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 /*
 122  * Table to convert BTRFS_RAID_* to the error code if minimum number of devices
 123  * condition is not met. Zero means there's no corresponding
 124  * BTRFS_ERROR_DEV_*_NOT_MET value.
 125  */
 126 const int btrfs_raid_mindev_error[BTRFS_NR_RAID_TYPES] = {
 127         [BTRFS_RAID_RAID10] = BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET,
 128         [BTRFS_RAID_RAID1]  = BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET,
 129         [BTRFS_RAID_DUP]    = 0,
 130         [BTRFS_RAID_RAID0]  = 0,
 131         [BTRFS_RAID_SINGLE] = 0,
 132         [BTRFS_RAID_RAID5]  = BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET,
 133         [BTRFS_RAID_RAID6]  = BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET,
 134 };
 135
 136 static int init_first_rw_device(struct btrfs_trans_handle *trans,
 137                                 struct btrfs_root *root,
 138                                 struct btrfs_device *device);
 139 static int btrfs_relocate_sys_chunks(struct btrfs_root *root);
 140 static void __btrfs_reset_dev_stats(struct btrfs_device *dev);
 141 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev);
 142 static void btrfs_dev_stat_print_on_load(struct btrfs_device *device);
 143 static void btrfs_close_one_device(struct btrfs_device *device);
 144
 145 DEFINE_MUTEX(uuid_mutex);
 146 static LIST_HEAD(fs_uuids);
 147 struct list_head *btrfs_get_fs_uuids(void)
 148 {
 149         return &fs_uuids;
 150 }
 151
 152 static struct btrfs_fs_devices *__alloc_fs_devices(void)
 153 {
 154         struct btrfs_fs_devices *fs_devs;
 155
 156         fs_devs = kzalloc(sizeof(*fs_devs), GFP_KERNEL);
 157         if (!fs_devs)
 158                 return ERR_PTR(-ENOMEM);
 159
 160         mutex_init(&fs_devs->device_list_mutex);
 161
 162         INIT_LIST_HEAD(&fs_devs->devices);
 163         INIT_LIST_HEAD(&fs_devs->resized_devices);
 164         INIT_LIST_HEAD(&fs_devs->alloc_list);
 165         INIT_LIST_HEAD(&fs_devs->list);
 166
 167         return fs_devs;
 168 }
 169
 170 /**
 171  * alloc_fs_devices - allocate struct btrfs_fs_devices
 172  * @fsid:       a pointer to UUID for this FS.  If NULL a new UUID is
 173  *              generated.
 174  *
 175  * Return: a pointer to a new &struct btrfs_fs_devices on success;
 176  * ERR_PTR() on error.  Returned struct is not linked onto any lists and
 177  * can be destroyed with kfree() right away.
 178  */
 179 static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid)
 180 {
 181         struct btrfs_fs_devices *fs_devs;
 182
 183         fs_devs = __alloc_fs_devices();
 184         if (IS_ERR(fs_devs))
 185                 return fs_devs;
 186
 187         if (fsid)
 188                 memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE);
 189         else
 190                 generate_random_uuid(fs_devs->fsid);
 191
 192         return fs_devs;
 193 }
 194
 195 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
 196 {
 197         struct btrfs_device *device;
 198         WARN_ON(fs_devices->opened);
 199         while (!list_empty(&fs_devices->devices)) {
 200                 device = list_entry(fs_devices->devices.next,
 201                                     struct btrfs_device, dev_list);
 202                 list_del(&device->dev_list);
 203                 rcu_string_free(device->name);
 204                 kfree(device);
 205         }
 206         kfree(fs_devices);
 207 }
 208
 209 static void btrfs_kobject_uevent(struct block_device *bdev,
 210                                  enum kobject_action action)
 211 {
 212         int ret;
 213
 214         ret = kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, action);
 215         if (ret)
 216                 pr_warn("BTRFS: Sending event '%d' to kobject: '%s' (%p): failed\n",
 217                         action,
 218                         kobject_name(&disk_to_dev(bdev->bd_disk)->kobj),
 219                         &disk_to_dev(bdev->bd_disk)->kobj);
 220 }
 221
 222 void btrfs_cleanup_fs_uuids(void)
 223 {
 224         struct btrfs_fs_devices *fs_devices;
 225
 226         while (!list_empty(&fs_uuids)) {
 227                 fs_devices = list_entry(fs_uuids.next,
 228                                         struct btrfs_fs_devices, list);
 229                 list_del(&fs_devices->list);
 230                 free_fs_devices(fs_devices);
 231         }
 232 }
 233
 234 static struct btrfs_device *__alloc_device(void)
 235 {
 236         struct btrfs_device *dev;
 237
 238         dev = kzalloc(sizeof(*dev), GFP_KERNEL);
 239         if (!dev)
 240                 return ERR_PTR(-ENOMEM);
 241
 242         INIT_LIST_HEAD(&dev->dev_list);
 243         INIT_LIST_HEAD(&dev->dev_alloc_list);
 244         INIT_LIST_HEAD(&dev->resized_list);
 245
 246         spin_lock_init(&dev->io_lock);
 247
 248         spin_lock_init(&dev->reada_lock);
 249         atomic_set(&dev->reada_in_flight, 0);
 250         atomic_set(&dev->dev_stats_ccnt, 0);
 251         btrfs_device_data_ordered_init(dev);
 252         INIT_RADIX_TREE(&dev->reada_zones, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
 253         INIT_RADIX_TREE(&dev->reada_extents, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
 254
 255         return dev;
 256 }
 257
 258 static noinline struct btrfs_device *__find_device(struct list_head *head,
 259                                                    u64 devid, u8 *uuid)
 260 {
 261         struct btrfs_device *dev;
 262
 263         list_for_each_entry(dev, head, dev_list) {
 264                 if (dev->devid == devid &&
 265                     (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
 266                         return dev;
 267                 }
 268         }
 269         return NULL;
 270 }
 271
 272 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
 273 {
 274         struct btrfs_fs_devices *fs_devices;
 275
 276         list_for_each_entry(fs_devices, &fs_uuids, list) {
 277                 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
 278                         return fs_devices;
 279         }
 280         return NULL;
 281 }
 282
 283 static int
 284 btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder,
 285                       int flush, struct block_device **bdev,
 286                       struct buffer_head **bh)
 287 {
 288         int ret;
 289
 290         *bdev = blkdev_get_by_path(device_path, flags, holder);
 291
 292         if (IS_ERR(*bdev)) {
 293                 ret = PTR_ERR(*bdev);
 294                 goto error;
 295         }
 296
 297         if (flush)
 298                 filemap_write_and_wait((*bdev)->bd_inode->i_mapping);
 299         ret = set_blocksize(*bdev, 4096);
 300         if (ret) {
 301                 blkdev_put(*bdev, flags);
 302                 goto error;
 303         }
 304         invalidate_bdev(*bdev);
 305         *bh = btrfs_read_dev_super(*bdev);
 306         if (IS_ERR(*bh)) {
 307                 ret = PTR_ERR(*bh);
 308                 blkdev_put(*bdev, flags);
 309                 goto error;
 310         }
 311
 312         return 0;
 313
 314 error:
 315         *bdev = NULL;
 316         *bh = NULL;
 317         return ret;
 318 }
 319
 320 static void requeue_list(struct btrfs_pending_bios *pending_bios,
 321                         struct bio *head, struct bio *tail)
 322 {
 323
 324         struct bio *old_head;
 325
 326         old_head = pending_bios->head;
 327         pending_bios->head = head;
 328         if (pending_bios->tail)
 329                 tail->bi_next = old_head;
 330         else
 331                 pending_bios->tail = tail;
 332 }
 333
 334 /*
 335  * we try to collect pending bios for a device so we don't get a large
 336  * number of procs sending bios down to the same device.  This greatly
 337  * improves the schedulers ability to collect and merge the bios.
 338  *
 339  * But, it also turns into a long list of bios to process and that is sure
 340  * to eventually make the worker thread block.  The solution here is to
 341  * make some progress and then put this work struct back at the end of
 342  * the list if the block device is congested.  This way, multiple devices
 343  * can make progress from a single worker thread.
 344  */
 345 static noinline void run_scheduled_bios(struct btrfs_device *device)
 346 {
 347         struct bio *pending;
 348         struct backing_dev_info *bdi;
 349         struct btrfs_fs_info *fs_info;
 350         struct btrfs_pending_bios *pending_bios;
 351         struct bio *tail;
 352         struct bio *cur;
 353         int again = 0;
 354         unsigned long num_run;
 355         unsigned long batch_run = 0;
 356         unsigned long limit;
 357         unsigned long last_waited = 0;
 358         int force_reg = 0;
 359         int sync_pending = 0;
 360         struct blk_plug plug;
 361
 362         /*
 363          * this function runs all the bios we've collected for
 364          * a particular device.  We don't want to wander off to
 365          * another device without first sending all of these down.
 366          * So, setup a plug here and finish it off before we return
 367          */
 368         blk_start_plug(&plug);
 369
 370         bdi = blk_get_backing_dev_info(device->bdev);
 371         fs_info = device->dev_root->fs_info;
 372         limit = btrfs_async_submit_limit(fs_info);
 373         limit = limit * 2 / 3;
 374
 375 loop:
 376         spin_lock(&device->io_lock);
 377
 378 loop_lock:
 379         num_run = 0;
 380
 381         /* take all the bios off the list at once and process them
 382          * later on (without the lock held).  But, remember the
 383          * tail and other pointers so the bios can be properly reinserted
 384          * into the list if we hit congestion
 385          */
 386         if (!force_reg && device->pending_sync_bios.head) {
 387                 pending_bios = &device->pending_sync_bios;
 388                 force_reg = 1;
 389         } else {
 390                 pending_bios = &device->pending_bios;
 391                 force_reg = 0;
 392         }
 393
 394         pending = pending_bios->head;
 395         tail = pending_bios->tail;
 396         WARN_ON(pending && !tail);
 397
 398         /*
 399          * if pending was null this time around, no bios need processing
 400          * at all and we can stop.  Otherwise it'll loop back up again
 401          * and do an additional check so no bios are missed.
 402          *
 403          * device->running_pending is used to synchronize with the
 404          * schedule_bio code.
 405          */
 406         if (device->pending_sync_bios.head == NULL &&
 407             device->pending_bios.head == NULL) {
 408                 again = 0;
 409                 device->running_pending = 0;
 410         } else {
 411                 again = 1;
 412                 device->running_pending = 1;
 413         }
 414
 415         pending_bios->head = NULL;
 416         pending_bios->tail = NULL;
 417
 418         spin_unlock(&device->io_lock);
 419
 420         while (pending) {
 421
 422                 rmb();
 423                 /* we want to work on both lists, but do more bios on the
 424                  * sync list than the regular list
 425                  */
 426                 if ((num_run > 32 &&
 427                     pending_bios != &device->pending_sync_bios &&
 428                     device->pending_sync_bios.head) ||
 429                    (num_run > 64 && pending_bios == &device->pending_sync_bios &&
 430                     device->pending_bios.head)) {
 431                         spin_lock(&device->io_lock);
 432                         requeue_list(pending_bios, pending, tail);
 433                         goto loop_lock;
 434                 }
 435
 436                 cur = pending;
 437                 pending = pending->bi_next;
 438                 cur->bi_next = NULL;
 439
 440                 /*
 441                  * atomic_dec_return implies a barrier for waitqueue_active
 442                  */
 443                 if (atomic_dec_return(&fs_info->nr_async_bios) < limit &&
 444                     waitqueue_active(&fs_info->async_submit_wait))
 445                         wake_up(&fs_info->async_submit_wait);
 446
 447                 BUG_ON(atomic_read(&cur->__bi_cnt) == 0);
 448
 449                 /*
 450                  * if we're doing the sync list, record that our
 451                  * plug has some sync requests on it
 452                  *
 453                  * If we're doing the regular list and there are
 454                  * sync requests sitting around, unplug before
 455                  * we add more
 456                  */
 457                 if (pending_bios == &device->pending_sync_bios) {
 458                         sync_pending = 1;
 459                 } else if (sync_pending) {
 460                         blk_finish_plug(&plug);
 461                         blk_start_plug(&plug);
 462                         sync_pending = 0;
 463                 }
 464
 465                 btrfsic_submit_bio(cur->bi_rw, cur);
 466                 num_run++;
 467                 batch_run++;
 468
 469                 cond_resched();
 470
 471                 /*
 472                  * we made progress, there is more work to do and the bdi
 473                  * is now congested.  Back off and let other work structs
 474                  * run instead
 475                  */
 476                 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
 477                     fs_info->fs_devices->open_devices > 1) {
 478                         struct io_context *ioc;
 479
 480                         ioc = current->io_context;
 481
 482                         /*
 483                          * the main goal here is that we don't want to
 484                          * block if we're going to be able to submit
 485                          * more requests without blocking.
 486                          *
 487                          * This code does two great things, it pokes into
 488                          * the elevator code from a filesystem _and_
 489                          * it makes assumptions about how batching works.
 490                          */
 491                         if (ioc && ioc->nr_batch_requests > 0 &&
 492                             time_before(jiffies, ioc->last_waited + HZ/50UL) &&
 493                             (last_waited == 0 ||
 494                              ioc->last_waited == last_waited)) {
 495                                 /*
 496                                  * we want to go through our batch of
 497                                  * requests and stop.  So, we copy out
 498                                  * the ioc->last_waited time and test
 499                                  * against it before looping
 500                                  */
 501                                 last_waited = ioc->last_waited;
 502                                 cond_resched();
 503                                 continue;
 504                         }
 505                         spin_lock(&device->io_lock);
 506                         requeue_list(pending_bios, pending, tail);
 507                         device->running_pending = 1;
 508
 509                         spin_unlock(&device->io_lock);
 510                         btrfs_queue_work(fs_info->submit_workers,
 511                                          &device->work);
 512                         goto done;
 513                 }
 514                 /* unplug every 64 requests just for good measure */
 515                 if (batch_run % 64 == 0) {
 516                         blk_finish_plug(&plug);
 517                         blk_start_plug(&plug);
 518                         sync_pending = 0;
 519                 }
 520         }
 521
 522         cond_resched();
 523         if (again)
 524                 goto loop;
 525
 526         spin_lock(&device->io_lock);
 527         if (device->pending_bios.head || device->pending_sync_bios.head)
 528                 goto loop_lock;
 529         spin_unlock(&device->io_lock);
 530
 531 done:
 532         blk_finish_plug(&plug);
 533 }
 534
 535 static void pending_bios_fn(struct btrfs_work *work)
 536 {
 537         struct btrfs_device *device;
 538
 539         device = container_of(work, struct btrfs_device, work);
 540         run_scheduled_bios(device);
 541 }
 542
 543
 544 void btrfs_free_stale_device(struct btrfs_device *cur_dev)
 545 {
 546         struct btrfs_fs_devices *fs_devs;
 547         struct btrfs_device *dev;
 548
 549         if (!cur_dev->name)
 550                 return;
 551
 552         list_for_each_entry(fs_devs, &fs_uuids, list) {
 553                 int del = 1;
 554
 555                 if (fs_devs->opened)
 556                         continue;
 557                 if (fs_devs->seeding)
 558                         continue;
 559
 560                 list_for_each_entry(dev, &fs_devs->devices, dev_list) {
 561
 562                         if (dev == cur_dev)
 563                                 continue;
 564                         if (!dev->name)
 565                                 continue;
 566
 567                         /*
 568                          * Todo: This won't be enough. What if the same device
 569                          * comes back (with new uuid and) with its mapper path?
 570                          * But for now, this does help as mostly an admin will
 571                          * either use mapper or non mapper path throughout.
 572                          */
 573                         rcu_read_lock();
 574                         del = strcmp(rcu_str_deref(dev->name),
 575                                                 rcu_str_deref(cur_dev->name));
 576                         rcu_read_unlock();
 577                         if (!del)
 578                                 break;
 579                 }
 580
 581                 if (!del) {
 582                         /* delete the stale device */
 583                         if (fs_devs->num_devices == 1) {
 584                                 btrfs_sysfs_remove_fsid(fs_devs);
 585                                 list_del(&fs_devs->list);
 586                                 free_fs_devices(fs_devs);
 587                         } else {
 588                                 fs_devs->num_devices--;
 589                                 list_del(&dev->dev_list);
 590                                 rcu_string_free(dev->name);
 591                                 kfree(dev);
 592                         }
 593                         break;
 594                 }
 595         }
 596 }
 597
 598 /*
 599  * Add new device to list of registered devices
 600  *
 601  * Returns:
 602  * 1   - first time device is seen
 603  * 0   - device already known
 604  * < 0 - error
 605  */
 606 static noinline int device_list_add(const char *path,
 607                            struct btrfs_super_block *disk_super,
 608                            u64 devid, struct btrfs_fs_devices **fs_devices_ret)
 609 {
 610         struct btrfs_device *device;
 611         struct btrfs_fs_devices *fs_devices;
 612         struct rcu_string *name;
 613         int ret = 0;
 614         u64 found_transid = btrfs_super_generation(disk_super);
 615
 616         fs_devices = find_fsid(disk_super->fsid);
 617         if (!fs_devices) {
 618                 fs_devices = alloc_fs_devices(disk_super->fsid);
 619                 if (IS_ERR(fs_devices))
 620                         return PTR_ERR(fs_devices);
 621
 622                 list_add(&fs_devices->list, &fs_uuids);
 623
 624                 device = NULL;
 625         } else {
 626                 device = __find_device(&fs_devices->devices, devid,
 627                                        disk_super->dev_item.uuid);
 628         }
 629
 630         if (!device) {
 631                 if (fs_devices->opened)
 632                         return -EBUSY;
 633
 634                 device = btrfs_alloc_device(NULL, &devid,
 635                                             disk_super->dev_item.uuid);
 636                 if (IS_ERR(device)) {
 637                         /* we can safely leave the fs_devices entry around */
 638                         return PTR_ERR(device);
 639                 }
 640
 641                 name = rcu_string_strdup(path, GFP_NOFS);
 642                 if (!name) {
 643                         kfree(device);
 644                         return -ENOMEM;
 645                 }
 646                 rcu_assign_pointer(device->name, name);
 647
 648                 mutex_lock(&fs_devices->device_list_mutex);
 649                 list_add_rcu(&device->dev_list, &fs_devices->devices);
 650                 fs_devices->num_devices++;
 651                 mutex_unlock(&fs_devices->device_list_mutex);
 652
 653                 ret = 1;
 654                 device->fs_devices = fs_devices;
 655         } else if (!device->name || strcmp(device->name->str, path)) {
 656                 /*
 657                  * When FS is already mounted.
 658                  * 1. If you are here and if the device->name is NULL that
 659                  *    means this device was missing at time of FS mount.
 660                  * 2. If you are here and if the device->name is different
 661                  *    from 'path' that means either
 662                  *      a. The same device disappeared and reappeared with
 663                  *         different name. or
 664                  *      b. The missing-disk-which-was-replaced, has
 665                  *         reappeared now.
 666                  *
 667                  * We must allow 1 and 2a above. But 2b would be a spurious
 668                  * and unintentional.
 669                  *
 670                  * Further in case of 1 and 2a above, the disk at 'path'
 671                  * would have missed some transaction when it was away and
 672                  * in case of 2a the stale bdev has to be updated as well.
 673                  * 2b must not be allowed at all time.
 674                  */
 675
 676                 /*
 677                  * For now, we do allow update to btrfs_fs_device through the
 678                  * btrfs dev scan cli after FS has been mounted.  We're still
 679                  * tracking a problem where systems fail mount by subvolume id
 680                  * when we reject replacement on a mounted FS.
 681                  */
 682                 if (!fs_devices->opened && found_transid < device->generation) {
 683                         /*
 684                          * That is if the FS is _not_ mounted and if you
 685                          * are here, that means there is more than one
 686                          * disk with same uuid and devid.We keep the one
 687                          * with larger generation number or the last-in if
 688                          * generation are equal.
 689                          */
 690                         return -EEXIST;
 691                 }
 692
 693                 name = rcu_string_strdup(path, GFP_NOFS);
 694                 if (!name)
 695                         return -ENOMEM;
 696                 rcu_string_free(device->name);
 697                 rcu_assign_pointer(device->name, name);
 698                 if (device->missing) {
 699                         fs_devices->missing_devices--;
 700                         device->missing = 0;
 701                 }
 702         }
 703
 704         /*
 705          * Unmount does not free the btrfs_device struct but would zero
 706          * generation along with most of the other members. So just update
 707          * it back. We need it to pick the disk with largest generation
 708          * (as above).
 709          */
 710         if (!fs_devices->opened)
 711                 device->generation = found_transid;
 712
 713         /*
 714          * if there is new btrfs on an already registered device,
 715          * then remove the stale device entry.
