
2788                                 goto out;
2789                         }
2790                 }
2791         }
2792         ret = btrfs_free_chunk(trans, root, chunk_objectid, chunk_offset);
2793         if (ret) {
2794                 btrfs_abort_transaction(trans, root, ret);
2795                 goto out;
2796         }
2797
2798         trace_btrfs_chunk_free(root, map, chunk_offset, em->len);
2799
2800         if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2801                 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
2802                 if (ret) {
2803                         btrfs_abort_transaction(trans, root, ret);
2804                         goto out;
2805                 }
2806         }
2807
2808         ret = btrfs_remove_block_group(trans, extent_root, chunk_offset, em);
2809         if (ret) {
2810                 btrfs_abort_transaction(trans, extent_root, ret);
2811                 goto out;
2812         }
2813
2814 out:
2815         /* once for us */
2816         free_extent_map(em);
2817         return ret;
2818 }
2819
2820 static int btrfs_relocate_chunk(struct btrfs_root *root, u64 chunk_offset)
2821 {
2822         struct btrfs_root *extent_root;
2823         struct btrfs_trans_handle *trans;
2824         int ret;
2825
2826         root = root->fs_info->chunk_root;
2827         extent_root = root->fs_info->extent_root;
2828
2829         /*
2830          * Prevent races with automatic removal of unused block groups.
2831          * After we relocate and before we remove the chunk with offset
2832          * chunk_offset, automatic removal of the block group can kick in,
2833          * resulting in a failure when calling btrfs_remove_chunk() below.
2834          *
2835          * Make sure to acquire this mutex before doing a tree search (dev
2836          * or chunk trees) to find chunks. Otherwise the cleaner kthread might
2837          * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after
2838          * we release the path used to search the chunk/dev tree and before
2839          * the current task acquires this mutex and calls us.
2840          */
2841         ASSERT(mutex_is_locked(&root->fs_info->delete_unused_bgs_mutex));
2842
2843         ret = btrfs_can_relocate(extent_root, chunk_offset);
2844         if (ret)
2845                 return -ENOSPC;
2846
2847         /* step one, relocate all the extents inside this chunk */
2848         btrfs_scrub_pause(root);
2849         ret = btrfs_relocate_block_group(extent_root, chunk_offset);
2850         btrfs_scrub_continue(root);
2851         if (ret)
2852                 return ret;
2853
2854         trans = btrfs_start_trans_remove_block_group(root->fs_info,
2855                                                      chunk_offset);
2856         if (IS_ERR(trans)) {
2857                 ret = PTR_ERR(trans);
2858                 btrfs_std_error(root->fs_info, ret, NULL);
2859                 return ret;
2860         }
2861
2862         /*
2863          * step two, delete the device extents and the
2864          * chunk tree entries
2865          */
2866         ret = btrfs_remove_chunk(trans, root, chunk_offset);
2867         btrfs_end_transaction(trans, root);
2868         return ret;
2869 }
2870
2871 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
2872 {
2873         struct btrfs_root *chunk_root = root->fs_info->chunk_root;
2874         struct btrfs_path *path;
2875         struct extent_buffer *leaf;
2876         struct btrfs_chunk *chunk;
2877         struct btrfs_key key;
2878         struct btrfs_key found_key;
2879         u64 chunk_type;
2880         bool retried = false;
2881         int failed = 0;
2882         int ret;
2883
2884         path = btrfs_alloc_path();
2885         if (!path)
2886                 return -ENOMEM;
2887
2888 again:
2889         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2890         key.offset = (u64)-1;
2891         key.type = BTRFS_CHUNK_ITEM_KEY;
2892
2893         while (1) {
2894                 mutex_lock(&root->fs_info->delete_unused_bgs_mutex);
2895                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2896                 if (ret < 0) {
2897                         mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
2898                         goto error;
2899                 }
2900                 BUG_ON(ret == 0); /* Corruption */
2901
2902                 ret = btrfs_previous_item(chunk_root, path, key.objectid,
2903                                           key.type);
2904                 if (ret)
2905                         mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
2906                 if (ret < 0)
2907                         goto error;
2908                 if (ret > 0)
2909                         break;
2910
2911                 leaf = path->nodes[0];
2912                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2913
2914                 chunk = btrfs_item_ptr(leaf, path->slots[0],
2915                                        struct btrfs_chunk);
2916                 chunk_type = btrfs_chunk_type(leaf, chunk);
2917                 btrfs_release_path(path);
2918
2919                 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
2920                         ret = btrfs_relocate_chunk(chunk_root,
2921                                                    found_key.offset);
2922                         if (ret == -ENOSPC)
2923                                 failed++;
2924                         else
2925                                 BUG_ON(ret);
2926                 }
2927                 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
2928
2929                 if (found_key.offset == 0)
2930                         break;
2931                 key.offset = found_key.offset - 1;
2932         }
2933         ret = 0;
2934         if (failed && !retried) {
2935                 failed = 0;
2936                 retried = true;
2937                 goto again;
2938         } else if (WARN_ON(failed && retried)) {
2939                 ret = -ENOSPC;
2940         }
2941 error:
2942         btrfs_free_path(path);
2943         return ret;
2944 }
2945
2946 static int insert_balance_item(struct btrfs_root *root,
2947                                struct btrfs_balance_control *bctl)
2948 {
2949         struct btrfs_trans_handle *trans;
2950         struct btrfs_balance_item *item;
2951         struct btrfs_disk_balance_args disk_bargs;
2952         struct btrfs_path *path;
2953         struct extent_buffer *leaf;
2954         struct btrfs_key key;
2955         int ret, err;
2956
2957         path = btrfs_alloc_path();
2958         if (!path)
2959                 return -ENOMEM;
2960
2961         trans = btrfs_start_transaction(root, 0);
2962         if (IS_ERR(trans)) {
2963                 btrfs_free_path(path);
2964                 return PTR_ERR(trans);
2965         }
2966
2967         key.objectid = BTRFS_BALANCE_OBJECTID;
2968         key.type = BTRFS_BALANCE_ITEM_KEY;
2969         key.offset = 0;
2970
2971         ret = btrfs_insert_empty_item(trans, root, path, &key,
2972                                       sizeof(*item));
2973         if (ret)
2974                 goto out;
2975
2976         leaf = path->nodes[0];
2977         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
2978
2979         memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
2980
2981         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
2982         btrfs_set_balance_data(leaf, item, &disk_bargs);
2983         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
2984         btrfs_set_balance_meta(leaf, item, &disk_bargs);
2985         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
2986         btrfs_set_balance_sys(leaf, item, &disk_bargs);
2987
2988         btrfs_set_balance_flags(leaf, item, bctl->flags);
2989
2990         btrfs_mark_buffer_dirty(leaf);
2991 out:
2992         btrfs_free_path(path);
2993         err = btrfs_commit_transaction(trans, root);
2994         if (err && !ret)
2995                 ret = err;
2996         return ret;
2997 }
2998
2999 static int del_balance_item(struct btrfs_root *root)
3000 {
3001         struct btrfs_trans_handle *trans;
3002         struct btrfs_path *path;
3003         struct btrfs_key key;
3004         int ret, err;
3005
3006         path = btrfs_alloc_path();
3007         if (!path)
3008                 return -ENOMEM;
3009
3010         trans = btrfs_start_transaction(root, 0);
3011         if (IS_ERR(trans)) {
3012                 btrfs_free_path(path);
3013                 return PTR_ERR(trans);
3014         }
3015
3016         key.objectid = BTRFS_BALANCE_OBJECTID;
3017         key.type = BTRFS_BALANCE_ITEM_KEY;
3018         key.offset = 0;
3019
3020         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3021         if (ret < 0)
3022                 goto out;
3023         if (ret > 0) {
3024                 ret = -ENOENT;
3025                 goto out;
3026         }
3027
3028         ret = btrfs_del_item(trans, root, path);
3029 out:
3030         btrfs_free_path(path);
3031         err = btrfs_commit_transaction(trans, root);
3032         if (err && !ret)
3033                 ret = err;
3034         return ret;
3035 }
3036
3037 /*
3038  * This is a heuristic used to reduce the number of chunks balanced on
3039  * resume after balance was interrupted.
3040  */
3041 static void update_balance_args(struct btrfs_balance_control *bctl)
3042 {
3043         /*
3044          * Turn on soft mode for chunk types that were being converted.
3045          */
3046         if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
3047                 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
3048         if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
3049                 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
3050         if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
3051                 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
3052
3053         /*
3054          * Turn on usage filter if is not already used.  The idea is
3055          * that chunks that we have already balanced should be
3056          * reasonably full.  Don't do it for chunks that are being
3057          * converted - that will keep us from relocating unconverted
3058          * (albeit full) chunks.
3059          */
3060         if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3061             !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3062             !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3063                 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
3064                 bctl->data.usage = 90;
3065         }
3066         if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3067             !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3068             !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3069                 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
3070                 bctl->sys.usage = 90;
3071         }
3072         if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3073             !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3074             !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3075                 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
3076                 bctl->meta.usage = 90;
3077         }
3078 }
3079
3080 /*
3081  * Should be called with both balance and volume mutexes held to
3082  * serialize other volume operations (add_dev/rm_dev/resize) with
3083  * restriper.  Same goes for unset_balance_control.
3084  */
3085 static void set_balance_control(struct btrfs_balance_control *bctl)
3086 {
3087         struct btrfs_fs_info *fs_info = bctl->fs_info;
3088
3089         BUG_ON(fs_info->balance_ctl);
3090
3091         spin_lock(&fs_info->balance_lock);
3092         fs_info->balance_ctl = bctl;
3093         spin_unlock(&fs_info->balance_lock);
3094 }
3095
3096 static void unset_balance_control(struct btrfs_fs_info *fs_info)
3097 {
3098         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3099
3100         BUG_ON(!fs_info->balance_ctl);
3101
3102         spin_lock(&fs_info->balance_lock);
3103         fs_info->balance_ctl = NULL;
3104         spin_unlock(&fs_info->balance_lock);
3105
3106         kfree(bctl);
3107 }
3108
3109 /*
3110  * Balance filters.  Return 1 if chunk should be filtered out
3111  * (should not be balanced).
3112  */
3113 static int chunk_profiles_filter(u64 chunk_type,
3114                                  struct btrfs_balance_args *bargs)
3115 {
3116         chunk_type = chunk_to_extended(chunk_type) &
3117                                 BTRFS_EXTENDED_PROFILE_MASK;
3118
3119         if (bargs->profiles & chunk_type)
3120                 return 0;
3121
3122         return 1;
3123 }
3124
3125 static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
3126                               struct btrfs_balance_args *bargs)
3127 {
3128         struct btrfs_block_group_cache *cache;
3129         u64 chunk_used;
3130         u64 user_thresh_min;
3131         u64 user_thresh_max;
3132         int ret = 1;
3133
3134         cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3135         chunk_used = btrfs_block_group_used(&cache->item);
3136
3137         if (bargs->usage_min == 0)
3138                 user_thresh_min = 0;
3139         else
3140                 user_thresh_min = div_factor_fine(cache->key.offset,
3141                                         bargs->usage_min);
3142
3143         if (bargs->usage_max == 0)
3144                 user_thresh_max = 1;
3145         else if (bargs->usage_max > 100)
3146                 user_thresh_max = cache->key.offset;
3147         else
3148                 user_thresh_max = div_factor_fine(cache->key.offset,
3149                                         bargs->usage_max);
3150
3151         if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max)
3152                 ret = 0;
3153
3154         btrfs_put_block_group(cache);
3155         return ret;
3156 }
3157
3158 static int chunk_usage_filter(struct btrfs_fs_info *fs_info,
3159                 u64 chunk_offset, struct btrfs_balance_args *bargs)
3160 {
3161         struct btrfs_block_group_cache *cache;
3162         u64 chunk_used, user_thresh;
3163         int ret = 1;
3164
3165         cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3166         chunk_used = btrfs_block_group_used(&cache->item);
3167
3168         if (bargs->usage_min == 0)
3169                 user_thresh = 1;
3170         else if (bargs->usage > 100)
3171                 user_thresh = cache->key.offset;
3172         else
3173                 user_thresh = div_factor_fine(cache->key.offset,
3174                                               bargs->usage);
3175
3176         if (chunk_used < user_thresh)
3177                 ret = 0;
3178
3179         btrfs_put_block_group(cache);
3180         return ret;
3181 }
3182
3183 static int chunk_devid_filter(struct extent_buffer *leaf,
3184                               struct btrfs_chunk *chunk,
3185                               struct btrfs_balance_args *bargs)
3186 {
3187         struct btrfs_stripe *stripe;
3188         int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3189         int i;
3190
3191         for (i = 0; i < num_stripes; i++) {
3192                 stripe = btrfs_stripe_nr(chunk, i);
3193                 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
3194                         return 0;
3195         }
3196
3197         return 1;
3198 }
3199
3200 /* [pstart, pend) */
3201 static int chunk_drange_filter(struct extent_buffer *leaf,
3202                                struct btrfs_chunk *chunk,
3203                                u64 chunk_offset,
3204                                struct btrfs_balance_args *bargs)
3205 {
3206         struct btrfs_stripe *stripe;
3207         int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3208         u64 stripe_offset;
3209         u64 stripe_length;
3210         int factor;
3211         int i;
3212
3213         if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
3214                 return 0;
3215
3216         if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP |
3217              BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) {
3218                 factor = num_stripes / 2;
3219         } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) {
3220                 factor = num_stripes - 1;
3221         } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) {
3222                 factor = num_stripes - 2;
3223         } else {
3224                 factor = num_stripes;
3225         }
3226
3227         for (i = 0; i < num_stripes; i++) {
3228                 stripe = btrfs_stripe_nr(chunk, i);
3229                 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
3230                         continue;
3231
3232                 stripe_offset = btrfs_stripe_offset(leaf, stripe);
3233                 stripe_length = btrfs_chunk_length(leaf, chunk);
3234                 stripe_length = div_u64(stripe_length, factor);
3235
3236                 if (stripe_offset < bargs->pend &&
3237                     stripe_offset + stripe_length > bargs->pstart)
3238                         return 0;
3239         }
3240
3241         return 1;
3242 }
3243
3244 /* [vstart, vend) */
3245 static int chunk_vrange_filter(struct extent_buffer *leaf,
3246                                struct btrfs_chunk *chunk,
3247                                u64 chunk_offset,
3248                                struct btrfs_balance_args *bargs)
3249 {
3250         if (chunk_offset < bargs->vend &&
3251             chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
3252                 /* at least part of the chunk is inside this vrange */
3253                 return 0;
3254
3255         return 1;
3256 }
3257
3258 static int chunk_stripes_range_filter(struct extent_buffer *leaf,
3259                                struct btrfs_chunk *chunk,
3260                                struct btrfs_balance_args *bargs)
3261 {
3262         int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3263
3264         if (bargs->stripes_min <= num_stripes
3265                         && num_stripes <= bargs->stripes_max)
3266                 return 0;
3267
3268         return 1;
3269 }
3270
3271 static int chunk_soft_convert_filter(u64 chunk_type,
3272                                      struct btrfs_balance_args *bargs)
3273 {
3274         if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
3275                 return 0;
3276
3277         chunk_type = chunk_to_extended(chunk_type) &
3278                                 BTRFS_EXTENDED_PROFILE_MASK;
3279
3280         if (bargs->target == chunk_type)
3281                 return 1;
3282
3283         return 0;
3284 }
3285
3286 static int should_balance_chunk(struct btrfs_root *root,
3287                                 struct extent_buffer *leaf,
3288                                 struct btrfs_chunk *chunk, u64 chunk_offset)
3289 {
3290         struct btrfs_balance_control *bctl = root->fs_info->balance_ctl;
3291         struct btrfs_balance_args *bargs = NULL;
3292         u64 chunk_type = btrfs_chunk_type(leaf, chunk);
3293
3294         /* type filter */
3295         if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
3296               (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
3297                 return 0;
3298         }
3299
3300         if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3301                 bargs = &bctl->data;
3302         else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3303                 bargs = &bctl->sys;
3304         else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3305                 bargs = &bctl->meta;
3306
3307         /* profiles filter */
3308         if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
3309             chunk_profiles_filter(chunk_type, bargs)) {
3310                 return 0;
3311         }
3312
3313         /* usage filter */
3314         if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
3315             chunk_usage_filter(bctl->fs_info, chunk_offset, bargs)) {
3316                 return 0;
3317         } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3318             chunk_usage_range_filter(bctl->fs_info, chunk_offset, bargs)) {
3319                 return 0;
3320         }
3321
3322         /* devid filter */
3323         if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
3324             chunk_devid_filter(leaf, chunk, bargs)) {
3325                 return 0;
3326         }
3327
3328         /* drange filter, makes sense only with devid filter */
3329         if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
3330             chunk_drange_filter(leaf, chunk, chunk_offset, bargs)) {
3331                 return 0;
3332         }
3333
3334         /* vrange filter */
3335         if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
3336             chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
3337                 return 0;
3338         }
3339
3340         /* stripes filter */
3341         if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) &&
3342             chunk_stripes_range_filter(leaf, chunk, bargs)) {
3343                 return 0;
3344         }
3345
3346         /* soft profile changing mode */
3347         if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
3348             chunk_soft_convert_filter(chunk_type, bargs)) {
3349                 return 0;
3350         }
3351
3352         /*
3353          * limited by count, must be the last filter
3354          */
3355         if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) {
3356                 if (bargs->limit == 0)
3357                         return 0;
3358                 else
3359                         bargs->limit--;
3360         } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) {
3361                 /*
3362                  * Same logic as the 'limit' filter; the minimum cannot be
3363                  * determined here because we do not have the global informatoin
3364                  * about the count of all chunks that satisfy the filters.
