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_WAIT);
 236         INIT_RADIX_TREE(&dev->reada_extents, GFP_NOFS & ~__GFP_WAIT);
 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,
1977                                         rcu_str_deref(srcdev->name));
1978         }
1979
1980         if (srcdev->bdev)
1981                 fs_devices->open_devices--;
1982 }
1983
1984 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_fs_info *fs_info,
1985                                       struct btrfs_device *srcdev)
1986 {
1987         struct btrfs_fs_devices *fs_devices = srcdev->fs_devices;
1988
1989         call_rcu(&srcdev->rcu, free_device);
1990
1991         /*
1992          * unless fs_devices is seed fs, num_devices shouldn't go
1993          * zero
1994          */
1995         BUG_ON(!fs_devices->num_devices && !fs_devices->seeding);
1996
1997         /* if this is no devs we rather delete the fs_devices */
1998         if (!fs_devices->num_devices) {
1999                 struct btrfs_fs_devices *tmp_fs_devices;
2000
2001                 tmp_fs_devices = fs_info->fs_devices;
2002                 while (tmp_fs_devices) {
2003                         if (tmp_fs_devices->seed == fs_devices) {
2004                                 tmp_fs_devices->seed = fs_devices->seed;
2005                                 break;
2006                         }
2007                         tmp_fs_devices = tmp_fs_devices->seed;
2008                 }
2009                 fs_devices->seed = NULL;
2010                 __btrfs_close_devices(fs_devices);
2011                 free_fs_devices(fs_devices);
2012         }
2013 }
2014
2015 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
2016                                       struct btrfs_device *tgtdev)
2017 {
2018         struct btrfs_device *next_device;
2019
2020         mutex_lock(&uuid_mutex);
2021         WARN_ON(!tgtdev);
2022         mutex_lock(&fs_info->fs_devices->device_list_mutex);
2023
2024         btrfs_sysfs_rm_device_link(fs_info->fs_devices, tgtdev);
2025
2026         if (tgtdev->bdev) {
2027                 btrfs_scratch_superblocks(tgtdev->bdev,
2028                                         rcu_str_deref(tgtdev->name));
2029                 fs_info->fs_devices->open_devices--;
2030         }
2031         fs_info->fs_devices->num_devices--;
2032
2033         next_device = list_entry(fs_info->fs_devices->devices.next,
2034                                  struct btrfs_device, dev_list);
2035         if (tgtdev->bdev == fs_info->sb->s_bdev)
2036                 fs_info->sb->s_bdev = next_device->bdev;
2037         if (tgtdev->bdev == fs_info->fs_devices->latest_bdev)
2038                 fs_info->fs_devices->latest_bdev = next_device->bdev;
2039         list_del_rcu(&tgtdev->dev_list);
2040
2041         call_rcu(&tgtdev->rcu, free_device);
2042
2043         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2044         mutex_unlock(&uuid_mutex);
2045 }
2046
2047 static int btrfs_find_device_by_path(struct btrfs_root *root, char *device_path,
2048                                      struct btrfs_device **device)
2049 {
2050         int ret = 0;
2051         struct btrfs_super_block *disk_super;
2052         u64 devid;
2053         u8 *dev_uuid;
2054         struct block_device *bdev;
2055         struct buffer_head *bh;
2056
2057         *device = NULL;
2058         ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
2059                                     root->fs_info->bdev_holder, 0, &bdev, &bh);
2060         if (ret)
2061                 return ret;
2062         disk_super = (struct btrfs_super_block *)bh->b_data;
2063         devid = btrfs_stack_device_id(&disk_super->dev_item);
2064         dev_uuid = disk_super->dev_item.uuid;
2065         *device = btrfs_find_device(root->fs_info, devid, dev_uuid,
2066                                     disk_super->fsid);
2067         brelse(bh);
2068         if (!*device)
2069                 ret = -ENOENT;
2070         blkdev_put(bdev, FMODE_READ);
2071         return ret;
2072 }
2073
2074 int btrfs_find_device_missing_or_by_path(struct btrfs_root *root,
2075                                          char *device_path,
2076                                          struct btrfs_device **device)
2077 {
2078         *device = NULL;
2079         if (strcmp(device_path, "missing") == 0) {
2080                 struct list_head *devices;
2081                 struct btrfs_device *tmp;
2082
2083                 devices = &root->fs_info->fs_devices->devices;
2084                 /*
2085                  * It is safe to read the devices since the volume_mutex
2086                  * is held by the caller.
2087                  */
2088                 list_for_each_entry(tmp, devices, dev_list) {
2089                         if (tmp->in_fs_metadata && !tmp->bdev) {
2090                                 *device = tmp;
2091                                 break;
2092                         }
2093                 }
2094
2095                 if (!*device)
2096                         return BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
2097
2098                 return 0;
2099         } else {
2100                 return btrfs_find_device_by_path(root, device_path, device);
2101         }
2102 }
2103
2104 /*
2105  * does all the dirty work required for changing file system's UUID.
2106  */
2107 static int btrfs_prepare_sprout(struct btrfs_root *root)
2108 {
2109         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2110         struct btrfs_fs_devices *old_devices;
2111         struct btrfs_fs_devices *seed_devices;
2112         struct btrfs_super_block *disk_super = root->fs_info->super_copy;
2113         struct btrfs_device *device;
2114         u64 super_flags;
2115
2116         BUG_ON(!mutex_is_locked(&uuid_mutex));
2117         if (!fs_devices->seeding)
2118                 return -EINVAL;
2119
2120         seed_devices = __alloc_fs_devices();
2121         if (IS_ERR(seed_devices))
2122                 return PTR_ERR(seed_devices);
2123
2124         old_devices = clone_fs_devices(fs_devices);
2125         if (IS_ERR(old_devices)) {
2126                 kfree(seed_devices);
2127                 return PTR_ERR(old_devices);
2128         }
2129
2130         list_add(&old_devices->list, &fs_uuids);
2131
2132         memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
2133         seed_devices->opened = 1;
2134         INIT_LIST_HEAD(&seed_devices->devices);
2135         INIT_LIST_HEAD(&seed_devices->alloc_list);
2136         mutex_init(&seed_devices->device_list_mutex);
2137
2138         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2139         list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
2140                               synchronize_rcu);
2141         list_for_each_entry(device, &seed_devices->devices, dev_list)
2142                 device->fs_devices = seed_devices;
2143
2144         lock_chunks(root);
2145         list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
2146         unlock_chunks(root);
2147
2148         fs_devices->seeding = 0;
2149         fs_devices->num_devices = 0;
2150         fs_devices->open_devices = 0;
2151         fs_devices->missing_devices = 0;
2152         fs_devices->rotating = 0;
2153         fs_devices->seed = seed_devices;
2154
2155         generate_random_uuid(fs_devices->fsid);
2156         memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2157         memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2158         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2159
2160         super_flags = btrfs_super_flags(disk_super) &
2161                       ~BTRFS_SUPER_FLAG_SEEDING;
2162         btrfs_set_super_flags(disk_super, super_flags);
2163
2164         return 0;
2165 }
2166
2167 /*
2168  * strore the expected generation for seed devices in device items.
2169  */
2170 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
2171                                struct btrfs_root *root)
2172 {
2173         struct btrfs_path *path;
2174         struct extent_buffer *leaf;
2175         struct btrfs_dev_item *dev_item;
2176         struct btrfs_device *device;
2177         struct btrfs_key key;
2178         u8 fs_uuid[BTRFS_UUID_SIZE];
2179         u8 dev_uuid[BTRFS_UUID_SIZE];
2180         u64 devid;
2181         int ret;
2182
2183         path = btrfs_alloc_path();
2184         if (!path)
2185                 return -ENOMEM;
2186
2187         root = root->fs_info->chunk_root;
2188         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2189         key.offset = 0;
2190         key.type = BTRFS_DEV_ITEM_KEY;
2191
2192         while (1) {
2193                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2194                 if (ret < 0)
2195                         goto error;
2196
2197                 leaf = path->nodes[0];
2198 next_slot:
2199                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2200                         ret = btrfs_next_leaf(root, path);
2201                         if (ret > 0)
2202                                 break;
2203                         if (ret < 0)
2204                                 goto error;
2205                         leaf = path->nodes[0];
2206                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2207                         btrfs_release_path(path);
2208                         continue;
2209                 }
2210
2211                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2212                 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
2213                     key.type != BTRFS_DEV_ITEM_KEY)
2214                         break;
2215
2216                 dev_item = btrfs_item_ptr(leaf, path->slots[0],
2217                                           struct btrfs_dev_item);
2218                 devid = btrfs_device_id(leaf, dev_item);
2219                 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
2220                                    BTRFS_UUID_SIZE);
2221                 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
2222                                    BTRFS_UUID_SIZE);
2223                 device = btrfs_find_device(root->fs_info, devid, dev_uuid,
2224                                            fs_uuid);
2225                 BUG_ON(!device); /* Logic error */
2226
2227                 if (device->fs_devices->seeding) {
2228                         btrfs_set_device_generation(leaf, dev_item,
2229                                                     device->generation);
2230                         btrfs_mark_buffer_dirty(leaf);
2231                 }
2232
2233                 path->slots[0]++;
2234                 goto next_slot;
2235         }
2236         ret = 0;
2237 error:
2238         btrfs_free_path(path);
2239         return ret;
2240 }
2241
2242 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
2243 {
2244         struct request_queue *q;
2245         struct btrfs_trans_handle *trans;
2246         struct btrfs_device *device;
2247         struct block_device *bdev;
2248         struct list_head *devices;
2249         struct super_block *sb = root->fs_info->sb;
2250         struct rcu_string *name;
2251         u64 tmp;
2252         int seeding_dev = 0;
2253         int ret = 0;
2254
2255         if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
2256                 return -EROFS;
2257
2258         bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2259                                   root->fs_info->bdev_holder);
2260         if (IS_ERR(bdev))
2261                 return PTR_ERR(bdev);
2262
2263         if (root->fs_info->fs_devices->seeding) {
2264                 seeding_dev = 1;
2265                 down_write(&sb->s_umount);
2266                 mutex_lock(&uuid_mutex);
2267         }
2268
2269         filemap_write_and_wait(bdev->bd_inode->i_mapping);
2270
2271         devices = &root->fs_info->fs_devices->devices;
2272
2273         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2274         list_for_each_entry(device, devices, dev_list) {
2275                 if (device->bdev == bdev) {
2276                         ret = -EEXIST;
2277                         mutex_unlock(
2278                                 &root->fs_info->fs_devices->device_list_mutex);
2279                         goto error;
2280                 }
2281         }
2282         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2283
2284         device = btrfs_alloc_device(root->fs_info, NULL, NULL);
2285         if (IS_ERR(device)) {
2286                 /* we can safely leave the fs_devices entry around */
2287                 ret = PTR_ERR(device);
2288                 goto error;
2289         }
2290
2291         name = rcu_string_strdup(device_path, GFP_NOFS);
2292         if (!name) {
2293                 kfree(device);
2294                 ret = -ENOMEM;
2295                 goto error;
2296         }
2297         rcu_assign_pointer(device->name, name);
2298
2299         trans = btrfs_start_transaction(root, 0);
2300         if (IS_ERR(trans)) {
2301                 rcu_string_free(device->name);
2302                 kfree(device);
2303                 ret = PTR_ERR(trans);
2304                 goto error;
2305         }
2306
2307         q = bdev_get_queue(bdev);
2308         if (blk_queue_discard(q))
2309                 device->can_discard = 1;
2310         device->writeable = 1;
2311         device->generation = trans->transid;
2312         device->io_width = root->sectorsize;
2313         device->io_align = root->sectorsize;
2314         device->sector_size = root->sectorsize;
2315         device->total_bytes = i_size_read(bdev->bd_inode);
2316         device->disk_total_bytes = device->total_bytes;
2317         device->commit_total_bytes = device->total_bytes;
2318         device->dev_root = root->fs_info->dev_root;
2319         device->bdev = bdev;
2320         device->in_fs_metadata = 1;
2321         device->is_tgtdev_for_dev_replace = 0;
2322         device->mode = FMODE_EXCL;
2323         device->dev_stats_valid = 1;
2324         set_blocksize(device->bdev, 4096);
2325
2326         if (seeding_dev) {
2327                 sb->s_flags &= ~MS_RDONLY;
2328                 ret = btrfs_prepare_sprout(root);
2329                 BUG_ON(ret); /* -ENOMEM */
2330         }
2331
2332         device->fs_devices = root->fs_info->fs_devices;
2333
2334         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2335         lock_chunks(root);
2336         list_add_rcu(&device->dev_list, &root->fs_info->fs_devices->devices);
2337         list_add(&device->dev_alloc_list,
2338                  &root->fs_info->fs_devices->alloc_list);
2339         root->fs_info->fs_devices->num_devices++;
2340         root->fs_info->fs_devices->open_devices++;
2341         root->fs_info->fs_devices->rw_devices++;
2342         root->fs_info->fs_devices->total_devices++;
2343         root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
2344
2345         spin_lock(&root->fs_info->free_chunk_lock);
2346         root->fs_info->free_chunk_space += device->total_bytes;
2347         spin_unlock(&root->fs_info->free_chunk_lock);
2348
2349         if (!blk_queue_nonrot(bdev_get_queue(bdev)))
2350                 root->fs_info->fs_devices->rotating = 1;
2351
2352         tmp = btrfs_super_total_bytes(root->fs_info->super_copy);
2353         btrfs_set_super_total_bytes(root->fs_info->super_copy,
2354                                     tmp + device->total_bytes);
2355
2356         tmp = btrfs_super_num_devices(root->fs_info->super_copy);
2357         btrfs_set_super_num_devices(root->fs_info->super_copy,
2358                                     tmp + 1);
2359
2360         /* add sysfs device entry */
2361         btrfs_sysfs_add_device_link(root->fs_info->fs_devices, device);
2362
2363         /*
2364          * we've got more storage, clear any full flags on the space
2365          * infos
2366          */
2367         btrfs_clear_space_info_full(root->fs_info);
2368
2369         unlock_chunks(root);
2370         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2371
2372         if (seeding_dev) {
2373                 lock_chunks(root);
2374                 ret = init_first_rw_device(trans, root, device);
2375                 unlock_chunks(root);
2376                 if (ret) {
2377                         btrfs_abort_transaction(trans, root, ret);
2378                         goto error_trans;
2379                 }
2380         }
2381
2382         ret = btrfs_add_device(trans, root, device);
2383         if (ret) {
2384                 btrfs_abort_transaction(trans, root, ret);
2385                 goto error_trans;
2386         }
2387
2388         if (seeding_dev) {
2389                 char fsid_buf[BTRFS_UUID_UNPARSED_SIZE];
2390
2391                 ret = btrfs_finish_sprout(trans, root);
2392                 if (ret) {
2393                         btrfs_abort_transaction(trans, root, ret);
2394                         goto error_trans;
2395                 }
2396
2397                 /* Sprouting would change fsid of the mounted root,
2398                  * so rename the fsid on the sysfs
2399                  */
2400                 snprintf(fsid_buf, BTRFS_UUID_UNPARSED_SIZE, "%pU",
2401                                                 root->fs_info->fsid);
2402                 if (kobject_rename(&root->fs_info->fs_devices->fsid_kobj,
2403                                                                 fsid_buf))
2404                         btrfs_warn(root->fs_info,
2405                                 "sysfs: failed to create fsid for sprout");
2406         }
2407
2408         root->fs_info->num_tolerated_disk_barrier_failures =
2409                 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
2410         ret = btrfs_commit_transaction(trans, root);
2411
2412         if (seeding_dev) {
2413                 mutex_unlock(&uuid_mutex);
2414                 up_write(&sb->s_umount);
2415
2416                 if (ret) /* transaction commit */
2417                         return ret;
2418
2419                 ret = btrfs_relocate_sys_chunks(root);
2420                 if (ret < 0)
2421                         btrfs_std_error(root->fs_info, ret,
2422                                     "Failed to relocate sys chunks after "
2423                                     "device initialization. This can be fixed "
2424                                     "using the \"btrfs balance\" command.");
2425                 trans = btrfs_attach_transaction(root);
2426                 if (IS_ERR(trans)) {
2427                         if (PTR_ERR(trans) == -ENOENT)
2428                                 return 0;
2429                         return PTR_ERR(trans);
2430                 }
2431                 ret = btrfs_commit_transaction(trans, root);
2432         }
2433
2434         /* Update ctime/mtime for libblkid */
2435         update_dev_time(device_path);
2436         return ret;
2437
2438 error_trans:
2439         btrfs_end_transaction(trans, root);
2440         rcu_string_free(device->name);
2441         btrfs_sysfs_rm_device_link(root->fs_info->fs_devices, device);
2442         kfree(device);
2443 error:
2444         blkdev_put(bdev, FMODE_EXCL);
2445         if (seeding_dev) {
2446                 mutex_unlock(&uuid_mutex);
2447                 up_write(&sb->s_umount);
2448         }
2449         return ret;
2450 }
2451
2452 int btrfs_init_dev_replace_tgtdev(struct btrfs_root *root, char *device_path,
2453                                   struct btrfs_device *srcdev,
2454                                   struct btrfs_device **device_out)
2455 {
2456         struct request_queue *q;
2457         struct btrfs_device *device;
2458         struct block_device *bdev;
2459         struct btrfs_fs_info *fs_info = root->fs_info;
2460         struct list_head *devices;
2461         struct rcu_string *name;
2462         u64 devid = BTRFS_DEV_REPLACE_DEVID;
2463         int ret = 0;
2464
2465         *device_out = NULL;
2466         if (fs_info->fs_devices->seeding) {
2467                 btrfs_err(fs_info, "the filesystem is a seed filesystem!");
2468                 return -EINVAL;
2469         }
2470
2471         bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2472                                   fs_info->bdev_holder);
2473         if (IS_ERR(bdev)) {
2474                 btrfs_err(fs_info, "target device %s is invalid!", device_path);
2475                 return PTR_ERR(bdev);
2476         }
2477
2478         filemap_write_and_wait(bdev->bd_inode->i_mapping);
2479
2480         devices = &fs_info->fs_devices->devices;
2481         list_for_each_entry(device, devices, dev_list) {
2482                 if (device->bdev == bdev) {
2483                         btrfs_err(fs_info, "target device is in the filesystem!");
2484                         ret = -EEXIST;
2485                         goto error;
2486                 }
2487         }
2488
2489
2490         if (i_size_read(bdev->bd_inode) <
2491             btrfs_device_get_total_bytes(srcdev)) {
2492                 btrfs_err(fs_info, "target device is smaller than source device!");
2493                 ret = -EINVAL;
2494                 goto error;
2495         }
2496
2497
2498         device = btrfs_alloc_device(NULL, &devid, NULL);
2499         if (IS_ERR(device)) {
2500                 ret = PTR_ERR(device);
2501                 goto error;
2502         }
2503
2504         name = rcu_string_strdup(device_path, GFP_NOFS);
2505         if (!name) {
2506                 kfree(device);
2507                 ret = -ENOMEM;
2508                 goto error;
2509         }
2510         rcu_assign_pointer(device->name, name);
2511
2512         q = bdev_get_queue(bdev);
2513         if (blk_queue_discard(q))
2514                 device->can_discard = 1;
2515         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2516         device->writeable = 1;
2517         device->generation = 0;
2518         device->io_width = root->sectorsize;
2519         device->io_align = root->sectorsize;
2520         device->sector_size = root->sectorsize;
2521         device->total_bytes = btrfs_device_get_total_bytes(srcdev);
2522         device->disk_total_bytes = btrfs_device_get_disk_total_bytes(srcdev);
2523         device->bytes_used = btrfs_device_get_bytes_used(srcdev);
2524         ASSERT(list_empty(&srcdev->resized_list));
2525         device->commit_total_bytes = srcdev->commit_total_bytes;
2526         device->commit_bytes_used = device->bytes_used;
2527         device->dev_root = fs_info->dev_root;
2528         device->bdev = bdev;
2529         device->in_fs_metadata = 1;
2530         device->is_tgtdev_for_dev_replace = 1;
2531         device->mode = FMODE_EXCL;
2532         device->dev_stats_valid = 1;
2533         set_blocksize(device->bdev, 4096);
2534         device->fs_devices = fs_info->fs_devices;
2535         list_add(&device->dev_list, &fs_info->fs_devices->devices);
2536         fs_info->fs_devices->num_devices++;
2537         fs_info->fs_devices->open_devices++;
2538         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2539
2540         *device_out = device;
2541         return ret;
2542
2543 error:
2544         blkdev_put(bdev, FMODE_EXCL);
2545         return ret;
2546 }
2547
2548 void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info *fs_info,
2549                                               struct btrfs_device *tgtdev)
2550 {
2551         WARN_ON(fs_info->fs_devices->rw_devices == 0);
2552         tgtdev->io_width = fs_info->dev_root->sectorsize;
2553         tgtdev->io_align = fs_info->dev_root->sectorsize;
2554         tgtdev->sector_size = fs_info->dev_root->sectorsize;
2555         tgtdev->dev_root = fs_info->dev_root;
2556         tgtdev->in_fs_metadata = 1;
2557 }
2558
2559 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2560                                         struct btrfs_device *device)
2561 {
2562         int ret;
2563         struct btrfs_path *path;
2564         struct btrfs_root *root;
2565         struct btrfs_dev_item *dev_item;
2566         struct extent_buffer *leaf;
2567         struct btrfs_key key;
2568
2569         root = device->dev_root->fs_info->chunk_root;
2570
2571         path = btrfs_alloc_path();
2572         if (!path)
2573                 return -ENOMEM;
2574
2575         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2576         key.type = BTRFS_DEV_ITEM_KEY;
2577         key.offset = device->devid;
2578
2579         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2580         if (ret < 0)
2581                 goto out;
2582
2583         if (ret > 0) {
2584                 ret = -ENOENT;
2585                 goto out;
2586         }
2587
2588         leaf = path->nodes[0];
2589         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2590
2591         btrfs_set_device_id(leaf, dev_item, device->devid);
2592         btrfs_set_device_type(leaf, dev_item, device->type);
2593         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2594         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2595         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2596         btrfs_set_device_total_bytes(leaf, dev_item,
2597                                      btrfs_device_get_disk_total_bytes(device));
2598         btrfs_set_device_bytes_used(leaf, dev_item,
2599                                     btrfs_device_get_bytes_used(device));
2600         btrfs_mark_buffer_dirty(leaf);
2601
2602 out:
2603         btrfs_free_path(path);
2604         return ret;
2605 }
2606
2607 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2608                       struct btrfs_device *device, u64 new_size)
2609 {
2610         struct btrfs_super_block *super_copy =
2611                 device->dev_root->fs_info->super_copy;
2612         struct btrfs_fs_devices *fs_devices;
2613         u64 old_total;
2614         u64 diff;
2615
2616         if (!device->writeable)
2617                 return -EACCES;
2618
2619         lock_chunks(device->dev_root);
2620         old_total = btrfs_super_total_bytes(super_copy);
2621         diff = new_size - device->total_bytes;
2622
2623         if (new_size <= device->total_bytes ||
2624             device->is_tgtdev_for_dev_replace) {
2625                 unlock_chunks(device->dev_root);
2626                 return -EINVAL;
2627         }
2628
2629         fs_devices = device->dev_root->fs_info->fs_devices;
2630
2631         btrfs_set_super_total_bytes(super_copy, old_total + diff);
2632         device->fs_devices->total_rw_bytes += diff;
2633
2634         btrfs_device_set_total_bytes(device, new_size);
2635         btrfs_device_set_disk_total_bytes(device, new_size);
2636         btrfs_clear_space_info_full(device->dev_root->fs_info);
2637         if (list_empty(&device->resized_list))
2638                 list_add_tail(&device->resized_list,
2639                               &fs_devices->resized_devices);
2640         unlock_chunks(device->dev_root);
2641
2642         return btrfs_update_device(trans, device);
2643 }
2644
2645 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
2646                             struct btrfs_root *root, u64 chunk_objectid,
2647                             u64 chunk_offset)
2648 {
2649         int ret;
2650         struct btrfs_path *path;
2651         struct btrfs_key key;
2652
2653         root = root->fs_info->chunk_root;
2654         path = btrfs_alloc_path();
2655         if (!path)
2656                 return -ENOMEM;
2657
2658         key.objectid = chunk_objectid;
2659         key.offset = chunk_offset;
2660         key.type = BTRFS_CHUNK_ITEM_KEY;
2661
2662         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2663         if (ret < 0)
2664                 goto out;
2665         else if (ret > 0) { /* Logic error or corruption */
2666                 btrfs_std_error(root->fs_info, -ENOENT,
2667                             "Failed lookup while freeing chunk.");
2668                 ret = -ENOENT;
2669                 goto out;
2670         }
2671
2672         ret = btrfs_del_item(trans, root, path);
2673         if (ret < 0)
2674                 btrfs_std_error(root->fs_info, ret,
2675                             "Failed to delete chunk item.");
2676 out:
2677         btrfs_free_path(path);
2678         return ret;
2679 }
2680
2681 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
2682                         chunk_offset)
2683 {
2684         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
2685         struct btrfs_disk_key *disk_key;
2686         struct btrfs_chunk *chunk;
2687         u8 *ptr;
2688         int ret = 0;
2689         u32 num_stripes;
2690         u32 array_size;
2691         u32 len = 0;
2692         u32 cur;
2693         struct btrfs_key key;
2694
2695         lock_chunks(root);
2696         array_size = btrfs_super_sys_array_size(super_copy);
2697
2698         ptr = super_copy->sys_chunk_array;
2699         cur = 0;
2700
2701         while (cur < array_size) {
2702                 disk_key = (struct btrfs_disk_key *)ptr;
2703                 btrfs_disk_key_to_cpu(&key, disk_key);
2704
2705                 len = sizeof(*disk_key);
2706
2707                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2708                         chunk = (struct btrfs_chunk *)(ptr + len);
2709                         num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2710                         len += btrfs_chunk_item_size(num_stripes);
2711                 } else {
2712                         ret = -EIO;
2713                         break;
2714                 }
2715                 if (key.objectid == chunk_objectid &&
2716                     key.offset == chunk_offset) {
2717                         memmove(ptr, ptr + len, array_size - (cur + len));
2718                         array_size -= len;
2719                         btrfs_set_super_sys_array_size(super_copy, array_size);
2720                 } else {
2721                         ptr += len;
2722                         cur += len;
2723                 }
2724         }
2725         unlock_chunks(root);
2726         return ret;
2727 }
2728
2729 int btrfs_remove_chunk(struct btrfs_trans_handle *trans,
2730                        struct btrfs_root *root, u64 chunk_offset)
2731 {
2732         struct extent_map_tree *em_tree;
2733         struct extent_map *em;
2734         struct btrfs_root *extent_root = root->fs_info->extent_root;
2735         struct map_lookup *map;
2736         u64 dev_extent_len = 0;
2737         u64 chunk_objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2738         int i, ret = 0;
2739
2740         /* Just in case */
2741         root = root->fs_info->chunk_root;
2742         em_tree = &root->fs_info->mapping_tree.map_tree;
2743
2744         read_lock(&em_tree->lock);
2745         em = lookup_extent_mapping(em_tree, chunk_offset, 1);
2746         read_unlock(&em_tree->lock);
2747
2748         if (!em || em->start > chunk_offset ||
2749             em->start + em->len < chunk_offset) {
2750                 /*
2751                  * This is a logic error, but we don't want to just rely on the
2752                  * user having built with ASSERT enabled, so if ASSERT doens't
2753                  * do anything we still error out.
