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 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 static void btrfs_close_one_device(struct btrfs_device *device);
 129
 130 DEFINE_MUTEX(uuid_mutex);
 131 static LIST_HEAD(fs_uuids);
 132 struct list_head *btrfs_get_fs_uuids(void)
 133 {
 134         return &fs_uuids;
 135 }
 136
 137 static struct btrfs_fs_devices *__alloc_fs_devices(void)
 138 {
 139         struct btrfs_fs_devices *fs_devs;
 140
 141         fs_devs = kzalloc(sizeof(*fs_devs), GFP_KERNEL);
 142         if (!fs_devs)
 143                 return ERR_PTR(-ENOMEM);
 144
 145         mutex_init(&fs_devs->device_list_mutex);
 146
 147         INIT_LIST_HEAD(&fs_devs->devices);
 148         INIT_LIST_HEAD(&fs_devs->resized_devices);
 149         INIT_LIST_HEAD(&fs_devs->alloc_list);
 150         INIT_LIST_HEAD(&fs_devs->list);
 151
 152         return fs_devs;
 153 }
 154
 155 /**
 156  * alloc_fs_devices - allocate struct btrfs_fs_devices
 157  * @fsid:       a pointer to UUID for this FS.  If NULL a new UUID is
 158  *              generated.
 159  *
 160  * Return: a pointer to a new &struct btrfs_fs_devices on success;
 161  * ERR_PTR() on error.  Returned struct is not linked onto any lists and
 162  * can be destroyed with kfree() right away.
 163  */
 164 static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid)
 165 {
 166         struct btrfs_fs_devices *fs_devs;
 167
 168         fs_devs = __alloc_fs_devices();
 169         if (IS_ERR(fs_devs))
 170                 return fs_devs;
 171
 172         if (fsid)
 173                 memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE);
 174         else
 175                 generate_random_uuid(fs_devs->fsid);
 176
 177         return fs_devs;
 178 }
 179
 180 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
 181 {
 182         struct btrfs_device *device;
 183         WARN_ON(fs_devices->opened);
 184         while (!list_empty(&fs_devices->devices)) {
 185                 device = list_entry(fs_devices->devices.next,
 186                                     struct btrfs_device, dev_list);
 187                 list_del(&device->dev_list);
 188                 rcu_string_free(device->name);
 189                 kfree(device);
 190         }
 191         kfree(fs_devices);
 192 }
 193
 194 static void btrfs_kobject_uevent(struct block_device *bdev,
 195                                  enum kobject_action action)
 196 {
 197         int ret;
 198
 199         ret = kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, action);
 200         if (ret)
 201                 pr_warn("BTRFS: Sending event '%d' to kobject: '%s' (%p): failed\n",
 202                         action,
 203                         kobject_name(&disk_to_dev(bdev->bd_disk)->kobj),
 204                         &disk_to_dev(bdev->bd_disk)->kobj);
 205 }
 206
 207 void btrfs_cleanup_fs_uuids(void)
 208 {
 209         struct btrfs_fs_devices *fs_devices;
 210
 211         while (!list_empty(&fs_uuids)) {
 212                 fs_devices = list_entry(fs_uuids.next,
 213                                         struct btrfs_fs_devices, list);
 214                 list_del(&fs_devices->list);
 215                 free_fs_devices(fs_devices);
 216         }
 217 }
 218
 219 static struct btrfs_device *__alloc_device(void)
 220 {
 221         struct btrfs_device *dev;
 222
 223         dev = kzalloc(sizeof(*dev), GFP_KERNEL);
 224         if (!dev)
 225                 return ERR_PTR(-ENOMEM);
 226
 227         INIT_LIST_HEAD(&dev->dev_list);
 228         INIT_LIST_HEAD(&dev->dev_alloc_list);
 229         INIT_LIST_HEAD(&dev->resized_list);
 230
 231         spin_lock_init(&dev->io_lock);
 232
 233         spin_lock_init(&dev->reada_lock);
 234         atomic_set(&dev->reada_in_flight, 0);
 235         atomic_set(&dev->dev_stats_ccnt, 0);
 236         btrfs_device_data_ordered_init(dev);
 237         INIT_RADIX_TREE(&dev->reada_zones, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
 238         INIT_RADIX_TREE(&dev->reada_extents, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
 239
 240         return dev;
 241 }
 242
 243 static noinline struct btrfs_device *__find_device(struct list_head *head,
 244                                                    u64 devid, u8 *uuid)
 245 {
 246         struct btrfs_device *dev;
 247
 248         list_for_each_entry(dev, head, dev_list) {
 249                 if (dev->devid == devid &&
 250                     (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
 251                         return dev;
 252                 }
 253         }
 254         return NULL;
 255 }
 256
 257 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
 258 {
 259         struct btrfs_fs_devices *fs_devices;
 260
 261         list_for_each_entry(fs_devices, &fs_uuids, list) {
 262                 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
 263                         return fs_devices;
 264         }
 265         return NULL;
 266 }
 267
 268 static int
 269 btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder,
 270                       int flush, struct block_device **bdev,
 271                       struct buffer_head **bh)
 272 {
 273         int ret;
 274
 275         *bdev = blkdev_get_by_path(device_path, flags, holder);
 276
 277         if (IS_ERR(*bdev)) {
 278                 ret = PTR_ERR(*bdev);
 279                 goto error;
 280         }
 281
 282         if (flush)
 283                 filemap_write_and_wait((*bdev)->bd_inode->i_mapping);
 284         ret = set_blocksize(*bdev, 4096);
 285         if (ret) {
 286                 blkdev_put(*bdev, flags);
 287                 goto error;
 288         }
 289         invalidate_bdev(*bdev);
 290         *bh = btrfs_read_dev_super(*bdev);
 291         if (IS_ERR(*bh)) {
 292                 ret = PTR_ERR(*bh);
 293                 blkdev_put(*bdev, flags);
 294                 goto error;
 295         }
 296
 297         return 0;
 298
 299 error:
 300         *bdev = NULL;
 301         *bh = NULL;
 302         return ret;
 303 }
 304
 305 static void requeue_list(struct btrfs_pending_bios *pending_bios,
 306                         struct bio *head, struct bio *tail)
 307 {
 308
 309         struct bio *old_head;
 310
 311         old_head = pending_bios->head;
 312         pending_bios->head = head;
 313         if (pending_bios->tail)
 314                 tail->bi_next = old_head;
 315         else
 316                 pending_bios->tail = tail;
 317 }
 318
 319 /*
 320  * we try to collect pending bios for a device so we don't get a large
 321  * number of procs sending bios down to the same device.  This greatly
 322  * improves the schedulers ability to collect and merge the bios.
 323  *
 324  * But, it also turns into a long list of bios to process and that is sure
 325  * to eventually make the worker thread block.  The solution here is to
 326  * make some progress and then put this work struct back at the end of
 327  * the list if the block device is congested.  This way, multiple devices
 328  * can make progress from a single worker thread.
 329  */
 330 static noinline void run_scheduled_bios(struct btrfs_device *device)
 331 {
 332         struct bio *pending;
 333         struct backing_dev_info *bdi;
 334         struct btrfs_fs_info *fs_info;
 335         struct btrfs_pending_bios *pending_bios;
 336         struct bio *tail;
 337         struct bio *cur;
 338         int again = 0;
 339         unsigned long num_run;
 340         unsigned long batch_run = 0;
 341         unsigned long limit;
 342         unsigned long last_waited = 0;
 343         int force_reg = 0;
 344         int sync_pending = 0;
 345         struct blk_plug plug;
 346
 347         /*
 348          * this function runs all the bios we've collected for
 349          * a particular device.  We don't want to wander off to
 350          * another device without first sending all of these down.
 351          * So, setup a plug here and finish it off before we return
 352          */
 353         blk_start_plug(&plug);
 354
 355         bdi = blk_get_backing_dev_info(device->bdev);
 356         fs_info = device->dev_root->fs_info;
 357         limit = btrfs_async_submit_limit(fs_info);
 358         limit = limit * 2 / 3;
 359
 360 loop:
 361         spin_lock(&device->io_lock);
 362
 363 loop_lock:
 364         num_run = 0;
 365
 366         /* take all the bios off the list at once and process them
 367          * later on (without the lock held).  But, remember the
 368          * tail and other pointers so the bios can be properly reinserted
 369          * into the list if we hit congestion
 370          */
 371         if (!force_reg && device->pending_sync_bios.head) {
 372                 pending_bios = &device->pending_sync_bios;
 373                 force_reg = 1;
 374         } else {
 375                 pending_bios = &device->pending_bios;
 376                 force_reg = 0;
 377         }
 378
 379         pending = pending_bios->head;
 380         tail = pending_bios->tail;
 381         WARN_ON(pending && !tail);
 382
 383         /*
 384          * if pending was null this time around, no bios need processing
 385          * at all and we can stop.  Otherwise it'll loop back up again
 386          * and do an additional check so no bios are missed.
 387          *
 388          * device->running_pending is used to synchronize with the
 389          * schedule_bio code.
 390          */
 391         if (device->pending_sync_bios.head == NULL &&
 392             device->pending_bios.head == NULL) {
 393                 again = 0;
 394                 device->running_pending = 0;
 395         } else {
 396                 again = 1;
 397                 device->running_pending = 1;
 398         }
 399
 400         pending_bios->head = NULL;
 401         pending_bios->tail = NULL;
 402
 403         spin_unlock(&device->io_lock);
 404
 405         while (pending) {
 406
 407                 rmb();
 408                 /* we want to work on both lists, but do more bios on the
 409                  * sync list than the regular list
 410                  */
 411                 if ((num_run > 32 &&
 412                     pending_bios != &device->pending_sync_bios &&
 413                     device->pending_sync_bios.head) ||
 414                    (num_run > 64 && pending_bios == &device->pending_sync_bios &&
 415                     device->pending_bios.head)) {
 416                         spin_lock(&device->io_lock);
 417                         requeue_list(pending_bios, pending, tail);
 418                         goto loop_lock;
 419                 }
 420
 421                 cur = pending;
 422                 pending = pending->bi_next;
 423                 cur->bi_next = NULL;
 424
 425                 /*
 426                  * atomic_dec_return implies a barrier for waitqueue_active
 427                  */
 428                 if (atomic_dec_return(&fs_info->nr_async_bios) < limit &&
 429                     waitqueue_active(&fs_info->async_submit_wait))
 430                         wake_up(&fs_info->async_submit_wait);
 431
 432                 BUG_ON(atomic_read(&cur->__bi_cnt) == 0);
 433
 434                 /*
 435                  * if we're doing the sync list, record that our
 436                  * plug has some sync requests on it
 437                  *
 438                  * If we're doing the regular list and there are
 439                  * sync requests sitting around, unplug before
 440                  * we add more
 441                  */
 442                 if (pending_bios == &device->pending_sync_bios) {
 443                         sync_pending = 1;
 444                 } else if (sync_pending) {
 445                         blk_finish_plug(&plug);
 446                         blk_start_plug(&plug);
 447                         sync_pending = 0;
 448                 }
 449
 450                 btrfsic_submit_bio(cur->bi_rw, cur);
 451                 num_run++;
 452                 batch_run++;
 453
 454                 cond_resched();
 455
 456                 /*
 457                  * we made progress, there is more work to do and the bdi
 458                  * is now congested.  Back off and let other work structs
 459                  * run instead
 460                  */
 461                 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
 462                     fs_info->fs_devices->open_devices > 1) {
 463                         struct io_context *ioc;
 464
 465                         ioc = current->io_context;
 466
 467                         /*
 468                          * the main goal here is that we don't want to
 469                          * block if we're going to be able to submit
 470                          * more requests without blocking.
 471                          *
 472                          * This code does two great things, it pokes into
 473                          * the elevator code from a filesystem _and_
 474                          * it makes assumptions about how batching works.
 475                          */
 476                         if (ioc && ioc->nr_batch_requests > 0 &&
 477                             time_before(jiffies, ioc->last_waited + HZ/50UL) &&
 478                             (last_waited == 0 ||
 479                              ioc->last_waited == last_waited)) {
 480                                 /*
 481                                  * we want to go through our batch of
 482                                  * requests and stop.  So, we copy out
 483                                  * the ioc->last_waited time and test
 484                                  * against it before looping
 485                                  */
 486                                 last_waited = ioc->last_waited;
 487                                 cond_resched();
 488                                 continue;
 489                         }
 490                         spin_lock(&device->io_lock);
 491                         requeue_list(pending_bios, pending, tail);
 492                         device->running_pending = 1;
 493
 494                         spin_unlock(&device->io_lock);
 495                         btrfs_queue_work(fs_info->submit_workers,
 496                                          &device->work);
 497                         goto done;
 498                 }
 499                 /* unplug every 64 requests just for good measure */
 500                 if (batch_run % 64 == 0) {
 501                         blk_finish_plug(&plug);
 502                         blk_start_plug(&plug);
 503                         sync_pending = 0;
 504                 }
 505         }
 506
 507         cond_resched();
 508         if (again)
 509                 goto loop;
 510
 511         spin_lock(&device->io_lock);
 512         if (device->pending_bios.head || device->pending_sync_bios.head)
 513                 goto loop_lock;
 514         spin_unlock(&device->io_lock);
 515
 516 done:
 517         blk_finish_plug(&plug);
 518 }
 519
 520 static void pending_bios_fn(struct btrfs_work *work)
 521 {
 522         struct btrfs_device *device;
 523
 524         device = container_of(work, struct btrfs_device, work);
 525         run_scheduled_bios(device);
 526 }
 527
 528
 529 void btrfs_free_stale_device(struct btrfs_device *cur_dev)
 530 {
 531         struct btrfs_fs_devices *fs_devs;
 532         struct btrfs_device *dev;
 533
 534         if (!cur_dev->name)
 535                 return;
 536
 537         list_for_each_entry(fs_devs, &fs_uuids, list) {
 538                 int del = 1;
 539
 540                 if (fs_devs->opened)
 541                         continue;
 542                 if (fs_devs->seeding)
 543                         continue;
 544
 545                 list_for_each_entry(dev, &fs_devs->devices, dev_list) {
 546
 547                         if (dev == cur_dev)
 548                                 continue;
 549                         if (!dev->name)
 550                                 continue;
 551
 552                         /*
 553                          * Todo: This won't be enough. What if the same device
 554                          * comes back (with new uuid and) with its mapper path?
 555                          * But for now, this does help as mostly an admin will
 556                          * either use mapper or non mapper path throughout.
 557                          */
 558                         rcu_read_lock();
 559                         del = strcmp(rcu_str_deref(dev->name),
 560                                                 rcu_str_deref(cur_dev->name));
 561                         rcu_read_unlock();
 562                         if (!del)
 563                                 break;
 564                 }
 565
 566                 if (!del) {
 567                         /* delete the stale device */
 568                         if (fs_devs->num_devices == 1) {
 569                                 btrfs_sysfs_remove_fsid(fs_devs);
 570                                 list_del(&fs_devs->list);
 571                                 free_fs_devices(fs_devs);
 572                         } else {
 573                                 fs_devs->num_devices--;
 574                                 list_del(&dev->dev_list);
 575                                 rcu_string_free(dev->name);
 576                                 kfree(dev);
 577                         }
 578                         break;
 579                 }
 580         }
 581 }
 582
 583 /*
 584  * Add new device to list of registered devices
 585  *
 586  * Returns:
 587  * 1   - first time device is seen
 588  * 0   - device already known
 589  * < 0 - error
 590  */
 591 static noinline int device_list_add(const char *path,
 592                            struct btrfs_super_block *disk_super,
 593                            u64 devid, struct btrfs_fs_devices **fs_devices_ret)
 594 {
 595         struct btrfs_device *device;
 596         struct btrfs_fs_devices *fs_devices;
 597         struct rcu_string *name;
 598         int ret = 0;
 599         u64 found_transid = btrfs_super_generation(disk_super);
 600
 601         fs_devices = find_fsid(disk_super->fsid);
 602         if (!fs_devices) {
 603                 fs_devices = alloc_fs_devices(disk_super->fsid);
 604                 if (IS_ERR(fs_devices))
 605                         return PTR_ERR(fs_devices);
 606
 607                 list_add(&fs_devices->list, &fs_uuids);
 608
 609                 device = NULL;
 610         } else {
 611                 device = __find_device(&fs_devices->devices, devid,
 612                                        disk_super->dev_item.uuid);
 613         }
 614
 615         if (!device) {
 616                 if (fs_devices->opened)
 617                         return -EBUSY;
 618
 619                 device = btrfs_alloc_device(NULL, &devid,
 620                                             disk_super->dev_item.uuid);
 621                 if (IS_ERR(device)) {
 622                         /* we can safely leave the fs_devices entry around */
 623                         return PTR_ERR(device);
 624                 }
 625
 626                 name = rcu_string_strdup(path, GFP_NOFS);
 627                 if (!name) {
 628                         kfree(device);
 629                         return -ENOMEM;
 630                 }
 631                 rcu_assign_pointer(device->name, name);
 632
 633                 mutex_lock(&fs_devices->device_list_mutex);
 634                 list_add_rcu(&device->dev_list, &fs_devices->devices);
 635                 fs_devices->num_devices++;
 636                 mutex_unlock(&fs_devices->device_list_mutex);
 637
 638                 ret = 1;
 639                 device->fs_devices = fs_devices;
 640         } else if (!device->name || strcmp(device->name->str, path)) {
 641                 /*
 642                  * When FS is already mounted.
 643                  * 1. If you are here and if the device->name is NULL that
 644                  *    means this device was missing at time of FS mount.
 645                  * 2. If you are here and if the device->name is different
 646                  *    from 'path' that means either
 647                  *      a. The same device disappeared and reappeared with
 648                  *         different name. or
 649                  *      b. The missing-disk-which-was-replaced, has
 650                  *         reappeared now.
 651                  *
 652                  * We must allow 1 and 2a above. But 2b would be a spurious
 653                  * and unintentional.
 654                  *
 655                  * Further in case of 1 and 2a above, the disk at 'path'
 656                  * would have missed some transaction when it was away and
 657                  * in case of 2a the stale bdev has to be updated as well.
 658                  * 2b must not be allowed at all time.
 659                  */
 660
 661                 /*
 662                  * For now, we do allow update to btrfs_fs_device through the
 663                  * btrfs dev scan cli after FS has been mounted.  We're still
 664                  * tracking a problem where systems fail mount by subvolume id
 665                  * when we reject replacement on a mounted FS.
 666                  */
 667                 if (!fs_devices->opened && found_transid < device->generation) {
 668                         /*
 669                          * That is if the FS is _not_ mounted and if you
 670                          * are here, that means there is more than one
 671                          * disk with same uuid and devid.We keep the one
 672                          * with larger generation number or the last-in if
 673                          * generation are equal.
 674                          */
 675                         return -EEXIST;
 676                 }
 677
 678                 name = rcu_string_strdup(path, GFP_NOFS);
 679                 if (!name)
 680                         return -ENOMEM;
 681                 rcu_string_free(device->name);
 682                 rcu_assign_pointer(device->name, name);
 683                 if (device->missing) {
 684                         fs_devices->missing_devices--;
 685                         device->missing = 0;
 686                 }
 687         }
 688
 689         /*
 690          * Unmount does not free the btrfs_device struct but would zero
 691          * generation along with most of the other members. So just update
 692          * it back. We need it to pick the disk with largest generation
 693          * (as above).
 694          */
 695         if (!fs_devices->opened)
 696                 device->generation = found_transid;
 697
 698         /*
 699          * if there is new btrfs on an already registered device,
 700          * then remove the stale device entry.
 701          */
 702         btrfs_free_stale_device(device);
 703
 704         *fs_devices_ret = fs_devices;
 705
 706         return ret;
 707 }
 708
 709 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
 710 {
 711         struct btrfs_fs_devices *fs_devices;
 712         struct btrfs_device *device;
 713         struct btrfs_device *orig_dev;
 714
 715         fs_devices = alloc_fs_devices(orig->fsid);
 716         if (IS_ERR(fs_devices))
 717                 return fs_devices;
 718
 719         mutex_lock(&orig->device_list_mutex);
 720         fs_devices->total_devices = orig->total_devices;
 721
 722         /* We have held the volume lock, it is safe to get the devices. */
 723         list_for_each_entry(orig_dev, &orig->devices, dev_list) {
 724                 struct rcu_string *name;
 725
 726                 device = btrfs_alloc_device(NULL, &orig_dev->devid,
 727                                             orig_dev->uuid);
 728                 if (IS_ERR(device))
 729                         goto error;
 730
 731                 /*
 732                  * This is ok to do without rcu read locked because we hold the
 733                  * uuid mutex so nothing we touch in here is going to disappear.
 734                  */
 735                 if (orig_dev->name) {
 736                         name = rcu_string_strdup(orig_dev->name->str,
 737                                         GFP_KERNEL);
 738                         if (!name) {
 739                                 kfree(device);
 740                                 goto error;
 741                         }
 742                         rcu_assign_pointer(device->name, name);
 743                 }
 744
 745                 list_add(&device->dev_list, &fs_devices->devices);
 746                 device->fs_devices = fs_devices;
 747                 fs_devices->num_devices++;
 748         }
 749         mutex_unlock(&orig->device_list_mutex);
 750         return fs_devices;
 751 error:
 752         mutex_unlock(&orig->device_list_mutex);
 753         free_fs_devices(fs_devices);
 754         return ERR_PTR(-ENOMEM);
 755 }
 756
 757 void btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices, int step)
 758 {
 759         struct btrfs_device *device, *next;
 760         struct btrfs_device *latest_dev = NULL;
 761
 762         mutex_lock(&uuid_mutex);
 763 again:
 764         /* This is the initialized path, it is safe to release the devices. */
 765         list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
 766                 if (device->in_fs_metadata) {
 767                         if (!device->is_tgtdev_for_dev_replace &&
 768                             (!latest_dev ||
 769                              device->generation > latest_dev->generation)) {
 770                                 latest_dev = device;
 771                         }
 772                         continue;
 773                 }
 774
 775                 if (device->devid == BTRFS_DEV_REPLACE_DEVID) {
 776                         /*
 777                          * In the first step, keep the device which has
 778                          * the correct fsid and the devid that is used
 779                          * for the dev_replace procedure.
 780                          * In the second step, the dev_replace state is
 781                          * read from the device tree and it is known
 782                          * whether the procedure is really active or
 783                          * not, which means whether this device is
 784                          * used or whether it should be removed.
 785                          */
 786                         if (step == 0 || device->is_tgtdev_for_dev_replace) {
 787                                 continue;
 788                         }
 789                 }
 790                 if (device->bdev) {
 791                         blkdev_put(device->bdev, device->mode);
 792                         device->bdev = NULL;
 793                         fs_devices->open_devices--;
 794                 }
 795                 if (device->writeable) {
 796                         list_del_init(&device->dev_alloc_list);
 797                         device->writeable = 0;
 798                         if (!device->is_tgtdev_for_dev_replace)
 799                                 fs_devices->rw_devices--;
 800                 }
 801                 list_del_init(&device->dev_list);
 802                 fs_devices->num_devices--;
 803                 rcu_string_free(device->name);
 804                 kfree(device);
 805         }
 806
 807         if (fs_devices->seed) {
 808                 fs_devices = fs_devices->seed;
 809                 goto again;
 810         }
 811
 812         fs_devices->latest_bdev = latest_dev->bdev;
 813
 814         mutex_unlock(&uuid_mutex);
 815 }
 816
 817 static void __free_device(struct work_struct *work)
 818 {
 819         struct btrfs_device *device;
 820
 821         device = container_of(work, struct btrfs_device, rcu_work);
 822
 823         if (device->bdev)
 824                 blkdev_put(device->bdev, device->mode);
 825
 826         rcu_string_free(device->name);
 827         kfree(device);
 828 }
 829
 830 static void free_device(struct rcu_head *head)
 831 {
 832         struct btrfs_device *device;
 833
 834         device = container_of(head, struct btrfs_device, rcu);
 835
 836         INIT_WORK(&device->rcu_work, __free_device);
 837         schedule_work(&device->rcu_work);
 838 }
 839
 840 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
 841 {
 842         struct btrfs_device *device, *tmp;
 843
 844         if (--fs_devices->opened > 0)
 845                 return 0;
 846
 847         mutex_lock(&fs_devices->device_list_mutex);
 848         list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list) {
 849                 btrfs_close_one_device(device);
 850         }
 851         mutex_unlock(&fs_devices->device_list_mutex);
 852
 853         WARN_ON(fs_devices->open_devices);
 854         WARN_ON(fs_devices->rw_devices);
 855         fs_devices->opened = 0;
 856         fs_devices->seeding = 0;
 857
 858         return 0;
 859 }
 860
 861 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
 862 {
 863         struct btrfs_fs_devices *seed_devices = NULL;
 864         int ret;
 865
 866         mutex_lock(&uuid_mutex);
 867         ret = __btrfs_close_devices(fs_devices);
 868         if (!fs_devices->opened) {
 869                 seed_devices = fs_devices->seed;
 870                 fs_devices->seed = NULL;
 871         }
 872         mutex_unlock(&uuid_mutex);
 873
 874         while (seed_devices) {
 875                 fs_devices = seed_devices;
 876                 seed_devices = fs_devices->seed;
 877                 __btrfs_close_devices(fs_devices);
 878                 free_fs_devices(fs_devices);
 879         }
 880         /*
 881          * Wait for rcu kworkers under __btrfs_close_devices
 882          * to finish all blkdev_puts so device is really
 883          * free when umount is done.
 884          */
 885         rcu_barrier();
 886         return ret;
 887 }
 888
 889 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
 890                                 fmode_t flags, void *holder)
 891 {
 892         struct request_queue *q;
 893         struct block_device *bdev;
 894         struct list_head *head = &fs_devices->devices;
 895         struct btrfs_device *device;
 896         struct btrfs_device *latest_dev = NULL;
 897         struct buffer_head *bh;
 898         struct btrfs_super_block *disk_super;
 899         u64 devid;
 900         int seeding = 1;
 901         int ret = 0;
 902
 903         flags |= FMODE_EXCL;
 904
 905         list_for_each_entry(device, head, dev_list) {
 906                 if (device->bdev)
 907                         continue;
 908                 if (!device->name)
 909                         continue;
 910
 911                 /* Just open everything we can; ignore failures here */
 912                 if (btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1,
 913                                             &bdev, &bh))
 914                         continue;
 915
 916                 disk_super = (struct btrfs_super_block *)bh->b_data;
 917                 devid = btrfs_stack_device_id(&disk_super->dev_item);
 918                 if (devid != device->devid)
 919                         goto error_brelse;
 920
 921                 if (memcmp(device->uuid, disk_super->dev_item.uuid,
 922                            BTRFS_UUID_SIZE))
 923                         goto error_brelse;
 924
 925                 device->generation = btrfs_super_generation(disk_super);
 926                 if (!latest_dev ||
 927                     device->generation > latest_dev->generation)
 928                         latest_dev = device;
 929
 930                 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
 931                         device->writeable = 0;
 932                 } else {
 933                         device->writeable = !bdev_read_only(bdev);
 934                         seeding = 0;
 935                 }
 936
 937                 q = bdev_get_queue(bdev);
 938                 if (blk_queue_discard(q))
 939                         device->can_discard = 1;
 940
 941                 device->bdev = bdev;
 942                 device->in_fs_metadata = 0;
 943                 device->mode = flags;
 944
 945                 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
 946                         fs_devices->rotating = 1;
 947
 948                 fs_devices->open_devices++;
 949                 if (device->writeable &&
 950                     device->devid != BTRFS_DEV_REPLACE_DEVID) {
 951                         fs_devices->rw_devices++;
 952                         list_add(&device->dev_alloc_list,
 953                                  &fs_devices->alloc_list);
 954                 }
 955                 brelse(bh);
 956                 continue;
 957
 958 error_brelse:
 959                 brelse(bh);
 960                 blkdev_put(bdev, flags);
 961                 continue;
 962         }
 963         if (fs_devices->open_devices == 0) {
 964                 ret = -EINVAL;
 965                 goto out;
 966         }
 967         fs_devices->seeding = seeding;
 968         fs_devices->opened = 1;
 969         fs_devices->latest_bdev = latest_dev->bdev;
 970         fs_devices->total_rw_bytes = 0;
 971 out:
 972         return ret;
 973 }
 974
 975 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
 976                        fmode_t flags, void *holder)
 977 {
 978         int ret;
 979
 980         mutex_lock(&uuid_mutex);
 981         if (fs_devices->opened) {
 982                 fs_devices->opened++;
 983                 ret = 0;
 984         } else {
 985                 ret = __btrfs_open_devices(fs_devices, flags, holder);
 986         }
 987         mutex_unlock(&uuid_mutex);
 988         return ret;
 989 }
 990
 991 void btrfs_release_disk_super(struct page *page)
 992 {
 993         kunmap(page);
 994         put_page(page);
 995 }
 996
 997 int btrfs_read_disk_super(struct block_device *bdev, u64 bytenr,
 998                 struct page **page, struct btrfs_super_block **disk_super)
 999 {
1000         void *p;
1001         pgoff_t index;
1002
1003         /* make sure our super fits in the device */
1004         if (bytenr + PAGE_SIZE >= i_size_read(bdev->bd_inode))
1005                 return 1;
1006
1007         /* make sure our super fits in the page */
1008         if (sizeof(**disk_super) > PAGE_SIZE)
1009                 return 1;
1010
1011         /* make sure our super doesn't straddle pages on disk */
1012         index = bytenr >> PAGE_SHIFT;
1013         if ((bytenr + sizeof(**disk_super) - 1) >> PAGE_SHIFT != index)
1014                 return 1;
1015
1016         /* pull in the page with our super */
1017         *page = read_cache_page_gfp(bdev->bd_inode->i_mapping,
1018                                    index, GFP_KERNEL);
1019
1020         if (IS_ERR_OR_NULL(*page))
1021                 return 1;
1022
1023         p = kmap(*page);
1024
1025         /* align our pointer to the offset of the super block */
1026         *disk_super = p + (bytenr & ~PAGE_MASK);
1027
1028         if (btrfs_super_bytenr(*disk_super) != bytenr ||
1029             btrfs_super_magic(*disk_super) != BTRFS_MAGIC) {
1030                 btrfs_release_disk_super(*page);
1031                 return 1;
1032         }
1033
1034         if ((*disk_super)->label[0] &&
1035                 (*disk_super)->label[BTRFS_LABEL_SIZE - 1])
1036                 (*disk_super)->label[BTRFS_LABEL_SIZE - 1] = '\0';
1037
1038         return 0;
1039 }
1040
1041 /*
1042  * Look for a btrfs signature on a device. This may be called out of the mount path
1043  * and we are not allowed to call set_blocksize during the scan. The superblock
1044  * is read via pagecache
1045  */
1046 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
1047                           struct btrfs_fs_devices **fs_devices_ret)
1048 {
1049         struct btrfs_super_block *disk_super;
1050         struct block_device *bdev;
1051         struct page *page;
1052         int ret = -EINVAL;
1053         u64 devid;
1054         u64 transid;
1055         u64 total_devices;
1056         u64 bytenr;
1057
1058         /*
1059          * we would like to check all the supers, but that would make
1060          * a btrfs mount succeed after a mkfs from a different FS.