3365                  */
3366                 if (bargs->limit_max == 0)
3367                         return 0;
3368                 else
3369                         bargs->limit_max--;
3370         }
3371
3372         return 1;
3373 }
3374
3375 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
3376 {
3377         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3378         struct btrfs_root *chunk_root = fs_info->chunk_root;
3379         struct btrfs_root *dev_root = fs_info->dev_root;
3380         struct list_head *devices;
3381         struct btrfs_device *device;
3382         u64 old_size;
3383         u64 size_to_free;
3384         u64 chunk_type;
3385         struct btrfs_chunk *chunk;
3386         struct btrfs_path *path;
3387         struct btrfs_key key;
3388         struct btrfs_key found_key;
3389         struct btrfs_trans_handle *trans;
3390         struct extent_buffer *leaf;
3391         int slot;
3392         int ret;
3393         int enospc_errors = 0;
3394         bool counting = true;
3395         /* The single value limit and min/max limits use the same bytes in the */
3396         u64 limit_data = bctl->data.limit;
3397         u64 limit_meta = bctl->meta.limit;
3398         u64 limit_sys = bctl->sys.limit;
3399         u32 count_data = 0;
3400         u32 count_meta = 0;
3401         u32 count_sys = 0;
3402         int chunk_reserved = 0;
3403
3404         /* step one make some room on all the devices */
3405         devices = &fs_info->fs_devices->devices;
3406         list_for_each_entry(device, devices, dev_list) {
3407                 old_size = btrfs_device_get_total_bytes(device);
3408                 size_to_free = div_factor(old_size, 1);
3409                 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
3410                 if (!device->writeable ||
3411                     btrfs_device_get_total_bytes(device) -
3412                     btrfs_device_get_bytes_used(device) > size_to_free ||
3413                     device->is_tgtdev_for_dev_replace)
3414                         continue;
3415
3416                 ret = btrfs_shrink_device(device, old_size - size_to_free);
3417                 if (ret == -ENOSPC)
3418                         break;
3419                 BUG_ON(ret);
3420
3421                 trans = btrfs_start_transaction(dev_root, 0);
3422                 BUG_ON(IS_ERR(trans));
3423
3424                 ret = btrfs_grow_device(trans, device, old_size);
3425                 BUG_ON(ret);
3426
3427                 btrfs_end_transaction(trans, dev_root);
3428         }
3429
3430         /* step two, relocate all the chunks */
3431         path = btrfs_alloc_path();
3432         if (!path) {
3433                 ret = -ENOMEM;
3434                 goto error;
3435         }
3436
3437         /* zero out stat counters */
3438         spin_lock(&fs_info->balance_lock);
3439         memset(&bctl->stat, 0, sizeof(bctl->stat));
3440         spin_unlock(&fs_info->balance_lock);
3441 again:
3442         if (!counting) {
3443                 /*
3444                  * The single value limit and min/max limits use the same bytes
3445                  * in the
3446                  */
3447                 bctl->data.limit = limit_data;
3448                 bctl->meta.limit = limit_meta;
3449                 bctl->sys.limit = limit_sys;
3450         }
3451         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3452         key.offset = (u64)-1;
3453         key.type = BTRFS_CHUNK_ITEM_KEY;
3454
3455         while (1) {
3456                 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
3457                     atomic_read(&fs_info->balance_cancel_req)) {
3458                         ret = -ECANCELED;
3459                         goto error;
3460                 }
3461
3462                 mutex_lock(&fs_info->delete_unused_bgs_mutex);
3463                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3464                 if (ret < 0) {
3465                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3466                         goto error;
3467                 }
3468
3469                 /*
3470                  * this shouldn't happen, it means the last relocate
3471                  * failed
3472                  */
3473                 if (ret == 0)
3474                         BUG(); /* FIXME break ? */
3475
3476                 ret = btrfs_previous_item(chunk_root, path, 0,
3477                                           BTRFS_CHUNK_ITEM_KEY);
3478                 if (ret) {
3479                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3480                         ret = 0;
3481                         break;
3482                 }
3483
3484                 leaf = path->nodes[0];
3485                 slot = path->slots[0];
3486                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3487
3488                 if (found_key.objectid != key.objectid) {
3489                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3490                         break;
3491                 }
3492
3493                 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3494                 chunk_type = btrfs_chunk_type(leaf, chunk);
3495
3496                 if (!counting) {
3497                         spin_lock(&fs_info->balance_lock);
3498                         bctl->stat.considered++;
3499                         spin_unlock(&fs_info->balance_lock);
3500                 }
3501
3502                 ret = should_balance_chunk(chunk_root, leaf, chunk,
3503                                            found_key.offset);
3504
3505                 btrfs_release_path(path);
3506                 if (!ret) {
3507                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3508                         goto loop;
3509                 }
3510
3511                 if (counting) {
3512                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3513                         spin_lock(&fs_info->balance_lock);
3514                         bctl->stat.expected++;
3515                         spin_unlock(&fs_info->balance_lock);
3516
3517                         if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3518                                 count_data++;
3519                         else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3520                                 count_sys++;
3521                         else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3522                                 count_meta++;
3523
3524                         goto loop;
3525                 }
3526
3527                 /*
3528                  * Apply limit_min filter, no need to check if the LIMITS
3529                  * filter is used, limit_min is 0 by default
3530                  */
3531                 if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
3532                                         count_data < bctl->data.limit_min)
3533                                 || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) &&
3534                                         count_meta < bctl->meta.limit_min)
3535                                 || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) &&
3536                                         count_sys < bctl->sys.limit_min)) {
3537                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3538                         goto loop;
3539                 }
3540
3541                 if ((chunk_type & BTRFS_BLOCK_GROUP_DATA) && !chunk_reserved) {
3542                         trans = btrfs_start_transaction(chunk_root, 0);
3543                         if (IS_ERR(trans)) {
3544                                 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3545                                 ret = PTR_ERR(trans);
3546                                 goto error;
3547                         }
3548
3549                         ret = btrfs_force_chunk_alloc(trans, chunk_root,
3550                                                       BTRFS_BLOCK_GROUP_DATA);
3551                         btrfs_end_transaction(trans, chunk_root);
3552                         if (ret < 0) {
3553                                 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3554                                 goto error;
3555                         }
3556                         chunk_reserved = 1;
3557                 }
3558
3559                 ret = btrfs_relocate_chunk(chunk_root,
3560                                            found_key.offset);
3561                 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3562                 if (ret && ret != -ENOSPC)
3563                         goto error;
3564                 if (ret == -ENOSPC) {
3565                         enospc_errors++;
3566                 } else {
3567                         spin_lock(&fs_info->balance_lock);
3568                         bctl->stat.completed++;
3569                         spin_unlock(&fs_info->balance_lock);
3570                 }
3571 loop:
3572                 if (found_key.offset == 0)
3573                         break;
3574                 key.offset = found_key.offset - 1;
3575         }
3576
3577         if (counting) {
3578                 btrfs_release_path(path);
3579                 counting = false;
3580                 goto again;
3581         }
3582 error:
3583         btrfs_free_path(path);
3584         if (enospc_errors) {
3585                 btrfs_info(fs_info, "%d enospc errors during balance",
3586                        enospc_errors);
3587                 if (!ret)
3588                         ret = -ENOSPC;
3589         }
3590
3591         return ret;
3592 }
3593
3594 /**
3595  * alloc_profile_is_valid - see if a given profile is valid and reduced
3596  * @flags: profile to validate
3597  * @extended: if true @flags is treated as an extended profile
3598  */
3599 static int alloc_profile_is_valid(u64 flags, int extended)
3600 {
3601         u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
3602                                BTRFS_BLOCK_GROUP_PROFILE_MASK);
3603
3604         flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
3605
3606         /* 1) check that all other bits are zeroed */
3607         if (flags & ~mask)
3608                 return 0;
3609
3610         /* 2) see if profile is reduced */
3611         if (flags == 0)
3612                 return !extended; /* "0" is valid for usual profiles */
3613
3614         /* true if exactly one bit set */
3615         return (flags & (flags - 1)) == 0;
3616 }
3617
3618 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
3619 {
3620         /* cancel requested || normal exit path */
3621         return atomic_read(&fs_info->balance_cancel_req) ||
3622                 (atomic_read(&fs_info->balance_pause_req) == 0 &&
3623                  atomic_read(&fs_info->balance_cancel_req) == 0);
3624 }
3625
3626 static void __cancel_balance(struct btrfs_fs_info *fs_info)
3627 {
3628         int ret;
3629
3630         unset_balance_control(fs_info);
3631         ret = del_balance_item(fs_info->tree_root);
3632         if (ret)
3633                 btrfs_std_error(fs_info, ret, NULL);
3634
3635         atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3636 }
3637
3638 /* Non-zero return value signifies invalidity */
3639 static inline int validate_convert_profile(struct btrfs_balance_args *bctl_arg,
3640                 u64 allowed)
3641 {
3642         return ((bctl_arg->flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3643                 (!alloc_profile_is_valid(bctl_arg->target, 1) ||
3644                  (bctl_arg->target & ~allowed)));
3645 }
3646
3647 /*
3648  * Should be called with both balance and volume mutexes held
3649  */
3650 int btrfs_balance(struct btrfs_balance_control *bctl,
3651                   struct btrfs_ioctl_balance_args *bargs)
3652 {
3653         struct btrfs_fs_info *fs_info = bctl->fs_info;
3654         u64 allowed;
3655         int mixed = 0;
3656         int ret;
3657         u64 num_devices;
3658         unsigned seq;
3659
3660         if (btrfs_fs_closing(fs_info) ||
3661             atomic_read(&fs_info->balance_pause_req) ||
3662             atomic_read(&fs_info->balance_cancel_req)) {
3663                 ret = -EINVAL;
3664                 goto out;
3665         }
3666
3667         allowed = btrfs_super_incompat_flags(fs_info->super_copy);
3668         if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
3669                 mixed = 1;
3670
3671         /*
3672          * In case of mixed groups both data and meta should be picked,
3673          * and identical options should be given for both of them.
3674          */
3675         allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
3676         if (mixed && (bctl->flags & allowed)) {
3677                 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
3678                     !(bctl->flags & BTRFS_BALANCE_METADATA) ||
3679                     memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
3680                         btrfs_err(fs_info, "with mixed groups data and "
3681                                    "metadata balance options must be the same");
3682                         ret = -EINVAL;
3683                         goto out;
3684                 }
3685         }
3686
3687         num_devices = fs_info->fs_devices->num_devices;
3688         btrfs_dev_replace_lock(&fs_info->dev_replace);
3689         if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
3690                 BUG_ON(num_devices < 1);
3691                 num_devices--;
3692         }
3693         btrfs_dev_replace_unlock(&fs_info->dev_replace);
3694         allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE;
3695         if (num_devices == 1)
3696                 allowed |= BTRFS_BLOCK_GROUP_DUP;
3697         else if (num_devices > 1)
3698                 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1);
3699         if (num_devices > 2)
3700                 allowed |= BTRFS_BLOCK_GROUP_RAID5;
3701         if (num_devices > 3)
3702                 allowed |= (BTRFS_BLOCK_GROUP_RAID10 |
3703                             BTRFS_BLOCK_GROUP_RAID6);
3704         if (validate_convert_profile(&bctl->data, allowed)) {
3705                 btrfs_err(fs_info, "unable to start balance with target "
3706                            "data profile %llu",
3707                        bctl->data.target);
3708                 ret = -EINVAL;
3709                 goto out;
3710         }
3711         if (validate_convert_profile(&bctl->meta, allowed)) {
3712                 btrfs_err(fs_info,
3713                            "unable to start balance with target metadata profile %llu",
3714                        bctl->meta.target);
3715                 ret = -EINVAL;
3716                 goto out;
3717         }
3718         if (validate_convert_profile(&bctl->sys, allowed)) {
3719                 btrfs_err(fs_info,
3720                            "unable to start balance with target system profile %llu",
3721                        bctl->sys.target);
3722                 ret = -EINVAL;
3723                 goto out;
3724         }
3725
3726         /* allow dup'ed data chunks only in mixed mode */
3727         if (!mixed && (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3728             (bctl->data.target & BTRFS_BLOCK_GROUP_DUP)) {
3729                 btrfs_err(fs_info, "dup for data is not allowed");
3730                 ret = -EINVAL;
3731                 goto out;
3732         }
3733
3734         /* allow to reduce meta or sys integrity only if force set */
3735         allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3736                         BTRFS_BLOCK_GROUP_RAID10 |
3737                         BTRFS_BLOCK_GROUP_RAID5 |
3738                         BTRFS_BLOCK_GROUP_RAID6;
3739         do {
3740                 seq = read_seqbegin(&fs_info->profiles_lock);
3741
3742                 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3743                      (fs_info->avail_system_alloc_bits & allowed) &&
3744                      !(bctl->sys.target & allowed)) ||
3745                     ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3746                      (fs_info->avail_metadata_alloc_bits & allowed) &&
3747                      !(bctl->meta.target & allowed))) {
3748                         if (bctl->flags & BTRFS_BALANCE_FORCE) {
3749                                 btrfs_info(fs_info, "force reducing metadata integrity");
3750                         } else {
3751                                 btrfs_err(fs_info, "balance will reduce metadata "
3752                                            "integrity, use force if you want this");
3753                                 ret = -EINVAL;
3754                                 goto out;
3755                         }
3756                 }
3757         } while (read_seqretry(&fs_info->profiles_lock, seq));
3758
3759         if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3760                 fs_info->num_tolerated_disk_barrier_failures = min(
3761                         btrfs_calc_num_tolerated_disk_barrier_failures(fs_info),
3762                         btrfs_get_num_tolerated_disk_barrier_failures(
3763                                 bctl->sys.target));
3764         }
3765
3766         ret = insert_balance_item(fs_info->tree_root, bctl);
3767         if (ret && ret != -EEXIST)
3768                 goto out;
3769
3770         if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
3771                 BUG_ON(ret == -EEXIST);
3772                 set_balance_control(bctl);
3773         } else {
3774                 BUG_ON(ret != -EEXIST);
3775                 spin_lock(&fs_info->balance_lock);
3776                 update_balance_args(bctl);
3777                 spin_unlock(&fs_info->balance_lock);
3778         }
3779
3780         atomic_inc(&fs_info->balance_running);
3781         mutex_unlock(&fs_info->balance_mutex);
3782
3783         ret = __btrfs_balance(fs_info);
3784
3785         mutex_lock(&fs_info->balance_mutex);
3786         atomic_dec(&fs_info->balance_running);
3787
3788         if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3789                 fs_info->num_tolerated_disk_barrier_failures =
3790                         btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3791         }
3792
3793         if (bargs) {
3794                 memset(bargs, 0, sizeof(*bargs));
3795                 update_ioctl_balance_args(fs_info, 0, bargs);
3796         }
3797
3798         if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
3799             balance_need_close(fs_info)) {
3800                 __cancel_balance(fs_info);
3801         }
3802
3803         wake_up(&fs_info->balance_wait_q);
3804
3805         return ret;
3806 out:
3807         if (bctl->flags & BTRFS_BALANCE_RESUME)
3808                 __cancel_balance(fs_info);
3809         else {
3810                 kfree(bctl);
3811                 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3812         }
3813         return ret;
3814 }
3815
3816 static int balance_kthread(void *data)
3817 {
3818         struct btrfs_fs_info *fs_info = data;
3819         int ret = 0;
3820
3821         mutex_lock(&fs_info->volume_mutex);
3822         mutex_lock(&fs_info->balance_mutex);
3823
3824         if (fs_info->balance_ctl) {
3825                 btrfs_info(fs_info, "continuing balance");
3826                 ret = btrfs_balance(fs_info->balance_ctl, NULL);
3827         }
3828
3829         mutex_unlock(&fs_info->balance_mutex);
3830         mutex_unlock(&fs_info->volume_mutex);
3831
3832         return ret;
3833 }
3834
3835 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
3836 {
3837         struct task_struct *tsk;
3838
3839         spin_lock(&fs_info->balance_lock);
3840         if (!fs_info->balance_ctl) {
3841                 spin_unlock(&fs_info->balance_lock);
3842                 return 0;
3843         }
3844         spin_unlock(&fs_info->balance_lock);
3845
3846         if (btrfs_test_opt(fs_info->tree_root, SKIP_BALANCE)) {
3847                 btrfs_info(fs_info, "force skipping balance");
3848                 return 0;
3849         }
3850