2754                  */
2755                 ASSERT(0);
2756                 if (em)
2757                         free_extent_map(em);
2758                 return -EINVAL;
2759         }
2760         map = (struct map_lookup *)em->bdev;
2761         lock_chunks(root->fs_info->chunk_root);
2762         check_system_chunk(trans, extent_root, map->type);
2763         unlock_chunks(root->fs_info->chunk_root);
2764
2765         for (i = 0; i < map->num_stripes; i++) {
2766                 struct btrfs_device *device = map->stripes[i].dev;
2767                 ret = btrfs_free_dev_extent(trans, device,
2768                                             map->stripes[i].physical,
2769                                             &dev_extent_len);
2770                 if (ret) {
2771                         btrfs_abort_transaction(trans, root, ret);
2772                         goto out;
2773                 }
2774
2775                 if (device->bytes_used > 0) {
2776                         lock_chunks(root);
2777                         btrfs_device_set_bytes_used(device,
2778                                         device->bytes_used - dev_extent_len);
2779                         spin_lock(&root->fs_info->free_chunk_lock);
2780                         root->fs_info->free_chunk_space += dev_extent_len;
2781                         spin_unlock(&root->fs_info->free_chunk_lock);
2782                         btrfs_clear_space_info_full(root->fs_info);
2783                         unlock_chunks(root);
2784                 }
2785
2786                 if (map->stripes[i].dev) {
2787                         ret = btrfs_update_device(trans, map->stripes[i].dev);
2788                         if (ret) {
2789                                 btrfs_abort_transaction(trans, root, ret);
2790                                 goto out;
2791                         }
2792                 }
2793         }
2794         ret = btrfs_free_chunk(trans, root, chunk_objectid, chunk_offset);
2795         if (ret) {
2796                 btrfs_abort_transaction(trans, root, ret);
2797                 goto out;
2798         }
2799
2800         trace_btrfs_chunk_free(root, map, chunk_offset, em->len);
2801
2802         if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2803                 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
2804                 if (ret) {
2805                         btrfs_abort_transaction(trans, root, ret);
2806                         goto out;
2807                 }
2808         }
2809
2810         ret = btrfs_remove_block_group(trans, extent_root, chunk_offset, em);
2811         if (ret) {
2812                 btrfs_abort_transaction(trans, extent_root, ret);
2813                 goto out;
2814         }
2815
2816 out:
2817         /* once for us */
2818         free_extent_map(em);
2819         return ret;
2820 }
2821
2822 static int btrfs_relocate_chunk(struct btrfs_root *root, u64 chunk_offset)
2823 {
2824         struct btrfs_root *extent_root;
2825         struct btrfs_trans_handle *trans;
2826         int ret;
2827
2828         root = root->fs_info->chunk_root;
2829         extent_root = root->fs_info->extent_root;
2830
2831         /*
2832          * Prevent races with automatic removal of unused block groups.
2833          * After we relocate and before we remove the chunk with offset
2834          * chunk_offset, automatic removal of the block group can kick in,
2835          * resulting in a failure when calling btrfs_remove_chunk() below.
2836          *
2837          * Make sure to acquire this mutex before doing a tree search (dev
2838          * or chunk trees) to find chunks. Otherwise the cleaner kthread might
2839          * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after
2840          * we release the path used to search the chunk/dev tree and before
2841          * the current task acquires this mutex and calls us.
2842          */
2843         ASSERT(mutex_is_locked(&root->fs_info->delete_unused_bgs_mutex));
2844
2845         ret = btrfs_can_relocate(extent_root, chunk_offset);
2846         if (ret)
2847                 return -ENOSPC;
2848
2849         /* step one, relocate all the extents inside this chunk */
2850         btrfs_scrub_pause(root);
2851         ret = btrfs_relocate_block_group(extent_root, chunk_offset);
2852         btrfs_scrub_continue(root);
2853         if (ret)
2854                 return ret;
2855
2856         trans = btrfs_start_trans_remove_block_group(root->fs_info,
2857                                                      chunk_offset);
2858         if (IS_ERR(trans)) {
2859                 ret = PTR_ERR(trans);
2860                 btrfs_std_error(root->fs_info, ret, NULL);
2861                 return ret;
2862         }
2863
2864         /*
2865          * step two, delete the device extents and the
2866          * chunk tree entries
2867          */
2868         ret = btrfs_remove_chunk(trans, root, chunk_offset);
2869         btrfs_end_transaction(trans, root);
2870         return ret;
2871 }
2872
2873 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
2874 {
2875         struct btrfs_root *chunk_root = root->fs_info->chunk_root;
2876         struct btrfs_path *path;
2877         struct extent_buffer *leaf;
2878         struct btrfs_chunk *chunk;
2879         struct btrfs_key key;
2880         struct btrfs_key found_key;
2881         u64 chunk_type;
2882         bool retried = false;
2883         int failed = 0;
2884         int ret;
2885
2886         path = btrfs_alloc_path();
2887         if (!path)
2888                 return -ENOMEM;
2889
2890 again:
2891         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2892         key.offset = (u64)-1;
2893         key.type = BTRFS_CHUNK_ITEM_KEY;
2894
2895         while (1) {
2896                 mutex_lock(&root->fs_info->delete_unused_bgs_mutex);
2897                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2898                 if (ret < 0) {
2899                         mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
2900                         goto error;
2901                 }
2902                 BUG_ON(ret == 0); /* Corruption */
2903
2904                 ret = btrfs_previous_item(chunk_root, path, key.objectid,
2905                                           key.type);
2906                 if (ret)
2907                         mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
2908                 if (ret < 0)
2909                         goto error;
2910                 if (ret > 0)
2911                         break;
2912
2913                 leaf = path->nodes[0];
2914                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2915
2916                 chunk = btrfs_item_ptr(leaf, path->slots[0],
2917                                        struct btrfs_chunk);
2918                 chunk_type = btrfs_chunk_type(leaf, chunk);
2919                 btrfs_release_path(path);
2920
2921                 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
2922                         ret = btrfs_relocate_chunk(chunk_root,
2923                                                    found_key.offset);
2924                         if (ret == -ENOSPC)
2925                                 failed++;
2926                         else
2927                                 BUG_ON(ret);
2928                 }
2929                 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
2930
2931                 if (found_key.offset == 0)
2932                         break;
2933                 key.offset = found_key.offset - 1;
2934         }
2935         ret = 0;
2936         if (failed && !retried) {
2937                 failed = 0;
2938                 retried = true;
2939                 goto again;
2940         } else if (WARN_ON(failed && retried)) {
2941                 ret = -ENOSPC;
2942         }
2943 error:
2944         btrfs_free_path(path);
2945         return ret;
2946 }
2947
2948 static int insert_balance_item(struct btrfs_root *root,
2949                                struct btrfs_balance_control *bctl)
2950 {
2951         struct btrfs_trans_handle *trans;
2952         struct btrfs_balance_item *item;
2953         struct btrfs_disk_balance_args disk_bargs;
2954         struct btrfs_path *path;
2955         struct extent_buffer *leaf;
2956         struct btrfs_key key;
2957         int ret, err;
2958
2959         path = btrfs_alloc_path();
2960         if (!path)
2961                 return -ENOMEM;
2962
2963         trans = btrfs_start_transaction(root, 0);
2964         if (IS_ERR(trans)) {
2965                 btrfs_free_path(path);
2966                 return PTR_ERR(trans);
2967         }
2968
2969         key.objectid = BTRFS_BALANCE_OBJECTID;
2970         key.type = BTRFS_BALANCE_ITEM_KEY;
2971         key.offset = 0;
2972
2973         ret = btrfs_insert_empty_item(trans, root, path, &key,
2974                                       sizeof(*item));
2975         if (ret)
2976                 goto out;
2977
2978         leaf = path->nodes[0];
2979         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
2980
2981         memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
2982
2983         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
2984         btrfs_set_balance_data(leaf, item, &disk_bargs);
2985         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
2986         btrfs_set_balance_meta(leaf, item, &disk_bargs);
2987         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
2988         btrfs_set_balance_sys(leaf, item, &disk_bargs);
2989
2990         btrfs_set_balance_flags(leaf, item, bctl->flags);
2991
2992         btrfs_mark_buffer_dirty(leaf);
2993 out:
2994         btrfs_free_path(path);
2995         err = btrfs_commit_transaction(trans, root);
2996         if (err && !ret)
2997                 ret = err;
2998         return ret;
2999 }
3000
3001 static int del_balance_item(struct btrfs_root *root)
3002 {
3003         struct btrfs_trans_handle *trans;
3004         struct btrfs_path *path;
3005         struct btrfs_key key;
3006         int ret, err;
3007
3008         path = btrfs_alloc_path();
3009         if (!path)
3010                 return -ENOMEM;
3011
3012         trans = btrfs_start_transaction(root, 0);
3013         if (IS_ERR(trans)) {
3014                 btrfs_free_path(path);
3015                 return PTR_ERR(trans);
3016         }
3017
3018         key.objectid = BTRFS_BALANCE_OBJECTID;
3019         key.type = BTRFS_BALANCE_ITEM_KEY;
3020         key.offset = 0;
3021
3022         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3023         if (ret < 0)
3024                 goto out;
3025         if (ret > 0) {
3026                 ret = -ENOENT;
3027                 goto out;
3028         }
3029
3030         ret = btrfs_del_item(trans, root, path);
3031 out:
3032         btrfs_free_path(path);
3033         err = btrfs_commit_transaction(trans, root);
3034         if (err && !ret)
3035                 ret = err;
3036         return ret;
3037 }
3038
3039 /*
3040  * This is a heuristic used to reduce the number of chunks balanced on
3041  * resume after balance was interrupted.
3042  */
3043 static void update_balance_args(struct btrfs_balance_control *bctl)
3044 {
3045         /*
3046          * Turn on soft mode for chunk types that were being converted.
3047          */
3048         if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
3049                 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
3050         if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
3051                 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
3052         if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
3053                 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
3054
3055         /*
3056          * Turn on usage filter if is not already used.  The idea is
3057          * that chunks that we have already balanced should be
3058          * reasonably full.  Don't do it for chunks that are being
3059          * converted - that will keep us from relocating unconverted
3060          * (albeit full) chunks.
3061          */
3062         if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3063             !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3064             !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3065                 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
3066                 bctl->data.usage = 90;
3067         }
3068         if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3069             !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3070             !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3071                 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
3072                 bctl->sys.usage = 90;
3073         }
3074         if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3075             !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3076             !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3077                 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
3078                 bctl->meta.usage = 90;
3079         }
3080 }
3081
3082 /*
3083  * Should be called with both balance and volume mutexes held to
3084  * serialize other volume operations (add_dev/rm_dev/resize) with
3085  * restriper.  Same goes for unset_balance_control.
3086  */
3087 static void set_balance_control(struct btrfs_balance_control *bctl)
3088 {
3089         struct btrfs_fs_info *fs_info = bctl->fs_info;
3090
3091         BUG_ON(fs_info->balance_ctl);
3092
3093         spin_lock(&fs_info->balance_lock);
3094         fs_info->balance_ctl = bctl;
3095         spin_unlock(&fs_info->balance_lock);
3096 }
3097
3098 static void unset_balance_control(struct btrfs_fs_info *fs_info)
3099 {
3100         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3101
3102         BUG_ON(!fs_info->balance_ctl);
3103
3104         spin_lock(&fs_info->balance_lock);
3105         fs_info->balance_ctl = NULL;
3106         spin_unlock(&fs_info->balance_lock);
3107
3108         kfree(bctl);
3109 }
3110
3111 /*
3112  * Balance filters.  Return 1 if chunk should be filtered out
3113  * (should not be balanced).
3114  */
3115 static int chunk_profiles_filter(u64 chunk_type,
3116                                  struct btrfs_balance_args *bargs)
3117 {
3118         chunk_type = chunk_to_extended(chunk_type) &
3119                                 BTRFS_EXTENDED_PROFILE_MASK;
3120
3121         if (bargs->profiles & chunk_type)
3122                 return 0;
3123
3124         return 1;
3125 }
3126
3127 static int chunk_usage_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
3128                               struct btrfs_balance_args *bargs)
3129 {
3130         struct btrfs_block_group_cache *cache;
3131         u64 chunk_used;
3132         u64 user_thresh_min;
3133         u64 user_thresh_max;
3134         int ret = 1;
3135
3136         cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3137         chunk_used = btrfs_block_group_used(&cache->item);
3138
3139         if (bargs->usage_min == 0)
3140                 user_thresh_min = 0;
3141         else
3142                 user_thresh_min = div_factor_fine(cache->key.offset,
3143                                         bargs->usage_min);
3144
3145         if (bargs->usage_max == 0)
3146                 user_thresh_max = 1;
3147         else if (bargs->usage_max > 100)
3148                 user_thresh_max = cache->key.offset;
3149         else
3150                 user_thresh_max = div_factor_fine(cache->key.offset,
3151                                         bargs->usage_max);
3152
3153         if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max)
3154                 ret = 0;
3155
3156         btrfs_put_block_group(cache);
3157         return ret;
3158 }
3159
3160 static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info,
3161                 u64 chunk_offset, struct btrfs_balance_args *bargs)
3162 {
3163         struct btrfs_block_group_cache *cache;
3164         u64 chunk_used, user_thresh;
3165         int ret = 1;
3166
3167         cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3168         chunk_used = btrfs_block_group_used(&cache->item);
3169
3170         if (bargs->usage_min == 0)
3171                 user_thresh = 1;
3172         else if (bargs->usage > 100)
3173                 user_thresh = cache->key.offset;
3174         else
3175                 user_thresh = div_factor_fine(cache->key.offset,
3176                                               bargs->usage);
3177
3178         if (chunk_used < user_thresh)
3179                 ret = 0;
3180
3181         btrfs_put_block_group(cache);
3182         return ret;
3183 }
3184
3185 static int chunk_devid_filter(struct extent_buffer *leaf,
3186                               struct btrfs_chunk *chunk,
3187                               struct btrfs_balance_args *bargs)
3188 {
3189         struct btrfs_stripe *stripe;
3190         int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3191         int i;
3192
3193         for (i = 0; i < num_stripes; i++) {
3194                 stripe = btrfs_stripe_nr(chunk, i);
3195                 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
3196                         return 0;
3197         }
3198
3199         return 1;
3200 }
3201
3202 /* [pstart, pend) */
3203 static int chunk_drange_filter(struct extent_buffer *leaf,
3204                                struct btrfs_chunk *chunk,
3205                                u64 chunk_offset,
3206                                struct btrfs_balance_args *bargs)
3207 {
3208         struct btrfs_stripe *stripe;
3209         int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3210         u64 stripe_offset;
3211         u64 stripe_length;
3212         int factor;
3213         int i;
3214
3215         if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
3216                 return 0;
3217
3218         if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP |
3219              BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) {
3220                 factor = num_stripes / 2;
3221         } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) {
3222                 factor = num_stripes - 1;
3223         } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) {
3224                 factor = num_stripes - 2;
3225         } else {
3226                 factor = num_stripes;
3227         }
3228
3229         for (i = 0; i < num_stripes; i++) {
3230                 stripe = btrfs_stripe_nr(chunk, i);
3231                 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
3232                         continue;
3233
3234                 stripe_offset = btrfs_stripe_offset(leaf, stripe);
3235                 stripe_length = btrfs_chunk_length(leaf, chunk);
3236                 stripe_length = div_u64(stripe_length, factor);
3237
3238                 if (stripe_offset < bargs->pend &&
3239                     stripe_offset + stripe_length > bargs->pstart)
3240                         return 0;
3241         }
3242
3243         return 1;
3244 }
3245
3246 /* [vstart, vend) */
3247 static int chunk_vrange_filter(struct extent_buffer *leaf,
3248                                struct btrfs_chunk *chunk,
3249                                u64 chunk_offset,
3250                                struct btrfs_balance_args *bargs)
3251 {
3252         if (chunk_offset < bargs->vend &&
3253             chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
3254                 /* at least part of the chunk is inside this vrange */
3255                 return 0;
3256
3257         return 1;
3258 }
3259
3260 static int chunk_stripes_range_filter(struct extent_buffer *leaf,
3261                                struct btrfs_chunk *chunk,
3262                                struct btrfs_balance_args *bargs)
3263 {
3264         int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3265
3266         if (bargs->stripes_min <= num_stripes
3267                         && num_stripes <= bargs->stripes_max)
3268                 return 0;
3269
3270         return 1;
3271 }
3272
3273 static int chunk_soft_convert_filter(u64 chunk_type,
3274                                      struct btrfs_balance_args *bargs)
3275 {
3276         if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
3277                 return 0;
3278
3279         chunk_type = chunk_to_extended(chunk_type) &
3280                                 BTRFS_EXTENDED_PROFILE_MASK;
3281
3282         if (bargs->target == chunk_type)
3283                 return 1;
3284
3285         return 0;
3286 }
3287
3288 static int should_balance_chunk(struct btrfs_root *root,
3289                                 struct extent_buffer *leaf,
3290                                 struct btrfs_chunk *chunk, u64 chunk_offset)
3291 {
3292         struct btrfs_balance_control *bctl = root->fs_info->balance_ctl;
3293         struct btrfs_balance_args *bargs = NULL;
3294         u64 chunk_type = btrfs_chunk_type(leaf, chunk);
3295
3296         /* type filter */
3297         if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
3298               (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
3299                 return 0;
3300         }
3301
3302         if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3303                 bargs = &bctl->data;
3304         else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3305                 bargs = &bctl->sys;
3306         else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3307                 bargs = &bctl->meta;
3308
3309         /* profiles filter */
3310         if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
3311             chunk_profiles_filter(chunk_type, bargs)) {
3312                 return 0;
3313         }
3314
3315         /* usage filter */
3316         if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
3317             chunk_usage_filter(bctl->fs_info, chunk_offset, bargs)) {
3318                 return 0;
3319         } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3320             chunk_usage_range_filter(bctl->fs_info, chunk_offset, bargs)) {
3321                 return 0;
3322         }
3323
3324         /* devid filter */
3325         if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
3326             chunk_devid_filter(leaf, chunk, bargs)) {
3327                 return 0;
3328         }
3329
3330         /* drange filter, makes sense only with devid filter */
3331         if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
3332             chunk_drange_filter(leaf, chunk, chunk_offset, bargs)) {
3333                 return 0;
3334         }
3335
3336         /* vrange filter */
3337         if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
3338             chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
3339                 return 0;
3340         }
3341
3342         /* stripes filter */
3343         if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) &&
3344             chunk_stripes_range_filter(leaf, chunk, bargs)) {
3345                 return 0;
3346         }
3347
3348         /* soft profile changing mode */
3349         if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
3350             chunk_soft_convert_filter(chunk_type, bargs)) {
3351                 return 0;
3352         }
3353
3354         /*
3355          * limited by count, must be the last filter
3356          */
3357         if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) {
3358                 if (bargs->limit == 0)
3359                         return 0;
3360                 else
3361                         bargs->limit--;
3362         } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) {
3363                 /*
3364                  * Same logic as the 'limit' filter; the minimum cannot be
3365                  * determined here because we do not have the global informatoin
3366                  * about the count of all chunks that satisfy the filters.