1061          * So, we need to add a special mount option to scan for
1062          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
1063          */
1064         bytenr = btrfs_sb_offset(0);
1065         flags |= FMODE_EXCL;
1066         mutex_lock(&uuid_mutex);
1067
1068         bdev = blkdev_get_by_path(path, flags, holder);
1069         if (IS_ERR(bdev)) {
1070                 ret = PTR_ERR(bdev);
1071                 goto error;
1072         }
1073
1074         if (btrfs_read_disk_super(bdev, bytenr, &page, &disk_super))
1075                 goto error_bdev_put;
1076
1077         devid = btrfs_stack_device_id(&disk_super->dev_item);
1078         transid = btrfs_super_generation(disk_super);
1079         total_devices = btrfs_super_num_devices(disk_super);
1080
1081         ret = device_list_add(path, disk_super, devid, fs_devices_ret);
1082         if (ret > 0) {
1083                 if (disk_super->label[0]) {
1084                         printk(KERN_INFO "BTRFS: device label %s ", disk_super->label);
1085                 } else {
1086                         printk(KERN_INFO "BTRFS: device fsid %pU ", disk_super->fsid);
1087                 }
1088
1089                 printk(KERN_CONT "devid %llu transid %llu %s\n", devid, transid, path);
1090                 ret = 0;
1091         }
1092         if (!ret && fs_devices_ret)
1093                 (*fs_devices_ret)->total_devices = total_devices;
1094
1095         btrfs_release_disk_super(page);
1096
1097 error_bdev_put:
1098         blkdev_put(bdev, flags);
1099 error:
1100         mutex_unlock(&uuid_mutex);
1101         return ret;
1102 }
1103
1104 /* helper to account the used device space in the range */
1105 int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
1106                                    u64 end, u64 *length)
1107 {
1108         struct btrfs_key key;
1109         struct btrfs_root *root = device->dev_root;
1110         struct btrfs_dev_extent *dev_extent;
1111         struct btrfs_path *path;
1112         u64 extent_end;
1113         int ret;
1114         int slot;
1115         struct extent_buffer *l;
1116
1117         *length = 0;
1118
1119         if (start >= device->total_bytes || device->is_tgtdev_for_dev_replace)
1120                 return 0;
1121
1122         path = btrfs_alloc_path();
1123         if (!path)
1124                 return -ENOMEM;
1125         path->reada = READA_FORWARD;
1126
1127         key.objectid = device->devid;
1128         key.offset = start;
1129         key.type = BTRFS_DEV_EXTENT_KEY;
1130
1131         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1132         if (ret < 0)
1133                 goto out;
1134         if (ret > 0) {
1135                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1136                 if (ret < 0)
1137                         goto out;
1138         }
1139
1140         while (1) {
1141                 l = path->nodes[0];
1142                 slot = path->slots[0];
1143                 if (slot >= btrfs_header_nritems(l)) {
1144                         ret = btrfs_next_leaf(root, path);
1145                         if (ret == 0)
1146                                 continue;
1147                         if (ret < 0)
1148                                 goto out;
1149
1150                         break;
1151                 }
1152                 btrfs_item_key_to_cpu(l, &key, slot);
1153
1154                 if (key.objectid < device->devid)
1155                         goto next;
1156
1157                 if (key.objectid > device->devid)
1158                         break;
1159
1160                 if (key.type != BTRFS_DEV_EXTENT_KEY)
1161                         goto next;
1162
1163                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1164                 extent_end = key.offset + btrfs_dev_extent_length(l,
1165                                                                   dev_extent);
1166                 if (key.offset <= start && extent_end > end) {
1167                         *length = end - start + 1;
1168                         break;
1169                 } else if (key.offset <= start && extent_end > start)
1170                         *length += extent_end - start;
1171                 else if (key.offset > start && extent_end <= end)
1172                         *length += extent_end - key.offset;
1173                 else if (key.offset > start && key.offset <= end) {
1174                         *length += end - key.offset + 1;
1175                         break;
1176                 } else if (key.offset > end)
1177                         break;
1178
1179 next:
1180                 path->slots[0]++;
1181         }
1182         ret = 0;
1183 out:
1184         btrfs_free_path(path);
1185         return ret;
1186 }
1187
1188 static int contains_pending_extent(struct btrfs_transaction *transaction,
1189                                    struct btrfs_device *device,
1190                                    u64 *start, u64 len)
1191 {
1192         struct btrfs_fs_info *fs_info = device->dev_root->fs_info;
1193         struct extent_map *em;
1194         struct list_head *search_list = &fs_info->pinned_chunks;
1195         int ret = 0;
1196         u64 physical_start = *start;
1197
1198         if (transaction)
1199                 search_list = &transaction->pending_chunks;
1200 again:
1201         list_for_each_entry(em, search_list, list) {
1202                 struct map_lookup *map;
1203                 int i;
1204
1205                 map = em->map_lookup;
1206                 for (i = 0; i < map->num_stripes; i++) {
1207                         u64 end;
1208
1209                         if (map->stripes[i].dev != device)
1210                                 continue;
1211                         if (map->stripes[i].physical >= physical_start + len ||
1212                             map->stripes[i].physical + em->orig_block_len <=
1213                             physical_start)
1214                                 continue;
1215                         /*
1216                          * Make sure that while processing the pinned list we do
1217                          * not override our *start with a lower value, because
1218                          * we can have pinned chunks that fall within this
1219                          * device hole and that have lower physical addresses
1220                          * than the pending chunks we processed before. If we
1221                          * do not take this special care we can end up getting
1222                          * 2 pending chunks that start at the same physical
1223                          * device offsets because the end offset of a pinned
1224                          * chunk can be equal to the start offset of some
1225                          * pending chunk.
1226                          */
1227                         end = map->stripes[i].physical + em->orig_block_len;
1228                         if (end > *start) {
1229                                 *start = end;
1230                                 ret = 1;
1231                         }
1232                 }
1233         }
1234         if (search_list != &fs_info->pinned_chunks) {
1235                 search_list = &fs_info->pinned_chunks;
1236                 goto again;
1237         }
1238
1239         return ret;
1240 }
1241
1242
1243 /*
1244  * find_free_dev_extent_start - find free space in the specified device
1245  * @device:       the device which we search the free space in
1246  * @num_bytes:    the size of the free space that we need
1247  * @search_start: the position from which to begin the search
1248  * @start:        store the start of the free space.
1249  * @len:          the size of the free space. that we find, or the size
1250  *                of the max free space if we don't find suitable free space
1251  *
1252  * this uses a pretty simple search, the expectation is that it is
1253  * called very infrequently and that a given device has a small number
1254  * of extents
1255  *
1256  * @start is used to store the start of the free space if we find. But if we
1257  * don't find suitable free space, it will be used to store the start position
1258  * of the max free space.
1259  *
1260  * @len is used to store the size of the free space that we find.
1261  * But if we don't find suitable free space, it is used to store the size of
1262  * the max free space.
1263  */
1264 int find_free_dev_extent_start(struct btrfs_transaction *transaction,
1265                                struct btrfs_device *device, u64 num_bytes,
1266                                u64 search_start, u64 *start, u64 *len)
1267 {
1268         struct btrfs_key key;
1269         struct btrfs_root *root = device->dev_root;
1270         struct btrfs_dev_extent *dev_extent;
1271         struct btrfs_path *path;
1272         u64 hole_size;
1273         u64 max_hole_start;
1274         u64 max_hole_size;
1275         u64 extent_end;
1276         u64 search_end = device->total_bytes;
1277         int ret;
1278         int slot;
1279         struct extent_buffer *l;
1280         u64 min_search_start;
1281
1282         /*
1283          * We don't want to overwrite the superblock on the drive nor any area
1284          * used by the boot loader (grub for example), so we make sure to start
1285          * at an offset of at least 1MB.
1286          */
1287         min_search_start = max(root->fs_info->alloc_start, 1024ull * 1024);
1288         search_start = max(search_start, min_search_start);
1289
1290         path = btrfs_alloc_path();
1291         if (!path)
1292                 return -ENOMEM;
1293
1294         max_hole_start = search_start;
1295         max_hole_size = 0;
1296
1297 again:
1298         if (search_start >= search_end || device->is_tgtdev_for_dev_replace) {
1299                 ret = -ENOSPC;
1300                 goto out;
1301         }
1302
1303         path->reada = READA_FORWARD;
1304         path->search_commit_root = 1;
1305         path->skip_locking = 1;
1306
1307         key.objectid = device->devid;
1308         key.offset = search_start;
1309         key.type = BTRFS_DEV_EXTENT_KEY;
1310
1311         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1312         if (ret < 0)
1313                 goto out;
1314         if (ret > 0) {
1315                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1316                 if (ret < 0)
1317                         goto out;
1318         }
1319
1320         while (1) {
1321                 l = path->nodes[0];
1322                 slot = path->slots[0];
1323                 if (slot >= btrfs_header_nritems(l)) {
1324                         ret = btrfs_next_leaf(root, path);
1325                         if (ret == 0)
1326                                 continue;
1327                         if (ret < 0)
1328                                 goto out;
1329
1330                         break;
1331                 }
1332                 btrfs_item_key_to_cpu(l, &key, slot);
1333
1334                 if (key.objectid < device->devid)
1335                         goto next;
1336
1337                 if (key.objectid > device->devid)
1338                         break;
1339
1340                 if (key.type != BTRFS_DEV_EXTENT_KEY)
1341                         goto next;
1342
1343                 if (key.offset > search_start) {
1344                         hole_size = key.offset - search_start;
1345
1346                         /*
1347                          * Have to check before we set max_hole_start, otherwise
1348                          * we could end up sending back this offset anyway.
1349                          */
1350                         if (contains_pending_extent(transaction, device,
1351                                                     &search_start,
1352                                                     hole_size)) {
1353                                 if (key.offset >= search_start) {
1354                                         hole_size = key.offset - search_start;
1355                                 } else {
1356                                         WARN_ON_ONCE(1);
1357                                         hole_size = 0;
1358                                 }
1359                         }
1360
1361                         if (hole_size > max_hole_size) {
1362                                 max_hole_start = search_start;
1363                                 max_hole_size = hole_size;
1364                         }
1365
1366                         /*
1367                          * If this free space is greater than which we need,
1368                          * it must be the max free space that we have found
1369                          * until now, so max_hole_start must point to the start
1370                          * of this free space and the length of this free space
1371                          * is stored in max_hole_size. Thus, we return
1372                          * max_hole_start and max_hole_size and go back to the
1373                          * caller.
1374                          */
1375                         if (hole_size >= num_bytes) {
1376                                 ret = 0;
1377                                 goto out;
1378                         }
1379                 }
1380
1381                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1382                 extent_end = key.offset + btrfs_dev_extent_length(l,
1383                                                                   dev_extent);
1384                 if (extent_end > search_start)
1385                         search_start = extent_end;
1386 next:
1387                 path->slots[0]++;
1388                 cond_resched();
1389         }
1390
1391         /*
1392          * At this point, search_start should be the end of
1393          * allocated dev extents, and when shrinking the device,
1394          * search_end may be smaller than search_start.
1395          */
1396         if (search_end > search_start) {
1397                 hole_size = search_end - search_start;
1398
1399                 if (contains_pending_extent(transaction, device, &search_start,
1400                                             hole_size)) {
1401                         btrfs_release_path(path);
1402                         goto again;
1403                 }
1404
1405                 if (hole_size > max_hole_size) {
1406                         max_hole_start = search_start;
1407                         max_hole_size = hole_size;
1408                 }
1409         }
1410
1411         /* See above. */
1412         if (max_hole_size < num_bytes)
1413                 ret = -ENOSPC;
1414         else
1415                 ret = 0;
1416
1417 out:
1418         btrfs_free_path(path);
1419         *start = max_hole_start;
1420         if (len)
1421                 *len = max_hole_size;
1422         return ret;
1423 }
1424
1425 int find_free_dev_extent(struct btrfs_trans_handle *trans,
1426                          struct btrfs_device *device, u64 num_bytes,
1427                          u64 *start, u64 *len)
1428 {
1429         /* FIXME use last free of some kind */
1430         return find_free_dev_extent_start(trans->transaction, device,
1431                                           num_bytes, 0, start, len);
1432 }
1433
1434 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1435                           struct btrfs_device *device,
1436                           u64 start, u64 *dev_extent_len)
1437 {
1438         int ret;
1439         struct btrfs_path *path;
1440         struct btrfs_root *root = device->dev_root;
1441         struct btrfs_key key;
1442         struct btrfs_key found_key;
1443         struct extent_buffer *leaf = NULL;
1444         struct btrfs_dev_extent *extent = NULL;
1445
1446         path = btrfs_alloc_path();
1447         if (!path)
1448                 return -ENOMEM;
1449
1450         key.objectid = device->devid;
1451         key.offset = start;
1452         key.type = BTRFS_DEV_EXTENT_KEY;
1453 again:
1454         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1455         if (ret > 0) {
1456                 ret = btrfs_previous_item(root, path, key.objectid,
1457                                           BTRFS_DEV_EXTENT_KEY);
1458                 if (ret)
1459                         goto out;
1460                 leaf = path->nodes[0];
1461                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1462                 extent = btrfs_item_ptr(leaf, path->slots[0],
1463                                         struct btrfs_dev_extent);
1464                 BUG_ON(found_key.offset > start || found_key.offset +
1465                        btrfs_dev_extent_length(leaf, extent) < start);
1466                 key = found_key;
1467                 btrfs_release_path(path);
1468                 goto again;
1469         } else if (ret == 0) {
1470                 leaf = path->nodes[0];
1471                 extent = btrfs_item_ptr(leaf, path->slots[0],
1472                                         struct btrfs_dev_extent);
1473         } else {
1474                 btrfs_std_error(root->fs_info, ret, "Slot search failed");
1475                 goto out;
1476         }
1477
1478         *dev_extent_len = btrfs_dev_extent_length(leaf, extent);
1479
1480         ret = btrfs_del_item(trans, root, path);
1481         if (ret) {
1482                 btrfs_std_error(root->fs_info, ret,
1483                             "Failed to remove dev extent item");
1484         } else {
1485                 set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags);
1486         }
1487 out:
1488         btrfs_free_path(path);
1489         return ret;
1490 }
1491
1492 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1493                                   struct btrfs_device *device,
1494                                   u64 chunk_tree, u64 chunk_objectid,
1495                                   u64 chunk_offset, u64 start, u64 num_bytes)
1496 {
1497         int ret;
1498         struct btrfs_path *path;
1499         struct btrfs_root *root = device->dev_root;
1500         struct btrfs_dev_extent *extent;
1501         struct extent_buffer *leaf;
1502         struct btrfs_key key;
1503
1504         WARN_ON(!device->in_fs_metadata);
1505         WARN_ON(device->is_tgtdev_for_dev_replace);
1506         path = btrfs_alloc_path();
1507         if (!path)
1508                 return -ENOMEM;
1509
1510         key.objectid = device->devid;
1511         key.offset = start;
1512         key.type = BTRFS_DEV_EXTENT_KEY;
1513         ret = btrfs_insert_empty_item(trans, root, path, &key,
1514                                       sizeof(*extent));
1515         if (ret)
1516                 goto out;
1517
1518         leaf = path->nodes[0];
1519         extent = btrfs_item_ptr(leaf, path->slots[0],
1520                                 struct btrfs_dev_extent);
1521         btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
1522         btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
1523         btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1524
1525         write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
1526                     btrfs_dev_extent_chunk_tree_uuid(extent), BTRFS_UUID_SIZE);
1527
1528         btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1529         btrfs_mark_buffer_dirty(leaf);
1530 out:
1531         btrfs_free_path(path);
1532         return ret;
1533 }
1534
1535 static u64 find_next_chunk(struct btrfs_fs_info *fs_info)
1536 {
1537         struct extent_map_tree *em_tree;
1538         struct extent_map *em;
1539         struct rb_node *n;
1540         u64 ret = 0;
1541
1542         em_tree = &fs_info->mapping_tree.map_tree;
1543         read_lock(&em_tree->lock);
1544         n = rb_last(&em_tree->map);
1545         if (n) {
1546                 em = rb_entry(n, struct extent_map, rb_node);
1547                 ret = em->start + em->len;
1548         }
1549         read_unlock(&em_tree->lock);
1550
1551         return ret;
1552 }
1553
1554 static noinline int find_next_devid(struct btrfs_fs_info *fs_info,
1555                                     u64 *devid_ret)
1556 {
1557         int ret;
1558         struct btrfs_key key;
1559         struct btrfs_key found_key;
1560         struct btrfs_path *path;
1561
1562         path = btrfs_alloc_path();
1563         if (!path)
1564                 return -ENOMEM;
1565
1566         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1567         key.type = BTRFS_DEV_ITEM_KEY;
1568         key.offset = (u64)-1;
1569
1570         ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
1571         if (ret < 0)
1572                 goto error;
1573
1574         BUG_ON(ret == 0); /* Corruption */
1575
1576         ret = btrfs_previous_item(fs_info->chunk_root, path,
1577                                   BTRFS_DEV_ITEMS_OBJECTID,
1578                                   BTRFS_DEV_ITEM_KEY);
1579         if (ret) {
1580                 *devid_ret = 1;
1581         } else {
1582                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1583                                       path->slots[0]);
1584                 *devid_ret = found_key.offset + 1;
1585         }
1586         ret = 0;
1587 error:
1588         btrfs_free_path(path);
1589         return ret;
1590 }
1591
1592 /*
1593  * the device information is stored in the chunk root
1594  * the btrfs_device struct should be fully filled in
1595  */
1596 static int btrfs_add_device(struct btrfs_trans_handle *trans,
1597                             struct btrfs_root *root,
1598                             struct btrfs_device *device)
1599 {
1600         int ret;
1601         struct btrfs_path *path;
1602         struct btrfs_dev_item *dev_item;
1603         struct extent_buffer *leaf;
1604         struct btrfs_key key;
1605         unsigned long ptr;
1606
1607         root = root->fs_info->chunk_root;
1608
1609         path = btrfs_alloc_path();
1610         if (!path)
1611                 return -ENOMEM;
1612
1613         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1614         key.type = BTRFS_DEV_ITEM_KEY;
1615         key.offset = device->devid;
1616
1617         ret = btrfs_insert_empty_item(trans, root, path, &key,
1618                                       sizeof(*dev_item));
1619         if (ret)
1620                 goto out;
1621
1622         leaf = path->nodes[0];
1623         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1624
1625         btrfs_set_device_id(leaf, dev_item, device->devid);
1626         btrfs_set_device_generation(leaf, dev_item, 0);
1627         btrfs_set_device_type(leaf, dev_item, device->type);
1628         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1629         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1630         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1631         btrfs_set_device_total_bytes(leaf, dev_item,
1632                                      btrfs_device_get_disk_total_bytes(device));
1633         btrfs_set_device_bytes_used(leaf, dev_item,
1634                                     btrfs_device_get_bytes_used(device));
1635         btrfs_set_device_group(leaf, dev_item, 0);
1636         btrfs_set_device_seek_speed(leaf, dev_item, 0);
1637         btrfs_set_device_bandwidth(leaf, dev_item, 0);
1638         btrfs_set_device_start_offset(leaf, dev_item, 0);
1639
1640         ptr = btrfs_device_uuid(dev_item);
1641         write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1642         ptr = btrfs_device_fsid(dev_item);
1643         write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1644         btrfs_mark_buffer_dirty(leaf);
1645
1646         ret = 0;
1647 out:
1648         btrfs_free_path(path);
1649         return ret;
1650 }
1651
1652 /*
1653  * Function to update ctime/mtime for a given device path.
1654  * Mainly used for ctime/mtime based probe like libblkid.
1655  */
1656 static void update_dev_time(char *path_name)
1657 {
1658         struct file *filp;
1659
1660         filp = filp_open(path_name, O_RDWR, 0);
1661         if (IS_ERR(filp))
1662                 return;
1663         file_update_time(filp);
1664         filp_close(filp, NULL);
1665 }
1666
1667 static int btrfs_rm_dev_item(struct btrfs_root *root,
1668                              struct btrfs_device *device)
1669 {
1670         int ret;
1671         struct btrfs_path *path;
1672         struct btrfs_key key;
1673         struct btrfs_trans_handle *trans;
1674
1675         root = root->fs_info->chunk_root;
1676
1677         path = btrfs_alloc_path();
1678         if (!path)
1679                 return -ENOMEM;
1680
1681         trans = btrfs_start_transaction(root, 0);
1682         if (IS_ERR(trans)) {
1683                 btrfs_free_path(path);
1684                 return PTR_ERR(trans);
1685         }
1686         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1687         key.type = BTRFS_DEV_ITEM_KEY;
1688         key.offset = device->devid;
1689
1690         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1691         if (ret < 0)
1692                 goto out;
1693
1694         if (ret > 0) {
1695                 ret = -ENOENT;
1696                 goto out;
1697         }
1698
1699         ret = btrfs_del_item(trans, root, path);
1700         if (ret)
1701                 goto out;
1702 out:
1703         btrfs_free_path(path);
1704         btrfs_commit_transaction(trans, root);
1705         return ret;
1706 }
1707
1708 static int __check_raid_min_devices(struct btrfs_fs_info *fs_info)
1709 {
1710         u64 all_avail;
1711         u64 num_devices;
1712         unsigned seq;
1713
1714         num_devices = fs_info->fs_devices->num_devices;
1715         btrfs_dev_replace_lock(&fs_info->dev_replace, 0);
1716         if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
1717                 WARN_ON(num_devices < 1);
1718                 num_devices--;
1719         }
1720         btrfs_dev_replace_unlock(&fs_info->dev_replace, 0);
1721
1722         do {
1723                 seq = read_seqbegin(&fs_info->profiles_lock);
1724
1725                 all_avail = fs_info->avail_data_alloc_bits |
1726                             fs_info->avail_system_alloc_bits |
1727                             fs_info->avail_metadata_alloc_bits;
1728         } while (read_seqretry(&fs_info->profiles_lock, seq));
1729
1730         if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) && num_devices <= 4) {
1731                 return BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET;
1732         }
1733
1734         if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) && num_devices <= 2) {
1735                 return BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET;
1736         }
1737
1738         if ((all_avail & BTRFS_BLOCK_GROUP_RAID5) &&
1739             fs_info->fs_devices->rw_devices <= 2) {
1740                 return BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET;
1741         }
1742
1743         if ((all_avail & BTRFS_BLOCK_GROUP_RAID6) &&
1744             fs_info->fs_devices->rw_devices <= 3) {
1745                 return BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET;
1746         }
1747
1748         return 0;
1749 }
1750
1751 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
1752 {
1753         struct btrfs_device *device;
1754         struct btrfs_device *next_device;
1755         struct block_device *bdev;
1756         struct buffer_head *bh = NULL;
1757         struct btrfs_super_block *disk_super;
1758         struct btrfs_fs_devices *cur_devices;
1759         u64 devid;
1760         u64 num_devices;
1761         u8 *dev_uuid;
1762         int ret = 0;
1763         bool clear_super = false;
1764
1765         mutex_lock(&uuid_mutex);
1766
1767         ret = __check_raid_min_devices(root->fs_info);
1768         if (ret)
1769                 goto out;
1770
1771         if (strcmp(device_path, "missing") == 0) {
1772                 struct list_head *devices;
1773                 struct btrfs_device *tmp;
1774
1775                 device = NULL;
1776                 devices = &root->fs_info->fs_devices->devices;
1777                 /*
1778                  * It is safe to read the devices since the volume_mutex
1779                  * is held.
1780                  */
1781                 list_for_each_entry(tmp, devices, dev_list) {
1782                         if (tmp->in_fs_metadata &&
1783                             !tmp->is_tgtdev_for_dev_replace &&
1784                             !tmp->bdev) {
1785                                 device = tmp;
1786                                 break;
1787                         }
1788                 }
1789                 bdev = NULL;
1790                 bh = NULL;
1791                 disk_super = NULL;
1792                 if (!device) {
1793                         ret = BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
1794                         goto out;
1795                 }
1796         } else {
1797                 ret = btrfs_get_bdev_and_sb(device_path,
1798                                             FMODE_WRITE | FMODE_EXCL,
1799                                             root->fs_info->bdev_holder, 0,
1800                                             &bdev, &bh);
1801                 if (ret)
1802                         goto out;
1803                 disk_super = (struct btrfs_super_block *)bh->b_data;
1804                 devid = btrfs_stack_device_id(&disk_super->dev_item);
1805                 dev_uuid = disk_super->dev_item.uuid;
1806                 device = btrfs_find_device(root->fs_info, devid, dev_uuid,
1807                                            disk_super->fsid);
1808                 if (!device) {
1809                         ret = -ENOENT;
1810                         goto error_brelse;
1811                 }
1812         }
1813
1814         if (device->is_tgtdev_for_dev_replace) {
1815                 ret = BTRFS_ERROR_DEV_TGT_REPLACE;
1816                 goto error_brelse;
1817         }
1818
1819         if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1820                 ret = BTRFS_ERROR_DEV_ONLY_WRITABLE;
1821                 goto error_brelse;
1822         }
1823
1824         if (device->writeable) {
1825                 lock_chunks(root);
1826                 list_del_init(&device->dev_alloc_list);
1827                 device->fs_devices->rw_devices--;
1828                 unlock_chunks(root);
1829                 clear_super = true;
1830         }
1831
1832         mutex_unlock(&uuid_mutex);
1833         ret = btrfs_shrink_device(device, 0);
1834         mutex_lock(&uuid_mutex);
1835         if (ret)
1836                 goto error_undo;
1837
1838         /*
1839          * TODO: the superblock still includes this device in its num_devices
1840          * counter although write_all_supers() is not locked out. This
1841          * could give a filesystem state which requires a degraded mount.
1842          */
1843         ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1844         if (ret)
1845                 goto error_undo;
1846
1847         device->in_fs_metadata = 0;
1848         btrfs_scrub_cancel_dev(root->fs_info, device);
1849
1850         /*
1851          * the device list mutex makes sure that we don't change
1852          * the device list while someone else is writing out all
1853          * the device supers. Whoever is writing all supers, should
1854          * lock the device list mutex before getting the number of
1855          * devices in the super block (super_copy). Conversely,
1856          * whoever updates the number of devices in the super block
1857          * (super_copy) should hold the device list mutex.
1858          */
1859
1860         cur_devices = device->fs_devices;
1861         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1862         list_del_rcu(&device->dev_list);
1863
1864         device->fs_devices->num_devices--;
1865         device->fs_devices->total_devices--;
1866
1867         if (device->missing)
1868                 device->fs_devices->missing_devices--;
1869
1870         next_device = list_entry(root->fs_info->fs_devices->devices.next,
1871                                  struct btrfs_device, dev_list);
1872         if (device->bdev == root->fs_info->sb->s_bdev)
1873                 root->fs_info->sb->s_bdev = next_device->bdev;
1874         if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1875                 root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1876
1877         if (device->bdev) {
1878                 device->fs_devices->open_devices--;
1879                 /* remove sysfs entry */
1880                 btrfs_sysfs_rm_device_link(root->fs_info->fs_devices, device);
1881         }
1882
1883         call_rcu(&device->rcu, free_device);
1884
1885         num_devices = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
1886         btrfs_set_super_num_devices(root->fs_info->super_copy, num_devices);
1887         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1888
1889         if (cur_devices->open_devices == 0) {
1890                 struct btrfs_fs_devices *fs_devices;
1891                 fs_devices = root->fs_info->fs_devices;
1892                 while (fs_devices) {
1893                         if (fs_devices->seed == cur_devices) {
1894                                 fs_devices->seed = cur_devices->seed;
1895                                 break;
1896                         }
1897                         fs_devices = fs_devices->seed;
1898                 }
1899                 cur_devices->seed = NULL;
1900                 __btrfs_close_devices(cur_devices);
1901                 free_fs_devices(cur_devices);
1902         }
1903
1904         root->fs_info->num_tolerated_disk_barrier_failures =
1905                 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
1906
1907         /*
1908          * at this point, the device is zero sized.  We want to
1909          * remove it from the devices list and zero out the old super
1910          */
1911         if (clear_super && disk_super) {
1912                 u64 bytenr;
1913                 int i;
1914
1915                 /* make sure this device isn't detected as part of
1916                  * the FS anymore
1917                  */
1918                 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1919                 set_buffer_dirty(bh);
1920                 sync_dirty_buffer(bh);
1921
1922                 /* clear the mirror copies of super block on the disk
1923                  * being removed, 0th copy is been taken care above and
1924                  * the below would take of the rest
1925                  */
1926                 for (i = 1; i < BTRFS_SUPER_MIRROR_MAX; i++) {
1927                         bytenr = btrfs_sb_offset(i);
1928                         if (bytenr + BTRFS_SUPER_INFO_SIZE >=
1929                                         i_size_read(bdev->bd_inode))
1930                                 break;
1931
1932                         brelse(bh);
1933                         bh = __bread(bdev, bytenr / 4096,
1934                                         BTRFS_SUPER_INFO_SIZE);
1935                         if (!bh)
1936                                 continue;
1937
1938                         disk_super = (struct btrfs_super_block *)bh->b_data;
1939
1940                         if (btrfs_super_bytenr(disk_super) != bytenr ||
1941                                 btrfs_super_magic(disk_super) != BTRFS_MAGIC) {
1942                                 continue;
1943                         }
1944                         memset(&disk_super->magic, 0,
1945                                                 sizeof(disk_super->magic));
1946                         set_buffer_dirty(bh);
1947                         sync_dirty_buffer(bh);
1948                 }
1949         }
1950
1951         ret = 0;
1952
1953         if (bdev) {
1954                 /* Notify udev that device has changed */
1955                 btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
1956
1957                 /* Update ctime/mtime for device path for libblkid */
1958                 update_dev_time(device_path);
1959         }
1960
1961 error_brelse:
1962         brelse(bh);
1963         if (bdev)
1964                 blkdev_put(bdev, FMODE_READ | FMODE_EXCL);
1965 out:
1966         mutex_unlock(&uuid_mutex);
1967         return ret;
1968 error_undo:
1969         if (device->writeable) {
1970                 lock_chunks(root);
1971                 list_add(&device->dev_alloc_list,
1972                          &root->fs_info->fs_devices->alloc_list);
1973                 device->fs_devices->rw_devices++;
1974                 unlock_chunks(root);
1975         }
1976         goto error_brelse;
1977 }
1978
1979 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_fs_info *fs_info,
1980                                         struct btrfs_device *srcdev)
1981 {
1982         struct btrfs_fs_devices *fs_devices;
1983
1984         WARN_ON(!mutex_is_locked(&fs_info->fs_devices->device_list_mutex));
1985
1986         /*
1987          * in case of fs with no seed, srcdev->fs_devices will point
1988          * to fs_devices of fs_info. However when the dev being replaced is
1989          * a seed dev it will point to the seed's local fs_devices. In short
1990          * srcdev will have its correct fs_devices in both the cases.