 716          */
 717         if (ret > 0)
 718                 btrfs_free_stale_device(device);
 719
 720         *fs_devices_ret = fs_devices;
 721
 722         return ret;
 723 }
 724
 725 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
 726 {
 727         struct btrfs_fs_devices *fs_devices;
 728         struct btrfs_device *device;
 729         struct btrfs_device *orig_dev;
 730
 731         fs_devices = alloc_fs_devices(orig->fsid);
 732         if (IS_ERR(fs_devices))
 733                 return fs_devices;
 734
 735         mutex_lock(&orig->device_list_mutex);
 736         fs_devices->total_devices = orig->total_devices;
 737
 738         /* We have held the volume lock, it is safe to get the devices. */
 739         list_for_each_entry(orig_dev, &orig->devices, dev_list) {
 740                 struct rcu_string *name;
 741
 742                 device = btrfs_alloc_device(NULL, &orig_dev->devid,
 743                                             orig_dev->uuid);
 744                 if (IS_ERR(device))
 745                         goto error;
 746
 747                 /*
 748                  * This is ok to do without rcu read locked because we hold the
 749                  * uuid mutex so nothing we touch in here is going to disappear.
 750                  */
 751                 if (orig_dev->name) {
 752                         name = rcu_string_strdup(orig_dev->name->str,
 753                                         GFP_KERNEL);
 754                         if (!name) {
 755                                 kfree(device);
 756                                 goto error;
 757                         }
 758                         rcu_assign_pointer(device->name, name);
 759                 }
 760
 761                 list_add(&device->dev_list, &fs_devices->devices);
 762                 device->fs_devices = fs_devices;
 763                 fs_devices->num_devices++;
 764         }
 765         mutex_unlock(&orig->device_list_mutex);
 766         return fs_devices;
 767 error:
 768         mutex_unlock(&orig->device_list_mutex);
 769         free_fs_devices(fs_devices);
 770         return ERR_PTR(-ENOMEM);
 771 }
 772
 773 void btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices, int step)
 774 {
 775         struct btrfs_device *device, *next;
 776         struct btrfs_device *latest_dev = NULL;
 777
 778         mutex_lock(&uuid_mutex);
 779 again:
 780         /* This is the initialized path, it is safe to release the devices. */
 781         list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
 782                 if (device->in_fs_metadata) {
 783                         if (!device->is_tgtdev_for_dev_replace &&
 784                             (!latest_dev ||
 785                              device->generation > latest_dev->generation)) {
 786                                 latest_dev = device;
 787                         }
 788                         continue;
 789                 }
 790
 791                 if (device->devid == BTRFS_DEV_REPLACE_DEVID) {
 792                         /*
 793                          * In the first step, keep the device which has
 794                          * the correct fsid and the devid that is used
 795                          * for the dev_replace procedure.
 796                          * In the second step, the dev_replace state is
 797                          * read from the device tree and it is known
 798                          * whether the procedure is really active or
 799                          * not, which means whether this device is
 800                          * used or whether it should be removed.
 801                          */
 802                         if (step == 0 || device->is_tgtdev_for_dev_replace) {
 803                                 continue;
 804                         }
 805                 }
 806                 if (device->bdev) {
 807                         blkdev_put(device->bdev, device->mode);
 808                         device->bdev = NULL;
 809                         fs_devices->open_devices--;
 810                 }
 811                 if (device->writeable) {
 812                         list_del_init(&device->dev_alloc_list);
 813                         device->writeable = 0;
 814                         if (!device->is_tgtdev_for_dev_replace)
 815                                 fs_devices->rw_devices--;
 816                 }
 817                 list_del_init(&device->dev_list);
 818                 fs_devices->num_devices--;
 819                 rcu_string_free(device->name);
 820                 kfree(device);
 821         }
 822
 823         if (fs_devices->seed) {
 824                 fs_devices = fs_devices->seed;
 825                 goto again;
 826         }
 827
 828         fs_devices->latest_bdev = latest_dev->bdev;
 829
 830         mutex_unlock(&uuid_mutex);
 831 }
 832
 833 static void __free_device(struct work_struct *work)
 834 {
 835         struct btrfs_device *device;
 836
 837         device = container_of(work, struct btrfs_device, rcu_work);
 838
 839         if (device->bdev)
 840                 blkdev_put(device->bdev, device->mode);
 841
 842         rcu_string_free(device->name);
 843         kfree(device);
 844 }
 845
 846 static void free_device(struct rcu_head *head)
 847 {
 848         struct btrfs_device *device;
 849
 850         device = container_of(head, struct btrfs_device, rcu);
 851
 852         INIT_WORK(&device->rcu_work, __free_device);
 853         schedule_work(&device->rcu_work);
 854 }
 855
 856 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
 857 {
 858         struct btrfs_device *device, *tmp;
 859
 860         if (--fs_devices->opened > 0)
 861                 return 0;
 862
 863         mutex_lock(&fs_devices->device_list_mutex);
 864         list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list) {
 865                 btrfs_close_one_device(device);
 866         }
 867         mutex_unlock(&fs_devices->device_list_mutex);
 868
 869         WARN_ON(fs_devices->open_devices);
 870         WARN_ON(fs_devices->rw_devices);
 871         fs_devices->opened = 0;
 872         fs_devices->seeding = 0;
 873
 874         return 0;
 875 }
 876
 877 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
 878 {
 879         struct btrfs_fs_devices *seed_devices = NULL;
 880         int ret;
 881
 882         mutex_lock(&uuid_mutex);
 883         ret = __btrfs_close_devices(fs_devices);
 884         if (!fs_devices->opened) {
 885                 seed_devices = fs_devices->seed;
 886                 fs_devices->seed = NULL;
 887         }
 888         mutex_unlock(&uuid_mutex);
 889
 890         while (seed_devices) {
 891                 fs_devices = seed_devices;
 892                 seed_devices = fs_devices->seed;
 893                 __btrfs_close_devices(fs_devices);
 894                 free_fs_devices(fs_devices);
 895         }
 896         /*
 897          * Wait for rcu kworkers under __btrfs_close_devices
 898          * to finish all blkdev_puts so device is really
 899          * free when umount is done.
 900          */
 901         rcu_barrier();
 902         return ret;
 903 }
 904
 905 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
 906                                 fmode_t flags, void *holder)
 907 {
 908         struct request_queue *q;
 909         struct block_device *bdev;
 910         struct list_head *head = &fs_devices->devices;
 911         struct btrfs_device *device;
 912         struct btrfs_device *latest_dev = NULL;
 913         struct buffer_head *bh;
 914         struct btrfs_super_block *disk_super;
 915         u64 devid;
 916         int seeding = 1;
 917         int ret = 0;
 918
 919         flags |= FMODE_EXCL;
 920
 921         list_for_each_entry(device, head, dev_list) {
 922                 if (device->bdev)
 923                         continue;
 924                 if (!device->name)
 925                         continue;
 926
 927                 /* Just open everything we can; ignore failures here */
 928                 if (btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1,
 929                                             &bdev, &bh))
 930                         continue;
 931
 932                 disk_super = (struct btrfs_super_block *)bh->b_data;
 933                 devid = btrfs_stack_device_id(&disk_super->dev_item);
 934                 if (devid != device->devid)
 935                         goto error_brelse;
 936
 937                 if (memcmp(device->uuid, disk_super->dev_item.uuid,
 938                            BTRFS_UUID_SIZE))
 939                         goto error_brelse;
 940
 941                 device->generation = btrfs_super_generation(disk_super);
 942                 if (!latest_dev ||
 943                     device->generation > latest_dev->generation)
 944                         latest_dev = device;
 945
 946                 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
 947                         device->writeable = 0;
 948                 } else {
 949                         device->writeable = !bdev_read_only(bdev);
 950                         seeding = 0;
 951                 }
 952
 953                 q = bdev_get_queue(bdev);
 954                 if (blk_queue_discard(q))
 955                         device->can_discard = 1;
 956
 957                 device->bdev = bdev;
 958                 device->in_fs_metadata = 0;
 959                 device->mode = flags;
 960
 961                 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
 962                         fs_devices->rotating = 1;
 963
 964                 fs_devices->open_devices++;
 965                 if (device->writeable &&
 966                     device->devid != BTRFS_DEV_REPLACE_DEVID) {
 967                         fs_devices->rw_devices++;
 968                         list_add(&device->dev_alloc_list,
 969                                  &fs_devices->alloc_list);
 970                 }
 971                 brelse(bh);
 972                 continue;
 973
 974 error_brelse:
 975                 brelse(bh);
 976                 blkdev_put(bdev, flags);
 977                 continue;
 978         }
 979         if (fs_devices->open_devices == 0) {
 980                 ret = -EINVAL;
 981                 goto out;
 982         }
 983         fs_devices->seeding = seeding;
 984         fs_devices->opened = 1;
 985         fs_devices->latest_bdev = latest_dev->bdev;
 986         fs_devices->total_rw_bytes = 0;
 987 out:
 988         return ret;
 989 }
 990
 991 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
 992                        fmode_t flags, void *holder)
 993 {
 994         int ret;
 995
 996         mutex_lock(&uuid_mutex);
 997         if (fs_devices->opened) {
 998                 fs_devices->opened++;
 999                 ret = 0;
1000         } else {
1001                 ret = __btrfs_open_devices(fs_devices, flags, holder);
1002         }
1003         mutex_unlock(&uuid_mutex);
1004         return ret;
1005 }
1006
1007 void btrfs_release_disk_super(struct page *page)
1008 {
1009         kunmap(page);
1010         put_page(page);
1011 }
1012
1013 int btrfs_read_disk_super(struct block_device *bdev, u64 bytenr,
1014                 struct page **page, struct btrfs_super_block **disk_super)
1015 {
1016         void *p;
1017         pgoff_t index;
1018
1019         /* make sure our super fits in the device */
1020         if (bytenr + PAGE_SIZE >= i_size_read(bdev->bd_inode))
1021                 return 1;
1022
1023         /* make sure our super fits in the page */
1024         if (sizeof(**disk_super) > PAGE_SIZE)
1025                 return 1;
1026
1027         /* make sure our super doesn't straddle pages on disk */
1028         index = bytenr >> PAGE_SHIFT;
1029         if ((bytenr + sizeof(**disk_super) - 1) >> PAGE_SHIFT != index)
1030                 return 1;
1031
1032         /* pull in the page with our super */
1033         *page = read_cache_page_gfp(bdev->bd_inode->i_mapping,
1034                                    index, GFP_KERNEL);
1035
1036         if (IS_ERR_OR_NULL(*page))
1037                 return 1;
1038
1039         p = kmap(*page);
1040
1041         /* align our pointer to the offset of the super block */
1042         *disk_super = p + (bytenr & ~PAGE_MASK);
1043
1044         if (btrfs_super_bytenr(*disk_super) != bytenr ||
1045             btrfs_super_magic(*disk_super) != BTRFS_MAGIC) {
1046                 btrfs_release_disk_super(*page);
1047                 return 1;
1048         }
1049
1050         if ((*disk_super)->label[0] &&
1051                 (*disk_super)->label[BTRFS_LABEL_SIZE - 1])
1052                 (*disk_super)->label[BTRFS_LABEL_SIZE - 1] = '\0';
1053
1054         return 0;
1055 }
1056
1057 /*
1058  * Look for a btrfs signature on a device. This may be called out of the mount path
1059  * and we are not allowed to call set_blocksize during the scan. The superblock
1060  * is read via pagecache
1061  */
1062 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
1063                           struct btrfs_fs_devices **fs_devices_ret)
1064 {
1065         struct btrfs_super_block *disk_super;
1066         struct block_device *bdev;
1067         struct page *page;
1068         int ret = -EINVAL;
1069         u64 devid;
1070         u64 transid;
1071         u64 total_devices;
1072         u64 bytenr;
1073
1074         /*
1075          * we would like to check all the supers, but that would make
1076          * a btrfs mount succeed after a mkfs from a different FS.
1077          * So, we need to add a special mount option to scan for
1078          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
1079          */
1080         bytenr = btrfs_sb_offset(0);
1081         flags |= FMODE_EXCL;
1082         mutex_lock(&uuid_mutex);
1083
1084         bdev = blkdev_get_by_path(path, flags, holder);
1085         if (IS_ERR(bdev)) {
1086                 ret = PTR_ERR(bdev);
1087                 goto error;
1088         }
1089
1090         if (btrfs_read_disk_super(bdev, bytenr, &page, &disk_super))
1091                 goto error_bdev_put;
1092
1093         devid = btrfs_stack_device_id(&disk_super->dev_item);
1094         transid = btrfs_super_generation(disk_super);
1095         total_devices = btrfs_super_num_devices(disk_super);
1096
1097         ret = device_list_add(path, disk_super, devid, fs_devices_ret);
1098         if (ret > 0) {
1099                 if (disk_super->label[0]) {
1100                         printk(KERN_INFO "BTRFS: device label %s ", disk_super->label);
1101                 } else {
1102                         printk(KERN_INFO "BTRFS: device fsid %pU ", disk_super->fsid);
1103                 }
1104
1105                 printk(KERN_CONT "devid %llu transid %llu %s\n", devid, transid, path);
1106                 ret = 0;
1107         }
1108         if (!ret && fs_devices_ret)
1109                 (*fs_devices_ret)->total_devices = total_devices;
1110
1111         btrfs_release_disk_super(page);
1112
1113 error_bdev_put:
1114         blkdev_put(bdev, flags);
1115 error:
1116         mutex_unlock(&uuid_mutex);
1117         return ret;
1118 }
1119
1120 /* helper to account the used device space in the range */
1121 int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
1122                                    u64 end, u64 *length)
1123 {
1124         struct btrfs_key key;
1125         struct btrfs_root *root = device->dev_root;
1126         struct btrfs_dev_extent *dev_extent;
1127         struct btrfs_path *path;
1128         u64 extent_end;
1129         int ret;
1130         int slot;
1131         struct extent_buffer *l;
1132
1133         *length = 0;
1134
1135         if (start >= device->total_bytes || device->is_tgtdev_for_dev_replace)
1136                 return 0;
1137
1138         path = btrfs_alloc_path();
1139         if (!path)
1140                 return -ENOMEM;
1141         path->reada = READA_FORWARD;
1142
1143         key.objectid = device->devid;
1144         key.offset = start;
1145         key.type = BTRFS_DEV_EXTENT_KEY;
1146
1147         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1148         if (ret < 0)
1149                 goto out;
1150         if (ret > 0) {
1151                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1152                 if (ret < 0)
1153                         goto out;
1154         }
1155
1156         while (1) {
1157                 l = path->nodes[0];
1158                 slot = path->slots[0];
1159                 if (slot >= btrfs_header_nritems(l)) {
1160                         ret = btrfs_next_leaf(root, path);
1161                         if (ret == 0)
1162                                 continue;
1163                         if (ret < 0)
1164                                 goto out;
1165
1166                         break;
1167                 }
1168                 btrfs_item_key_to_cpu(l, &key, slot);
1169
1170                 if (key.objectid < device->devid)
1171                         goto next;
1172
1173                 if (key.objectid > device->devid)
1174                         break;
1175
1176                 if (key.type != BTRFS_DEV_EXTENT_KEY)
1177                         goto next;
1178
1179                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1180                 extent_end = key.offset + btrfs_dev_extent_length(l,
1181                                                                   dev_extent);
1182                 if (key.offset <= start && extent_end > end) {
1183                         *length = end - start + 1;
1184                         break;
1185                 } else if (key.offset <= start && extent_end > start)
1186                         *length += extent_end - start;
1187                 else if (key.offset > start && extent_end <= end)
1188                         *length += extent_end - key.offset;
1189                 else if (key.offset > start && key.offset <= end) {
1190                         *length += end - key.offset + 1;
1191                         break;
1192                 } else if (key.offset > end)
1193                         break;
1194
1195 next:
1196                 path->slots[0]++;
1197         }
1198         ret = 0;
1199 out:
1200         btrfs_free_path(path);
1201         return ret;
1202 }
1203
1204 static int contains_pending_extent(struct btrfs_transaction *transaction,
1205                                    struct btrfs_device *device,
1206                                    u64 *start, u64 len)
1207 {
1208         struct btrfs_fs_info *fs_info = device->dev_root->fs_info;
1209         struct extent_map *em;
1210         struct list_head *search_list = &fs_info->pinned_chunks;
1211         int ret = 0;
1212         u64 physical_start = *start;
1213
1214         if (transaction)
1215                 search_list = &transaction->pending_chunks;
1216 again:
1217         list_for_each_entry(em, search_list, list) {
1218                 struct map_lookup *map;
1219                 int i;
1220
1221                 map = em->map_lookup;
1222                 for (i = 0; i < map->num_stripes; i++) {
1223                         u64 end;
1224
1225                         if (map->stripes[i].dev != device)
1226                                 continue;
1227                         if (map->stripes[i].physical >= physical_start + len ||
1228                             map->stripes[i].physical + em->orig_block_len <=
1229                             physical_start)
1230                                 continue;
1231                         /*
1232                          * Make sure that while processing the pinned list we do
1233                          * not override our *start with a lower value, because
1234                          * we can have pinned chunks that fall within this
1235                          * device hole and that have lower physical addresses
1236                          * than the pending chunks we processed before. If we
1237                          * do not take this special care we can end up getting
1238                          * 2 pending chunks that start at the same physical
1239                          * device offsets because the end offset of a pinned
1240                          * chunk can be equal to the start offset of some
1241                          * pending chunk.
1242                          */
1243                         end = map->stripes[i].physical + em->orig_block_len;
1244                         if (end > *start) {
1245                                 *start = end;
1246                                 ret = 1;
1247                         }
1248                 }
1249         }
1250         if (search_list != &fs_info->pinned_chunks) {
1251                 search_list = &fs_info->pinned_chunks;
1252                 goto again;
1253         }
1254
1255         return ret;
1256 }
1257
1258
1259 /*
1260  * find_free_dev_extent_start - find free space in the specified device
1261  * @device:       the device which we search the free space in
1262  * @num_bytes:    the size of the free space that we need
1263  * @search_start: the position from which to begin the search
1264  * @start:        store the start of the free space.
1265  * @len:          the size of the free space. that we find, or the size
1266  *                of the max free space if we don't find suitable free space
1267  *
1268  * this uses a pretty simple search, the expectation is that it is
1269  * called very infrequently and that a given device has a small number
1270  * of extents
1271  *
1272  * @start is used to store the start of the free space if we find. But if we
1273  * don't find suitable free space, it will be used to store the start position
1274  * of the max free space.
1275  *
1276  * @len is used to store the size of the free space that we find.
1277  * But if we don't find suitable free space, it is used to store the size of
1278  * the max free space.
1279  */
1280 int find_free_dev_extent_start(struct btrfs_transaction *transaction,
1281                                struct btrfs_device *device, u64 num_bytes,
1282                                u64 search_start, u64 *start, u64 *len)
1283 {
1284         struct btrfs_key key;
1285         struct btrfs_root *root = device->dev_root;
1286         struct btrfs_dev_extent *dev_extent;
1287         struct btrfs_path *path;
1288         u64 hole_size;
1289         u64 max_hole_start;
1290         u64 max_hole_size;
1291         u64 extent_end;
1292         u64 search_end = device->total_bytes;
1293         int ret;
1294         int slot;
1295         struct extent_buffer *l;
1296         u64 min_search_start;
1297
1298         /*
1299          * We don't want to overwrite the superblock on the drive nor any area
1300          * used by the boot loader (grub for example), so we make sure to start
1301          * at an offset of at least 1MB.
1302          */
1303         min_search_start = max(root->fs_info->alloc_start, 1024ull * 1024);
1304         search_start = max(search_start, min_search_start);
1305
1306         path = btrfs_alloc_path();
1307         if (!path)
1308                 return -ENOMEM;
1309
1310         max_hole_start = search_start;
1311         max_hole_size = 0;
1312
1313 again:
1314         if (search_start >= search_end || device->is_tgtdev_for_dev_replace) {
1315                 ret = -ENOSPC;
1316                 goto out;
1317         }
1318
1319         path->reada = READA_FORWARD;
1320         path->search_commit_root = 1;
1321         path->skip_locking = 1;
1322
1323         key.objectid = device->devid;
1324         key.offset = search_start;
1325         key.type = BTRFS_DEV_EXTENT_KEY;
1326
1327         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1328         if (ret < 0)
1329                 goto out;
1330         if (ret > 0) {
1331                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1332                 if (ret < 0)
1333                         goto out;
1334         }
1335
1336         while (1) {
1337                 l = path->nodes[0];
1338                 slot = path->slots[0];
1339                 if (slot >= btrfs_header_nritems(l)) {
1340                         ret = btrfs_next_leaf(root, path);
1341                         if (ret == 0)
1342                                 continue;
1343                         if (ret < 0)
1344                                 goto out;
1345
1346                         break;
1347                 }
1348                 btrfs_item_key_to_cpu(l, &key, slot);
1349
1350                 if (key.objectid < device->devid)
1351                         goto next;
1352
1353                 if (key.objectid > device->devid)
1354                         break;
1355
1356                 if (key.type != BTRFS_DEV_EXTENT_KEY)
1357                         goto next;
1358
1359                 if (key.offset > search_start) {
1360                         hole_size = key.offset - search_start;
1361
1362                         /*
1363                          * Have to check before we set max_hole_start, otherwise
1364                          * we could end up sending back this offset anyway.