3851         tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
3852         return PTR_ERR_OR_ZERO(tsk);
3853 }
3854
3855 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
3856 {
3857         struct btrfs_balance_control *bctl;
3858         struct btrfs_balance_item *item;
3859         struct btrfs_disk_balance_args disk_bargs;
3860         struct btrfs_path *path;
3861         struct extent_buffer *leaf;
3862         struct btrfs_key key;
3863         int ret;
3864
3865         path = btrfs_alloc_path();
3866         if (!path)
3867                 return -ENOMEM;
3868
3869         key.objectid = BTRFS_BALANCE_OBJECTID;
3870         key.type = BTRFS_BALANCE_ITEM_KEY;
3871         key.offset = 0;
3872
3873         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
3874         if (ret < 0)
3875                 goto out;
3876         if (ret > 0) { /* ret = -ENOENT; */
3877                 ret = 0;
3878                 goto out;
3879         }
3880
3881         bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
3882         if (!bctl) {
3883                 ret = -ENOMEM;
3884                 goto out;
3885         }
3886
3887         leaf = path->nodes[0];
3888         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3889
3890         bctl->fs_info = fs_info;
3891         bctl->flags = btrfs_balance_flags(leaf, item);
3892         bctl->flags |= BTRFS_BALANCE_RESUME;
3893
3894         btrfs_balance_data(leaf, item, &disk_bargs);
3895         btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
3896         btrfs_balance_meta(leaf, item, &disk_bargs);
3897         btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
3898         btrfs_balance_sys(leaf, item, &disk_bargs);
3899         btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
3900
3901         WARN_ON(atomic_xchg(&fs_info->mutually_exclusive_operation_running, 1));
3902
3903         mutex_lock(&fs_info->volume_mutex);
3904         mutex_lock(&fs_info->balance_mutex);
3905
3906         set_balance_control(bctl);
3907
3908         mutex_unlock(&fs_info->balance_mutex);
3909         mutex_unlock(&fs_info->volume_mutex);
3910 out:
3911         btrfs_free_path(path);
3912         return ret;
3913 }
3914
3915 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
3916 {
3917         int ret = 0;
3918
3919         mutex_lock(&fs_info->balance_mutex);
3920         if (!fs_info->balance_ctl) {
3921                 mutex_unlock(&fs_info->balance_mutex);
3922                 return -ENOTCONN;
3923         }
3924
3925         if (atomic_read(&fs_info->balance_running)) {
3926                 atomic_inc(&fs_info->balance_pause_req);
3927                 mutex_unlock(&fs_info->balance_mutex);
3928
3929                 wait_event(fs_info->balance_wait_q,
3930                            atomic_read(&fs_info->balance_running) == 0);
3931
3932                 mutex_lock(&fs_info->balance_mutex);
3933                 /* we are good with balance_ctl ripped off from under us */
3934                 BUG_ON(atomic_read(&fs_info->balance_running));
3935                 atomic_dec(&fs_info->balance_pause_req);
3936         } else {
3937                 ret = -ENOTCONN;
3938         }
3939
3940         mutex_unlock(&fs_info->balance_mutex);
3941         return ret;
3942 }
3943
3944 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
3945 {
3946         if (fs_info->sb->s_flags & MS_RDONLY)
3947                 return -EROFS;
3948
3949         mutex_lock(&fs_info->balance_mutex);
3950         if (!fs_info->balance_ctl) {
3951                 mutex_unlock(&fs_info->balance_mutex);
3952                 return -ENOTCONN;
3953         }
3954
3955         atomic_inc(&fs_info->balance_cancel_req);
3956         /*
3957          * if we are running just wait and return, balance item is
3958          * deleted in btrfs_balance in this case
3959          */
3960         if (atomic_read(&fs_info->balance_running)) {
3961                 mutex_unlock(&fs_info->balance_mutex);
3962                 wait_event(fs_info->balance_wait_q,
3963                            atomic_read(&fs_info->balance_running) == 0);
3964                 mutex_lock(&fs_info->balance_mutex);
3965         } else {
3966                 /* __cancel_balance needs volume_mutex */
3967                 mutex_unlock(&fs_info->balance_mutex);
3968                 mutex_lock(&fs_info->volume_mutex);
3969                 mutex_lock(&fs_info->balance_mutex);
3970
3971                 if (fs_info->balance_ctl)
3972                         __cancel_balance(fs_info);
3973
3974                 mutex_unlock(&fs_info->volume_mutex);
3975         }
3976
3977         BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running));
3978         atomic_dec(&fs_info->balance_cancel_req);
3979         mutex_unlock(&fs_info->balance_mutex);
3980         return 0;
3981 }
3982
3983 static int btrfs_uuid_scan_kthread(void *data)
3984 {
3985         struct btrfs_fs_info *fs_info = data;
3986         struct btrfs_root *root = fs_info->tree_root;
3987         struct btrfs_key key;
3988         struct btrfs_key max_key;
3989         struct btrfs_path *path = NULL;
3990         int ret = 0;
3991         struct extent_buffer *eb;
3992         int slot;
3993         struct btrfs_root_item root_item;
3994         u32 item_size;
3995         struct btrfs_trans_handle *trans = NULL;
3996
3997         path = btrfs_alloc_path();
3998         if (!path) {
3999                 ret = -ENOMEM;
4000                 goto out;
4001         }
4002
4003         key.objectid = 0;
4004         key.type = BTRFS_ROOT_ITEM_KEY;
4005         key.offset = 0;
4006
4007         max_key.objectid = (u64)-1;
4008         max_key.type = BTRFS_ROOT_ITEM_KEY;
4009         max_key.offset = (u64)-1;
4010
4011         while (1) {
4012                 ret = btrfs_search_forward(root, &key, path, 0);
4013                 if (ret) {
4014                         if (ret > 0)
4015                                 ret = 0;
4016                         break;
4017                 }
4018
4019                 if (key.type != BTRFS_ROOT_ITEM_KEY ||
4020                     (key.objectid < BTRFS_FIRST_FREE_OBJECTID &&
4021                      key.objectid != BTRFS_FS_TREE_OBJECTID) ||
4022                     key.objectid > BTRFS_LAST_FREE_OBJECTID)
4023                         goto skip;
4024
4025                 eb = path->nodes[0];
4026                 slot = path->slots[0];
4027                 item_size = btrfs_item_size_nr(eb, slot);
4028                 if (item_size < sizeof(root_item))
4029                         goto skip;
4030
4031                 read_extent_buffer(eb, &root_item,
4032                                    btrfs_item_ptr_offset(eb, slot),
4033                                    (int)sizeof(root_item));
4034                 if (btrfs_root_refs(&root_item) == 0)
4035                         goto skip;
4036
4037                 if (!btrfs_is_empty_uuid(root_item.uuid) ||
4038                     !btrfs_is_empty_uuid(root_item.received_uuid)) {
4039                         if (trans)
4040                                 goto update_tree;
4041
4042                         btrfs_release_path(path);
4043                         /*
4044                          * 1 - subvol uuid item
4045                          * 1 - received_subvol uuid item
4046                          */
4047                         trans = btrfs_start_transaction(fs_info->uuid_root, 2);
4048                         if (IS_ERR(trans)) {
4049                                 ret = PTR_ERR(trans);
4050                                 break;
4051                         }
4052                         continue;
4053                 } else {
4054                         goto skip;
4055                 }
4056 update_tree:
4057                 if (!btrfs_is_empty_uuid(root_item.uuid)) {
4058                         ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
4059                                                   root_item.uuid,
4060                                                   BTRFS_UUID_KEY_SUBVOL,
4061                                                   key.objectid);
4062                         if (ret < 0) {
4063                                 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4064                                         ret);
4065                                 break;
4066                         }
4067                 }
4068
4069                 if (!btrfs_is_empty_uuid(root_item.received_uuid)) {
4070                         ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
4071                                                   root_item.received_uuid,
4072                                                  BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4073                                                   key.objectid);
4074                         if (ret < 0) {
4075                                 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4076                                         ret);
4077                                 break;
4078                         }
4079                 }
4080
4081 skip:
4082                 if (trans) {
4083                         ret = btrfs_end_transaction(trans, fs_info->uuid_root);
4084                         trans = NULL;
4085                         if (ret)
4086                                 break;
4087                 }
4088
4089                 btrfs_release_path(path);
4090                 if (key.offset < (u64)-1) {
4091                         key.offset++;
4092                 } else if (key.type < BTRFS_ROOT_ITEM_KEY) {
4093                         key.offset = 0;
4094                         key.type = BTRFS_ROOT_ITEM_KEY;
4095                 } else if (key.objectid < (u64)-1) {
4096                         key.offset = 0;
4097                         key.type = BTRFS_ROOT_ITEM_KEY;
4098                         key.objectid++;
4099                 } else {
4100                         break;
4101                 }
4102                 cond_resched();
4103         }
4104
4105 out:
4106         btrfs_free_path(path);
4107         if (trans && !IS_ERR(trans))
4108                 btrfs_end_transaction(trans, fs_info->uuid_root);
4109         if (ret)
4110                 btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret);
4111         else
4112                 fs_info->update_uuid_tree_gen = 1;
4113         up(&fs_info->uuid_tree_rescan_sem);
4114         return 0;
4115 }
4116
4117 /*
4118  * Callback for btrfs_uuid_tree_iterate().
4119  * returns:
4120  * 0    check succeeded, the entry is not outdated.
4121  * < 0  if an error occured.
4122  * > 0  if the check failed, which means the caller shall remove the entry.
4123  */
4124 static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info *fs_info,
4125                                        u8 *uuid, u8 type, u64 subid)
4126 {
4127         struct btrfs_key key;
4128         int ret = 0;
4129         struct btrfs_root *subvol_root;
4130
4131         if (type != BTRFS_UUID_KEY_SUBVOL &&
4132             type != BTRFS_UUID_KEY_RECEIVED_SUBVOL)
4133                 goto out;
4134
4135         key.objectid = subid;
4136         key.type = BTRFS_ROOT_ITEM_KEY;
4137         key.offset = (u64)-1;
4138         subvol_root = btrfs_read_fs_root_no_name(fs_info, &key);
4139         if (IS_ERR(subvol_root)) {
4140                 ret = PTR_ERR(subvol_root);
4141                 if (ret == -ENOENT)
4142                         ret = 1;
4143                 goto out;
4144         }
4145
4146         switch (type) {
4147         case BTRFS_UUID_KEY_SUBVOL:
4148                 if (memcmp(uuid, subvol_root->root_item.uuid, BTRFS_UUID_SIZE))
4149                         ret = 1;
4150                 break;
4151         case BTRFS_UUID_KEY_RECEIVED_SUBVOL:
4152                 if (memcmp(uuid, subvol_root->root_item.received_uuid,
4153                            BTRFS_UUID_SIZE))
4154                         ret = 1;
4155                 break;
4156         }
4157
4158 out:
4159         return ret;
4160 }
4161
4162 static int btrfs_uuid_rescan_kthread(void *data)
4163 {
4164         struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data;
4165         int ret;
4166
4167         /*
4168          * 1st step is to iterate through the existing UUID tree and
4169          * to delete all entries that contain outdated data.
4170          * 2nd step is to add all missing entries to the UUID tree.
4171          */
4172         ret = btrfs_uuid_tree_iterate(fs_info, btrfs_check_uuid_tree_entry);
4173         if (ret < 0) {
4174                 btrfs_warn(fs_info, "iterating uuid_tree failed %d", ret);
4175                 up(&fs_info->uuid_tree_rescan_sem);
4176                 return ret;
4177         }
4178         return btrfs_uuid_scan_kthread(data);
4179 }
4180
4181 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info)
4182 {
4183         struct btrfs_trans_handle *trans;
4184         struct btrfs_root *tree_root = fs_info->tree_root;
4185         struct btrfs_root *uuid_root;
4186         struct task_struct *task;
4187         int ret;
4188
4189         /*
4190          * 1 - root node
4191          * 1 - root item
4192          */
4193         trans = btrfs_start_transaction(tree_root, 2);
4194         if (IS_ERR(trans))
4195                 return PTR_ERR(trans);
4196
4197         uuid_root = btrfs_create_tree(trans, fs_info,
4198                                       BTRFS_UUID_TREE_OBJECTID);
4199         if (IS_ERR(uuid_root)) {
4200                 ret = PTR_ERR(uuid_root);
4201                 btrfs_abort_transaction(trans, tree_root, ret);
4202                 return ret;
4203         }
4204
4205         fs_info->uuid_root = uuid_root;
4206
4207         ret = btrfs_commit_transaction(trans, tree_root);
4208         if (ret)
4209                 return ret;
4210
4211         down(&fs_info->uuid_tree_rescan_sem);
4212         task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid");
4213         if (IS_ERR(task)) {
4214                 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4215                 btrfs_warn(fs_info, "failed to start uuid_scan task");
4216                 up(&fs_info->uuid_tree_rescan_sem);
4217                 return PTR_ERR(task);
4218         }
4219
4220         return 0;
4221 }
4222
4223 int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
4224 {
4225         struct task_struct *task;
4226
4227         down(&fs_info->uuid_tree_rescan_sem);
4228         task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
4229         if (IS_ERR(task)) {
4230                 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4231                 btrfs_warn(fs_info, "failed to start uuid_rescan task");
4232                 up(&fs_info->uuid_tree_rescan_sem);
4233                 return PTR_ERR(task);
4234         }
4235
4236         return 0;
4237 }
4238
4239 /*
4240  * shrinking a device means finding all of the device extents past
4241  * the new size, and then following the back refs to the chunks.
4242  * The chunk relocation code actually frees the device extent
4243  */
4244 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
4245 {
4246         struct btrfs_trans_handle *trans;
4247         struct btrfs_root *root = device->dev_root;
4248         struct btrfs_dev_extent *dev_extent = NULL;
4249         struct btrfs_path *path;
4250         u64 length;
4251         u64 chunk_offset;
4252         int ret;
4253         int slot;
4254         int failed = 0;
4255         bool retried = false;
4256         bool checked_pending_chunks = false;
4257         struct extent_buffer *l;
4258         struct btrfs_key key;
4259         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
4260         u64 old_total = btrfs_super_total_bytes(super_copy);
4261         u64 old_size = btrfs_device_get_total_bytes(device);
4262         u64 diff = old_size - new_size;
4263
4264         if (device->is_tgtdev_for_dev_replace)
4265                 return -EINVAL;
4266
4267         path = btrfs_alloc_path();
4268         if (!path)
4269                 return -ENOMEM;
4270
4271         path->reada = 2;
4272
4273         lock_chunks(root);
4274
4275         btrfs_device_set_total_bytes(device, new_size);
4276         if (device->writeable) {
4277                 device->fs_devices->total_rw_bytes -= diff;
4278                 spin_lock(&root->fs_info->free_chunk_lock);
4279                 root->fs_info->free_chunk_space -= diff;
4280                 spin_unlock(&root->fs_info->free_chunk_lock);
4281         }
4282         unlock_chunks(root);
4283
4284 again:
4285         key.objectid = device->devid;
4286         key.offset = (u64)-1;
4287         key.type = BTRFS_DEV_EXTENT_KEY;
4288
4289         do {
4290                 mutex_lock(&root->fs_info->delete_unused_bgs_mutex);
4291                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4292                 if (ret < 0) {
4293                         mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4294                         goto done;
4295                 }
4296
4297                 ret = btrfs_previous_item(root, path, 0, key.type);
4298                 if (ret)
4299                         mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4300                 if (ret < 0)
4301                         goto done;
4302                 if (ret) {
4303                         ret = 0;
4304                         btrfs_release_path(path);
4305                         break;
4306                 }
4307
4308                 l = path->nodes[0];
4309                 slot = path->slots[0];
4310                 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
4311
4312                 if (key.objectid != device->devid) {
4313                         mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4314                         btrfs_release_path(path);
4315                         break;
4316                 }
4317
4318                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
4319                 length = btrfs_dev_extent_length(l, dev_extent);
4320
4321                 if (key.offset + length <= new_size) {
4322                         mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4323                         btrfs_release_path(path);
4324                         break;
4325                 }
4326
4327                 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
4328                 btrfs_release_path(path);
4329
4330                 ret = btrfs_relocate_chunk(root, chunk_offset);
4331                 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4332                 if (ret && ret != -ENOSPC)
4333                         goto done;
4334                 if (ret == -ENOSPC)
4335                         failed++;
4336         } while (key.offset-- > 0);
4337
4338         if (failed && !retried) {
4339                 failed = 0;
4340                 retried = true;
4341                 goto again;
4342         } else if (failed && retried) {
4343                 ret = -ENOSPC;
4344                 goto done;
4345         }
4346
4347         /* Shrinking succeeded, else we would be at "done". */
4348         trans = btrfs_start_transaction(root, 0);
4349         if (IS_ERR(trans)) {
4350                 ret = PTR_ERR(trans);
4351                 goto done;
4352         }
4353
4354         lock_chunks(root);
4355
4356         /*
4357          * We checked in the above loop all device extents that were already in
4358          * the device tree. However before we have updated the device's
4359          * total_bytes to the new size, we might have had chunk allocations that
4360          * have not complete yet (new block groups attached to transaction
4361          * handles), and therefore their device extents were not yet in the
4362          * device tree and we missed them in the loop above. So if we have any
4363          * pending chunk using a device extent that overlaps the device range
4364          * that we can not use anymore, commit the current transaction and
4365          * repeat the search on the device tree - this way we guarantee we will
4366          * not have chunks using device extents that end beyond 'new_size'.