3367                  */
3368                 if (bargs->limit_max == 0)
3369                         return 0;
3370                 else
3371                         bargs->limit_max--;
3372         }
3373
3374         return 1;
3375 }
3376
3377 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
3378 {
3379         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3380         struct btrfs_root *chunk_root = fs_info->chunk_root;
3381         struct btrfs_root *dev_root = fs_info->dev_root;
3382         struct list_head *devices;
3383         struct btrfs_device *device;
3384         u64 old_size;
3385         u64 size_to_free;
3386         u64 chunk_type;
3387         struct btrfs_chunk *chunk;
3388         struct btrfs_path *path;
3389         struct btrfs_key key;
3390         struct btrfs_key found_key;
3391         struct btrfs_trans_handle *trans;
3392         struct extent_buffer *leaf;
3393         int slot;
3394         int ret;
3395         int enospc_errors = 0;
3396         bool counting = true;
3397         /* The single value limit and min/max limits use the same bytes in the */
3398         u64 limit_data = bctl->data.limit;
3399         u64 limit_meta = bctl->meta.limit;
3400         u64 limit_sys = bctl->sys.limit;
3401         u32 count_data = 0;
3402         u32 count_meta = 0;
3403         u32 count_sys = 0;
3404         int chunk_reserved = 0;
3405
3406         /* step one make some room on all the devices */
3407         devices = &fs_info->fs_devices->devices;
3408         list_for_each_entry(device, devices, dev_list) {
3409                 old_size = btrfs_device_get_total_bytes(device);
3410                 size_to_free = div_factor(old_size, 1);
3411                 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
3412                 if (!device->writeable ||
3413                     btrfs_device_get_total_bytes(device) -
3414                     btrfs_device_get_bytes_used(device) > size_to_free ||
3415                     device->is_tgtdev_for_dev_replace)
3416                         continue;
3417
3418                 ret = btrfs_shrink_device(device, old_size - size_to_free);
3419                 if (ret == -ENOSPC)
3420                         break;
3421                 BUG_ON(ret);
3422
3423                 trans = btrfs_start_transaction(dev_root, 0);
3424                 BUG_ON(IS_ERR(trans));
3425
3426                 ret = btrfs_grow_device(trans, device, old_size);
3427                 BUG_ON(ret);
3428
3429                 btrfs_end_transaction(trans, dev_root);
3430         }
3431
3432         /* step two, relocate all the chunks */
3433         path = btrfs_alloc_path();
3434         if (!path) {
3435                 ret = -ENOMEM;
3436                 goto error;
3437         }
3438
3439         /* zero out stat counters */
3440         spin_lock(&fs_info->balance_lock);
3441         memset(&bctl->stat, 0, sizeof(bctl->stat));
3442         spin_unlock(&fs_info->balance_lock);
3443 again:
3444         if (!counting) {
3445                 /*
3446                  * The single value limit and min/max limits use the same bytes
3447                  * in the
3448                  */
3449                 bctl->data.limit = limit_data;
3450                 bctl->meta.limit = limit_meta;
3451                 bctl->sys.limit = limit_sys;
3452         }
3453         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3454         key.offset = (u64)-1;
3455         key.type = BTRFS_CHUNK_ITEM_KEY;
3456
3457         while (1) {
3458                 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
3459                     atomic_read(&fs_info->balance_cancel_req)) {
3460                         ret = -ECANCELED;
3461                         goto error;
3462                 }
3463
3464                 mutex_lock(&fs_info->delete_unused_bgs_mutex);
3465                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3466                 if (ret < 0) {
3467                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3468                         goto error;
3469                 }
3470
3471                 /*
3472                  * this shouldn't happen, it means the last relocate
3473                  * failed
3474                  */
3475                 if (ret == 0)
3476                         BUG(); /* FIXME break ? */
3477
3478                 ret = btrfs_previous_item(chunk_root, path, 0,
3479                                           BTRFS_CHUNK_ITEM_KEY);
3480                 if (ret) {
3481                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3482                         ret = 0;
3483                         break;
3484                 }
3485
3486                 leaf = path->nodes[0];
3487                 slot = path->slots[0];
3488                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3489
3490                 if (found_key.objectid != key.objectid) {
3491                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3492                         break;
3493                 }
3494
3495                 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3496                 chunk_type = btrfs_chunk_type(leaf, chunk);
3497
3498                 if (!counting) {
3499                         spin_lock(&fs_info->balance_lock);
3500                         bctl->stat.considered++;
3501                         spin_unlock(&fs_info->balance_lock);
3502                 }
3503
3504                 ret = should_balance_chunk(chunk_root, leaf, chunk,
3505                                            found_key.offset);
3506
3507                 btrfs_release_path(path);
3508                 if (!ret) {
3509                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3510                         goto loop;
3511                 }
3512
3513                 if (counting) {
3514                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3515                         spin_lock(&fs_info->balance_lock);
3516                         bctl->stat.expected++;
3517                         spin_unlock(&fs_info->balance_lock);
3518
3519                         if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3520                                 count_data++;
3521                         else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3522                                 count_sys++;
3523                         else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3524                                 count_meta++;
3525
3526                         goto loop;
3527                 }
3528
3529                 /*
3530                  * Apply limit_min filter, no need to check if the LIMITS
3531                  * filter is used, limit_min is 0 by default
3532                  */
3533                 if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
3534                                         count_data < bctl->data.limit_min)
3535                                 || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) &&
3536                                         count_meta < bctl->meta.limit_min)
3537                                 || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) &&
3538                                         count_sys < bctl->sys.limit_min)) {
3539                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3540                         goto loop;
3541                 }
3542
3543                 if ((chunk_type & BTRFS_BLOCK_GROUP_DATA) && !chunk_reserved) {
3544                         trans = btrfs_start_transaction(chunk_root, 0);
3545                         if (IS_ERR(trans)) {
3546                                 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3547                                 ret = PTR_ERR(trans);
3548                                 goto error;
3549                         }
3550
3551                         ret = btrfs_force_chunk_alloc(trans, chunk_root,
3552                                                       BTRFS_BLOCK_GROUP_DATA);
3553                         if (ret < 0) {
3554                                 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3555                                 goto error;
3556                         }
3557
3558                         btrfs_end_transaction(trans, chunk_root);
3559                         chunk_reserved = 1;
3560                 }
3561
3562                 ret = btrfs_relocate_chunk(chunk_root,
3563                                            found_key.offset);
3564                 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3565                 if (ret && ret != -ENOSPC)
3566                         goto error;
3567                 if (ret == -ENOSPC) {
3568                         enospc_errors++;
3569                 } else {
3570                         spin_lock(&fs_info->balance_lock);
3571                         bctl->stat.completed++;
3572                         spin_unlock(&fs_info->balance_lock);
3573                 }
3574 loop:
3575                 if (found_key.offset == 0)
3576                         break;
3577                 key.offset = found_key.offset - 1;
3578         }
3579
3580         if (counting) {
3581                 btrfs_release_path(path);
3582                 counting = false;
3583                 goto again;
3584         }
3585 error:
3586         btrfs_free_path(path);
3587         if (enospc_errors) {
3588                 btrfs_info(fs_info, "%d enospc errors during balance",
3589                        enospc_errors);
3590                 if (!ret)
3591                         ret = -ENOSPC;
3592         }
3593
3594         return ret;
3595 }
3596
3597 /**
3598  * alloc_profile_is_valid - see if a given profile is valid and reduced
3599  * @flags: profile to validate
3600  * @extended: if true @flags is treated as an extended profile
3601  */
3602 static int alloc_profile_is_valid(u64 flags, int extended)
3603 {
3604         u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
3605                                BTRFS_BLOCK_GROUP_PROFILE_MASK);
3606
3607         flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
3608
3609         /* 1) check that all other bits are zeroed */
3610         if (flags & ~mask)
3611                 return 0;
3612
3613         /* 2) see if profile is reduced */
3614         if (flags == 0)
3615                 return !extended; /* "0" is valid for usual profiles */
3616
3617         /* true if exactly one bit set */
3618         return (flags & (flags - 1)) == 0;
3619 }
3620
3621 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
3622 {
3623         /* cancel requested || normal exit path */
3624         return atomic_read(&fs_info->balance_cancel_req) ||
3625                 (atomic_read(&fs_info->balance_pause_req) == 0 &&
3626                  atomic_read(&fs_info->balance_cancel_req) == 0);
3627 }
3628
3629 static void __cancel_balance(struct btrfs_fs_info *fs_info)
3630 {
3631         int ret;
3632
3633         unset_balance_control(fs_info);
3634         ret = del_balance_item(fs_info->tree_root);
3635         if (ret)
3636                 btrfs_std_error(fs_info, ret, NULL);
3637
3638         atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3639 }
3640
3641 /* Non-zero return value signifies invalidity */
3642 static inline int validate_convert_profile(struct btrfs_balance_args *bctl_arg,
3643                 u64 allowed)
3644 {
3645         return ((bctl_arg->flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3646                 (!alloc_profile_is_valid(bctl_arg->target, 1) ||
3647                  (bctl_arg->target & ~allowed)));
3648 }
3649
3650 /*
3651  * Should be called with both balance and volume mutexes held
3652  */
3653 int btrfs_balance(struct btrfs_balance_control *bctl,
3654                   struct btrfs_ioctl_balance_args *bargs)
3655 {
3656         struct btrfs_fs_info *fs_info = bctl->fs_info;
3657         u64 allowed;
3658         int mixed = 0;
3659         int ret;
3660         u64 num_devices;
3661         unsigned seq;
3662
3663         if (btrfs_fs_closing(fs_info) ||
3664             atomic_read(&fs_info->balance_pause_req) ||
3665             atomic_read(&fs_info->balance_cancel_req)) {
3666                 ret = -EINVAL;
3667                 goto out;
3668         }
3669
3670         allowed = btrfs_super_incompat_flags(fs_info->super_copy);
3671         if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
3672                 mixed = 1;
3673
3674         /*
3675          * In case of mixed groups both data and meta should be picked,
3676          * and identical options should be given for both of them.
3677          */
3678         allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
3679         if (mixed && (bctl->flags & allowed)) {
3680                 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
3681                     !(bctl->flags & BTRFS_BALANCE_METADATA) ||
3682                     memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
3683                         btrfs_err(fs_info, "with mixed groups data and "
3684                                    "metadata balance options must be the same");
3685                         ret = -EINVAL;
3686                         goto out;
3687                 }
3688         }
3689
3690         num_devices = fs_info->fs_devices->num_devices;
3691         btrfs_dev_replace_lock(&fs_info->dev_replace);
3692         if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
3693                 BUG_ON(num_devices < 1);
3694                 num_devices--;
3695         }
3696         btrfs_dev_replace_unlock(&fs_info->dev_replace);
3697         allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE;
3698         if (num_devices == 1)
3699                 allowed |= BTRFS_BLOCK_GROUP_DUP;
3700         else if (num_devices > 1)
3701                 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1);
3702         if (num_devices > 2)
3703                 allowed |= BTRFS_BLOCK_GROUP_RAID5;
3704         if (num_devices > 3)
3705                 allowed |= (BTRFS_BLOCK_GROUP_RAID10 |
3706                             BTRFS_BLOCK_GROUP_RAID6);
3707         if (validate_convert_profile(&bctl->data, allowed)) {
3708                 btrfs_err(fs_info, "unable to start balance with target "
3709                            "data profile %llu",
3710                        bctl->data.target);
3711                 ret = -EINVAL;
3712                 goto out;
3713         }
3714         if (validate_convert_profile(&bctl->meta, allowed)) {
3715                 btrfs_err(fs_info,
3716                            "unable to start balance with target metadata profile %llu",
3717                        bctl->meta.target);
3718                 ret = -EINVAL;
3719                 goto out;
3720         }
3721         if (validate_convert_profile(&bctl->sys, allowed)) {
3722                 btrfs_err(fs_info,
3723                            "unable to start balance with target system profile %llu",
3724                        bctl->sys.target);
3725                 ret = -EINVAL;
3726                 goto out;
3727         }
3728
3729         /* allow dup'ed data chunks only in mixed mode */
3730         if (!mixed && (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3731             (bctl->data.target & BTRFS_BLOCK_GROUP_DUP)) {
3732                 btrfs_err(fs_info, "dup for data is not allowed");
3733                 ret = -EINVAL;
3734                 goto out;
3735         }
3736
3737         /* allow to reduce meta or sys integrity only if force set */
3738         allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3739                         BTRFS_BLOCK_GROUP_RAID10 |
3740                         BTRFS_BLOCK_GROUP_RAID5 |
3741                         BTRFS_BLOCK_GROUP_RAID6;
3742         do {
3743                 seq = read_seqbegin(&fs_info->profiles_lock);
3744
3745                 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3746                      (fs_info->avail_system_alloc_bits & allowed) &&
3747                      !(bctl->sys.target & allowed)) ||
3748                     ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3749                      (fs_info->avail_metadata_alloc_bits & allowed) &&
3750                      !(bctl->meta.target & allowed))) {
3751                         if (bctl->flags & BTRFS_BALANCE_FORCE) {
3752                                 btrfs_info(fs_info, "force reducing metadata integrity");
3753                         } else {
3754                                 btrfs_err(fs_info, "balance will reduce metadata "
3755                                            "integrity, use force if you want this");
3756                                 ret = -EINVAL;
3757                                 goto out;
3758                         }
3759                 }
3760         } while (read_seqretry(&fs_info->profiles_lock, seq));
3761
3762         if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3763                 fs_info->num_tolerated_disk_barrier_failures = min(
3764                         btrfs_calc_num_tolerated_disk_barrier_failures(fs_info),
3765                         btrfs_get_num_tolerated_disk_barrier_failures(
3766                                 bctl->sys.target));
3767         }
3768
3769         ret = insert_balance_item(fs_info->tree_root, bctl);
3770         if (ret && ret != -EEXIST)
3771                 goto out;
3772
3773         if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
3774                 BUG_ON(ret == -EEXIST);
3775                 set_balance_control(bctl);
3776         } else {
3777                 BUG_ON(ret != -EEXIST);
3778                 spin_lock(&fs_info->balance_lock);
3779                 update_balance_args(bctl);
3780                 spin_unlock(&fs_info->balance_lock);
3781         }
3782
3783         atomic_inc(&fs_info->balance_running);
3784         mutex_unlock(&fs_info->balance_mutex);
3785
3786         ret = __btrfs_balance(fs_info);
3787
3788         mutex_lock(&fs_info->balance_mutex);
3789         atomic_dec(&fs_info->balance_running);
3790
3791         if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3792                 fs_info->num_tolerated_disk_barrier_failures =
3793                         btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3794         }
3795
3796         if (bargs) {
3797                 memset(bargs, 0, sizeof(*bargs));
3798                 update_ioctl_balance_args(fs_info, 0, bargs);
3799         }
3800
3801         if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
3802             balance_need_close(fs_info)) {
3803                 __cancel_balance(fs_info);
3804         }
3805
3806         wake_up(&fs_info->balance_wait_q);
3807
3808         return ret;
3809 out:
3810         if (bctl->flags & BTRFS_BALANCE_RESUME)
3811                 __cancel_balance(fs_info);
3812         else {
3813                 kfree(bctl);
3814                 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3815         }
3816         return ret;
3817 }
3818
3819 static int balance_kthread(void *data)
3820 {
3821         struct btrfs_fs_info *fs_info = data;
3822         int ret = 0;
3823
3824         mutex_lock(&fs_info->volume_mutex);
3825         mutex_lock(&fs_info->balance_mutex);
3826
3827         if (fs_info->balance_ctl) {
3828                 btrfs_info(fs_info, "continuing balance");
3829                 ret = btrfs_balance(fs_info->balance_ctl, NULL);
3830         }
3831
3832         mutex_unlock(&fs_info->balance_mutex);
3833         mutex_unlock(&fs_info->volume_mutex);
3834
3835         return ret;
3836 }
3837
3838 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
3839 {
3840         struct task_struct *tsk;
3841
3842         spin_lock(&fs_info->balance_lock);
3843         if (!fs_info->balance_ctl) {
3844                 spin_unlock(&fs_info->balance_lock);
3845                 return 0;
3846         }
3847         spin_unlock(&fs_info->balance_lock);
3848
3849         if (btrfs_test_opt(fs_info->tree_root, SKIP_BALANCE)) {
3850                 btrfs_info(fs_info, "force skipping balance");
3851                 return 0;
3852         }
3853
3854         tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
3855         return PTR_ERR_OR_ZERO(tsk);
3856 }
3857
3858 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
3859 {
3860         struct btrfs_balance_control *bctl;
3861         struct btrfs_balance_item *item;
3862         struct btrfs_disk_balance_args disk_bargs;
3863         struct btrfs_path *path;
3864         struct extent_buffer *leaf;
3865         struct btrfs_key key;
3866         int ret;
3867
3868         path = btrfs_alloc_path();
3869         if (!path)
3870                 return -ENOMEM;
3871
3872         key.objectid = BTRFS_BALANCE_OBJECTID;
3873         key.type = BTRFS_BALANCE_ITEM_KEY;
3874         key.offset = 0;
3875
3876         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
3877         if (ret < 0)
3878                 goto out;
3879         if (ret > 0) { /* ret = -ENOENT; */
3880                 ret = 0;
3881                 goto out;
3882         }
3883
3884         bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
3885         if (!bctl) {
3886                 ret = -ENOMEM;
3887                 goto out;
3888         }
3889
3890         leaf = path->nodes[0];
3891         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3892
3893         bctl->fs_info = fs_info;
3894         bctl->flags = btrfs_balance_flags(leaf, item);
3895         bctl->flags |= BTRFS_BALANCE_RESUME;
3896
3897         btrfs_balance_data(leaf, item, &disk_bargs);
3898         btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
3899         btrfs_balance_meta(leaf, item, &disk_bargs);
3900         btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
3901         btrfs_balance_sys(leaf, item, &disk_bargs);
3902         btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
3903
3904         WARN_ON(atomic_xchg(&fs_info->mutually_exclusive_operation_running, 1));
3905
3906         mutex_lock(&fs_info->volume_mutex);
3907         mutex_lock(&fs_info->balance_mutex);
3908
3909         set_balance_control(bctl);
3910
3911         mutex_unlock(&fs_info->balance_mutex);
3912         mutex_unlock(&fs_info->volume_mutex);
3913 out:
3914         btrfs_free_path(path);
3915         return ret;
3916 }
3917
3918 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
3919 {
3920         int ret = 0;
3921
3922         mutex_lock(&fs_info->balance_mutex);
3923         if (!fs_info->balance_ctl) {
3924                 mutex_unlock(&fs_info->balance_mutex);
3925                 return -ENOTCONN;
3926         }
3927
3928         if (atomic_read(&fs_info->balance_running)) {
3929                 atomic_inc(&fs_info->balance_pause_req);
3930                 mutex_unlock(&fs_info->balance_mutex);
3931
3932                 wait_event(fs_info->balance_wait_q,
3933                            atomic_read(&fs_info->balance_running) == 0);
3934
3935                 mutex_lock(&fs_info->balance_mutex);
3936                 /* we are good with balance_ctl ripped off from under us */
3937                 BUG_ON(atomic_read(&fs_info->balance_running));
3938                 atomic_dec(&fs_info->balance_pause_req);
3939         } else {
3940                 ret = -ENOTCONN;
3941         }
3942
3943         mutex_unlock(&fs_info->balance_mutex);
3944         return ret;
3945 }
3946
3947 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
3948 {
3949         if (fs_info->sb->s_flags & MS_RDONLY)
3950                 return -EROFS;
3951
3952         mutex_lock(&fs_info->balance_mutex);
3953         if (!fs_info->balance_ctl) {
3954                 mutex_unlock(&fs_info->balance_mutex);
3955                 return -ENOTCONN;
3956         }
3957
3958         atomic_inc(&fs_info->balance_cancel_req);
3959         /*
3960          * if we are running just wait and return, balance item is
3961          * deleted in btrfs_balance in this case
3962          */
3963         if (atomic_read(&fs_info->balance_running)) {
3964                 mutex_unlock(&fs_info->balance_mutex);
3965                 wait_event(fs_info->balance_wait_q,
3966                            atomic_read(&fs_info->balance_running) == 0);
3967                 mutex_lock(&fs_info->balance_mutex);
3968         } else {
3969                 /* __cancel_balance needs volume_mutex */
3970                 mutex_unlock(&fs_info->balance_mutex);