1991          */
1992         fs_devices = srcdev->fs_devices;
1993
1994         list_del_rcu(&srcdev->dev_list);
1995         list_del_rcu(&srcdev->dev_alloc_list);
1996         fs_devices->num_devices--;
1997         if (srcdev->missing)
1998                 fs_devices->missing_devices--;
1999
2000         if (srcdev->writeable) {
2001                 fs_devices->rw_devices--;
2002                 /* zero out the old super if it is writable */
2003                 btrfs_scratch_superblocks(srcdev->bdev, srcdev->name->str);
2004         }
2005
2006         if (srcdev->bdev)
2007                 fs_devices->open_devices--;
2008 }
2009
2010 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_fs_info *fs_info,
2011                                       struct btrfs_device *srcdev)
2012 {
2013         struct btrfs_fs_devices *fs_devices = srcdev->fs_devices;
2014
2015         call_rcu(&srcdev->rcu, free_device);
2016
2017         /*
2018          * unless fs_devices is seed fs, num_devices shouldn't go
2019          * zero
2020          */
2021         BUG_ON(!fs_devices->num_devices && !fs_devices->seeding);
2022
2023         /* if this is no devs we rather delete the fs_devices */
2024         if (!fs_devices->num_devices) {
2025                 struct btrfs_fs_devices *tmp_fs_devices;
2026
2027                 tmp_fs_devices = fs_info->fs_devices;
2028                 while (tmp_fs_devices) {
2029                         if (tmp_fs_devices->seed == fs_devices) {
2030                                 tmp_fs_devices->seed = fs_devices->seed;
2031                                 break;
2032                         }
2033                         tmp_fs_devices = tmp_fs_devices->seed;
2034                 }
2035                 fs_devices->seed = NULL;
2036                 __btrfs_close_devices(fs_devices);
2037                 free_fs_devices(fs_devices);
2038         }
2039 }
2040
2041 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
2042                                       struct btrfs_device *tgtdev)
2043 {
2044         struct btrfs_device *next_device;
2045
2046         mutex_lock(&uuid_mutex);
2047         WARN_ON(!tgtdev);
2048         mutex_lock(&fs_info->fs_devices->device_list_mutex);
2049
2050         btrfs_sysfs_rm_device_link(fs_info->fs_devices, tgtdev);
2051
2052         if (tgtdev->bdev) {
2053                 btrfs_scratch_superblocks(tgtdev->bdev, tgtdev->name->str);
2054                 fs_info->fs_devices->open_devices--;
2055         }
2056         fs_info->fs_devices->num_devices--;
2057
2058         next_device = list_entry(fs_info->fs_devices->devices.next,
2059                                  struct btrfs_device, dev_list);
2060         if (tgtdev->bdev == fs_info->sb->s_bdev)
2061                 fs_info->sb->s_bdev = next_device->bdev;
2062         if (tgtdev->bdev == fs_info->fs_devices->latest_bdev)
2063                 fs_info->fs_devices->latest_bdev = next_device->bdev;
2064         list_del_rcu(&tgtdev->dev_list);
2065
2066         call_rcu(&tgtdev->rcu, free_device);
2067
2068         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2069         mutex_unlock(&uuid_mutex);
2070 }
2071
2072 static int btrfs_find_device_by_path(struct btrfs_root *root, char *device_path,
2073                                      struct btrfs_device **device)
2074 {
2075         int ret = 0;
2076         struct btrfs_super_block *disk_super;
2077         u64 devid;
2078         u8 *dev_uuid;
2079         struct block_device *bdev;
2080         struct buffer_head *bh;
2081
2082         *device = NULL;
2083         ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
2084                                     root->fs_info->bdev_holder, 0, &bdev, &bh);
2085         if (ret)
2086                 return ret;
2087         disk_super = (struct btrfs_super_block *)bh->b_data;
2088         devid = btrfs_stack_device_id(&disk_super->dev_item);
2089         dev_uuid = disk_super->dev_item.uuid;
2090         *device = btrfs_find_device(root->fs_info, devid, dev_uuid,
2091                                     disk_super->fsid);
2092         brelse(bh);
2093         if (!*device)
2094                 ret = -ENOENT;
2095         blkdev_put(bdev, FMODE_READ);
2096         return ret;
2097 }
2098
2099 int btrfs_find_device_missing_or_by_path(struct btrfs_root *root,
2100                                          char *device_path,
2101                                          struct btrfs_device **device)
2102 {
2103         *device = NULL;
2104         if (strcmp(device_path, "missing") == 0) {
2105                 struct list_head *devices;
2106                 struct btrfs_device *tmp;
2107
2108                 devices = &root->fs_info->fs_devices->devices;
2109                 /*
2110                  * It is safe to read the devices since the volume_mutex
2111                  * is held by the caller.
2112                  */
2113                 list_for_each_entry(tmp, devices, dev_list) {
2114                         if (tmp->in_fs_metadata && !tmp->bdev) {
2115                                 *device = tmp;
2116                                 break;
2117                         }
2118                 }
2119
2120                 if (!*device)
2121                         return BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
2122
2123                 return 0;
2124         } else {
2125                 return btrfs_find_device_by_path(root, device_path, device);
2126         }
2127 }
2128
2129 int btrfs_find_device_by_user_input(struct btrfs_root *root, u64 srcdevid,
2130                                          char *srcdev_name,
2131                                          struct btrfs_device **device)
2132 {
2133         int ret;
2134
2135         if (srcdevid) {
2136                 ret = 0;
2137                 *device = btrfs_find_device(root->fs_info, srcdevid, NULL,
2138                                             NULL);
2139                 if (!*device)
2140                         ret = -ENOENT;
2141         } else {
2142                 ret = btrfs_find_device_missing_or_by_path(root, srcdev_name,
2143                                                            device);
2144         }
2145         return ret;
2146 }
2147
2148 /*
2149  * does all the dirty work required for changing file system's UUID.
2150  */
2151 static int btrfs_prepare_sprout(struct btrfs_root *root)
2152 {
2153         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2154         struct btrfs_fs_devices *old_devices;
2155         struct btrfs_fs_devices *seed_devices;
2156         struct btrfs_super_block *disk_super = root->fs_info->super_copy;
2157         struct btrfs_device *device;
2158         u64 super_flags;
2159
2160         BUG_ON(!mutex_is_locked(&uuid_mutex));
2161         if (!fs_devices->seeding)
2162                 return -EINVAL;
2163
2164         seed_devices = __alloc_fs_devices();
2165         if (IS_ERR(seed_devices))
2166                 return PTR_ERR(seed_devices);
2167
2168         old_devices = clone_fs_devices(fs_devices);
2169         if (IS_ERR(old_devices)) {
2170                 kfree(seed_devices);
2171                 return PTR_ERR(old_devices);
2172         }
2173
2174         list_add(&old_devices->list, &fs_uuids);
2175
2176         memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
2177         seed_devices->opened = 1;
2178         INIT_LIST_HEAD(&seed_devices->devices);
2179         INIT_LIST_HEAD(&seed_devices->alloc_list);
2180         mutex_init(&seed_devices->device_list_mutex);
2181
2182         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2183         list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
2184                               synchronize_rcu);
2185         list_for_each_entry(device, &seed_devices->devices, dev_list)
2186                 device->fs_devices = seed_devices;
2187
2188         lock_chunks(root);
2189         list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
2190         unlock_chunks(root);
2191
2192         fs_devices->seeding = 0;
2193         fs_devices->num_devices = 0;
2194         fs_devices->open_devices = 0;
2195         fs_devices->missing_devices = 0;
2196         fs_devices->rotating = 0;
2197         fs_devices->seed = seed_devices;
2198
2199         generate_random_uuid(fs_devices->fsid);
2200         memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2201         memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2202         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2203
2204         super_flags = btrfs_super_flags(disk_super) &
2205                       ~BTRFS_SUPER_FLAG_SEEDING;
2206         btrfs_set_super_flags(disk_super, super_flags);
2207
2208         return 0;
2209 }
2210
2211 /*
2212  * strore the expected generation for seed devices in device items.
2213  */
2214 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
2215                                struct btrfs_root *root)
2216 {
2217         struct btrfs_path *path;
2218         struct extent_buffer *leaf;
2219         struct btrfs_dev_item *dev_item;
2220         struct btrfs_device *device;
2221         struct btrfs_key key;
2222         u8 fs_uuid[BTRFS_UUID_SIZE];
2223         u8 dev_uuid[BTRFS_UUID_SIZE];
2224         u64 devid;
2225         int ret;
2226
2227         path = btrfs_alloc_path();
2228         if (!path)
2229                 return -ENOMEM;
2230
2231         root = root->fs_info->chunk_root;
2232         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2233         key.offset = 0;
2234         key.type = BTRFS_DEV_ITEM_KEY;
2235
2236         while (1) {
2237                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2238                 if (ret < 0)
2239                         goto error;
2240
2241                 leaf = path->nodes[0];
2242 next_slot:
2243                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2244                         ret = btrfs_next_leaf(root, path);
2245                         if (ret > 0)
2246                                 break;
2247                         if (ret < 0)
2248                                 goto error;
2249                         leaf = path->nodes[0];
2250                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2251                         btrfs_release_path(path);
2252                         continue;
2253                 }
2254
2255                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2256                 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
2257                     key.type != BTRFS_DEV_ITEM_KEY)
2258                         break;
2259
2260                 dev_item = btrfs_item_ptr(leaf, path->slots[0],
2261                                           struct btrfs_dev_item);
2262                 devid = btrfs_device_id(leaf, dev_item);
2263                 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
2264                                    BTRFS_UUID_SIZE);
2265                 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
2266                                    BTRFS_UUID_SIZE);
2267                 device = btrfs_find_device(root->fs_info, devid, dev_uuid,
2268                                            fs_uuid);
2269                 BUG_ON(!device); /* Logic error */
2270
2271                 if (device->fs_devices->seeding) {
2272                         btrfs_set_device_generation(leaf, dev_item,
2273                                                     device->generation);
2274                         btrfs_mark_buffer_dirty(leaf);
2275                 }
2276
2277                 path->slots[0]++;
2278                 goto next_slot;
2279         }
2280         ret = 0;
2281 error:
2282         btrfs_free_path(path);
2283         return ret;
2284 }
2285
2286 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
2287 {
2288         struct request_queue *q;
2289         struct btrfs_trans_handle *trans;
2290         struct btrfs_device *device;
2291         struct block_device *bdev;
2292         struct list_head *devices;
2293         struct super_block *sb = root->fs_info->sb;
2294         struct rcu_string *name;
2295         u64 tmp;
2296         int seeding_dev = 0;
2297         int ret = 0;
2298
2299         if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
2300                 return -EROFS;
2301
2302         bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2303                                   root->fs_info->bdev_holder);
2304         if (IS_ERR(bdev))
2305                 return PTR_ERR(bdev);
2306
2307         if (root->fs_info->fs_devices->seeding) {
2308                 seeding_dev = 1;
2309                 down_write(&sb->s_umount);
2310                 mutex_lock(&uuid_mutex);
2311         }
2312
2313         filemap_write_and_wait(bdev->bd_inode->i_mapping);
2314
2315         devices = &root->fs_info->fs_devices->devices;
2316
2317         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2318         list_for_each_entry(device, devices, dev_list) {
2319                 if (device->bdev == bdev) {
2320                         ret = -EEXIST;
2321                         mutex_unlock(
2322                                 &root->fs_info->fs_devices->device_list_mutex);
2323                         goto error;
2324                 }
2325         }
2326         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2327
2328         device = btrfs_alloc_device(root->fs_info, NULL, NULL);
2329         if (IS_ERR(device)) {
2330                 /* we can safely leave the fs_devices entry around */
2331                 ret = PTR_ERR(device);
2332                 goto error;
2333         }
2334
2335         name = rcu_string_strdup(device_path, GFP_KERNEL);
2336         if (!name) {
2337                 kfree(device);
2338                 ret = -ENOMEM;
2339                 goto error;
2340         }
2341         rcu_assign_pointer(device->name, name);
2342
2343         trans = btrfs_start_transaction(root, 0);
2344         if (IS_ERR(trans)) {
2345                 rcu_string_free(device->name);
2346                 kfree(device);
2347                 ret = PTR_ERR(trans);
2348                 goto error;
2349         }
2350
2351         q = bdev_get_queue(bdev);
2352         if (blk_queue_discard(q))
2353                 device->can_discard = 1;
2354         device->writeable = 1;
2355         device->generation = trans->transid;
2356         device->io_width = root->sectorsize;
2357         device->io_align = root->sectorsize;
2358         device->sector_size = root->sectorsize;
2359         device->total_bytes = i_size_read(bdev->bd_inode);
2360         device->disk_total_bytes = device->total_bytes;
2361         device->commit_total_bytes = device->total_bytes;
2362         device->dev_root = root->fs_info->dev_root;
2363         device->bdev = bdev;
2364         device->in_fs_metadata = 1;
2365         device->is_tgtdev_for_dev_replace = 0;
2366         device->mode = FMODE_EXCL;
2367         device->dev_stats_valid = 1;
2368         set_blocksize(device->bdev, 4096);
2369
2370         if (seeding_dev) {
2371                 sb->s_flags &= ~MS_RDONLY;
2372                 ret = btrfs_prepare_sprout(root);
2373                 BUG_ON(ret); /* -ENOMEM */
2374         }
2375
2376         device->fs_devices = root->fs_info->fs_devices;
2377
2378         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2379         lock_chunks(root);
2380         list_add_rcu(&device->dev_list, &root->fs_info->fs_devices->devices);
2381         list_add(&device->dev_alloc_list,
2382                  &root->fs_info->fs_devices->alloc_list);
2383         root->fs_info->fs_devices->num_devices++;
2384         root->fs_info->fs_devices->open_devices++;
2385         root->fs_info->fs_devices->rw_devices++;
2386         root->fs_info->fs_devices->total_devices++;
2387         root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
2388
2389         spin_lock(&root->fs_info->free_chunk_lock);
2390         root->fs_info->free_chunk_space += device->total_bytes;
2391         spin_unlock(&root->fs_info->free_chunk_lock);
2392
2393         if (!blk_queue_nonrot(bdev_get_queue(bdev)))
2394                 root->fs_info->fs_devices->rotating = 1;
2395
2396         tmp = btrfs_super_total_bytes(root->fs_info->super_copy);
2397         btrfs_set_super_total_bytes(root->fs_info->super_copy,
2398                                     tmp + device->total_bytes);
2399
2400         tmp = btrfs_super_num_devices(root->fs_info->super_copy);
2401         btrfs_set_super_num_devices(root->fs_info->super_copy,
2402                                     tmp + 1);
2403
2404         /* add sysfs device entry */
2405         btrfs_sysfs_add_device_link(root->fs_info->fs_devices, device);
2406
2407         /*
2408          * we've got more storage, clear any full flags on the space
2409          * infos
2410          */
2411         btrfs_clear_space_info_full(root->fs_info);
2412
2413         unlock_chunks(root);
2414         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2415
2416         if (seeding_dev) {
2417                 lock_chunks(root);
2418                 ret = init_first_rw_device(trans, root, device);
2419                 unlock_chunks(root);
2420                 if (ret) {
2421                         btrfs_abort_transaction(trans, root, ret);
2422                         goto error_trans;
2423                 }
2424         }
2425
2426         ret = btrfs_add_device(trans, root, device);
2427         if (ret) {
2428                 btrfs_abort_transaction(trans, root, ret);
2429                 goto error_trans;
2430         }
2431
2432         if (seeding_dev) {
2433                 char fsid_buf[BTRFS_UUID_UNPARSED_SIZE];
2434
2435                 ret = btrfs_finish_sprout(trans, root);
2436                 if (ret) {
2437                         btrfs_abort_transaction(trans, root, ret);
2438                         goto error_trans;
2439                 }
2440
2441                 /* Sprouting would change fsid of the mounted root,
2442                  * so rename the fsid on the sysfs
2443                  */
2444                 snprintf(fsid_buf, BTRFS_UUID_UNPARSED_SIZE, "%pU",
2445                                                 root->fs_info->fsid);
2446                 if (kobject_rename(&root->fs_info->fs_devices->fsid_kobj,
2447                                                                 fsid_buf))
2448                         btrfs_warn(root->fs_info,
2449                                 "sysfs: failed to create fsid for sprout");
2450         }
2451
2452         root->fs_info->num_tolerated_disk_barrier_failures =
2453                 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
2454         ret = btrfs_commit_transaction(trans, root);
2455
2456         if (seeding_dev) {
2457                 mutex_unlock(&uuid_mutex);
2458                 up_write(&sb->s_umount);
2459
2460                 if (ret) /* transaction commit */
2461                         return ret;
2462
2463                 ret = btrfs_relocate_sys_chunks(root);
2464                 if (ret < 0)
2465                         btrfs_std_error(root->fs_info, ret,
2466                                     "Failed to relocate sys chunks after "
2467                                     "device initialization. This can be fixed "
2468                                     "using the \"btrfs balance\" command.");
2469                 trans = btrfs_attach_transaction(root);
2470                 if (IS_ERR(trans)) {
2471                         if (PTR_ERR(trans) == -ENOENT)
2472                                 return 0;
2473                         return PTR_ERR(trans);
2474                 }
2475                 ret = btrfs_commit_transaction(trans, root);
2476         }
2477
2478         /* Update ctime/mtime for libblkid */
2479         update_dev_time(device_path);
2480         return ret;
2481
2482 error_trans:
2483         btrfs_end_transaction(trans, root);
2484         rcu_string_free(device->name);
2485         btrfs_sysfs_rm_device_link(root->fs_info->fs_devices, device);
2486         kfree(device);
2487 error:
2488         blkdev_put(bdev, FMODE_EXCL);
2489         if (seeding_dev) {
2490                 mutex_unlock(&uuid_mutex);
2491                 up_write(&sb->s_umount);
2492         }
2493         return ret;
2494 }
2495
2496 int btrfs_init_dev_replace_tgtdev(struct btrfs_root *root, char *device_path,
2497                                   struct btrfs_device *srcdev,
2498                                   struct btrfs_device **device_out)
2499 {
2500         struct request_queue *q;
2501         struct btrfs_device *device;
2502         struct block_device *bdev;
2503         struct btrfs_fs_info *fs_info = root->fs_info;
2504         struct list_head *devices;
2505         struct rcu_string *name;
2506         u64 devid = BTRFS_DEV_REPLACE_DEVID;
2507         int ret = 0;
2508
2509         *device_out = NULL;
2510         if (fs_info->fs_devices->seeding) {
2511                 btrfs_err(fs_info, "the filesystem is a seed filesystem!");
2512                 return -EINVAL;
2513         }
2514
2515         bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2516                                   fs_info->bdev_holder);
2517         if (IS_ERR(bdev)) {
2518                 btrfs_err(fs_info, "target device %s is invalid!", device_path);
2519                 return PTR_ERR(bdev);
2520         }
2521
2522         filemap_write_and_wait(bdev->bd_inode->i_mapping);
2523
2524         devices = &fs_info->fs_devices->devices;
2525         list_for_each_entry(device, devices, dev_list) {
2526                 if (device->bdev == bdev) {
2527                         btrfs_err(fs_info, "target device is in the filesystem!");
2528                         ret = -EEXIST;
2529                         goto error;
2530                 }
2531         }
2532
2533
2534         if (i_size_read(bdev->bd_inode) <
2535             btrfs_device_get_total_bytes(srcdev)) {
2536                 btrfs_err(fs_info, "target device is smaller than source device!");
2537                 ret = -EINVAL;
2538                 goto error;
2539         }
2540
2541
2542         device = btrfs_alloc_device(NULL, &devid, NULL);
2543         if (IS_ERR(device)) {
2544                 ret = PTR_ERR(device);
2545                 goto error;
2546         }
2547
2548         name = rcu_string_strdup(device_path, GFP_NOFS);
2549         if (!name) {
2550                 kfree(device);
2551                 ret = -ENOMEM;
2552                 goto error;
2553         }
2554         rcu_assign_pointer(device->name, name);
2555
2556         q = bdev_get_queue(bdev);
2557         if (blk_queue_discard(q))
2558                 device->can_discard = 1;
2559         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2560         device->writeable = 1;
2561         device->generation = 0;
2562         device->io_width = root->sectorsize;
2563         device->io_align = root->sectorsize;
2564         device->sector_size = root->sectorsize;
2565         device->total_bytes = btrfs_device_get_total_bytes(srcdev);
2566         device->disk_total_bytes = btrfs_device_get_disk_total_bytes(srcdev);
2567         device->bytes_used = btrfs_device_get_bytes_used(srcdev);
2568         ASSERT(list_empty(&srcdev->resized_list));
2569         device->commit_total_bytes = srcdev->commit_total_bytes;
2570         device->commit_bytes_used = device->bytes_used;
2571         device->dev_root = fs_info->dev_root;
2572         device->bdev = bdev;
2573         device->in_fs_metadata = 1;
2574         device->is_tgtdev_for_dev_replace = 1;
2575         device->mode = FMODE_EXCL;
2576         device->dev_stats_valid = 1;
2577         set_blocksize(device->bdev, 4096);
2578         device->fs_devices = fs_info->fs_devices;
2579         list_add(&device->dev_list, &fs_info->fs_devices->devices);
2580         fs_info->fs_devices->num_devices++;
2581         fs_info->fs_devices->open_devices++;
2582         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2583
2584         *device_out = device;
2585         return ret;
2586
2587 error:
2588         blkdev_put(bdev, FMODE_EXCL);
2589         return ret;
2590 }
2591
2592 void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info *fs_info,
2593                                               struct btrfs_device *tgtdev)
2594 {
2595         WARN_ON(fs_info->fs_devices->rw_devices == 0);
2596         tgtdev->io_width = fs_info->dev_root->sectorsize;
2597         tgtdev->io_align = fs_info->dev_root->sectorsize;
2598         tgtdev->sector_size = fs_info->dev_root->sectorsize;
2599         tgtdev->dev_root = fs_info->dev_root;
2600         tgtdev->in_fs_metadata = 1;
2601 }
2602
2603 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2604                                         struct btrfs_device *device)
2605 {
2606         int ret;
2607         struct btrfs_path *path;
2608         struct btrfs_root *root;
2609         struct btrfs_dev_item *dev_item;
2610         struct extent_buffer *leaf;
2611         struct btrfs_key key;
2612
2613         root = device->dev_root->fs_info->chunk_root;
2614
2615         path = btrfs_alloc_path();
2616         if (!path)
2617                 return -ENOMEM;
2618
2619         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2620         key.type = BTRFS_DEV_ITEM_KEY;
2621         key.offset = device->devid;
2622
2623         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2624         if (ret < 0)
2625                 goto out;
2626
2627         if (ret > 0) {
2628                 ret = -ENOENT;
2629                 goto out;
2630         }
2631
2632         leaf = path->nodes[0];
2633         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2634
2635         btrfs_set_device_id(leaf, dev_item, device->devid);
2636         btrfs_set_device_type(leaf, dev_item, device->type);
2637         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2638         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2639         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2640         btrfs_set_device_total_bytes(leaf, dev_item,
2641                                      btrfs_device_get_disk_total_bytes(device));
2642         btrfs_set_device_bytes_used(leaf, dev_item,
2643                                     btrfs_device_get_bytes_used(device));
2644         btrfs_mark_buffer_dirty(leaf);
2645
2646 out:
2647         btrfs_free_path(path);
2648         return ret;
2649 }
2650
2651 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2652                       struct btrfs_device *device, u64 new_size)
2653 {
2654         struct btrfs_super_block *super_copy =
2655                 device->dev_root->fs_info->super_copy;
2656         struct btrfs_fs_devices *fs_devices;
2657         u64 old_total;
2658         u64 diff;
2659
2660         if (!device->writeable)
2661                 return -EACCES;
2662
2663         lock_chunks(device->dev_root);
2664         old_total = btrfs_super_total_bytes(super_copy);
2665         diff = new_size - device->total_bytes;
2666
2667         if (new_size <= device->total_bytes ||
2668             device->is_tgtdev_for_dev_replace) {
2669                 unlock_chunks(device->dev_root);
2670                 return -EINVAL;
2671         }
2672
2673         fs_devices = device->dev_root->fs_info->fs_devices;
2674
2675         btrfs_set_super_total_bytes(super_copy, old_total + diff);
2676         device->fs_devices->total_rw_bytes += diff;
2677
2678         btrfs_device_set_total_bytes(device, new_size);
2679         btrfs_device_set_disk_total_bytes(device, new_size);
2680         btrfs_clear_space_info_full(device->dev_root->fs_info);
2681         if (list_empty(&device->resized_list))
2682                 list_add_tail(&device->resized_list,
2683                               &fs_devices->resized_devices);
2684         unlock_chunks(device->dev_root);
2685
2686         return btrfs_update_device(trans, device);
2687 }
2688
2689 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
2690                             struct btrfs_root *root, u64 chunk_objectid,
2691                             u64 chunk_offset)
2692 {
2693         int ret;
2694         struct btrfs_path *path;
2695         struct btrfs_key key;
2696
2697         root = root->fs_info->chunk_root;
2698         path = btrfs_alloc_path();
2699         if (!path)
2700                 return -ENOMEM;
2701
2702         key.objectid = chunk_objectid;
2703         key.offset = chunk_offset;
2704         key.type = BTRFS_CHUNK_ITEM_KEY;
2705
2706         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2707         if (ret < 0)
2708                 goto out;
2709         else if (ret > 0) { /* Logic error or corruption */
2710                 btrfs_std_error(root->fs_info, -ENOENT,
2711                             "Failed lookup while freeing chunk.");
2712                 ret = -ENOENT;
2713                 goto out;
2714         }
2715
2716         ret = btrfs_del_item(trans, root, path);
2717         if (ret < 0)
2718                 btrfs_std_error(root->fs_info, ret,
2719                             "Failed to delete chunk item.");
2720 out:
2721         btrfs_free_path(path);
2722         return ret;
2723 }
2724
2725 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
2726                         chunk_offset)
2727 {
2728         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
2729         struct btrfs_disk_key *disk_key;
2730         struct btrfs_chunk *chunk;
2731         u8 *ptr;
2732         int ret = 0;
2733         u32 num_stripes;
2734         u32 array_size;
2735         u32 len = 0;
2736         u32 cur;
2737         struct btrfs_key key;
2738
2739         lock_chunks(root);
2740         array_size = btrfs_super_sys_array_size(super_copy);
2741
2742         ptr = super_copy->sys_chunk_array;
2743         cur = 0;
2744
2745         while (cur < array_size) {
2746                 disk_key = (struct btrfs_disk_key *)ptr;
2747                 btrfs_disk_key_to_cpu(&key, disk_key);
2748
2749                 len = sizeof(*disk_key);
2750
2751                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2752                         chunk = (struct btrfs_chunk *)(ptr + len);
2753                         num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2754                         len += btrfs_chunk_item_size(num_stripes);
2755                 } else {
2756                         ret = -EIO;
2757                         break;
2758                 }
2759                 if (key.objectid == chunk_objectid &&
2760                     key.offset == chunk_offset) {
2761                         memmove(ptr, ptr + len, array_size - (cur + len));
2762                         array_size -= len;
2763                         btrfs_set_super_sys_array_size(super_copy, array_size);
2764                 } else {
2765                         ptr += len;
2766                         cur += len;
2767                 }
2768         }
2769         unlock_chunks(root);
2770         return ret;
2771 }
2772
2773 int btrfs_remove_chunk(struct btrfs_trans_handle *trans,
2774                        struct btrfs_root *root, u64 chunk_offset)
2775 {
2776         struct extent_map_tree *em_tree;
2777         struct extent_map *em;
2778         struct btrfs_root *extent_root = root->fs_info->extent_root;
2779         struct map_lookup *map;
2780         u64 dev_extent_len = 0;
2781         u64 chunk_objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2782         int i, ret = 0;
2783
2784         /* Just in case */
2785         root = root->fs_info->chunk_root;
2786         em_tree = &root->fs_info->mapping_tree.map_tree;
2787
2788         read_lock(&em_tree->lock);
2789         em = lookup_extent_mapping(em_tree, chunk_offset, 1);
2790         read_unlock(&em_tree->lock);
2791
2792         if (!em || em->start > chunk_offset ||
2793             em->start + em->len < chunk_offset) {
2794                 /*
2795                  * This is a logic error, but we don't want to just rely on the
2796                  * user having built with ASSERT enabled, so if ASSERT doesn't
2797                  * do anything we still error out.