4367          */
4368         if (!checked_pending_chunks) {
4369                 u64 start = new_size;
4370                 u64 len = old_size - new_size;
4371
4372                 if (contains_pending_extent(trans->transaction, device,
4373                                             &start, len)) {
4374                         unlock_chunks(root);
4375                         checked_pending_chunks = true;
4376                         failed = 0;
4377                         retried = false;
4378                         ret = btrfs_commit_transaction(trans, root);
4379                         if (ret)
4380                                 goto done;
4381                         goto again;
4382                 }
4383         }
4384
4385         btrfs_device_set_disk_total_bytes(device, new_size);
4386         if (list_empty(&device->resized_list))
4387                 list_add_tail(&device->resized_list,
4388                               &root->fs_info->fs_devices->resized_devices);
4389
4390         WARN_ON(diff > old_total);
4391         btrfs_set_super_total_bytes(super_copy, old_total - diff);
4392         unlock_chunks(root);
4393
4394         /* Now btrfs_update_device() will change the on-disk size. */
4395         ret = btrfs_update_device(trans, device);
4396         btrfs_end_transaction(trans, root);
4397 done:
4398         btrfs_free_path(path);
4399         if (ret) {
4400                 lock_chunks(root);
4401                 btrfs_device_set_total_bytes(device, old_size);
4402                 if (device->writeable)
4403                         device->fs_devices->total_rw_bytes += diff;
4404                 spin_lock(&root->fs_info->free_chunk_lock);
4405                 root->fs_info->free_chunk_space += diff;
4406                 spin_unlock(&root->fs_info->free_chunk_lock);
4407                 unlock_chunks(root);
4408         }
4409         return ret;
4410 }
4411
4412 static int btrfs_add_system_chunk(struct btrfs_root *root,
4413                            struct btrfs_key *key,
4414                            struct btrfs_chunk *chunk, int item_size)
4415 {
4416         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
4417         struct btrfs_disk_key disk_key;
4418         u32 array_size;
4419         u8 *ptr;
4420
4421         lock_chunks(root);
4422         array_size = btrfs_super_sys_array_size(super_copy);
4423         if (array_size + item_size + sizeof(disk_key)
4424                         > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
4425                 unlock_chunks(root);
4426                 return -EFBIG;
4427         }
4428
4429         ptr = super_copy->sys_chunk_array + array_size;
4430         btrfs_cpu_key_to_disk(&disk_key, key);
4431         memcpy(ptr, &disk_key, sizeof(disk_key));
4432         ptr += sizeof(disk_key);
4433         memcpy(ptr, chunk, item_size);
4434         item_size += sizeof(disk_key);
4435         btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
4436         unlock_chunks(root);
4437
4438         return 0;
4439 }
4440
4441 /*
4442  * sort the devices in descending order by max_avail, total_avail
4443  */
4444 static int btrfs_cmp_device_info(const void *a, const void *b)
4445 {
4446         const struct btrfs_device_info *di_a = a;
4447         const struct btrfs_device_info *di_b = b;
4448
4449         if (di_a->max_avail > di_b->max_avail)
4450                 return -1;
4451         if (di_a->max_avail < di_b->max_avail)
4452                 return 1;
4453         if (di_a->total_avail > di_b->total_avail)
4454                 return -1;
4455         if (di_a->total_avail < di_b->total_avail)
4456                 return 1;
4457         return 0;
4458 }
4459
4460 static u32 find_raid56_stripe_len(u32 data_devices, u32 dev_stripe_target)
4461 {
4462         /* TODO allow them to set a preferred stripe size */
4463         return 64 * 1024;
4464 }
4465
4466 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
4467 {
4468         if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK))
4469                 return;
4470
4471         btrfs_set_fs_incompat(info, RAID56);
4472 }
4473
4474 #define BTRFS_MAX_DEVS(r) ((BTRFS_LEAF_DATA_SIZE(r)             \
4475                         - sizeof(struct btrfs_item)             \
4476                         - sizeof(struct btrfs_chunk))           \
4477                         / sizeof(struct btrfs_stripe) + 1)
4478
4479 #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE        \
4480                                 - 2 * sizeof(struct btrfs_disk_key)     \
4481                                 - 2 * sizeof(struct btrfs_chunk))       \
4482                                 / sizeof(struct btrfs_stripe) + 1)
4483
4484 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4485                                struct btrfs_root *extent_root, u64 start,
4486                                u64 type)
4487 {
4488         struct btrfs_fs_info *info = extent_root->fs_info;
4489         struct btrfs_fs_devices *fs_devices = info->fs_devices;
4490         struct list_head *cur;
4491         struct map_lookup *map = NULL;
4492         struct extent_map_tree *em_tree;
4493         struct extent_map *em;
4494         struct btrfs_device_info *devices_info = NULL;
4495         u64 total_avail;
4496         int num_stripes;        /* total number of stripes to allocate */
4497         int data_stripes;       /* number of stripes that count for
4498                                    block group size */
4499         int sub_stripes;        /* sub_stripes info for map */
4500         int dev_stripes;        /* stripes per dev */
4501         int devs_max;           /* max devs to use */
4502         int devs_min;           /* min devs needed */
4503         int devs_increment;     /* ndevs has to be a multiple of this */
4504         int ncopies;            /* how many copies to data has */
4505         int ret;
4506         u64 max_stripe_size;
4507         u64 max_chunk_size;
4508         u64 stripe_size;
4509         u64 num_bytes;
4510         u64 raid_stripe_len = BTRFS_STRIPE_LEN;
4511         int ndevs;
4512         int i;
4513         int j;
4514         int index;
4515
4516         BUG_ON(!alloc_profile_is_valid(type, 0));
4517
4518         if (list_empty(&fs_devices->alloc_list))
4519                 return -ENOSPC;
4520
4521         index = __get_raid_index(type);
4522
4523         sub_stripes = btrfs_raid_array[index].sub_stripes;
4524         dev_stripes = btrfs_raid_array[index].dev_stripes;
4525         devs_max = btrfs_raid_array[index].devs_max;
4526         devs_min = btrfs_raid_array[index].devs_min;
4527         devs_increment = btrfs_raid_array[index].devs_increment;
4528         ncopies = btrfs_raid_array[index].ncopies;
4529
4530         if (type & BTRFS_BLOCK_GROUP_DATA) {
4531                 max_stripe_size = 1024 * 1024 * 1024;
4532                 max_chunk_size = 10 * max_stripe_size;
4533                 if (!devs_max)
4534                         devs_max = BTRFS_MAX_DEVS(info->chunk_root);
4535         } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
4536                 /* for larger filesystems, use larger metadata chunks */
4537                 if (fs_devices->total_rw_bytes > 50ULL * 1024 * 1024 * 1024)
4538                         max_stripe_size = 1024 * 1024 * 1024;
4539                 else
4540                         max_stripe_size = 256 * 1024 * 1024;
4541                 max_chunk_size = max_stripe_size;
4542                 if (!devs_max)
4543                         devs_max = BTRFS_MAX_DEVS(info->chunk_root);
4544         } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
4545                 max_stripe_size = 32 * 1024 * 1024;
4546                 max_chunk_size = 2 * max_stripe_size;
4547                 if (!devs_max)
4548                         devs_max = BTRFS_MAX_DEVS_SYS_CHUNK;
4549         } else {
4550                 btrfs_err(info, "invalid chunk type 0x%llx requested",
4551                        type);
4552                 BUG_ON(1);
4553         }
4554
4555         /* we don't want a chunk larger than 10% of writeable space */
4556         max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
4557                              max_chunk_size);
4558
4559         devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info),
4560                                GFP_NOFS);
4561         if (!devices_info)
4562                 return -ENOMEM;
4563
4564         cur = fs_devices->alloc_list.next;
4565
4566         /*
4567          * in the first pass through the devices list, we gather information
4568          * about the available holes on each device.
4569          */
4570         ndevs = 0;
4571         while (cur != &fs_devices->alloc_list) {
4572                 struct btrfs_device *device;
4573                 u64 max_avail;
4574                 u64 dev_offset;
4575
4576                 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
4577
4578                 cur = cur->next;
4579
4580                 if (!device->writeable) {
4581                         WARN(1, KERN_ERR
4582                                "BTRFS: read-only device in alloc_list\n");
4583                         continue;
4584                 }
4585
4586                 if (!device->in_fs_metadata ||
4587                     device->is_tgtdev_for_dev_replace)
4588                         continue;
4589
4590                 if (device->total_bytes > device->bytes_used)
4591                         total_avail = device->total_bytes - device->bytes_used;
4592                 else
4593                         total_avail = 0;
4594
4595                 /* If there is no space on this device, skip it. */
4596                 if (total_avail == 0)
4597                         continue;
4598
4599                 ret = find_free_dev_extent(trans, device,
4600                                            max_stripe_size * dev_stripes,
4601                                            &dev_offset, &max_avail);
4602                 if (ret && ret != -ENOSPC)
4603                         goto error;
4604
4605                 if (ret == 0)
4606                         max_avail = max_stripe_size * dev_stripes;
4607
4608                 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes)
4609                         continue;
4610
4611                 if (ndevs == fs_devices->rw_devices) {
4612                         WARN(1, "%s: found more than %llu devices\n",
4613                              __func__, fs_devices->rw_devices);
4614                         break;
4615                 }
4616                 devices_info[ndevs].dev_offset = dev_offset;
4617                 devices_info[ndevs].max_avail = max_avail;
4618                 devices_info[ndevs].total_avail = total_avail;
4619                 devices_info[ndevs].dev = device;
4620                 ++ndevs;
4621         }
4622
4623         /*
4624          * now sort the devices by hole size / available space
4625          */
4626         sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
4627              btrfs_cmp_device_info, NULL);
4628
4629         /* round down to number of usable stripes */
4630         ndevs -= ndevs % devs_increment;
4631
4632         if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) {
4633                 ret = -ENOSPC;
4634                 goto error;
4635         }
4636
4637         if (devs_max && ndevs > devs_max)
4638                 ndevs = devs_max;
4639         /*
4640          * the primary goal is to maximize the number of stripes, so use as many
4641          * devices as possible, even if the stripes are not maximum sized.
4642          */
4643         stripe_size = devices_info[ndevs-1].max_avail;
4644         num_stripes = ndevs * dev_stripes;
4645
4646         /*
4647          * this will have to be fixed for RAID1 and RAID10 over
4648          * more drives
4649          */
4650         data_stripes = num_stripes / ncopies;
4651
4652         if (type & BTRFS_BLOCK_GROUP_RAID5) {
4653                 raid_stripe_len = find_raid56_stripe_len(ndevs - 1,
4654                                  btrfs_super_stripesize(info->super_copy));
4655                 data_stripes = num_stripes - 1;
4656         }
4657         if (type & BTRFS_BLOCK_GROUP_RAID6) {
4658                 raid_stripe_len = find_raid56_stripe_len(ndevs - 2,
4659                                  btrfs_super_stripesize(info->super_copy));
4660                 data_stripes = num_stripes - 2;
4661         }
4662
4663         /*
4664          * Use the number of data stripes to figure out how big this chunk
4665          * is really going to be in terms of logical address space,
4666          * and compare that answer with the max chunk size
4667          */
4668         if (stripe_size * data_stripes > max_chunk_size) {
4669                 u64 mask = (1ULL << 24) - 1;
4670
4671                 stripe_size = div_u64(max_chunk_size, data_stripes);
4672
4673                 /* bump the answer up to a 16MB boundary */
4674                 stripe_size = (stripe_size + mask) & ~mask;
4675
4676                 /* but don't go higher than the limits we found
4677                  * while searching for free extents
4678                  */
4679                 if (stripe_size > devices_info[ndevs-1].max_avail)
4680                         stripe_size = devices_info[ndevs-1].max_avail;
4681         }
4682
4683         stripe_size = div_u64(stripe_size, dev_stripes);
4684
4685         /* align to BTRFS_STRIPE_LEN */
4686         stripe_size = div_u64(stripe_size, raid_stripe_len);
4687         stripe_size *= raid_stripe_len;
4688
4689         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
4690         if (!map) {
4691                 ret = -ENOMEM;
4692                 goto error;
4693         }
4694         map->num_stripes = num_stripes;
4695
4696         for (i = 0; i < ndevs; ++i) {
4697                 for (j = 0; j < dev_stripes; ++j) {
4698                         int s = i * dev_stripes + j;
4699                         map->stripes[s].dev = devices_info[i].dev;
4700                         map->stripes[s].physical = devices_info[i].dev_offset +
4701                                                    j * stripe_size;
4702                 }
4703         }
4704         map->sector_size = extent_root->sectorsize;
4705         map->stripe_len = raid_stripe_len;
4706         map->io_align = raid_stripe_len;
4707         map->io_width = raid_stripe_len;
4708         map->type = type;
4709         map->sub_stripes = sub_stripes;
4710
4711         num_bytes = stripe_size * data_stripes;
4712
4713         trace_btrfs_chunk_alloc(info->chunk_root, map, start, num_bytes);
4714
4715         em = alloc_extent_map();
4716         if (!em) {
4717                 kfree(map);
4718                 ret = -ENOMEM;
4719                 goto error;
4720         }
4721         set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
4722         em->bdev = (struct block_device *)map;
4723         em->start = start;
4724         em->len = num_bytes;
4725         em->block_start = 0;
4726         em->block_len = em->len;
4727         em->orig_block_len = stripe_size;
4728
4729         em_tree = &extent_root->fs_info->mapping_tree.map_tree;
4730         write_lock(&em_tree->lock);
4731         ret = add_extent_mapping(em_tree, em, 0);
4732         if (!ret) {
4733                 list_add_tail(&em->list, &trans->transaction->pending_chunks);
4734                 atomic_inc(&em->refs);
4735         }
4736         write_unlock(&em_tree->lock);
4737         if (ret) {
4738                 free_extent_map(em);
4739                 goto error;
4740         }
4741
4742         ret = btrfs_make_block_group(trans, extent_root, 0, type,
4743                                      BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4744                                      start, num_bytes);
4745         if (ret)
4746                 goto error_del_extent;
4747
4748         for (i = 0; i < map->num_stripes; i++) {
4749                 num_bytes = map->stripes[i].dev->bytes_used + stripe_size;
4750                 btrfs_device_set_bytes_used(map->stripes[i].dev, num_bytes);
4751         }
4752
4753         spin_lock(&extent_root->fs_info->free_chunk_lock);
4754         extent_root->fs_info->free_chunk_space -= (stripe_size *
4755                                                    map->num_stripes);
4756         spin_unlock(&extent_root->fs_info->free_chunk_lock);
4757
4758         free_extent_map(em);
4759         check_raid56_incompat_flag(extent_root->fs_info, type);
4760
4761         kfree(devices_info);
4762         return 0;
4763
4764 error_del_extent:
4765         write_lock(&em_tree->lock);
4766         remove_extent_mapping(em_tree, em);
4767         write_unlock(&em_tree->lock);
4768
4769         /* One for our allocation */
4770         free_extent_map(em);
4771         /* One for the tree reference */
4772         free_extent_map(em);
4773         /* One for the pending_chunks list reference */
4774         free_extent_map(em);
4775 error:
4776         kfree(devices_info);
4777         return ret;
4778 }
4779
4780 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans,
4781                                 struct btrfs_root *extent_root,
4782                                 u64 chunk_offset, u64 chunk_size)
4783 {
4784         struct btrfs_key key;
4785         struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
4786         struct btrfs_device *device;
4787         struct btrfs_chunk *chunk;
4788         struct btrfs_stripe *stripe;
4789         struct extent_map_tree *em_tree;
4790         struct extent_map *em;
4791         struct map_lookup *map;
4792         size_t item_size;
4793         u64 dev_offset;
4794         u64 stripe_size;
4795         int i = 0;
4796         int ret;
4797
4798         em_tree = &extent_root->fs_info->mapping_tree.map_tree;
4799         read_lock(&em_tree->lock);
4800         em = lookup_extent_mapping(em_tree, chunk_offset, chunk_size);
4801         read_unlock(&em_tree->lock);
4802
4803         if (!em) {
4804                 btrfs_crit(extent_root->fs_info, "unable to find logical "
4805                            "%Lu len %Lu", chunk_offset, chunk_size);
4806                 return -EINVAL;
4807         }
4808
4809         if (em->start != chunk_offset || em->len != chunk_size) {
4810                 btrfs_crit(extent_root->fs_info, "found a bad mapping, wanted"
4811                           " %Lu-%Lu, found %Lu-%Lu", chunk_offset,
4812                           chunk_size, em->start, em->len);
4813                 free_extent_map(em);
4814                 return -EINVAL;
4815         }
4816
4817         map = (struct map_lookup *)em->bdev;
4818         item_size = btrfs_chunk_item_size(map->num_stripes);
4819         stripe_size = em->orig_block_len;
4820
4821         chunk = kzalloc(item_size, GFP_NOFS);
4822         if (!chunk) {
4823                 ret = -ENOMEM;
4824                 goto out;
4825         }
4826
4827         for (i = 0; i < map->num_stripes; i++) {
4828                 device = map->stripes[i].dev;
4829                 dev_offset = map->stripes[i].physical;
4830
4831                 ret = btrfs_update_device(trans, device);
4832                 if (ret)
4833                         goto out;
4834                 ret = btrfs_alloc_dev_extent(trans, device,
4835                                              chunk_root->root_key.objectid,
4836                                              BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4837                                              chunk_offset, dev_offset,
4838                                              stripe_size);
4839                 if (ret)
4840                         goto out;
4841         }
4842
4843         stripe = &chunk->stripe;
4844         for (i = 0; i < map->num_stripes; i++) {
4845                 device = map->stripes[i].dev;
4846                 dev_offset = map->stripes[i].physical;
4847
4848                 btrfs_set_stack_stripe_devid(stripe, device->devid);
4849                 btrfs_set_stack_stripe_offset(stripe, dev_offset);
4850                 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
4851                 stripe++;
4852         }
4853
4854         btrfs_set_stack_chunk_length(chunk, chunk_size);
4855         btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
4856         btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
4857         btrfs_set_stack_chunk_type(chunk, map->type);
4858         btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
4859         btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
4860         btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
4861         btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
4862         btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
4863
4864         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
4865         key.type = BTRFS_CHUNK_ITEM_KEY;
4866         key.offset = chunk_offset;
4867
4868         ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
4869         if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
4870                 /*
4871                  * TODO: Cleanup of inserted chunk root in case of
4872                  * failure.
4873                  */
4874                 ret = btrfs_add_system_chunk(chunk_root, &key, chunk,
4875                                              item_size);
4876         }
4877
4878 out:
4879         kfree(chunk);
4880         free_extent_map(em);
4881         return ret;
4882 }
4883
4884 /*
4885  * Chunk allocation falls into two parts. The first part does works
4886  * that make the new allocated chunk useable, but not do any operation
4887  * that modifies the chunk tree. The second part does the works that
4888  * require modifying the chunk tree. This division is important for the
4889  * bootstrap process of adding storage to a seed btrfs.
4890  */
4891 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4892                       struct btrfs_root *extent_root, u64 type)
4893 {
4894         u64 chunk_offset;
4895
4896         ASSERT(mutex_is_locked(&extent_root->fs_info->chunk_mutex));
4897         chunk_offset = find_next_chunk(extent_root->fs_info);
4898         return __btrfs_alloc_chunk(trans, extent_root, chunk_offset, type);
4899 }
4900
4901 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
4902                                          struct btrfs_root *root,
4903                                          struct btrfs_device *device)
4904 {
4905         u64 chunk_offset;
4906         u64 sys_chunk_offset;
4907         u64 alloc_profile;
4908         struct btrfs_fs_info *fs_info = root->fs_info;
4909         struct btrfs_root *extent_root = fs_info->extent_root;
4910         int ret;
4911
4912         chunk_offset = find_next_chunk(fs_info);
4913         alloc_profile = btrfs_get_alloc_profile(extent_root, 0);
4914         ret = __btrfs_alloc_chunk(trans, extent_root, chunk_offset,
4915                                   alloc_profile);
4916         if (ret)
4917                 return ret;
4918
4919         sys_chunk_offset = find_next_chunk(root->fs_info);
4920         alloc_profile = btrfs_get_alloc_profile(fs_info->chunk_root, 0);
4921         ret = __btrfs_alloc_chunk(trans, extent_root, sys_chunk_offset,
4922                                   alloc_profile);
4923         return ret;
4924 }
4925
4926 static inline int btrfs_chunk_max_errors(struct map_lookup *map)
4927 {
4928         int max_errors;
4929
4930         if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
4931                          BTRFS_BLOCK_GROUP_RAID10 |
4932                          BTRFS_BLOCK_GROUP_RAID5 |
4933                          BTRFS_BLOCK_GROUP_DUP)) {
4934                 max_errors = 1;
4935         } else if (map->type & BTRFS_BLOCK_GROUP_RAID6) {
4936                 max_errors = 2;
4937         } else {
4938                 max_errors = 0;
4939         }
4940
4941         return max_errors;
4942 }
4943
4944 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
4945 {
4946         struct extent_map *em;
4947         struct map_lookup *map;
4948         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
4949         int readonly = 0;
4950         int miss_ndevs = 0;
4951         int i;
4952
4953         read_lock(&map_tree->map_tree.lock);
4954         em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
4955         read_unlock(&map_tree->map_tree.lock);
4956         if (!em)
4957                 return 1;
4958
4959         map = (struct map_lookup *)em->bdev;
4960         for (i = 0; i < map->num_stripes; i++) {
4961                 if (map->stripes[i].dev->missing) {
4962                         miss_ndevs++;
4963                         continue;
4964                 }
4965
4966                 if (!map->stripes[i].dev->writeable) {
4967                         readonly = 1;
4968                         goto end;
4969                 }
4970         }
4971
4972         /*
4973          * If the number of missing devices is larger than max errors,
4974          * we can not write the data into that chunk successfully, so
4975          * set it readonly.