3971                 mutex_lock(&fs_info->volume_mutex);
3972                 mutex_lock(&fs_info->balance_mutex);
3973
3974                 if (fs_info->balance_ctl)
3975                         __cancel_balance(fs_info);
3976
3977                 mutex_unlock(&fs_info->volume_mutex);
3978         }
3979
3980         BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running));
3981         atomic_dec(&fs_info->balance_cancel_req);
3982         mutex_unlock(&fs_info->balance_mutex);
3983         return 0;
3984 }
3985
3986 static int btrfs_uuid_scan_kthread(void *data)
3987 {
3988         struct btrfs_fs_info *fs_info = data;
3989         struct btrfs_root *root = fs_info->tree_root;
3990         struct btrfs_key key;
3991         struct btrfs_key max_key;
3992         struct btrfs_path *path = NULL;
3993         int ret = 0;
3994         struct extent_buffer *eb;
3995         int slot;
3996         struct btrfs_root_item root_item;
3997         u32 item_size;
3998         struct btrfs_trans_handle *trans = NULL;
3999
4000         path = btrfs_alloc_path();
4001         if (!path) {
4002                 ret = -ENOMEM;
4003                 goto out;
4004         }
4005
4006         key.objectid = 0;
4007         key.type = BTRFS_ROOT_ITEM_KEY;
4008         key.offset = 0;
4009
4010         max_key.objectid = (u64)-1;
4011         max_key.type = BTRFS_ROOT_ITEM_KEY;
4012         max_key.offset = (u64)-1;
4013
4014         while (1) {
4015                 ret = btrfs_search_forward(root, &key, path, 0);
4016                 if (ret) {
4017                         if (ret > 0)
4018                                 ret = 0;
4019                         break;
4020                 }
4021
4022                 if (key.type != BTRFS_ROOT_ITEM_KEY ||
4023                     (key.objectid < BTRFS_FIRST_FREE_OBJECTID &&
4024                      key.objectid != BTRFS_FS_TREE_OBJECTID) ||
4025                     key.objectid > BTRFS_LAST_FREE_OBJECTID)
4026                         goto skip;
4027
4028                 eb = path->nodes[0];
4029                 slot = path->slots[0];
4030                 item_size = btrfs_item_size_nr(eb, slot);
4031                 if (item_size < sizeof(root_item))
4032                         goto skip;
4033
4034                 read_extent_buffer(eb, &root_item,
4035                                    btrfs_item_ptr_offset(eb, slot),
4036                                    (int)sizeof(root_item));
4037                 if (btrfs_root_refs(&root_item) == 0)
4038                         goto skip;
4039
4040                 if (!btrfs_is_empty_uuid(root_item.uuid) ||
4041                     !btrfs_is_empty_uuid(root_item.received_uuid)) {
4042                         if (trans)
4043                                 goto update_tree;
4044
4045                         btrfs_release_path(path);
4046                         /*
4047                          * 1 - subvol uuid item
4048                          * 1 - received_subvol uuid item
4049                          */
4050                         trans = btrfs_start_transaction(fs_info->uuid_root, 2);
4051                         if (IS_ERR(trans)) {
4052                                 ret = PTR_ERR(trans);
4053                                 break;
4054                         }
4055                         continue;
4056                 } else {
4057                         goto skip;
4058                 }
4059 update_tree:
4060                 if (!btrfs_is_empty_uuid(root_item.uuid)) {
4061                         ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
4062                                                   root_item.uuid,
4063                                                   BTRFS_UUID_KEY_SUBVOL,
4064                                                   key.objectid);
4065                         if (ret < 0) {
4066                                 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4067                                         ret);
4068                                 break;
4069                         }
4070                 }
4071
4072                 if (!btrfs_is_empty_uuid(root_item.received_uuid)) {
4073                         ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
4074                                                   root_item.received_uuid,
4075                                                  BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4076                                                   key.objectid);
4077                         if (ret < 0) {
4078                                 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4079                                         ret);
4080                                 break;
4081                         }
4082                 }
4083
4084 skip:
4085                 if (trans) {
4086                         ret = btrfs_end_transaction(trans, fs_info->uuid_root);
4087                         trans = NULL;
4088                         if (ret)
4089                                 break;
4090                 }
4091
4092                 btrfs_release_path(path);
4093                 if (key.offset < (u64)-1) {
4094                         key.offset++;
4095                 } else if (key.type < BTRFS_ROOT_ITEM_KEY) {
4096                         key.offset = 0;
4097                         key.type = BTRFS_ROOT_ITEM_KEY;
4098                 } else if (key.objectid < (u64)-1) {
4099                         key.offset = 0;
4100                         key.type = BTRFS_ROOT_ITEM_KEY;
4101                         key.objectid++;
4102                 } else {
4103                         break;
4104                 }
4105                 cond_resched();
4106         }
4107
4108 out:
4109         btrfs_free_path(path);
4110         if (trans && !IS_ERR(trans))
4111                 btrfs_end_transaction(trans, fs_info->uuid_root);
4112         if (ret)
4113                 btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret);
4114         else
4115                 fs_info->update_uuid_tree_gen = 1;
4116         up(&fs_info->uuid_tree_rescan_sem);
4117         return 0;
4118 }
4119
4120 /*
4121  * Callback for btrfs_uuid_tree_iterate().
4122  * returns:
4123  * 0    check succeeded, the entry is not outdated.
4124  * < 0  if an error occured.
4125  * > 0  if the check failed, which means the caller shall remove the entry.
4126  */
4127 static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info *fs_info,
4128                                        u8 *uuid, u8 type, u64 subid)
4129 {
4130         struct btrfs_key key;
4131         int ret = 0;
4132         struct btrfs_root *subvol_root;
4133
4134         if (type != BTRFS_UUID_KEY_SUBVOL &&
4135             type != BTRFS_UUID_KEY_RECEIVED_SUBVOL)
4136                 goto out;
4137
4138         key.objectid = subid;
4139         key.type = BTRFS_ROOT_ITEM_KEY;
4140         key.offset = (u64)-1;
4141         subvol_root = btrfs_read_fs_root_no_name(fs_info, &key);
4142         if (IS_ERR(subvol_root)) {
4143                 ret = PTR_ERR(subvol_root);
4144                 if (ret == -ENOENT)
4145                         ret = 1;
4146                 goto out;
4147         }
4148
4149         switch (type) {
4150         case BTRFS_UUID_KEY_SUBVOL:
4151                 if (memcmp(uuid, subvol_root->root_item.uuid, BTRFS_UUID_SIZE))
4152                         ret = 1;
4153                 break;
4154         case BTRFS_UUID_KEY_RECEIVED_SUBVOL:
4155                 if (memcmp(uuid, subvol_root->root_item.received_uuid,
4156                            BTRFS_UUID_SIZE))
4157                         ret = 1;
4158                 break;
4159         }
4160
4161 out:
4162         return ret;
4163 }
4164
4165 static int btrfs_uuid_rescan_kthread(void *data)
4166 {
4167         struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data;
4168         int ret;
4169
4170         /*
4171          * 1st step is to iterate through the existing UUID tree and
4172          * to delete all entries that contain outdated data.
4173          * 2nd step is to add all missing entries to the UUID tree.
4174          */
4175         ret = btrfs_uuid_tree_iterate(fs_info, btrfs_check_uuid_tree_entry);
4176         if (ret < 0) {
4177                 btrfs_warn(fs_info, "iterating uuid_tree failed %d", ret);
4178                 up(&fs_info->uuid_tree_rescan_sem);
4179                 return ret;
4180         }
4181         return btrfs_uuid_scan_kthread(data);
4182 }
4183
4184 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info)
4185 {
4186         struct btrfs_trans_handle *trans;
4187         struct btrfs_root *tree_root = fs_info->tree_root;
4188         struct btrfs_root *uuid_root;
4189         struct task_struct *task;
4190         int ret;
4191
4192         /*
4193          * 1 - root node
4194          * 1 - root item
4195          */
4196         trans = btrfs_start_transaction(tree_root, 2);
4197         if (IS_ERR(trans))
4198                 return PTR_ERR(trans);
4199
4200         uuid_root = btrfs_create_tree(trans, fs_info,
4201                                       BTRFS_UUID_TREE_OBJECTID);
4202         if (IS_ERR(uuid_root)) {
4203                 ret = PTR_ERR(uuid_root);
4204                 btrfs_abort_transaction(trans, tree_root, ret);
4205                 return ret;
4206         }
4207
4208         fs_info->uuid_root = uuid_root;
4209
4210         ret = btrfs_commit_transaction(trans, tree_root);
4211         if (ret)
4212                 return ret;
4213
4214         down(&fs_info->uuid_tree_rescan_sem);
4215         task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid");
4216         if (IS_ERR(task)) {
4217                 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4218                 btrfs_warn(fs_info, "failed to start uuid_scan task");
4219                 up(&fs_info->uuid_tree_rescan_sem);
4220                 return PTR_ERR(task);
4221         }
4222
4223         return 0;
4224 }
4225
4226 int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
4227 {
4228         struct task_struct *task;
4229
4230         down(&fs_info->uuid_tree_rescan_sem);
4231         task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
4232         if (IS_ERR(task)) {
4233                 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4234                 btrfs_warn(fs_info, "failed to start uuid_rescan task");
4235                 up(&fs_info->uuid_tree_rescan_sem);
4236                 return PTR_ERR(task);
4237         }
4238
4239         return 0;
4240 }
4241
4242 /*
4243  * shrinking a device means finding all of the device extents past
4244  * the new size, and then following the back refs to the chunks.
4245  * The chunk relocation code actually frees the device extent
4246  */
4247 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
4248 {
4249         struct btrfs_trans_handle *trans;
4250         struct btrfs_root *root = device->dev_root;
4251         struct btrfs_dev_extent *dev_extent = NULL;
4252         struct btrfs_path *path;
4253         u64 length;
4254         u64 chunk_offset;
4255         int ret;
4256         int slot;
4257         int failed = 0;
4258         bool retried = false;
4259         bool checked_pending_chunks = false;
4260         struct extent_buffer *l;
4261         struct btrfs_key key;
4262         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
4263         u64 old_total = btrfs_super_total_bytes(super_copy);
4264         u64 old_size = btrfs_device_get_total_bytes(device);
4265         u64 diff = old_size - new_size;
4266
4267         if (device->is_tgtdev_for_dev_replace)
4268                 return -EINVAL;
4269
4270         path = btrfs_alloc_path();
4271         if (!path)
4272                 return -ENOMEM;
4273
4274         path->reada = 2;
4275
4276         lock_chunks(root);
4277
4278         btrfs_device_set_total_bytes(device, new_size);
4279         if (device->writeable) {
4280                 device->fs_devices->total_rw_bytes -= diff;
4281                 spin_lock(&root->fs_info->free_chunk_lock);
4282                 root->fs_info->free_chunk_space -= diff;
4283                 spin_unlock(&root->fs_info->free_chunk_lock);
4284         }
4285         unlock_chunks(root);
4286
4287 again:
4288         key.objectid = device->devid;
4289         key.offset = (u64)-1;
4290         key.type = BTRFS_DEV_EXTENT_KEY;
4291
4292         do {
4293                 mutex_lock(&root->fs_info->delete_unused_bgs_mutex);
4294                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4295                 if (ret < 0) {
4296                         mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4297                         goto done;
4298                 }
4299
4300                 ret = btrfs_previous_item(root, path, 0, key.type);
4301                 if (ret)
4302                         mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4303                 if (ret < 0)
4304                         goto done;
4305                 if (ret) {
4306                         ret = 0;
4307                         btrfs_release_path(path);
4308                         break;
4309                 }
4310
4311                 l = path->nodes[0];
4312                 slot = path->slots[0];
4313                 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
4314
4315                 if (key.objectid != device->devid) {
4316                         mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4317                         btrfs_release_path(path);
4318                         break;
4319                 }
4320
4321                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
4322                 length = btrfs_dev_extent_length(l, dev_extent);
4323
4324                 if (key.offset + length <= new_size) {
4325                         mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4326                         btrfs_release_path(path);
4327                         break;
4328                 }
4329
4330                 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
4331                 btrfs_release_path(path);
4332
4333                 ret = btrfs_relocate_chunk(root, chunk_offset);
4334                 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4335                 if (ret && ret != -ENOSPC)
4336                         goto done;
4337                 if (ret == -ENOSPC)
4338                         failed++;
4339         } while (key.offset-- > 0);
4340
4341         if (failed && !retried) {
4342                 failed = 0;
4343                 retried = true;
4344                 goto again;
4345         } else if (failed && retried) {
4346                 ret = -ENOSPC;
4347                 goto done;
4348         }
4349
4350         /* Shrinking succeeded, else we would be at "done". */
4351         trans = btrfs_start_transaction(root, 0);
4352         if (IS_ERR(trans)) {
4353                 ret = PTR_ERR(trans);
4354                 goto done;
4355         }
4356
4357         lock_chunks(root);
4358
4359         /*
4360          * We checked in the above loop all device extents that were already in
4361          * the device tree. However before we have updated the device's
4362          * total_bytes to the new size, we might have had chunk allocations that
4363          * have not complete yet (new block groups attached to transaction
4364          * handles), and therefore their device extents were not yet in the
4365          * device tree and we missed them in the loop above. So if we have any
4366          * pending chunk using a device extent that overlaps the device range
4367          * that we can not use anymore, commit the current transaction and
4368          * repeat the search on the device tree - this way we guarantee we will
4369          * not have chunks using device extents that end beyond 'new_size'.
4370          */
4371         if (!checked_pending_chunks) {
4372                 u64 start = new_size;
4373                 u64 len = old_size - new_size;
4374
4375                 if (contains_pending_extent(trans->transaction, device,
4376                                             &start, len)) {
4377                         unlock_chunks(root);
4378                         checked_pending_chunks = true;
4379                         failed = 0;
4380                         retried = false;
4381                         ret = btrfs_commit_transaction(trans, root);
4382                         if (ret)
4383                                 goto done;
4384                         goto again;
4385                 }
4386         }
4387
4388         btrfs_device_set_disk_total_bytes(device, new_size);
4389         if (list_empty(&device->resized_list))
4390                 list_add_tail(&device->resized_list,
4391                               &root->fs_info->fs_devices->resized_devices);
4392
4393         WARN_ON(diff > old_total);
4394         btrfs_set_super_total_bytes(super_copy, old_total - diff);
4395         unlock_chunks(root);
4396
4397         /* Now btrfs_update_device() will change the on-disk size. */
4398         ret = btrfs_update_device(trans, device);
4399         btrfs_end_transaction(trans, root);
4400 done:
4401         btrfs_free_path(path);
4402         if (ret) {
4403                 lock_chunks(root);
4404                 btrfs_device_set_total_bytes(device, old_size);
4405                 if (device->writeable)
4406                         device->fs_devices->total_rw_bytes += diff;
4407                 spin_lock(&root->fs_info->free_chunk_lock);
4408                 root->fs_info->free_chunk_space += diff;
4409                 spin_unlock(&root->fs_info->free_chunk_lock);
4410                 unlock_chunks(root);
4411         }
4412         return ret;
4413 }
4414
4415 static int btrfs_add_system_chunk(struct btrfs_root *root,
4416                            struct btrfs_key *key,
4417                            struct btrfs_chunk *chunk, int item_size)
4418 {
4419         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
4420         struct btrfs_disk_key disk_key;
4421         u32 array_size;
4422         u8 *ptr;
4423
4424         lock_chunks(root);
4425         array_size = btrfs_super_sys_array_size(super_copy);
4426         if (array_size + item_size + sizeof(disk_key)
4427                         > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
4428                 unlock_chunks(root);
4429                 return -EFBIG;
4430         }
4431
4432         ptr = super_copy->sys_chunk_array + array_size;
4433         btrfs_cpu_key_to_disk(&disk_key, key);
4434         memcpy(ptr, &disk_key, sizeof(disk_key));
4435         ptr += sizeof(disk_key);
4436         memcpy(ptr, chunk, item_size);
4437         item_size += sizeof(disk_key);
4438         btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
4439         unlock_chunks(root);
4440
4441         return 0;
4442 }
4443
4444 /*
4445  * sort the devices in descending order by max_avail, total_avail
4446  */
4447 static int btrfs_cmp_device_info(const void *a, const void *b)
4448 {
4449         const struct btrfs_device_info *di_a = a;
4450         const struct btrfs_device_info *di_b = b;
4451
4452         if (di_a->max_avail > di_b->max_avail)
4453                 return -1;
4454         if (di_a->max_avail < di_b->max_avail)
4455                 return 1;
4456         if (di_a->total_avail > di_b->total_avail)
4457                 return -1;
4458         if (di_a->total_avail < di_b->total_avail)
4459                 return 1;
4460         return 0;
4461 }
4462
4463 static u32 find_raid56_stripe_len(u32 data_devices, u32 dev_stripe_target)
4464 {
4465         /* TODO allow them to set a preferred stripe size */
4466         return 64 * 1024;
4467 }
4468
4469 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
4470 {
4471         if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK))
4472                 return;
4473
4474         btrfs_set_fs_incompat(info, RAID56);
4475 }
4476
4477 #define BTRFS_MAX_DEVS(r) ((BTRFS_LEAF_DATA_SIZE(r)             \
4478                         - sizeof(struct btrfs_item)             \
4479                         - sizeof(struct btrfs_chunk))           \
4480                         / sizeof(struct btrfs_stripe) + 1)
4481
4482 #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE        \
4483                                 - 2 * sizeof(struct btrfs_disk_key)     \
4484                                 - 2 * sizeof(struct btrfs_chunk))       \
4485                                 / sizeof(struct btrfs_stripe) + 1)
4486
4487 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4488                                struct btrfs_root *extent_root, u64 start,
4489                                u64 type)
4490 {
4491         struct btrfs_fs_info *info = extent_root->fs_info;
4492         struct btrfs_fs_devices *fs_devices = info->fs_devices;
4493         struct list_head *cur;
4494         struct map_lookup *map = NULL;
4495         struct extent_map_tree *em_tree;
4496         struct extent_map *em;
4497         struct btrfs_device_info *devices_info = NULL;
4498         u64 total_avail;
4499         int num_stripes;        /* total number of stripes to allocate */
4500         int data_stripes;       /* number of stripes that count for
4501                                    block group size */
4502         int sub_stripes;        /* sub_stripes info for map */
4503         int dev_stripes;        /* stripes per dev */
4504         int devs_max;           /* max devs to use */
4505         int devs_min;           /* min devs needed */
4506         int devs_increment;     /* ndevs has to be a multiple of this */
4507         int ncopies;            /* how many copies to data has */
4508         int ret;
4509         u64 max_stripe_size;
4510         u64 max_chunk_size;
4511         u64 stripe_size;
4512         u64 num_bytes;
4513         u64 raid_stripe_len = BTRFS_STRIPE_LEN;
4514         int ndevs;
4515         int i;
4516         int j;
4517         int index;
4518
4519         BUG_ON(!alloc_profile_is_valid(type, 0));
4520
4521         if (list_empty(&fs_devices->alloc_list))
4522                 return -ENOSPC;
4523
4524         index = __get_raid_index(type);
4525
4526         sub_stripes = btrfs_raid_array[index].sub_stripes;
4527         dev_stripes = btrfs_raid_array[index].dev_stripes;
4528         devs_max = btrfs_raid_array[index].devs_max;
4529         devs_min = btrfs_raid_array[index].devs_min;
4530         devs_increment = btrfs_raid_array[index].devs_increment;
4531         ncopies = btrfs_raid_array[index].ncopies;
4532
4533         if (type & BTRFS_BLOCK_GROUP_DATA) {
4534                 max_stripe_size = 1024 * 1024 * 1024;
4535                 max_chunk_size = 10 * max_stripe_size;
4536                 if (!devs_max)
4537                         devs_max = BTRFS_MAX_DEVS(info->chunk_root);
4538         } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
4539                 /* for larger filesystems, use larger metadata chunks */
4540                 if (fs_devices->total_rw_bytes > 50ULL * 1024 * 1024 * 1024)
4541                         max_stripe_size = 1024 * 1024 * 1024;
4542                 else
4543                         max_stripe_size = 256 * 1024 * 1024;
4544                 max_chunk_size = max_stripe_size;
4545                 if (!devs_max)
4546                         devs_max = BTRFS_MAX_DEVS(info->chunk_root);
4547         } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
4548                 max_stripe_size = 32 * 1024 * 1024;
4549                 max_chunk_size = 2 * max_stripe_size;
4550                 if (!devs_max)
4551                         devs_max = BTRFS_MAX_DEVS_SYS_CHUNK;
4552         } else {
4553                 btrfs_err(info, "invalid chunk type 0x%llx requested",
4554                        type);
4555                 BUG_ON(1);
4556         }
4557
4558         /* we don't want a chunk larger than 10% of writeable space */
4559         max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
4560                              max_chunk_size);
4561
4562         devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info),
4563                                GFP_NOFS);
4564         if (!devices_info)
4565                 return -ENOMEM;
4566
4567         cur = fs_devices->alloc_list.next;
4568
4569         /*
4570          * in the first pass through the devices list, we gather information
4571          * about the available holes on each device.