1365                          */
1366                         if (contains_pending_extent(transaction, device,
1367                                                     &search_start,
1368                                                     hole_size)) {
1369                                 if (key.offset >= search_start) {
1370                                         hole_size = key.offset - search_start;
1371                                 } else {
1372                                         WARN_ON_ONCE(1);
1373                                         hole_size = 0;
1374                                 }
1375                         }
1376
1377                         if (hole_size > max_hole_size) {
1378                                 max_hole_start = search_start;
1379                                 max_hole_size = hole_size;
1380                         }
1381
1382                         /*
1383                          * If this free space is greater than which we need,
1384                          * it must be the max free space that we have found
1385                          * until now, so max_hole_start must point to the start
1386                          * of this free space and the length of this free space
1387                          * is stored in max_hole_size. Thus, we return
1388                          * max_hole_start and max_hole_size and go back to the
1389                          * caller.
1390                          */
1391                         if (hole_size >= num_bytes) {
1392                                 ret = 0;
1393                                 goto out;
1394                         }
1395                 }
1396
1397                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1398                 extent_end = key.offset + btrfs_dev_extent_length(l,
1399                                                                   dev_extent);
1400                 if (extent_end > search_start)
1401                         search_start = extent_end;
1402 next:
1403                 path->slots[0]++;
1404                 cond_resched();
1405         }
1406
1407         /*
1408          * At this point, search_start should be the end of
1409          * allocated dev extents, and when shrinking the device,
1410          * search_end may be smaller than search_start.
1411          */
1412         if (search_end > search_start) {
1413                 hole_size = search_end - search_start;
1414
1415                 if (contains_pending_extent(transaction, device, &search_start,
1416                                             hole_size)) {
1417                         btrfs_release_path(path);
1418                         goto again;
1419                 }
1420
1421                 if (hole_size > max_hole_size) {
1422                         max_hole_start = search_start;
1423                         max_hole_size = hole_size;
1424                 }
1425         }
1426
1427         /* See above. */
1428         if (max_hole_size < num_bytes)
1429                 ret = -ENOSPC;
1430         else
1431                 ret = 0;
1432
1433 out:
1434         btrfs_free_path(path);
1435         *start = max_hole_start;
1436         if (len)
1437                 *len = max_hole_size;
1438         return ret;
1439 }
1440
1441 int find_free_dev_extent(struct btrfs_trans_handle *trans,
1442                          struct btrfs_device *device, u64 num_bytes,
1443                          u64 *start, u64 *len)
1444 {
1445         /* FIXME use last free of some kind */
1446         return find_free_dev_extent_start(trans->transaction, device,
1447                                           num_bytes, 0, start, len);
1448 }
1449
1450 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1451                           struct btrfs_device *device,
1452                           u64 start, u64 *dev_extent_len)
1453 {
1454         int ret;
1455         struct btrfs_path *path;
1456         struct btrfs_root *root = device->dev_root;
1457         struct btrfs_key key;
1458         struct btrfs_key found_key;
1459         struct extent_buffer *leaf = NULL;
1460         struct btrfs_dev_extent *extent = NULL;
1461
1462         path = btrfs_alloc_path();
1463         if (!path)
1464                 return -ENOMEM;
1465
1466         key.objectid = device->devid;
1467         key.offset = start;
1468         key.type = BTRFS_DEV_EXTENT_KEY;
1469 again:
1470         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1471         if (ret > 0) {
1472                 ret = btrfs_previous_item(root, path, key.objectid,
1473                                           BTRFS_DEV_EXTENT_KEY);
1474                 if (ret)
1475                         goto out;
1476                 leaf = path->nodes[0];
1477                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1478                 extent = btrfs_item_ptr(leaf, path->slots[0],
1479                                         struct btrfs_dev_extent);
1480                 BUG_ON(found_key.offset > start || found_key.offset +
1481                        btrfs_dev_extent_length(leaf, extent) < start);
1482                 key = found_key;
1483                 btrfs_release_path(path);
1484                 goto again;
1485         } else if (ret == 0) {
1486                 leaf = path->nodes[0];
1487                 extent = btrfs_item_ptr(leaf, path->slots[0],
1488                                         struct btrfs_dev_extent);
1489         } else {
1490                 btrfs_handle_fs_error(root->fs_info, ret, "Slot search failed");
1491                 goto out;
1492         }
1493
1494         *dev_extent_len = btrfs_dev_extent_length(leaf, extent);
1495
1496         ret = btrfs_del_item(trans, root, path);
1497         if (ret) {
1498                 btrfs_handle_fs_error(root->fs_info, ret,
1499                             "Failed to remove dev extent item");
1500         } else {
1501                 set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags);
1502         }
1503 out:
1504         btrfs_free_path(path);
1505         return ret;
1506 }
1507
1508 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1509                                   struct btrfs_device *device,
1510                                   u64 chunk_tree, u64 chunk_objectid,
1511                                   u64 chunk_offset, u64 start, u64 num_bytes)
1512 {
1513         int ret;
1514         struct btrfs_path *path;
1515         struct btrfs_root *root = device->dev_root;
1516         struct btrfs_dev_extent *extent;
1517         struct extent_buffer *leaf;
1518         struct btrfs_key key;
1519
1520         WARN_ON(!device->in_fs_metadata);
1521         WARN_ON(device->is_tgtdev_for_dev_replace);
1522         path = btrfs_alloc_path();
1523         if (!path)
1524                 return -ENOMEM;
1525
1526         key.objectid = device->devid;
1527         key.offset = start;
1528         key.type = BTRFS_DEV_EXTENT_KEY;
1529         ret = btrfs_insert_empty_item(trans, root, path, &key,
1530                                       sizeof(*extent));
1531         if (ret)
1532                 goto out;
1533
1534         leaf = path->nodes[0];
1535         extent = btrfs_item_ptr(leaf, path->slots[0],
1536                                 struct btrfs_dev_extent);
1537         btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
1538         btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
1539         btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1540
1541         write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
1542                     btrfs_dev_extent_chunk_tree_uuid(extent), BTRFS_UUID_SIZE);
1543
1544         btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1545         btrfs_mark_buffer_dirty(leaf);
1546 out:
1547         btrfs_free_path(path);
1548         return ret;
1549 }
1550
1551 static u64 find_next_chunk(struct btrfs_fs_info *fs_info)
1552 {
1553         struct extent_map_tree *em_tree;
1554         struct extent_map *em;
1555         struct rb_node *n;
1556         u64 ret = 0;
1557
1558         em_tree = &fs_info->mapping_tree.map_tree;
1559         read_lock(&em_tree->lock);
1560         n = rb_last(&em_tree->map);
1561         if (n) {
1562                 em = rb_entry(n, struct extent_map, rb_node);
1563                 ret = em->start + em->len;
1564         }
1565         read_unlock(&em_tree->lock);
1566
1567         return ret;
1568 }
1569
1570 static noinline int find_next_devid(struct btrfs_fs_info *fs_info,
1571                                     u64 *devid_ret)
1572 {
1573         int ret;
1574         struct btrfs_key key;
1575         struct btrfs_key found_key;
1576         struct btrfs_path *path;
1577
1578         path = btrfs_alloc_path();
1579         if (!path)
1580                 return -ENOMEM;
1581
1582         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1583         key.type = BTRFS_DEV_ITEM_KEY;
1584         key.offset = (u64)-1;
1585
1586         ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
1587         if (ret < 0)
1588                 goto error;
1589
1590         BUG_ON(ret == 0); /* Corruption */
1591
1592         ret = btrfs_previous_item(fs_info->chunk_root, path,
1593                                   BTRFS_DEV_ITEMS_OBJECTID,
1594                                   BTRFS_DEV_ITEM_KEY);
1595         if (ret) {
1596                 *devid_ret = 1;
1597         } else {
1598                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1599                                       path->slots[0]);
1600                 *devid_ret = found_key.offset + 1;
1601         }
1602         ret = 0;
1603 error:
1604         btrfs_free_path(path);
1605         return ret;
1606 }
1607
1608 /*
1609  * the device information is stored in the chunk root
1610  * the btrfs_device struct should be fully filled in
1611  */
1612 static int btrfs_add_device(struct btrfs_trans_handle *trans,
1613                             struct btrfs_root *root,
1614                             struct btrfs_device *device)
1615 {
1616         int ret;
1617         struct btrfs_path *path;
1618         struct btrfs_dev_item *dev_item;
1619         struct extent_buffer *leaf;
1620         struct btrfs_key key;
1621         unsigned long ptr;
1622
1623         root = root->fs_info->chunk_root;
1624
1625         path = btrfs_alloc_path();
1626         if (!path)
1627                 return -ENOMEM;
1628
1629         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1630         key.type = BTRFS_DEV_ITEM_KEY;
1631         key.offset = device->devid;
1632
1633         ret = btrfs_insert_empty_item(trans, root, path, &key,
1634                                       sizeof(*dev_item));
1635         if (ret)
1636                 goto out;
1637
1638         leaf = path->nodes[0];
1639         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1640
1641         btrfs_set_device_id(leaf, dev_item, device->devid);
1642         btrfs_set_device_generation(leaf, dev_item, 0);
1643         btrfs_set_device_type(leaf, dev_item, device->type);
1644         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1645         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1646         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1647         btrfs_set_device_total_bytes(leaf, dev_item,
1648                                      btrfs_device_get_disk_total_bytes(device));
1649         btrfs_set_device_bytes_used(leaf, dev_item,
1650                                     btrfs_device_get_bytes_used(device));
1651         btrfs_set_device_group(leaf, dev_item, 0);
1652         btrfs_set_device_seek_speed(leaf, dev_item, 0);
1653         btrfs_set_device_bandwidth(leaf, dev_item, 0);
1654         btrfs_set_device_start_offset(leaf, dev_item, 0);
1655
1656         ptr = btrfs_device_uuid(dev_item);
1657         write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1658         ptr = btrfs_device_fsid(dev_item);
1659         write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1660         btrfs_mark_buffer_dirty(leaf);
1661
1662         ret = 0;
1663 out:
1664         btrfs_free_path(path);
1665         return ret;
1666 }
1667
1668 /*
1669  * Function to update ctime/mtime for a given device path.
1670  * Mainly used for ctime/mtime based probe like libblkid.
1671  */
1672 static void update_dev_time(char *path_name)
1673 {
1674         struct file *filp;
1675
1676         filp = filp_open(path_name, O_RDWR, 0);
1677         if (IS_ERR(filp))
1678                 return;
1679         file_update_time(filp);
1680         filp_close(filp, NULL);
1681 }
1682
1683 static int btrfs_rm_dev_item(struct btrfs_root *root,
1684                              struct btrfs_device *device)
1685 {
1686         int ret;
1687         struct btrfs_path *path;
1688         struct btrfs_key key;
1689         struct btrfs_trans_handle *trans;
1690
1691         root = root->fs_info->chunk_root;
1692
1693         path = btrfs_alloc_path();
1694         if (!path)
1695                 return -ENOMEM;
1696
1697         trans = btrfs_start_transaction(root, 0);
1698         if (IS_ERR(trans)) {
1699                 btrfs_free_path(path);
1700                 return PTR_ERR(trans);
1701         }
1702         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1703         key.type = BTRFS_DEV_ITEM_KEY;
1704         key.offset = device->devid;
1705
1706         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1707         if (ret < 0)
1708                 goto out;
1709
1710         if (ret > 0) {
1711                 ret = -ENOENT;
1712                 goto out;
1713         }
1714
1715         ret = btrfs_del_item(trans, root, path);
1716         if (ret)
1717                 goto out;
1718 out:
1719         btrfs_free_path(path);
1720         btrfs_commit_transaction(trans, root);
1721         return ret;
1722 }
1723
1724 /*
1725  * Verify that @num_devices satisfies the RAID profile constraints in the whole
1726  * filesystem. It's up to the caller to adjust that number regarding eg. device
1727  * replace.
1728  */
1729 static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info,
1730                 u64 num_devices)
1731 {
1732         u64 all_avail;
1733         unsigned seq;
1734         int i;
1735
1736         do {
1737                 seq = read_seqbegin(&fs_info->profiles_lock);
1738
1739                 all_avail = fs_info->avail_data_alloc_bits |
1740                             fs_info->avail_system_alloc_bits |
1741                             fs_info->avail_metadata_alloc_bits;
1742         } while (read_seqretry(&fs_info->profiles_lock, seq));
1743
1744         for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
1745                 if (!(all_avail & btrfs_raid_group[i]))
1746                         continue;
1747
1748                 if (num_devices < btrfs_raid_array[i].devs_min) {
1749                         int ret = btrfs_raid_mindev_error[i];
1750
1751                         if (ret)
1752                                 return ret;
1753                 }
1754         }
1755
1756         return 0;
1757 }
1758
1759 struct btrfs_device *btrfs_find_next_active_device(struct btrfs_fs_devices *fs_devs,
1760                                         struct btrfs_device *device)
1761 {
1762         struct btrfs_device *next_device;
1763
1764         list_for_each_entry(next_device, &fs_devs->devices, dev_list) {
1765                 if (next_device != device &&
1766                         !next_device->missing && next_device->bdev)
1767                         return next_device;
1768         }
1769
1770         return NULL;
1771 }
1772
1773 /*
1774  * Helper function to check if the given device is part of s_bdev / latest_bdev
1775  * and replace it with the provided or the next active device, in the context
1776  * where this function called, there should be always be another device (or
1777  * this_dev) which is active.
1778  */
1779 void btrfs_assign_next_active_device(struct btrfs_fs_info *fs_info,
1780                 struct btrfs_device *device, struct btrfs_device *this_dev)
1781 {
1782         struct btrfs_device *next_device;
1783
1784         if (this_dev)
1785                 next_device = this_dev;
1786         else
1787                 next_device = btrfs_find_next_active_device(fs_info->fs_devices,
1788                                                                 device);
1789         ASSERT(next_device);
1790
1791         if (fs_info->sb->s_bdev &&
1792                         (fs_info->sb->s_bdev == device->bdev))
1793                 fs_info->sb->s_bdev = next_device->bdev;
1794
1795         if (fs_info->fs_devices->latest_bdev == device->bdev)
1796                 fs_info->fs_devices->latest_bdev = next_device->bdev;
1797 }
1798
1799 int btrfs_rm_device(struct btrfs_root *root, char *device_path, u64 devid)
1800 {
1801         struct btrfs_device *device;
1802         struct btrfs_fs_devices *cur_devices;
1803         u64 num_devices;
1804         int ret = 0;
1805         bool clear_super = false;
1806         char *dev_name = NULL;
1807
1808         mutex_lock(&uuid_mutex);
1809
1810         num_devices = root->fs_info->fs_devices->num_devices;
1811         btrfs_dev_replace_lock(&root->fs_info->dev_replace, 0);
1812         if (btrfs_dev_replace_is_ongoing(&root->fs_info->dev_replace)) {
1813                 WARN_ON(num_devices < 1);
1814                 num_devices--;
1815         }
1816         btrfs_dev_replace_unlock(&root->fs_info->dev_replace, 0);
1817
1818         ret = btrfs_check_raid_min_devices(root->fs_info, num_devices - 1);
1819         if (ret)
1820                 goto out;
1821
1822         ret = btrfs_find_device_by_devspec(root, devid, device_path,
1823                                 &device);
1824         if (ret)
1825                 goto out;
1826
1827         if (device->is_tgtdev_for_dev_replace) {
1828                 ret = BTRFS_ERROR_DEV_TGT_REPLACE;
1829                 goto out;
1830         }
1831
1832         if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1833                 ret = BTRFS_ERROR_DEV_ONLY_WRITABLE;
1834                 goto out;
1835         }
1836
1837         if (device->writeable) {
1838                 lock_chunks(root);
1839                 list_del_init(&device->dev_alloc_list);
1840                 device->fs_devices->rw_devices--;
1841                 unlock_chunks(root);
1842                 dev_name = kstrdup(device->name->str, GFP_KERNEL);
1843                 if (!dev_name) {
1844                         ret = -ENOMEM;
1845                         goto error_undo;
1846                 }
1847                 clear_super = true;
1848         }
1849
1850         mutex_unlock(&uuid_mutex);
1851         ret = btrfs_shrink_device(device, 0);
1852         mutex_lock(&uuid_mutex);
1853         if (ret)
1854                 goto error_undo;
1855
1856         /*
1857          * TODO: the superblock still includes this device in its num_devices
1858          * counter although write_all_supers() is not locked out. This
1859          * could give a filesystem state which requires a degraded mount.
1860          */
1861         ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1862         if (ret)
1863                 goto error_undo;
1864
1865         device->in_fs_metadata = 0;
1866         btrfs_scrub_cancel_dev(root->fs_info, device);
1867
1868         /*
1869          * the device list mutex makes sure that we don't change
1870          * the device list while someone else is writing out all
1871          * the device supers. Whoever is writing all supers, should
1872          * lock the device list mutex before getting the number of
1873          * devices in the super block (super_copy). Conversely,
1874          * whoever updates the number of devices in the super block
1875          * (super_copy) should hold the device list mutex.
1876          */
1877
1878         cur_devices = device->fs_devices;
1879         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1880         list_del_rcu(&device->dev_list);
1881
1882         device->fs_devices->num_devices--;
1883         device->fs_devices->total_devices--;
1884
1885         if (device->missing)
1886                 device->fs_devices->missing_devices--;
1887
1888         btrfs_assign_next_active_device(root->fs_info, device, NULL);
1889
1890         if (device->bdev) {
1891                 device->fs_devices->open_devices--;
1892                 /* remove sysfs entry */
1893                 btrfs_sysfs_rm_device_link(root->fs_info->fs_devices, device);
1894         }
1895
1896         call_rcu(&device->rcu, free_device);
1897
1898         num_devices = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
1899         btrfs_set_super_num_devices(root->fs_info->super_copy, num_devices);
1900         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1901
1902         if (cur_devices->open_devices == 0) {
1903                 struct btrfs_fs_devices *fs_devices;
1904                 fs_devices = root->fs_info->fs_devices;
1905                 while (fs_devices) {
1906                         if (fs_devices->seed == cur_devices) {
1907                                 fs_devices->seed = cur_devices->seed;
1908                                 break;
1909                         }
1910                         fs_devices = fs_devices->seed;
1911                 }
1912                 cur_devices->seed = NULL;
1913                 __btrfs_close_devices(cur_devices);
1914                 free_fs_devices(cur_devices);
1915         }
1916
1917         root->fs_info->num_tolerated_disk_barrier_failures =
1918                 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
1919
1920         /*
1921          * at this point, the device is zero sized.  We want to
1922          * remove it from the devices list and zero out the old super
1923          */
1924         if (clear_super) {
1925                 struct block_device *bdev;
1926
1927                 bdev = blkdev_get_by_path(dev_name, FMODE_READ | FMODE_EXCL,
1928                                                 root->fs_info->bdev_holder);
1929                 if (!IS_ERR(bdev)) {
1930                         btrfs_scratch_superblocks(bdev, dev_name);
1931                         blkdev_put(bdev, FMODE_READ | FMODE_EXCL);
1932                 }
1933         }
1934
1935 out:
1936         kfree(dev_name);
1937
1938         mutex_unlock(&uuid_mutex);
1939         return ret;
1940
1941 error_undo:
1942         if (device->writeable) {
1943                 lock_chunks(root);
1944                 list_add(&device->dev_alloc_list,
1945                          &root->fs_info->fs_devices->alloc_list);
1946                 device->fs_devices->rw_devices++;
1947                 unlock_chunks(root);
1948         }
1949         goto out;
1950 }
1951
1952 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_fs_info *fs_info,
1953                                         struct btrfs_device *srcdev)
1954 {
1955         struct btrfs_fs_devices *fs_devices;
1956
1957         WARN_ON(!mutex_is_locked(&fs_info->fs_devices->device_list_mutex));
1958
1959         /*
1960          * in case of fs with no seed, srcdev->fs_devices will point
1961          * to fs_devices of fs_info. However when the dev being replaced is
1962          * a seed dev it will point to the seed's local fs_devices. In short
1963          * srcdev will have its correct fs_devices in both the cases.