2798                  */
2799                 ASSERT(0);
2800                 if (em)
2801                         free_extent_map(em);
2802                 return -EINVAL;
2803         }
2804         map = em->map_lookup;
2805         lock_chunks(root->fs_info->chunk_root);
2806         check_system_chunk(trans, extent_root, map->type);
2807         unlock_chunks(root->fs_info->chunk_root);
2808
2809         for (i = 0; i < map->num_stripes; i++) {
2810                 struct btrfs_device *device = map->stripes[i].dev;
2811                 ret = btrfs_free_dev_extent(trans, device,
2812                                             map->stripes[i].physical,
2813                                             &dev_extent_len);
2814                 if (ret) {
2815                         btrfs_abort_transaction(trans, root, ret);
2816                         goto out;
2817                 }
2818
2819                 if (device->bytes_used > 0) {
2820                         lock_chunks(root);
2821                         btrfs_device_set_bytes_used(device,
2822                                         device->bytes_used - dev_extent_len);
2823                         spin_lock(&root->fs_info->free_chunk_lock);
2824                         root->fs_info->free_chunk_space += dev_extent_len;
2825                         spin_unlock(&root->fs_info->free_chunk_lock);
2826                         btrfs_clear_space_info_full(root->fs_info);
2827                         unlock_chunks(root);
2828                 }
2829
2830                 if (map->stripes[i].dev) {
2831                         ret = btrfs_update_device(trans, map->stripes[i].dev);
2832                         if (ret) {
2833                                 btrfs_abort_transaction(trans, root, ret);
2834                                 goto out;
2835                         }
2836                 }
2837         }
2838         ret = btrfs_free_chunk(trans, root, chunk_objectid, chunk_offset);
2839         if (ret) {
2840                 btrfs_abort_transaction(trans, root, ret);
2841                 goto out;
2842         }
2843
2844         trace_btrfs_chunk_free(root, map, chunk_offset, em->len);
2845
2846         if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2847                 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
2848                 if (ret) {
2849                         btrfs_abort_transaction(trans, root, ret);
2850                         goto out;
2851                 }
2852         }
2853
2854         ret = btrfs_remove_block_group(trans, extent_root, chunk_offset, em);
2855         if (ret) {
2856                 btrfs_abort_transaction(trans, extent_root, ret);
2857                 goto out;
2858         }
2859
2860 out:
2861         /* once for us */
2862         free_extent_map(em);
2863         return ret;
2864 }
2865
2866 static int btrfs_relocate_chunk(struct btrfs_root *root, u64 chunk_offset)
2867 {
2868         struct btrfs_root *extent_root;
2869         struct btrfs_trans_handle *trans;
2870         int ret;
2871
2872         root = root->fs_info->chunk_root;
2873         extent_root = root->fs_info->extent_root;
2874
2875         /*
2876          * Prevent races with automatic removal of unused block groups.
2877          * After we relocate and before we remove the chunk with offset
2878          * chunk_offset, automatic removal of the block group can kick in,
2879          * resulting in a failure when calling btrfs_remove_chunk() below.
2880          *
2881          * Make sure to acquire this mutex before doing a tree search (dev
2882          * or chunk trees) to find chunks. Otherwise the cleaner kthread might
2883          * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after
2884          * we release the path used to search the chunk/dev tree and before
2885          * the current task acquires this mutex and calls us.
2886          */
2887         ASSERT(mutex_is_locked(&root->fs_info->delete_unused_bgs_mutex));
2888
2889         ret = btrfs_can_relocate(extent_root, chunk_offset);
2890         if (ret)
2891                 return -ENOSPC;
2892
2893         /* step one, relocate all the extents inside this chunk */
2894         btrfs_scrub_pause(root);
2895         ret = btrfs_relocate_block_group(extent_root, chunk_offset);
2896         btrfs_scrub_continue(root);
2897         if (ret)
2898                 return ret;
2899
2900         trans = btrfs_start_trans_remove_block_group(root->fs_info,
2901                                                      chunk_offset);
2902         if (IS_ERR(trans)) {
2903                 ret = PTR_ERR(trans);
2904                 btrfs_std_error(root->fs_info, ret, NULL);
2905                 return ret;
2906         }
2907
2908         /*
2909          * step two, delete the device extents and the
2910          * chunk tree entries
2911          */
2912         ret = btrfs_remove_chunk(trans, root, chunk_offset);
2913         btrfs_end_transaction(trans, root);
2914         return ret;
2915 }
2916
2917 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
2918 {
2919         struct btrfs_root *chunk_root = root->fs_info->chunk_root;
2920         struct btrfs_path *path;
2921         struct extent_buffer *leaf;
2922         struct btrfs_chunk *chunk;
2923         struct btrfs_key key;
2924         struct btrfs_key found_key;
2925         u64 chunk_type;
2926         bool retried = false;
2927         int failed = 0;
2928         int ret;
2929
2930         path = btrfs_alloc_path();
2931         if (!path)
2932                 return -ENOMEM;
2933
2934 again:
2935         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2936         key.offset = (u64)-1;
2937         key.type = BTRFS_CHUNK_ITEM_KEY;
2938
2939         while (1) {
2940                 mutex_lock(&root->fs_info->delete_unused_bgs_mutex);
2941                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2942                 if (ret < 0) {
2943                         mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
2944                         goto error;
2945                 }
2946                 BUG_ON(ret == 0); /* Corruption */
2947
2948                 ret = btrfs_previous_item(chunk_root, path, key.objectid,
2949                                           key.type);
2950                 if (ret)
2951                         mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
2952                 if (ret < 0)
2953                         goto error;
2954                 if (ret > 0)
2955                         break;
2956
2957                 leaf = path->nodes[0];
2958                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2959
2960                 chunk = btrfs_item_ptr(leaf, path->slots[0],
2961                                        struct btrfs_chunk);
2962                 chunk_type = btrfs_chunk_type(leaf, chunk);
2963                 btrfs_release_path(path);
2964
2965                 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
2966                         ret = btrfs_relocate_chunk(chunk_root,
2967                                                    found_key.offset);
2968                         if (ret == -ENOSPC)
2969                                 failed++;
2970                         else
2971                                 BUG_ON(ret);
2972                 }
2973                 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
2974
2975                 if (found_key.offset == 0)
2976                         break;
2977                 key.offset = found_key.offset - 1;
2978         }
2979         ret = 0;
2980         if (failed && !retried) {
2981                 failed = 0;
2982                 retried = true;
2983                 goto again;
2984         } else if (WARN_ON(failed && retried)) {
2985                 ret = -ENOSPC;
2986         }
2987 error:
2988         btrfs_free_path(path);
2989         return ret;
2990 }
2991
2992 static int insert_balance_item(struct btrfs_root *root,
2993                                struct btrfs_balance_control *bctl)
2994 {
2995         struct btrfs_trans_handle *trans;
2996         struct btrfs_balance_item *item;
2997         struct btrfs_disk_balance_args disk_bargs;
2998         struct btrfs_path *path;
2999         struct extent_buffer *leaf;
3000         struct btrfs_key key;
3001         int ret, err;
3002
3003         path = btrfs_alloc_path();
3004         if (!path)
3005                 return -ENOMEM;
3006
3007         trans = btrfs_start_transaction(root, 0);
3008         if (IS_ERR(trans)) {
3009                 btrfs_free_path(path);
3010                 return PTR_ERR(trans);
3011         }
3012
3013         key.objectid = BTRFS_BALANCE_OBJECTID;
3014         key.type = BTRFS_TEMPORARY_ITEM_KEY;
3015         key.offset = 0;
3016
3017         ret = btrfs_insert_empty_item(trans, root, path, &key,
3018                                       sizeof(*item));
3019         if (ret)
3020                 goto out;
3021
3022         leaf = path->nodes[0];
3023         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3024
3025         memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
3026
3027         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
3028         btrfs_set_balance_data(leaf, item, &disk_bargs);
3029         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
3030         btrfs_set_balance_meta(leaf, item, &disk_bargs);
3031         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
3032         btrfs_set_balance_sys(leaf, item, &disk_bargs);
3033
3034         btrfs_set_balance_flags(leaf, item, bctl->flags);
3035
3036         btrfs_mark_buffer_dirty(leaf);
3037 out:
3038         btrfs_free_path(path);
3039         err = btrfs_commit_transaction(trans, root);
3040         if (err && !ret)
3041                 ret = err;
3042         return ret;
3043 }
3044
3045 static int del_balance_item(struct btrfs_root *root)
3046 {
3047         struct btrfs_trans_handle *trans;
3048         struct btrfs_path *path;
3049         struct btrfs_key key;
3050         int ret, err;
3051
3052         path = btrfs_alloc_path();
3053         if (!path)
3054                 return -ENOMEM;
3055
3056         trans = btrfs_start_transaction(root, 0);
3057         if (IS_ERR(trans)) {
3058                 btrfs_free_path(path);
3059                 return PTR_ERR(trans);
3060         }
3061
3062         key.objectid = BTRFS_BALANCE_OBJECTID;
3063         key.type = BTRFS_TEMPORARY_ITEM_KEY;
3064         key.offset = 0;
3065
3066         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3067         if (ret < 0)
3068                 goto out;
3069         if (ret > 0) {
3070                 ret = -ENOENT;
3071                 goto out;
3072         }
3073
3074         ret = btrfs_del_item(trans, root, path);
3075 out:
3076         btrfs_free_path(path);
3077         err = btrfs_commit_transaction(trans, root);
3078         if (err && !ret)
3079                 ret = err;
3080         return ret;
3081 }
3082
3083 /*
3084  * This is a heuristic used to reduce the number of chunks balanced on
3085  * resume after balance was interrupted.
3086  */
3087 static void update_balance_args(struct btrfs_balance_control *bctl)
3088 {
3089         /*
3090          * Turn on soft mode for chunk types that were being converted.
3091          */
3092         if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
3093                 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
3094         if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
3095                 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
3096         if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
3097                 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
3098
3099         /*
3100          * Turn on usage filter if is not already used.  The idea is
3101          * that chunks that we have already balanced should be
3102          * reasonably full.  Don't do it for chunks that are being
3103          * converted - that will keep us from relocating unconverted
3104          * (albeit full) chunks.
3105          */
3106         if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3107             !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3108             !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3109                 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
3110                 bctl->data.usage = 90;
3111         }
3112         if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3113             !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3114             !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3115                 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
3116                 bctl->sys.usage = 90;
3117         }
3118         if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3119             !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3120             !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3121                 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
3122                 bctl->meta.usage = 90;
3123         }
3124 }
3125
3126 /*
3127  * Should be called with both balance and volume mutexes held to
3128  * serialize other volume operations (add_dev/rm_dev/resize) with
3129  * restriper.  Same goes for unset_balance_control.
3130  */
3131 static void set_balance_control(struct btrfs_balance_control *bctl)
3132 {
3133         struct btrfs_fs_info *fs_info = bctl->fs_info;
3134
3135         BUG_ON(fs_info->balance_ctl);
3136
3137         spin_lock(&fs_info->balance_lock);
3138         fs_info->balance_ctl = bctl;
3139         spin_unlock(&fs_info->balance_lock);
3140 }
3141
3142 static void unset_balance_control(struct btrfs_fs_info *fs_info)
3143 {
3144         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3145
3146         BUG_ON(!fs_info->balance_ctl);
3147
3148         spin_lock(&fs_info->balance_lock);
3149         fs_info->balance_ctl = NULL;
3150         spin_unlock(&fs_info->balance_lock);
3151
3152         kfree(bctl);
3153 }
3154
3155 /*
3156  * Balance filters.  Return 1 if chunk should be filtered out
3157  * (should not be balanced).
3158  */
3159 static int chunk_profiles_filter(u64 chunk_type,
3160                                  struct btrfs_balance_args *bargs)
3161 {
3162         chunk_type = chunk_to_extended(chunk_type) &
3163                                 BTRFS_EXTENDED_PROFILE_MASK;
3164
3165         if (bargs->profiles & chunk_type)
3166                 return 0;
3167
3168         return 1;
3169 }
3170
3171 static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
3172                               struct btrfs_balance_args *bargs)
3173 {
3174         struct btrfs_block_group_cache *cache;
3175         u64 chunk_used;
3176         u64 user_thresh_min;
3177         u64 user_thresh_max;
3178         int ret = 1;
3179
3180         cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3181         chunk_used = btrfs_block_group_used(&cache->item);
3182
3183         if (bargs->usage_min == 0)
3184                 user_thresh_min = 0;
3185         else
3186                 user_thresh_min = div_factor_fine(cache->key.offset,
3187                                         bargs->usage_min);
3188
3189         if (bargs->usage_max == 0)
3190                 user_thresh_max = 1;
3191         else if (bargs->usage_max > 100)
3192                 user_thresh_max = cache->key.offset;
3193         else
3194                 user_thresh_max = div_factor_fine(cache->key.offset,
3195                                         bargs->usage_max);
3196
3197         if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max)
3198                 ret = 0;
3199
3200         btrfs_put_block_group(cache);
3201         return ret;
3202 }
3203
3204 static int chunk_usage_filter(struct btrfs_fs_info *fs_info,
3205                 u64 chunk_offset, struct btrfs_balance_args *bargs)
3206 {
3207         struct btrfs_block_group_cache *cache;
3208         u64 chunk_used, user_thresh;
3209         int ret = 1;
3210
3211         cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3212         chunk_used = btrfs_block_group_used(&cache->item);
3213
3214         if (bargs->usage_min == 0)
3215                 user_thresh = 1;
3216         else if (bargs->usage > 100)
3217                 user_thresh = cache->key.offset;
3218         else
3219                 user_thresh = div_factor_fine(cache->key.offset,
3220                                               bargs->usage);
3221
3222         if (chunk_used < user_thresh)
3223                 ret = 0;
3224
3225         btrfs_put_block_group(cache);
3226         return ret;
3227 }
3228
3229 static int chunk_devid_filter(struct extent_buffer *leaf,
3230                               struct btrfs_chunk *chunk,
3231                               struct btrfs_balance_args *bargs)
3232 {
3233         struct btrfs_stripe *stripe;
3234         int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3235         int i;
3236
3237         for (i = 0; i < num_stripes; i++) {
3238                 stripe = btrfs_stripe_nr(chunk, i);
3239                 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
3240                         return 0;
3241         }
3242
3243         return 1;
3244 }
3245
3246 /* [pstart, pend) */
3247 static int chunk_drange_filter(struct extent_buffer *leaf,
3248                                struct btrfs_chunk *chunk,
3249                                u64 chunk_offset,
3250                                struct btrfs_balance_args *bargs)
3251 {
3252         struct btrfs_stripe *stripe;
3253         int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3254         u64 stripe_offset;
3255         u64 stripe_length;
3256         int factor;
3257         int i;
3258
3259         if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
3260                 return 0;
3261
3262         if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP |
3263              BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) {
3264                 factor = num_stripes / 2;
3265         } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) {
3266                 factor = num_stripes - 1;
3267         } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) {
3268                 factor = num_stripes - 2;
3269         } else {
3270                 factor = num_stripes;
3271         }
3272
3273         for (i = 0; i < num_stripes; i++) {
3274                 stripe = btrfs_stripe_nr(chunk, i);
3275                 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
3276                         continue;
3277
3278                 stripe_offset = btrfs_stripe_offset(leaf, stripe);
3279                 stripe_length = btrfs_chunk_length(leaf, chunk);
3280                 stripe_length = div_u64(stripe_length, factor);
3281
3282                 if (stripe_offset < bargs->pend &&
3283                     stripe_offset + stripe_length > bargs->pstart)
3284                         return 0;
3285         }
3286
3287         return 1;
3288 }
3289
3290 /* [vstart, vend) */
3291 static int chunk_vrange_filter(struct extent_buffer *leaf,
3292                                struct btrfs_chunk *chunk,
3293                                u64 chunk_offset,
3294                                struct btrfs_balance_args *bargs)
3295 {
3296         if (chunk_offset < bargs->vend &&
3297             chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
3298                 /* at least part of the chunk is inside this vrange */
3299                 return 0;
3300
3301         return 1;
3302 }
3303
3304 static int chunk_stripes_range_filter(struct extent_buffer *leaf,
3305                                struct btrfs_chunk *chunk,
3306                                struct btrfs_balance_args *bargs)
3307 {
3308         int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3309
3310         if (bargs->stripes_min <= num_stripes
3311                         && num_stripes <= bargs->stripes_max)
3312                 return 0;
3313
3314         return 1;
3315 }
3316
3317 static int chunk_soft_convert_filter(u64 chunk_type,
3318                                      struct btrfs_balance_args *bargs)
3319 {
3320         if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
3321                 return 0;
3322
3323         chunk_type = chunk_to_extended(chunk_type) &
3324                                 BTRFS_EXTENDED_PROFILE_MASK;
3325
3326         if (bargs->target == chunk_type)
3327                 return 1;
3328
3329         return 0;
3330 }
3331
3332 static int should_balance_chunk(struct btrfs_root *root,
3333                                 struct extent_buffer *leaf,
3334                                 struct btrfs_chunk *chunk, u64 chunk_offset)
3335 {
3336         struct btrfs_balance_control *bctl = root->fs_info->balance_ctl;
3337         struct btrfs_balance_args *bargs = NULL;
3338         u64 chunk_type = btrfs_chunk_type(leaf, chunk);
3339
3340         /* type filter */
3341         if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
3342               (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
3343                 return 0;
3344         }
3345
3346         if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3347                 bargs = &bctl->data;
3348         else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3349                 bargs = &bctl->sys;
3350         else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3351                 bargs = &bctl->meta;
3352
3353         /* profiles filter */
3354         if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
3355             chunk_profiles_filter(chunk_type, bargs)) {
3356                 return 0;
3357         }
3358
3359         /* usage filter */
3360         if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
3361             chunk_usage_filter(bctl->fs_info, chunk_offset, bargs)) {
3362                 return 0;
3363         } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3364             chunk_usage_range_filter(bctl->fs_info, chunk_offset, bargs)) {
3365                 return 0;
3366         }
3367
3368         /* devid filter */
3369         if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
3370             chunk_devid_filter(leaf, chunk, bargs)) {
3371                 return 0;
3372         }
3373
3374         /* drange filter, makes sense only with devid filter */
3375         if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
3376             chunk_drange_filter(leaf, chunk, chunk_offset, bargs)) {
3377                 return 0;
3378         }
3379
3380         /* vrange filter */
3381         if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
3382             chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
3383                 return 0;
3384         }
3385
3386         /* stripes filter */
3387         if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) &&
3388             chunk_stripes_range_filter(leaf, chunk, bargs)) {
3389                 return 0;
3390         }
3391
3392         /* soft profile changing mode */
3393         if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
3394             chunk_soft_convert_filter(chunk_type, bargs)) {
3395                 return 0;
3396         }
3397
3398         /*
3399          * limited by count, must be the last filter
3400          */
3401         if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) {
3402                 if (bargs->limit == 0)
3403                         return 0;
3404                 else
3405                         bargs->limit--;
3406         } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) {
3407                 /*
3408                  * Same logic as the 'limit' filter; the minimum cannot be
3409                  * determined here because we do not have the global informatoin
3410                  * about the count of all chunks that satisfy the filters.
3411                  */
3412                 if (bargs->limit_max == 0)
3413                         return 0;
3414                 else
3415                         bargs->limit_max--;
3416         }
3417
3418         return 1;
3419 }
3420
3421 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
3422 {
3423         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3424         struct btrfs_root *chunk_root = fs_info->chunk_root;
3425         struct btrfs_root *dev_root = fs_info->dev_root;
3426         struct list_head *devices;
3427         struct btrfs_device *device;
3428         u64 old_size;
3429         u64 size_to_free;
3430         u64 chunk_type;
3431         struct btrfs_chunk *chunk;
3432         struct btrfs_path *path;
3433         struct btrfs_key key;
3434         struct btrfs_key found_key;
3435         struct btrfs_trans_handle *trans;
3436         struct extent_buffer *leaf;
3437         int slot;
3438         int ret;
3439         int enospc_errors = 0;
3440         bool counting = true;
3441         /* The single value limit and min/max limits use the same bytes in the */
3442         u64 limit_data = bctl->data.limit;
3443         u64 limit_meta = bctl->meta.limit;
3444         u64 limit_sys = bctl->sys.limit;
3445         u32 count_data = 0;
3446         u32 count_meta = 0;
3447         u32 count_sys = 0;
3448         int chunk_reserved = 0;
3449
3450         /* step one make some room on all the devices */
3451         devices = &fs_info->fs_devices->devices;
3452         list_for_each_entry(device, devices, dev_list) {
3453                 old_size = btrfs_device_get_total_bytes(device);
3454                 size_to_free = div_factor(old_size, 1);
3455                 size_to_free = min_t(u64, size_to_free, SZ_1M);
3456                 if (!device->writeable ||
3457                     btrfs_device_get_total_bytes(device) -
3458                     btrfs_device_get_bytes_used(device) > size_to_free ||
3459                     device->is_tgtdev_for_dev_replace)
3460                         continue;
3461
3462                 ret = btrfs_shrink_device(device, old_size - size_to_free);
3463                 if (ret == -ENOSPC)
3464                         break;
3465                 BUG_ON(ret);
3466
3467                 trans = btrfs_start_transaction(dev_root, 0);
3468                 BUG_ON(IS_ERR(trans));
3469
3470                 ret = btrfs_grow_device(trans, device, old_size);
3471                 BUG_ON(ret);
3472
3473                 btrfs_end_transaction(trans, dev_root);
3474         }
3475
3476         /* step two, relocate all the chunks */
3477         path = btrfs_alloc_path();
3478         if (!path) {
3479                 ret = -ENOMEM;
3480                 goto error;
3481         }
3482
3483         /* zero out stat counters */
3484         spin_lock(&fs_info->balance_lock);
3485         memset(&bctl->stat, 0, sizeof(bctl->stat));
3486         spin_unlock(&fs_info->balance_lock);
3487 again:
3488         if (!counting) {
3489                 /*
3490                  * The single value limit and min/max limits use the same bytes
3491                  * in the
3492                  */
3493                 bctl->data.limit = limit_data;
3494                 bctl->meta.limit = limit_meta;
3495                 bctl->sys.limit = limit_sys;
3496         }
3497         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3498         key.offset = (u64)-1;
3499         key.type = BTRFS_CHUNK_ITEM_KEY;
3500
3501         while (1) {
3502                 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
3503                     atomic_read(&fs_info->balance_cancel_req)) {
3504                         ret = -ECANCELED;
3505                         goto error;
3506                 }
3507
3508                 mutex_lock(&fs_info->delete_unused_bgs_mutex);
3509                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3510                 if (ret < 0) {
3511                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3512                         goto error;
3513                 }
3514
3515                 /*
3516                  * this shouldn't happen, it means the last relocate
3517                  * failed
3518                  */
3519                 if (ret == 0)
3520                         BUG(); /* FIXME break ? */
3521
3522                 ret = btrfs_previous_item(chunk_root, path, 0,
3523                                           BTRFS_CHUNK_ITEM_KEY);
3524                 if (ret) {
3525                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3526                         ret = 0;
3527                         break;
3528                 }
3529
3530                 leaf = path->nodes[0];
3531                 slot = path->slots[0];
3532                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3533
3534                 if (found_key.objectid != key.objectid) {
3535                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3536                         break;
3537                 }
3538
3539                 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3540                 chunk_type = btrfs_chunk_type(leaf, chunk);
3541
3542                 if (!counting) {
3543                         spin_lock(&fs_info->balance_lock);
3544                         bctl->stat.considered++;
3545                         spin_unlock(&fs_info->balance_lock);
3546                 }
3547
3548                 ret = should_balance_chunk(chunk_root, leaf, chunk,
3549                                            found_key.offset);
3550
3551                 btrfs_release_path(path);
3552                 if (!ret) {
3553                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3554                         goto loop;
3555                 }
3556
3557                 if (counting) {
3558                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3559                         spin_lock(&fs_info->balance_lock);
3560                         bctl->stat.expected++;
3561                         spin_unlock(&fs_info->balance_lock);
3562
3563                         if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3564                                 count_data++;
3565                         else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3566                                 count_sys++;
3567                         else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3568                                 count_meta++;
3569
3570                         goto loop;
3571                 }
3572
3573                 /*
3574                  * Apply limit_min filter, no need to check if the LIMITS
3575                  * filter is used, limit_min is 0 by default
3576                  */
3577                 if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
3578                                         count_data < bctl->data.limit_min)
3579                                 || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) &&
3580                                         count_meta < bctl->meta.limit_min)
3581                                 || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) &&
3582                                         count_sys < bctl->sys.limit_min)) {
3583                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3584                         goto loop;
3585                 }
3586
3587                 if ((chunk_type & BTRFS_BLOCK_GROUP_DATA) && !chunk_reserved) {
3588                         trans = btrfs_start_transaction(chunk_root, 0);
3589                         if (IS_ERR(trans)) {
3590                                 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3591                                 ret = PTR_ERR(trans);
3592                                 goto error;
3593                         }
3594
3595                         ret = btrfs_force_chunk_alloc(trans, chunk_root,
3596                                                       BTRFS_BLOCK_GROUP_DATA);
3597                         btrfs_end_transaction(trans, chunk_root);
3598                         if (ret < 0) {
3599                                 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3600                                 goto error;
3601                         }
3602                         chunk_reserved = 1;
3603                 }
3604
3605                 ret = btrfs_relocate_chunk(chunk_root,
3606                                            found_key.offset);
3607                 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3608                 if (ret && ret != -ENOSPC)
3609                         goto error;
3610                 if (ret == -ENOSPC) {
3611                         enospc_errors++;
3612                 } else {
3613                         spin_lock(&fs_info->balance_lock);
3614                         bctl->stat.completed++;
3615                         spin_unlock(&fs_info->balance_lock);
3616                 }
3617 loop:
3618                 if (found_key.offset == 0)
3619                         break;
3620                 key.offset = found_key.offset - 1;
3621         }
3622
3623         if (counting) {
3624                 btrfs_release_path(path);
3625                 counting = false;
3626                 goto again;
3627         }
3628 error:
3629         btrfs_free_path(path);
3630         if (enospc_errors) {
3631                 btrfs_info(fs_info, "%d enospc errors during balance",
3632                        enospc_errors);
3633                 if (!ret)
3634                         ret = -ENOSPC;
3635         }
3636
3637         return ret;
3638 }
3639
3640 /**
3641  * alloc_profile_is_valid - see if a given profile is valid and reduced
3642  * @flags: profile to validate
3643  * @extended: if true @flags is treated as an extended profile
3644  */
3645 static int alloc_profile_is_valid(u64 flags, int extended)
3646 {
3647         u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
3648                                BTRFS_BLOCK_GROUP_PROFILE_MASK);
3649
3650         flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
3651
3652         /* 1) check that all other bits are zeroed */
3653         if (flags & ~mask)
3654                 return 0;
3655
3656         /* 2) see if profile is reduced */
3657         if (flags == 0)
3658                 return !extended; /* "0" is valid for usual profiles */
3659
3660         /* true if exactly one bit set */
3661         return (flags & (flags - 1)) == 0;
3662 }
3663
3664 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
3665 {
3666         /* cancel requested || normal exit path */
3667         return atomic_read(&fs_info->balance_cancel_req) ||
3668                 (atomic_read(&fs_info->balance_pause_req) == 0 &&
3669                  atomic_read(&fs_info->balance_cancel_req) == 0);
3670 }
3671
3672 static void __cancel_balance(struct btrfs_fs_info *fs_info)
3673 {
3674         int ret;
3675
3676         unset_balance_control(fs_info);
3677         ret = del_balance_item(fs_info->tree_root);
3678         if (ret)
3679                 btrfs_std_error(fs_info, ret, NULL);
3680
3681         atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3682 }
3683
3684 /* Non-zero return value signifies invalidity */
3685 static inline int validate_convert_profile(struct btrfs_balance_args *bctl_arg,
3686                 u64 allowed)
3687 {
3688         return ((bctl_arg->flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3689                 (!alloc_profile_is_valid(bctl_arg->target, 1) ||
3690                  (bctl_arg->target & ~allowed)));
3691 }
3692
3693 /*
3694  * Should be called with both balance and volume mutexes held
3695  */
3696 int btrfs_balance(struct btrfs_balance_control *bctl,
3697                   struct btrfs_ioctl_balance_args *bargs)
3698 {
3699         struct btrfs_fs_info *fs_info = bctl->fs_info;
3700         u64 allowed;
3701         int mixed = 0;
3702         int ret;
3703         u64 num_devices;
3704         unsigned seq;
3705
3706         if (btrfs_fs_closing(fs_info) ||
3707             atomic_read(&fs_info->balance_pause_req) ||
3708             atomic_read(&fs_info->balance_cancel_req)) {
3709                 ret = -EINVAL;
3710                 goto out;
3711         }
3712
3713         allowed = btrfs_super_incompat_flags(fs_info->super_copy);
3714         if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
3715                 mixed = 1;
3716
3717         /*
3718          * In case of mixed groups both data and meta should be picked,
3719          * and identical options should be given for both of them.