4572          */
4573         ndevs = 0;
4574         while (cur != &fs_devices->alloc_list) {
4575                 struct btrfs_device *device;
4576                 u64 max_avail;
4577                 u64 dev_offset;
4578
4579                 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
4580
4581                 cur = cur->next;
4582
4583                 if (!device->writeable) {
4584                         WARN(1, KERN_ERR
4585                                "BTRFS: read-only device in alloc_list\n");
4586                         continue;
4587                 }
4588
4589                 if (!device->in_fs_metadata ||
4590                     device->is_tgtdev_for_dev_replace)
4591                         continue;
4592
4593                 if (device->total_bytes > device->bytes_used)
4594                         total_avail = device->total_bytes - device->bytes_used;
4595                 else
4596                         total_avail = 0;
4597
4598                 /* If there is no space on this device, skip it. */
4599                 if (total_avail == 0)
4600                         continue;
4601
4602                 ret = find_free_dev_extent(trans, device,
4603                                            max_stripe_size * dev_stripes,
4604                                            &dev_offset, &max_avail);
4605                 if (ret && ret != -ENOSPC)
4606                         goto error;
4607
4608                 if (ret == 0)
4609                         max_avail = max_stripe_size * dev_stripes;
4610
4611                 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes)
4612                         continue;
4613
4614                 if (ndevs == fs_devices->rw_devices) {
4615                         WARN(1, "%s: found more than %llu devices\n",
4616                              __func__, fs_devices->rw_devices);
4617                         break;
4618                 }
4619                 devices_info[ndevs].dev_offset = dev_offset;
4620                 devices_info[ndevs].max_avail = max_avail;
4621                 devices_info[ndevs].total_avail = total_avail;
4622                 devices_info[ndevs].dev = device;
4623                 ++ndevs;
4624         }
4625
4626         /*
4627          * now sort the devices by hole size / available space
4628          */
4629         sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
4630              btrfs_cmp_device_info, NULL);
4631
4632         /* round down to number of usable stripes */
4633         ndevs -= ndevs % devs_increment;
4634
4635         if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) {
4636                 ret = -ENOSPC;
4637                 goto error;
4638         }
4639
4640         if (devs_max && ndevs > devs_max)
4641                 ndevs = devs_max;
4642         /*
4643          * the primary goal is to maximize the number of stripes, so use as many
4644          * devices as possible, even if the stripes are not maximum sized.
4645          */
4646         stripe_size = devices_info[ndevs-1].max_avail;
4647         num_stripes = ndevs * dev_stripes;
4648
4649         /*
4650          * this will have to be fixed for RAID1 and RAID10 over
4651          * more drives
4652          */
4653         data_stripes = num_stripes / ncopies;
4654
4655         if (type & BTRFS_BLOCK_GROUP_RAID5) {
4656                 raid_stripe_len = find_raid56_stripe_len(ndevs - 1,
4657                                  btrfs_super_stripesize(info->super_copy));
4658                 data_stripes = num_stripes - 1;
4659         }
4660         if (type & BTRFS_BLOCK_GROUP_RAID6) {
4661                 raid_stripe_len = find_raid56_stripe_len(ndevs - 2,
4662                                  btrfs_super_stripesize(info->super_copy));
4663                 data_stripes = num_stripes - 2;
4664         }
4665
4666         /*
4667          * Use the number of data stripes to figure out how big this chunk
4668          * is really going to be in terms of logical address space,
4669          * and compare that answer with the max chunk size
4670          */
4671         if (stripe_size * data_stripes > max_chunk_size) {
4672                 u64 mask = (1ULL << 24) - 1;
4673
4674                 stripe_size = div_u64(max_chunk_size, data_stripes);
4675
4676                 /* bump the answer up to a 16MB boundary */
4677                 stripe_size = (stripe_size + mask) & ~mask;
4678
4679                 /* but don't go higher than the limits we found
4680                  * while searching for free extents
4681                  */
4682                 if (stripe_size > devices_info[ndevs-1].max_avail)
4683                         stripe_size = devices_info[ndevs-1].max_avail;
4684         }
4685
4686         stripe_size = div_u64(stripe_size, dev_stripes);
4687
4688         /* align to BTRFS_STRIPE_LEN */
4689         stripe_size = div_u64(stripe_size, raid_stripe_len);
4690         stripe_size *= raid_stripe_len;
4691
4692         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
4693         if (!map) {
4694                 ret = -ENOMEM;
4695                 goto error;
4696         }
4697         map->num_stripes = num_stripes;
4698
4699         for (i = 0; i < ndevs; ++i) {
4700                 for (j = 0; j < dev_stripes; ++j) {
4701                         int s = i * dev_stripes + j;
4702                         map->stripes[s].dev = devices_info[i].dev;
4703                         map->stripes[s].physical = devices_info[i].dev_offset +
4704                                                    j * stripe_size;
4705                 }
4706         }
4707         map->sector_size = extent_root->sectorsize;
4708         map->stripe_len = raid_stripe_len;
4709         map->io_align = raid_stripe_len;
4710         map->io_width = raid_stripe_len;
4711         map->type = type;
4712         map->sub_stripes = sub_stripes;
4713
4714         num_bytes = stripe_size * data_stripes;
4715
4716         trace_btrfs_chunk_alloc(info->chunk_root, map, start, num_bytes);
4717
4718         em = alloc_extent_map();
4719         if (!em) {
4720                 kfree(map);
4721                 ret = -ENOMEM;
4722                 goto error;
4723         }
4724         set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
4725         em->bdev = (struct block_device *)map;
4726         em->start = start;
4727         em->len = num_bytes;
4728         em->block_start = 0;
4729         em->block_len = em->len;
4730         em->orig_block_len = stripe_size;
4731
4732         em_tree = &extent_root->fs_info->mapping_tree.map_tree;
4733         write_lock(&em_tree->lock);
4734         ret = add_extent_mapping(em_tree, em, 0);
4735         if (!ret) {
4736                 list_add_tail(&em->list, &trans->transaction->pending_chunks);
4737                 atomic_inc(&em->refs);
4738         }
4739         write_unlock(&em_tree->lock);
4740         if (ret) {
4741                 free_extent_map(em);
4742                 goto error;
4743         }
4744
4745         ret = btrfs_make_block_group(trans, extent_root, 0, type,
4746                                      BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4747                                      start, num_bytes);
4748         if (ret)
4749                 goto error_del_extent;
4750
4751         for (i = 0; i < map->num_stripes; i++) {
4752                 num_bytes = map->stripes[i].dev->bytes_used + stripe_size;
4753                 btrfs_device_set_bytes_used(map->stripes[i].dev, num_bytes);
4754         }
4755
4756         spin_lock(&extent_root->fs_info->free_chunk_lock);
4757         extent_root->fs_info->free_chunk_space -= (stripe_size *
4758                                                    map->num_stripes);
4759         spin_unlock(&extent_root->fs_info->free_chunk_lock);
4760
4761         free_extent_map(em);
4762         check_raid56_incompat_flag(extent_root->fs_info, type);
4763
4764         kfree(devices_info);
4765         return 0;
4766
4767 error_del_extent:
4768         write_lock(&em_tree->lock);
4769         remove_extent_mapping(em_tree, em);
4770         write_unlock(&em_tree->lock);
4771
4772         /* One for our allocation */
4773         free_extent_map(em);
4774         /* One for the tree reference */
4775         free_extent_map(em);
4776         /* One for the pending_chunks list reference */
4777         free_extent_map(em);
4778 error:
4779         kfree(devices_info);
4780         return ret;
4781 }
4782
4783 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans,
4784                                 struct btrfs_root *extent_root,
4785                                 u64 chunk_offset, u64 chunk_size)
4786 {
4787         struct btrfs_key key;
4788         struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
4789         struct btrfs_device *device;
4790         struct btrfs_chunk *chunk;
4791         struct btrfs_stripe *stripe;
4792         struct extent_map_tree *em_tree;
4793         struct extent_map *em;
4794         struct map_lookup *map;
4795         size_t item_size;
4796         u64 dev_offset;
4797         u64 stripe_size;
4798         int i = 0;
4799         int ret;
4800
4801         em_tree = &extent_root->fs_info->mapping_tree.map_tree;
4802         read_lock(&em_tree->lock);
4803         em = lookup_extent_mapping(em_tree, chunk_offset, chunk_size);
4804         read_unlock(&em_tree->lock);
4805
4806         if (!em) {
4807                 btrfs_crit(extent_root->fs_info, "unable to find logical "
4808                            "%Lu len %Lu", chunk_offset, chunk_size);
4809                 return -EINVAL;
4810         }
4811
4812         if (em->start != chunk_offset || em->len != chunk_size) {
4813                 btrfs_crit(extent_root->fs_info, "found a bad mapping, wanted"
4814                           " %Lu-%Lu, found %Lu-%Lu", chunk_offset,
4815                           chunk_size, em->start, em->len);
4816                 free_extent_map(em);
4817                 return -EINVAL;
4818         }
4819
4820         map = (struct map_lookup *)em->bdev;
4821         item_size = btrfs_chunk_item_size(map->num_stripes);
4822         stripe_size = em->orig_block_len;
4823
4824         chunk = kzalloc(item_size, GFP_NOFS);
4825         if (!chunk) {
4826                 ret = -ENOMEM;
4827                 goto out;
4828         }
4829
4830         for (i = 0; i < map->num_stripes; i++) {
4831                 device = map->stripes[i].dev;
4832                 dev_offset = map->stripes[i].physical;
4833
4834                 ret = btrfs_update_device(trans, device);
4835                 if (ret)
4836                         goto out;
4837                 ret = btrfs_alloc_dev_extent(trans, device,
4838                                              chunk_root->root_key.objectid,
4839                                              BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4840                                              chunk_offset, dev_offset,
4841                                              stripe_size);
4842                 if (ret)
4843                         goto out;
4844         }
4845
4846         stripe = &chunk->stripe;
4847         for (i = 0; i < map->num_stripes; i++) {
4848                 device = map->stripes[i].dev;
4849                 dev_offset = map->stripes[i].physical;
4850
4851                 btrfs_set_stack_stripe_devid(stripe, device->devid);
4852                 btrfs_set_stack_stripe_offset(stripe, dev_offset);
4853                 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
4854                 stripe++;
4855         }
4856
4857         btrfs_set_stack_chunk_length(chunk, chunk_size);
4858         btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
4859         btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
4860         btrfs_set_stack_chunk_type(chunk, map->type);
4861         btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
4862         btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
4863         btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
4864         btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
4865         btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
4866
4867         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
4868         key.type = BTRFS_CHUNK_ITEM_KEY;
4869         key.offset = chunk_offset;
4870
4871         ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
4872         if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
4873                 /*
4874                  * TODO: Cleanup of inserted chunk root in case of
4875                  * failure.
4876                  */
4877                 ret = btrfs_add_system_chunk(chunk_root, &key, chunk,
4878                                              item_size);
4879         }
4880
4881 out:
4882         kfree(chunk);
4883         free_extent_map(em);
4884         return ret;
4885 }
4886
4887 /*
4888  * Chunk allocation falls into two parts. The first part does works
4889  * that make the new allocated chunk useable, but not do any operation
4890  * that modifies the chunk tree. The second part does the works that
4891  * require modifying the chunk tree. This division is important for the
4892  * bootstrap process of adding storage to a seed btrfs.
4893  */
4894 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4895                       struct btrfs_root *extent_root, u64 type)
4896 {
4897         u64 chunk_offset;
4898
4899         ASSERT(mutex_is_locked(&extent_root->fs_info->chunk_mutex));
4900         chunk_offset = find_next_chunk(extent_root->fs_info);
4901         return __btrfs_alloc_chunk(trans, extent_root, chunk_offset, type);
4902 }
4903
4904 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
4905                                          struct btrfs_root *root,
4906                                          struct btrfs_device *device)
4907 {
4908         u64 chunk_offset;
4909         u64 sys_chunk_offset;
4910         u64 alloc_profile;
4911         struct btrfs_fs_info *fs_info = root->fs_info;
4912         struct btrfs_root *extent_root = fs_info->extent_root;
4913         int ret;
4914
4915         chunk_offset = find_next_chunk(fs_info);
4916         alloc_profile = btrfs_get_alloc_profile(extent_root, 0);
4917         ret = __btrfs_alloc_chunk(trans, extent_root, chunk_offset,
4918                                   alloc_profile);
4919         if (ret)
4920                 return ret;
4921
4922         sys_chunk_offset = find_next_chunk(root->fs_info);
4923         alloc_profile = btrfs_get_alloc_profile(fs_info->chunk_root, 0);
4924         ret = __btrfs_alloc_chunk(trans, extent_root, sys_chunk_offset,
4925                                   alloc_profile);
4926         return ret;
4927 }
4928
4929 static inline int btrfs_chunk_max_errors(struct map_lookup *map)
4930 {
4931         int max_errors;
4932
4933         if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
4934                          BTRFS_BLOCK_GROUP_RAID10 |
4935                          BTRFS_BLOCK_GROUP_RAID5 |
4936                          BTRFS_BLOCK_GROUP_DUP)) {
4937                 max_errors = 1;
4938         } else if (map->type & BTRFS_BLOCK_GROUP_RAID6) {
4939                 max_errors = 2;
4940         } else {
4941                 max_errors = 0;
4942         }
4943
4944         return max_errors;
4945 }
4946
4947 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
4948 {
4949         struct extent_map *em;
4950         struct map_lookup *map;
4951         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
4952         int readonly = 0;
4953         int miss_ndevs = 0;
4954         int i;
4955
4956         read_lock(&map_tree->map_tree.lock);
4957         em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
4958         read_unlock(&map_tree->map_tree.lock);
4959         if (!em)
4960                 return 1;
4961
4962         map = (struct map_lookup *)em->bdev;
4963         for (i = 0; i < map->num_stripes; i++) {
4964                 if (map->stripes[i].dev->missing) {
4965                         miss_ndevs++;
4966                         continue;
4967                 }
4968
4969                 if (!map->stripes[i].dev->writeable) {
4970                         readonly = 1;
4971                         goto end;
4972                 }
4973         }
4974
4975         /*
4976          * If the number of missing devices is larger than max errors,
4977          * we can not write the data into that chunk successfully, so
4978          * set it readonly.
4979          */
4980         if (miss_ndevs > btrfs_chunk_max_errors(map))
4981                 readonly = 1;
4982 end:
4983         free_extent_map(em);
4984         return readonly;
4985 }
4986
4987 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
4988 {
4989         extent_map_tree_init(&tree->map_tree);
4990 }
4991
4992 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
4993 {
4994         struct extent_map *em;
4995
4996         while (1) {
4997                 write_lock(&tree->map_tree.lock);
4998                 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
4999                 if (em)
5000                         remove_extent_mapping(&tree->map_tree, em);
5001                 write_unlock(&tree->map_tree.lock);
5002                 if (!em)
5003                         break;
5004                 /* once for us */
5005                 free_extent_map(em);
5006                 /* once for the tree */
5007                 free_extent_map(em);
5008         }
5009 }
5010
5011 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5012 {
5013         struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
5014         struct extent_map *em;
5015         struct map_lookup *map;
5016         struct extent_map_tree *em_tree = &map_tree->map_tree;
5017         int ret;
5018
5019         read_lock(&em_tree->lock);
5020         em = lookup_extent_mapping(em_tree, logical, len);
5021         read_unlock(&em_tree->lock);
5022
5023         /*
5024          * We could return errors for these cases, but that could get ugly and
5025          * we'd probably do the same thing which is just not do anything else
5026          * and exit, so return 1 so the callers don't try to use other copies.
5027          */
5028         if (!em) {
5029                 btrfs_crit(fs_info, "No mapping for %Lu-%Lu", logical,
5030                             logical+len);
5031                 return 1;
5032         }
5033
5034         if (em->start > logical || em->start + em->len < logical) {
5035                 btrfs_crit(fs_info, "Invalid mapping for %Lu-%Lu, got "
5036                             "%Lu-%Lu", logical, logical+len, em->start,
5037                             em->start + em->len);
5038                 free_extent_map(em);
5039                 return 1;
5040         }
5041
5042         map = (struct map_lookup *)em->bdev;
5043         if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
5044                 ret = map->num_stripes;
5045         else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5046                 ret = map->sub_stripes;
5047         else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
5048                 ret = 2;
5049         else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5050                 ret = 3;
5051         else
5052                 ret = 1;
5053         free_extent_map(em);
5054
5055         btrfs_dev_replace_lock(&fs_info->dev_replace);
5056         if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace))
5057                 ret++;
5058         btrfs_dev_replace_unlock(&fs_info->dev_replace);
5059
5060         return ret;
5061 }
5062
5063 unsigned long btrfs_full_stripe_len(struct btrfs_root *root,
5064                                     struct btrfs_mapping_tree *map_tree,
5065                                     u64 logical)
5066 {
5067         struct extent_map *em;
5068         struct map_lookup *map;
5069         struct extent_map_tree *em_tree = &map_tree->map_tree;
5070         unsigned long len = root->sectorsize;
5071
5072         read_lock(&em_tree->lock);
5073         em = lookup_extent_mapping(em_tree, logical, len);
5074         read_unlock(&em_tree->lock);
5075         BUG_ON(!em);
5076
5077         BUG_ON(em->start > logical || em->start + em->len < logical);
5078         map = (struct map_lookup *)em->bdev;
5079         if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5080                 len = map->stripe_len * nr_data_stripes(map);
5081         free_extent_map(em);
5082         return len;
5083 }
5084
5085 int btrfs_is_parity_mirror(struct btrfs_mapping_tree *map_tree,
5086                            u64 logical, u64 len, int mirror_num)
5087 {
5088         struct extent_map *em;
5089         struct map_lookup *map;
5090         struct extent_map_tree *em_tree = &map_tree->map_tree;
5091         int ret = 0;
5092
5093         read_lock(&em_tree->lock);
5094         em = lookup_extent_mapping(em_tree, logical, len);
5095         read_unlock(&em_tree->lock);
5096         BUG_ON(!em);
5097
5098         BUG_ON(em->start > logical || em->start + em->len < logical);
5099         map = (struct map_lookup *)em->bdev;
5100         if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5101                 ret = 1;
5102         free_extent_map(em);
5103         return ret;
5104 }
5105
5106 static int find_live_mirror(struct btrfs_fs_info *fs_info,
5107                             struct map_lookup *map, int first, int num,
5108                             int optimal, int dev_replace_is_ongoing)
5109 {
5110         int i;
5111         int tolerance;
5112         struct btrfs_device *srcdev;
5113
5114         if (dev_replace_is_ongoing &&
5115             fs_info->dev_replace.cont_reading_from_srcdev_mode ==
5116              BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
5117                 srcdev = fs_info->dev_replace.srcdev;
5118         else
5119                 srcdev = NULL;
5120
5121         /*
5122          * try to avoid the drive that is the source drive for a
5123          * dev-replace procedure, only choose it if no other non-missing
5124          * mirror is available
5125          */
5126         for (tolerance = 0; tolerance < 2; tolerance++) {
5127                 if (map->stripes[optimal].dev->bdev &&
5128                     (tolerance || map->stripes[optimal].dev != srcdev))
5129                         return optimal;
5130                 for (i = first; i < first + num; i++) {
5131                         if (map->stripes[i].dev->bdev &&
5132                             (tolerance || map->stripes[i].dev != srcdev))
5133                                 return i;
5134                 }
5135         }
5136
5137         /* we couldn't find one that doesn't fail.  Just return something
5138          * and the io error handling code will clean up eventually
5139          */
5140         return optimal;
5141 }
5142
5143 static inline int parity_smaller(u64 a, u64 b)
5144 {
5145         return a > b;
5146 }
5147
5148 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
5149 static void sort_parity_stripes(struct btrfs_bio *bbio, int num_stripes)
5150 {
5151         struct btrfs_bio_stripe s;
5152         int i;
5153         u64 l;
5154         int again = 1;
5155
5156         while (again) {
5157                 again = 0;
5158                 for (i = 0; i < num_stripes - 1; i++) {
5159                         if (parity_smaller(bbio->raid_map[i],
5160                                            bbio->raid_map[i+1])) {
5161                                 s = bbio->stripes[i];
5162                                 l = bbio->raid_map[i];
5163                                 bbio->stripes[i] = bbio->stripes[i+1];
5164                                 bbio->raid_map[i] = bbio->raid_map[i+1];
5165                                 bbio->stripes[i+1] = s;
5166                                 bbio->raid_map[i+1] = l;
5167
5168                                 again = 1;
5169                         }
5170                 }
5171         }
5172 }
5173
5174 static struct btrfs_bio *alloc_btrfs_bio(int total_stripes, int real_stripes)
5175 {
5176         struct btrfs_bio *bbio = kzalloc(
5177                  /* the size of the btrfs_bio */
5178                 sizeof(struct btrfs_bio) +
5179                 /* plus the variable array for the stripes */
5180                 sizeof(struct btrfs_bio_stripe) * (total_stripes) +
5181                 /* plus the variable array for the tgt dev */
5182                 sizeof(int) * (real_stripes) +
5183                 /*
5184                  * plus the raid_map, which includes both the tgt dev
5185                  * and the stripes
5186                  */
5187                 sizeof(u64) * (total_stripes),
5188                 GFP_NOFS|__GFP_NOFAIL);
5189
5190         atomic_set(&bbio->error, 0);
5191         atomic_set(&bbio->refs, 1);
5192
5193         return bbio;
5194 }
5195
5196 void btrfs_get_bbio(struct btrfs_bio *bbio)
5197 {
5198         WARN_ON(!atomic_read(&bbio->refs));
5199         atomic_inc(&bbio->refs);
5200 }
5201
5202 void btrfs_put_bbio(struct btrfs_bio *bbio)
5203 {
5204         if (!bbio)
5205                 return;
5206         if (atomic_dec_and_test(&bbio->refs))
5207                 kfree(bbio);
5208 }
5209
5210 static int __btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
5211                              u64 logical, u64 *length,
5212                              struct btrfs_bio **bbio_ret,
5213                              int mirror_num, int need_raid_map)
5214 {
5215         struct extent_map *em;
5216         struct map_lookup *map;
5217         struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
5218         struct extent_map_tree *em_tree = &map_tree->map_tree;
5219         u64 offset;
5220         u64 stripe_offset;
5221         u64 stripe_end_offset;
5222         u64 stripe_nr;
5223         u64 stripe_nr_orig;
5224         u64 stripe_nr_end;
5225         u64 stripe_len;
5226         u32 stripe_index;
5227         int i;
5228         int ret = 0;
5229         int num_stripes;
5230         int max_errors = 0;
5231         int tgtdev_indexes = 0;
5232         struct btrfs_bio *bbio = NULL;
5233         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
5234         int dev_replace_is_ongoing = 0;
5235         int num_alloc_stripes;
5236         int patch_the_first_stripe_for_dev_replace = 0;
5237         u64 physical_to_patch_in_first_stripe = 0;
5238         u64 raid56_full_stripe_start = (u64)-1;
5239
5240         read_lock(&em_tree->lock);
5241         em = lookup_extent_mapping(em_tree, logical, *length);
5242         read_unlock(&em_tree->lock);
5243
5244         if (!em) {
5245                 btrfs_crit(fs_info, "unable to find logical %llu len %llu",
5246                         logical, *length);
5247                 return -EINVAL;
5248         }
5249
5250         if (em->start > logical || em->start + em->len < logical) {
5251                 btrfs_crit(fs_info, "found a bad mapping, wanted %Lu, "
5252                            "found %Lu-%Lu", logical, em->start,
5253                            em->start + em->len);
5254                 free_extent_map(em);
5255                 return -EINVAL;
5256         }
5257
5258         map = (struct map_lookup *)em->bdev;
5259         offset = logical - em->start;
5260
5261         stripe_len = map->stripe_len;
5262         stripe_nr = offset;
5263         /*
5264          * stripe_nr counts the total number of stripes we have to stride
5265          * to get to this block
5266          */
5267         stripe_nr = div64_u64(stripe_nr, stripe_len);
5268
5269         stripe_offset = stripe_nr * stripe_len;
5270         BUG_ON(offset < stripe_offset);
5271
5272         /* stripe_offset is the offset of this block in its stripe*/
5273         stripe_offset = offset - stripe_offset;
5274
5275         /* if we're here for raid56, we need to know the stripe aligned start */
5276         if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5277                 unsigned long full_stripe_len = stripe_len * nr_data_stripes(map);
5278                 raid56_full_stripe_start = offset;
5279
5280                 /* allow a write of a full stripe, but make sure we don't
5281                  * allow straddling of stripes
5282                  */
5283                 raid56_full_stripe_start = div64_u64(raid56_full_stripe_start,
5284                                 full_stripe_len);
5285                 raid56_full_stripe_start *= full_stripe_len;
5286         }
5287
5288         if (rw & REQ_DISCARD) {
5289                 /* we don't discard raid56 yet */
5290                 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5291                         ret = -EOPNOTSUPP;
5292                         goto out;
5293                 }
5294                 *length = min_t(u64, em->len - offset, *length);
5295         } else if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
5296                 u64 max_len;
5297                 /* For writes to RAID[56], allow a full stripeset across all disks.