1964          */
1965         fs_devices = srcdev->fs_devices;
1966
1967         list_del_rcu(&srcdev->dev_list);
1968         list_del_rcu(&srcdev->dev_alloc_list);
1969         fs_devices->num_devices--;
1970         if (srcdev->missing)
1971                 fs_devices->missing_devices--;
1972
1973         if (srcdev->writeable)
1974                 fs_devices->rw_devices--;
1975
1976         if (srcdev->bdev)
1977                 fs_devices->open_devices--;
1978 }
1979
1980 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_fs_info *fs_info,
1981                                       struct btrfs_device *srcdev)
1982 {
1983         struct btrfs_fs_devices *fs_devices = srcdev->fs_devices;
1984
1985         if (srcdev->writeable) {
1986                 /* zero out the old super if it is writable */
1987                 btrfs_scratch_superblocks(srcdev->bdev, srcdev->name->str);
1988         }
1989         call_rcu(&srcdev->rcu, free_device);
1990
1991         /*
1992          * unless fs_devices is seed fs, num_devices shouldn't go
1993          * zero
1994          */
1995         BUG_ON(!fs_devices->num_devices && !fs_devices->seeding);
1996
1997         /* if this is no devs we rather delete the fs_devices */
1998         if (!fs_devices->num_devices) {
1999                 struct btrfs_fs_devices *tmp_fs_devices;
2000
2001                 tmp_fs_devices = fs_info->fs_devices;
2002                 while (tmp_fs_devices) {
2003                         if (tmp_fs_devices->seed == fs_devices) {
2004                                 tmp_fs_devices->seed = fs_devices->seed;
2005                                 break;
2006                         }
2007                         tmp_fs_devices = tmp_fs_devices->seed;
2008                 }
2009                 fs_devices->seed = NULL;
2010                 __btrfs_close_devices(fs_devices);
2011                 free_fs_devices(fs_devices);
2012         }
2013 }
2014
2015 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
2016                                       struct btrfs_device *tgtdev)
2017 {
2018         mutex_lock(&uuid_mutex);
2019         WARN_ON(!tgtdev);
2020         mutex_lock(&fs_info->fs_devices->device_list_mutex);
2021
2022         btrfs_sysfs_rm_device_link(fs_info->fs_devices, tgtdev);
2023
2024         if (tgtdev->bdev)
2025                 fs_info->fs_devices->open_devices--;
2026
2027         fs_info->fs_devices->num_devices--;
2028
2029         btrfs_assign_next_active_device(fs_info, tgtdev, NULL);
2030
2031         list_del_rcu(&tgtdev->dev_list);
2032
2033         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2034         mutex_unlock(&uuid_mutex);
2035
2036         /*
2037          * The update_dev_time() with in btrfs_scratch_superblocks()
2038          * may lead to a call to btrfs_show_devname() which will try
2039          * to hold device_list_mutex. And here this device
2040          * is already out of device list, so we don't have to hold
2041          * the device_list_mutex lock.
2042          */
2043         btrfs_scratch_superblocks(tgtdev->bdev, tgtdev->name->str);
2044         call_rcu(&tgtdev->rcu, free_device);
2045 }
2046
2047 static int btrfs_find_device_by_path(struct btrfs_root *root, char *device_path,
2048                                      struct btrfs_device **device)
2049 {
2050         int ret = 0;
2051         struct btrfs_super_block *disk_super;
2052         u64 devid;
2053         u8 *dev_uuid;
2054         struct block_device *bdev;
2055         struct buffer_head *bh;
2056
2057         *device = NULL;
2058         ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
2059                                     root->fs_info->bdev_holder, 0, &bdev, &bh);
2060         if (ret)
2061                 return ret;
2062         disk_super = (struct btrfs_super_block *)bh->b_data;
2063         devid = btrfs_stack_device_id(&disk_super->dev_item);
2064         dev_uuid = disk_super->dev_item.uuid;
2065         *device = btrfs_find_device(root->fs_info, devid, dev_uuid,
2066                                     disk_super->fsid);
2067         brelse(bh);
2068         if (!*device)
2069                 ret = -ENOENT;
2070         blkdev_put(bdev, FMODE_READ);
2071         return ret;
2072 }
2073
2074 int btrfs_find_device_missing_or_by_path(struct btrfs_root *root,
2075                                          char *device_path,
2076                                          struct btrfs_device **device)
2077 {
2078         *device = NULL;
2079         if (strcmp(device_path, "missing") == 0) {
2080                 struct list_head *devices;
2081                 struct btrfs_device *tmp;
2082
2083                 devices = &root->fs_info->fs_devices->devices;
2084                 /*
2085                  * It is safe to read the devices since the volume_mutex
2086                  * is held by the caller.
2087                  */
2088                 list_for_each_entry(tmp, devices, dev_list) {
2089                         if (tmp->in_fs_metadata && !tmp->bdev) {
2090                                 *device = tmp;
2091                                 break;
2092                         }
2093                 }
2094
2095                 if (!*device)
2096                         return BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
2097
2098                 return 0;
2099         } else {
2100                 return btrfs_find_device_by_path(root, device_path, device);
2101         }
2102 }
2103
2104 /*
2105  * Lookup a device given by device id, or the path if the id is 0.
2106  */
2107 int btrfs_find_device_by_devspec(struct btrfs_root *root, u64 devid,
2108                                          char *devpath,
2109                                          struct btrfs_device **device)
2110 {
2111         int ret;
2112
2113         if (devid) {
2114                 ret = 0;
2115                 *device = btrfs_find_device(root->fs_info, devid, NULL,
2116                                             NULL);
2117                 if (!*device)
2118                         ret = -ENOENT;
2119         } else {
2120                 if (!devpath || !devpath[0])
2121                         return -EINVAL;
2122
2123                 ret = btrfs_find_device_missing_or_by_path(root, devpath,
2124                                                            device);
2125         }
2126         return ret;
2127 }
2128
2129 /*
2130  * does all the dirty work required for changing file system's UUID.
2131  */
2132 static int btrfs_prepare_sprout(struct btrfs_root *root)
2133 {
2134         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2135         struct btrfs_fs_devices *old_devices;
2136         struct btrfs_fs_devices *seed_devices;
2137         struct btrfs_super_block *disk_super = root->fs_info->super_copy;
2138         struct btrfs_device *device;
2139         u64 super_flags;
2140
2141         BUG_ON(!mutex_is_locked(&uuid_mutex));
2142         if (!fs_devices->seeding)
2143                 return -EINVAL;
2144
2145         seed_devices = __alloc_fs_devices();
2146         if (IS_ERR(seed_devices))
2147                 return PTR_ERR(seed_devices);
2148
2149         old_devices = clone_fs_devices(fs_devices);
2150         if (IS_ERR(old_devices)) {
2151                 kfree(seed_devices);
2152                 return PTR_ERR(old_devices);
2153         }
2154
2155         list_add(&old_devices->list, &fs_uuids);
2156
2157         memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
2158         seed_devices->opened = 1;
2159         INIT_LIST_HEAD(&seed_devices->devices);
2160         INIT_LIST_HEAD(&seed_devices->alloc_list);
2161         mutex_init(&seed_devices->device_list_mutex);
2162
2163         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2164         list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
2165                               synchronize_rcu);
2166         list_for_each_entry(device, &seed_devices->devices, dev_list)
2167                 device->fs_devices = seed_devices;
2168
2169         lock_chunks(root);
2170         list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
2171         unlock_chunks(root);
2172
2173         fs_devices->seeding = 0;
2174         fs_devices->num_devices = 0;
2175         fs_devices->open_devices = 0;
2176         fs_devices->missing_devices = 0;
2177         fs_devices->rotating = 0;
2178         fs_devices->seed = seed_devices;
2179
2180         generate_random_uuid(fs_devices->fsid);
2181         memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2182         memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2183         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2184
2185         super_flags = btrfs_super_flags(disk_super) &
2186                       ~BTRFS_SUPER_FLAG_SEEDING;
2187         btrfs_set_super_flags(disk_super, super_flags);
2188
2189         return 0;
2190 }
2191
2192 /*
2193  * Store the expected generation for seed devices in device items.
2194  */
2195 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
2196                                struct btrfs_root *root)
2197 {
2198         struct btrfs_path *path;
2199         struct extent_buffer *leaf;
2200         struct btrfs_dev_item *dev_item;
2201         struct btrfs_device *device;
2202         struct btrfs_key key;
2203         u8 fs_uuid[BTRFS_UUID_SIZE];
2204         u8 dev_uuid[BTRFS_UUID_SIZE];
2205         u64 devid;
2206         int ret;
2207
2208         path = btrfs_alloc_path();
2209         if (!path)
2210                 return -ENOMEM;
2211
2212         root = root->fs_info->chunk_root;
2213         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2214         key.offset = 0;
2215         key.type = BTRFS_DEV_ITEM_KEY;
2216
2217         while (1) {
2218                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2219                 if (ret < 0)
2220                         goto error;
2221
2222                 leaf = path->nodes[0];
2223 next_slot:
2224                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2225                         ret = btrfs_next_leaf(root, path);
2226                         if (ret > 0)
2227                                 break;
2228                         if (ret < 0)
2229                                 goto error;
2230                         leaf = path->nodes[0];
2231                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2232                         btrfs_release_path(path);
2233                         continue;
2234                 }
2235
2236                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2237                 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
2238                     key.type != BTRFS_DEV_ITEM_KEY)
2239                         break;
2240
2241                 dev_item = btrfs_item_ptr(leaf, path->slots[0],
2242                                           struct btrfs_dev_item);
2243                 devid = btrfs_device_id(leaf, dev_item);
2244                 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
2245                                    BTRFS_UUID_SIZE);
2246                 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
2247                                    BTRFS_UUID_SIZE);
2248                 device = btrfs_find_device(root->fs_info, devid, dev_uuid,
2249                                            fs_uuid);
2250                 BUG_ON(!device); /* Logic error */
2251
2252                 if (device->fs_devices->seeding) {
2253                         btrfs_set_device_generation(leaf, dev_item,
2254                                                     device->generation);
2255                         btrfs_mark_buffer_dirty(leaf);
2256                 }
2257
2258                 path->slots[0]++;
2259                 goto next_slot;
2260         }
2261         ret = 0;
2262 error:
2263         btrfs_free_path(path);
2264         return ret;
2265 }
2266
2267 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
2268 {
2269         struct request_queue *q;
2270         struct btrfs_trans_handle *trans;
2271         struct btrfs_device *device;
2272         struct block_device *bdev;
2273         struct list_head *devices;
2274         struct super_block *sb = root->fs_info->sb;
2275         struct rcu_string *name;
2276         u64 tmp;
2277         int seeding_dev = 0;
2278         int ret = 0;
2279
2280         if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
2281                 return -EROFS;
2282
2283         bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2284                                   root->fs_info->bdev_holder);
2285         if (IS_ERR(bdev))
2286                 return PTR_ERR(bdev);
2287
2288         if (root->fs_info->fs_devices->seeding) {
2289                 seeding_dev = 1;
2290                 down_write(&sb->s_umount);
2291                 mutex_lock(&uuid_mutex);
2292         }
2293
2294         filemap_write_and_wait(bdev->bd_inode->i_mapping);
2295
2296         devices = &root->fs_info->fs_devices->devices;
2297
2298         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2299         list_for_each_entry(device, devices, dev_list) {
2300                 if (device->bdev == bdev) {
2301                         ret = -EEXIST;
2302                         mutex_unlock(
2303                                 &root->fs_info->fs_devices->device_list_mutex);
2304                         goto error;
2305                 }
2306         }
2307         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2308
2309         device = btrfs_alloc_device(root->fs_info, NULL, NULL);
2310         if (IS_ERR(device)) {
2311                 /* we can safely leave the fs_devices entry around */
2312                 ret = PTR_ERR(device);
2313                 goto error;
2314         }
2315
2316         name = rcu_string_strdup(device_path, GFP_KERNEL);
2317         if (!name) {
2318                 kfree(device);
2319                 ret = -ENOMEM;
2320                 goto error;
2321         }
2322         rcu_assign_pointer(device->name, name);
2323
2324         trans = btrfs_start_transaction(root, 0);
2325         if (IS_ERR(trans)) {
2326                 rcu_string_free(device->name);
2327                 kfree(device);
2328                 ret = PTR_ERR(trans);
2329                 goto error;
2330         }
2331
2332         q = bdev_get_queue(bdev);
2333         if (blk_queue_discard(q))
2334                 device->can_discard = 1;
2335         device->writeable = 1;
2336         device->generation = trans->transid;
2337         device->io_width = root->sectorsize;
2338         device->io_align = root->sectorsize;
2339         device->sector_size = root->sectorsize;
2340         device->total_bytes = i_size_read(bdev->bd_inode);
2341         device->disk_total_bytes = device->total_bytes;
2342         device->commit_total_bytes = device->total_bytes;
2343         device->dev_root = root->fs_info->dev_root;
2344         device->bdev = bdev;
2345         device->in_fs_metadata = 1;
2346         device->is_tgtdev_for_dev_replace = 0;
2347         device->mode = FMODE_EXCL;
2348         device->dev_stats_valid = 1;
2349         set_blocksize(device->bdev, 4096);
2350
2351         if (seeding_dev) {
2352                 sb->s_flags &= ~MS_RDONLY;
2353                 ret = btrfs_prepare_sprout(root);
2354                 BUG_ON(ret); /* -ENOMEM */
2355         }
2356
2357         device->fs_devices = root->fs_info->fs_devices;
2358
2359         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2360         lock_chunks(root);
2361         list_add_rcu(&device->dev_list, &root->fs_info->fs_devices->devices);
2362         list_add(&device->dev_alloc_list,
2363                  &root->fs_info->fs_devices->alloc_list);
2364         root->fs_info->fs_devices->num_devices++;
2365         root->fs_info->fs_devices->open_devices++;
2366         root->fs_info->fs_devices->rw_devices++;
2367         root->fs_info->fs_devices->total_devices++;
2368         root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
2369
2370         spin_lock(&root->fs_info->free_chunk_lock);
2371         root->fs_info->free_chunk_space += device->total_bytes;
2372         spin_unlock(&root->fs_info->free_chunk_lock);
2373
2374         if (!blk_queue_nonrot(bdev_get_queue(bdev)))
2375                 root->fs_info->fs_devices->rotating = 1;
2376
2377         tmp = btrfs_super_total_bytes(root->fs_info->super_copy);
2378         btrfs_set_super_total_bytes(root->fs_info->super_copy,
2379                                     tmp + device->total_bytes);
2380
2381         tmp = btrfs_super_num_devices(root->fs_info->super_copy);
2382         btrfs_set_super_num_devices(root->fs_info->super_copy,
2383                                     tmp + 1);
2384
2385         /* add sysfs device entry */
2386         btrfs_sysfs_add_device_link(root->fs_info->fs_devices, device);
2387
2388         /*
2389          * we've got more storage, clear any full flags on the space
2390          * infos
2391          */
2392         btrfs_clear_space_info_full(root->fs_info);
2393
2394         unlock_chunks(root);
2395         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2396
2397         if (seeding_dev) {
2398                 lock_chunks(root);
2399                 ret = init_first_rw_device(trans, root, device);
2400                 unlock_chunks(root);
2401                 if (ret) {
2402                         btrfs_abort_transaction(trans, root, ret);
2403                         goto error_trans;
2404                 }
2405         }
2406
2407         ret = btrfs_add_device(trans, root, device);
2408         if (ret) {
2409                 btrfs_abort_transaction(trans, root, ret);
2410                 goto error_trans;
2411         }
2412
2413         if (seeding_dev) {
2414                 char fsid_buf[BTRFS_UUID_UNPARSED_SIZE];
2415
2416                 ret = btrfs_finish_sprout(trans, root);
2417                 if (ret) {
2418                         btrfs_abort_transaction(trans, root, ret);
2419                         goto error_trans;
2420                 }
2421
2422                 /* Sprouting would change fsid of the mounted root,
2423                  * so rename the fsid on the sysfs
2424                  */
2425                 snprintf(fsid_buf, BTRFS_UUID_UNPARSED_SIZE, "%pU",
2426                                                 root->fs_info->fsid);
2427                 if (kobject_rename(&root->fs_info->fs_devices->fsid_kobj,
2428                                                                 fsid_buf))
2429                         btrfs_warn(root->fs_info,
2430                                 "sysfs: failed to create fsid for sprout");
2431         }
2432
2433         root->fs_info->num_tolerated_disk_barrier_failures =
2434                 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
2435         ret = btrfs_commit_transaction(trans, root);
2436
2437         if (seeding_dev) {
2438                 mutex_unlock(&uuid_mutex);
2439                 up_write(&sb->s_umount);
2440
2441                 if (ret) /* transaction commit */
2442                         return ret;
2443
2444                 ret = btrfs_relocate_sys_chunks(root);
2445                 if (ret < 0)
2446                         btrfs_handle_fs_error(root->fs_info, ret,
2447                                     "Failed to relocate sys chunks after "
2448                                     "device initialization. This can be fixed "
2449                                     "using the \"btrfs balance\" command.");
2450                 trans = btrfs_attach_transaction(root);
2451                 if (IS_ERR(trans)) {
2452                         if (PTR_ERR(trans) == -ENOENT)
2453                                 return 0;
2454                         return PTR_ERR(trans);
2455                 }
2456                 ret = btrfs_commit_transaction(trans, root);
2457         }
2458
2459         /* Update ctime/mtime for libblkid */
2460         update_dev_time(device_path);
2461         return ret;
2462
2463 error_trans:
2464         btrfs_end_transaction(trans, root);
2465         rcu_string_free(device->name);
2466         btrfs_sysfs_rm_device_link(root->fs_info->fs_devices, device);
2467         kfree(device);
2468 error:
2469         blkdev_put(bdev, FMODE_EXCL);
2470         if (seeding_dev) {
2471                 mutex_unlock(&uuid_mutex);
2472                 up_write(&sb->s_umount);
2473         }
2474         return ret;
2475 }
2476
2477 int btrfs_init_dev_replace_tgtdev(struct btrfs_root *root, char *device_path,
2478                                   struct btrfs_device *srcdev,
2479                                   struct btrfs_device **device_out)
2480 {
2481         struct request_queue *q;
2482         struct btrfs_device *device;
2483         struct block_device *bdev;
2484         struct btrfs_fs_info *fs_info = root->fs_info;
2485         struct list_head *devices;
2486         struct rcu_string *name;
2487         u64 devid = BTRFS_DEV_REPLACE_DEVID;
2488         int ret = 0;
2489
2490         *device_out = NULL;
2491         if (fs_info->fs_devices->seeding) {
2492                 btrfs_err(fs_info, "the filesystem is a seed filesystem!");
2493                 return -EINVAL;
2494         }
2495
2496         bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2497                                   fs_info->bdev_holder);
2498         if (IS_ERR(bdev)) {
2499                 btrfs_err(fs_info, "target device %s is invalid!", device_path);
2500                 return PTR_ERR(bdev);
2501         }
2502
2503         filemap_write_and_wait(bdev->bd_inode->i_mapping);
2504
2505         devices = &fs_info->fs_devices->devices;
2506         list_for_each_entry(device, devices, dev_list) {
2507                 if (device->bdev == bdev) {
2508                         btrfs_err(fs_info, "target device is in the filesystem!");
2509                         ret = -EEXIST;
2510                         goto error;
2511                 }
2512         }
2513
2514
2515         if (i_size_read(bdev->bd_inode) <
2516             btrfs_device_get_total_bytes(srcdev)) {
2517                 btrfs_err(fs_info, "target device is smaller than source device!");
2518                 ret = -EINVAL;
2519                 goto error;
2520         }
2521
2522
2523         device = btrfs_alloc_device(NULL, &devid, NULL);
2524         if (IS_ERR(device)) {
2525                 ret = PTR_ERR(device);
2526                 goto error;
2527         }
2528
2529         name = rcu_string_strdup(device_path, GFP_NOFS);
2530         if (!name) {
2531                 kfree(device);
2532                 ret = -ENOMEM;
2533                 goto error;
2534         }
2535         rcu_assign_pointer(device->name, name);
2536
2537         q = bdev_get_queue(bdev);
2538         if (blk_queue_discard(q))
2539                 device->can_discard = 1;
2540         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2541         device->writeable = 1;
2542         device->generation = 0;
2543         device->io_width = root->sectorsize;
2544         device->io_align = root->sectorsize;
2545         device->sector_size = root->sectorsize;
2546         device->total_bytes = btrfs_device_get_total_bytes(srcdev);
2547         device->disk_total_bytes = btrfs_device_get_disk_total_bytes(srcdev);
2548         device->bytes_used = btrfs_device_get_bytes_used(srcdev);
2549         ASSERT(list_empty(&srcdev->resized_list));
2550         device->commit_total_bytes = srcdev->commit_total_bytes;
2551         device->commit_bytes_used = device->bytes_used;
2552         device->dev_root = fs_info->dev_root;
2553         device->bdev = bdev;
2554         device->in_fs_metadata = 1;
2555         device->is_tgtdev_for_dev_replace = 1;
2556         device->mode = FMODE_EXCL;
2557         device->dev_stats_valid = 1;
2558         set_blocksize(device->bdev, 4096);
2559         device->fs_devices = fs_info->fs_devices;
2560         list_add(&device->dev_list, &fs_info->fs_devices->devices);
2561         fs_info->fs_devices->num_devices++;
2562         fs_info->fs_devices->open_devices++;
2563         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2564
2565         *device_out = device;
2566         return ret;
2567
2568 error:
2569         blkdev_put(bdev, FMODE_EXCL);
2570         return ret;
2571 }
2572
2573 void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info *fs_info,
2574                                               struct btrfs_device *tgtdev)
2575 {
2576         WARN_ON(fs_info->fs_devices->rw_devices == 0);
2577         tgtdev->io_width = fs_info->dev_root->sectorsize;
2578         tgtdev->io_align = fs_info->dev_root->sectorsize;
2579         tgtdev->sector_size = fs_info->dev_root->sectorsize;
2580         tgtdev->dev_root = fs_info->dev_root;
2581         tgtdev->in_fs_metadata = 1;
2582 }
2583
2584 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2585                                         struct btrfs_device *device)
2586 {
2587         int ret;
2588         struct btrfs_path *path;
2589         struct btrfs_root *root;
2590         struct btrfs_dev_item *dev_item;
2591         struct extent_buffer *leaf;
2592         struct btrfs_key key;
2593
2594         root = device->dev_root->fs_info->chunk_root;
2595
2596         path = btrfs_alloc_path();
2597         if (!path)
2598                 return -ENOMEM;
2599
2600         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2601         key.type = BTRFS_DEV_ITEM_KEY;
2602         key.offset = device->devid;
2603
2604         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2605         if (ret < 0)
2606                 goto out;
2607
2608         if (ret > 0) {
2609                 ret = -ENOENT;
2610                 goto out;
2611         }
2612
2613         leaf = path->nodes[0];
2614         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2615
2616         btrfs_set_device_id(leaf, dev_item, device->devid);
2617         btrfs_set_device_type(leaf, dev_item, device->type);
2618         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2619         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2620         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2621         btrfs_set_device_total_bytes(leaf, dev_item,
2622                                      btrfs_device_get_disk_total_bytes(device));
2623         btrfs_set_device_bytes_used(leaf, dev_item,
2624                                     btrfs_device_get_bytes_used(device));
2625         btrfs_mark_buffer_dirty(leaf);
2626
2627 out:
2628         btrfs_free_path(path);
2629         return ret;
2630 }
2631
2632 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2633                       struct btrfs_device *device, u64 new_size)
2634 {
2635         struct btrfs_super_block *super_copy =
2636                 device->dev_root->fs_info->super_copy;
2637         struct btrfs_fs_devices *fs_devices;
2638         u64 old_total;
2639         u64 diff;
2640
2641         if (!device->writeable)
2642                 return -EACCES;
2643
2644         lock_chunks(device->dev_root);
2645         old_total = btrfs_super_total_bytes(super_copy);
2646         diff = new_size - device->total_bytes;
2647
2648         if (new_size <= device->total_bytes ||
2649             device->is_tgtdev_for_dev_replace) {
2650                 unlock_chunks(device->dev_root);
2651                 return -EINVAL;
2652         }
2653
2654         fs_devices = device->dev_root->fs_info->fs_devices;
2655
2656         btrfs_set_super_total_bytes(super_copy, old_total + diff);
2657         device->fs_devices->total_rw_bytes += diff;
2658
2659         btrfs_device_set_total_bytes(device, new_size);
2660         btrfs_device_set_disk_total_bytes(device, new_size);
2661         btrfs_clear_space_info_full(device->dev_root->fs_info);
2662         if (list_empty(&device->resized_list))
2663                 list_add_tail(&device->resized_list,
2664                               &fs_devices->resized_devices);
2665         unlock_chunks(device->dev_root);
2666
2667         return btrfs_update_device(trans, device);
2668 }
2669
2670 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
2671                             struct btrfs_root *root, u64 chunk_objectid,
2672                             u64 chunk_offset)
2673 {
2674         int ret;
2675         struct btrfs_path *path;
2676         struct btrfs_key key;
2677
2678         root = root->fs_info->chunk_root;
2679         path = btrfs_alloc_path();
2680         if (!path)
2681                 return -ENOMEM;
2682
2683         key.objectid = chunk_objectid;
2684         key.offset = chunk_offset;
2685         key.type = BTRFS_CHUNK_ITEM_KEY;
2686
2687         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2688         if (ret < 0)
2689                 goto out;
2690         else if (ret > 0) { /* Logic error or corruption */
2691                 btrfs_handle_fs_error(root->fs_info, -ENOENT,
2692                             "Failed lookup while freeing chunk.");
2693                 ret = -ENOENT;
2694                 goto out;
2695         }
2696
2697         ret = btrfs_del_item(trans, root, path);
2698         if (ret < 0)
2699                 btrfs_handle_fs_error(root->fs_info, ret,
2700                             "Failed to delete chunk item.");
2701 out:
2702         btrfs_free_path(path);
2703         return ret;
2704 }
2705
2706 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
2707                         chunk_offset)
2708 {
2709         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
2710         struct btrfs_disk_key *disk_key;
2711         struct btrfs_chunk *chunk;
2712         u8 *ptr;
2713         int ret = 0;
2714         u32 num_stripes;
2715         u32 array_size;
2716         u32 len = 0;
2717         u32 cur;
2718         struct btrfs_key key;
2719
2720         lock_chunks(root);
2721         array_size = btrfs_super_sys_array_size(super_copy);
2722
2723         ptr = super_copy->sys_chunk_array;
2724         cur = 0;
2725
2726         while (cur < array_size) {
2727                 disk_key = (struct btrfs_disk_key *)ptr;
2728                 btrfs_disk_key_to_cpu(&key, disk_key);
2729
2730                 len = sizeof(*disk_key);
2731
2732                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2733                         chunk = (struct btrfs_chunk *)(ptr + len);
2734                         num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2735                         len += btrfs_chunk_item_size(num_stripes);
2736                 } else {
2737                         ret = -EIO;
2738                         break;
2739                 }
2740                 if (key.objectid == chunk_objectid &&
2741                     key.offset == chunk_offset) {
2742                         memmove(ptr, ptr + len, array_size - (cur + len));
2743                         array_size -= len;
2744                         btrfs_set_super_sys_array_size(super_copy, array_size);
2745                 } else {
2746                         ptr += len;
2747                         cur += len;
2748                 }
2749         }
2750         unlock_chunks(root);
2751         return ret;
2752 }
2753
2754 int btrfs_remove_chunk(struct btrfs_trans_handle *trans,
2755                        struct btrfs_root *root, u64 chunk_offset)
2756 {
2757         struct extent_map_tree *em_tree;
2758         struct extent_map *em;
2759         struct btrfs_root *extent_root = root->fs_info->extent_root;
2760         struct map_lookup *map;
2761         u64 dev_extent_len = 0;
2762         u64 chunk_objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2763         int i, ret = 0;
2764
2765         /* Just in case */
2766         root = root->fs_info->chunk_root;
2767         em_tree = &root->fs_info->mapping_tree.map_tree;
2768
2769         read_lock(&em_tree->lock);
2770         em = lookup_extent_mapping(em_tree, chunk_offset, 1);
2771         read_unlock(&em_tree->lock);
2772
2773         if (!em || em->start > chunk_offset ||
2774             em->start + em->len < chunk_offset) {
2775                 /*
2776                  * This is a logic error, but we don't want to just rely on the
2777                  * user having built with ASSERT enabled, so if ASSERT doesn't
2778                  * do anything we still error out.