3720          */
3721         allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
3722         if (mixed && (bctl->flags & allowed)) {
3723                 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
3724                     !(bctl->flags & BTRFS_BALANCE_METADATA) ||
3725                     memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
3726                         btrfs_err(fs_info, "with mixed groups data and "
3727                                    "metadata balance options must be the same");
3728                         ret = -EINVAL;
3729                         goto out;
3730                 }
3731         }
3732
3733         num_devices = fs_info->fs_devices->num_devices;
3734         btrfs_dev_replace_lock(&fs_info->dev_replace, 0);
3735         if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
3736                 BUG_ON(num_devices < 1);
3737                 num_devices--;
3738         }
3739         btrfs_dev_replace_unlock(&fs_info->dev_replace, 0);
3740         allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE;
3741         if (num_devices == 1)
3742                 allowed |= BTRFS_BLOCK_GROUP_DUP;
3743         else if (num_devices > 1)
3744                 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1);
3745         if (num_devices > 2)
3746                 allowed |= BTRFS_BLOCK_GROUP_RAID5;
3747         if (num_devices > 3)
3748                 allowed |= (BTRFS_BLOCK_GROUP_RAID10 |
3749                             BTRFS_BLOCK_GROUP_RAID6);
3750         if (validate_convert_profile(&bctl->data, allowed)) {
3751                 btrfs_err(fs_info, "unable to start balance with target "
3752                            "data profile %llu",
3753                        bctl->data.target);
3754                 ret = -EINVAL;
3755                 goto out;
3756         }
3757         if (validate_convert_profile(&bctl->meta, allowed)) {
3758                 btrfs_err(fs_info,
3759                            "unable to start balance with target metadata profile %llu",
3760                        bctl->meta.target);
3761                 ret = -EINVAL;
3762                 goto out;
3763         }
3764         if (validate_convert_profile(&bctl->sys, allowed)) {
3765                 btrfs_err(fs_info,
3766                            "unable to start balance with target system profile %llu",
3767                        bctl->sys.target);
3768                 ret = -EINVAL;
3769                 goto out;
3770         }
3771
3772         /* allow to reduce meta or sys integrity only if force set */
3773         allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3774                         BTRFS_BLOCK_GROUP_RAID10 |
3775                         BTRFS_BLOCK_GROUP_RAID5 |
3776                         BTRFS_BLOCK_GROUP_RAID6;
3777         do {
3778                 seq = read_seqbegin(&fs_info->profiles_lock);
3779
3780                 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3781                      (fs_info->avail_system_alloc_bits & allowed) &&
3782                      !(bctl->sys.target & allowed)) ||
3783                     ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3784                      (fs_info->avail_metadata_alloc_bits & allowed) &&
3785                      !(bctl->meta.target & allowed))) {
3786                         if (bctl->flags & BTRFS_BALANCE_FORCE) {
3787                                 btrfs_info(fs_info, "force reducing metadata integrity");
3788                         } else {
3789                                 btrfs_err(fs_info, "balance will reduce metadata "
3790                                            "integrity, use force if you want this");
3791                                 ret = -EINVAL;
3792                                 goto out;
3793                         }
3794                 }
3795         } while (read_seqretry(&fs_info->profiles_lock, seq));
3796
3797         if (btrfs_get_num_tolerated_disk_barrier_failures(bctl->meta.target) <
3798                 btrfs_get_num_tolerated_disk_barrier_failures(bctl->data.target)) {
3799                 btrfs_warn(fs_info,
3800         "metadata profile 0x%llx has lower redundancy than data profile 0x%llx",
3801                         bctl->meta.target, bctl->data.target);
3802         }
3803
3804         if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3805                 fs_info->num_tolerated_disk_barrier_failures = min(
3806                         btrfs_calc_num_tolerated_disk_barrier_failures(fs_info),
3807                         btrfs_get_num_tolerated_disk_barrier_failures(
3808                                 bctl->sys.target));
3809         }
3810
3811         ret = insert_balance_item(fs_info->tree_root, bctl);
3812         if (ret && ret != -EEXIST)
3813                 goto out;
3814
3815         if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
3816                 BUG_ON(ret == -EEXIST);
3817                 set_balance_control(bctl);
3818         } else {
3819                 BUG_ON(ret != -EEXIST);
3820                 spin_lock(&fs_info->balance_lock);
3821                 update_balance_args(bctl);
3822                 spin_unlock(&fs_info->balance_lock);
3823         }
3824
3825         atomic_inc(&fs_info->balance_running);
3826         mutex_unlock(&fs_info->balance_mutex);
3827
3828         ret = __btrfs_balance(fs_info);
3829
3830         mutex_lock(&fs_info->balance_mutex);
3831         atomic_dec(&fs_info->balance_running);
3832
3833         if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3834                 fs_info->num_tolerated_disk_barrier_failures =
3835                         btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3836         }
3837
3838         if (bargs) {
3839                 memset(bargs, 0, sizeof(*bargs));
3840                 update_ioctl_balance_args(fs_info, 0, bargs);
3841         }
3842
3843         if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
3844             balance_need_close(fs_info)) {
3845                 __cancel_balance(fs_info);
3846         }
3847
3848         wake_up(&fs_info->balance_wait_q);
3849
3850         return ret;
3851 out:
3852         if (bctl->flags & BTRFS_BALANCE_RESUME)
3853                 __cancel_balance(fs_info);
3854         else {
3855                 kfree(bctl);
3856                 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3857         }
3858         return ret;
3859 }
3860
3861 static int balance_kthread(void *data)
3862 {
3863         struct btrfs_fs_info *fs_info = data;
3864         int ret = 0;
3865
3866         mutex_lock(&fs_info->volume_mutex);
3867         mutex_lock(&fs_info->balance_mutex);
3868
3869         if (fs_info->balance_ctl) {
3870                 btrfs_info(fs_info, "continuing balance");
3871                 ret = btrfs_balance(fs_info->balance_ctl, NULL);
3872         }
3873
3874         mutex_unlock(&fs_info->balance_mutex);
3875         mutex_unlock(&fs_info->volume_mutex);
3876
3877         return ret;
3878 }
3879
3880 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
3881 {
3882         struct task_struct *tsk;
3883
3884         spin_lock(&fs_info->balance_lock);
3885         if (!fs_info->balance_ctl) {
3886                 spin_unlock(&fs_info->balance_lock);
3887                 return 0;
3888         }
3889         spin_unlock(&fs_info->balance_lock);
3890
3891         if (btrfs_test_opt(fs_info->tree_root, SKIP_BALANCE)) {
3892                 btrfs_info(fs_info, "force skipping balance");
3893                 return 0;
3894         }
3895
3896         tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
3897         return PTR_ERR_OR_ZERO(tsk);
3898 }
3899
3900 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
3901 {
3902         struct btrfs_balance_control *bctl;
3903         struct btrfs_balance_item *item;
3904         struct btrfs_disk_balance_args disk_bargs;
3905         struct btrfs_path *path;
3906         struct extent_buffer *leaf;
3907         struct btrfs_key key;
3908         int ret;
3909
3910         path = btrfs_alloc_path();
3911         if (!path)
3912                 return -ENOMEM;
3913
3914         key.objectid = BTRFS_BALANCE_OBJECTID;
3915         key.type = BTRFS_TEMPORARY_ITEM_KEY;
3916         key.offset = 0;
3917
3918         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
3919         if (ret < 0)
3920                 goto out;
3921         if (ret > 0) { /* ret = -ENOENT; */
3922                 ret = 0;
3923                 goto out;
3924         }
3925
3926         bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
3927         if (!bctl) {
3928                 ret = -ENOMEM;
3929                 goto out;
3930         }
3931
3932         leaf = path->nodes[0];
3933         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3934
3935         bctl->fs_info = fs_info;
3936         bctl->flags = btrfs_balance_flags(leaf, item);
3937         bctl->flags |= BTRFS_BALANCE_RESUME;
3938
3939         btrfs_balance_data(leaf, item, &disk_bargs);
3940         btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
3941         btrfs_balance_meta(leaf, item, &disk_bargs);
3942         btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
3943         btrfs_balance_sys(leaf, item, &disk_bargs);
3944         btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
3945
3946         WARN_ON(atomic_xchg(&fs_info->mutually_exclusive_operation_running, 1));
3947
3948         mutex_lock(&fs_info->volume_mutex);
3949         mutex_lock(&fs_info->balance_mutex);
3950
3951         set_balance_control(bctl);
3952
3953         mutex_unlock(&fs_info->balance_mutex);
3954         mutex_unlock(&fs_info->volume_mutex);
3955 out:
3956         btrfs_free_path(path);
3957         return ret;
3958 }
3959
3960 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
3961 {
3962         int ret = 0;
3963
3964         mutex_lock(&fs_info->balance_mutex);
3965         if (!fs_info->balance_ctl) {
3966                 mutex_unlock(&fs_info->balance_mutex);
3967                 return -ENOTCONN;
3968         }
3969
3970         if (atomic_read(&fs_info->balance_running)) {
3971                 atomic_inc(&fs_info->balance_pause_req);
3972                 mutex_unlock(&fs_info->balance_mutex);
3973
3974                 wait_event(fs_info->balance_wait_q,
3975                            atomic_read(&fs_info->balance_running) == 0);
3976
3977                 mutex_lock(&fs_info->balance_mutex);
3978                 /* we are good with balance_ctl ripped off from under us */
3979                 BUG_ON(atomic_read(&fs_info->balance_running));
3980                 atomic_dec(&fs_info->balance_pause_req);
3981         } else {
3982                 ret = -ENOTCONN;
3983         }
3984
3985         mutex_unlock(&fs_info->balance_mutex);
3986         return ret;
3987 }
3988
3989 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
3990 {
3991         if (fs_info->sb->s_flags & MS_RDONLY)
3992                 return -EROFS;
3993
3994         mutex_lock(&fs_info->balance_mutex);
3995         if (!fs_info->balance_ctl) {
3996                 mutex_unlock(&fs_info->balance_mutex);
3997                 return -ENOTCONN;
3998         }
3999
4000         atomic_inc(&fs_info->balance_cancel_req);
4001         /*
4002          * if we are running just wait and return, balance item is
4003          * deleted in btrfs_balance in this case
4004          */
4005         if (atomic_read(&fs_info->balance_running)) {
4006                 mutex_unlock(&fs_info->balance_mutex);
4007                 wait_event(fs_info->balance_wait_q,
4008                            atomic_read(&fs_info->balance_running) == 0);
4009                 mutex_lock(&fs_info->balance_mutex);
4010         } else {
4011                 /* __cancel_balance needs volume_mutex */
4012                 mutex_unlock(&fs_info->balance_mutex);
4013                 mutex_lock(&fs_info->volume_mutex);
4014                 mutex_lock(&fs_info->balance_mutex);
4015
4016                 if (fs_info->balance_ctl)
4017                         __cancel_balance(fs_info);
4018
4019                 mutex_unlock(&fs_info->volume_mutex);
4020         }
4021
4022         BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running));
4023         atomic_dec(&fs_info->balance_cancel_req);
4024         mutex_unlock(&fs_info->balance_mutex);
4025         return 0;
4026 }
4027
4028 static int btrfs_uuid_scan_kthread(void *data)
4029 {
4030         struct btrfs_fs_info *fs_info = data;
4031         struct btrfs_root *root = fs_info->tree_root;
4032         struct btrfs_key key;
4033         struct btrfs_key max_key;
4034         struct btrfs_path *path = NULL;
4035         int ret = 0;
4036         struct extent_buffer *eb;
4037         int slot;
4038         struct btrfs_root_item root_item;
4039         u32 item_size;
4040         struct btrfs_trans_handle *trans = NULL;
4041
4042         path = btrfs_alloc_path();
4043         if (!path) {
4044                 ret = -ENOMEM;
4045                 goto out;
4046         }
4047
4048         key.objectid = 0;
4049         key.type = BTRFS_ROOT_ITEM_KEY;
4050         key.offset = 0;
4051
4052         max_key.objectid = (u64)-1;
4053         max_key.type = BTRFS_ROOT_ITEM_KEY;
4054         max_key.offset = (u64)-1;
4055
4056         while (1) {
4057                 ret = btrfs_search_forward(root, &key, path, 0);
4058                 if (ret) {
4059                         if (ret > 0)
4060                                 ret = 0;
4061                         break;
4062                 }
4063
4064                 if (key.type != BTRFS_ROOT_ITEM_KEY ||
4065                     (key.objectid < BTRFS_FIRST_FREE_OBJECTID &&
4066                      key.objectid != BTRFS_FS_TREE_OBJECTID) ||
4067                     key.objectid > BTRFS_LAST_FREE_OBJECTID)
4068                         goto skip;
4069
4070                 eb = path->nodes[0];
4071                 slot = path->slots[0];
4072                 item_size = btrfs_item_size_nr(eb, slot);
4073                 if (item_size < sizeof(root_item))
4074                         goto skip;
4075
4076                 read_extent_buffer(eb, &root_item,
4077                                    btrfs_item_ptr_offset(eb, slot),
4078                                    (int)sizeof(root_item));
4079                 if (btrfs_root_refs(&root_item) == 0)
4080                         goto skip;
4081
4082                 if (!btrfs_is_empty_uuid(root_item.uuid) ||
4083                     !btrfs_is_empty_uuid(root_item.received_uuid)) {
4084                         if (trans)
4085                                 goto update_tree;
4086
4087                         btrfs_release_path(path);
4088                         /*
4089                          * 1 - subvol uuid item
4090                          * 1 - received_subvol uuid item
4091                          */
4092                         trans = btrfs_start_transaction(fs_info->uuid_root, 2);
4093                         if (IS_ERR(trans)) {
4094                                 ret = PTR_ERR(trans);
4095                                 break;
4096                         }
4097                         continue;
4098                 } else {
4099                         goto skip;
4100                 }
4101 update_tree:
4102                 if (!btrfs_is_empty_uuid(root_item.uuid)) {
4103                         ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
4104                                                   root_item.uuid,
4105                                                   BTRFS_UUID_KEY_SUBVOL,
4106                                                   key.objectid);
4107                         if (ret < 0) {
4108                                 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4109                                         ret);
4110                                 break;
4111                         }
4112                 }
4113
4114                 if (!btrfs_is_empty_uuid(root_item.received_uuid)) {
4115                         ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
4116                                                   root_item.received_uuid,
4117                                                  BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4118                                                   key.objectid);
4119                         if (ret < 0) {
4120                                 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4121                                         ret);
4122                                 break;
4123                         }
4124                 }
4125
4126 skip:
4127                 if (trans) {
4128                         ret = btrfs_end_transaction(trans, fs_info->uuid_root);
4129                         trans = NULL;
4130                         if (ret)
4131                                 break;
4132                 }
4133
4134                 btrfs_release_path(path);
4135                 if (key.offset < (u64)-1) {
4136                         key.offset++;
4137                 } else if (key.type < BTRFS_ROOT_ITEM_KEY) {
4138                         key.offset = 0;
4139                         key.type = BTRFS_ROOT_ITEM_KEY;
4140                 } else if (key.objectid < (u64)-1) {
4141                         key.offset = 0;
4142                         key.type = BTRFS_ROOT_ITEM_KEY;
4143                         key.objectid++;
4144                 } else {
4145                         break;
4146                 }
4147                 cond_resched();
4148         }
4149
4150 out:
4151         btrfs_free_path(path);
4152         if (trans && !IS_ERR(trans))
4153                 btrfs_end_transaction(trans, fs_info->uuid_root);
4154         if (ret)
4155                 btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret);
4156         else
4157                 fs_info->update_uuid_tree_gen = 1;
4158         up(&fs_info->uuid_tree_rescan_sem);
4159         return 0;
4160 }
4161
4162 /*
4163  * Callback for btrfs_uuid_tree_iterate().
4164  * returns:
4165  * 0    check succeeded, the entry is not outdated.
4166  * < 0  if an error occurred.
4167  * > 0  if the check failed, which means the caller shall remove the entry.
4168  */
4169 static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info *fs_info,
4170                                        u8 *uuid, u8 type, u64 subid)
4171 {
4172         struct btrfs_key key;
4173         int ret = 0;
4174         struct btrfs_root *subvol_root;
4175
4176         if (type != BTRFS_UUID_KEY_SUBVOL &&
4177             type != BTRFS_UUID_KEY_RECEIVED_SUBVOL)
4178                 goto out;
4179
4180         key.objectid = subid;
4181         key.type = BTRFS_ROOT_ITEM_KEY;
4182         key.offset = (u64)-1;
4183         subvol_root = btrfs_read_fs_root_no_name(fs_info, &key);
4184         if (IS_ERR(subvol_root)) {
4185                 ret = PTR_ERR(subvol_root);
4186                 if (ret == -ENOENT)
4187                         ret = 1;
4188                 goto out;
4189         }
4190
4191         switch (type) {
4192         case BTRFS_UUID_KEY_SUBVOL:
4193                 if (memcmp(uuid, subvol_root->root_item.uuid, BTRFS_UUID_SIZE))
4194                         ret = 1;
4195                 break;
4196         case BTRFS_UUID_KEY_RECEIVED_SUBVOL:
4197                 if (memcmp(uuid, subvol_root->root_item.received_uuid,
4198                            BTRFS_UUID_SIZE))
4199                         ret = 1;
4200                 break;
4201         }
4202
4203 out:
4204         return ret;
4205 }
4206
4207 static int btrfs_uuid_rescan_kthread(void *data)
4208 {
4209         struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data;
4210         int ret;
4211
4212         /*
4213          * 1st step is to iterate through the existing UUID tree and
4214          * to delete all entries that contain outdated data.
4215          * 2nd step is to add all missing entries to the UUID tree.
4216          */
4217         ret = btrfs_uuid_tree_iterate(fs_info, btrfs_check_uuid_tree_entry);
4218         if (ret < 0) {
4219                 btrfs_warn(fs_info, "iterating uuid_tree failed %d", ret);
4220                 up(&fs_info->uuid_tree_rescan_sem);
4221                 return ret;
4222         }
4223         return btrfs_uuid_scan_kthread(data);
4224 }
4225
4226 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info)
4227 {
4228         struct btrfs_trans_handle *trans;
4229         struct btrfs_root *tree_root = fs_info->tree_root;
4230         struct btrfs_root *uuid_root;
4231         struct task_struct *task;
4232         int ret;
4233
4234         /*
4235          * 1 - root node
4236          * 1 - root item
4237          */
4238         trans = btrfs_start_transaction(tree_root, 2);
4239         if (IS_ERR(trans))
4240                 return PTR_ERR(trans);
4241
4242         uuid_root = btrfs_create_tree(trans, fs_info,
4243                                       BTRFS_UUID_TREE_OBJECTID);
4244         if (IS_ERR(uuid_root)) {
4245                 ret = PTR_ERR(uuid_root);
4246                 btrfs_abort_transaction(trans, tree_root, ret);
4247                 return ret;
4248         }
4249
4250         fs_info->uuid_root = uuid_root;
4251
4252         ret = btrfs_commit_transaction(trans, tree_root);
4253         if (ret)
4254                 return ret;
4255
4256         down(&fs_info->uuid_tree_rescan_sem);
4257         task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid");
4258         if (IS_ERR(task)) {
4259                 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4260                 btrfs_warn(fs_info, "failed to start uuid_scan task");
4261                 up(&fs_info->uuid_tree_rescan_sem);
4262                 return PTR_ERR(task);
4263         }
4264
4265         return 0;
4266 }
4267
4268 int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
4269 {
4270         struct task_struct *task;
4271
4272         down(&fs_info->uuid_tree_rescan_sem);
4273         task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
4274         if (IS_ERR(task)) {
4275                 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4276                 btrfs_warn(fs_info, "failed to start uuid_rescan task");
4277                 up(&fs_info->uuid_tree_rescan_sem);
4278                 return PTR_ERR(task);
4279         }
4280
4281         return 0;
4282 }
4283
4284 /*
4285  * shrinking a device means finding all of the device extents past
4286  * the new size, and then following the back refs to the chunks.
4287  * The chunk relocation code actually frees the device extent
4288  */
4289 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
4290 {
4291         struct btrfs_trans_handle *trans;
4292         struct btrfs_root *root = device->dev_root;
4293         struct btrfs_dev_extent *dev_extent = NULL;
4294         struct btrfs_path *path;
4295         u64 length;
4296         u64 chunk_offset;
4297         int ret;
4298         int slot;
4299         int failed = 0;
4300         bool retried = false;
4301         bool checked_pending_chunks = false;
4302         struct extent_buffer *l;
4303         struct btrfs_key key;
4304         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
4305         u64 old_total = btrfs_super_total_bytes(super_copy);
4306         u64 old_size = btrfs_device_get_total_bytes(device);
4307         u64 diff = old_size - new_size;
4308
4309         if (device->is_tgtdev_for_dev_replace)
4310                 return -EINVAL;
4311
4312         path = btrfs_alloc_path();
4313         if (!path)
4314                 return -ENOMEM;
4315
4316         path->reada = READA_FORWARD;
4317
4318         lock_chunks(root);
4319
4320         btrfs_device_set_total_bytes(device, new_size);
4321         if (device->writeable) {
4322                 device->fs_devices->total_rw_bytes -= diff;
4323                 spin_lock(&root->fs_info->free_chunk_lock);
4324                 root->fs_info->free_chunk_space -= diff;
4325                 spin_unlock(&root->fs_info->free_chunk_lock);
4326         }
4327         unlock_chunks(root);
4328
4329 again:
4330         key.objectid = device->devid;
4331         key.offset = (u64)-1;
4332         key.type = BTRFS_DEV_EXTENT_KEY;
4333
4334         do {
4335                 mutex_lock(&root->fs_info->delete_unused_bgs_mutex);
4336                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4337                 if (ret < 0) {
4338                         mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4339                         goto done;
4340                 }
4341
4342                 ret = btrfs_previous_item(root, path, 0, key.type);
4343                 if (ret)
4344                         mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4345                 if (ret < 0)
4346                         goto done;
4347                 if (ret) {
4348                         ret = 0;
4349                         btrfs_release_path(path);
4350                         break;
4351                 }
4352
4353                 l = path->nodes[0];
4354                 slot = path->slots[0];
4355                 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
4356
4357                 if (key.objectid != device->devid) {
4358                         mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4359                         btrfs_release_path(path);
4360                         break;
4361                 }
4362
4363                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
4364                 length = btrfs_dev_extent_length(l, dev_extent);
4365
4366                 if (key.offset + length <= new_size) {
4367                         mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4368                         btrfs_release_path(path);
4369                         break;
4370                 }
4371
4372                 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
4373                 btrfs_release_path(path);
4374
4375                 ret = btrfs_relocate_chunk(root, chunk_offset);
4376                 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4377                 if (ret && ret != -ENOSPC)
4378                         goto done;
4379                 if (ret == -ENOSPC)
4380                         failed++;
4381         } while (key.offset-- > 0);
4382
4383         if (failed && !retried) {
4384                 failed = 0;
4385                 retried = true;
4386                 goto again;
4387         } else if (failed && retried) {
4388                 ret = -ENOSPC;
4389                 goto done;
4390         }
4391
4392         /* Shrinking succeeded, else we would be at "done". */
4393         trans = btrfs_start_transaction(root, 0);
4394         if (IS_ERR(trans)) {
4395                 ret = PTR_ERR(trans);
4396                 goto done;
4397         }
4398
4399         lock_chunks(root);
4400
4401         /*
4402          * We checked in the above loop all device extents that were already in
4403          * the device tree. However before we have updated the device's
4404          * total_bytes to the new size, we might have had chunk allocations that
4405          * have not complete yet (new block groups attached to transaction
4406          * handles), and therefore their device extents were not yet in the
4407          * device tree and we missed them in the loop above. So if we have any
4408          * pending chunk using a device extent that overlaps the device range
4409          * that we can not use anymore, commit the current transaction and
4410          * repeat the search on the device tree - this way we guarantee we will
4411          * not have chunks using device extents that end beyond 'new_size'.
4412          */
4413         if (!checked_pending_chunks) {
4414                 u64 start = new_size;
4415                 u64 len = old_size - new_size;
4416
4417                 if (contains_pending_extent(trans->transaction, device,
4418                                             &start, len)) {
4419                         unlock_chunks(root);
4420                         checked_pending_chunks = true;
4421                         failed = 0;
4422                         retried = false;
4423                         ret = btrfs_commit_transaction(trans, root);
4424                         if (ret)
4425                                 goto done;
4426                         goto again;
4427                 }
4428         }
4429
4430         btrfs_device_set_disk_total_bytes(device, new_size);
4431         if (list_empty(&device->resized_list))
4432                 list_add_tail(&device->resized_list,
4433                               &root->fs_info->fs_devices->resized_devices);
4434
4435         WARN_ON(diff > old_total);
4436         btrfs_set_super_total_bytes(super_copy, old_total - diff);
4437         unlock_chunks(root);
4438
4439         /* Now btrfs_update_device() will change the on-disk size. */
4440         ret = btrfs_update_device(trans, device);
4441         btrfs_end_transaction(trans, root);
4442 done:
4443         btrfs_free_path(path);
4444         if (ret) {
4445                 lock_chunks(root);
4446                 btrfs_device_set_total_bytes(device, old_size);
4447                 if (device->writeable)
4448                         device->fs_devices->total_rw_bytes += diff;
4449                 spin_lock(&root->fs_info->free_chunk_lock);
4450                 root->fs_info->free_chunk_space += diff;
4451                 spin_unlock(&root->fs_info->free_chunk_lock);
4452                 unlock_chunks(root);
4453         }
4454         return ret;
4455 }
4456
4457 static int btrfs_add_system_chunk(struct btrfs_root *root,
4458                            struct btrfs_key *key,
4459                            struct btrfs_chunk *chunk, int item_size)
4460 {
4461         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
4462         struct btrfs_disk_key disk_key;
4463         u32 array_size;
4464         u8 *ptr;
4465
4466         lock_chunks(root);
4467         array_size = btrfs_super_sys_array_size(super_copy);
4468         if (array_size + item_size + sizeof(disk_key)
4469                         > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
4470                 unlock_chunks(root);
4471                 return -EFBIG;
4472         }
4473
4474         ptr = super_copy->sys_chunk_array + array_size;
4475         btrfs_cpu_key_to_disk(&disk_key, key);
4476         memcpy(ptr, &disk_key, sizeof(disk_key));
4477         ptr += sizeof(disk_key);
4478         memcpy(ptr, chunk, item_size);
4479         item_size += sizeof(disk_key);
4480         btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
4481         unlock_chunks(root);
4482
4483         return 0;
4484 }
4485
4486 /*
4487  * sort the devices in descending order by max_avail, total_avail
4488  */
4489 static int btrfs_cmp_device_info(const void *a, const void *b)
4490 {
4491         const struct btrfs_device_info *di_a = a;
4492         const struct btrfs_device_info *di_b = b;
4493
4494         if (di_a->max_avail > di_b->max_avail)
4495                 return -1;
4496         if (di_a->max_avail < di_b->max_avail)
4497                 return 1;
4498         if (di_a->total_avail > di_b->total_avail)
4499                 return -1;
4500         if (di_a->total_avail < di_b->total_avail)
4501                 return 1;
4502         return 0;
4503 }
4504
4505 static u32 find_raid56_stripe_len(u32 data_devices, u32 dev_stripe_target)
4506 {
4507         /* TODO allow them to set a preferred stripe size */
4508         return SZ_64K;
4509 }
4510
4511 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
4512 {
4513         if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK))
4514                 return;
4515
4516         btrfs_set_fs_incompat(info, RAID56);
4517 }
4518
4519 #define BTRFS_MAX_DEVS(r) ((BTRFS_LEAF_DATA_SIZE(r)             \
4520                         - sizeof(struct btrfs_item)             \
4521                         - sizeof(struct btrfs_chunk))           \
4522                         / sizeof(struct btrfs_stripe) + 1)
4523
4524 #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE        \
4525                                 - 2 * sizeof(struct btrfs_disk_key)     \
4526                                 - 2 * sizeof(struct btrfs_chunk))       \
4527                                 / sizeof(struct btrfs_stripe) + 1)
4528
4529 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4530                                struct btrfs_root *extent_root, u64 start,
4531                                u64 type)
4532 {
4533         struct btrfs_fs_info *info = extent_root->fs_info;
4534         struct btrfs_fs_devices *fs_devices = info->fs_devices;
4535         struct list_head *cur;
4536         struct map_lookup *map = NULL;
4537         struct extent_map_tree *em_tree;
4538         struct extent_map *em;
4539         struct btrfs_device_info *devices_info = NULL;
4540         u64 total_avail;
4541         int num_stripes;        /* total number of stripes to allocate */
4542         int data_stripes;       /* number of stripes that count for
4543                                    block group size */
4544         int sub_stripes;        /* sub_stripes info for map */
4545         int dev_stripes;        /* stripes per dev */
4546         int devs_max;           /* max devs to use */
4547         int devs_min;           /* min devs needed */
4548         int devs_increment;     /* ndevs has to be a multiple of this */
4549         int ncopies;            /* how many copies to data has */
4550         int ret;
4551         u64 max_stripe_size;
4552         u64 max_chunk_size;
4553         u64 stripe_size;
4554         u64 num_bytes;
4555         u64 raid_stripe_len = BTRFS_STRIPE_LEN;
4556         int ndevs;
4557         int i;
4558         int j;
4559         int index;
4560
4561         BUG_ON(!alloc_profile_is_valid(type, 0));
4562
4563         if (list_empty(&fs_devices->alloc_list))
4564                 return -ENOSPC;
4565
4566         index = __get_raid_index(type);
4567
4568         sub_stripes = btrfs_raid_array[index].sub_stripes;
4569         dev_stripes = btrfs_raid_array[index].dev_stripes;
4570         devs_max = btrfs_raid_array[index].devs_max;
4571         devs_min = btrfs_raid_array[index].devs_min;
4572         devs_increment = btrfs_raid_array[index].devs_increment;
4573         ncopies = btrfs_raid_array[index].ncopies;
4574
4575         if (type & BTRFS_BLOCK_GROUP_DATA) {
4576                 max_stripe_size = SZ_1G;
4577                 max_chunk_size = 10 * max_stripe_size;
4578                 if (!devs_max)
4579                         devs_max = BTRFS_MAX_DEVS(info->chunk_root);
4580         } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
4581                 /* for larger filesystems, use larger metadata chunks */
4582                 if (fs_devices->total_rw_bytes > 50ULL * SZ_1G)
4583                         max_stripe_size = SZ_1G;
4584                 else
4585                         max_stripe_size = SZ_256M;
4586                 max_chunk_size = max_stripe_size;
4587                 if (!devs_max)
4588                         devs_max = BTRFS_MAX_DEVS(info->chunk_root);
4589         } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
4590                 max_stripe_size = SZ_32M;
4591                 max_chunk_size = 2 * max_stripe_size;
4592                 if (!devs_max)
4593                         devs_max = BTRFS_MAX_DEVS_SYS_CHUNK;
4594         } else {
4595                 btrfs_err(info, "invalid chunk type 0x%llx requested",
4596                        type);
4597                 BUG_ON(1);
4598         }
4599
4600         /* we don't want a chunk larger than 10% of writeable space */
4601         max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
4602                              max_chunk_size);
4603
4604         devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info),
4605                                GFP_NOFS);
4606         if (!devices_info)
4607                 return -ENOMEM;
4608
4609         cur = fs_devices->alloc_list.next;
4610
4611         /*
4612          * in the first pass through the devices list, we gather information
4613          * about the available holes on each device.