5298                    For other RAID types and for RAID[56] reads, just allow a single
5299                    stripe (on a single disk). */
5300                 if ((map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
5301                     (rw & REQ_WRITE)) {
5302                         max_len = stripe_len * nr_data_stripes(map) -
5303                                 (offset - raid56_full_stripe_start);
5304                 } else {
5305                         /* we limit the length of each bio to what fits in a stripe */
5306                         max_len = stripe_len - stripe_offset;
5307                 }
5308                 *length = min_t(u64, em->len - offset, max_len);
5309         } else {
5310                 *length = em->len - offset;
5311         }
5312
5313         /* This is for when we're called from btrfs_merge_bio_hook() and all
5314            it cares about is the length */
5315         if (!bbio_ret)
5316                 goto out;
5317
5318         btrfs_dev_replace_lock(dev_replace);
5319         dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
5320         if (!dev_replace_is_ongoing)
5321                 btrfs_dev_replace_unlock(dev_replace);
5322
5323         if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
5324             !(rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) &&
5325             dev_replace->tgtdev != NULL) {
5326                 /*
5327                  * in dev-replace case, for repair case (that's the only
5328                  * case where the mirror is selected explicitly when
5329                  * calling btrfs_map_block), blocks left of the left cursor
5330                  * can also be read from the target drive.
5331                  * For REQ_GET_READ_MIRRORS, the target drive is added as
5332                  * the last one to the array of stripes. For READ, it also
5333                  * needs to be supported using the same mirror number.
5334                  * If the requested block is not left of the left cursor,
5335                  * EIO is returned. This can happen because btrfs_num_copies()
5336                  * returns one more in the dev-replace case.
5337                  */
5338                 u64 tmp_length = *length;
5339                 struct btrfs_bio *tmp_bbio = NULL;
5340                 int tmp_num_stripes;
5341                 u64 srcdev_devid = dev_replace->srcdev->devid;
5342                 int index_srcdev = 0;
5343                 int found = 0;
5344                 u64 physical_of_found = 0;
5345
5346                 ret = __btrfs_map_block(fs_info, REQ_GET_READ_MIRRORS,
5347                              logical, &tmp_length, &tmp_bbio, 0, 0);
5348                 if (ret) {
5349                         WARN_ON(tmp_bbio != NULL);
5350                         goto out;
5351                 }
5352
5353                 tmp_num_stripes = tmp_bbio->num_stripes;
5354                 if (mirror_num > tmp_num_stripes) {
5355                         /*
5356                          * REQ_GET_READ_MIRRORS does not contain this
5357                          * mirror, that means that the requested area
5358                          * is not left of the left cursor
5359                          */
5360                         ret = -EIO;
5361                         btrfs_put_bbio(tmp_bbio);
5362                         goto out;
5363                 }
5364
5365                 /*
5366                  * process the rest of the function using the mirror_num
5367                  * of the source drive. Therefore look it up first.
5368                  * At the end, patch the device pointer to the one of the
5369                  * target drive.
5370                  */
5371                 for (i = 0; i < tmp_num_stripes; i++) {
5372                         if (tmp_bbio->stripes[i].dev->devid == srcdev_devid) {
5373                                 /*
5374                                  * In case of DUP, in order to keep it
5375                                  * simple, only add the mirror with the
5376                                  * lowest physical address
5377                                  */
5378                                 if (found &&
5379                                     physical_of_found <=
5380                                      tmp_bbio->stripes[i].physical)
5381                                         continue;
5382                                 index_srcdev = i;
5383                                 found = 1;
5384                                 physical_of_found =
5385                                         tmp_bbio->stripes[i].physical;
5386                         }
5387                 }
5388
5389                 if (found) {
5390                         mirror_num = index_srcdev + 1;
5391                         patch_the_first_stripe_for_dev_replace = 1;
5392                         physical_to_patch_in_first_stripe = physical_of_found;
5393                 } else {
5394                         WARN_ON(1);
5395                         ret = -EIO;
5396                         btrfs_put_bbio(tmp_bbio);
5397                         goto out;
5398                 }
5399
5400                 btrfs_put_bbio(tmp_bbio);
5401         } else if (mirror_num > map->num_stripes) {
5402                 mirror_num = 0;
5403         }
5404
5405         num_stripes = 1;
5406         stripe_index = 0;
5407         stripe_nr_orig = stripe_nr;
5408         stripe_nr_end = ALIGN(offset + *length, map->stripe_len);
5409         stripe_nr_end = div_u64(stripe_nr_end, map->stripe_len);
5410         stripe_end_offset = stripe_nr_end * map->stripe_len -
5411                             (offset + *length);
5412
5413         if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5414                 if (rw & REQ_DISCARD)
5415                         num_stripes = min_t(u64, map->num_stripes,
5416                                             stripe_nr_end - stripe_nr_orig);
5417                 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5418                                 &stripe_index);
5419                 if (!(rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)))
5420                         mirror_num = 1;
5421         } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
5422                 if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS))
5423                         num_stripes = map->num_stripes;
5424                 else if (mirror_num)
5425                         stripe_index = mirror_num - 1;
5426                 else {
5427                         stripe_index = find_live_mirror(fs_info, map, 0,
5428                                             map->num_stripes,
5429                                             current->pid % map->num_stripes,
5430                                             dev_replace_is_ongoing);
5431                         mirror_num = stripe_index + 1;
5432                 }
5433
5434         } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
5435                 if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) {
5436                         num_stripes = map->num_stripes;
5437                 } else if (mirror_num) {
5438                         stripe_index = mirror_num - 1;
5439                 } else {
5440                         mirror_num = 1;
5441                 }
5442
5443         } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5444                 u32 factor = map->num_stripes / map->sub_stripes;
5445
5446                 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
5447                 stripe_index *= map->sub_stripes;
5448
5449                 if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS))
5450                         num_stripes = map->sub_stripes;
5451                 else if (rw & REQ_DISCARD)
5452                         num_stripes = min_t(u64, map->sub_stripes *
5453                                             (stripe_nr_end - stripe_nr_orig),
5454                                             map->num_stripes);
5455                 else if (mirror_num)
5456                         stripe_index += mirror_num - 1;
5457                 else {
5458                         int old_stripe_index = stripe_index;
5459                         stripe_index = find_live_mirror(fs_info, map,
5460                                               stripe_index,
5461                                               map->sub_stripes, stripe_index +
5462                                               current->pid % map->sub_stripes,
5463                                               dev_replace_is_ongoing);
5464                         mirror_num = stripe_index - old_stripe_index + 1;
5465                 }
5466
5467         } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5468                 if (need_raid_map &&
5469                     ((rw & (REQ_WRITE | REQ_GET_READ_MIRRORS)) ||
5470                      mirror_num > 1)) {
5471                         /* push stripe_nr back to the start of the full stripe */
5472                         stripe_nr = div_u64(raid56_full_stripe_start,
5473                                         stripe_len * nr_data_stripes(map));
5474
5475                         /* RAID[56] write or recovery. Return all stripes */
5476                         num_stripes = map->num_stripes;
5477                         max_errors = nr_parity_stripes(map);
5478
5479                         *length = map->stripe_len;
5480                         stripe_index = 0;
5481                         stripe_offset = 0;
5482                 } else {
5483                         /*
5484                          * Mirror #0 or #1 means the original data block.
5485                          * Mirror #2 is RAID5 parity block.
5486                          * Mirror #3 is RAID6 Q block.
5487                          */
5488                         stripe_nr = div_u64_rem(stripe_nr,
5489                                         nr_data_stripes(map), &stripe_index);
5490                         if (mirror_num > 1)
5491                                 stripe_index = nr_data_stripes(map) +
5492                                                 mirror_num - 2;
5493
5494                         /* We distribute the parity blocks across stripes */
5495                         div_u64_rem(stripe_nr + stripe_index, map->num_stripes,
5496                                         &stripe_index);
5497                         if (!(rw & (REQ_WRITE | REQ_DISCARD |
5498                                     REQ_GET_READ_MIRRORS)) && mirror_num <= 1)
5499                                 mirror_num = 1;
5500                 }
5501         } else {
5502                 /*
5503                  * after this, stripe_nr is the number of stripes on this
5504                  * device we have to walk to find the data, and stripe_index is
5505                  * the number of our device in the stripe array
5506                  */
5507                 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5508                                 &stripe_index);
5509                 mirror_num = stripe_index + 1;
5510         }
5511         BUG_ON(stripe_index >= map->num_stripes);
5512
5513         num_alloc_stripes = num_stripes;
5514         if (dev_replace_is_ongoing) {
5515                 if (rw & (REQ_WRITE | REQ_DISCARD))
5516                         num_alloc_stripes <<= 1;
5517                 if (rw & REQ_GET_READ_MIRRORS)
5518                         num_alloc_stripes++;
5519                 tgtdev_indexes = num_stripes;
5520         }
5521
5522         bbio = alloc_btrfs_bio(num_alloc_stripes, tgtdev_indexes);
5523         if (!bbio) {
5524                 ret = -ENOMEM;
5525                 goto out;
5526         }
5527         if (dev_replace_is_ongoing)
5528                 bbio->tgtdev_map = (int *)(bbio->stripes + num_alloc_stripes);
5529
5530         /* build raid_map */
5531         if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK &&
5532             need_raid_map && ((rw & (REQ_WRITE | REQ_GET_READ_MIRRORS)) ||
5533             mirror_num > 1)) {
5534                 u64 tmp;
5535                 unsigned rot;
5536
5537                 bbio->raid_map = (u64 *)((void *)bbio->stripes +
5538                                  sizeof(struct btrfs_bio_stripe) *
5539                                  num_alloc_stripes +
5540                                  sizeof(int) * tgtdev_indexes);
5541
5542                 /* Work out the disk rotation on this stripe-set */
5543                 div_u64_rem(stripe_nr, num_stripes, &rot);
5544
5545                 /* Fill in the logical address of each stripe */
5546                 tmp = stripe_nr * nr_data_stripes(map);
5547                 for (i = 0; i < nr_data_stripes(map); i++)
5548                         bbio->raid_map[(i+rot) % num_stripes] =
5549                                 em->start + (tmp + i) * map->stripe_len;
5550
5551                 bbio->raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE;
5552                 if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5553                         bbio->raid_map[(i+rot+1) % num_stripes] =
5554                                 RAID6_Q_STRIPE;
5555         }
5556
5557         if (rw & REQ_DISCARD) {
5558                 u32 factor = 0;
5559                 u32 sub_stripes = 0;
5560                 u64 stripes_per_dev = 0;
5561                 u32 remaining_stripes = 0;
5562                 u32 last_stripe = 0;
5563
5564                 if (map->type &
5565                     (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID10)) {
5566                         if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5567                                 sub_stripes = 1;
5568                         else
5569                                 sub_stripes = map->sub_stripes;
5570
5571                         factor = map->num_stripes / sub_stripes;
5572                         stripes_per_dev = div_u64_rem(stripe_nr_end -
5573                                                       stripe_nr_orig,
5574                                                       factor,
5575                                                       &remaining_stripes);
5576                         div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
5577                         last_stripe *= sub_stripes;
5578                 }
5579
5580                 for (i = 0; i < num_stripes; i++) {
5581                         bbio->stripes[i].physical =
5582                                 map->stripes[stripe_index].physical +
5583                                 stripe_offset + stripe_nr * map->stripe_len;
5584                         bbio->stripes[i].dev = map->stripes[stripe_index].dev;
5585
5586                         if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5587                                          BTRFS_BLOCK_GROUP_RAID10)) {
5588                                 bbio->stripes[i].length = stripes_per_dev *
5589                                                           map->stripe_len;
5590
5591                                 if (i / sub_stripes < remaining_stripes)
5592                                         bbio->stripes[i].length +=
5593                                                 map->stripe_len;
5594
5595                                 /*
5596                                  * Special for the first stripe and
5597                                  * the last stripe:
5598                                  *
5599                                  * |-------|...|-------|
5600                                  *     |----------|
5601                                  *    off     end_off
5602                                  */
5603                                 if (i < sub_stripes)
5604                                         bbio->stripes[i].length -=
5605                                                 stripe_offset;
5606
5607                                 if (stripe_index >= last_stripe &&
5608                                     stripe_index <= (last_stripe +
5609                                                      sub_stripes - 1))
5610                                         bbio->stripes[i].length -=
5611                                                 stripe_end_offset;
5612
5613                                 if (i == sub_stripes - 1)
5614                                         stripe_offset = 0;
5615                         } else
5616                                 bbio->stripes[i].length = *length;
5617
5618                         stripe_index++;
5619                         if (stripe_index == map->num_stripes) {
5620                                 /* This could only happen for RAID0/10 */
5621                                 stripe_index = 0;
5622                                 stripe_nr++;
5623                         }
5624                 }
5625         } else {
5626                 for (i = 0; i < num_stripes; i++) {
5627                         bbio->stripes[i].physical =
5628                                 map->stripes[stripe_index].physical +
5629                                 stripe_offset +
5630                                 stripe_nr * map->stripe_len;
5631                         bbio->stripes[i].dev =
5632                                 map->stripes[stripe_index].dev;
5633                         stripe_index++;
5634                 }
5635         }
5636
5637         if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS))
5638                 max_errors = btrfs_chunk_max_errors(map);
5639
5640         if (bbio->raid_map)
5641                 sort_parity_stripes(bbio, num_stripes);
5642
5643         tgtdev_indexes = 0;
5644         if (dev_replace_is_ongoing && (rw & (REQ_WRITE | REQ_DISCARD)) &&
5645             dev_replace->tgtdev != NULL) {
5646                 int index_where_to_add;
5647                 u64 srcdev_devid = dev_replace->srcdev->devid;
5648
5649                 /*
5650                  * duplicate the write operations while the dev replace
5651                  * procedure is running. Since the copying of the old disk
5652                  * to the new disk takes place at run time while the
5653                  * filesystem is mounted writable, the regular write
5654                  * operations to the old disk have to be duplicated to go
5655                  * to the new disk as well.
5656                  * Note that device->missing is handled by the caller, and
5657                  * that the write to the old disk is already set up in the
5658                  * stripes array.
5659                  */
5660                 index_where_to_add = num_stripes;
5661                 for (i = 0; i < num_stripes; i++) {
5662                         if (bbio->stripes[i].dev->devid == srcdev_devid) {
5663                                 /* write to new disk, too */
5664                                 struct btrfs_bio_stripe *new =
5665                                         bbio->stripes + index_where_to_add;
5666                                 struct btrfs_bio_stripe *old =
5667                                         bbio->stripes + i;
5668
5669                                 new->physical = old->physical;
5670                                 new->length = old->length;
5671                                 new->dev = dev_replace->tgtdev;
5672                                 bbio->tgtdev_map[i] = index_where_to_add;
5673                                 index_where_to_add++;
5674                                 max_errors++;
5675                                 tgtdev_indexes++;
5676                         }
5677                 }
5678                 num_stripes = index_where_to_add;
5679         } else if (dev_replace_is_ongoing && (rw & REQ_GET_READ_MIRRORS) &&
5680                    dev_replace->tgtdev != NULL) {
5681                 u64 srcdev_devid = dev_replace->srcdev->devid;
5682                 int index_srcdev = 0;
5683                 int found = 0;
5684                 u64 physical_of_found = 0;
5685
5686                 /*
5687                  * During the dev-replace procedure, the target drive can
5688                  * also be used to read data in case it is needed to repair
5689                  * a corrupt block elsewhere. This is possible if the
5690                  * requested area is left of the left cursor. In this area,
5691                  * the target drive is a full copy of the source drive.