2779                  */
2780                 ASSERT(0);
2781                 if (em)
2782                         free_extent_map(em);
2783                 return -EINVAL;
2784         }
2785         map = em->map_lookup;
2786         lock_chunks(root->fs_info->chunk_root);
2787         check_system_chunk(trans, extent_root, map->type);
2788         unlock_chunks(root->fs_info->chunk_root);
2789
2790         for (i = 0; i < map->num_stripes; i++) {
2791                 struct btrfs_device *device = map->stripes[i].dev;
2792                 ret = btrfs_free_dev_extent(trans, device,
2793                                             map->stripes[i].physical,
2794                                             &dev_extent_len);
2795                 if (ret) {
2796                         btrfs_abort_transaction(trans, root, ret);
2797                         goto out;
2798                 }
2799
2800                 if (device->bytes_used > 0) {
2801                         lock_chunks(root);
2802                         btrfs_device_set_bytes_used(device,
2803                                         device->bytes_used - dev_extent_len);
2804                         spin_lock(&root->fs_info->free_chunk_lock);
2805                         root->fs_info->free_chunk_space += dev_extent_len;
2806                         spin_unlock(&root->fs_info->free_chunk_lock);
2807                         btrfs_clear_space_info_full(root->fs_info);
2808                         unlock_chunks(root);
2809                 }
2810
2811                 if (map->stripes[i].dev) {
2812                         ret = btrfs_update_device(trans, map->stripes[i].dev);
2813                         if (ret) {
2814                                 btrfs_abort_transaction(trans, root, ret);
2815                                 goto out;
2816                         }
2817                 }
2818         }
2819         ret = btrfs_free_chunk(trans, root, chunk_objectid, chunk_offset);
2820         if (ret) {
2821                 btrfs_abort_transaction(trans, root, ret);
2822                 goto out;
2823         }
2824
2825         trace_btrfs_chunk_free(root, map, chunk_offset, em->len);
2826
2827         if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2828                 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
2829                 if (ret) {
2830                         btrfs_abort_transaction(trans, root, ret);
2831                         goto out;
2832                 }
2833         }
2834
2835         ret = btrfs_remove_block_group(trans, extent_root, chunk_offset, em);
2836         if (ret) {
2837                 btrfs_abort_transaction(trans, extent_root, ret);
2838                 goto out;
2839         }
2840
2841 out:
2842         /* once for us */
2843         free_extent_map(em);
2844         return ret;
2845 }
2846
2847 static int btrfs_relocate_chunk(struct btrfs_root *root, u64 chunk_offset)
2848 {
2849         struct btrfs_root *extent_root;
2850         struct btrfs_trans_handle *trans;
2851         int ret;
2852
2853         root = root->fs_info->chunk_root;
2854         extent_root = root->fs_info->extent_root;
2855
2856         /*
2857          * Prevent races with automatic removal of unused block groups.
2858          * After we relocate and before we remove the chunk with offset
2859          * chunk_offset, automatic removal of the block group can kick in,
2860          * resulting in a failure when calling btrfs_remove_chunk() below.
2861          *
2862          * Make sure to acquire this mutex before doing a tree search (dev
2863          * or chunk trees) to find chunks. Otherwise the cleaner kthread might
2864          * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after
2865          * we release the path used to search the chunk/dev tree and before
2866          * the current task acquires this mutex and calls us.
2867          */
2868         ASSERT(mutex_is_locked(&root->fs_info->delete_unused_bgs_mutex));
2869
2870         ret = btrfs_can_relocate(extent_root, chunk_offset);
2871         if (ret)
2872                 return -ENOSPC;
2873
2874         /* step one, relocate all the extents inside this chunk */
2875         btrfs_scrub_pause(root);
2876         ret = btrfs_relocate_block_group(extent_root, chunk_offset);
2877         btrfs_scrub_continue(root);
2878         if (ret)
2879                 return ret;
2880
2881         trans = btrfs_start_trans_remove_block_group(root->fs_info,
2882                                                      chunk_offset);
2883         if (IS_ERR(trans)) {
2884                 ret = PTR_ERR(trans);
2885                 btrfs_handle_fs_error(root->fs_info, ret, NULL);
2886                 return ret;
2887         }
2888
2889         /*
2890          * step two, delete the device extents and the
2891          * chunk tree entries
2892          */
2893         ret = btrfs_remove_chunk(trans, root, chunk_offset);
2894         btrfs_end_transaction(trans, root);
2895         return ret;
2896 }
2897
2898 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
2899 {
2900         struct btrfs_root *chunk_root = root->fs_info->chunk_root;
2901         struct btrfs_path *path;
2902         struct extent_buffer *leaf;
2903         struct btrfs_chunk *chunk;
2904         struct btrfs_key key;
2905         struct btrfs_key found_key;
2906         u64 chunk_type;
2907         bool retried = false;
2908         int failed = 0;
2909         int ret;
2910
2911         path = btrfs_alloc_path();
2912         if (!path)
2913                 return -ENOMEM;
2914
2915 again:
2916         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2917         key.offset = (u64)-1;
2918         key.type = BTRFS_CHUNK_ITEM_KEY;
2919
2920         while (1) {
2921                 mutex_lock(&root->fs_info->delete_unused_bgs_mutex);
2922                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2923                 if (ret < 0) {
2924                         mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
2925                         goto error;
2926                 }
2927                 BUG_ON(ret == 0); /* Corruption */
2928
2929                 ret = btrfs_previous_item(chunk_root, path, key.objectid,
2930                                           key.type);
2931                 if (ret)
2932                         mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
2933                 if (ret < 0)
2934                         goto error;
2935                 if (ret > 0)
2936                         break;
2937
2938                 leaf = path->nodes[0];
2939                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2940
2941                 chunk = btrfs_item_ptr(leaf, path->slots[0],
2942                                        struct btrfs_chunk);
2943                 chunk_type = btrfs_chunk_type(leaf, chunk);
2944                 btrfs_release_path(path);
2945
2946                 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
2947                         ret = btrfs_relocate_chunk(chunk_root,
2948                                                    found_key.offset);
2949                         if (ret == -ENOSPC)
2950                                 failed++;
2951                         else
2952                                 BUG_ON(ret);
2953                 }
2954                 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
2955
2956                 if (found_key.offset == 0)
2957                         break;
2958                 key.offset = found_key.offset - 1;
2959         }
2960         ret = 0;
2961         if (failed && !retried) {
2962                 failed = 0;
2963                 retried = true;
2964                 goto again;
2965         } else if (WARN_ON(failed && retried)) {
2966                 ret = -ENOSPC;
2967         }
2968 error:
2969         btrfs_free_path(path);
2970         return ret;
2971 }
2972
2973 static int insert_balance_item(struct btrfs_root *root,
2974                                struct btrfs_balance_control *bctl)
2975 {
2976         struct btrfs_trans_handle *trans;
2977         struct btrfs_balance_item *item;
2978         struct btrfs_disk_balance_args disk_bargs;
2979         struct btrfs_path *path;
2980         struct extent_buffer *leaf;
2981         struct btrfs_key key;
2982         int ret, err;
2983
2984         path = btrfs_alloc_path();
2985         if (!path)
2986                 return -ENOMEM;
2987
2988         trans = btrfs_start_transaction(root, 0);
2989         if (IS_ERR(trans)) {
2990                 btrfs_free_path(path);
2991                 return PTR_ERR(trans);
2992         }
2993
2994         key.objectid = BTRFS_BALANCE_OBJECTID;
2995         key.type = BTRFS_TEMPORARY_ITEM_KEY;
2996         key.offset = 0;
2997
2998         ret = btrfs_insert_empty_item(trans, root, path, &key,
2999                                       sizeof(*item));
3000         if (ret)
3001                 goto out;
3002
3003         leaf = path->nodes[0];
3004         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3005
3006         memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
3007
3008         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
3009         btrfs_set_balance_data(leaf, item, &disk_bargs);
3010         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
3011         btrfs_set_balance_meta(leaf, item, &disk_bargs);
3012         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
3013         btrfs_set_balance_sys(leaf, item, &disk_bargs);
3014
3015         btrfs_set_balance_flags(leaf, item, bctl->flags);
3016
3017         btrfs_mark_buffer_dirty(leaf);
3018 out:
3019         btrfs_free_path(path);
3020         err = btrfs_commit_transaction(trans, root);
3021         if (err && !ret)
3022                 ret = err;
3023         return ret;
3024 }
3025
3026 static int del_balance_item(struct btrfs_root *root)
3027 {
3028         struct btrfs_trans_handle *trans;
3029         struct btrfs_path *path;
3030         struct btrfs_key key;
3031         int ret, err;
3032
3033         path = btrfs_alloc_path();
3034         if (!path)
3035                 return -ENOMEM;
3036
3037         trans = btrfs_start_transaction(root, 0);
3038         if (IS_ERR(trans)) {
3039                 btrfs_free_path(path);
3040                 return PTR_ERR(trans);
3041         }
3042
3043         key.objectid = BTRFS_BALANCE_OBJECTID;
3044         key.type = BTRFS_TEMPORARY_ITEM_KEY;
3045         key.offset = 0;
3046
3047         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3048         if (ret < 0)
3049                 goto out;
3050         if (ret > 0) {
3051                 ret = -ENOENT;
3052                 goto out;
3053         }
3054
3055         ret = btrfs_del_item(trans, root, path);
3056 out:
3057         btrfs_free_path(path);
3058         err = btrfs_commit_transaction(trans, root);
3059         if (err && !ret)
3060                 ret = err;
3061         return ret;
3062 }
3063
3064 /*
3065  * This is a heuristic used to reduce the number of chunks balanced on
3066  * resume after balance was interrupted.
3067  */
3068 static void update_balance_args(struct btrfs_balance_control *bctl)
3069 {
3070         /*
3071          * Turn on soft mode for chunk types that were being converted.
3072          */
3073         if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
3074                 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
3075         if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
3076                 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
3077         if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
3078                 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
3079
3080         /*
3081          * Turn on usage filter if is not already used.  The idea is
3082          * that chunks that we have already balanced should be
3083          * reasonably full.  Don't do it for chunks that are being
3084          * converted - that will keep us from relocating unconverted
3085          * (albeit full) chunks.
3086          */
3087         if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3088             !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3089             !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3090                 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
3091                 bctl->data.usage = 90;
3092         }
3093         if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3094             !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3095             !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3096                 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
3097                 bctl->sys.usage = 90;
3098         }
3099         if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3100             !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3101             !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3102                 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
3103                 bctl->meta.usage = 90;
3104         }
3105 }
3106
3107 /*
3108  * Should be called with both balance and volume mutexes held to
3109  * serialize other volume operations (add_dev/rm_dev/resize) with
3110  * restriper.  Same goes for unset_balance_control.
3111  */
3112 static void set_balance_control(struct btrfs_balance_control *bctl)
3113 {
3114         struct btrfs_fs_info *fs_info = bctl->fs_info;
3115
3116         BUG_ON(fs_info->balance_ctl);
3117
3118         spin_lock(&fs_info->balance_lock);
3119         fs_info->balance_ctl = bctl;
3120         spin_unlock(&fs_info->balance_lock);
3121 }
3122
3123 static void unset_balance_control(struct btrfs_fs_info *fs_info)
3124 {
3125         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3126
3127         BUG_ON(!fs_info->balance_ctl);
3128
3129         spin_lock(&fs_info->balance_lock);
3130         fs_info->balance_ctl = NULL;
3131         spin_unlock(&fs_info->balance_lock);
3132
3133         kfree(bctl);
3134 }
3135
3136 /*
3137  * Balance filters.  Return 1 if chunk should be filtered out
3138  * (should not be balanced).