4614          */
4615         ndevs = 0;
4616         while (cur != &fs_devices->alloc_list) {
4617                 struct btrfs_device *device;
4618                 u64 max_avail;
4619                 u64 dev_offset;
4620
4621                 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
4622
4623                 cur = cur->next;
4624
4625                 if (!device->writeable) {
4626                         WARN(1, KERN_ERR
4627                                "BTRFS: read-only device in alloc_list\n");
4628                         continue;
4629                 }
4630
4631                 if (!device->in_fs_metadata ||
4632                     device->is_tgtdev_for_dev_replace)
4633                         continue;
4634
4635                 if (device->total_bytes > device->bytes_used)
4636                         total_avail = device->total_bytes - device->bytes_used;
4637                 else
4638                         total_avail = 0;
4639
4640                 /* If there is no space on this device, skip it. */
4641                 if (total_avail == 0)
4642                         continue;
4643
4644                 ret = find_free_dev_extent(trans, device,
4645                                            max_stripe_size * dev_stripes,
4646                                            &dev_offset, &max_avail);
4647                 if (ret && ret != -ENOSPC)
4648                         goto error;
4649
4650                 if (ret == 0)
4651                         max_avail = max_stripe_size * dev_stripes;
4652
4653                 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes)
4654                         continue;
4655
4656                 if (ndevs == fs_devices->rw_devices) {
4657                         WARN(1, "%s: found more than %llu devices\n",
4658                              __func__, fs_devices->rw_devices);
4659                         break;
4660                 }
4661                 devices_info[ndevs].dev_offset = dev_offset;
4662                 devices_info[ndevs].max_avail = max_avail;
4663                 devices_info[ndevs].total_avail = total_avail;
4664                 devices_info[ndevs].dev = device;
4665                 ++ndevs;
4666         }
4667
4668         /*
4669          * now sort the devices by hole size / available space
4670          */
4671         sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
4672              btrfs_cmp_device_info, NULL);
4673
4674         /* round down to number of usable stripes */
4675         ndevs -= ndevs % devs_increment;
4676
4677         if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) {
4678                 ret = -ENOSPC;
4679                 goto error;
4680         }
4681
4682         if (devs_max && ndevs > devs_max)
4683                 ndevs = devs_max;
4684         /*
4685          * the primary goal is to maximize the number of stripes, so use as many
4686          * devices as possible, even if the stripes are not maximum sized.
4687          */
4688         stripe_size = devices_info[ndevs-1].max_avail;
4689         num_stripes = ndevs * dev_stripes;
4690
4691         /*
4692          * this will have to be fixed for RAID1 and RAID10 over
4693          * more drives
4694          */
4695         data_stripes = num_stripes / ncopies;
4696
4697         if (type & BTRFS_BLOCK_GROUP_RAID5) {
4698                 raid_stripe_len = find_raid56_stripe_len(ndevs - 1,
4699                                  btrfs_super_stripesize(info->super_copy));
4700                 data_stripes = num_stripes - 1;
4701         }
4702         if (type & BTRFS_BLOCK_GROUP_RAID6) {
4703                 raid_stripe_len = find_raid56_stripe_len(ndevs - 2,
4704                                  btrfs_super_stripesize(info->super_copy));
4705                 data_stripes = num_stripes - 2;
4706         }
4707
4708         /*
4709          * Use the number of data stripes to figure out how big this chunk
4710          * is really going to be in terms of logical address space,
4711          * and compare that answer with the max chunk size
4712          */
4713         if (stripe_size * data_stripes > max_chunk_size) {
4714                 u64 mask = (1ULL << 24) - 1;
4715
4716                 stripe_size = div_u64(max_chunk_size, data_stripes);
4717
4718                 /* bump the answer up to a 16MB boundary */
4719                 stripe_size = (stripe_size + mask) & ~mask;
4720
4721                 /* but don't go higher than the limits we found
4722                  * while searching for free extents
4723                  */
4724                 if (stripe_size > devices_info[ndevs-1].max_avail)
4725                         stripe_size = devices_info[ndevs-1].max_avail;
4726         }
4727
4728         stripe_size = div_u64(stripe_size, dev_stripes);
4729
4730         /* align to BTRFS_STRIPE_LEN */
4731         stripe_size = div_u64(stripe_size, raid_stripe_len);
4732         stripe_size *= raid_stripe_len;
4733
4734         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
4735         if (!map) {
4736                 ret = -ENOMEM;
4737                 goto error;
4738         }
4739         map->num_stripes = num_stripes;
4740
4741         for (i = 0; i < ndevs; ++i) {
4742                 for (j = 0; j < dev_stripes; ++j) {
4743                         int s = i * dev_stripes + j;
4744                         map->stripes[s].dev = devices_info[i].dev;
4745                         map->stripes[s].physical = devices_info[i].dev_offset +
4746                                                    j * stripe_size;
4747                 }
4748         }
4749         map->sector_size = extent_root->sectorsize;
4750         map->stripe_len = raid_stripe_len;
4751         map->io_align = raid_stripe_len;
4752         map->io_width = raid_stripe_len;
4753         map->type = type;
4754         map->sub_stripes = sub_stripes;
4755
4756         num_bytes = stripe_size * data_stripes;
4757
4758         trace_btrfs_chunk_alloc(info->chunk_root, map, start, num_bytes);
4759
4760         em = alloc_extent_map();
4761         if (!em) {
4762                 kfree(map);
4763                 ret = -ENOMEM;
4764                 goto error;
4765         }
4766         set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
4767         em->map_lookup = map;
4768         em->start = start;
4769         em->len = num_bytes;
4770         em->block_start = 0;
4771         em->block_len = em->len;
4772         em->orig_block_len = stripe_size;
4773
4774         em_tree = &extent_root->fs_info->mapping_tree.map_tree;
4775         write_lock(&em_tree->lock);
4776         ret = add_extent_mapping(em_tree, em, 0);
4777         if (!ret) {
4778                 list_add_tail(&em->list, &trans->transaction->pending_chunks);
4779                 atomic_inc(&em->refs);
4780         }
4781         write_unlock(&em_tree->lock);
4782         if (ret) {
4783                 free_extent_map(em);
4784                 goto error;
4785         }
4786
4787         ret = btrfs_make_block_group(trans, extent_root, 0, type,
4788                                      BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4789                                      start, num_bytes);
4790         if (ret)
4791                 goto error_del_extent;
4792
4793         for (i = 0; i < map->num_stripes; i++) {
4794                 num_bytes = map->stripes[i].dev->bytes_used + stripe_size;
4795                 btrfs_device_set_bytes_used(map->stripes[i].dev, num_bytes);
4796         }
4797
4798         spin_lock(&extent_root->fs_info->free_chunk_lock);
4799         extent_root->fs_info->free_chunk_space -= (stripe_size *
4800                                                    map->num_stripes);
4801         spin_unlock(&extent_root->fs_info->free_chunk_lock);
4802
4803         free_extent_map(em);
4804         check_raid56_incompat_flag(extent_root->fs_info, type);
4805
4806         kfree(devices_info);
4807         return 0;
4808
4809 error_del_extent:
4810         write_lock(&em_tree->lock);
4811         remove_extent_mapping(em_tree, em);
4812         write_unlock(&em_tree->lock);
4813
4814         /* One for our allocation */
4815         free_extent_map(em);
4816         /* One for the tree reference */
4817         free_extent_map(em);
4818         /* One for the pending_chunks list reference */
4819         free_extent_map(em);
4820 error:
4821         kfree(devices_info);
4822         return ret;
4823 }
4824
4825 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans,
4826                                 struct btrfs_root *extent_root,
4827                                 u64 chunk_offset, u64 chunk_size)
4828 {
4829         struct btrfs_key key;
4830         struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
4831         struct btrfs_device *device;
4832         struct btrfs_chunk *chunk;
4833         struct btrfs_stripe *stripe;
4834         struct extent_map_tree *em_tree;
4835         struct extent_map *em;
4836         struct map_lookup *map;
4837         size_t item_size;
4838         u64 dev_offset;
4839         u64 stripe_size;
4840         int i = 0;
4841         int ret = 0;
4842
4843         em_tree = &extent_root->fs_info->mapping_tree.map_tree;
4844         read_lock(&em_tree->lock);
4845         em = lookup_extent_mapping(em_tree, chunk_offset, chunk_size);
4846         read_unlock(&em_tree->lock);
4847
4848         if (!em) {
4849                 btrfs_crit(extent_root->fs_info, "unable to find logical "
4850                            "%Lu len %Lu", chunk_offset, chunk_size);
4851                 return -EINVAL;
4852         }
4853
4854         if (em->start != chunk_offset || em->len != chunk_size) {
4855                 btrfs_crit(extent_root->fs_info, "found a bad mapping, wanted"
4856                           " %Lu-%Lu, found %Lu-%Lu", chunk_offset,
4857                           chunk_size, em->start, em->len);
4858                 free_extent_map(em);
4859                 return -EINVAL;
4860         }
4861
4862         map = em->map_lookup;
4863         item_size = btrfs_chunk_item_size(map->num_stripes);
4864         stripe_size = em->orig_block_len;
4865
4866         chunk = kzalloc(item_size, GFP_NOFS);
4867         if (!chunk) {
4868                 ret = -ENOMEM;
4869                 goto out;
4870         }
4871
4872         /*
4873          * Take the device list mutex to prevent races with the final phase of
4874          * a device replace operation that replaces the device object associated
4875          * with the map's stripes, because the device object's id can change
4876          * at any time during that final phase of the device replace operation
4877          * (dev-replace.c:btrfs_dev_replace_finishing()).
4878          */
4879         mutex_lock(&chunk_root->fs_info->fs_devices->device_list_mutex);
4880         for (i = 0; i < map->num_stripes; i++) {
4881                 device = map->stripes[i].dev;
4882                 dev_offset = map->stripes[i].physical;
4883
4884                 ret = btrfs_update_device(trans, device);
4885                 if (ret)
4886                         break;
4887                 ret = btrfs_alloc_dev_extent(trans, device,
4888                                              chunk_root->root_key.objectid,
4889                                              BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4890                                              chunk_offset, dev_offset,
4891                                              stripe_size);
4892                 if (ret)
4893                         break;
4894         }
4895         if (ret) {
4896                 mutex_unlock(&chunk_root->fs_info->fs_devices->device_list_mutex);
4897                 goto out;
4898         }
4899
4900         stripe = &chunk->stripe;
4901         for (i = 0; i < map->num_stripes; i++) {
4902                 device = map->stripes[i].dev;
4903                 dev_offset = map->stripes[i].physical;
4904
4905                 btrfs_set_stack_stripe_devid(stripe, device->devid);
4906                 btrfs_set_stack_stripe_offset(stripe, dev_offset);
4907                 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
4908                 stripe++;
4909         }
4910         mutex_unlock(&chunk_root->fs_info->fs_devices->device_list_mutex);
4911
4912         btrfs_set_stack_chunk_length(chunk, chunk_size);
4913         btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
4914         btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
4915         btrfs_set_stack_chunk_type(chunk, map->type);
4916         btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
4917         btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
4918         btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
4919         btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
4920         btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
4921
4922         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
4923         key.type = BTRFS_CHUNK_ITEM_KEY;
4924         key.offset = chunk_offset;
4925
4926         ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
4927         if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
4928                 /*
4929                  * TODO: Cleanup of inserted chunk root in case of
4930                  * failure.
4931                  */
4932                 ret = btrfs_add_system_chunk(chunk_root, &key, chunk,
4933                                              item_size);
4934         }
4935
4936 out:
4937         kfree(chunk);
4938         free_extent_map(em);
4939         return ret;
4940 }
4941
4942 /*
4943  * Chunk allocation falls into two parts. The first part does works
4944  * that make the new allocated chunk useable, but not do any operation
4945  * that modifies the chunk tree. The second part does the works that
4946  * require modifying the chunk tree. This division is important for the
4947  * bootstrap process of adding storage to a seed btrfs.
4948  */
4949 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4950                       struct btrfs_root *extent_root, u64 type)
4951 {
4952         u64 chunk_offset;
4953
4954         ASSERT(mutex_is_locked(&extent_root->fs_info->chunk_mutex));
4955         chunk_offset = find_next_chunk(extent_root->fs_info);
4956         return __btrfs_alloc_chunk(trans, extent_root, chunk_offset, type);
4957 }
4958
4959 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
4960                                          struct btrfs_root *root,
4961                                          struct btrfs_device *device)
4962 {
4963         u64 chunk_offset;
4964         u64 sys_chunk_offset;
4965         u64 alloc_profile;
4966         struct btrfs_fs_info *fs_info = root->fs_info;
4967         struct btrfs_root *extent_root = fs_info->extent_root;
4968         int ret;
4969
4970         chunk_offset = find_next_chunk(fs_info);
4971         alloc_profile = btrfs_get_alloc_profile(extent_root, 0);
4972         ret = __btrfs_alloc_chunk(trans, extent_root, chunk_offset,
4973                                   alloc_profile);
4974         if (ret)
4975                 return ret;
4976
4977         sys_chunk_offset = find_next_chunk(root->fs_info);
4978         alloc_profile = btrfs_get_alloc_profile(fs_info->chunk_root, 0);
4979         ret = __btrfs_alloc_chunk(trans, extent_root, sys_chunk_offset,
4980                                   alloc_profile);
4981         return ret;
4982 }
4983
4984 static inline int btrfs_chunk_max_errors(struct map_lookup *map)
4985 {
4986         int max_errors;
4987
4988         if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
4989                          BTRFS_BLOCK_GROUP_RAID10 |
4990                          BTRFS_BLOCK_GROUP_RAID5 |
4991                          BTRFS_BLOCK_GROUP_DUP)) {
4992                 max_errors = 1;
4993         } else if (map->type & BTRFS_BLOCK_GROUP_RAID6) {
4994                 max_errors = 2;
4995         } else {
4996                 max_errors = 0;
4997         }
4998
4999         return max_errors;
5000 }
5001
5002 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
5003 {
5004         struct extent_map *em;
5005         struct map_lookup *map;
5006         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
5007         int readonly = 0;
5008         int miss_ndevs = 0;
5009         int i;
5010
5011         read_lock(&map_tree->map_tree.lock);
5012         em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
5013         read_unlock(&map_tree->map_tree.lock);
5014         if (!em)
5015                 return 1;
5016
5017         map = em->map_lookup;
5018         for (i = 0; i < map->num_stripes; i++) {
5019                 if (map->stripes[i].dev->missing) {
5020                         miss_ndevs++;
5021                         continue;
5022                 }
5023
5024                 if (!map->stripes[i].dev->writeable) {
5025                         readonly = 1;
5026                         goto end;
5027                 }
5028         }
5029
5030         /*
5031          * If the number of missing devices is larger than max errors,
5032          * we can not write the data into that chunk successfully, so
5033          * set it readonly.
5034          */
5035         if (miss_ndevs > btrfs_chunk_max_errors(map))
5036                 readonly = 1;
5037 end:
5038         free_extent_map(em);
5039         return readonly;
5040 }
5041
5042 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
5043 {
5044         extent_map_tree_init(&tree->map_tree);
5045 }
5046
5047 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
5048 {
5049         struct extent_map *em;
5050
5051         while (1) {
5052                 write_lock(&tree->map_tree.lock);
5053                 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
5054                 if (em)
5055                         remove_extent_mapping(&tree->map_tree, em);
5056                 write_unlock(&tree->map_tree.lock);
5057                 if (!em)
5058                         break;
5059                 /* once for us */
5060                 free_extent_map(em);
5061                 /* once for the tree */
5062                 free_extent_map(em);
5063         }
5064 }
5065
5066 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5067 {
5068         struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
5069         struct extent_map *em;
5070         struct map_lookup *map;
5071         struct extent_map_tree *em_tree = &map_tree->map_tree;
5072         int ret;
5073
5074         read_lock(&em_tree->lock);
5075         em = lookup_extent_mapping(em_tree, logical, len);
5076         read_unlock(&em_tree->lock);
5077
5078         /*
5079          * We could return errors for these cases, but that could get ugly and
5080          * we'd probably do the same thing which is just not do anything else
5081          * and exit, so return 1 so the callers don't try to use other copies.
5082          */
5083         if (!em) {
5084                 btrfs_crit(fs_info, "No mapping for %Lu-%Lu", logical,
5085                             logical+len);
5086                 return 1;
5087         }
5088
5089         if (em->start > logical || em->start + em->len < logical) {
5090                 btrfs_crit(fs_info, "Invalid mapping for %Lu-%Lu, got "
5091                             "%Lu-%Lu", logical, logical+len, em->start,
5092                             em->start + em->len);
5093                 free_extent_map(em);
5094                 return 1;
5095         }
5096
5097         map = em->map_lookup;
5098         if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
5099                 ret = map->num_stripes;
5100         else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5101                 ret = map->sub_stripes;
5102         else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
5103                 ret = 2;
5104         else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5105                 ret = 3;
5106         else
5107                 ret = 1;
5108         free_extent_map(em);
5109
5110         btrfs_dev_replace_lock(&fs_info->dev_replace, 0);
5111         if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace))
5112                 ret++;
5113         btrfs_dev_replace_unlock(&fs_info->dev_replace, 0);
5114
5115         return ret;
5116 }
5117
5118 unsigned long btrfs_full_stripe_len(struct btrfs_root *root,
5119                                     struct btrfs_mapping_tree *map_tree,
5120                                     u64 logical)
5121 {
5122         struct extent_map *em;
5123         struct map_lookup *map;
5124         struct extent_map_tree *em_tree = &map_tree->map_tree;
5125         unsigned long len = root->sectorsize;
5126
5127         read_lock(&em_tree->lock);
5128         em = lookup_extent_mapping(em_tree, logical, len);
5129         read_unlock(&em_tree->lock);
5130         BUG_ON(!em);
5131
5132         BUG_ON(em->start > logical || em->start + em->len < logical);
5133         map = em->map_lookup;
5134         if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5135                 len = map->stripe_len * nr_data_stripes(map);
5136         free_extent_map(em);
5137         return len;
5138 }
5139
5140 int btrfs_is_parity_mirror(struct btrfs_mapping_tree *map_tree,
5141                            u64 logical, u64 len, int mirror_num)
5142 {
5143         struct extent_map *em;
5144         struct map_lookup *map;
5145         struct extent_map_tree *em_tree = &map_tree->map_tree;
5146         int ret = 0;
5147
5148         read_lock(&em_tree->lock);
5149         em = lookup_extent_mapping(em_tree, logical, len);
5150         read_unlock(&em_tree->lock);
5151         BUG_ON(!em);
5152
5153         BUG_ON(em->start > logical || em->start + em->len < logical);
5154         map = em->map_lookup;
5155         if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5156                 ret = 1;
5157         free_extent_map(em);
5158         return ret;
5159 }
5160
5161 static int find_live_mirror(struct btrfs_fs_info *fs_info,
5162                             struct map_lookup *map, int first, int num,
5163                             int optimal, int dev_replace_is_ongoing)
5164 {
5165         int i;
5166         int tolerance;
5167         struct btrfs_device *srcdev;
5168
5169         if (dev_replace_is_ongoing &&
5170             fs_info->dev_replace.cont_reading_from_srcdev_mode ==
5171              BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
5172                 srcdev = fs_info->dev_replace.srcdev;
5173         else
5174                 srcdev = NULL;
5175
5176         /*
5177          * try to avoid the drive that is the source drive for a
5178          * dev-replace procedure, only choose it if no other non-missing
5179          * mirror is available
5180          */
5181         for (tolerance = 0; tolerance < 2; tolerance++) {
5182                 if (map->stripes[optimal].dev->bdev &&
5183                     (tolerance || map->stripes[optimal].dev != srcdev))
5184                         return optimal;
5185                 for (i = first; i < first + num; i++) {
5186                         if (map->stripes[i].dev->bdev &&
5187                             (tolerance || map->stripes[i].dev != srcdev))
5188                                 return i;
5189                 }
5190         }
5191
5192         /* we couldn't find one that doesn't fail.  Just return something
5193          * and the io error handling code will clean up eventually
5194          */
5195         return optimal;
5196 }
5197
5198 static inline int parity_smaller(u64 a, u64 b)
5199 {
5200         return a > b;
5201 }
5202
5203 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
5204 static void sort_parity_stripes(struct btrfs_bio *bbio, int num_stripes)
5205 {
5206         struct btrfs_bio_stripe s;
5207         int i;
5208         u64 l;
5209         int again = 1;
5210
5211         while (again) {
5212                 again = 0;
5213                 for (i = 0; i < num_stripes - 1; i++) {
5214                         if (parity_smaller(bbio->raid_map[i],
5215                                            bbio->raid_map[i+1])) {
5216                                 s = bbio->stripes[i];
5217                                 l = bbio->raid_map[i];
5218                                 bbio->stripes[i] = bbio->stripes[i+1];
5219                                 bbio->raid_map[i] = bbio->raid_map[i+1];
5220                                 bbio->stripes[i+1] = s;
5221                                 bbio->raid_map[i+1] = l;
5222
5223                                 again = 1;
5224                         }
5225                 }
5226         }
5227 }
5228
5229 static struct btrfs_bio *alloc_btrfs_bio(int total_stripes, int real_stripes)
5230 {
5231         struct btrfs_bio *bbio = kzalloc(
5232                  /* the size of the btrfs_bio */
5233                 sizeof(struct btrfs_bio) +
5234                 /* plus the variable array for the stripes */
5235                 sizeof(struct btrfs_bio_stripe) * (total_stripes) +
5236                 /* plus the variable array for the tgt dev */
5237                 sizeof(int) * (real_stripes) +
5238                 /*
5239                  * plus the raid_map, which includes both the tgt dev
5240                  * and the stripes
5241                  */
5242                 sizeof(u64) * (total_stripes),
5243                 GFP_NOFS|__GFP_NOFAIL);
5244
5245         atomic_set(&bbio->error, 0);
5246         atomic_set(&bbio->refs, 1);
5247
5248         return bbio;
5249 }
5250
5251 void btrfs_get_bbio(struct btrfs_bio *bbio)
5252 {
5253         WARN_ON(!atomic_read(&bbio->refs));
5254         atomic_inc(&bbio->refs);
5255 }
5256
5257 void btrfs_put_bbio(struct btrfs_bio *bbio)
5258 {
5259         if (!bbio)
5260                 return;
5261         if (atomic_dec_and_test(&bbio->refs))
5262                 kfree(bbio);
5263 }
5264
5265 static int __btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
5266                              u64 logical, u64 *length,
5267                              struct btrfs_bio **bbio_ret,
5268                              int mirror_num, int need_raid_map)
5269 {
5270         struct extent_map *em;
5271         struct map_lookup *map;
5272         struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
5273         struct extent_map_tree *em_tree = &map_tree->map_tree;
5274         u64 offset;
5275         u64 stripe_offset;
5276         u64 stripe_end_offset;
5277         u64 stripe_nr;
5278         u64 stripe_nr_orig;
5279         u64 stripe_nr_end;
5280         u64 stripe_len;
5281         u32 stripe_index;
5282         int i;
5283         int ret = 0;
5284         int num_stripes;
5285         int max_errors = 0;
5286         int tgtdev_indexes = 0;
5287         struct btrfs_bio *bbio = NULL;
5288         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
5289         int dev_replace_is_ongoing = 0;
5290         int num_alloc_stripes;
5291         int patch_the_first_stripe_for_dev_replace = 0;
5292         u64 physical_to_patch_in_first_stripe = 0;
5293         u64 raid56_full_stripe_start = (u64)-1;
5294
5295         read_lock(&em_tree->lock);
5296         em = lookup_extent_mapping(em_tree, logical, *length);
5297         read_unlock(&em_tree->lock);
5298
5299         if (!em) {
5300                 btrfs_crit(fs_info, "unable to find logical %llu len %llu",
5301                         logical, *length);
5302                 return -EINVAL;
5303         }
5304
5305         if (em->start > logical || em->start + em->len < logical) {
5306                 btrfs_crit(fs_info, "found a bad mapping, wanted %Lu, "
5307                            "found %Lu-%Lu", logical, em->start,
5308                            em->start + em->len);
5309                 free_extent_map(em);
5310                 return -EINVAL;
5311         }
5312
5313         map = em->map_lookup;
5314         offset = logical - em->start;
5315
5316         stripe_len = map->stripe_len;
5317         stripe_nr = offset;
5318         /*
5319          * stripe_nr counts the total number of stripes we have to stride
5320          * to get to this block
5321          */
5322         stripe_nr = div64_u64(stripe_nr, stripe_len);
5323
5324         stripe_offset = stripe_nr * stripe_len;
5325         BUG_ON(offset < stripe_offset);
5326
5327         /* stripe_offset is the offset of this block in its stripe*/
5328         stripe_offset = offset - stripe_offset;
5329
5330         /* if we're here for raid56, we need to know the stripe aligned start */
5331         if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5332                 unsigned long full_stripe_len = stripe_len * nr_data_stripes(map);
5333                 raid56_full_stripe_start = offset;
5334
5335                 /* allow a write of a full stripe, but make sure we don't
5336                  * allow straddling of stripes
5337                  */
5338                 raid56_full_stripe_start = div64_u64(raid56_full_stripe_start,
5339                                 full_stripe_len);
5340                 raid56_full_stripe_start *= full_stripe_len;
5341         }
5342
5343         if (rw & REQ_DISCARD) {
5344                 /* we don't discard raid56 yet */
5345                 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5346                         ret = -EOPNOTSUPP;
5347                         goto out;
5348                 }
5349                 *length = min_t(u64, em->len - offset, *length);
5350         } else if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
5351                 u64 max_len;
5352                 /* For writes to RAID[56], allow a full stripeset across all disks.
5353                    For other RAID types and for RAID[56] reads, just allow a single
5354                    stripe (on a single disk). */
5355                 if ((map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
5356                     (rw & REQ_WRITE)) {
5357                         max_len = stripe_len * nr_data_stripes(map) -
5358                                 (offset - raid56_full_stripe_start);
5359                 } else {
5360                         /* we limit the length of each bio to what fits in a stripe */
5361                         max_len = stripe_len - stripe_offset;
5362                 }
5363                 *length = min_t(u64, em->len - offset, max_len);
5364         } else {
5365                 *length = em->len - offset;
5366         }
5367
5368         /* This is for when we're called from btrfs_merge_bio_hook() and all
5369            it cares about is the length */
5370         if (!bbio_ret)
5371                 goto out;
5372
5373         btrfs_dev_replace_lock(dev_replace, 0);
5374         dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
5375         if (!dev_replace_is_ongoing)
5376                 btrfs_dev_replace_unlock(dev_replace, 0);
5377         else
5378                 btrfs_dev_replace_set_lock_blocking(dev_replace);
5379
5380         if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
5381             !(rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) &&
5382             dev_replace->tgtdev != NULL) {
5383                 /*
5384                  * in dev-replace case, for repair case (that's the only
5385                  * case where the mirror is selected explicitly when
5386                  * calling btrfs_map_block), blocks left of the left cursor
5387                  * can also be read from the target drive.
5388                  * For REQ_GET_READ_MIRRORS, the target drive is added as
5389                  * the last one to the array of stripes. For READ, it also
5390                  * needs to be supported using the same mirror number.
5391                  * If the requested block is not left of the left cursor,
5392                  * EIO is returned. This can happen because btrfs_num_copies()
5393                  * returns one more in the dev-replace case.
5394                  */
5395                 u64 tmp_length = *length;
5396                 struct btrfs_bio *tmp_bbio = NULL;
5397                 int tmp_num_stripes;
5398                 u64 srcdev_devid = dev_replace->srcdev->devid;
5399                 int index_srcdev = 0;
5400                 int found = 0;
5401                 u64 physical_of_found = 0;
5402
5403                 ret = __btrfs_map_block(fs_info, REQ_GET_READ_MIRRORS,
5404                              logical, &tmp_length, &tmp_bbio, 0, 0);
5405                 if (ret) {
5406                         WARN_ON(tmp_bbio != NULL);
5407                         goto out;
5408                 }
5409
5410                 tmp_num_stripes = tmp_bbio->num_stripes;
5411                 if (mirror_num > tmp_num_stripes) {
5412                         /*
5413                          * REQ_GET_READ_MIRRORS does not contain this
5414                          * mirror, that means that the requested area
5415                          * is not left of the left cursor
5416                          */
5417                         ret = -EIO;
5418                         btrfs_put_bbio(tmp_bbio);
5419                         goto out;
5420                 }
5421
5422                 /*
5423                  * process the rest of the function using the mirror_num
5424                  * of the source drive. Therefore look it up first.