4976          */
4977         if (miss_ndevs > btrfs_chunk_max_errors(map))
4978                 readonly = 1;
4979 end:
4980         free_extent_map(em);
4981         return readonly;
4982 }
4983
4984 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
4985 {
4986         extent_map_tree_init(&tree->map_tree);
4987 }
4988
4989 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
4990 {
4991         struct extent_map *em;
4992
4993         while (1) {
4994                 write_lock(&tree->map_tree.lock);
4995                 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
4996                 if (em)
4997                         remove_extent_mapping(&tree->map_tree, em);
4998                 write_unlock(&tree->map_tree.lock);
4999                 if (!em)
5000                         break;
5001                 /* once for us */
5002                 free_extent_map(em);
5003                 /* once for the tree */
5004                 free_extent_map(em);
5005         }
5006 }
5007
5008 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5009 {
5010         struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
5011         struct extent_map *em;
5012         struct map_lookup *map;
5013         struct extent_map_tree *em_tree = &map_tree->map_tree;
5014         int ret;
5015
5016         read_lock(&em_tree->lock);
5017         em = lookup_extent_mapping(em_tree, logical, len);
5018         read_unlock(&em_tree->lock);
5019
5020         /*
5021          * We could return errors for these cases, but that could get ugly and
5022          * we'd probably do the same thing which is just not do anything else
5023          * and exit, so return 1 so the callers don't try to use other copies.
5024          */
5025         if (!em) {
5026                 btrfs_crit(fs_info, "No mapping for %Lu-%Lu", logical,
5027                             logical+len);
5028                 return 1;
5029         }
5030
5031         if (em->start > logical || em->start + em->len < logical) {
5032                 btrfs_crit(fs_info, "Invalid mapping for %Lu-%Lu, got "
5033                             "%Lu-%Lu", logical, logical+len, em->start,
5034                             em->start + em->len);
5035                 free_extent_map(em);
5036                 return 1;
5037         }
5038
5039         map = (struct map_lookup *)em->bdev;
5040         if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
5041                 ret = map->num_stripes;
5042         else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5043                 ret = map->sub_stripes;
5044         else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
5045                 ret = 2;
5046         else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5047                 ret = 3;
5048         else
5049                 ret = 1;
5050         free_extent_map(em);
5051
5052         btrfs_dev_replace_lock(&fs_info->dev_replace);
5053         if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace))
5054                 ret++;
5055         btrfs_dev_replace_unlock(&fs_info->dev_replace);
5056
5057         return ret;
5058 }
5059
5060 unsigned long btrfs_full_stripe_len(struct btrfs_root *root,
5061                                     struct btrfs_mapping_tree *map_tree,
5062                                     u64 logical)
5063 {
5064         struct extent_map *em;
5065         struct map_lookup *map;
5066         struct extent_map_tree *em_tree = &map_tree->map_tree;
5067         unsigned long len = root->sectorsize;
5068
5069         read_lock(&em_tree->lock);
5070         em = lookup_extent_mapping(em_tree, logical, len);
5071         read_unlock(&em_tree->lock);
5072         BUG_ON(!em);
5073
5074         BUG_ON(em->start > logical || em->start + em->len < logical);
5075         map = (struct map_lookup *)em->bdev;
5076         if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5077                 len = map->stripe_len * nr_data_stripes(map);
5078         free_extent_map(em);
5079         return len;
5080 }
5081
5082 int btrfs_is_parity_mirror(struct btrfs_mapping_tree *map_tree,
5083                            u64 logical, u64 len, int mirror_num)
5084 {
5085         struct extent_map *em;
5086         struct map_lookup *map;
5087         struct extent_map_tree *em_tree = &map_tree->map_tree;
5088         int ret = 0;
5089
5090         read_lock(&em_tree->lock);
5091         em = lookup_extent_mapping(em_tree, logical, len);
5092         read_unlock(&em_tree->lock);
5093         BUG_ON(!em);
5094
5095         BUG_ON(em->start > logical || em->start + em->len < logical);
5096         map = (struct map_lookup *)em->bdev;
5097         if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5098                 ret = 1;
5099         free_extent_map(em);
5100         return ret;
5101 }
5102
5103 static int find_live_mirror(struct btrfs_fs_info *fs_info,
5104                             struct map_lookup *map, int first, int num,
5105                             int optimal, int dev_replace_is_ongoing)
5106 {
5107         int i;
5108         int tolerance;
5109         struct btrfs_device *srcdev;
5110
5111         if (dev_replace_is_ongoing &&
5112             fs_info->dev_replace.cont_reading_from_srcdev_mode ==
5113              BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
5114                 srcdev = fs_info->dev_replace.srcdev;
5115         else
5116                 srcdev = NULL;
5117
5118         /*
5119          * try to avoid the drive that is the source drive for a
5120          * dev-replace procedure, only choose it if no other non-missing
5121          * mirror is available
5122          */
5123         for (tolerance = 0; tolerance < 2; tolerance++) {
5124                 if (map->stripes[optimal].dev->bdev &&
5125                     (tolerance || map->stripes[optimal].dev != srcdev))
5126                         return optimal;
5127                 for (i = first; i < first + num; i++) {
5128                         if (map->stripes[i].dev->bdev &&
5129                             (tolerance || map->stripes[i].dev != srcdev))
5130                                 return i;
5131                 }
5132         }
5133
5134         /* we couldn't find one that doesn't fail.  Just return something
5135          * and the io error handling code will clean up eventually
5136          */
5137         return optimal;
5138 }
5139
5140 static inline int parity_smaller(u64 a, u64 b)
5141 {
5142         return a > b;
5143 }
5144
5145 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
5146 static void sort_parity_stripes(struct btrfs_bio *bbio, int num_stripes)
5147 {
5148         struct btrfs_bio_stripe s;
5149         int i;
5150         u64 l;
5151         int again = 1;
5152
5153         while (again) {
5154                 again = 0;
5155                 for (i = 0; i < num_stripes - 1; i++) {
5156                         if (parity_smaller(bbio->raid_map[i],
5157                                            bbio->raid_map[i+1])) {
5158                                 s = bbio->stripes[i];
5159                                 l = bbio->raid_map[i];
5160                                 bbio->stripes[i] = bbio->stripes[i+1];
5161                                 bbio->raid_map[i] = bbio->raid_map[i+1];
5162                                 bbio->stripes[i+1] = s;
5163                                 bbio->raid_map[i+1] = l;
5164
5165                                 again = 1;
5166                         }
5167                 }
5168         }
5169 }
5170
5171 static struct btrfs_bio *alloc_btrfs_bio(int total_stripes, int real_stripes)
5172 {
5173         struct btrfs_bio *bbio = kzalloc(
5174                  /* the size of the btrfs_bio */
5175                 sizeof(struct btrfs_bio) +
5176                 /* plus the variable array for the stripes */
5177                 sizeof(struct btrfs_bio_stripe) * (total_stripes) +
5178                 /* plus the variable array for the tgt dev */
5179                 sizeof(int) * (real_stripes) +
5180                 /*
5181                  * plus the raid_map, which includes both the tgt dev
5182                  * and the stripes
5183                  */
5184                 sizeof(u64) * (total_stripes),
5185                 GFP_NOFS|__GFP_NOFAIL);
5186
5187         atomic_set(&bbio->error, 0);
5188         atomic_set(&bbio->refs, 1);
5189
5190         return bbio;
5191 }
5192
5193 void btrfs_get_bbio(struct btrfs_bio *bbio)
5194 {
5195         WARN_ON(!atomic_read(&bbio->refs));
5196         atomic_inc(&bbio->refs);
5197 }
5198
5199 void btrfs_put_bbio(struct btrfs_bio *bbio)
5200 {
5201         if (!bbio)
5202                 return;
5203         if (atomic_dec_and_test(&bbio->refs))
5204                 kfree(bbio);
5205 }
5206
5207 static int __btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
5208                              u64 logical, u64 *length,
5209                              struct btrfs_bio **bbio_ret,
5210                              int mirror_num, int need_raid_map)
5211 {
5212         struct extent_map *em;
5213         struct map_lookup *map;
5214         struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
5215         struct extent_map_tree *em_tree = &map_tree->map_tree;
5216         u64 offset;
5217         u64 stripe_offset;
5218         u64 stripe_end_offset;
5219         u64 stripe_nr;
5220         u64 stripe_nr_orig;
5221         u64 stripe_nr_end;
5222         u64 stripe_len;
5223         u32 stripe_index;
5224         int i;
5225         int ret = 0;
5226         int num_stripes;
5227         int max_errors = 0;
5228         int tgtdev_indexes = 0;
5229         struct btrfs_bio *bbio = NULL;
5230         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
5231         int dev_replace_is_ongoing = 0;
5232         int num_alloc_stripes;
5233         int patch_the_first_stripe_for_dev_replace = 0;
5234         u64 physical_to_patch_in_first_stripe = 0;
5235         u64 raid56_full_stripe_start = (u64)-1;
5236
5237         read_lock(&em_tree->lock);
5238         em = lookup_extent_mapping(em_tree, logical, *length);
5239         read_unlock(&em_tree->lock);
5240
5241         if (!em) {
5242                 btrfs_crit(fs_info, "unable to find logical %llu len %llu",
5243                         logical, *length);
5244                 return -EINVAL;
5245         }
5246
5247         if (em->start > logical || em->start + em->len < logical) {
5248                 btrfs_crit(fs_info, "found a bad mapping, wanted %Lu, "
5249                            "found %Lu-%Lu", logical, em->start,
5250                            em->start + em->len);
5251                 free_extent_map(em);
5252                 return -EINVAL;
5253         }
5254
5255         map = (struct map_lookup *)em->bdev;
5256         offset = logical - em->start;
5257
5258         stripe_len = map->stripe_len;
5259         stripe_nr = offset;
5260         /*
5261          * stripe_nr counts the total number of stripes we have to stride
5262          * to get to this block
5263          */
5264         stripe_nr = div64_u64(stripe_nr, stripe_len);
5265
5266         stripe_offset = stripe_nr * stripe_len;
5267         BUG_ON(offset < stripe_offset);
5268
5269         /* stripe_offset is the offset of this block in its stripe*/
5270         stripe_offset = offset - stripe_offset;
5271
5272         /* if we're here for raid56, we need to know the stripe aligned start */
5273         if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5274                 unsigned long full_stripe_len = stripe_len * nr_data_stripes(map);
5275                 raid56_full_stripe_start = offset;
5276
5277                 /* allow a write of a full stripe, but make sure we don't
5278                  * allow straddling of stripes
5279                  */
5280                 raid56_full_stripe_start = div64_u64(raid56_full_stripe_start,
5281                                 full_stripe_len);
5282                 raid56_full_stripe_start *= full_stripe_len;
5283         }
5284
5285         if (rw & REQ_DISCARD) {
5286                 /* we don't discard raid56 yet */
5287                 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5288                         ret = -EOPNOTSUPP;
5289                         goto out;
5290                 }
5291                 *length = min_t(u64, em->len - offset, *length);
5292         } else if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
5293                 u64 max_len;
5294                 /* For writes to RAID[56], allow a full stripeset across all disks.
5295                    For other RAID types and for RAID[56] reads, just allow a single
5296                    stripe (on a single disk). */
5297                 if ((map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
5298                     (rw & REQ_WRITE)) {
5299                         max_len = stripe_len * nr_data_stripes(map) -
5300                                 (offset - raid56_full_stripe_start);
5301                 } else {
5302                         /* we limit the length of each bio to what fits in a stripe */
5303                         max_len = stripe_len - stripe_offset;
5304                 }
5305                 *length = min_t(u64, em->len - offset, max_len);
5306         } else {
5307                 *length = em->len - offset;
5308         }
5309
5310         /* This is for when we're called from btrfs_merge_bio_hook() and all
5311            it cares about is the length */
5312         if (!bbio_ret)
5313                 goto out;
5314
5315         btrfs_dev_replace_lock(dev_replace);
5316         dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
5317         if (!dev_replace_is_ongoing)
5318                 btrfs_dev_replace_unlock(dev_replace);
5319
5320         if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
5321             !(rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) &&
5322             dev_replace->tgtdev != NULL) {
5323                 /*
5324                  * in dev-replace case, for repair case (that's the only
5325                  * case where the mirror is selected explicitly when
5326                  * calling btrfs_map_block), blocks left of the left cursor
5327                  * can also be read from the target drive.
5328                  * For REQ_GET_READ_MIRRORS, the target drive is added as
5329                  * the last one to the array of stripes. For READ, it also
5330                  * needs to be supported using the same mirror number.
5331                  * If the requested block is not left of the left cursor,
5332                  * EIO is returned. This can happen because btrfs_num_copies()
5333                  * returns one more in the dev-replace case.
5334                  */
5335                 u64 tmp_length = *length;
5336                 struct btrfs_bio *tmp_bbio = NULL;
5337                 int tmp_num_stripes;
5338                 u64 srcdev_devid = dev_replace->srcdev->devid;
5339                 int index_srcdev = 0;
5340                 int found = 0;
5341                 u64 physical_of_found = 0;
5342
5343                 ret = __btrfs_map_block(fs_info, REQ_GET_READ_MIRRORS,
5344                              logical, &tmp_length, &tmp_bbio, 0, 0);
5345                 if (ret) {
5346                         WARN_ON(tmp_bbio != NULL);
5347                         goto out;
5348                 }
5349
5350                 tmp_num_stripes = tmp_bbio->num_stripes;
5351                 if (mirror_num > tmp_num_stripes) {
5352                         /*
5353                          * REQ_GET_READ_MIRRORS does not contain this
5354                          * mirror, that means that the requested area
5355                          * is not left of the left cursor
5356                          */
5357                         ret = -EIO;
5358                         btrfs_put_bbio(tmp_bbio);
5359                         goto out;
5360                 }
5361
5362                 /*
5363                  * process the rest of the function using the mirror_num
5364                  * of the source drive. Therefore look it up first.
5365                  * At the end, patch the device pointer to the one of the
5366                  * target drive.
5367                  */
5368                 for (i = 0; i < tmp_num_stripes; i++) {
5369                         if (tmp_bbio->stripes[i].dev->devid == srcdev_devid) {
5370                                 /*
5371                                  * In case of DUP, in order to keep it
5372                                  * simple, only add the mirror with the
5373                                  * lowest physical address
5374                                  */
5375                                 if (found &&
5376                                     physical_of_found <=
5377                                      tmp_bbio->stripes[i].physical)
5378                                         continue;
5379                                 index_srcdev = i;
5380                                 found = 1;
5381                                 physical_of_found =
5382                                         tmp_bbio->stripes[i].physical;
5383                         }
5384                 }
5385
5386                 if (found) {
5387                         mirror_num = index_srcdev + 1;
5388                         patch_the_first_stripe_for_dev_replace = 1;
5389                         physical_to_patch_in_first_stripe = physical_of_found;
5390                 } else {
5391                         WARN_ON(1);
5392                         ret = -EIO;
5393                         btrfs_put_bbio(tmp_bbio);
5394                         goto out;
5395                 }
5396
5397                 btrfs_put_bbio(tmp_bbio);
5398         } else if (mirror_num > map->num_stripes) {
5399                 mirror_num = 0;
5400         }
5401
5402         num_stripes = 1;
5403         stripe_index = 0;
5404         stripe_nr_orig = stripe_nr;
5405         stripe_nr_end = ALIGN(offset + *length, map->stripe_len);
5406         stripe_nr_end = div_u64(stripe_nr_end, map->stripe_len);
5407         stripe_end_offset = stripe_nr_end * map->stripe_len -
5408                             (offset + *length);
5409
5410         if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5411                 if (rw & REQ_DISCARD)
5412                         num_stripes = min_t(u64, map->num_stripes,
5413                                             stripe_nr_end - stripe_nr_orig);
5414                 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5415                                 &stripe_index);
5416                 if (!(rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)))
5417                         mirror_num = 1;
5418         } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
5419                 if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS))
5420                         num_stripes = map->num_stripes;
5421                 else if (mirror_num)
5422                         stripe_index = mirror_num - 1;
5423                 else {
5424                         stripe_index = find_live_mirror(fs_info, map, 0,
5425                                             map->num_stripes,
5426                                             current->pid % map->num_stripes,
5427                                             dev_replace_is_ongoing);
5428                         mirror_num = stripe_index + 1;
5429                 }
5430
5431         } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
5432                 if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) {
5433                         num_stripes = map->num_stripes;
5434                 } else if (mirror_num) {
5435                         stripe_index = mirror_num - 1;
5436                 } else {
5437                         mirror_num = 1;
5438                 }
5439
5440         } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5441                 u32 factor = map->num_stripes / map->sub_stripes;
5442
5443                 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
5444                 stripe_index *= map->sub_stripes;
5445
5446                 if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS))
5447                         num_stripes = map->sub_stripes;
5448                 else if (rw & REQ_DISCARD)
5449                         num_stripes = min_t(u64, map->sub_stripes *
5450                                             (stripe_nr_end - stripe_nr_orig),
5451                                             map->num_stripes);
5452                 else if (mirror_num)
5453                         stripe_index += mirror_num - 1;
5454                 else {
5455                         int old_stripe_index = stripe_index;
5456                         stripe_index = find_live_mirror(fs_info, map,
5457                                               stripe_index,
5458                                               map->sub_stripes, stripe_index +
5459                                               current->pid % map->sub_stripes,
5460                                               dev_replace_is_ongoing);
5461                         mirror_num = stripe_index - old_stripe_index + 1;
5462                 }
5463
5464         } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5465                 if (need_raid_map &&
5466                     ((rw & (REQ_WRITE | REQ_GET_READ_MIRRORS)) ||
5467                      mirror_num > 1)) {
5468                         /* push stripe_nr back to the start of the full stripe */
5469                         stripe_nr = div_u64(raid56_full_stripe_start,
5470                                         stripe_len * nr_data_stripes(map));
5471
5472                         /* RAID[56] write or recovery. Return all stripes */
5473                         num_stripes = map->num_stripes;
5474                         max_errors = nr_parity_stripes(map);
5475
5476                         *length = map->stripe_len;
5477                         stripe_index = 0;
5478                         stripe_offset = 0;
5479                 } else {
5480                         /*
5481                          * Mirror #0 or #1 means the original data block.