5692                  */
5693                 for (i = 0; i < num_stripes; i++) {
5694                         if (bbio->stripes[i].dev->devid == srcdev_devid) {
5695                                 /*
5696                                  * In case of DUP, in order to keep it
5697                                  * simple, only add the mirror with the
5698                                  * lowest physical address
5699                                  */
5700                                 if (found &&
5701                                     physical_of_found <=
5702                                      bbio->stripes[i].physical)
5703                                         continue;
5704                                 index_srcdev = i;
5705                                 found = 1;
5706                                 physical_of_found = bbio->stripes[i].physical;
5707                         }
5708                 }
5709                 if (found) {
5710                         if (physical_of_found + map->stripe_len <=
5711                             dev_replace->cursor_left) {
5712                                 struct btrfs_bio_stripe *tgtdev_stripe =
5713                                         bbio->stripes + num_stripes;
5714
5715                                 tgtdev_stripe->physical = physical_of_found;
5716                                 tgtdev_stripe->length =
5717                                         bbio->stripes[index_srcdev].length;
5718                                 tgtdev_stripe->dev = dev_replace->tgtdev;
5719                                 bbio->tgtdev_map[index_srcdev] = num_stripes;
5720
5721                                 tgtdev_indexes++;
5722                                 num_stripes++;
5723                         }
5724                 }
5725         }
5726
5727         *bbio_ret = bbio;
5728         bbio->map_type = map->type;
5729         bbio->num_stripes = num_stripes;
5730         bbio->max_errors = max_errors;
5731         bbio->mirror_num = mirror_num;
5732         bbio->num_tgtdevs = tgtdev_indexes;
5733
5734         /*
5735          * this is the case that REQ_READ && dev_replace_is_ongoing &&
5736          * mirror_num == num_stripes + 1 && dev_replace target drive is
5737          * available as a mirror
5738          */
5739         if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) {
5740                 WARN_ON(num_stripes > 1);
5741                 bbio->stripes[0].dev = dev_replace->tgtdev;
5742                 bbio->stripes[0].physical = physical_to_patch_in_first_stripe;
5743                 bbio->mirror_num = map->num_stripes + 1;
5744         }
5745 out:
5746         if (dev_replace_is_ongoing)
5747                 btrfs_dev_replace_unlock(dev_replace);
5748         free_extent_map(em);
5749         return ret;
5750 }
5751
5752 int btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
5753                       u64 logical, u64 *length,
5754                       struct btrfs_bio **bbio_ret, int mirror_num)
5755 {
5756         return __btrfs_map_block(fs_info, rw, logical, length, bbio_ret,
5757                                  mirror_num, 0);
5758 }
5759
5760 /* For Scrub/replace */
5761 int btrfs_map_sblock(struct btrfs_fs_info *fs_info, int rw,
5762                      u64 logical, u64 *length,
5763                      struct btrfs_bio **bbio_ret, int mirror_num,
5764                      int need_raid_map)
5765 {
5766         return __btrfs_map_block(fs_info, rw, logical, length, bbio_ret,
5767                                  mirror_num, need_raid_map);
5768 }
5769
5770 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
5771                      u64 chunk_start, u64 physical, u64 devid,
5772                      u64 **logical, int *naddrs, int *stripe_len)
5773 {
5774         struct extent_map_tree *em_tree = &map_tree->map_tree;
5775         struct extent_map *em;
5776         struct map_lookup *map;
5777         u64 *buf;
5778         u64 bytenr;
5779         u64 length;
5780         u64 stripe_nr;
5781         u64 rmap_len;
5782         int i, j, nr = 0;
5783
5784         read_lock(&em_tree->lock);
5785         em = lookup_extent_mapping(em_tree, chunk_start, 1);
5786         read_unlock(&em_tree->lock);
5787
5788         if (!em) {
5789                 printk(KERN_ERR "BTRFS: couldn't find em for chunk %Lu\n",
5790                        chunk_start);
5791                 return -EIO;
5792         }
5793
5794         if (em->start != chunk_start) {
5795                 printk(KERN_ERR "BTRFS: bad chunk start, em=%Lu, wanted=%Lu\n",
5796                        em->start, chunk_start);
5797                 free_extent_map(em);
5798                 return -EIO;
5799         }
5800         map = (struct map_lookup *)em->bdev;
5801
5802         length = em->len;
5803         rmap_len = map->stripe_len;
5804
5805         if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5806                 length = div_u64(length, map->num_stripes / map->sub_stripes);
5807         else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5808                 length = div_u64(length, map->num_stripes);
5809         else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5810                 length = div_u64(length, nr_data_stripes(map));
5811                 rmap_len = map->stripe_len * nr_data_stripes(map);
5812         }
5813
5814         buf = kcalloc(map->num_stripes, sizeof(u64), GFP_NOFS);
5815         BUG_ON(!buf); /* -ENOMEM */
5816
5817         for (i = 0; i < map->num_stripes; i++) {
5818                 if (devid && map->stripes[i].dev->devid != devid)
5819                         continue;
5820                 if (map->stripes[i].physical > physical ||
5821                     map->stripes[i].physical + length <= physical)
5822                         continue;
5823
5824                 stripe_nr = physical - map->stripes[i].physical;
5825                 stripe_nr = div_u64(stripe_nr, map->stripe_len);
5826
5827                 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5828                         stripe_nr = stripe_nr * map->num_stripes + i;
5829                         stripe_nr = div_u64(stripe_nr, map->sub_stripes);
5830                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5831                         stripe_nr = stripe_nr * map->num_stripes + i;
5832                 } /* else if RAID[56], multiply by nr_data_stripes().
5833                    * Alternatively, just use rmap_len below instead of
5834                    * map->stripe_len */
5835
5836                 bytenr = chunk_start + stripe_nr * rmap_len;
5837                 WARN_ON(nr >= map->num_stripes);
5838                 for (j = 0; j < nr; j++) {
5839                         if (buf[j] == bytenr)
5840                                 break;
5841                 }
5842                 if (j == nr) {
5843                         WARN_ON(nr >= map->num_stripes);
5844                         buf[nr++] = bytenr;
5845                 }
5846         }
5847
5848         *logical = buf;
5849         *naddrs = nr;
5850         *stripe_len = rmap_len;
5851
5852         free_extent_map(em);
5853         return 0;
5854 }
5855
5856 static inline void btrfs_end_bbio(struct btrfs_bio *bbio, struct bio *bio)
5857 {
5858         bio->bi_private = bbio->private;
5859         bio->bi_end_io = bbio->end_io;
5860         bio_endio(bio);
5861
5862         btrfs_put_bbio(bbio);
5863 }
5864
5865 static void btrfs_end_bio(struct bio *bio)
5866 {
5867         struct btrfs_bio *bbio = bio->bi_private;
5868         int is_orig_bio = 0;
5869
5870         if (bio->bi_error) {
5871                 atomic_inc(&bbio->error);
5872                 if (bio->bi_error == -EIO || bio->bi_error == -EREMOTEIO) {
5873                         unsigned int stripe_index =
5874                                 btrfs_io_bio(bio)->stripe_index;
5875                         struct btrfs_device *dev;
5876
5877                         BUG_ON(stripe_index >= bbio->num_stripes);
5878                         dev = bbio->stripes[stripe_index].dev;
5879                         if (dev->bdev) {
5880                                 if (bio->bi_rw & WRITE)
5881                                         btrfs_dev_stat_inc(dev,
5882                                                 BTRFS_DEV_STAT_WRITE_ERRS);
5883                                 else
5884                                         btrfs_dev_stat_inc(dev,
5885                                                 BTRFS_DEV_STAT_READ_ERRS);
5886                                 if ((bio->bi_rw & WRITE_FLUSH) == WRITE_FLUSH)
5887                                         btrfs_dev_stat_inc(dev,
5888                                                 BTRFS_DEV_STAT_FLUSH_ERRS);
5889                                 btrfs_dev_stat_print_on_error(dev);
5890                         }
5891                 }
5892         }
5893
5894         if (bio == bbio->orig_bio)
5895                 is_orig_bio = 1;
5896
5897         btrfs_bio_counter_dec(bbio->fs_info);
5898
5899         if (atomic_dec_and_test(&bbio->stripes_pending)) {
5900                 if (!is_orig_bio) {
5901                         bio_put(bio);
5902                         bio = bbio->orig_bio;
5903                 }
5904
5905                 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
5906                 /* only send an error to the higher layers if it is
5907                  * beyond the tolerance of the btrfs bio
5908                  */
5909                 if (atomic_read(&bbio->error) > bbio->max_errors) {
5910                         bio->bi_error = -EIO;
5911                 } else {
5912                         /*
5913                          * this bio is actually up to date, we didn't
5914                          * go over the max number of errors
5915                          */
5916                         bio->bi_error = 0;
5917                 }
5918
5919                 btrfs_end_bbio(bbio, bio);
5920         } else if (!is_orig_bio) {
5921                 bio_put(bio);
5922         }
5923 }
5924
5925 /*
5926  * see run_scheduled_bios for a description of why bios are collected for
5927  * async submit.
5928  *
5929  * This will add one bio to the pending list for a device and make sure
5930  * the work struct is scheduled.
5931  */
5932 static noinline void btrfs_schedule_bio(struct btrfs_root *root,
5933                                         struct btrfs_device *device,
5934                                         int rw, struct bio *bio)
5935 {
5936         int should_queue = 1;
5937         struct btrfs_pending_bios *pending_bios;
5938
5939         if (device->missing || !device->bdev) {
5940                 bio_io_error(bio);
5941                 return;
5942         }
5943
5944         /* don't bother with additional async steps for reads, right now */
5945         if (!(rw & REQ_WRITE)) {
5946                 bio_get(bio);
5947                 btrfsic_submit_bio(rw, bio);
5948                 bio_put(bio);
5949                 return;
5950         }
5951
5952         /*
5953          * nr_async_bios allows us to reliably return congestion to the
5954          * higher layers.  Otherwise, the async bio makes it appear we have
5955          * made progress against dirty pages when we've really just put it
5956          * on a queue for later
5957          */
5958         atomic_inc(&root->fs_info->nr_async_bios);
5959         WARN_ON(bio->bi_next);
5960         bio->bi_next = NULL;
5961         bio->bi_rw |= rw;
5962
5963         spin_lock(&device->io_lock);
5964         if (bio->bi_rw & REQ_SYNC)
5965                 pending_bios = &device->pending_sync_bios;
5966         else
5967                 pending_bios = &device->pending_bios;
5968
5969         if (pending_bios->tail)
5970                 pending_bios->tail->bi_next = bio;
5971
5972         pending_bios->tail = bio;
5973         if (!pending_bios->head)
5974                 pending_bios->head = bio;
5975         if (device->running_pending)
5976                 should_queue = 0;
5977
5978         spin_unlock(&device->io_lock);
5979
5980         if (should_queue)
5981                 btrfs_queue_work(root->fs_info->submit_workers,
5982                                  &device->work);
5983 }
5984
5985 static void submit_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio,
5986                               struct bio *bio, u64 physical, int dev_nr,
5987                               int rw, int async)
5988 {
5989         struct btrfs_device *dev = bbio->stripes[dev_nr].dev;
5990
5991         bio->bi_private = bbio;
5992         btrfs_io_bio(bio)->stripe_index = dev_nr;
5993         bio->bi_end_io = btrfs_end_bio;
5994         bio->bi_iter.bi_sector = physical >> 9;
5995 #ifdef DEBUG
5996         {
5997                 struct rcu_string *name;
5998
5999                 rcu_read_lock();
6000                 name = rcu_dereference(dev->name);
6001                 pr_debug("btrfs_map_bio: rw %d, sector=%llu, dev=%lu "
6002                          "(%s id %llu), size=%u\n", rw,
6003                          (u64)bio->bi_iter.bi_sector, (u_long)dev->bdev->bd_dev,
6004                          name->str, dev->devid, bio->bi_iter.bi_size);
6005                 rcu_read_unlock();
6006         }
6007 #endif
6008         bio->bi_bdev = dev->bdev;
6009
6010         btrfs_bio_counter_inc_noblocked(root->fs_info);
6011
6012         if (async)
6013                 btrfs_schedule_bio(root, dev, rw, bio);
6014         else
6015                 btrfsic_submit_bio(rw, bio);
6016 }
6017
6018 static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical)
6019 {
6020         atomic_inc(&bbio->error);
6021         if (atomic_dec_and_test(&bbio->stripes_pending)) {
6022                 /* Shoud be the original bio. */
6023                 WARN_ON(bio != bbio->orig_bio);
6024
6025                 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
6026                 bio->bi_iter.bi_sector = logical >> 9;
6027                 bio->bi_error = -EIO;
6028                 btrfs_end_bbio(bbio, bio);
6029         }
6030 }
6031
6032 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
6033                   int mirror_num, int async_submit)
6034 {
6035         struct btrfs_device *dev;
6036         struct bio *first_bio = bio;
6037         u64 logical = (u64)bio->bi_iter.bi_sector << 9;
6038         u64 length = 0;
6039         u64 map_length;
6040         int ret;
6041         int dev_nr;
6042         int total_devs;
6043         struct btrfs_bio *bbio = NULL;
6044
6045         length = bio->bi_iter.bi_size;
6046         map_length = length;
6047
6048         btrfs_bio_counter_inc_blocked(root->fs_info);
6049         ret = __btrfs_map_block(root->fs_info, rw, logical, &map_length, &bbio,
6050                               mirror_num, 1);
6051         if (ret) {
6052                 btrfs_bio_counter_dec(root->fs_info);
6053                 return ret;
6054         }
6055
6056         total_devs = bbio->num_stripes;
6057         bbio->orig_bio = first_bio;
6058         bbio->private = first_bio->bi_private;
6059         bbio->end_io = first_bio->bi_end_io;
6060         bbio->fs_info = root->fs_info;
6061         atomic_set(&bbio->stripes_pending, bbio->num_stripes);
6062
6063         if (bbio->raid_map) {
6064                 /* In this case, map_length has been set to the length of
6065                    a single stripe; not the whole write */
6066                 if (rw & WRITE) {
6067                         ret = raid56_parity_write(root, bio, bbio, map_length);
6068                 } else {
6069                         ret = raid56_parity_recover(root, bio, bbio, map_length,
6070                                                     mirror_num, 1);
6071                 }
6072
6073                 btrfs_bio_counter_dec(root->fs_info);
6074                 return ret;
6075         }
6076
6077         if (map_length < length) {
6078                 btrfs_crit(root->fs_info, "mapping failed logical %llu bio len %llu len %llu",
6079                         logical, length, map_length);
6080                 BUG();
6081         }
6082
6083         for (dev_nr = 0; dev_nr < total_devs; dev_nr++) {
6084                 dev = bbio->stripes[dev_nr].dev;
6085                 if (!dev || !dev->bdev || (rw & WRITE && !dev->writeable)) {
6086                         bbio_error(bbio, first_bio, logical);
6087                         continue;
6088                 }
6089
6090                 if (dev_nr < total_devs - 1) {
6091                         bio = btrfs_bio_clone(first_bio, GFP_NOFS);
6092                         BUG_ON(!bio); /* -ENOMEM */
6093                 } else
6094                         bio = first_bio;
6095
6096                 submit_stripe_bio(root, bbio, bio,
6097                                   bbio->stripes[dev_nr].physical, dev_nr, rw,
6098                                   async_submit);
6099         }
6100         btrfs_bio_counter_dec(root->fs_info);
6101         return 0;
6102 }
6103
6104 struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid,
6105                                        u8 *uuid, u8 *fsid)
6106 {
6107         struct btrfs_device *device;
6108         struct btrfs_fs_devices *cur_devices;
6109
6110         cur_devices = fs_info->fs_devices;
6111         while (cur_devices) {
6112                 if (!fsid ||
6113                     !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
6114                         device = __find_device(&cur_devices->devices,
6115                                                devid, uuid);
6116                         if (device)
6117                                 return device;
6118                 }
6119                 cur_devices = cur_devices->seed;
6120         }
6121         return NULL;
6122 }
6123
6124 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
6125                                             struct btrfs_fs_devices *fs_devices,
6126                                             u64 devid, u8 *dev_uuid)
6127 {
6128         struct btrfs_device *device;
6129
6130         device = btrfs_alloc_device(NULL, &devid, dev_uuid);
6131         if (IS_ERR(device))
6132                 return NULL;
6133
6134         list_add(&device->dev_list, &fs_devices->devices);
6135         device->fs_devices = fs_devices;
6136         fs_devices->num_devices++;
6137
6138         device->missing = 1;
6139         fs_devices->missing_devices++;
6140
6141         return device;
6142 }
6143
6144 /**
6145  * btrfs_alloc_device - allocate struct btrfs_device
6146  * @fs_info:    used only for generating a new devid, can be NULL if
6147  *              devid is provided (i.e. @devid != NULL).
6148  * @devid:      a pointer to devid for this device.  If NULL a new devid
6149  *              is generated.
6150  * @uuid:       a pointer to UUID for this device.  If NULL a new UUID
6151  *              is generated.
6152  *
6153  * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
6154  * on error.  Returned struct is not linked onto any lists and can be
6155  * destroyed with kfree() right away.
6156  */
6157 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
6158                                         const u64 *devid,
6159                                         const u8 *uuid)
6160 {
6161         struct btrfs_device *dev;
6162         u64 tmp;
6163
6164         if (WARN_ON(!devid && !fs_info))
6165                 return ERR_PTR(-EINVAL);
6166
6167         dev = __alloc_device();
6168         if (IS_ERR(dev))
6169                 return dev;
6170
6171         if (devid)
6172                 tmp = *devid;
6173         else {
6174                 int ret;
6175
6176                 ret = find_next_devid(fs_info, &tmp);
6177                 if (ret) {
6178                         kfree(dev);
6179                         return ERR_PTR(ret);
6180                 }
6181         }
6182         dev->devid = tmp;
6183
6184         if (uuid)
6185                 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE);
6186         else
6187                 generate_random_uuid(dev->uuid);
6188
6189         btrfs_init_work(&dev->work, btrfs_submit_helper,
6190                         pending_bios_fn, NULL, NULL);
6191
6192         return dev;
6193 }
6194
6195 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
6196                           struct extent_buffer *leaf,
6197                           struct btrfs_chunk *chunk)
6198 {
6199         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
6200         struct map_lookup *map;
6201         struct extent_map *em;
6202         u64 logical;
6203         u64 length;
6204         u64 devid;
6205         u8 uuid[BTRFS_UUID_SIZE];
6206         int num_stripes;
6207         int ret;
6208         int i;
6209
6210         logical = key->offset;
6211         length = btrfs_chunk_length(leaf, chunk);
6212
6213         read_lock(&map_tree->map_tree.lock);
6214         em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
6215         read_unlock(&map_tree->map_tree.lock);
6216
6217         /* already mapped? */
6218         if (em && em->start <= logical && em->start + em->len > logical) {
6219                 free_extent_map(em);
6220                 return 0;
6221         } else if (em) {
6222                 free_extent_map(em);
6223         }
6224
6225         em = alloc_extent_map();
6226         if (!em)
6227                 return -ENOMEM;
6228         num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6229         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
6230         if (!map) {
6231                 free_extent_map(em);
6232                 return -ENOMEM;
6233         }
6234
6235         set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
6236         em->bdev = (struct block_device *)map;
6237         em->start = logical;
6238         em->len = length;
6239         em->orig_start = 0;
6240         em->block_start = 0;
6241         em->block_len = em->len;
6242
6243         map->num_stripes = num_stripes;
6244         map->io_width = btrfs_chunk_io_width(leaf, chunk);
6245         map->io_align = btrfs_chunk_io_align(leaf, chunk);
6246         map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
6247         map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6248         map->type = btrfs_chunk_type(leaf, chunk);
6249         map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
6250         for (i = 0; i < num_stripes; i++) {
6251                 map->stripes[i].physical =
6252                         btrfs_stripe_offset_nr(leaf, chunk, i);
6253                 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
6254                 read_extent_buffer(leaf, uuid, (unsigned long)
6255                                    btrfs_stripe_dev_uuid_nr(chunk, i),
6256                                    BTRFS_UUID_SIZE);
6257                 map->stripes[i].dev = btrfs_find_device(root->fs_info, devid,
6258                                                         uuid, NULL);
6259                 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
6260                         free_extent_map(em);
6261                         return -EIO;
6262                 }
6263                 if (!map->stripes[i].dev) {
6264                         map->stripes[i].dev =
6265                                 add_missing_dev(root, root->fs_info->fs_devices,
6266                                                 devid, uuid);
6267                         if (!map->stripes[i].dev) {
6268                                 free_extent_map(em);
6269                                 return -EIO;
6270                         }
6271                         btrfs_warn(root->fs_info, "devid %llu uuid %pU is missing",
6272                                                 devid, uuid);
6273                 }
6274                 map->stripes[i].dev->in_fs_metadata = 1;
6275         }
6276
6277         write_lock(&map_tree->map_tree.lock);
6278         ret = add_extent_mapping(&map_tree->map_tree, em, 0);
6279         write_unlock(&map_tree->map_tree.lock);
6280         BUG_ON(ret); /* Tree corruption */
6281         free_extent_map(em);
6282
6283         return 0;
6284 }
6285
6286 static void fill_device_from_item(struct extent_buffer *leaf,
6287                                  struct btrfs_dev_item *dev_item,
6288                                  struct btrfs_device *device)
6289 {
6290         unsigned long ptr;
6291
6292         device->devid = btrfs_device_id(leaf, dev_item);
6293         device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
6294         device->total_bytes = device->disk_total_bytes;
6295         device->commit_total_bytes = device->disk_total_bytes;
6296         device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
6297         device->commit_bytes_used = device->bytes_used;
6298         device->type = btrfs_device_type(leaf, dev_item);
6299         device->io_align = btrfs_device_io_align(leaf, dev_item);
6300         device->io_width = btrfs_device_io_width(leaf, dev_item);
6301         device->sector_size = btrfs_device_sector_size(leaf, dev_item);
6302         WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
6303         device->is_tgtdev_for_dev_replace = 0;
6304
6305         ptr = btrfs_device_uuid(dev_item);
6306         read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
6307 }
6308
6309 static struct btrfs_fs_devices *open_seed_devices(struct btrfs_root *root,
6310                                                   u8 *fsid)
6311 {
6312         struct btrfs_fs_devices *fs_devices;
6313         int ret;
6314
6315         BUG_ON(!mutex_is_locked(&uuid_mutex));
6316
6317         fs_devices = root->fs_info->fs_devices->seed;
6318         while (fs_devices) {
6319                 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE))
6320                         return fs_devices;
6321
6322                 fs_devices = fs_devices->seed;
6323         }
6324
6325         fs_devices = find_fsid(fsid);
6326         if (!fs_devices) {
6327                 if (!btrfs_test_opt(root, DEGRADED))
6328                         return ERR_PTR(-ENOENT);
6329
6330                 fs_devices = alloc_fs_devices(fsid);
6331                 if (IS_ERR(fs_devices))
6332                         return fs_devices;
6333
6334                 fs_devices->seeding = 1;
6335                 fs_devices->opened = 1;
6336                 return fs_devices;
6337         }
6338
6339         fs_devices = clone_fs_devices(fs_devices);
6340         if (IS_ERR(fs_devices))
6341                 return fs_devices;
6342
6343         ret = __btrfs_open_devices(fs_devices, FMODE_READ,
6344                                    root->fs_info->bdev_holder);
6345         if (ret) {
6346                 free_fs_devices(fs_devices);
6347                 fs_devices = ERR_PTR(ret);
6348                 goto out;
6349         }
6350
6351         if (!fs_devices->seeding) {
6352                 __btrfs_close_devices(fs_devices);
6353                 free_fs_devices(fs_devices);
6354                 fs_devices = ERR_PTR(-EINVAL);
6355                 goto out;
6356         }
6357
6358         fs_devices->seed = root->fs_info->fs_devices->seed;
6359         root->fs_info->fs_devices->seed = fs_devices;
6360 out:
6361         return fs_devices;
6362 }
6363
6364 static int read_one_dev(struct btrfs_root *root,
6365                         struct extent_buffer *leaf,
6366                         struct btrfs_dev_item *dev_item)
6367 {
6368         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
6369         struct btrfs_device *device;
6370         u64 devid;
6371         int ret;
6372         u8 fs_uuid[BTRFS_UUID_SIZE];
6373         u8 dev_uuid[BTRFS_UUID_SIZE];
6374
6375         devid = btrfs_device_id(leaf, dev_item);
6376         read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
6377                            BTRFS_UUID_SIZE);
6378         read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
6379                            BTRFS_UUID_SIZE);
6380
6381         if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
6382                 fs_devices = open_seed_devices(root, fs_uuid);
6383                 if (IS_ERR(fs_devices))
6384                         return PTR_ERR(fs_devices);
6385         }
6386
6387         device = btrfs_find_device(root->fs_info, devid, dev_uuid, fs_uuid);
6388         if (!device) {
6389                 if (!btrfs_test_opt(root, DEGRADED))
6390                         return -EIO;
6391
6392                 device = add_missing_dev(root, fs_devices, devid, dev_uuid);
6393                 if (!device)
6394                         return -ENOMEM;
6395                 btrfs_warn(root->fs_info, "devid %llu uuid %pU missing",
6396                                 devid, dev_uuid);
6397         } else {
6398                 if (!device->bdev && !btrfs_test_opt(root, DEGRADED))
6399                         return -EIO;
6400
6401                 if(!device->bdev && !device->missing) {
6402                         /*
6403                          * this happens when a device that was properly setup
6404                          * in the device info lists suddenly goes bad.