5425                  * At the end, patch the device pointer to the one of the
5426                  * target drive.
5427                  */
5428                 for (i = 0; i < tmp_num_stripes; i++) {
5429                         if (tmp_bbio->stripes[i].dev->devid != srcdev_devid)
5430                                 continue;
5431
5432                         /*
5433                          * In case of DUP, in order to keep it simple, only add
5434                          * the mirror with the lowest physical address
5435                          */
5436                         if (found &&
5437                             physical_of_found <= tmp_bbio->stripes[i].physical)
5438                                 continue;
5439
5440                         index_srcdev = i;
5441                         found = 1;
5442                         physical_of_found = tmp_bbio->stripes[i].physical;
5443                 }
5444
5445                 btrfs_put_bbio(tmp_bbio);
5446
5447                 if (!found) {
5448                         WARN_ON(1);
5449                         ret = -EIO;
5450                         goto out;
5451                 }
5452
5453                 mirror_num = index_srcdev + 1;
5454                 patch_the_first_stripe_for_dev_replace = 1;
5455                 physical_to_patch_in_first_stripe = physical_of_found;
5456         } else if (mirror_num > map->num_stripes) {
5457                 mirror_num = 0;
5458         }
5459
5460         num_stripes = 1;
5461         stripe_index = 0;
5462         stripe_nr_orig = stripe_nr;
5463         stripe_nr_end = ALIGN(offset + *length, map->stripe_len);
5464         stripe_nr_end = div_u64(stripe_nr_end, map->stripe_len);
5465         stripe_end_offset = stripe_nr_end * map->stripe_len -
5466                             (offset + *length);
5467
5468         if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5469                 if (rw & REQ_DISCARD)
5470                         num_stripes = min_t(u64, map->num_stripes,
5471                                             stripe_nr_end - stripe_nr_orig);
5472                 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5473                                 &stripe_index);
5474                 if (!(rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)))
5475                         mirror_num = 1;
5476         } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
5477                 if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS))
5478                         num_stripes = map->num_stripes;
5479                 else if (mirror_num)
5480                         stripe_index = mirror_num - 1;
5481                 else {
5482                         stripe_index = find_live_mirror(fs_info, map, 0,
5483                                             map->num_stripes,
5484                                             current->pid % map->num_stripes,
5485                                             dev_replace_is_ongoing);
5486                         mirror_num = stripe_index + 1;
5487                 }
5488
5489         } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
5490                 if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) {
5491                         num_stripes = map->num_stripes;
5492                 } else if (mirror_num) {
5493                         stripe_index = mirror_num - 1;
5494                 } else {
5495                         mirror_num = 1;
5496                 }
5497
5498         } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5499                 u32 factor = map->num_stripes / map->sub_stripes;
5500
5501                 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
5502                 stripe_index *= map->sub_stripes;
5503
5504                 if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS))
5505                         num_stripes = map->sub_stripes;
5506                 else if (rw & REQ_DISCARD)
5507                         num_stripes = min_t(u64, map->sub_stripes *
5508                                             (stripe_nr_end - stripe_nr_orig),
5509                                             map->num_stripes);
5510                 else if (mirror_num)
5511                         stripe_index += mirror_num - 1;
5512                 else {
5513                         int old_stripe_index = stripe_index;
5514                         stripe_index = find_live_mirror(fs_info, map,
5515                                               stripe_index,
5516                                               map->sub_stripes, stripe_index +
5517                                               current->pid % map->sub_stripes,
5518                                               dev_replace_is_ongoing);
5519                         mirror_num = stripe_index - old_stripe_index + 1;
5520                 }
5521
5522         } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5523                 if (need_raid_map &&
5524                     ((rw & (REQ_WRITE | REQ_GET_READ_MIRRORS)) ||
5525                      mirror_num > 1)) {
5526                         /* push stripe_nr back to the start of the full stripe */
5527                         stripe_nr = div_u64(raid56_full_stripe_start,
5528                                         stripe_len * nr_data_stripes(map));
5529
5530                         /* RAID[56] write or recovery. Return all stripes */
5531                         num_stripes = map->num_stripes;
5532                         max_errors = nr_parity_stripes(map);
5533
5534                         *length = map->stripe_len;
5535                         stripe_index = 0;
5536                         stripe_offset = 0;
5537                 } else {
5538                         /*
5539                          * Mirror #0 or #1 means the original data block.
5540                          * Mirror #2 is RAID5 parity block.
5541                          * Mirror #3 is RAID6 Q block.
5542                          */
5543                         stripe_nr = div_u64_rem(stripe_nr,
5544                                         nr_data_stripes(map), &stripe_index);
5545                         if (mirror_num > 1)
5546                                 stripe_index = nr_data_stripes(map) +
5547                                                 mirror_num - 2;
5548
5549                         /* We distribute the parity blocks across stripes */
5550                         div_u64_rem(stripe_nr + stripe_index, map->num_stripes,
5551                                         &stripe_index);
5552                         if (!(rw & (REQ_WRITE | REQ_DISCARD |
5553                                     REQ_GET_READ_MIRRORS)) && mirror_num <= 1)
5554                                 mirror_num = 1;
5555                 }
5556         } else {
5557                 /*
5558                  * after this, stripe_nr is the number of stripes on this
5559                  * device we have to walk to find the data, and stripe_index is
5560                  * the number of our device in the stripe array
5561                  */
5562                 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5563                                 &stripe_index);
5564                 mirror_num = stripe_index + 1;
5565         }
5566         BUG_ON(stripe_index >= map->num_stripes);
5567
5568         num_alloc_stripes = num_stripes;
5569         if (dev_replace_is_ongoing) {
5570                 if (rw & (REQ_WRITE | REQ_DISCARD))
5571                         num_alloc_stripes <<= 1;
5572                 if (rw & REQ_GET_READ_MIRRORS)
5573                         num_alloc_stripes++;
5574                 tgtdev_indexes = num_stripes;
5575         }
5576
5577         bbio = alloc_btrfs_bio(num_alloc_stripes, tgtdev_indexes);
5578         if (!bbio) {
5579                 ret = -ENOMEM;
5580                 goto out;
5581         }
5582         if (dev_replace_is_ongoing)
5583                 bbio->tgtdev_map = (int *)(bbio->stripes + num_alloc_stripes);
5584
5585         /* build raid_map */
5586         if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK &&
5587             need_raid_map && ((rw & (REQ_WRITE | REQ_GET_READ_MIRRORS)) ||
5588             mirror_num > 1)) {
5589                 u64 tmp;
5590                 unsigned rot;
5591
5592                 bbio->raid_map = (u64 *)((void *)bbio->stripes +
5593                                  sizeof(struct btrfs_bio_stripe) *
5594                                  num_alloc_stripes +
5595                                  sizeof(int) * tgtdev_indexes);
5596
5597                 /* Work out the disk rotation on this stripe-set */
5598                 div_u64_rem(stripe_nr, num_stripes, &rot);
5599
5600                 /* Fill in the logical address of each stripe */
5601                 tmp = stripe_nr * nr_data_stripes(map);
5602                 for (i = 0; i < nr_data_stripes(map); i++)
5603                         bbio->raid_map[(i+rot) % num_stripes] =
5604                                 em->start + (tmp + i) * map->stripe_len;
5605
5606                 bbio->raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE;
5607                 if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5608                         bbio->raid_map[(i+rot+1) % num_stripes] =
5609                                 RAID6_Q_STRIPE;
5610         }
5611
5612         if (rw & REQ_DISCARD) {
5613                 u32 factor = 0;
5614                 u32 sub_stripes = 0;
5615                 u64 stripes_per_dev = 0;
5616                 u32 remaining_stripes = 0;
5617                 u32 last_stripe = 0;
5618
5619                 if (map->type &
5620                     (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID10)) {
5621                         if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5622                                 sub_stripes = 1;
5623                         else
5624                                 sub_stripes = map->sub_stripes;
5625
5626                         factor = map->num_stripes / sub_stripes;
5627                         stripes_per_dev = div_u64_rem(stripe_nr_end -
5628                                                       stripe_nr_orig,
5629                                                       factor,
5630                                                       &remaining_stripes);
5631                         div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
5632                         last_stripe *= sub_stripes;
5633                 }
5634
5635                 for (i = 0; i < num_stripes; i++) {
5636                         bbio->stripes[i].physical =
5637                                 map->stripes[stripe_index].physical +
5638                                 stripe_offset + stripe_nr * map->stripe_len;
5639                         bbio->stripes[i].dev = map->stripes[stripe_index].dev;
5640
5641                         if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5642                                          BTRFS_BLOCK_GROUP_RAID10)) {
5643                                 bbio->stripes[i].length = stripes_per_dev *
5644                                                           map->stripe_len;
5645
5646                                 if (i / sub_stripes < remaining_stripes)
5647                                         bbio->stripes[i].length +=
5648                                                 map->stripe_len;
5649
5650                                 /*
5651                                  * Special for the first stripe and
5652                                  * the last stripe:
5653                                  *
5654                                  * |-------|...|-------|
5655                                  *     |----------|
5656                                  *    off     end_off
5657                                  */
5658                                 if (i < sub_stripes)
5659                                         bbio->stripes[i].length -=
5660                                                 stripe_offset;
5661
5662                                 if (stripe_index >= last_stripe &&
5663                                     stripe_index <= (last_stripe +
5664                                                      sub_stripes - 1))
5665                                         bbio->stripes[i].length -=
5666                                                 stripe_end_offset;
5667
5668                                 if (i == sub_stripes - 1)
5669                                         stripe_offset = 0;
5670                         } else
5671                                 bbio->stripes[i].length = *length;
5672
5673                         stripe_index++;
5674                         if (stripe_index == map->num_stripes) {
5675                                 /* This could only happen for RAID0/10 */
5676                                 stripe_index = 0;
5677                                 stripe_nr++;
5678                         }
5679                 }
5680         } else {
5681                 for (i = 0; i < num_stripes; i++) {
5682                         bbio->stripes[i].physical =
5683                                 map->stripes[stripe_index].physical +
5684                                 stripe_offset +
5685                                 stripe_nr * map->stripe_len;
5686                         bbio->stripes[i].dev =
5687                                 map->stripes[stripe_index].dev;
5688                         stripe_index++;
5689                 }
5690         }
5691
5692         if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS))
5693                 max_errors = btrfs_chunk_max_errors(map);
5694
5695         if (bbio->raid_map)
5696                 sort_parity_stripes(bbio, num_stripes);
5697
5698         tgtdev_indexes = 0;
5699         if (dev_replace_is_ongoing && (rw & (REQ_WRITE | REQ_DISCARD)) &&
5700             dev_replace->tgtdev != NULL) {
5701                 int index_where_to_add;
5702                 u64 srcdev_devid = dev_replace->srcdev->devid;
5703
5704                 /*
5705                  * duplicate the write operations while the dev replace
5706                  * procedure is running. Since the copying of the old disk
5707                  * to the new disk takes place at run time while the
5708                  * filesystem is mounted writable, the regular write
5709                  * operations to the old disk have to be duplicated to go
5710                  * to the new disk as well.
5711                  * Note that device->missing is handled by the caller, and
5712                  * that the write to the old disk is already set up in the
5713                  * stripes array.
5714                  */
5715                 index_where_to_add = num_stripes;
5716                 for (i = 0; i < num_stripes; i++) {
5717                         if (bbio->stripes[i].dev->devid == srcdev_devid) {
5718                                 /* write to new disk, too */
5719                                 struct btrfs_bio_stripe *new =
5720                                         bbio->stripes + index_where_to_add;
5721                                 struct btrfs_bio_stripe *old =
5722                                         bbio->stripes + i;
5723
5724                                 new->physical = old->physical;
5725                                 new->length = old->length;
5726                                 new->dev = dev_replace->tgtdev;
5727                                 bbio->tgtdev_map[i] = index_where_to_add;
5728                                 index_where_to_add++;
5729                                 max_errors++;
5730                                 tgtdev_indexes++;
5731                         }
5732                 }
5733                 num_stripes = index_where_to_add;
5734         } else if (dev_replace_is_ongoing && (rw & REQ_GET_READ_MIRRORS) &&
5735                    dev_replace->tgtdev != NULL) {
5736                 u64 srcdev_devid = dev_replace->srcdev->devid;
5737                 int index_srcdev = 0;
5738                 int found = 0;
5739                 u64 physical_of_found = 0;
5740
5741                 /*
5742                  * During the dev-replace procedure, the target drive can
5743                  * also be used to read data in case it is needed to repair
5744                  * a corrupt block elsewhere. This is possible if the
5745                  * requested area is left of the left cursor. In this area,
5746                  * the target drive is a full copy of the source drive.
5747                  */
5748                 for (i = 0; i < num_stripes; i++) {
5749                         if (bbio->stripes[i].dev->devid == srcdev_devid) {
5750                                 /*
5751                                  * In case of DUP, in order to keep it
5752                                  * simple, only add the mirror with the
5753                                  * lowest physical address
5754                                  */
5755                                 if (found &&
5756                                     physical_of_found <=
5757                                      bbio->stripes[i].physical)
5758                                         continue;
5759                                 index_srcdev = i;
5760                                 found = 1;
5761                                 physical_of_found = bbio->stripes[i].physical;
5762                         }
5763                 }
5764                 if (found) {
5765                         if (physical_of_found + map->stripe_len <=
5766                             dev_replace->cursor_left) {
5767                                 struct btrfs_bio_stripe *tgtdev_stripe =
5768                                         bbio->stripes + num_stripes;
5769
5770                                 tgtdev_stripe->physical = physical_of_found;
5771                                 tgtdev_stripe->length =
5772                                         bbio->stripes[index_srcdev].length;
5773                                 tgtdev_stripe->dev = dev_replace->tgtdev;
5774                                 bbio->tgtdev_map[index_srcdev] = num_stripes;
5775
5776                                 tgtdev_indexes++;
5777                                 num_stripes++;
5778                         }
5779                 }
5780         }
5781
5782         *bbio_ret = bbio;
5783         bbio->map_type = map->type;
5784         bbio->num_stripes = num_stripes;
5785         bbio->max_errors = max_errors;
5786         bbio->mirror_num = mirror_num;
5787         bbio->num_tgtdevs = tgtdev_indexes;
5788
5789         /*
5790          * this is the case that REQ_READ && dev_replace_is_ongoing &&
5791          * mirror_num == num_stripes + 1 && dev_replace target drive is
5792          * available as a mirror
5793          */
5794         if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) {
5795                 WARN_ON(num_stripes > 1);
5796                 bbio->stripes[0].dev = dev_replace->tgtdev;
5797                 bbio->stripes[0].physical = physical_to_patch_in_first_stripe;
5798                 bbio->mirror_num = map->num_stripes + 1;
5799         }
5800 out:
5801         if (dev_replace_is_ongoing) {
5802                 btrfs_dev_replace_clear_lock_blocking(dev_replace);
5803                 btrfs_dev_replace_unlock(dev_replace, 0);
5804         }
5805         free_extent_map(em);
5806         return ret;
5807 }
5808
5809 int btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
5810                       u64 logical, u64 *length,
5811                       struct btrfs_bio **bbio_ret, int mirror_num)
5812 {
5813         return __btrfs_map_block(fs_info, rw, logical, length, bbio_ret,
5814                                  mirror_num, 0);
5815 }
5816
5817 /* For Scrub/replace */
5818 int btrfs_map_sblock(struct btrfs_fs_info *fs_info, int rw,
5819                      u64 logical, u64 *length,
5820                      struct btrfs_bio **bbio_ret, int mirror_num,
5821                      int need_raid_map)
5822 {
5823         return __btrfs_map_block(fs_info, rw, logical, length, bbio_ret,
5824                                  mirror_num, need_raid_map);
5825 }
5826
5827 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
5828                      u64 chunk_start, u64 physical, u64 devid,
5829                      u64 **logical, int *naddrs, int *stripe_len)
5830 {
5831         struct extent_map_tree *em_tree = &map_tree->map_tree;
5832         struct extent_map *em;
5833         struct map_lookup *map;
5834         u64 *buf;
5835         u64 bytenr;
5836         u64 length;
5837         u64 stripe_nr;
5838         u64 rmap_len;
5839         int i, j, nr = 0;
5840
5841         read_lock(&em_tree->lock);
5842         em = lookup_extent_mapping(em_tree, chunk_start, 1);
5843         read_unlock(&em_tree->lock);
5844
5845         if (!em) {
5846                 printk(KERN_ERR "BTRFS: couldn't find em for chunk %Lu\n",
5847                        chunk_start);
5848                 return -EIO;
5849         }
5850
5851         if (em->start != chunk_start) {
5852                 printk(KERN_ERR "BTRFS: bad chunk start, em=%Lu, wanted=%Lu\n",
5853                        em->start, chunk_start);
5854                 free_extent_map(em);
5855                 return -EIO;
5856         }
5857         map = em->map_lookup;
5858
5859         length = em->len;
5860         rmap_len = map->stripe_len;
5861
5862         if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5863                 length = div_u64(length, map->num_stripes / map->sub_stripes);
5864         else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5865                 length = div_u64(length, map->num_stripes);
5866         else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5867                 length = div_u64(length, nr_data_stripes(map));
5868                 rmap_len = map->stripe_len * nr_data_stripes(map);
5869         }
5870
5871         buf = kcalloc(map->num_stripes, sizeof(u64), GFP_NOFS);
5872         BUG_ON(!buf); /* -ENOMEM */
5873
5874         for (i = 0; i < map->num_stripes; i++) {
5875                 if (devid && map->stripes[i].dev->devid != devid)
5876                         continue;
5877                 if (map->stripes[i].physical > physical ||
5878                     map->stripes[i].physical + length <= physical)
5879                         continue;
5880
5881                 stripe_nr = physical - map->stripes[i].physical;
5882                 stripe_nr = div_u64(stripe_nr, map->stripe_len);
5883
5884                 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5885                         stripe_nr = stripe_nr * map->num_stripes + i;
5886                         stripe_nr = div_u64(stripe_nr, map->sub_stripes);
5887                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5888                         stripe_nr = stripe_nr * map->num_stripes + i;
5889                 } /* else if RAID[56], multiply by nr_data_stripes().
5890                    * Alternatively, just use rmap_len below instead of
5891                    * map->stripe_len */
5892
5893                 bytenr = chunk_start + stripe_nr * rmap_len;
5894                 WARN_ON(nr >= map->num_stripes);
5895                 for (j = 0; j < nr; j++) {
5896                         if (buf[j] == bytenr)
5897                                 break;
5898                 }
5899                 if (j == nr) {
5900                         WARN_ON(nr >= map->num_stripes);
5901                         buf[nr++] = bytenr;
5902                 }
5903         }
5904
5905         *logical = buf;
5906         *naddrs = nr;
5907         *stripe_len = rmap_len;
5908
5909         free_extent_map(em);
5910         return 0;
5911 }
5912
5913 static inline void btrfs_end_bbio(struct btrfs_bio *bbio, struct bio *bio)
5914 {
5915         bio->bi_private = bbio->private;
5916         bio->bi_end_io = bbio->end_io;
5917         bio_endio(bio);
5918
5919         btrfs_put_bbio(bbio);
5920 }
5921
5922 static void btrfs_end_bio(struct bio *bio)
5923 {
5924         struct btrfs_bio *bbio = bio->bi_private;
5925         int is_orig_bio = 0;
5926
5927         if (bio->bi_error) {
5928                 atomic_inc(&bbio->error);
5929                 if (bio->bi_error == -EIO || bio->bi_error == -EREMOTEIO) {
5930                         unsigned int stripe_index =
5931                                 btrfs_io_bio(bio)->stripe_index;
5932                         struct btrfs_device *dev;
5933
5934                         BUG_ON(stripe_index >= bbio->num_stripes);
5935                         dev = bbio->stripes[stripe_index].dev;
5936                         if (dev->bdev) {
5937                                 if (bio->bi_rw & WRITE)
5938                                         btrfs_dev_stat_inc(dev,
5939                                                 BTRFS_DEV_STAT_WRITE_ERRS);
5940                                 else
5941                                         btrfs_dev_stat_inc(dev,
5942                                                 BTRFS_DEV_STAT_READ_ERRS);
5943                                 if ((bio->bi_rw & WRITE_FLUSH) == WRITE_FLUSH)
5944                                         btrfs_dev_stat_inc(dev,
5945                                                 BTRFS_DEV_STAT_FLUSH_ERRS);
5946                                 btrfs_dev_stat_print_on_error(dev);
5947                         }
5948                 }
5949         }
5950
5951         if (bio == bbio->orig_bio)
5952                 is_orig_bio = 1;
5953
5954         btrfs_bio_counter_dec(bbio->fs_info);
5955
5956         if (atomic_dec_and_test(&bbio->stripes_pending)) {
5957                 if (!is_orig_bio) {
5958                         bio_put(bio);
5959                         bio = bbio->orig_bio;
5960                 }
5961
5962                 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
5963                 /* only send an error to the higher layers if it is
5964                  * beyond the tolerance of the btrfs bio
5965                  */
5966                 if (atomic_read(&bbio->error) > bbio->max_errors) {
5967                         bio->bi_error = -EIO;
5968                 } else {
5969                         /*
5970                          * this bio is actually up to date, we didn't
5971                          * go over the max number of errors
5972                          */
5973                         bio->bi_error = 0;
5974                 }
5975
5976                 btrfs_end_bbio(bbio, bio);
5977         } else if (!is_orig_bio) {
5978                 bio_put(bio);
5979         }
5980 }
5981
5982 /*
5983  * see run_scheduled_bios for a description of why bios are collected for
5984  * async submit.
5985  *
5986  * This will add one bio to the pending list for a device and make sure
5987  * the work struct is scheduled.
5988  */
5989 static noinline void btrfs_schedule_bio(struct btrfs_root *root,
5990                                         struct btrfs_device *device,
5991                                         int rw, struct bio *bio)
5992 {
5993         int should_queue = 1;
5994         struct btrfs_pending_bios *pending_bios;
5995
5996         if (device->missing || !device->bdev) {
5997                 bio_io_error(bio);
5998                 return;
5999         }
6000
6001         /* don't bother with additional async steps for reads, right now */
6002         if (!(rw & REQ_WRITE)) {
6003                 bio_get(bio);
6004                 btrfsic_submit_bio(rw, bio);
6005                 bio_put(bio);
6006                 return;
6007         }
6008
6009         /*
6010          * nr_async_bios allows us to reliably return congestion to the
6011          * higher layers.  Otherwise, the async bio makes it appear we have
6012          * made progress against dirty pages when we've really just put it
6013          * on a queue for later
6014          */
6015         atomic_inc(&root->fs_info->nr_async_bios);
6016         WARN_ON(bio->bi_next);
6017         bio->bi_next = NULL;
6018         bio->bi_rw |= rw;
6019
6020         spin_lock(&device->io_lock);
6021         if (bio->bi_rw & REQ_SYNC)
6022                 pending_bios = &device->pending_sync_bios;
6023         else
6024                 pending_bios = &device->pending_bios;
6025
6026         if (pending_bios->tail)
6027                 pending_bios->tail->bi_next = bio;
6028
6029         pending_bios->tail = bio;
6030         if (!pending_bios->head)
6031                 pending_bios->head = bio;
6032         if (device->running_pending)
6033                 should_queue = 0;
6034
6035         spin_unlock(&device->io_lock);
6036
6037         if (should_queue)
6038                 btrfs_queue_work(root->fs_info->submit_workers,
6039                                  &device->work);
6040 }
6041
6042 static void submit_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio,
6043                               struct bio *bio, u64 physical, int dev_nr,
6044                               int rw, int async)
6045 {
6046         struct btrfs_device *dev = bbio->stripes[dev_nr].dev;
6047
6048         bio->bi_private = bbio;
6049         btrfs_io_bio(bio)->stripe_index = dev_nr;
6050         bio->bi_end_io = btrfs_end_bio;
6051         bio->bi_iter.bi_sector = physical >> 9;
6052 #ifdef DEBUG
6053         {
6054                 struct rcu_string *name;
6055
6056                 rcu_read_lock();
6057                 name = rcu_dereference(dev->name);
6058                 pr_debug("btrfs_map_bio: rw %d, sector=%llu, dev=%lu "
6059                          "(%s id %llu), size=%u\n", rw,
6060                          (u64)bio->bi_iter.bi_sector, (u_long)dev->bdev->bd_dev,
6061                          name->str, dev->devid, bio->bi_iter.bi_size);
6062                 rcu_read_unlock();
6063         }
6064 #endif
6065         bio->bi_bdev = dev->bdev;
6066
6067         btrfs_bio_counter_inc_noblocked(root->fs_info);
6068
6069         if (async)
6070                 btrfs_schedule_bio(root, dev, rw, bio);
6071         else
6072                 btrfsic_submit_bio(rw, bio);
6073 }
6074
6075 static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical)
6076 {
6077         atomic_inc(&bbio->error);
6078         if (atomic_dec_and_test(&bbio->stripes_pending)) {
6079                 /* Shoud be the original bio. */
6080                 WARN_ON(bio != bbio->orig_bio);
6081
6082                 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
6083                 bio->bi_iter.bi_sector = logical >> 9;
6084                 bio->bi_error = -EIO;
6085                 btrfs_end_bbio(bbio, bio);
6086         }
6087 }
6088
6089 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
6090                   int mirror_num, int async_submit)
6091 {
6092         struct btrfs_device *dev;
6093         struct bio *first_bio = bio;
6094         u64 logical = (u64)bio->bi_iter.bi_sector << 9;
6095         u64 length = 0;
6096         u64 map_length;
6097         int ret;
6098         int dev_nr;
6099         int total_devs;
6100         struct btrfs_bio *bbio = NULL;
6101
6102         length = bio->bi_iter.bi_size;
6103         map_length = length;
6104
6105         btrfs_bio_counter_inc_blocked(root->fs_info);
6106         ret = __btrfs_map_block(root->fs_info, rw, logical, &map_length, &bbio,
6107                               mirror_num, 1);
6108         if (ret) {
6109                 btrfs_bio_counter_dec(root->fs_info);
6110                 return ret;
6111         }
6112
6113         total_devs = bbio->num_stripes;
6114         bbio->orig_bio = first_bio;
6115         bbio->private = first_bio->bi_private;
6116         bbio->end_io = first_bio->bi_end_io;
6117         bbio->fs_info = root->fs_info;
6118         atomic_set(&bbio->stripes_pending, bbio->num_stripes);
6119
6120         if ((bbio->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
6121             ((rw & WRITE) || (mirror_num > 1))) {
6122                 /* In this case, map_length has been set to the length of
6123                    a single stripe; not the whole write */
6124                 if (rw & WRITE) {
6125                         ret = raid56_parity_write(root, bio, bbio, map_length);
6126                 } else {
6127                         ret = raid56_parity_recover(root, bio, bbio, map_length,
6128                                                     mirror_num, 1);
6129                 }
6130
6131                 btrfs_bio_counter_dec(root->fs_info);
6132                 return ret;
6133         }
6134
6135         if (map_length < length) {
6136                 btrfs_crit(root->fs_info, "mapping failed logical %llu bio len %llu len %llu",
6137                         logical, length, map_length);
6138                 BUG();
6139         }
6140
6141         for (dev_nr = 0; dev_nr < total_devs; dev_nr++) {
6142                 dev = bbio->stripes[dev_nr].dev;
6143                 if (!dev || !dev->bdev || (rw & WRITE && !dev->writeable)) {
6144                         bbio_error(bbio, first_bio, logical);
6145                         continue;
6146                 }
6147
6148                 if (dev_nr < total_devs - 1) {
6149                         bio = btrfs_bio_clone(first_bio, GFP_NOFS);
6150                         BUG_ON(!bio); /* -ENOMEM */
6151                 } else
6152                         bio = first_bio;
6153
6154                 submit_stripe_bio(root, bbio, bio,
6155                                   bbio->stripes[dev_nr].physical, dev_nr, rw,
6156                                   async_submit);
6157         }
6158         btrfs_bio_counter_dec(root->fs_info);
6159         return 0;
6160 }
6161
6162 struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid,
6163                                        u8 *uuid, u8 *fsid)
6164 {
6165         struct btrfs_device *device;
6166         struct btrfs_fs_devices *cur_devices;
6167
6168         cur_devices = fs_info->fs_devices;
6169         while (cur_devices) {
6170                 if (!fsid ||
6171                     !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
6172                         device = __find_device(&cur_devices->devices,
6173                                                devid, uuid);
6174                         if (device)
6175                                 return device;
6176                 }
6177                 cur_devices = cur_devices->seed;
6178         }
6179         return NULL;
6180 }
6181
6182 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
6183                                             struct btrfs_fs_devices *fs_devices,
6184                                             u64 devid, u8 *dev_uuid)
6185 {
6186         struct btrfs_device *device;
6187
6188         device = btrfs_alloc_device(NULL, &devid, dev_uuid);
6189         if (IS_ERR(device))
6190                 return NULL;
6191
6192         list_add(&device->dev_list, &fs_devices->devices);
6193         device->fs_devices = fs_devices;
6194         fs_devices->num_devices++;
6195
6196         device->missing = 1;
6197         fs_devices->missing_devices++;
6198
6199         return device;
6200 }
6201
6202 /**
6203  * btrfs_alloc_device - allocate struct btrfs_device
6204  * @fs_info:    used only for generating a new devid, can be NULL if
6205  *              devid is provided (i.e. @devid != NULL).
6206  * @devid:      a pointer to devid for this device.  If NULL a new devid
6207  *              is generated.
6208  * @uuid:       a pointer to UUID for this device.  If NULL a new UUID
6209  *              is generated.
6210  *
6211  * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
6212  * on error.  Returned struct is not linked onto any lists and can be
6213  * destroyed with kfree() right away.