5482                          * Mirror #2 is RAID5 parity block.
5483                          * Mirror #3 is RAID6 Q block.
5484                          */
5485                         stripe_nr = div_u64_rem(stripe_nr,
5486                                         nr_data_stripes(map), &stripe_index);
5487                         if (mirror_num > 1)
5488                                 stripe_index = nr_data_stripes(map) +
5489                                                 mirror_num - 2;
5490
5491                         /* We distribute the parity blocks across stripes */
5492                         div_u64_rem(stripe_nr + stripe_index, map->num_stripes,
5493                                         &stripe_index);
5494                         if (!(rw & (REQ_WRITE | REQ_DISCARD |
5495                                     REQ_GET_READ_MIRRORS)) && mirror_num <= 1)
5496                                 mirror_num = 1;
5497                 }
5498         } else {
5499                 /*
5500                  * after this, stripe_nr is the number of stripes on this
5501                  * device we have to walk to find the data, and stripe_index is
5502                  * the number of our device in the stripe array
5503                  */
5504                 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5505                                 &stripe_index);
5506                 mirror_num = stripe_index + 1;
5507         }
5508         BUG_ON(stripe_index >= map->num_stripes);
5509
5510         num_alloc_stripes = num_stripes;
5511         if (dev_replace_is_ongoing) {
5512                 if (rw & (REQ_WRITE | REQ_DISCARD))
5513                         num_alloc_stripes <<= 1;
5514                 if (rw & REQ_GET_READ_MIRRORS)
5515                         num_alloc_stripes++;
5516                 tgtdev_indexes = num_stripes;
5517         }
5518
5519         bbio = alloc_btrfs_bio(num_alloc_stripes, tgtdev_indexes);
5520         if (!bbio) {
5521                 ret = -ENOMEM;
5522                 goto out;
5523         }
5524         if (dev_replace_is_ongoing)
5525                 bbio->tgtdev_map = (int *)(bbio->stripes + num_alloc_stripes);
5526
5527         /* build raid_map */
5528         if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK &&
5529             need_raid_map && ((rw & (REQ_WRITE | REQ_GET_READ_MIRRORS)) ||
5530             mirror_num > 1)) {
5531                 u64 tmp;
5532                 unsigned rot;
5533
5534                 bbio->raid_map = (u64 *)((void *)bbio->stripes +
5535                                  sizeof(struct btrfs_bio_stripe) *
5536                                  num_alloc_stripes +
5537                                  sizeof(int) * tgtdev_indexes);
5538
5539                 /* Work out the disk rotation on this stripe-set */
5540                 div_u64_rem(stripe_nr, num_stripes, &rot);
5541
5542                 /* Fill in the logical address of each stripe */
5543                 tmp = stripe_nr * nr_data_stripes(map);
5544                 for (i = 0; i < nr_data_stripes(map); i++)
5545                         bbio->raid_map[(i+rot) % num_stripes] =
5546                                 em->start + (tmp + i) * map->stripe_len;
5547
5548                 bbio->raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE;
5549                 if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5550                         bbio->raid_map[(i+rot+1) % num_stripes] =
5551                                 RAID6_Q_STRIPE;
5552         }
5553
5554         if (rw & REQ_DISCARD) {
5555                 u32 factor = 0;
5556                 u32 sub_stripes = 0;
5557                 u64 stripes_per_dev = 0;
5558                 u32 remaining_stripes = 0;
5559                 u32 last_stripe = 0;
5560
5561                 if (map->type &
5562                     (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID10)) {
5563                         if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5564                                 sub_stripes = 1;
5565                         else
5566                                 sub_stripes = map->sub_stripes;
5567
5568                         factor = map->num_stripes / sub_stripes;
5569                         stripes_per_dev = div_u64_rem(stripe_nr_end -
5570                                                       stripe_nr_orig,
5571                                                       factor,
5572                                                       &remaining_stripes);
5573                         div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
5574                         last_stripe *= sub_stripes;
5575                 }
5576
5577                 for (i = 0; i < num_stripes; i++) {
5578                         bbio->stripes[i].physical =
5579                                 map->stripes[stripe_index].physical +
5580                                 stripe_offset + stripe_nr * map->stripe_len;
5581                         bbio->stripes[i].dev = map->stripes[stripe_index].dev;
5582
5583                         if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5584                                          BTRFS_BLOCK_GROUP_RAID10)) {
5585                                 bbio->stripes[i].length = stripes_per_dev *
5586                                                           map->stripe_len;
5587
5588                                 if (i / sub_stripes < remaining_stripes)
5589                                         bbio->stripes[i].length +=
5590                                                 map->stripe_len;
5591
5592                                 /*
5593                                  * Special for the first stripe and
5594                                  * the last stripe:
5595                                  *
5596                                  * |-------|...|-------|
5597                                  *     |----------|
5598                                  *    off     end_off
5599                                  */
5600                                 if (i < sub_stripes)
5601                                         bbio->stripes[i].length -=
5602                                                 stripe_offset;
5603
5604                                 if (stripe_index >= last_stripe &&
5605                                     stripe_index <= (last_stripe +
5606                                                      sub_stripes - 1))
5607                                         bbio->stripes[i].length -=
5608                                                 stripe_end_offset;
5609
5610                                 if (i == sub_stripes - 1)
5611                                         stripe_offset = 0;
5612                         } else
5613                                 bbio->stripes[i].length = *length;
5614
5615                         stripe_index++;
5616                         if (stripe_index == map->num_stripes) {
5617                                 /* This could only happen for RAID0/10 */
5618                                 stripe_index = 0;
5619                                 stripe_nr++;
5620                         }
5621                 }
5622         } else {
5623                 for (i = 0; i < num_stripes; i++) {
5624                         bbio->stripes[i].physical =
5625                                 map->stripes[stripe_index].physical +
5626                                 stripe_offset +
5627                                 stripe_nr * map->stripe_len;
5628                         bbio->stripes[i].dev =
5629                                 map->stripes[stripe_index].dev;
5630                         stripe_index++;
5631                 }
5632         }
5633
5634         if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS))
5635                 max_errors = btrfs_chunk_max_errors(map);
5636
5637         if (bbio->raid_map)
5638                 sort_parity_stripes(bbio, num_stripes);
5639
5640         tgtdev_indexes = 0;
5641         if (dev_replace_is_ongoing && (rw & (REQ_WRITE | REQ_DISCARD)) &&
5642             dev_replace->tgtdev != NULL) {
5643                 int index_where_to_add;
5644                 u64 srcdev_devid = dev_replace->srcdev->devid;
5645
5646                 /*
5647                  * duplicate the write operations while the dev replace
5648                  * procedure is running. Since the copying of the old disk
5649                  * to the new disk takes place at run time while the
5650                  * filesystem is mounted writable, the regular write
5651                  * operations to the old disk have to be duplicated to go
5652                  * to the new disk as well.
5653                  * Note that device->missing is handled by the caller, and
5654                  * that the write to the old disk is already set up in the
5655                  * stripes array.
5656                  */
5657                 index_where_to_add = num_stripes;
5658                 for (i = 0; i < num_stripes; i++) {
5659                         if (bbio->stripes[i].dev->devid == srcdev_devid) {
5660                                 /* write to new disk, too */
5661                                 struct btrfs_bio_stripe *new =
5662                                         bbio->stripes + index_where_to_add;
5663                                 struct btrfs_bio_stripe *old =
5664                                         bbio->stripes + i;
5665
5666                                 new->physical = old->physical;
5667                                 new->length = old->length;
5668                                 new->dev = dev_replace->tgtdev;
5669                                 bbio->tgtdev_map[i] = index_where_to_add;
5670                                 index_where_to_add++;
5671                                 max_errors++;
5672                                 tgtdev_indexes++;
5673                         }
5674                 }
5675                 num_stripes = index_where_to_add;
5676         } else if (dev_replace_is_ongoing && (rw & REQ_GET_READ_MIRRORS) &&
5677                    dev_replace->tgtdev != NULL) {
5678                 u64 srcdev_devid = dev_replace->srcdev->devid;
5679                 int index_srcdev = 0;
5680                 int found = 0;
5681                 u64 physical_of_found = 0;
5682
5683                 /*
5684                  * During the dev-replace procedure, the target drive can
5685                  * also be used to read data in case it is needed to repair
5686                  * a corrupt block elsewhere. This is possible if the
5687                  * requested area is left of the left cursor. In this area,
5688                  * the target drive is a full copy of the source drive.
5689                  */
5690                 for (i = 0; i < num_stripes; i++) {
5691                         if (bbio->stripes[i].dev->devid == srcdev_devid) {
5692                                 /*
5693                                  * In case of DUP, in order to keep it
5694                                  * simple, only add the mirror with the
5695                                  * lowest physical address
5696                                  */
5697                                 if (found &&
5698                                     physical_of_found <=
5699                                      bbio->stripes[i].physical)
5700                                         continue;
5701                                 index_srcdev = i;
5702                                 found = 1;
5703                                 physical_of_found = bbio->stripes[i].physical;
5704                         }
5705                 }
5706                 if (found) {
5707                         if (physical_of_found + map->stripe_len <=
5708                             dev_replace->cursor_left) {
5709                                 struct btrfs_bio_stripe *tgtdev_stripe =
5710                                         bbio->stripes + num_stripes;
5711
5712                                 tgtdev_stripe->physical = physical_of_found;
5713                                 tgtdev_stripe->length =
5714                                         bbio->stripes[index_srcdev].length;
5715                                 tgtdev_stripe->dev = dev_replace->tgtdev;
5716                                 bbio->tgtdev_map[index_srcdev] = num_stripes;
5717
5718                                 tgtdev_indexes++;
5719                                 num_stripes++;
5720                         }
5721                 }
5722         }
5723
5724         *bbio_ret = bbio;
5725         bbio->map_type = map->type;
5726         bbio->num_stripes = num_stripes;
5727         bbio->max_errors = max_errors;
5728         bbio->mirror_num = mirror_num;
5729         bbio->num_tgtdevs = tgtdev_indexes;
5730
5731         /*
5732          * this is the case that REQ_READ && dev_replace_is_ongoing &&
5733          * mirror_num == num_stripes + 1 && dev_replace target drive is
5734          * available as a mirror
5735          */
5736         if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) {
5737                 WARN_ON(num_stripes > 1);
5738                 bbio->stripes[0].dev = dev_replace->tgtdev;
5739                 bbio->stripes[0].physical = physical_to_patch_in_first_stripe;
5740                 bbio->mirror_num = map->num_stripes + 1;
5741         }
5742 out:
5743         if (dev_replace_is_ongoing)
5744                 btrfs_dev_replace_unlock(dev_replace);
5745         free_extent_map(em);
5746         return ret;
5747 }
5748
5749 int btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
5750                       u64 logical, u64 *length,
5751                       struct btrfs_bio **bbio_ret, int mirror_num)
5752 {
5753         return __btrfs_map_block(fs_info, rw, logical, length, bbio_ret,
5754                                  mirror_num, 0);
5755 }
5756
5757 /* For Scrub/replace */
5758 int btrfs_map_sblock(struct btrfs_fs_info *fs_info, int rw,
5759                      u64 logical, u64 *length,
5760                      struct btrfs_bio **bbio_ret, int mirror_num,
5761                      int need_raid_map)
5762 {
5763         return __btrfs_map_block(fs_info, rw, logical, length, bbio_ret,
5764                                  mirror_num, need_raid_map);
5765 }
5766
5767 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
5768                      u64 chunk_start, u64 physical, u64 devid,
5769                      u64 **logical, int *naddrs, int *stripe_len)
5770 {
5771         struct extent_map_tree *em_tree = &map_tree->map_tree;
5772         struct extent_map *em;
5773         struct map_lookup *map;
5774         u64 *buf;
5775         u64 bytenr;
5776         u64 length;
5777         u64 stripe_nr;
5778         u64 rmap_len;
5779         int i, j, nr = 0;
5780
5781         read_lock(&em_tree->lock);
5782         em = lookup_extent_mapping(em_tree, chunk_start, 1);
5783         read_unlock(&em_tree->lock);
5784
5785         if (!em) {
5786                 printk(KERN_ERR "BTRFS: couldn't find em for chunk %Lu\n",
5787                        chunk_start);
5788                 return -EIO;
5789         }
5790
5791         if (em->start != chunk_start) {
5792                 printk(KERN_ERR "BTRFS: bad chunk start, em=%Lu, wanted=%Lu\n",
5793                        em->start, chunk_start);
5794                 free_extent_map(em);
5795                 return -EIO;
5796         }
5797         map = (struct map_lookup *)em->bdev;
5798
5799         length = em->len;
5800         rmap_len = map->stripe_len;
5801
5802         if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5803                 length = div_u64(length, map->num_stripes / map->sub_stripes);
5804         else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5805                 length = div_u64(length, map->num_stripes);
5806         else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5807                 length = div_u64(length, nr_data_stripes(map));
5808                 rmap_len = map->stripe_len * nr_data_stripes(map);
5809         }
5810
5811         buf = kcalloc(map->num_stripes, sizeof(u64), GFP_NOFS);
5812         BUG_ON(!buf); /* -ENOMEM */
5813
5814         for (i = 0; i < map->num_stripes; i++) {
5815                 if (devid && map->stripes[i].dev->devid != devid)
5816                         continue;
5817                 if (map->stripes[i].physical > physical ||
5818                     map->stripes[i].physical + length <= physical)
5819                         continue;
5820
5821                 stripe_nr = physical - map->stripes[i].physical;
5822                 stripe_nr = div_u64(stripe_nr, map->stripe_len);
5823
5824                 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5825                         stripe_nr = stripe_nr * map->num_stripes + i;
5826                         stripe_nr = div_u64(stripe_nr, map->sub_stripes);
5827                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5828                         stripe_nr = stripe_nr * map->num_stripes + i;
5829                 } /* else if RAID[56], multiply by nr_data_stripes().
5830                    * Alternatively, just use rmap_len below instead of
5831                    * map->stripe_len */
5832
5833                 bytenr = chunk_start + stripe_nr * rmap_len;
5834                 WARN_ON(nr >= map->num_stripes);
5835                 for (j = 0; j < nr; j++) {
5836                         if (buf[j] == bytenr)
5837                                 break;
5838                 }
5839                 if (j == nr) {
5840                         WARN_ON(nr >= map->num_stripes);
5841                         buf[nr++] = bytenr;
5842                 }
5843         }
5844
5845         *logical = buf;
5846         *naddrs = nr;
5847         *stripe_len = rmap_len;
5848
5849         free_extent_map(em);
5850         return 0;
5851 }
5852
5853 static inline void btrfs_end_bbio(struct btrfs_bio *bbio, struct bio *bio)
5854 {
5855         bio->bi_private = bbio->private;
5856         bio->bi_end_io = bbio->end_io;
5857         bio_endio(bio);
5858
5859         btrfs_put_bbio(bbio);
5860 }
5861
5862 static void btrfs_end_bio(struct bio *bio)
5863 {
5864         struct btrfs_bio *bbio = bio->bi_private;
5865         int is_orig_bio = 0;
5866
5867         if (bio->bi_error) {
5868                 atomic_inc(&bbio->error);
5869                 if (bio->bi_error == -EIO || bio->bi_error == -EREMOTEIO) {
5870                         unsigned int stripe_index =
5871                                 btrfs_io_bio(bio)->stripe_index;
5872                         struct btrfs_device *dev;
5873
5874                         BUG_ON(stripe_index >= bbio->num_stripes);
5875                         dev = bbio->stripes[stripe_index].dev;
5876                         if (dev->bdev) {
5877                                 if (bio->bi_rw & WRITE)
5878                                         btrfs_dev_stat_inc(dev,
5879                                                 BTRFS_DEV_STAT_WRITE_ERRS);
5880                                 else
5881                                         btrfs_dev_stat_inc(dev,
5882                                                 BTRFS_DEV_STAT_READ_ERRS);
5883                                 if ((bio->bi_rw & WRITE_FLUSH) == WRITE_FLUSH)
5884                                         btrfs_dev_stat_inc(dev,
5885                                                 BTRFS_DEV_STAT_FLUSH_ERRS);
5886                                 btrfs_dev_stat_print_on_error(dev);
5887                         }
5888                 }
5889         }
5890
5891         if (bio == bbio->orig_bio)
5892                 is_orig_bio = 1;
5893
5894         btrfs_bio_counter_dec(bbio->fs_info);
5895
5896         if (atomic_dec_and_test(&bbio->stripes_pending)) {
5897                 if (!is_orig_bio) {
5898                         bio_put(bio);
5899                         bio = bbio->orig_bio;
5900                 }
5901
5902                 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
5903                 /* only send an error to the higher layers if it is
5904                  * beyond the tolerance of the btrfs bio
5905                  */
5906                 if (atomic_read(&bbio->error) > bbio->max_errors) {
5907                         bio->bi_error = -EIO;
5908                 } else {
5909                         /*
5910                          * this bio is actually up to date, we didn't
5911                          * go over the max number of errors
5912                          */
5913                         bio->bi_error = 0;
5914                 }
5915
5916                 btrfs_end_bbio(bbio, bio);
5917         } else if (!is_orig_bio) {
5918                 bio_put(bio);
5919         }
5920 }
5921
5922 /*
5923  * see run_scheduled_bios for a description of why bios are collected for
5924  * async submit.