6405                          * device->bdev is NULL, and so we have to set
6406                          * device->missing to one here
6407                          */
6408                         device->fs_devices->missing_devices++;
6409                         device->missing = 1;
6410                 }
6411
6412                 /* Move the device to its own fs_devices */
6413                 if (device->fs_devices != fs_devices) {
6414                         ASSERT(device->missing);
6415
6416                         list_move(&device->dev_list, &fs_devices->devices);
6417                         device->fs_devices->num_devices--;
6418                         fs_devices->num_devices++;
6419
6420                         device->fs_devices->missing_devices--;
6421                         fs_devices->missing_devices++;
6422
6423                         device->fs_devices = fs_devices;
6424                 }
6425         }
6426
6427         if (device->fs_devices != root->fs_info->fs_devices) {
6428                 BUG_ON(device->writeable);
6429                 if (device->generation !=
6430                     btrfs_device_generation(leaf, dev_item))
6431                         return -EINVAL;
6432         }
6433
6434         fill_device_from_item(leaf, dev_item, device);
6435         device->in_fs_metadata = 1;
6436         if (device->writeable && !device->is_tgtdev_for_dev_replace) {
6437                 device->fs_devices->total_rw_bytes += device->total_bytes;
6438                 spin_lock(&root->fs_info->free_chunk_lock);
6439                 root->fs_info->free_chunk_space += device->total_bytes -
6440                         device->bytes_used;
6441                 spin_unlock(&root->fs_info->free_chunk_lock);
6442         }
6443         ret = 0;
6444         return ret;
6445 }
6446
6447 int btrfs_read_sys_array(struct btrfs_root *root)
6448 {
6449         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
6450         struct extent_buffer *sb;
6451         struct btrfs_disk_key *disk_key;
6452         struct btrfs_chunk *chunk;
6453         u8 *array_ptr;
6454         unsigned long sb_array_offset;
6455         int ret = 0;
6456         u32 num_stripes;
6457         u32 array_size;
6458         u32 len = 0;
6459         u32 cur_offset;
6460         struct btrfs_key key;
6461
6462         ASSERT(BTRFS_SUPER_INFO_SIZE <= root->nodesize);
6463         /*
6464          * This will create extent buffer of nodesize, superblock size is
6465          * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will
6466          * overallocate but we can keep it as-is, only the first page is used.
6467          */
6468         sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET);
6469         if (!sb)
6470                 return -ENOMEM;
6471         btrfs_set_buffer_uptodate(sb);
6472         btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
6473         /*
6474          * The sb extent buffer is artifical and just used to read the system array.
6475          * btrfs_set_buffer_uptodate() call does not properly mark all it's
6476          * pages up-to-date when the page is larger: extent does not cover the
6477          * whole page and consequently check_page_uptodate does not find all
6478          * the page's extents up-to-date (the hole beyond sb),
6479          * write_extent_buffer then triggers a WARN_ON.
6480          *
6481          * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
6482          * but sb spans only this function. Add an explicit SetPageUptodate call
6483          * to silence the warning eg. on PowerPC 64.
6484          */
6485         if (PAGE_CACHE_SIZE > BTRFS_SUPER_INFO_SIZE)
6486                 SetPageUptodate(sb->pages[0]);
6487
6488         write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
6489         array_size = btrfs_super_sys_array_size(super_copy);
6490
6491         array_ptr = super_copy->sys_chunk_array;
6492         sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array);
6493         cur_offset = 0;
6494
6495         while (cur_offset < array_size) {
6496                 disk_key = (struct btrfs_disk_key *)array_ptr;
6497                 len = sizeof(*disk_key);
6498                 if (cur_offset + len > array_size)
6499                         goto out_short_read;
6500
6501                 btrfs_disk_key_to_cpu(&key, disk_key);
6502
6503                 array_ptr += len;
6504                 sb_array_offset += len;
6505                 cur_offset += len;
6506
6507                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
6508                         chunk = (struct btrfs_chunk *)sb_array_offset;
6509                         /*
6510                          * At least one btrfs_chunk with one stripe must be
6511                          * present, exact stripe count check comes afterwards
6512                          */
6513                         len = btrfs_chunk_item_size(1);
6514                         if (cur_offset + len > array_size)
6515                                 goto out_short_read;
6516
6517                         num_stripes = btrfs_chunk_num_stripes(sb, chunk);
6518                         len = btrfs_chunk_item_size(num_stripes);
6519                         if (cur_offset + len > array_size)
6520                                 goto out_short_read;
6521
6522                         ret = read_one_chunk(root, &key, sb, chunk);
6523                         if (ret)
6524                                 break;
6525                 } else {
6526                         ret = -EIO;
6527                         break;
6528                 }
6529                 array_ptr += len;
6530                 sb_array_offset += len;
6531                 cur_offset += len;
6532         }
6533         free_extent_buffer(sb);
6534         return ret;
6535
6536 out_short_read:
6537         printk(KERN_ERR "BTRFS: sys_array too short to read %u bytes at offset %u\n",
6538                         len, cur_offset);
6539         free_extent_buffer(sb);
6540         return -EIO;
6541 }
6542
6543 int btrfs_read_chunk_tree(struct btrfs_root *root)
6544 {
6545         struct btrfs_path *path;
6546         struct extent_buffer *leaf;
6547         struct btrfs_key key;
6548         struct btrfs_key found_key;
6549         int ret;
6550         int slot;
6551
6552         root = root->fs_info->chunk_root;
6553
6554         path = btrfs_alloc_path();
6555         if (!path)
6556                 return -ENOMEM;
6557
6558         mutex_lock(&uuid_mutex);
6559         lock_chunks(root);
6560
6561         /*
6562          * Read all device items, and then all the chunk items. All
6563          * device items are found before any chunk item (their object id
6564          * is smaller than the lowest possible object id for a chunk
6565          * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
6566          */
6567         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
6568         key.offset = 0;
6569         key.type = 0;
6570         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6571         if (ret < 0)
6572                 goto error;
6573         while (1) {
6574                 leaf = path->nodes[0];
6575                 slot = path->slots[0];
6576                 if (slot >= btrfs_header_nritems(leaf)) {
6577                         ret = btrfs_next_leaf(root, path);
6578                         if (ret == 0)
6579                                 continue;
6580                         if (ret < 0)
6581                                 goto error;
6582                         break;
6583                 }
6584                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6585                 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
6586                         struct btrfs_dev_item *dev_item;
6587                         dev_item = btrfs_item_ptr(leaf, slot,
6588                                                   struct btrfs_dev_item);
6589                         ret = read_one_dev(root, leaf, dev_item);
6590                         if (ret)
6591                                 goto error;
6592                 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
6593                         struct btrfs_chunk *chunk;
6594                         chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
6595                         ret = read_one_chunk(root, &found_key, leaf, chunk);
6596                         if (ret)
6597                                 goto error;
6598                 }
6599                 path->slots[0]++;
6600         }
6601         ret = 0;
6602 error:
6603         unlock_chunks(root);
6604         mutex_unlock(&uuid_mutex);
6605
6606         btrfs_free_path(path);
6607         return ret;
6608 }
6609
6610 void btrfs_init_devices_late(struct btrfs_fs_info *fs_info)
6611 {
6612         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6613         struct btrfs_device *device;
6614
6615         while (fs_devices) {
6616                 mutex_lock(&fs_devices->device_list_mutex);
6617                 list_for_each_entry(device, &fs_devices->devices, dev_list)
6618                         device->dev_root = fs_info->dev_root;
6619                 mutex_unlock(&fs_devices->device_list_mutex);
6620
6621                 fs_devices = fs_devices->seed;
6622         }
6623 }
6624
6625 static void __btrfs_reset_dev_stats(struct btrfs_device *dev)
6626 {
6627         int i;
6628
6629         for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6630                 btrfs_dev_stat_reset(dev, i);
6631 }
6632
6633 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
6634 {
6635         struct btrfs_key key;
6636         struct btrfs_key found_key;
6637         struct btrfs_root *dev_root = fs_info->dev_root;
6638         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6639         struct extent_buffer *eb;
6640         int slot;
6641         int ret = 0;
6642         struct btrfs_device *device;
6643         struct btrfs_path *path = NULL;
6644         int i;
6645
6646         path = btrfs_alloc_path();
6647         if (!path) {
6648                 ret = -ENOMEM;
6649                 goto out;
6650         }
6651
6652         mutex_lock(&fs_devices->device_list_mutex);
6653         list_for_each_entry(device, &fs_devices->devices, dev_list) {
6654                 int item_size;
6655                 struct btrfs_dev_stats_item *ptr;
6656
6657                 key.objectid = 0;
6658                 key.type = BTRFS_DEV_STATS_KEY;
6659                 key.offset = device->devid;
6660                 ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
6661                 if (ret) {
6662                         __btrfs_reset_dev_stats(device);
6663                         device->dev_stats_valid = 1;
6664                         btrfs_release_path(path);
6665                         continue;
6666                 }
6667                 slot = path->slots[0];
6668                 eb = path->nodes[0];
6669                 btrfs_item_key_to_cpu(eb, &found_key, slot);
6670                 item_size = btrfs_item_size_nr(eb, slot);
6671
6672                 ptr = btrfs_item_ptr(eb, slot,
6673                                      struct btrfs_dev_stats_item);
6674
6675                 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
6676                         if (item_size >= (1 + i) * sizeof(__le64))
6677                                 btrfs_dev_stat_set(device, i,
6678                                         btrfs_dev_stats_value(eb, ptr, i));
6679                         else
6680                                 btrfs_dev_stat_reset(device, i);
6681                 }
6682
6683                 device->dev_stats_valid = 1;
6684                 btrfs_dev_stat_print_on_load(device);
6685                 btrfs_release_path(path);
6686         }
6687         mutex_unlock(&fs_devices->device_list_mutex);
6688
6689 out:
6690         btrfs_free_path(path);
6691         return ret < 0 ? ret : 0;
6692 }
6693
6694 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
6695                                 struct btrfs_root *dev_root,
6696                                 struct btrfs_device *device)
6697 {
6698         struct btrfs_path *path;
6699         struct btrfs_key key;
6700         struct extent_buffer *eb;
6701         struct btrfs_dev_stats_item *ptr;
6702         int ret;
6703         int i;
6704
6705         key.objectid = 0;
6706         key.type = BTRFS_DEV_STATS_KEY;
6707         key.offset = device->devid;
6708
6709         path = btrfs_alloc_path();
6710         BUG_ON(!path);
6711         ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
6712         if (ret < 0) {
6713                 btrfs_warn_in_rcu(dev_root->fs_info,
6714                         "error %d while searching for dev_stats item for device %s",
6715                               ret, rcu_str_deref(device->name));
6716                 goto out;
6717         }
6718
6719         if (ret == 0 &&
6720             btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
6721                 /* need to delete old one and insert a new one */
6722                 ret = btrfs_del_item(trans, dev_root, path);
6723                 if (ret != 0) {
6724                         btrfs_warn_in_rcu(dev_root->fs_info,
6725                                 "delete too small dev_stats item for device %s failed %d",
6726                                       rcu_str_deref(device->name), ret);
6727                         goto out;
6728                 }
6729                 ret = 1;
6730         }
6731
6732         if (ret == 1) {
6733                 /* need to insert a new item */
6734                 btrfs_release_path(path);
6735                 ret = btrfs_insert_empty_item(trans, dev_root, path,
6736                                               &key, sizeof(*ptr));
6737                 if (ret < 0) {
6738                         btrfs_warn_in_rcu(dev_root->fs_info,
6739                                 "insert dev_stats item for device %s failed %d",
6740                                 rcu_str_deref(device->name), ret);
6741                         goto out;
6742                 }
6743         }
6744
6745         eb = path->nodes[0];
6746         ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
6747         for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6748                 btrfs_set_dev_stats_value(eb, ptr, i,
6749                                           btrfs_dev_stat_read(device, i));
6750         btrfs_mark_buffer_dirty(eb);
6751
6752 out:
6753         btrfs_free_path(path);
6754         return ret;
6755 }
6756
6757 /*
6758  * called from commit_transaction. Writes all changed device stats to disk.
6759  */
6760 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans,
6761                         struct btrfs_fs_info *fs_info)
6762 {
6763         struct btrfs_root *dev_root = fs_info->dev_root;
6764         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6765         struct btrfs_device *device;
6766         int stats_cnt;
6767         int ret = 0;
6768
6769         mutex_lock(&fs_devices->device_list_mutex);
6770         list_for_each_entry(device, &fs_devices->devices, dev_list) {
6771                 if (!device->dev_stats_valid || !btrfs_dev_stats_dirty(device))
6772                         continue;
6773
6774                 stats_cnt = atomic_read(&device->dev_stats_ccnt);
6775                 ret = update_dev_stat_item(trans, dev_root, device);
6776                 if (!ret)
6777                         atomic_sub(stats_cnt, &device->dev_stats_ccnt);
6778         }
6779         mutex_unlock(&fs_devices->device_list_mutex);
6780
6781         return ret;
6782 }
6783
6784 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
6785 {
6786         btrfs_dev_stat_inc(dev, index);
6787         btrfs_dev_stat_print_on_error(dev);
6788 }
6789
6790 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
6791 {
6792         if (!dev->dev_stats_valid)
6793                 return;
6794         btrfs_err_rl_in_rcu(dev->dev_root->fs_info,
6795                 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
6796                            rcu_str_deref(dev->name),
6797                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
6798                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
6799                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
6800                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
6801                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
6802 }
6803
6804 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
6805 {
6806         int i;
6807
6808         for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6809                 if (btrfs_dev_stat_read(dev, i) != 0)
6810                         break;
6811         if (i == BTRFS_DEV_STAT_VALUES_MAX)
6812                 return; /* all values == 0, suppress message */
6813
6814         btrfs_info_in_rcu(dev->dev_root->fs_info,
6815                 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
6816                rcu_str_deref(dev->name),
6817                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
6818                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
6819                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
6820                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
6821                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
6822 }
6823
6824 int btrfs_get_dev_stats(struct btrfs_root *root,
6825                         struct btrfs_ioctl_get_dev_stats *stats)
6826 {
6827         struct btrfs_device *dev;
6828         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
6829         int i;
6830
6831         mutex_lock(&fs_devices->device_list_mutex);
6832         dev = btrfs_find_device(root->fs_info, stats->devid, NULL, NULL);
6833         mutex_unlock(&fs_devices->device_list_mutex);
6834
6835         if (!dev) {
6836                 btrfs_warn(root->fs_info, "get dev_stats failed, device not found");
6837                 return -ENODEV;
6838         } else if (!dev->dev_stats_valid) {
6839                 btrfs_warn(root->fs_info, "get dev_stats failed, not yet valid");
6840                 return -ENODEV;
6841         } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
6842                 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
6843                         if (stats->nr_items > i)
6844                                 stats->values[i] =
6845                                         btrfs_dev_stat_read_and_reset(dev, i);
6846                         else
6847                                 btrfs_dev_stat_reset(dev, i);
6848                 }
6849         } else {
6850                 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6851                         if (stats->nr_items > i)
6852                                 stats->values[i] = btrfs_dev_stat_read(dev, i);
6853         }
6854         if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
6855                 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
6856         return 0;
6857 }
6858
6859 void btrfs_scratch_superblocks(struct block_device *bdev, char *device_path)
6860 {
6861         struct buffer_head *bh;
6862         struct btrfs_super_block *disk_super;
6863         int copy_num;
6864
6865         if (!bdev)
6866                 return;
6867
6868         for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX;
6869                 copy_num++) {
6870
6871                 if (btrfs_read_dev_one_super(bdev, copy_num, &bh))
6872                         continue;
6873
6874                 disk_super = (struct btrfs_super_block *)bh->b_data;
6875
6876                 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
6877                 set_buffer_dirty(bh);
6878                 sync_dirty_buffer(bh);
6879                 brelse(bh);
6880         }
6881
6882         /* Notify udev that device has changed */
6883         btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
6884
6885         /* Update ctime/mtime for device path for libblkid */
6886         update_dev_time(device_path);
6887 }
6888
6889 /*
6890  * Update the size of all devices, which is used for writing out the
6891  * super blocks.
6892  */
6893 void btrfs_update_commit_device_size(struct btrfs_fs_info *fs_info)
6894 {
6895         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6896         struct btrfs_device *curr, *next;
6897
6898         if (list_empty(&fs_devices->resized_devices))
6899                 return;
6900
6901         mutex_lock(&fs_devices->device_list_mutex);
6902         lock_chunks(fs_info->dev_root);
6903         list_for_each_entry_safe(curr, next, &fs_devices->resized_devices,
6904                                  resized_list) {
6905                 list_del_init(&curr->resized_list);
6906                 curr->commit_total_bytes = curr->disk_total_bytes;
6907         }
6908         unlock_chunks(fs_info->dev_root);
6909         mutex_unlock(&fs_devices->device_list_mutex);
6910 }
6911
6912 /* Must be invoked during the transaction commit */
6913 void btrfs_update_commit_device_bytes_used(struct btrfs_root *root,
6914                                         struct btrfs_transaction *transaction)
6915 {
6916         struct extent_map *em;
6917         struct map_lookup *map;
6918         struct btrfs_device *dev;
6919         int i;
6920
6921         if (list_empty(&transaction->pending_chunks))
6922                 return;
6923
6924         /* In order to kick the device replace finish process */
6925         lock_chunks(root);
6926         list_for_each_entry(em, &transaction->pending_chunks, list) {
6927                 map = (struct map_lookup *)em->bdev;
6928
6929                 for (i = 0; i < map->num_stripes; i++) {
6930                         dev = map->stripes[i].dev;
6931                         dev->commit_bytes_used = dev->bytes_used;
6932                 }
6933         }
6934         unlock_chunks(root);
6935 }
6936
6937 void btrfs_set_fs_info_ptr(struct btrfs_fs_info *fs_info)
6938 {
6939         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6940         while (fs_devices) {
6941                 fs_devices->fs_info = fs_info;
6942                 fs_devices = fs_devices->seed;
6943         }
6944 }
6945
6946 void btrfs_reset_fs_info_ptr(struct btrfs_fs_info *fs_info)
6947 {
6948         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6949         while (fs_devices) {
6950                 fs_devices->fs_info = NULL;
6951                 fs_devices = fs_devices->seed;
6952         }
6953 }
6954
6955 void btrfs_close_one_device(struct btrfs_device *device)
6956 {
6957         struct btrfs_fs_devices *fs_devices = device->fs_devices;
6958         struct btrfs_device *new_device;
6959         struct rcu_string *name;
6960
6961         if (device->bdev)
6962                 fs_devices->open_devices--;
6963
6964         if (device->writeable &&
6965             device->devid != BTRFS_DEV_REPLACE_DEVID) {
6966                 list_del_init(&device->dev_alloc_list);
6967                 fs_devices->rw_devices--;
6968         }
6969
6970         if (device->missing)
6971                 fs_devices->missing_devices--;
6972
6973         new_device = btrfs_alloc_device(NULL, &device->devid,
6974                                         device->uuid);
6975         BUG_ON(IS_ERR(new_device)); /* -ENOMEM */
6976
6977         /* Safe because we are under uuid_mutex */
6978         if (device->name) {
6979                 name = rcu_string_strdup(device->name->str, GFP_NOFS);
6980                 BUG_ON(!name); /* -ENOMEM */
6981                 rcu_assign_pointer(new_device->name, name);
6982         }
6983
6984         list_replace_rcu(&device->dev_list, &new_device->dev_list);
6985         new_device->fs_devices = device->fs_devices;
6986
6987         call_rcu(&device->rcu, free_device);
6988 }