3139  */
3140 static int chunk_profiles_filter(u64 chunk_type,
3141                                  struct btrfs_balance_args *bargs)
3142 {
3143         chunk_type = chunk_to_extended(chunk_type) &
3144                                 BTRFS_EXTENDED_PROFILE_MASK;
3145
3146         if (bargs->profiles & chunk_type)
3147                 return 0;
3148
3149         return 1;
3150 }
3151
3152 static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
3153                               struct btrfs_balance_args *bargs)
3154 {
3155         struct btrfs_block_group_cache *cache;
3156         u64 chunk_used;
3157         u64 user_thresh_min;
3158         u64 user_thresh_max;
3159         int ret = 1;
3160
3161         cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3162         chunk_used = btrfs_block_group_used(&cache->item);
3163
3164         if (bargs->usage_min == 0)
3165                 user_thresh_min = 0;
3166         else
3167                 user_thresh_min = div_factor_fine(cache->key.offset,
3168                                         bargs->usage_min);
3169
3170         if (bargs->usage_max == 0)
3171                 user_thresh_max = 1;
3172         else if (bargs->usage_max > 100)
3173                 user_thresh_max = cache->key.offset;
3174         else
3175                 user_thresh_max = div_factor_fine(cache->key.offset,
3176                                         bargs->usage_max);
3177
3178         if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max)
3179                 ret = 0;
3180
3181         btrfs_put_block_group(cache);
3182         return ret;
3183 }
3184
3185 static int chunk_usage_filter(struct btrfs_fs_info *fs_info,
3186                 u64 chunk_offset, struct btrfs_balance_args *bargs)
3187 {
3188         struct btrfs_block_group_cache *cache;
3189         u64 chunk_used, user_thresh;
3190         int ret = 1;
3191
3192         cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3193         chunk_used = btrfs_block_group_used(&cache->item);
3194
3195         if (bargs->usage_min == 0)
3196                 user_thresh = 1;
3197         else if (bargs->usage > 100)
3198                 user_thresh = cache->key.offset;
3199         else
3200                 user_thresh = div_factor_fine(cache->key.offset,
3201                                               bargs->usage);
3202
3203         if (chunk_used < user_thresh)
3204                 ret = 0;
3205
3206         btrfs_put_block_group(cache);
3207         return ret;
3208 }
3209
3210 static int chunk_devid_filter(struct extent_buffer *leaf,
3211                               struct btrfs_chunk *chunk,
3212                               struct btrfs_balance_args *bargs)
3213 {
3214         struct btrfs_stripe *stripe;
3215         int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3216         int i;
3217
3218         for (i = 0; i < num_stripes; i++) {
3219                 stripe = btrfs_stripe_nr(chunk, i);
3220                 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
3221                         return 0;
3222         }
3223
3224         return 1;
3225 }
3226
3227 /* [pstart, pend) */
3228 static int chunk_drange_filter(struct extent_buffer *leaf,
3229                                struct btrfs_chunk *chunk,
3230                                u64 chunk_offset,
3231                                struct btrfs_balance_args *bargs)
3232 {
3233         struct btrfs_stripe *stripe;
3234         int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3235         u64 stripe_offset;
3236         u64 stripe_length;
3237         int factor;
3238         int i;
3239
3240         if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
3241                 return 0;
3242
3243         if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP |
3244              BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) {
3245                 factor = num_stripes / 2;
3246         } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) {
3247                 factor = num_stripes - 1;
3248         } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) {
3249                 factor = num_stripes - 2;
3250         } else {
3251                 factor = num_stripes;
3252         }
3253
3254         for (i = 0; i < num_stripes; i++) {
3255                 stripe = btrfs_stripe_nr(chunk, i);
3256                 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
3257                         continue;
3258
3259                 stripe_offset = btrfs_stripe_offset(leaf, stripe);
3260                 stripe_length = btrfs_chunk_length(leaf, chunk);
3261                 stripe_length = div_u64(stripe_length, factor);
3262
3263                 if (stripe_offset < bargs->pend &&
3264                     stripe_offset + stripe_length > bargs->pstart)
3265                         return 0;
3266         }
3267
3268         return 1;
3269 }
3270
3271 /* [vstart, vend) */
3272 static int chunk_vrange_filter(struct extent_buffer *leaf,
3273                                struct btrfs_chunk *chunk,
3274                                u64 chunk_offset,
3275                                struct btrfs_balance_args *bargs)
3276 {
3277         if (chunk_offset < bargs->vend &&
3278             chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
3279                 /* at least part of the chunk is inside this vrange */
3280                 return 0;
3281
3282         return 1;
3283 }
3284
3285 static int chunk_stripes_range_filter(struct extent_buffer *leaf,
3286                                struct btrfs_chunk *chunk,
3287                                struct btrfs_balance_args *bargs)
3288 {
3289         int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3290
3291         if (bargs->stripes_min <= num_stripes
3292                         && num_stripes <= bargs->stripes_max)
3293                 return 0;
3294
3295         return 1;
3296 }
3297
3298 static int chunk_soft_convert_filter(u64 chunk_type,
3299                                      struct btrfs_balance_args *bargs)
3300 {
3301         if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
3302                 return 0;
3303
3304         chunk_type = chunk_to_extended(chunk_type) &
3305                                 BTRFS_EXTENDED_PROFILE_MASK;
3306
3307         if (bargs->target == chunk_type)
3308                 return 1;
3309
3310         return 0;
3311 }
3312
3313 static int should_balance_chunk(struct btrfs_root *root,
3314                                 struct extent_buffer *leaf,
3315                                 struct btrfs_chunk *chunk, u64 chunk_offset)
3316 {
3317         struct btrfs_balance_control *bctl = root->fs_info->balance_ctl;
3318         struct btrfs_balance_args *bargs = NULL;
3319         u64 chunk_type = btrfs_chunk_type(leaf, chunk);
3320
3321         /* type filter */
3322         if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
3323               (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
3324                 return 0;
3325         }
3326
3327         if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3328                 bargs = &bctl->data;
3329         else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3330                 bargs = &bctl->sys;
3331         else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3332                 bargs = &bctl->meta;
3333
3334         /* profiles filter */
3335         if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
3336             chunk_profiles_filter(chunk_type, bargs)) {
3337                 return 0;
3338         }
3339
3340         /* usage filter */
3341         if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
3342             chunk_usage_filter(bctl->fs_info, chunk_offset, bargs)) {
3343                 return 0;
3344         } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3345             chunk_usage_range_filter(bctl->fs_info, chunk_offset, bargs)) {
3346                 return 0;
3347         }
3348
3349         /* devid filter */
3350         if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
3351             chunk_devid_filter(leaf, chunk, bargs)) {
3352                 return 0;
3353         }
3354
3355         /* drange filter, makes sense only with devid filter */
3356         if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
3357             chunk_drange_filter(leaf, chunk, chunk_offset, bargs)) {
3358                 return 0;
3359         }
3360
3361         /* vrange filter */
3362         if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
3363             chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
3364                 return 0;
3365         }
3366
3367         /* stripes filter */
3368         if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) &&
3369             chunk_stripes_range_filter(leaf, chunk, bargs)) {
3370                 return 0;
3371         }
3372
3373         /* soft profile changing mode */
3374         if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
3375             chunk_soft_convert_filter(chunk_type, bargs)) {
3376                 return 0;
3377         }
3378
3379         /*
3380          * limited by count, must be the last filter
3381          */
3382         if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) {
3383                 if (bargs->limit == 0)
3384                         return 0;
3385                 else
3386                         bargs->limit--;
3387         } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) {
3388                 /*
3389                  * Same logic as the 'limit' filter; the minimum cannot be
3390                  * determined here because we do not have the global information
3391                  * about the count of all chunks that satisfy the filters.
3392                  */
3393                 if (bargs->limit_max == 0)
3394                         return 0;
3395                 else
3396                         bargs->limit_max--;
3397         }
3398
3399         return 1;
3400 }
3401
3402 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
3403 {
3404         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3405         struct btrfs_root *chunk_root = fs_info->chunk_root;
3406         struct btrfs_root *dev_root = fs_info->dev_root;
3407         struct list_head *devices;
3408         struct btrfs_device *device;
3409         u64 old_size;
3410         u64 size_to_free;
3411         u64 chunk_type;
3412         struct btrfs_chunk *chunk;
3413         struct btrfs_path *path;
3414         struct btrfs_key key;
3415         struct btrfs_key found_key;
3416         struct btrfs_trans_handle *trans;
3417         struct extent_buffer *leaf;
3418         int slot;
3419         int ret;
3420         int enospc_errors = 0;
3421         bool counting = true;
3422         /* The single value limit and min/max limits use the same bytes in the */
3423         u64 limit_data = bctl->data.limit;
3424         u64 limit_meta = bctl->meta.limit;
3425         u64 limit_sys = bctl->sys.limit;
3426         u32 count_data = 0;
3427         u32 count_meta = 0;
3428         u32 count_sys = 0;
3429         int chunk_reserved = 0;
3430         u64 bytes_used = 0;
3431
3432         /* step one make some room on all the devices */
3433         devices = &fs_info->fs_devices->devices;
3434         list_for_each_entry(device, devices, dev_list) {
3435                 old_size = btrfs_device_get_total_bytes(device);
3436                 size_to_free = div_factor(old_size, 1);
3437                 size_to_free = min_t(u64, size_to_free, SZ_1M);
3438                 if (!device->writeable ||
3439                     btrfs_device_get_total_bytes(device) -
3440                     btrfs_device_get_bytes_used(device) > size_to_free ||
3441                     device->is_tgtdev_for_dev_replace)
3442                         continue;
3443
3444                 ret = btrfs_shrink_device(device, old_size - size_to_free);
3445                 if (ret == -ENOSPC)
3446                         break;
3447                 BUG_ON(ret);
3448
3449                 trans = btrfs_start_transaction(dev_root, 0);
3450                 BUG_ON(IS_ERR(trans));
3451
3452                 ret = btrfs_grow_device(trans, device, old_size);
3453                 BUG_ON(ret);
3454
3455                 btrfs_end_transaction(trans, dev_root);
3456         }
3457
3458         /* step two, relocate all the chunks */
3459         path = btrfs_alloc_path();
3460         if (!path) {
3461                 ret = -ENOMEM;
3462                 goto error;
3463         }
3464
3465         /* zero out stat counters */
3466         spin_lock(&fs_info->balance_lock);
3467         memset(&bctl->stat, 0, sizeof(bctl->stat));
3468         spin_unlock(&fs_info->balance_lock);
3469 again:
3470         if (!counting) {
3471                 /*
3472                  * The single value limit and min/max limits use the same bytes
3473                  * in the
3474                  */
3475                 bctl->data.limit = limit_data;
3476                 bctl->meta.limit = limit_meta;
3477                 bctl->sys.limit = limit_sys;
3478         }
3479         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3480         key.offset = (u64)-1;
3481         key.type = BTRFS_CHUNK_ITEM_KEY;
3482
3483         while (1) {
3484                 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
3485                     atomic_read(&fs_info->balance_cancel_req)) {
3486                         ret = -ECANCELED;
3487                         goto error;
3488                 }
3489
3490                 mutex_lock(&fs_info->delete_unused_bgs_mutex);
3491                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3492                 if (ret < 0) {
3493                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3494                         goto error;
3495                 }
3496
3497                 /*
3498                  * this shouldn't happen, it means the last relocate
3499                  * failed
3500                  */
3501                 if (ret == 0)
3502                         BUG(); /* FIXME break ? */
3503
3504                 ret = btrfs_previous_item(chunk_root, path, 0,
3505                                           BTRFS_CHUNK_ITEM_KEY);
3506                 if (ret) {
3507                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3508                         ret = 0;
3509                         break;
3510                 }
3511
3512                 leaf = path->nodes[0];
3513                 slot = path->slots[0];
3514                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3515
3516                 if (found_key.objectid != key.objectid) {
3517                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3518                         break;
3519                 }
3520
3521                 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3522                 chunk_type = btrfs_chunk_type(leaf, chunk);
3523
3524                 if (!counting) {
3525                         spin_lock(&fs_info->balance_lock);
3526                         bctl->stat.considered++;
3527                         spin_unlock(&fs_info->balance_lock);
3528                 }
3529
3530                 ret = should_balance_chunk(chunk_root, leaf, chunk,
3531                                            found_key.offset);
3532
3533                 btrfs_release_path(path);
3534                 if (!ret) {
3535                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3536                         goto loop;
3537                 }
3538
3539                 if (counting) {
3540                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3541                         spin_lock(&fs_info->balance_lock);
3542                         bctl->stat.expected++;
3543                         spin_unlock(&fs_info->balance_lock);
3544
3545                         if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3546                                 count_data++;
3547                         else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3548                                 count_sys++;
3549                         else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3550                                 count_meta++;
3551
3552                         goto loop;
3553                 }
3554
3555                 /*
3556                  * Apply limit_min filter, no need to check if the LIMITS
3557                  * filter is used, limit_min is 0 by default
3558                  */
3559                 if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
3560                                         count_data < bctl->data.limit_min)
3561                                 || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) &&
3562                                         count_meta < bctl->meta.limit_min)
3563                                 || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) &&
3564                                         count_sys < bctl->sys.limit_min)) {
3565                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3566                         goto loop;
3567                 }
3568
3569                 ASSERT(fs_info->data_sinfo);
3570                 spin_lock(&fs_info->data_sinfo->lock);
3571                 bytes_used = fs_info->data_sinfo->bytes_used;
3572                 spin_unlock(&fs_info->data_sinfo->lock);
3573
3574                 if ((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
3575                     !chunk_reserved && !bytes_used) {
3576                         trans = btrfs_start_transaction(chunk_root, 0);
3577                         if (IS_ERR(trans)) {
3578                                 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3579                                 ret = PTR_ERR(trans);
3580                                 goto error;
3581                         }
3582
3583                         ret = btrfs_force_chunk_alloc(trans, chunk_root,
3584                                                       BTRFS_BLOCK_GROUP_DATA);
3585                         btrfs_end_transaction(trans, chunk_root);
3586                         if (ret < 0) {
3587                                 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3588                                 goto error;
3589                         }
3590                         chunk_reserved = 1;
3591                 }
3592
3593                 ret = btrfs_relocate_chunk(chunk_root,
3594                                            found_key.offset);
3595                 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3596                 if (ret && ret != -ENOSPC)
3597                         goto error;
3598                 if (ret == -ENOSPC) {
3599                         enospc_errors++;
3600                 } else {
3601                         spin_lock(&fs_info->balance_lock);
3602                         bctl->stat.completed++;
3603                         spin_unlock(&fs_info->balance_lock);
3604                 }
3605 loop:
3606                 if (found_key.offset == 0)
3607                         break;
3608                 key.offset = found_key.offset - 1;
3609         }
3610
3611         if (counting) {
3612                 btrfs_release_path(path);
3613                 counting = false;
3614                 goto again;
3615         }
3616 error:
3617         btrfs_free_path(path);
3618         if (enospc_errors) {
3619                 btrfs_info(fs_info, "%d enospc errors during balance",
3620                        enospc_errors);
3621                 if (!ret)
3622                         ret = -ENOSPC;
3623         }
3624
3625         return ret;
3626 }
3627
3628 /**
3629  * alloc_profile_is_valid - see if a given profile is valid and reduced
3630  * @flags: profile to validate
3631  * @extended: if true @flags is treated as an extended profile
3632  */
3633 static int alloc_profile_is_valid(u64 flags, int extended)
3634 {
3635         u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
3636                                BTRFS_BLOCK_GROUP_PROFILE_MASK);
3637
3638         flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
3639
3640         /* 1) check that all other bits are zeroed */
3641         if (flags & ~mask)
3642                 return 0;
3643
3644         /* 2) see if profile is reduced */
3645         if (flags == 0)
3646                 return !extended; /* "0" is valid for usual profiles */
3647
3648         /* true if exactly one bit set */
3649         return (flags & (flags - 1)) == 0;
3650 }
3651
3652 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
3653 {
3654         /* cancel requested || normal exit path */
3655         return atomic_read(&fs_info->balance_cancel_req) ||
3656                 (atomic_read(&fs_info->balance_pause_req) == 0 &&
3657                  atomic_read(&fs_info->balance_cancel_req) == 0);
3658 }
3659
3660 static void __cancel_balance(struct btrfs_fs_info *fs_info)
3661 {
3662         int ret;
3663
3664         unset_balance_control(fs_info);
3665         ret = del_balance_item(fs_info->tree_root);
3666         if (ret)
3667                 btrfs_handle_fs_error(fs_info, ret, NULL);
3668
3669         atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3670 }
3671
3672 /* Non-zero return value signifies invalidity */
3673 static inline int validate_convert_profile(struct btrfs_balance_args *bctl_arg,
3674                 u64 allowed)
3675 {
3676         return ((bctl_arg->flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3677                 (!alloc_profile_is_valid(bctl_arg->target, 1) ||
3678                  (bctl_arg->target & ~allowed)));
3679 }
3680
3681 /*
3682  * Should be called with both balance and volume mutexes held
3683  */
3684 int btrfs_balance(struct btrfs_balance_control *bctl,
3685                   struct btrfs_ioctl_balance_args *bargs)
3686 {
3687         struct btrfs_fs_info *fs_info = bctl->fs_info;
3688         u64 allowed;
3689         int mixed = 0;
3690         int ret;
3691         u64 num_devices;
3692         unsigned seq;
3693
3694         if (btrfs_fs_closing(fs_info) ||
3695             atomic_read(&fs_info->balance_pause_req) ||
3696             atomic_read(&fs_info->balance_cancel_req)) {
3697                 ret = -EINVAL;
3698                 goto out;
3699         }
3700
3701         allowed = btrfs_super_incompat_flags(fs_info->super_copy);
3702         if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
3703                 mixed = 1;
3704
3705         /*
3706          * In case of mixed groups both data and meta should be picked,
3707          * and identical options should be given for both of them.
3708          */
3709         allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
3710         if (mixed && (bctl->flags & allowed)) {
3711                 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
3712                     !(bctl->flags & BTRFS_BALANCE_METADATA) ||
3713                     memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
3714                         btrfs_err(fs_info, "with mixed groups data and "
3715                                    "metadata balance options must be the same");
3716                         ret = -EINVAL;
3717                         goto out;
3718                 }
3719         }
3720
3721         num_devices = fs_info->fs_devices->num_devices;
3722         btrfs_dev_replace_lock(&fs_info->dev_replace, 0);
3723         if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
3724                 BUG_ON(num_devices < 1);
3725                 num_devices--;
3726         }
3727         btrfs_dev_replace_unlock(&fs_info->dev_replace, 0);
3728         allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE | BTRFS_BLOCK_GROUP_DUP;
3729         if (num_devices > 1)
3730                 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1);
3731         if (num_devices > 2)
3732                 allowed |= BTRFS_BLOCK_GROUP_RAID5;
3733         if (num_devices > 3)
3734                 allowed |= (BTRFS_BLOCK_GROUP_RAID10 |
3735                             BTRFS_BLOCK_GROUP_RAID6);
3736         if (validate_convert_profile(&bctl->data, allowed)) {
3737                 btrfs_err(fs_info, "unable to start balance with target "
3738                            "data profile %llu",
3739                        bctl->data.target);
3740                 ret = -EINVAL;
3741                 goto out;
3742         }
3743         if (validate_convert_profile(&bctl->meta, allowed)) {
3744                 btrfs_err(fs_info,
3745                            "unable to start balance with target metadata profile %llu",
3746                        bctl->meta.target);
3747                 ret = -EINVAL;
3748                 goto out;
3749         }
3750         if (validate_convert_profile(&bctl->sys, allowed)) {
3751                 btrfs_err(fs_info,
3752                            "unable to start balance with target system profile %llu",
3753                        bctl->sys.target);
3754                 ret = -EINVAL;
3755                 goto out;
3756         }
3757
3758         /* allow to reduce meta or sys integrity only if force set */