5925  *
5926  * This will add one bio to the pending list for a device and make sure
5927  * the work struct is scheduled.
5928  */
5929 static noinline void btrfs_schedule_bio(struct btrfs_root *root,
5930                                         struct btrfs_device *device,
5931                                         int rw, struct bio *bio)
5932 {
5933         int should_queue = 1;
5934         struct btrfs_pending_bios *pending_bios;
5935
5936         if (device->missing || !device->bdev) {
5937                 bio_io_error(bio);
5938                 return;
5939         }
5940
5941         /* don't bother with additional async steps for reads, right now */
5942         if (!(rw & REQ_WRITE)) {
5943                 bio_get(bio);
5944                 btrfsic_submit_bio(rw, bio);
5945                 bio_put(bio);
5946                 return;
5947         }
5948
5949         /*
5950          * nr_async_bios allows us to reliably return congestion to the
5951          * higher layers.  Otherwise, the async bio makes it appear we have
5952          * made progress against dirty pages when we've really just put it
5953          * on a queue for later
5954          */
5955         atomic_inc(&root->fs_info->nr_async_bios);
5956         WARN_ON(bio->bi_next);
5957         bio->bi_next = NULL;
5958         bio->bi_rw |= rw;
5959
5960         spin_lock(&device->io_lock);
5961         if (bio->bi_rw & REQ_SYNC)
5962                 pending_bios = &device->pending_sync_bios;
5963         else
5964                 pending_bios = &device->pending_bios;
5965
5966         if (pending_bios->tail)
5967                 pending_bios->tail->bi_next = bio;
5968
5969         pending_bios->tail = bio;
5970         if (!pending_bios->head)
5971                 pending_bios->head = bio;
5972         if (device->running_pending)
5973                 should_queue = 0;
5974
5975         spin_unlock(&device->io_lock);
5976
5977         if (should_queue)
5978                 btrfs_queue_work(root->fs_info->submit_workers,
5979                                  &device->work);
5980 }
5981
5982 static void submit_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio,
5983                               struct bio *bio, u64 physical, int dev_nr,
5984                               int rw, int async)
5985 {
5986         struct btrfs_device *dev = bbio->stripes[dev_nr].dev;
5987
5988         bio->bi_private = bbio;
5989         btrfs_io_bio(bio)->stripe_index = dev_nr;
5990         bio->bi_end_io = btrfs_end_bio;
5991         bio->bi_iter.bi_sector = physical >> 9;
5992 #ifdef DEBUG
5993         {
5994                 struct rcu_string *name;
5995
5996                 rcu_read_lock();
5997                 name = rcu_dereference(dev->name);
5998                 pr_debug("btrfs_map_bio: rw %d, sector=%llu, dev=%lu "
5999                          "(%s id %llu), size=%u\n", rw,
6000                          (u64)bio->bi_iter.bi_sector, (u_long)dev->bdev->bd_dev,
6001                          name->str, dev->devid, bio->bi_iter.bi_size);
6002                 rcu_read_unlock();
6003         }
6004 #endif
6005         bio->bi_bdev = dev->bdev;
6006
6007         btrfs_bio_counter_inc_noblocked(root->fs_info);
6008
6009         if (async)
6010                 btrfs_schedule_bio(root, dev, rw, bio);
6011         else
6012                 btrfsic_submit_bio(rw, bio);
6013 }
6014
6015 static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical)
6016 {
6017         atomic_inc(&bbio->error);
6018         if (atomic_dec_and_test(&bbio->stripes_pending)) {
6019                 /* Shoud be the original bio. */
6020                 WARN_ON(bio != bbio->orig_bio);
6021
6022                 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
6023                 bio->bi_iter.bi_sector = logical >> 9;
6024                 bio->bi_error = -EIO;
6025                 btrfs_end_bbio(bbio, bio);
6026         }
6027 }
6028
6029 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
6030                   int mirror_num, int async_submit)
6031 {
6032         struct btrfs_device *dev;
6033         struct bio *first_bio = bio;
6034         u64 logical = (u64)bio->bi_iter.bi_sector << 9;
6035         u64 length = 0;
6036         u64 map_length;
6037         int ret;
6038         int dev_nr;
6039         int total_devs;
6040         struct btrfs_bio *bbio = NULL;
6041
6042         length = bio->bi_iter.bi_size;
6043         map_length = length;
6044
6045         btrfs_bio_counter_inc_blocked(root->fs_info);
6046         ret = __btrfs_map_block(root->fs_info, rw, logical, &map_length, &bbio,
6047                               mirror_num, 1);
6048         if (ret) {
6049                 btrfs_bio_counter_dec(root->fs_info);
6050                 return ret;
6051         }
6052
6053         total_devs = bbio->num_stripes;
6054         bbio->orig_bio = first_bio;
6055         bbio->private = first_bio->bi_private;
6056         bbio->end_io = first_bio->bi_end_io;
6057         bbio->fs_info = root->fs_info;
6058         atomic_set(&bbio->stripes_pending, bbio->num_stripes);
6059
6060         if (bbio->raid_map) {
6061                 /* In this case, map_length has been set to the length of
6062                    a single stripe; not the whole write */
6063                 if (rw & WRITE) {
6064                         ret = raid56_parity_write(root, bio, bbio, map_length);
6065                 } else {
6066                         ret = raid56_parity_recover(root, bio, bbio, map_length,
6067                                                     mirror_num, 1);
6068                 }
6069
6070                 btrfs_bio_counter_dec(root->fs_info);
6071                 return ret;
6072         }
6073
6074         if (map_length < length) {
6075                 btrfs_crit(root->fs_info, "mapping failed logical %llu bio len %llu len %llu",
6076                         logical, length, map_length);
6077                 BUG();
6078         }
6079
6080         for (dev_nr = 0; dev_nr < total_devs; dev_nr++) {
6081                 dev = bbio->stripes[dev_nr].dev;
6082                 if (!dev || !dev->bdev || (rw & WRITE && !dev->writeable)) {
6083                         bbio_error(bbio, first_bio, logical);
6084                         continue;
6085                 }
6086
6087                 if (dev_nr < total_devs - 1) {
6088                         bio = btrfs_bio_clone(first_bio, GFP_NOFS);
6089                         BUG_ON(!bio); /* -ENOMEM */
6090                 } else
6091                         bio = first_bio;
6092
6093                 submit_stripe_bio(root, bbio, bio,
6094                                   bbio->stripes[dev_nr].physical, dev_nr, rw,
6095                                   async_submit);
6096         }
6097         btrfs_bio_counter_dec(root->fs_info);
6098         return 0;
6099 }
6100
6101 struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid,
6102                                        u8 *uuid, u8 *fsid)
6103 {
6104         struct btrfs_device *device;
6105         struct btrfs_fs_devices *cur_devices;
6106
6107         cur_devices = fs_info->fs_devices;
6108         while (cur_devices) {
6109                 if (!fsid ||
6110                     !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
6111                         device = __find_device(&cur_devices->devices,
6112                                                devid, uuid);
6113                         if (device)
6114                                 return device;
6115                 }
6116                 cur_devices = cur_devices->seed;
6117         }
6118         return NULL;
6119 }
6120
6121 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
6122                                             struct btrfs_fs_devices *fs_devices,
6123                                             u64 devid, u8 *dev_uuid)
6124 {
6125         struct btrfs_device *device;
6126
6127         device = btrfs_alloc_device(NULL, &devid, dev_uuid);
6128         if (IS_ERR(device))
6129                 return NULL;
6130
6131         list_add(&device->dev_list, &fs_devices->devices);
6132         device->fs_devices = fs_devices;
6133         fs_devices->num_devices++;
6134
6135         device->missing = 1;
6136         fs_devices->missing_devices++;
6137
6138         return device;
6139 }
6140
6141 /**
6142  * btrfs_alloc_device - allocate struct btrfs_device
6143  * @fs_info:    used only for generating a new devid, can be NULL if
6144  *              devid is provided (i.e. @devid != NULL).
6145  * @devid:      a pointer to devid for this device.  If NULL a new devid
6146  *              is generated.
6147  * @uuid:       a pointer to UUID for this device.  If NULL a new UUID
6148  *              is generated.
6149  *
6150  * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
6151  * on error.  Returned struct is not linked onto any lists and can be
6152  * destroyed with kfree() right away.
6153  */
6154 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
6155                                         const u64 *devid,
6156                                         const u8 *uuid)
6157 {
6158         struct btrfs_device *dev;
6159         u64 tmp;
6160
6161         if (WARN_ON(!devid && !fs_info))
6162                 return ERR_PTR(-EINVAL);
6163
6164         dev = __alloc_device();
6165         if (IS_ERR(dev))
6166                 return dev;
6167
6168         if (devid)
6169                 tmp = *devid;
6170         else {
6171                 int ret;
6172
6173                 ret = find_next_devid(fs_info, &tmp);
6174                 if (ret) {
6175                         kfree(dev);
6176                         return ERR_PTR(ret);
6177                 }
6178         }
6179         dev->devid = tmp;
6180
6181         if (uuid)
6182                 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE);
6183         else
6184                 generate_random_uuid(dev->uuid);
6185
6186         btrfs_init_work(&dev->work, btrfs_submit_helper,
6187                         pending_bios_fn, NULL, NULL);
6188
6189         return dev;
6190 }
6191
6192 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
6193                           struct extent_buffer *leaf,
6194                           struct btrfs_chunk *chunk)
6195 {
6196         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
6197         struct map_lookup *map;
6198         struct extent_map *em;
6199         u64 logical;
6200         u64 length;
6201         u64 devid;
6202         u8 uuid[BTRFS_UUID_SIZE];
6203         int num_stripes;
6204         int ret;
6205         int i;
6206
6207         logical = key->offset;
6208         length = btrfs_chunk_length(leaf, chunk);
6209
6210         read_lock(&map_tree->map_tree.lock);
6211         em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
6212         read_unlock(&map_tree->map_tree.lock);
6213
6214         /* already mapped? */
6215         if (em && em->start <= logical && em->start + em->len > logical) {
6216                 free_extent_map(em);
6217                 return 0;
6218         } else if (em) {
6219                 free_extent_map(em);
6220         }
6221
6222         em = alloc_extent_map();
6223         if (!em)
6224                 return -ENOMEM;
6225         num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6226         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
6227         if (!map) {
6228                 free_extent_map(em);
6229                 return -ENOMEM;
6230         }
6231
6232         set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
6233         em->bdev = (struct block_device *)map;
6234         em->start = logical;
6235         em->len = length;
6236         em->orig_start = 0;
6237         em->block_start = 0;
6238         em->block_len = em->len;
6239
6240         map->num_stripes = num_stripes;
6241         map->io_width = btrfs_chunk_io_width(leaf, chunk);
6242         map->io_align = btrfs_chunk_io_align(leaf, chunk);
6243         map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
6244         map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6245         map->type = btrfs_chunk_type(leaf, chunk);
6246         map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
6247         for (i = 0; i < num_stripes; i++) {
6248                 map->stripes[i].physical =
6249                         btrfs_stripe_offset_nr(leaf, chunk, i);
6250                 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
6251                 read_extent_buffer(leaf, uuid, (unsigned long)
6252                                    btrfs_stripe_dev_uuid_nr(chunk, i),
6253                                    BTRFS_UUID_SIZE);
6254                 map->stripes[i].dev = btrfs_find_device(root->fs_info, devid,
6255                                                         uuid, NULL);
6256                 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
6257                         free_extent_map(em);
6258                         return -EIO;
6259                 }
6260                 if (!map->stripes[i].dev) {
6261                         map->stripes[i].dev =
6262                                 add_missing_dev(root, root->fs_info->fs_devices,
6263                                                 devid, uuid);
6264                         if (!map->stripes[i].dev) {
6265                                 free_extent_map(em);
6266                                 return -EIO;
6267                         }
6268                         btrfs_warn(root->fs_info, "devid %llu uuid %pU is missing",
6269                                                 devid, uuid);
6270                 }
6271                 map->stripes[i].dev->in_fs_metadata = 1;
6272         }
6273
6274         write_lock(&map_tree->map_tree.lock);
6275         ret = add_extent_mapping(&map_tree->map_tree, em, 0);
6276         write_unlock(&map_tree->map_tree.lock);
6277         BUG_ON(ret); /* Tree corruption */
6278         free_extent_map(em);
6279
6280         return 0;
6281 }
6282
6283 static void fill_device_from_item(struct extent_buffer *leaf,
6284                                  struct btrfs_dev_item *dev_item,
6285                                  struct btrfs_device *device)
6286 {
6287         unsigned long ptr;
6288
6289         device->devid = btrfs_device_id(leaf, dev_item);
6290         device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
6291         device->total_bytes = device->disk_total_bytes;
6292         device->commit_total_bytes = device->disk_total_bytes;
6293         device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
6294         device->commit_bytes_used = device->bytes_used;
6295         device->type = btrfs_device_type(leaf, dev_item);
6296         device->io_align = btrfs_device_io_align(leaf, dev_item);
6297         device->io_width = btrfs_device_io_width(leaf, dev_item);
6298         device->sector_size = btrfs_device_sector_size(leaf, dev_item);
6299         WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
6300         device->is_tgtdev_for_dev_replace = 0;
6301
6302         ptr = btrfs_device_uuid(dev_item);
6303         read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
6304 }
6305
6306 static struct btrfs_fs_devices *open_seed_devices(struct btrfs_root *root,
6307                                                   u8 *fsid)
6308 {
6309         struct btrfs_fs_devices *fs_devices;
6310         int ret;
6311
6312         BUG_ON(!mutex_is_locked(&uuid_mutex));
6313
6314         fs_devices = root->fs_info->fs_devices->seed;
6315         while (fs_devices) {
6316                 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE))
6317                         return fs_devices;
6318
6319                 fs_devices = fs_devices->seed;
6320         }
6321
6322         fs_devices = find_fsid(fsid);
6323         if (!fs_devices) {
6324                 if (!btrfs_test_opt(root, DEGRADED))
6325                         return ERR_PTR(-ENOENT);
6326
6327                 fs_devices = alloc_fs_devices(fsid);
6328                 if (IS_ERR(fs_devices))
6329                         return fs_devices;
6330
6331                 fs_devices->seeding = 1;
6332                 fs_devices->opened = 1;
6333                 return fs_devices;
6334         }
6335
6336         fs_devices = clone_fs_devices(fs_devices);
6337         if (IS_ERR(fs_devices))
6338                 return fs_devices;
6339
6340         ret = __btrfs_open_devices(fs_devices, FMODE_READ,
6341                                    root->fs_info->bdev_holder);
6342         if (ret) {
6343                 free_fs_devices(fs_devices);
6344                 fs_devices = ERR_PTR(ret);
6345                 goto out;
6346         }
6347
6348         if (!fs_devices->seeding) {
6349                 __btrfs_close_devices(fs_devices);
6350                 free_fs_devices(fs_devices);
6351                 fs_devices = ERR_PTR(-EINVAL);
6352                 goto out;
6353         }
6354
6355         fs_devices->seed = root->fs_info->fs_devices->seed;
6356         root->fs_info->fs_devices->seed = fs_devices;
6357 out:
6358         return fs_devices;
6359 }
6360
6361 static int read_one_dev(struct btrfs_root *root,
6362                         struct extent_buffer *leaf,
6363                         struct btrfs_dev_item *dev_item)
6364 {
6365         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
6366         struct btrfs_device *device;
6367         u64 devid;
6368         int ret;
6369         u8 fs_uuid[BTRFS_UUID_SIZE];
6370         u8 dev_uuid[BTRFS_UUID_SIZE];
6371
6372         devid = btrfs_device_id(leaf, dev_item);
6373         read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
6374                            BTRFS_UUID_SIZE);
6375         read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
6376                            BTRFS_UUID_SIZE);
6377
6378         if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
6379                 fs_devices = open_seed_devices(root, fs_uuid);
6380                 if (IS_ERR(fs_devices))
6381                         return PTR_ERR(fs_devices);
6382         }
6383
6384         device = btrfs_find_device(root->fs_info, devid, dev_uuid, fs_uuid);
6385         if (!device) {
6386                 if (!btrfs_test_opt(root, DEGRADED))
6387                         return -EIO;
6388
6389                 device = add_missing_dev(root, fs_devices, devid, dev_uuid);
6390                 if (!device)
6391                         return -ENOMEM;
6392                 btrfs_warn(root->fs_info, "devid %llu uuid %pU missing",
6393                                 devid, dev_uuid);
6394         } else {
6395                 if (!device->bdev && !btrfs_test_opt(root, DEGRADED))
6396                         return -EIO;
6397
6398                 if(!device->bdev && !device->missing) {
6399                         /*
6400                          * this happens when a device that was properly setup
6401                          * in the device info lists suddenly goes bad.
6402                          * device->bdev is NULL, and so we have to set
6403                          * device->missing to one here
6404                          */
6405                         device->fs_devices->missing_devices++;
6406                         device->missing = 1;
6407                 }
6408
6409                 /* Move the device to its own fs_devices */
6410                 if (device->fs_devices != fs_devices) {
6411                         ASSERT(device->missing);
6412
6413                         list_move(&device->dev_list, &fs_devices->devices);
6414                         device->fs_devices->num_devices--;
6415                         fs_devices->num_devices++;
6416
6417                         device->fs_devices->missing_devices--;
6418                         fs_devices->missing_devices++;
6419
6420                         device->fs_devices = fs_devices;
6421                 }
6422         }
6423
6424         if (device->fs_devices != root->fs_info->fs_devices) {
6425                 BUG_ON(device->writeable);
6426                 if (device->generation !=
6427                     btrfs_device_generation(leaf, dev_item))
6428                         return -EINVAL;
6429         }
6430
6431         fill_device_from_item(leaf, dev_item, device);
6432         device->in_fs_metadata = 1;
6433         if (device->writeable && !device->is_tgtdev_for_dev_replace) {
6434                 device->fs_devices->total_rw_bytes += device->total_bytes;
6435                 spin_lock(&root->fs_info->free_chunk_lock);
6436                 root->fs_info->free_chunk_space += device->total_bytes -
6437                         device->bytes_used;
6438                 spin_unlock(&root->fs_info->free_chunk_lock);
6439         }
6440         ret = 0;
6441         return ret;
6442 }
6443
6444 int btrfs_read_sys_array(struct btrfs_root *root)
6445 {
6446         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
6447         struct extent_buffer *sb;
6448         struct btrfs_disk_key *disk_key;
6449         struct btrfs_chunk *chunk;
6450         u8 *array_ptr;
6451         unsigned long sb_array_offset;
6452         int ret = 0;
6453         u32 num_stripes;
6454         u32 array_size;
6455         u32 len = 0;
6456         u32 cur_offset;
6457         struct btrfs_key key;
6458
6459         ASSERT(BTRFS_SUPER_INFO_SIZE <= root->nodesize);
6460         /*
6461          * This will create extent buffer of nodesize, superblock size is
6462          * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will
6463          * overallocate but we can keep it as-is, only the first page is used.