6214  */
6215 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
6216                                         const u64 *devid,
6217                                         const u8 *uuid)
6218 {
6219         struct btrfs_device *dev;
6220         u64 tmp;
6221
6222         if (WARN_ON(!devid && !fs_info))
6223                 return ERR_PTR(-EINVAL);
6224
6225         dev = __alloc_device();
6226         if (IS_ERR(dev))
6227                 return dev;
6228
6229         if (devid)
6230                 tmp = *devid;
6231         else {
6232                 int ret;
6233
6234                 ret = find_next_devid(fs_info, &tmp);
6235                 if (ret) {
6236                         kfree(dev);
6237                         return ERR_PTR(ret);
6238                 }
6239         }
6240         dev->devid = tmp;
6241
6242         if (uuid)
6243                 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE);
6244         else
6245                 generate_random_uuid(dev->uuid);
6246
6247         btrfs_init_work(&dev->work, btrfs_submit_helper,
6248                         pending_bios_fn, NULL, NULL);
6249
6250         return dev;
6251 }
6252
6253 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
6254                           struct extent_buffer *leaf,
6255                           struct btrfs_chunk *chunk)
6256 {
6257         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
6258         struct map_lookup *map;
6259         struct extent_map *em;
6260         u64 logical;
6261         u64 length;
6262         u64 stripe_len;
6263         u64 devid;
6264         u8 uuid[BTRFS_UUID_SIZE];
6265         int num_stripes;
6266         int ret;
6267         int i;
6268
6269         logical = key->offset;
6270         length = btrfs_chunk_length(leaf, chunk);
6271         stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6272         num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6273         /* Validation check */
6274         if (!num_stripes) {
6275                 btrfs_err(root->fs_info, "invalid chunk num_stripes: %u",
6276                           num_stripes);
6277                 return -EIO;
6278         }
6279         if (!IS_ALIGNED(logical, root->sectorsize)) {
6280                 btrfs_err(root->fs_info,
6281                           "invalid chunk logical %llu", logical);
6282                 return -EIO;
6283         }
6284         if (!length || !IS_ALIGNED(length, root->sectorsize)) {
6285                 btrfs_err(root->fs_info,
6286                         "invalid chunk length %llu", length);
6287                 return -EIO;
6288         }
6289         if (!is_power_of_2(stripe_len)) {
6290                 btrfs_err(root->fs_info, "invalid chunk stripe length: %llu",
6291                           stripe_len);
6292                 return -EIO;
6293         }
6294         if (~(BTRFS_BLOCK_GROUP_TYPE_MASK | BTRFS_BLOCK_GROUP_PROFILE_MASK) &
6295             btrfs_chunk_type(leaf, chunk)) {
6296                 btrfs_err(root->fs_info, "unrecognized chunk type: %llu",
6297                           ~(BTRFS_BLOCK_GROUP_TYPE_MASK |
6298                             BTRFS_BLOCK_GROUP_PROFILE_MASK) &
6299                           btrfs_chunk_type(leaf, chunk));
6300                 return -EIO;
6301         }
6302
6303         read_lock(&map_tree->map_tree.lock);
6304         em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
6305         read_unlock(&map_tree->map_tree.lock);
6306
6307         /* already mapped? */
6308         if (em && em->start <= logical && em->start + em->len > logical) {
6309                 free_extent_map(em);
6310                 return 0;
6311         } else if (em) {
6312                 free_extent_map(em);
6313         }
6314
6315         em = alloc_extent_map();
6316         if (!em)
6317                 return -ENOMEM;
6318         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
6319         if (!map) {
6320                 free_extent_map(em);
6321                 return -ENOMEM;
6322         }
6323
6324         set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
6325         em->map_lookup = map;
6326         em->start = logical;
6327         em->len = length;
6328         em->orig_start = 0;
6329         em->block_start = 0;
6330         em->block_len = em->len;
6331
6332         map->num_stripes = num_stripes;
6333         map->io_width = btrfs_chunk_io_width(leaf, chunk);
6334         map->io_align = btrfs_chunk_io_align(leaf, chunk);
6335         map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
6336         map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6337         map->type = btrfs_chunk_type(leaf, chunk);
6338         map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
6339         for (i = 0; i < num_stripes; i++) {
6340                 map->stripes[i].physical =
6341                         btrfs_stripe_offset_nr(leaf, chunk, i);
6342                 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
6343                 read_extent_buffer(leaf, uuid, (unsigned long)
6344                                    btrfs_stripe_dev_uuid_nr(chunk, i),
6345                                    BTRFS_UUID_SIZE);
6346                 map->stripes[i].dev = btrfs_find_device(root->fs_info, devid,
6347                                                         uuid, NULL);
6348                 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
6349                         free_extent_map(em);
6350                         return -EIO;
6351                 }
6352                 if (!map->stripes[i].dev) {
6353                         map->stripes[i].dev =
6354                                 add_missing_dev(root, root->fs_info->fs_devices,
6355                                                 devid, uuid);
6356                         if (!map->stripes[i].dev) {
6357                                 free_extent_map(em);
6358                                 return -EIO;
6359                         }
6360                         btrfs_warn(root->fs_info, "devid %llu uuid %pU is missing",
6361                                                 devid, uuid);
6362                 }
6363                 map->stripes[i].dev->in_fs_metadata = 1;
6364         }
6365
6366         write_lock(&map_tree->map_tree.lock);
6367         ret = add_extent_mapping(&map_tree->map_tree, em, 0);
6368         write_unlock(&map_tree->map_tree.lock);
6369         BUG_ON(ret); /* Tree corruption */
6370         free_extent_map(em);
6371
6372         return 0;
6373 }
6374
6375 static void fill_device_from_item(struct extent_buffer *leaf,
6376                                  struct btrfs_dev_item *dev_item,
6377                                  struct btrfs_device *device)
6378 {
6379         unsigned long ptr;
6380
6381         device->devid = btrfs_device_id(leaf, dev_item);
6382         device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
6383         device->total_bytes = device->disk_total_bytes;
6384         device->commit_total_bytes = device->disk_total_bytes;
6385         device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
6386         device->commit_bytes_used = device->bytes_used;
6387         device->type = btrfs_device_type(leaf, dev_item);
6388         device->io_align = btrfs_device_io_align(leaf, dev_item);
6389         device->io_width = btrfs_device_io_width(leaf, dev_item);
6390         device->sector_size = btrfs_device_sector_size(leaf, dev_item);
6391         WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
6392         device->is_tgtdev_for_dev_replace = 0;
6393
6394         ptr = btrfs_device_uuid(dev_item);
6395         read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
6396 }
6397
6398 static struct btrfs_fs_devices *open_seed_devices(struct btrfs_root *root,
6399                                                   u8 *fsid)
6400 {
6401         struct btrfs_fs_devices *fs_devices;
6402         int ret;
6403
6404         BUG_ON(!mutex_is_locked(&uuid_mutex));
6405
6406         fs_devices = root->fs_info->fs_devices->seed;
6407         while (fs_devices) {
6408                 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE))
6409                         return fs_devices;
6410
6411                 fs_devices = fs_devices->seed;
6412         }
6413
6414         fs_devices = find_fsid(fsid);
6415         if (!fs_devices) {
6416                 if (!btrfs_test_opt(root, DEGRADED))
6417                         return ERR_PTR(-ENOENT);
6418
6419                 fs_devices = alloc_fs_devices(fsid);
6420                 if (IS_ERR(fs_devices))
6421                         return fs_devices;
6422
6423                 fs_devices->seeding = 1;
6424                 fs_devices->opened = 1;
6425                 return fs_devices;
6426         }
6427
6428         fs_devices = clone_fs_devices(fs_devices);
6429         if (IS_ERR(fs_devices))
6430                 return fs_devices;
6431
6432         ret = __btrfs_open_devices(fs_devices, FMODE_READ,
6433                                    root->fs_info->bdev_holder);
6434         if (ret) {
6435                 free_fs_devices(fs_devices);
6436                 fs_devices = ERR_PTR(ret);
6437                 goto out;
6438         }
6439
6440         if (!fs_devices->seeding) {
6441                 __btrfs_close_devices(fs_devices);
6442                 free_fs_devices(fs_devices);
6443                 fs_devices = ERR_PTR(-EINVAL);
6444                 goto out;
6445         }
6446
6447         fs_devices->seed = root->fs_info->fs_devices->seed;
6448         root->fs_info->fs_devices->seed = fs_devices;
6449 out:
6450         return fs_devices;
6451 }
6452
6453 static int read_one_dev(struct btrfs_root *root,
6454                         struct extent_buffer *leaf,
6455                         struct btrfs_dev_item *dev_item)
6456 {
6457         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
6458         struct btrfs_device *device;
6459         u64 devid;
6460         int ret;
6461         u8 fs_uuid[BTRFS_UUID_SIZE];
6462         u8 dev_uuid[BTRFS_UUID_SIZE];
6463
6464         devid = btrfs_device_id(leaf, dev_item);
6465         read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
6466                            BTRFS_UUID_SIZE);
6467         read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
6468                            BTRFS_UUID_SIZE);
6469
6470         if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
6471                 fs_devices = open_seed_devices(root, fs_uuid);
6472                 if (IS_ERR(fs_devices))
6473                         return PTR_ERR(fs_devices);
6474         }
6475
6476         device = btrfs_find_device(root->fs_info, devid, dev_uuid, fs_uuid);
6477         if (!device) {
6478                 if (!btrfs_test_opt(root, DEGRADED))
6479                         return -EIO;
6480
6481                 device = add_missing_dev(root, fs_devices, devid, dev_uuid);
6482                 if (!device)
6483                         return -ENOMEM;
6484                 btrfs_warn(root->fs_info, "devid %llu uuid %pU missing",
6485                                 devid, dev_uuid);
6486         } else {
6487                 if (!device->bdev && !btrfs_test_opt(root, DEGRADED))
6488                         return -EIO;
6489
6490                 if(!device->bdev && !device->missing) {
6491                         /*
6492                          * this happens when a device that was properly setup
6493                          * in the device info lists suddenly goes bad.
6494                          * device->bdev is NULL, and so we have to set
6495                          * device->missing to one here
6496                          */
6497                         device->fs_devices->missing_devices++;
6498                         device->missing = 1;
6499                 }
6500
6501                 /* Move the device to its own fs_devices */
6502                 if (device->fs_devices != fs_devices) {
6503                         ASSERT(device->missing);
6504
6505                         list_move(&device->dev_list, &fs_devices->devices);
6506                         device->fs_devices->num_devices--;
6507                         fs_devices->num_devices++;
6508
6509                         device->fs_devices->missing_devices--;
6510                         fs_devices->missing_devices++;
6511
6512                         device->fs_devices = fs_devices;
6513                 }
6514         }
6515
6516         if (device->fs_devices != root->fs_info->fs_devices) {
6517                 BUG_ON(device->writeable);
6518                 if (device->generation !=
6519                     btrfs_device_generation(leaf, dev_item))
6520                         return -EINVAL;
6521         }
6522
6523         fill_device_from_item(leaf, dev_item, device);
6524         device->in_fs_metadata = 1;
6525         if (device->writeable && !device->is_tgtdev_for_dev_replace) {
6526                 device->fs_devices->total_rw_bytes += device->total_bytes;
6527                 spin_lock(&root->fs_info->free_chunk_lock);
6528                 root->fs_info->free_chunk_space += device->total_bytes -
6529                         device->bytes_used;
6530                 spin_unlock(&root->fs_info->free_chunk_lock);
6531         }
6532         ret = 0;
6533         return ret;
6534 }
6535
6536 int btrfs_read_sys_array(struct btrfs_root *root)
6537 {
6538         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
6539         struct extent_buffer *sb;
6540         struct btrfs_disk_key *disk_key;
6541         struct btrfs_chunk *chunk;
6542         u8 *array_ptr;
6543         unsigned long sb_array_offset;
6544         int ret = 0;
6545         u32 num_stripes;
6546         u32 array_size;
6547         u32 len = 0;
6548         u32 cur_offset;
6549         struct btrfs_key key;
6550
6551         ASSERT(BTRFS_SUPER_INFO_SIZE <= root->nodesize);
6552         /*
6553          * This will create extent buffer of nodesize, superblock size is
6554          * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will
6555          * overallocate but we can keep it as-is, only the first page is used.
6556          */
6557         sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET);
6558         if (!sb)
6559                 return -ENOMEM;
6560         set_extent_buffer_uptodate(sb);
6561         btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
6562         /*
6563          * The sb extent buffer is artifical and just used to read the system array.
6564          * set_extent_buffer_uptodate() call does not properly mark all it's
6565          * pages up-to-date when the page is larger: extent does not cover the
6566          * whole page and consequently check_page_uptodate does not find all
6567          * the page's extents up-to-date (the hole beyond sb),
6568          * write_extent_buffer then triggers a WARN_ON.
6569          *
6570          * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
6571          * but sb spans only this function. Add an explicit SetPageUptodate call
6572          * to silence the warning eg. on PowerPC 64.
6573          */
6574         if (PAGE_SIZE > BTRFS_SUPER_INFO_SIZE)
6575                 SetPageUptodate(sb->pages[0]);
6576
6577         write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
6578         array_size = btrfs_super_sys_array_size(super_copy);
6579
6580         array_ptr = super_copy->sys_chunk_array;
6581         sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array);
6582         cur_offset = 0;
6583
6584         while (cur_offset < array_size) {
6585                 disk_key = (struct btrfs_disk_key *)array_ptr;
6586                 len = sizeof(*disk_key);
6587                 if (cur_offset + len > array_size)
6588                         goto out_short_read;
6589
6590                 btrfs_disk_key_to_cpu(&key, disk_key);
6591
6592                 array_ptr += len;
6593                 sb_array_offset += len;
6594                 cur_offset += len;
6595
6596                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
6597                         chunk = (struct btrfs_chunk *)sb_array_offset;
6598                         /*
6599                          * At least one btrfs_chunk with one stripe must be
6600                          * present, exact stripe count check comes afterwards
6601                          */
6602                         len = btrfs_chunk_item_size(1);
6603                         if (cur_offset + len > array_size)
6604                                 goto out_short_read;
6605
6606                         num_stripes = btrfs_chunk_num_stripes(sb, chunk);
6607                         if (!num_stripes) {
6608                                 printk(KERN_ERR
6609             "BTRFS: invalid number of stripes %u in sys_array at offset %u\n",
6610                                         num_stripes, cur_offset);
6611                                 ret = -EIO;
6612                                 break;
6613                         }
6614
6615                         len = btrfs_chunk_item_size(num_stripes);
6616                         if (cur_offset + len > array_size)
6617                                 goto out_short_read;
6618
6619                         ret = read_one_chunk(root, &key, sb, chunk);
6620                         if (ret)
6621                                 break;
6622                 } else {
6623                         printk(KERN_ERR
6624                 "BTRFS: unexpected item type %u in sys_array at offset %u\n",
6625                                 (u32)key.type, cur_offset);
6626                         ret = -EIO;
6627                         break;
6628                 }
6629                 array_ptr += len;
6630                 sb_array_offset += len;
6631                 cur_offset += len;
6632         }
6633         free_extent_buffer(sb);
6634         return ret;
6635
6636 out_short_read:
6637         printk(KERN_ERR "BTRFS: sys_array too short to read %u bytes at offset %u\n",
6638                         len, cur_offset);
6639         free_extent_buffer(sb);
6640         return -EIO;
6641 }
6642
6643 int btrfs_read_chunk_tree(struct btrfs_root *root)
6644 {
6645         struct btrfs_path *path;
6646         struct extent_buffer *leaf;
6647         struct btrfs_key key;
6648         struct btrfs_key found_key;
6649         int ret;
6650         int slot;
6651
6652         root = root->fs_info->chunk_root;
6653
6654         path = btrfs_alloc_path();
6655         if (!path)
6656                 return -ENOMEM;
6657
6658         mutex_lock(&uuid_mutex);
6659         lock_chunks(root);
6660
6661         /*
6662          * Read all device items, and then all the chunk items. All
6663          * device items are found before any chunk item (their object id
6664          * is smaller than the lowest possible object id for a chunk
6665          * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
6666          */
6667         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
6668         key.offset = 0;
6669         key.type = 0;
6670         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6671         if (ret < 0)
6672                 goto error;
6673         while (1) {
6674                 leaf = path->nodes[0];
6675                 slot = path->slots[0];
6676                 if (slot >= btrfs_header_nritems(leaf)) {
6677                         ret = btrfs_next_leaf(root, path);
6678                         if (ret == 0)
6679                                 continue;
6680                         if (ret < 0)
6681                                 goto error;
6682                         break;
6683                 }
6684                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6685                 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
6686                         struct btrfs_dev_item *dev_item;
6687                         dev_item = btrfs_item_ptr(leaf, slot,
6688                                                   struct btrfs_dev_item);
6689                         ret = read_one_dev(root, leaf, dev_item);
6690                         if (ret)
6691                                 goto error;
6692                 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
6693                         struct btrfs_chunk *chunk;
6694                         chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
6695                         ret = read_one_chunk(root, &found_key, leaf, chunk);
6696                         if (ret)
6697                                 goto error;
6698                 }
6699                 path->slots[0]++;
6700         }
6701         ret = 0;
6702 error:
6703         unlock_chunks(root);
6704         mutex_unlock(&uuid_mutex);
6705
6706         btrfs_free_path(path);
6707         return ret;
6708 }
6709
6710 void btrfs_init_devices_late(struct btrfs_fs_info *fs_info)
6711 {
6712         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6713         struct btrfs_device *device;
6714
6715         while (fs_devices) {
6716                 mutex_lock(&fs_devices->device_list_mutex);
6717                 list_for_each_entry(device, &fs_devices->devices, dev_list)
6718                         device->dev_root = fs_info->dev_root;
6719                 mutex_unlock(&fs_devices->device_list_mutex);
6720
6721                 fs_devices = fs_devices->seed;
6722         }
6723 }
6724
6725 static void __btrfs_reset_dev_stats(struct btrfs_device *dev)
6726 {
6727         int i;
6728
6729         for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6730                 btrfs_dev_stat_reset(dev, i);
6731 }
6732
6733 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
6734 {
6735         struct btrfs_key key;
6736         struct btrfs_key found_key;
6737         struct btrfs_root *dev_root = fs_info->dev_root;
6738         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6739         struct extent_buffer *eb;
6740         int slot;
6741         int ret = 0;
6742         struct btrfs_device *device;
6743         struct btrfs_path *path = NULL;
6744         int i;
6745
6746         path = btrfs_alloc_path();
6747         if (!path) {
6748                 ret = -ENOMEM;
6749                 goto out;
6750         }
6751
6752         mutex_lock(&fs_devices->device_list_mutex);
6753         list_for_each_entry(device, &fs_devices->devices, dev_list) {
6754                 int item_size;
6755                 struct btrfs_dev_stats_item *ptr;
6756
6757                 key.objectid = BTRFS_DEV_STATS_OBJECTID;
6758                 key.type = BTRFS_PERSISTENT_ITEM_KEY;
6759                 key.offset = device->devid;
6760                 ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
6761                 if (ret) {
6762                         __btrfs_reset_dev_stats(device);
6763                         device->dev_stats_valid = 1;
6764                         btrfs_release_path(path);
6765                         continue;
6766                 }
6767                 slot = path->slots[0];
6768                 eb = path->nodes[0];
6769                 btrfs_item_key_to_cpu(eb, &found_key, slot);
6770                 item_size = btrfs_item_size_nr(eb, slot);
6771
6772                 ptr = btrfs_item_ptr(eb, slot,
6773                                      struct btrfs_dev_stats_item);
6774
6775                 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
6776                         if (item_size >= (1 + i) * sizeof(__le64))
6777                                 btrfs_dev_stat_set(device, i,
6778                                         btrfs_dev_stats_value(eb, ptr, i));
6779                         else
6780                                 btrfs_dev_stat_reset(device, i);
6781                 }
6782
6783                 device->dev_stats_valid = 1;
6784                 btrfs_dev_stat_print_on_load(device);
6785                 btrfs_release_path(path);
6786         }
6787         mutex_unlock(&fs_devices->device_list_mutex);
6788
6789 out:
6790         btrfs_free_path(path);
6791         return ret < 0 ? ret : 0;
6792 }
6793
6794 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
6795                                 struct btrfs_root *dev_root,
6796                                 struct btrfs_device *device)
6797 {
6798         struct btrfs_path *path;
6799         struct btrfs_key key;
6800         struct extent_buffer *eb;
6801         struct btrfs_dev_stats_item *ptr;
6802         int ret;
6803         int i;
6804
6805         key.objectid = BTRFS_DEV_STATS_OBJECTID;
6806         key.type = BTRFS_PERSISTENT_ITEM_KEY;
6807         key.offset = device->devid;
6808
6809         path = btrfs_alloc_path();
6810         BUG_ON(!path);
6811         ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
6812         if (ret < 0) {
6813                 btrfs_warn_in_rcu(dev_root->fs_info,
6814                         "error %d while searching for dev_stats item for device %s",
6815                               ret, rcu_str_deref(device->name));
6816                 goto out;
6817         }
6818
6819         if (ret == 0 &&
6820             btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
6821                 /* need to delete old one and insert a new one */
6822                 ret = btrfs_del_item(trans, dev_root, path);
6823                 if (ret != 0) {
6824                         btrfs_warn_in_rcu(dev_root->fs_info,
6825                                 "delete too small dev_stats item for device %s failed %d",
6826                                       rcu_str_deref(device->name), ret);
6827                         goto out;
6828                 }
6829                 ret = 1;
6830         }
6831
6832         if (ret == 1) {
6833                 /* need to insert a new item */
6834                 btrfs_release_path(path);
6835                 ret = btrfs_insert_empty_item(trans, dev_root, path,
6836                                               &key, sizeof(*ptr));
6837                 if (ret < 0) {
6838                         btrfs_warn_in_rcu(dev_root->fs_info,
6839                                 "insert dev_stats item for device %s failed %d",
6840                                 rcu_str_deref(device->name), ret);
6841                         goto out;
6842                 }
6843         }
6844
6845         eb = path->nodes[0];
6846         ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
6847         for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6848                 btrfs_set_dev_stats_value(eb, ptr, i,
6849                                           btrfs_dev_stat_read(device, i));
6850         btrfs_mark_buffer_dirty(eb);
6851
6852 out:
6853         btrfs_free_path(path);
6854         return ret;
6855 }
6856
6857 /*
6858  * called from commit_transaction. Writes all changed device stats to disk.
6859  */
6860 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans,
6861                         struct btrfs_fs_info *fs_info)
6862 {
6863         struct btrfs_root *dev_root = fs_info->dev_root;
6864         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6865         struct btrfs_device *device;
6866         int stats_cnt;
6867         int ret = 0;
6868
6869         mutex_lock(&fs_devices->device_list_mutex);
6870         list_for_each_entry(device, &fs_devices->devices, dev_list) {
6871                 if (!device->dev_stats_valid || !btrfs_dev_stats_dirty(device))
6872                         continue;
6873
6874                 stats_cnt = atomic_read(&device->dev_stats_ccnt);
6875                 ret = update_dev_stat_item(trans, dev_root, device);
6876                 if (!ret)
6877                         atomic_sub(stats_cnt, &device->dev_stats_ccnt);
6878         }
6879         mutex_unlock(&fs_devices->device_list_mutex);
6880
6881         return ret;
6882 }
6883
6884 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
6885 {
6886         btrfs_dev_stat_inc(dev, index);
6887         btrfs_dev_stat_print_on_error(dev);
6888 }
6889
6890 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
6891 {
6892         if (!dev->dev_stats_valid)
6893                 return;
6894         btrfs_err_rl_in_rcu(dev->dev_root->fs_info,
6895                 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
6896                            rcu_str_deref(dev->name),
6897                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
6898                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
6899                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
6900                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
6901                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
6902 }
6903
6904 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
6905 {
6906         int i;
6907
6908         for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6909                 if (btrfs_dev_stat_read(dev, i) != 0)
6910                         break;
6911         if (i == BTRFS_DEV_STAT_VALUES_MAX)
6912                 return; /* all values == 0, suppress message */
6913
6914         btrfs_info_in_rcu(dev->dev_root->fs_info,
6915                 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
6916                rcu_str_deref(dev->name),
6917                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
6918                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
6919                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
6920                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
6921                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
6922 }
6923
6924 int btrfs_get_dev_stats(struct btrfs_root *root,
6925                         struct btrfs_ioctl_get_dev_stats *stats)
6926 {
6927         struct btrfs_device *dev;
6928         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
6929         int i;
6930
6931         mutex_lock(&fs_devices->device_list_mutex);
6932         dev = btrfs_find_device(root->fs_info, stats->devid, NULL, NULL);
6933         mutex_unlock(&fs_devices->device_list_mutex);
6934
6935         if (!dev) {
6936                 btrfs_warn(root->fs_info, "get dev_stats failed, device not found");
6937                 return -ENODEV;
6938         } else if (!dev->dev_stats_valid) {
6939                 btrfs_warn(root->fs_info, "get dev_stats failed, not yet valid");
6940                 return -ENODEV;
6941         } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
6942                 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
6943                         if (stats->nr_items > i)
6944                                 stats->values[i] =
6945                                         btrfs_dev_stat_read_and_reset(dev, i);
6946                         else
6947                                 btrfs_dev_stat_reset(dev, i);
6948                 }
6949         } else {
6950                 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6951                         if (stats->nr_items > i)
6952                                 stats->values[i] = btrfs_dev_stat_read(dev, i);
6953         }
6954         if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
6955                 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
6956         return 0;
6957 }
6958
6959 void btrfs_scratch_superblocks(struct block_device *bdev, char *device_path)
6960 {
6961         struct buffer_head *bh;
6962         struct btrfs_super_block *disk_super;
6963         int copy_num;
6964
6965         if (!bdev)
6966                 return;
6967
6968         for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX;
6969                 copy_num++) {
6970
6971                 if (btrfs_read_dev_one_super(bdev, copy_num, &bh))
6972                         continue;
6973
6974                 disk_super = (struct btrfs_super_block *)bh->b_data;
6975
6976                 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
6977                 set_buffer_dirty(bh);
6978                 sync_dirty_buffer(bh);
6979                 brelse(bh);
6980         }
6981
6982         /* Notify udev that device has changed */
6983         btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
6984
6985         /* Update ctime/mtime for device path for libblkid */
6986         update_dev_time(device_path);
6987 }
6988
6989 /*
6990  * Update the size of all devices, which is used for writing out the
6991  * super blocks.
6992  */
6993 void btrfs_update_commit_device_size(struct btrfs_fs_info *fs_info)
6994 {
6995         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6996         struct btrfs_device *curr, *next;
6997
6998         if (list_empty(&fs_devices->resized_devices))
6999                 return;
7000
7001         mutex_lock(&fs_devices->device_list_mutex);
7002         lock_chunks(fs_info->dev_root);
7003         list_for_each_entry_safe(curr, next, &fs_devices->resized_devices,
7004                                  resized_list) {
7005                 list_del_init(&curr->resized_list);
7006                 curr->commit_total_bytes = curr->disk_total_bytes;
7007         }
7008         unlock_chunks(fs_info->dev_root);
7009         mutex_unlock(&fs_devices->device_list_mutex);
7010 }
7011
7012 /* Must be invoked during the transaction commit */
7013 void btrfs_update_commit_device_bytes_used(struct btrfs_root *root,
7014                                         struct btrfs_transaction *transaction)
7015 {
7016         struct extent_map *em;
7017         struct map_lookup *map;
7018         struct btrfs_device *dev;
7019         int i;
7020
7021         if (list_empty(&transaction->pending_chunks))
7022                 return;
7023
7024         /* In order to kick the device replace finish process */
7025         lock_chunks(root);
7026         list_for_each_entry(em, &transaction->pending_chunks, list) {
7027                 map = em->map_lookup;
7028
7029                 for (i = 0; i < map->num_stripes; i++) {
7030                         dev = map->stripes[i].dev;
7031                         dev->commit_bytes_used = dev->bytes_used;
7032                 }
7033         }
7034         unlock_chunks(root);
7035 }
7036
7037 void btrfs_set_fs_info_ptr(struct btrfs_fs_info *fs_info)
7038 {
7039         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7040         while (fs_devices) {
7041                 fs_devices->fs_info = fs_info;
7042                 fs_devices = fs_devices->seed;
7043         }
7044 }
7045
7046 void btrfs_reset_fs_info_ptr(struct btrfs_fs_info *fs_info)
7047 {
7048         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7049         while (fs_devices) {
7050                 fs_devices->fs_info = NULL;
7051                 fs_devices = fs_devices->seed;
7052         }
7053 }
7054
7055 static void btrfs_close_one_device(struct btrfs_device *device)
7056 {
7057         struct btrfs_fs_devices *fs_devices = device->fs_devices;
7058         struct btrfs_device *new_device;
7059         struct rcu_string *name;
7060
7061         if (device->bdev)
7062                 fs_devices->open_devices--;
7063
7064         if (device->writeable &&
7065             device->devid != BTRFS_DEV_REPLACE_DEVID) {
7066                 list_del_init(&device->dev_alloc_list);
7067                 fs_devices->rw_devices--;
7068         }
7069
7070         if (device->missing)
7071                 fs_devices->missing_devices--;
7072
7073         new_device = btrfs_alloc_device(NULL, &device->devid,
7074                                         device->uuid);
7075         BUG_ON(IS_ERR(new_device)); /* -ENOMEM */
7076
7077         /* Safe because we are under uuid_mutex */
7078         if (device->name) {
7079                 name = rcu_string_strdup(device->name->str, GFP_NOFS);
7080                 BUG_ON(!name); /* -ENOMEM */
7081                 rcu_assign_pointer(new_device->name, name);
7082         }
7083
7084         list_replace_rcu(&device->dev_list, &new_device->dev_list);
7085         new_device->fs_devices = device->fs_devices;
7086
7087         call_rcu(&device->rcu, free_device);
7088 }