3759         allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3760                         BTRFS_BLOCK_GROUP_RAID10 |
3761                         BTRFS_BLOCK_GROUP_RAID5 |
3762                         BTRFS_BLOCK_GROUP_RAID6;
3763         do {
3764                 seq = read_seqbegin(&fs_info->profiles_lock);
3765
3766                 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3767                      (fs_info->avail_system_alloc_bits & allowed) &&
3768                      !(bctl->sys.target & allowed)) ||
3769                     ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3770                      (fs_info->avail_metadata_alloc_bits & allowed) &&
3771                      !(bctl->meta.target & allowed))) {
3772                         if (bctl->flags & BTRFS_BALANCE_FORCE) {
3773                                 btrfs_info(fs_info, "force reducing metadata integrity");
3774                         } else {
3775                                 btrfs_err(fs_info, "balance will reduce metadata "
3776                                            "integrity, use force if you want this");
3777                                 ret = -EINVAL;
3778                                 goto out;
3779                         }
3780                 }
3781         } while (read_seqretry(&fs_info->profiles_lock, seq));
3782
3783         if (btrfs_get_num_tolerated_disk_barrier_failures(bctl->meta.target) <
3784                 btrfs_get_num_tolerated_disk_barrier_failures(bctl->data.target)) {
3785                 btrfs_warn(fs_info,
3786         "metadata profile 0x%llx has lower redundancy than data profile 0x%llx",
3787                         bctl->meta.target, bctl->data.target);
3788         }
3789
3790         if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3791                 fs_info->num_tolerated_disk_barrier_failures = min(
3792                         btrfs_calc_num_tolerated_disk_barrier_failures(fs_info),
3793                         btrfs_get_num_tolerated_disk_barrier_failures(
3794                                 bctl->sys.target));
3795         }
3796
3797         ret = insert_balance_item(fs_info->tree_root, bctl);
3798         if (ret && ret != -EEXIST)
3799                 goto out;
3800
3801         if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
3802                 BUG_ON(ret == -EEXIST);
3803                 set_balance_control(bctl);
3804         } else {
3805                 BUG_ON(ret != -EEXIST);
3806                 spin_lock(&fs_info->balance_lock);
3807                 update_balance_args(bctl);
3808                 spin_unlock(&fs_info->balance_lock);
3809         }
3810
3811         atomic_inc(&fs_info->balance_running);
3812         mutex_unlock(&fs_info->balance_mutex);
3813
3814         ret = __btrfs_balance(fs_info);
3815
3816         mutex_lock(&fs_info->balance_mutex);
3817         atomic_dec(&fs_info->balance_running);
3818
3819         if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3820                 fs_info->num_tolerated_disk_barrier_failures =
3821                         btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3822         }
3823
3824         if (bargs) {
3825                 memset(bargs, 0, sizeof(*bargs));
3826                 update_ioctl_balance_args(fs_info, 0, bargs);
3827         }
3828
3829         if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
3830             balance_need_close(fs_info)) {
3831                 __cancel_balance(fs_info);
3832         }
3833
3834         wake_up(&fs_info->balance_wait_q);
3835
3836         return ret;
3837 out:
3838         if (bctl->flags & BTRFS_BALANCE_RESUME)
3839                 __cancel_balance(fs_info);
3840         else {
3841                 kfree(bctl);
3842                 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3843         }
3844         return ret;
3845 }
3846
3847 static int balance_kthread(void *data)
3848 {
3849         struct btrfs_fs_info *fs_info = data;
3850         int ret = 0;
3851
3852         mutex_lock(&fs_info->volume_mutex);
3853         mutex_lock(&fs_info->balance_mutex);
3854
3855         if (fs_info->balance_ctl) {
3856                 btrfs_info(fs_info, "continuing balance");
3857                 ret = btrfs_balance(fs_info->balance_ctl, NULL);
3858         }
3859
3860         mutex_unlock(&fs_info->balance_mutex);
3861         mutex_unlock(&fs_info->volume_mutex);
3862
3863         return ret;
3864 }
3865
3866 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
3867 {
3868         struct task_struct *tsk;
3869
3870         spin_lock(&fs_info->balance_lock);
3871         if (!fs_info->balance_ctl) {
3872                 spin_unlock(&fs_info->balance_lock);
3873                 return 0;
3874         }
3875         spin_unlock(&fs_info->balance_lock);
3876
3877         if (btrfs_test_opt(fs_info->tree_root, SKIP_BALANCE)) {
3878                 btrfs_info(fs_info, "force skipping balance");
3879                 return 0;
3880         }
3881
3882         tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
3883         return PTR_ERR_OR_ZERO(tsk);
3884 }
3885
3886 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
3887 {
3888         struct btrfs_balance_control *bctl;
3889         struct btrfs_balance_item *item;
3890         struct btrfs_disk_balance_args disk_bargs;
3891         struct btrfs_path *path;
3892         struct extent_buffer *leaf;
3893         struct btrfs_key key;
3894         int ret;
3895
3896         path = btrfs_alloc_path();
3897         if (!path)
3898                 return -ENOMEM;
3899
3900         key.objectid = BTRFS_BALANCE_OBJECTID;
3901         key.type = BTRFS_TEMPORARY_ITEM_KEY;
3902         key.offset = 0;
3903
3904         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
3905         if (ret < 0)
3906                 goto out;
3907         if (ret > 0) { /* ret = -ENOENT; */
3908                 ret = 0;
3909                 goto out;
3910         }
3911
3912         bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
3913         if (!bctl) {
3914                 ret = -ENOMEM;
3915                 goto out;
3916         }
3917
3918         leaf = path->nodes[0];
3919         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3920
3921         bctl->fs_info = fs_info;
3922         bctl->flags = btrfs_balance_flags(leaf, item);
3923         bctl->flags |= BTRFS_BALANCE_RESUME;
3924
3925         btrfs_balance_data(leaf, item, &disk_bargs);
3926         btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
3927         btrfs_balance_meta(leaf, item, &disk_bargs);
3928         btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
3929         btrfs_balance_sys(leaf, item, &disk_bargs);
3930         btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
3931
3932         WARN_ON(atomic_xchg(&fs_info->mutually_exclusive_operation_running, 1));
3933
3934         mutex_lock(&fs_info->volume_mutex);
3935         mutex_lock(&fs_info->balance_mutex);
3936
3937         set_balance_control(bctl);
3938
3939         mutex_unlock(&fs_info->balance_mutex);
3940         mutex_unlock(&fs_info->volume_mutex);
3941 out:
3942         btrfs_free_path(path);
3943         return ret;
3944 }
3945
3946 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
3947 {
3948         int ret = 0;
3949
3950         mutex_lock(&fs_info->balance_mutex);
3951         if (!fs_info->balance_ctl) {
3952                 mutex_unlock(&fs_info->balance_mutex);
3953                 return -ENOTCONN;
3954         }
3955
3956         if (atomic_read(&fs_info->balance_running)) {
3957                 atomic_inc(&fs_info->balance_pause_req);
3958                 mutex_unlock(&fs_info->balance_mutex);
3959
3960                 wait_event(fs_info->balance_wait_q,
3961                            atomic_read(&fs_info->balance_running) == 0);
3962
3963                 mutex_lock(&fs_info->balance_mutex);
3964                 /* we are good with balance_ctl ripped off from under us */
3965                 BUG_ON(atomic_read(&fs_info->balance_running));
3966                 atomic_dec(&fs_info->balance_pause_req);
3967         } else {
3968                 ret = -ENOTCONN;
3969         }
3970
3971         mutex_unlock(&fs_info->balance_mutex);
3972         return ret;
3973 }
3974
3975 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
3976 {
3977         if (fs_info->sb->s_flags & MS_RDONLY)
3978                 return -EROFS;
3979
3980         mutex_lock(&fs_info->balance_mutex);
3981         if (!fs_info->balance_ctl) {
3982                 mutex_unlock(&fs_info->balance_mutex);
3983                 return -ENOTCONN;
3984         }
3985
3986         atomic_inc(&fs_info->balance_cancel_req);
3987         /*
3988          * if we are running just wait and return, balance item is
3989          * deleted in btrfs_balance in this case
3990          */
3991         if (atomic_read(&fs_info->balance_running)) {
3992                 mutex_unlock(&fs_info->balance_mutex);
3993                 wait_event(fs_info->balance_wait_q,
3994                            atomic_read(&fs_info->balance_running) == 0);
3995                 mutex_lock(&fs_info->balance_mutex);
3996         } else {
3997                 /* __cancel_balance needs volume_mutex */
3998                 mutex_unlock(&fs_info->balance_mutex);
3999                 mutex_lock(&fs_info->volume_mutex);
4000                 mutex_lock(&fs_info->balance_mutex);
4001
4002                 if (fs_info->balance_ctl)
4003                         __cancel_balance(fs_info);
4004
4005                 mutex_unlock(&fs_info->volume_mutex);
4006         }
4007
4008         BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running));
4009         atomic_dec(&fs_info->balance_cancel_req);
4010         mutex_unlock(&fs_info->balance_mutex);
4011         return 0;
4012 }
4013
4014 static int btrfs_uuid_scan_kthread(void *data)
4015 {
4016         struct btrfs_fs_info *fs_info = data;
4017         struct btrfs_root *root = fs_info->tree_root;
4018         struct btrfs_key key;
4019         struct btrfs_key max_key;
4020         struct btrfs_path *path = NULL;
4021         int ret = 0;
4022         struct extent_buffer *eb;
4023         int slot;
4024         struct btrfs_root_item root_item;
4025         u32 item_size;
4026         struct btrfs_trans_handle *trans = NULL;
4027
4028         path = btrfs_alloc_path();
4029         if (!path) {
4030                 ret = -ENOMEM;
4031                 goto out;
4032         }
4033
4034         key.objectid = 0;
4035         key.type = BTRFS_ROOT_ITEM_KEY;
4036         key.offset = 0;
4037
4038         max_key.objectid = (u64)-1;
4039         max_key.type = BTRFS_ROOT_ITEM_KEY;
4040         max_key.offset = (u64)-1;
4041
4042         while (1) {
4043                 ret = btrfs_search_forward(root, &key, path, 0);
4044                 if (ret) {
4045                         if (ret > 0)
4046                                 ret = 0;
4047                         break;
4048                 }
4049
4050                 if (key.type != BTRFS_ROOT_ITEM_KEY ||
4051                     (key.objectid < BTRFS_FIRST_FREE_OBJECTID &&
4052                      key.objectid != BTRFS_FS_TREE_OBJECTID) ||
4053                     key.objectid > BTRFS_LAST_FREE_OBJECTID)
4054                         goto skip;
4055
4056                 eb = path->nodes[0];
4057                 slot = path->slots[0];
4058                 item_size = btrfs_item_size_nr(eb, slot);
4059                 if (item_size < sizeof(root_item))
4060                         goto skip;
4061
4062                 read_extent_buffer(eb, &root_item,
4063                                    btrfs_item_ptr_offset(eb, slot),
4064                                    (int)sizeof(root_item));
4065                 if (btrfs_root_refs(&root_item) == 0)
4066                         goto skip;
4067
4068                 if (!btrfs_is_empty_uuid(root_item.uuid) ||
4069                     !btrfs_is_empty_uuid(root_item.received_uuid)) {
4070                         if (trans)
4071                                 goto update_tree;
4072
4073                         btrfs_release_path(path);
4074                         /*
4075                          * 1 - subvol uuid item
4076                          * 1 - received_subvol uuid item
4077                          */
4078                         trans = btrfs_start_transaction(fs_info->uuid_root, 2);
4079                         if (IS_ERR(trans)) {
4080                                 ret = PTR_ERR(trans);
4081                                 break;
4082                         }
4083                         continue;
4084                 } else {
4085                         goto skip;
4086                 }
4087 update_tree:
4088                 if (!btrfs_is_empty_uuid(root_item.uuid)) {
4089                         ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
4090                                                   root_item.uuid,
4091                                                   BTRFS_UUID_KEY_SUBVOL,
4092                                                   key.objectid);
4093                         if (ret < 0) {
4094                                 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4095                                         ret);
4096                                 break;
4097                         }
4098                 }
4099
4100                 if (!btrfs_is_empty_uuid(root_item.received_uuid)) {
4101                         ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
4102                                                   root_item.received_uuid,
4103                                                  BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4104                                                   key.objectid);
4105                         if (ret < 0) {
4106                                 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4107                                         ret);
4108                                 break;
4109                         }
4110                 }
4111
4112 skip:
4113                 if (trans) {
4114                         ret = btrfs_end_transaction(trans, fs_info->uuid_root);
4115                         trans = NULL;
4116                         if (ret)
4117                                 break;
4118                 }
4119
4120                 btrfs_release_path(path);
4121                 if (key.offset < (u64)-1) {
4122                         key.offset++;
4123                 } else if (key.type < BTRFS_ROOT_ITEM_KEY) {
4124                         key.offset = 0;
4125                         key.type = BTRFS_ROOT_ITEM_KEY;
4126                 } else if (key.objectid < (u64)-1) {
4127                         key.offset = 0;
4128                         key.type = BTRFS_ROOT_ITEM_KEY;
4129                         key.objectid++;
4130                 } else {
4131                         break;
4132                 }
4133                 cond_resched();
4134         }
4135
4136 out:
4137         btrfs_free_path(path);
4138         if (trans && !IS_ERR(trans))
4139                 btrfs_end_transaction(trans, fs_info->uuid_root);
4140         if (ret)
4141                 btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret);
4142         else
4143                 fs_info->update_uuid_tree_gen = 1;
4144         up(&fs_info->uuid_tree_rescan_sem);
4145         return 0;
4146 }
4147
4148 /*
4149  * Callback for btrfs_uuid_tree_iterate().
4150  * returns:
4151  * 0    check succeeded, the entry is not outdated.
4152  * < 0  if an error occurred.
4153  * > 0  if the check failed, which means the caller shall remove the entry.
4154  */
4155 static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info *fs_info,
4156                                        u8 *uuid, u8 type, u64 subid)
4157 {
4158         struct btrfs_key key;
4159         int ret = 0;
4160         struct btrfs_root *subvol_root;
4161
4162         if (type != BTRFS_UUID_KEY_SUBVOL &&
4163             type != BTRFS_UUID_KEY_RECEIVED_SUBVOL)
4164                 goto out;
4165
4166         key.objectid = subid;
4167         key.type = BTRFS_ROOT_ITEM_KEY;
4168         key.offset = (u64)-1;
4169         subvol_root = btrfs_read_fs_root_no_name(fs_info, &key);
4170         if (IS_ERR(subvol_root)) {
4171                 ret = PTR_ERR(subvol_root);
4172                 if (ret == -ENOENT)
4173                         ret = 1;
4174                 goto out;
4175         }
4176
4177         switch (type) {
4178         case BTRFS_UUID_KEY_SUBVOL:
4179                 if (memcmp(uuid, subvol_root->root_item.uuid, BTRFS_UUID_SIZE))
4180                         ret = 1;
4181                 break;
4182         case BTRFS_UUID_KEY_RECEIVED_SUBVOL:
4183                 if (memcmp(uuid, subvol_root->root_item.received_uuid,
4184                            BTRFS_UUID_SIZE))
4185                         ret = 1;
4186                 break;
4187         }
4188
4189 out:
4190         return ret;
4191 }
4192
4193 static int btrfs_uuid_rescan_kthread(void *data)
4194 {
4195         struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data;
4196         int ret;
4197
4198         /*
4199          * 1st step is to iterate through the existing UUID tree and
4200          * to delete all entries that contain outdated data.
4201          * 2nd step is to add all missing entries to the UUID tree.
4202          */
4203         ret = btrfs_uuid_tree_iterate(fs_info, btrfs_check_uuid_tree_entry);
4204         if (ret < 0) {
4205                 btrfs_warn(fs_info, "iterating uuid_tree failed %d", ret);
4206                 up(&fs_info->uuid_tree_rescan_sem);
4207                 return ret;
4208         }
4209         return btrfs_uuid_scan_kthread(data);
4210 }
4211
4212 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info)
4213 {
4214         struct btrfs_trans_handle *trans;
4215         struct btrfs_root *tree_root = fs_info->tree_root;
4216         struct btrfs_root *uuid_root;
4217         struct task_struct *task;
4218         int ret;
4219
4220         /*
4221          * 1 - root node
4222          * 1 - root item
4223          */
4224         trans = btrfs_start_transaction(tree_root, 2);
4225         if (IS_ERR(trans))
4226                 return PTR_ERR(trans);
4227
4228         uuid_root = btrfs_create_tree(trans, fs_info,
4229                                       BTRFS_UUID_TREE_OBJECTID);
4230         if (IS_ERR(uuid_root)) {
4231                 ret = PTR_ERR(uuid_root);
4232                 btrfs_abort_transaction(trans, tree_root, ret);
4233                 btrfs_end_transaction(trans, tree_root);
4234                 return ret;
4235         }
4236
4237         fs_info->uuid_root = uuid_root;
4238
4239         ret = btrfs_commit_transaction(trans, tree_root);
4240         if (ret)
4241                 return ret;
4242
4243         down(&fs_info->uuid_tree_rescan_sem);
4244         task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid");
4245         if (IS_ERR(task)) {
4246                 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4247                 btrfs_warn(fs_info, "failed to start uuid_scan task");
4248                 up(&fs_info->uuid_tree_rescan_sem);
4249                 return PTR_ERR(task);
4250         }
4251
4252         return 0;
4253 }
4254
4255 int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
4256 {
4257         struct task_struct *task;
4258
4259         down(&fs_info->uuid_tree_rescan_sem);
4260         task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
4261         if (IS_ERR(task)) {
4262                 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4263                 btrfs_warn(fs_info, "failed to start uuid_rescan task");
4264                 up(&fs_info->uuid_tree_rescan_sem);
4265                 return PTR_ERR(task);
4266         }
4267
4268         return 0;
4269 }
4270
4271 /*
4272  * shrinking a device means finding all of the device extents past
4273  * the new size, and then following the back refs to the chunks.
4274  * The chunk relocation code actually frees the device extent
4275  */
4276 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
4277 {
4278         struct btrfs_trans_handle *trans;
4279         struct btrfs_root *root = device->dev_root;
4280         struct btrfs_dev_extent *dev_extent = NULL;
4281         struct btrfs_path *path;
4282         u64 length;
4283         u64 chunk_offset;
4284         int ret;
4285         int slot;
4286         int failed = 0;
4287         bool retried = false;
4288         bool checked_pending_chunks = false;
4289         struct extent_buffer *l;
4290         struct btrfs_key key;
4291         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
4292         u64 old_total = btrfs_super_total_bytes(super_copy);
4293         u64 old_size = btrfs_device_get_total_bytes(device);
4294         u64 diff = old_size - new_size;
4295
4296         if (device->is_tgtdev_for_dev_replace)
4297                 return -EINVAL;
4298
4299         path = btrfs_alloc_path();
4300         if (!path)
4301                 return -ENOMEM;
4302
4303         path->reada = READA_FORWARD;
4304
4305         lock_chunks(root);
4306
4307         btrfs_device_set_total_bytes(device, new_size);
4308         if (device->writeable) {
4309                 device->fs_devices->total_rw_bytes -= diff;
4310                 spin_lock(&root->fs_info->free_chunk_lock);
4311                 root->fs_info->free_chunk_space -= diff;
4312                 spin_unlock(&root->fs_info->free_chunk_lock);
4313         }
4314         unlock_chunks(root);
4315
4316 again:
4317         key.objectid = device->devid;
4318         key.offset = (u64)-1;
4319         key.type = BTRFS_DEV_EXTENT_KEY;
4320
4321         do {
4322                 mutex_lock(&root->fs_info->delete_unused_bgs_mutex);
4323                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4324                 if (ret < 0) {
4325                         mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4326                         goto done;
4327                 }
4328
4329                 ret = btrfs_previous_item(root, path, 0, key.type);
4330                 if (ret)
4331                         mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4332                 if (ret < 0)
4333                         goto done;
4334                 if (ret) {
4335                         ret = 0;
4336                         btrfs_release_path(path);
4337                         break;
4338                 }
4339
4340                 l = path->nodes[0];
4341                 slot = path->slots[0];
4342                 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
4343
4344                 if (key.objectid != device->devid) {
4345                         mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4346                         btrfs_release_path(path);
4347                         break;
4348                 }
4349
4350                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
4351                 length = btrfs_dev_extent_length(l, dev_extent);
4352
4353                 if (key.offset + length <= new_size) {
4354                         mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4355                         btrfs_release_path(path);
4356                         break;
4357                 }
4358
4359                 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
4360                 btrfs_release_path(path);
4361
4362                 ret = btrfs_relocate_chunk(root, chunk_offset);
4363                 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4364                 if (ret && ret != -ENOSPC)
4365                         goto done;
4366                 if (ret == -ENOSPC)
4367                         failed++;
4368         } while (key.offset-- > 0);
4369
4370         if (failed && !retried) {
4371                 failed = 0;
4372                 retried = true;
4373                 goto again;
4374         } else if (failed && retried) {
4375                 ret = -ENOSPC;
4376                 goto done;
4377         }
4378
4379         /* Shrinking succeeded, else we would be at "done". */
4380         trans = btrfs_start_transaction(root, 0);
4381         if (IS_ERR(trans)) {
4382                 ret = PTR_ERR(trans);
4383                 goto done;
4384         }
4385
4386         lock_chunks(root);
4387
4388         /*
4389          * We checked in the above loop all device extents that were already in
4390          * the device tree. However before we have updated the device's
4391          * total_bytes to the new size, we might have had chunk allocations that
4392          * have not complete yet (new block groups attached to transaction
4393          * handles), and therefore their device extents were not yet in the
4394          * device tree and we missed them in the loop above. So if we have any
4395          * pending chunk using a device extent that overlaps the device range
4396          * that we can not use anymore, commit the current transaction and
4397          * repeat the search on the device tree - this way we guarantee we will
4398          * not have chunks using device extents that end beyond 'new_size'.