6464          */
6465         sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET);
6466         if (!sb)
6467                 return -ENOMEM;
6468         btrfs_set_buffer_uptodate(sb);
6469         btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
6470         /*
6471          * The sb extent buffer is artifical and just used to read the system array.
6472          * btrfs_set_buffer_uptodate() call does not properly mark all it's
6473          * pages up-to-date when the page is larger: extent does not cover the
6474          * whole page and consequently check_page_uptodate does not find all
6475          * the page's extents up-to-date (the hole beyond sb),
6476          * write_extent_buffer then triggers a WARN_ON.
6477          *
6478          * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
6479          * but sb spans only this function. Add an explicit SetPageUptodate call
6480          * to silence the warning eg. on PowerPC 64.
6481          */
6482         if (PAGE_CACHE_SIZE > BTRFS_SUPER_INFO_SIZE)
6483                 SetPageUptodate(sb->pages[0]);
6484
6485         write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
6486         array_size = btrfs_super_sys_array_size(super_copy);
6487
6488         array_ptr = super_copy->sys_chunk_array;
6489         sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array);
6490         cur_offset = 0;
6491
6492         while (cur_offset < array_size) {
6493                 disk_key = (struct btrfs_disk_key *)array_ptr;
6494                 len = sizeof(*disk_key);
6495                 if (cur_offset + len > array_size)
6496                         goto out_short_read;
6497
6498                 btrfs_disk_key_to_cpu(&key, disk_key);
6499
6500                 array_ptr += len;
6501                 sb_array_offset += len;
6502                 cur_offset += len;
6503
6504                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
6505                         chunk = (struct btrfs_chunk *)sb_array_offset;
6506                         /*
6507                          * At least one btrfs_chunk with one stripe must be
6508                          * present, exact stripe count check comes afterwards
6509                          */
6510                         len = btrfs_chunk_item_size(1);
6511                         if (cur_offset + len > array_size)
6512                                 goto out_short_read;
6513
6514                         num_stripes = btrfs_chunk_num_stripes(sb, chunk);
6515                         len = btrfs_chunk_item_size(num_stripes);
6516                         if (cur_offset + len > array_size)
6517                                 goto out_short_read;
6518
6519                         ret = read_one_chunk(root, &key, sb, chunk);
6520                         if (ret)
6521                                 break;
6522                 } else {
6523                         ret = -EIO;
6524                         break;
6525                 }
6526                 array_ptr += len;
6527                 sb_array_offset += len;
6528                 cur_offset += len;
6529         }
6530         free_extent_buffer(sb);
6531         return ret;
6532
6533 out_short_read:
6534         printk(KERN_ERR "BTRFS: sys_array too short to read %u bytes at offset %u\n",
6535                         len, cur_offset);
6536         free_extent_buffer(sb);
6537         return -EIO;
6538 }
6539
6540 int btrfs_read_chunk_tree(struct btrfs_root *root)
6541 {
6542         struct btrfs_path *path;
6543         struct extent_buffer *leaf;
6544         struct btrfs_key key;
6545         struct btrfs_key found_key;
6546         int ret;
6547         int slot;
6548
6549         root = root->fs_info->chunk_root;
6550
6551         path = btrfs_alloc_path();
6552         if (!path)
6553                 return -ENOMEM;
6554
6555         mutex_lock(&uuid_mutex);
6556         lock_chunks(root);
6557
6558         /*
6559          * Read all device items, and then all the chunk items. All
6560          * device items are found before any chunk item (their object id
6561          * is smaller than the lowest possible object id for a chunk
6562          * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
6563          */
6564         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
6565         key.offset = 0;
6566         key.type = 0;
6567         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6568         if (ret < 0)
6569                 goto error;
6570         while (1) {
6571                 leaf = path->nodes[0];
6572                 slot = path->slots[0];
6573                 if (slot >= btrfs_header_nritems(leaf)) {
6574                         ret = btrfs_next_leaf(root, path);
6575                         if (ret == 0)
6576                                 continue;
6577                         if (ret < 0)
6578                                 goto error;
6579                         break;
6580                 }
6581                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6582                 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
6583                         struct btrfs_dev_item *dev_item;
6584                         dev_item = btrfs_item_ptr(leaf, slot,
6585                                                   struct btrfs_dev_item);
6586                         ret = read_one_dev(root, leaf, dev_item);
6587                         if (ret)
6588                                 goto error;
6589                 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
6590                         struct btrfs_chunk *chunk;
6591                         chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
6592                         ret = read_one_chunk(root, &found_key, leaf, chunk);
6593                         if (ret)
6594                                 goto error;
6595                 }
6596                 path->slots[0]++;
6597         }
6598         ret = 0;
6599 error:
6600         unlock_chunks(root);
6601         mutex_unlock(&uuid_mutex);
6602
6603         btrfs_free_path(path);
6604         return ret;
6605 }
6606
6607 void btrfs_init_devices_late(struct btrfs_fs_info *fs_info)
6608 {
6609         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6610         struct btrfs_device *device;
6611
6612         while (fs_devices) {
6613                 mutex_lock(&fs_devices->device_list_mutex);
6614                 list_for_each_entry(device, &fs_devices->devices, dev_list)
6615                         device->dev_root = fs_info->dev_root;
6616                 mutex_unlock(&fs_devices->device_list_mutex);
6617
6618                 fs_devices = fs_devices->seed;
6619         }
6620 }
6621
6622 static void __btrfs_reset_dev_stats(struct btrfs_device *dev)
6623 {
6624         int i;
6625
6626         for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6627                 btrfs_dev_stat_reset(dev, i);
6628 }
6629
6630 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
6631 {
6632         struct btrfs_key key;
6633         struct btrfs_key found_key;
6634         struct btrfs_root *dev_root = fs_info->dev_root;
6635         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6636         struct extent_buffer *eb;
6637         int slot;
6638         int ret = 0;
6639         struct btrfs_device *device;
6640         struct btrfs_path *path = NULL;
6641         int i;
6642
6643         path = btrfs_alloc_path();
6644         if (!path) {
6645                 ret = -ENOMEM;
6646                 goto out;
6647         }
6648
6649         mutex_lock(&fs_devices->device_list_mutex);
6650         list_for_each_entry(device, &fs_devices->devices, dev_list) {
6651                 int item_size;
6652                 struct btrfs_dev_stats_item *ptr;
6653
6654                 key.objectid = 0;
6655                 key.type = BTRFS_DEV_STATS_KEY;
6656                 key.offset = device->devid;
6657                 ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
6658                 if (ret) {
6659                         __btrfs_reset_dev_stats(device);
6660                         device->dev_stats_valid = 1;
6661                         btrfs_release_path(path);
6662                         continue;
6663                 }
6664                 slot = path->slots[0];
6665                 eb = path->nodes[0];
6666                 btrfs_item_key_to_cpu(eb, &found_key, slot);
6667                 item_size = btrfs_item_size_nr(eb, slot);
6668
6669                 ptr = btrfs_item_ptr(eb, slot,
6670                                      struct btrfs_dev_stats_item);
6671
6672                 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
6673                         if (item_size >= (1 + i) * sizeof(__le64))
6674                                 btrfs_dev_stat_set(device, i,
6675                                         btrfs_dev_stats_value(eb, ptr, i));
6676                         else
6677                                 btrfs_dev_stat_reset(device, i);
6678                 }
6679
6680                 device->dev_stats_valid = 1;
6681                 btrfs_dev_stat_print_on_load(device);
6682                 btrfs_release_path(path);
6683         }
6684         mutex_unlock(&fs_devices->device_list_mutex);
6685
6686 out:
6687         btrfs_free_path(path);
6688         return ret < 0 ? ret : 0;
6689 }
6690
6691 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
6692                                 struct btrfs_root *dev_root,
6693                                 struct btrfs_device *device)
6694 {
6695         struct btrfs_path *path;
6696         struct btrfs_key key;
6697         struct extent_buffer *eb;
6698         struct btrfs_dev_stats_item *ptr;
6699         int ret;
6700         int i;
6701
6702         key.objectid = 0;
6703         key.type = BTRFS_DEV_STATS_KEY;
6704         key.offset = device->devid;
6705
6706         path = btrfs_alloc_path();
6707         BUG_ON(!path);
6708         ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
6709         if (ret < 0) {
6710                 btrfs_warn_in_rcu(dev_root->fs_info,
6711                         "error %d while searching for dev_stats item for device %s",
6712                               ret, rcu_str_deref(device->name));
6713                 goto out;
6714         }
6715
6716         if (ret == 0 &&
6717             btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
6718                 /* need to delete old one and insert a new one */
6719                 ret = btrfs_del_item(trans, dev_root, path);
6720                 if (ret != 0) {
6721                         btrfs_warn_in_rcu(dev_root->fs_info,
6722                                 "delete too small dev_stats item for device %s failed %d",
6723                                       rcu_str_deref(device->name), ret);
6724                         goto out;
6725                 }
6726                 ret = 1;
6727         }
6728
6729         if (ret == 1) {
6730                 /* need to insert a new item */
6731                 btrfs_release_path(path);
6732                 ret = btrfs_insert_empty_item(trans, dev_root, path,
6733                                               &key, sizeof(*ptr));
6734                 if (ret < 0) {
6735                         btrfs_warn_in_rcu(dev_root->fs_info,
6736                                 "insert dev_stats item for device %s failed %d",
6737                                 rcu_str_deref(device->name), ret);
6738                         goto out;
6739                 }
6740         }
6741
6742         eb = path->nodes[0];
6743         ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
6744         for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6745                 btrfs_set_dev_stats_value(eb, ptr, i,
6746                                           btrfs_dev_stat_read(device, i));
6747         btrfs_mark_buffer_dirty(eb);
6748
6749 out:
6750         btrfs_free_path(path);
6751         return ret;
6752 }
6753
6754 /*
6755  * called from commit_transaction. Writes all changed device stats to disk.
6756  */
6757 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans,
6758                         struct btrfs_fs_info *fs_info)
6759 {
6760         struct btrfs_root *dev_root = fs_info->dev_root;
6761         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6762         struct btrfs_device *device;
6763         int stats_cnt;
6764         int ret = 0;
6765
6766         mutex_lock(&fs_devices->device_list_mutex);
6767         list_for_each_entry(device, &fs_devices->devices, dev_list) {
6768                 if (!device->dev_stats_valid || !btrfs_dev_stats_dirty(device))
6769                         continue;
6770
6771                 stats_cnt = atomic_read(&device->dev_stats_ccnt);
6772                 ret = update_dev_stat_item(trans, dev_root, device);
6773                 if (!ret)
6774                         atomic_sub(stats_cnt, &device->dev_stats_ccnt);
6775         }
6776         mutex_unlock(&fs_devices->device_list_mutex);
6777
6778         return ret;
6779 }
6780
6781 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
6782 {
6783         btrfs_dev_stat_inc(dev, index);
6784         btrfs_dev_stat_print_on_error(dev);
6785 }
6786
6787 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
6788 {
6789         if (!dev->dev_stats_valid)
6790                 return;
6791         btrfs_err_rl_in_rcu(dev->dev_root->fs_info,
6792                 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
6793                            rcu_str_deref(dev->name),
6794                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
6795                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
6796                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
6797                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
6798                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
6799 }
6800
6801 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
6802 {
6803         int i;
6804
6805         for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6806                 if (btrfs_dev_stat_read(dev, i) != 0)
6807                         break;
6808         if (i == BTRFS_DEV_STAT_VALUES_MAX)
6809                 return; /* all values == 0, suppress message */
6810
6811         btrfs_info_in_rcu(dev->dev_root->fs_info,
6812                 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
6813                rcu_str_deref(dev->name),
6814                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
6815                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
6816                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
6817                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
6818                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
6819 }
6820
6821 int btrfs_get_dev_stats(struct btrfs_root *root,
6822                         struct btrfs_ioctl_get_dev_stats *stats)
6823 {
6824         struct btrfs_device *dev;
6825         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
6826         int i;
6827
6828         mutex_lock(&fs_devices->device_list_mutex);
6829         dev = btrfs_find_device(root->fs_info, stats->devid, NULL, NULL);
6830         mutex_unlock(&fs_devices->device_list_mutex);
6831
6832         if (!dev) {
6833                 btrfs_warn(root->fs_info, "get dev_stats failed, device not found");
6834                 return -ENODEV;
6835         } else if (!dev->dev_stats_valid) {
6836                 btrfs_warn(root->fs_info, "get dev_stats failed, not yet valid");
6837                 return -ENODEV;
6838         } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
6839                 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
6840                         if (stats->nr_items > i)
6841                                 stats->values[i] =
6842                                         btrfs_dev_stat_read_and_reset(dev, i);
6843                         else
6844                                 btrfs_dev_stat_reset(dev, i);
6845                 }
6846         } else {
6847                 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6848                         if (stats->nr_items > i)
6849                                 stats->values[i] = btrfs_dev_stat_read(dev, i);
6850         }
6851         if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
6852                 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
6853         return 0;
6854 }
6855
6856 void btrfs_scratch_superblocks(struct block_device *bdev, char *device_path)
6857 {
6858         struct buffer_head *bh;
6859         struct btrfs_super_block *disk_super;
6860         int copy_num;
6861
6862         if (!bdev)
6863                 return;
6864
6865         for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX;
6866                 copy_num++) {
6867
6868                 if (btrfs_read_dev_one_super(bdev, copy_num, &bh))
6869                         continue;
6870
6871                 disk_super = (struct btrfs_super_block *)bh->b_data;
6872
6873                 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
6874                 set_buffer_dirty(bh);
6875                 sync_dirty_buffer(bh);
6876                 brelse(bh);
6877         }
6878
6879         /* Notify udev that device has changed */
6880         btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
6881
6882         /* Update ctime/mtime for device path for libblkid */
6883         update_dev_time(device_path);
6884 }
6885
6886 /*
6887  * Update the size of all devices, which is used for writing out the
6888  * super blocks.
6889  */
6890 void btrfs_update_commit_device_size(struct btrfs_fs_info *fs_info)
6891 {
6892         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6893         struct btrfs_device *curr, *next;
6894
6895         if (list_empty(&fs_devices->resized_devices))
6896                 return;
6897
6898         mutex_lock(&fs_devices->device_list_mutex);
6899         lock_chunks(fs_info->dev_root);
6900         list_for_each_entry_safe(curr, next, &fs_devices->resized_devices,
6901                                  resized_list) {
6902                 list_del_init(&curr->resized_list);
6903                 curr->commit_total_bytes = curr->disk_total_bytes;
6904         }
6905         unlock_chunks(fs_info->dev_root);
6906         mutex_unlock(&fs_devices->device_list_mutex);
6907 }
6908
6909 /* Must be invoked during the transaction commit */
6910 void btrfs_update_commit_device_bytes_used(struct btrfs_root *root,
6911                                         struct btrfs_transaction *transaction)
6912 {
6913         struct extent_map *em;
6914         struct map_lookup *map;
6915         struct btrfs_device *dev;
6916         int i;
6917
6918         if (list_empty(&transaction->pending_chunks))
6919                 return;
6920
6921         /* In order to kick the device replace finish process */
6922         lock_chunks(root);
6923         list_for_each_entry(em, &transaction->pending_chunks, list) {
6924                 map = (struct map_lookup *)em->bdev;
6925
6926                 for (i = 0; i < map->num_stripes; i++) {
6927                         dev = map->stripes[i].dev;
6928                         dev->commit_bytes_used = dev->bytes_used;
6929                 }
6930         }
6931         unlock_chunks(root);
6932 }
6933
6934 void btrfs_set_fs_info_ptr(struct btrfs_fs_info *fs_info)
6935 {
6936         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6937         while (fs_devices) {
6938                 fs_devices->fs_info = fs_info;
6939                 fs_devices = fs_devices->seed;
6940         }
6941 }
6942
6943 void btrfs_reset_fs_info_ptr(struct btrfs_fs_info *fs_info)
6944 {
6945         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6946         while (fs_devices) {
6947                 fs_devices->fs_info = NULL;
6948                 fs_devices = fs_devices->seed;
6949         }
6950 }
6951
6952 void btrfs_close_one_device(struct btrfs_device *device)
6953 {
6954         struct btrfs_fs_devices *fs_devices = device->fs_devices;
6955         struct btrfs_device *new_device;
6956         struct rcu_string *name;
6957
6958         if (device->bdev)
6959                 fs_devices->open_devices--;
6960
6961         if (device->writeable &&
6962             device->devid != BTRFS_DEV_REPLACE_DEVID) {
6963                 list_del_init(&device->dev_alloc_list);
6964                 fs_devices->rw_devices--;
6965         }
6966
6967         if (device->missing)
6968                 fs_devices->missing_devices--;
6969
6970         new_device = btrfs_alloc_device(NULL, &device->devid,
6971                                         device->uuid);
6972         BUG_ON(IS_ERR(new_device)); /* -ENOMEM */
6973
6974         /* Safe because we are under uuid_mutex */
6975         if (device->name) {
6976                 name = rcu_string_strdup(device->name->str, GFP_NOFS);
6977                 BUG_ON(!name); /* -ENOMEM */
6978                 rcu_assign_pointer(new_device->name, name);
6979         }
6980
6981         list_replace_rcu(&device->dev_list, &new_device->dev_list);
6982         new_device->fs_devices = device->fs_devices;
6983
6984         call_rcu(&device->rcu, free_device);
6985 }