4399          */
4400         if (!checked_pending_chunks) {
4401                 u64 start = new_size;
4402                 u64 len = old_size - new_size;
4403
4404                 if (contains_pending_extent(trans->transaction, device,
4405                                             &start, len)) {
4406                         unlock_chunks(root);
4407                         checked_pending_chunks = true;
4408                         failed = 0;
4409                         retried = false;
4410                         ret = btrfs_commit_transaction(trans, root);
4411                         if (ret)
4412                                 goto done;
4413                         goto again;
4414                 }
4415         }
4416
4417         btrfs_device_set_disk_total_bytes(device, new_size);
4418         if (list_empty(&device->resized_list))
4419                 list_add_tail(&device->resized_list,
4420                               &root->fs_info->fs_devices->resized_devices);
4421
4422         WARN_ON(diff > old_total);
4423         btrfs_set_super_total_bytes(super_copy, old_total - diff);
4424         unlock_chunks(root);
4425
4426         /* Now btrfs_update_device() will change the on-disk size. */
4427         ret = btrfs_update_device(trans, device);
4428         btrfs_end_transaction(trans, root);
4429 done:
4430         btrfs_free_path(path);
4431         if (ret) {
4432                 lock_chunks(root);
4433                 btrfs_device_set_total_bytes(device, old_size);
4434                 if (device->writeable)
4435                         device->fs_devices->total_rw_bytes += diff;
4436                 spin_lock(&root->fs_info->free_chunk_lock);
4437                 root->fs_info->free_chunk_space += diff;
4438                 spin_unlock(&root->fs_info->free_chunk_lock);
4439                 unlock_chunks(root);
4440         }
4441         return ret;
4442 }
4443
4444 static int btrfs_add_system_chunk(struct btrfs_root *root,
4445                            struct btrfs_key *key,
4446                            struct btrfs_chunk *chunk, int item_size)
4447 {
4448         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
4449         struct btrfs_disk_key disk_key;
4450         u32 array_size;
4451         u8 *ptr;
4452
4453         lock_chunks(root);
4454         array_size = btrfs_super_sys_array_size(super_copy);
4455         if (array_size + item_size + sizeof(disk_key)
4456                         > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
4457                 unlock_chunks(root);
4458                 return -EFBIG;
4459         }
4460
4461         ptr = super_copy->sys_chunk_array + array_size;
4462         btrfs_cpu_key_to_disk(&disk_key, key);
4463         memcpy(ptr, &disk_key, sizeof(disk_key));
4464         ptr += sizeof(disk_key);
4465         memcpy(ptr, chunk, item_size);
4466         item_size += sizeof(disk_key);
4467         btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
4468         unlock_chunks(root);
4469
4470         return 0;
4471 }
4472
4473 /*
4474  * sort the devices in descending order by max_avail, total_avail
4475  */
4476 static int btrfs_cmp_device_info(const void *a, const void *b)
4477 {
4478         const struct btrfs_device_info *di_a = a;
4479         const struct btrfs_device_info *di_b = b;
4480
4481         if (di_a->max_avail > di_b->max_avail)
4482                 return -1;
4483         if (di_a->max_avail < di_b->max_avail)
4484                 return 1;
4485         if (di_a->total_avail > di_b->total_avail)
4486                 return -1;
4487         if (di_a->total_avail < di_b->total_avail)
4488                 return 1;
4489         return 0;
4490 }
4491
4492 static u32 find_raid56_stripe_len(u32 data_devices, u32 dev_stripe_target)
4493 {
4494         /* TODO allow them to set a preferred stripe size */
4495         return SZ_64K;
4496 }
4497
4498 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
4499 {
4500         if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK))
4501                 return;
4502
4503         btrfs_set_fs_incompat(info, RAID56);
4504 }
4505
4506 #define BTRFS_MAX_DEVS(r) ((BTRFS_LEAF_DATA_SIZE(r)             \
4507                         - sizeof(struct btrfs_item)             \
4508                         - sizeof(struct btrfs_chunk))           \
4509                         / sizeof(struct btrfs_stripe) + 1)
4510
4511 #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE        \
4512                                 - 2 * sizeof(struct btrfs_disk_key)     \
4513                                 - 2 * sizeof(struct btrfs_chunk))       \
4514                                 / sizeof(struct btrfs_stripe) + 1)
4515
4516 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4517                                struct btrfs_root *extent_root, u64 start,
4518                                u64 type)
4519 {
4520         struct btrfs_fs_info *info = extent_root->fs_info;
4521         struct btrfs_fs_devices *fs_devices = info->fs_devices;
4522         struct list_head *cur;
4523         struct map_lookup *map = NULL;
4524         struct extent_map_tree *em_tree;
4525         struct extent_map *em;
4526         struct btrfs_device_info *devices_info = NULL;
4527         u64 total_avail;
4528         int num_stripes;        /* total number of stripes to allocate */
4529         int data_stripes;       /* number of stripes that count for
4530                                    block group size */
4531         int sub_stripes;        /* sub_stripes info for map */
4532         int dev_stripes;        /* stripes per dev */
4533         int devs_max;           /* max devs to use */
4534         int devs_min;           /* min devs needed */
4535         int devs_increment;     /* ndevs has to be a multiple of this */
4536         int ncopies;            /* how many copies to data has */
4537         int ret;
4538         u64 max_stripe_size;
4539         u64 max_chunk_size;
4540         u64 stripe_size;
4541         u64 num_bytes;
4542         u64 raid_stripe_len = BTRFS_STRIPE_LEN;
4543         int ndevs;
4544         int i;
4545         int j;
4546         int index;
4547
4548         BUG_ON(!alloc_profile_is_valid(type, 0));
4549
4550         if (list_empty(&fs_devices->alloc_list))
4551                 return -ENOSPC;
4552
4553         index = __get_raid_index(type);
4554
4555         sub_stripes = btrfs_raid_array[index].sub_stripes;
4556         dev_stripes = btrfs_raid_array[index].dev_stripes;
4557         devs_max = btrfs_raid_array[index].devs_max;
4558         devs_min = btrfs_raid_array[index].devs_min;
4559         devs_increment = btrfs_raid_array[index].devs_increment;
4560         ncopies = btrfs_raid_array[index].ncopies;
4561
4562         if (type & BTRFS_BLOCK_GROUP_DATA) {
4563                 max_stripe_size = SZ_1G;
4564                 max_chunk_size = 10 * max_stripe_size;
4565                 if (!devs_max)
4566                         devs_max = BTRFS_MAX_DEVS(info->chunk_root);
4567         } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
4568                 /* for larger filesystems, use larger metadata chunks */
4569                 if (fs_devices->total_rw_bytes > 50ULL * SZ_1G)
4570                         max_stripe_size = SZ_1G;
4571                 else
4572                         max_stripe_size = SZ_256M;
4573                 max_chunk_size = max_stripe_size;
4574                 if (!devs_max)
4575                         devs_max = BTRFS_MAX_DEVS(info->chunk_root);
4576         } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
4577                 max_stripe_size = SZ_32M;
4578                 max_chunk_size = 2 * max_stripe_size;
4579                 if (!devs_max)
4580                         devs_max = BTRFS_MAX_DEVS_SYS_CHUNK;
4581         } else {
4582                 btrfs_err(info, "invalid chunk type 0x%llx requested",
4583                        type);
4584                 BUG_ON(1);
4585         }
4586
4587         /* we don't want a chunk larger than 10% of writeable space */
4588         max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
4589                              max_chunk_size);
4590
4591         devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info),
4592                                GFP_NOFS);
4593         if (!devices_info)
4594                 return -ENOMEM;
4595
4596         cur = fs_devices->alloc_list.next;
4597
4598         /*
4599          * in the first pass through the devices list, we gather information
4600          * about the available holes on each device.
4601          */
4602         ndevs = 0;
4603         while (cur != &fs_devices->alloc_list) {
4604                 struct btrfs_device *device;
4605                 u64 max_avail;
4606                 u64 dev_offset;
4607
4608                 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
4609
4610                 cur = cur->next;
4611
4612                 if (!device->writeable) {
4613                         WARN(1, KERN_ERR
4614                                "BTRFS: read-only device in alloc_list\n");
4615                         continue;
4616                 }
4617
4618                 if (!device->in_fs_metadata ||
4619                     device->is_tgtdev_for_dev_replace)
4620                         continue;
4621
4622                 if (device->total_bytes > device->bytes_used)
4623                         total_avail = device->total_bytes - device->bytes_used;
4624                 else
4625                         total_avail = 0;
4626
4627                 /* If there is no space on this device, skip it. */
4628                 if (total_avail == 0)
4629                         continue;
4630
4631                 ret = find_free_dev_extent(trans, device,
4632                                            max_stripe_size * dev_stripes,
4633                                            &dev_offset, &max_avail);
4634                 if (ret && ret != -ENOSPC)
4635                         goto error;
4636
4637                 if (ret == 0)
4638                         max_avail = max_stripe_size * dev_stripes;
4639
4640                 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes)
4641                         continue;
4642
4643                 if (ndevs == fs_devices->rw_devices) {
4644                         WARN(1, "%s: found more than %llu devices\n",
4645                              __func__, fs_devices->rw_devices);
4646                         break;
4647                 }
4648                 devices_info[ndevs].dev_offset = dev_offset;
4649                 devices_info[ndevs].max_avail = max_avail;
4650                 devices_info[ndevs].total_avail = total_avail;
4651                 devices_info[ndevs].dev = device;
4652                 ++ndevs;
4653         }
4654
4655         /*
4656          * now sort the devices by hole size / available space
4657          */
4658         sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
4659              btrfs_cmp_device_info, NULL);
4660
4661         /* round down to number of usable stripes */
4662         ndevs -= ndevs % devs_increment;
4663
4664         if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) {
4665                 ret = -ENOSPC;
4666                 goto error;
4667         }
4668
4669         if (devs_max && ndevs > devs_max)
4670                 ndevs = devs_max;
4671         /*
4672          * the primary goal is to maximize the number of stripes, so use as many
4673          * devices as possible, even if the stripes are not maximum sized.
4674          */
4675         stripe_size = devices_info[ndevs-1].max_avail;
4676         num_stripes = ndevs * dev_stripes;
4677
4678         /*
4679          * this will have to be fixed for RAID1 and RAID10 over
4680          * more drives
4681          */
4682         data_stripes = num_stripes / ncopies;
4683
4684         if (type & BTRFS_BLOCK_GROUP_RAID5) {
4685                 raid_stripe_len = find_raid56_stripe_len(ndevs - 1,
4686                                  btrfs_super_stripesize(info->super_copy));
4687                 data_stripes = num_stripes - 1;
4688         }
4689         if (type & BTRFS_BLOCK_GROUP_RAID6) {
4690                 raid_stripe_len = find_raid56_stripe_len(ndevs - 2,
4691                                  btrfs_super_stripesize(info->super_copy));
4692                 data_stripes = num_stripes - 2;
4693         }
4694
4695         /*
4696          * Use the number of data stripes to figure out how big this chunk
4697          * is really going to be in terms of logical address space,
4698          * and compare that answer with the max chunk size
4699          */
4700         if (stripe_size * data_stripes > max_chunk_size) {
4701                 u64 mask = (1ULL << 24) - 1;
4702
4703                 stripe_size = div_u64(max_chunk_size, data_stripes);
4704
4705                 /* bump the answer up to a 16MB boundary */
4706                 stripe_size = (stripe_size + mask) & ~mask;
4707
4708                 /* but don't go higher than the limits we found
4709                  * while searching for free extents
4710                  */
4711                 if (stripe_size > devices_info[ndevs-1].max_avail)
4712                         stripe_size = devices_info[ndevs-1].max_avail;
4713         }
4714
4715         stripe_size = div_u64(stripe_size, dev_stripes);
4716
4717         /* align to BTRFS_STRIPE_LEN */
4718         stripe_size = div_u64(stripe_size, raid_stripe_len);
4719         stripe_size *= raid_stripe_len;
4720
4721         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
4722         if (!map) {
4723                 ret = -ENOMEM;
4724                 goto error;
4725         }
4726         map->num_stripes = num_stripes;
4727
4728         for (i = 0; i < ndevs; ++i) {
4729                 for (j = 0; j < dev_stripes; ++j) {
4730                         int s = i * dev_stripes + j;
4731                         map->stripes[s].dev = devices_info[i].dev;
4732                         map->stripes[s].physical = devices_info[i].dev_offset +
4733                                                    j * stripe_size;
4734                 }
4735         }
4736         map->sector_size = extent_root->sectorsize;
4737         map->stripe_len = raid_stripe_len;
4738         map->io_align = raid_stripe_len;
4739         map->io_width = raid_stripe_len;
4740         map->type = type;
4741         map->sub_stripes = sub_stripes;
4742
4743         num_bytes = stripe_size * data_stripes;
4744
4745         trace_btrfs_chunk_alloc(info->chunk_root, map, start, num_bytes);
4746
4747         em = alloc_extent_map();
4748         if (!em) {
4749                 kfree(map);
4750                 ret = -ENOMEM;
4751                 goto error;
4752         }
4753         set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
4754         em->map_lookup = map;
4755         em->start = start;
4756         em->len = num_bytes;
4757         em->block_start = 0;
4758         em->block_len = em->len;
4759         em->orig_block_len = stripe_size;
4760
4761         em_tree = &extent_root->fs_info->mapping_tree.map_tree;
4762         write_lock(&em_tree->lock);
4763         ret = add_extent_mapping(em_tree, em, 0);
4764         if (!ret) {
4765                 list_add_tail(&em->list, &trans->transaction->pending_chunks);
4766                 atomic_inc(&em->refs);
4767         }
4768         write_unlock(&em_tree->lock);
4769         if (ret) {
4770                 free_extent_map(em);
4771                 goto error;
4772         }
4773
4774         ret = btrfs_make_block_group(trans, extent_root, 0, type,
4775                                      BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4776                                      start, num_bytes);
4777         if (ret)
4778                 goto error_del_extent;
4779
4780         for (i = 0; i < map->num_stripes; i++) {
4781                 num_bytes = map->stripes[i].dev->bytes_used + stripe_size;
4782                 btrfs_device_set_bytes_used(map->stripes[i].dev, num_bytes);
4783         }
4784
4785         spin_lock(&extent_root->fs_info->free_chunk_lock);
4786         extent_root->fs_info->free_chunk_space -= (stripe_size *
4787                                                    map->num_stripes);
4788         spin_unlock(&extent_root->fs_info->free_chunk_lock);
4789
4790         free_extent_map(em);
4791         check_raid56_incompat_flag(extent_root->fs_info, type);
4792
4793         kfree(devices_info);
4794         return 0;
4795
4796 error_del_extent:
4797         write_lock(&em_tree->lock);
4798         remove_extent_mapping(em_tree, em);
4799         write_unlock(&em_tree->lock);
4800
4801         /* One for our allocation */
4802         free_extent_map(em);
4803         /* One for the tree reference */
4804         free_extent_map(em);
4805         /* One for the pending_chunks list reference */
4806         free_extent_map(em);
4807 error:
4808         kfree(devices_info);
4809         return ret;
4810 }
4811
4812 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans,
4813                                 struct btrfs_root *extent_root,
4814                                 u64 chunk_offset, u64 chunk_size)
4815 {
4816         struct btrfs_key key;
4817         struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
4818         struct btrfs_device *device;
4819         struct btrfs_chunk *chunk;
4820         struct btrfs_stripe *stripe;
4821         struct extent_map_tree *em_tree;
4822         struct extent_map *em;
4823         struct map_lookup *map;
4824         size_t item_size;
4825         u64 dev_offset;
4826         u64 stripe_size;
4827         int i = 0;
4828         int ret = 0;
4829
4830         em_tree = &extent_root->fs_info->mapping_tree.map_tree;
4831         read_lock(&em_tree->lock);
4832         em = lookup_extent_mapping(em_tree, chunk_offset, chunk_size);
4833         read_unlock(&em_tree->lock);
4834
4835         if (!em) {
4836                 btrfs_crit(extent_root->fs_info, "unable to find logical "
4837                            "%Lu len %Lu", chunk_offset, chunk_size);
4838                 return -EINVAL;
4839         }
4840
4841         if (em->start != chunk_offset || em->len != chunk_size) {
4842                 btrfs_crit(extent_root->fs_info, "found a bad mapping, wanted"
4843                           " %Lu-%Lu, found %Lu-%Lu", chunk_offset,
4844                           chunk_size, em->start, em->len);
4845                 free_extent_map(em);
4846                 return -EINVAL;
4847         }
4848
4849         map = em->map_lookup;
4850         item_size = btrfs_chunk_item_size(map->num_stripes);
4851         stripe_size = em->orig_block_len;
4852
4853         chunk = kzalloc(item_size, GFP_NOFS);
4854         if (!chunk) {
4855                 ret = -ENOMEM;
4856                 goto out;
4857         }
4858
4859         /*
4860          * Take the device list mutex to prevent races with the final phase of
4861          * a device replace operation that replaces the device object associated
4862          * with the map's stripes, because the device object's id can change
4863          * at any time during that final phase of the device replace operation
4864          * (dev-replace.c:btrfs_dev_replace_finishing()).
4865          */
4866         mutex_lock(&chunk_root->fs_info->fs_devices->device_list_mutex);
4867         for (i = 0; i < map->num_stripes; i++) {
4868                 device = map->stripes[i].dev;
4869                 dev_offset = map->stripes[i].physical;
4870
4871                 ret = btrfs_update_device(trans, device);
4872                 if (ret)
4873                         break;
4874                 ret = btrfs_alloc_dev_extent(trans, device,
4875                                              chunk_root->root_key.objectid,
4876                                              BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4877                                              chunk_offset, dev_offset,
4878                                              stripe_size);
4879                 if (ret)
4880                         break;
4881         }
4882         if (ret) {
4883                 mutex_unlock(&chunk_root->fs_info->fs_devices->device_list_mutex);
4884                 goto out;
4885         }
4886
4887         stripe = &chunk->stripe;
4888         for (i = 0; i < map->num_stripes; i++) {
4889                 device = map->stripes[i].dev;
4890                 dev_offset = map->stripes[i].physical;
4891
4892                 btrfs_set_stack_stripe_devid(stripe, device->devid);
4893                 btrfs_set_stack_stripe_offset(stripe, dev_offset);
4894                 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
4895                 stripe++;
4896         }
4897         mutex_unlock(&chunk_root->fs_info->fs_devices->device_list_mutex);
4898
4899         btrfs_set_stack_chunk_length(chunk, chunk_size);
4900         btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
4901         btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
4902         btrfs_set_stack_chunk_type(chunk, map->type);
4903         btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
4904         btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
4905         btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
4906         btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
4907         btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
4908
4909         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
4910         key.type = BTRFS_CHUNK_ITEM_KEY;
4911         key.offset = chunk_offset;
4912
4913         ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
4914         if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
4915                 /*
4916                  * TODO: Cleanup of inserted chunk root in case of
4917                  * failure.
4918                  */
4919                 ret = btrfs_add_system_chunk(chunk_root, &key, chunk,
4920                                              item_size);
4921         }
4922
4923 out:
4924         kfree(chunk);
4925         free_extent_map(em);
4926         return ret;
4927 }
4928
4929 /*
4930  * Chunk allocation falls into two parts. The first part does works
4931  * that make the new allocated chunk useable, but not do any operation
4932  * that modifies the chunk tree. The second part does the works that
4933  * require modifying the chunk tree. This division is important for the
4934  * bootstrap process of adding storage to a seed btrfs.
4935  */
4936 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4937                       struct btrfs_root *extent_root, u64 type)
4938 {
4939         u64 chunk_offset;
4940
4941         ASSERT(mutex_is_locked(&extent_root->fs_info->chunk_mutex));
4942         chunk_offset = find_next_chunk(extent_root->fs_info);
4943         return __btrfs_alloc_chunk(trans, extent_root, chunk_offset, type);
4944 }
4945
4946 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
4947                                          struct btrfs_root *root,
4948                                          struct btrfs_device *device)
4949 {
4950         u64 chunk_offset;
4951         u64 sys_chunk_offset;
4952         u64 alloc_profile;
4953         struct btrfs_fs_info *fs_info = root->fs_info;
4954         struct btrfs_root *extent_root = fs_info->extent_root;
4955         int ret;
4956
4957         chunk_offset = find_next_chunk(fs_info);
4958         alloc_profile = btrfs_get_alloc_profile(extent_root, 0);
4959         ret = __btrfs_alloc_chunk(trans, extent_root, chunk_offset,
4960                                   alloc_profile);
4961         if (ret)
4962                 return ret;
4963
4964         sys_chunk_offset = find_next_chunk(root->fs_info);
4965         alloc_profile = btrfs_get_alloc_profile(fs_info->chunk_root, 0);
4966         ret = __btrfs_alloc_chunk(trans, extent_root, sys_chunk_offset,
4967                                   alloc_profile);
4968         return ret;
4969 }
4970
4971 static inline int btrfs_chunk_max_errors(struct map_lookup *map)
4972 {
4973         int max_errors;
4974
4975         if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
4976                          BTRFS_BLOCK_GROUP_RAID10 |
4977                          BTRFS_BLOCK_GROUP_RAID5 |
4978                          BTRFS_BLOCK_GROUP_DUP)) {
4979                 max_errors = 1;
4980         } else if (map->type & BTRFS_BLOCK_GROUP_RAID6) {
4981                 max_errors = 2;
4982         } else {
4983                 max_errors = 0;
4984         }
4985
4986         return max_errors;
4987 }
4988
4989 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
4990 {
4991         struct extent_map *em;
4992         struct map_lookup *map;
4993         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
4994         int readonly = 0;
4995         int miss_ndevs = 0;
4996         int i;
4997
4998         read_lock(&map_tree->map_tree.lock);
4999         em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
5000         read_unlock(&map_tree->map_tree.lock);
5001         if (!em)
5002                 return 1;
5003
5004         map = em->map_lookup;
5005         for (i = 0; i < map->num_stripes; i++) {
5006                 if (map->stripes[i].dev->missing) {
5007                         miss_ndevs++;
5008                         continue;
5009                 }
5010
5011                 if (!map->stripes[i].dev->writeable) {
5012                         readonly = 1;
5013                         goto end;
5014                 }
5015         }
5016
5017         /*
5018          * If the number of missing devices is larger than max errors,
5019          * we can not write the data into that chunk successfully, so
5020          * set it readonly.
5021          */
5022         if (miss_ndevs > btrfs_chunk_max_errors(map))
5023                 readonly = 1;
5024 end:
5025         free_extent_map(em);
5026         return readonly;
5027 }
5028
5029 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
5030 {
5031         extent_map_tree_init(&tree->map_tree);
5032 }
5033
5034 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
5035 {
5036         struct extent_map *em;
5037
5038         while (1) {
5039                 write_lock(&tree->map_tree.lock);
5040                 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
5041                 if (em)
5042                         remove_extent_mapping(&tree->map_tree, em);
5043                 write_unlock(&tree->map_tree.lock);
5044                 if (!em)
5045                         break;
5046                 /* once for us */
5047                 free_extent_map(em);
5048                 /* once for the tree */
5049                 free_extent_map(em);
5050         }
5051 }
5052
5053 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5054 {
5055         struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
5056         struct extent_map *em;
5057         struct map_lookup *map;
5058         struct extent_map_tree *em_tree = &map_tree->map_tree;
5059         int ret;
5060
5061         read_lock(&em_tree->lock);
5062         em = lookup_extent_mapping(em_tree, logical, len);
5063         read_unlock(&em_tree->lock);
5064
5065         /*
5066          * We could return errors for these cases, but that could get ugly and
5067          * we'd probably do the same thing which is just not do anything else
5068          * and exit, so return 1 so the callers don't try to use other copies.
5069          */
5070         if (!em) {
5071                 btrfs_crit(fs_info, "No mapping for %Lu-%Lu", logical,
5072                             logical+len);
5073                 return 1;
5074         }
5075
5076         if (em->start > logical || em->start + em->len < logical) {
5077                 btrfs_crit(fs_info, "Invalid mapping for %Lu-%Lu, got "
5078                             "%Lu-%Lu", logical, logical+len, em->start,
5079                             em->start + em->len);
5080                 free_extent_map(em);
5081                 return 1;
5082         }
5083
5084         map = em->map_lookup;
5085         if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
5086                 ret = map->num_stripes;
5087         else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5088                 ret = map->sub_stripes;
5089         else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
5090                 ret = 2;
5091         else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5092                 ret = 3;
5093         else
5094                 ret = 1;
5095         free_extent_map(em);
5096
5097         btrfs_dev_replace_lock(&fs_info->dev_replace, 0);
5098         if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace))
5099                 ret++;
5100         btrfs_dev_replace_unlock(&fs_info->dev_replace, 0);
5101
5102         return ret;
5103 }
5104
5105 unsigned long btrfs_full_stripe_len(struct btrfs_root *root,
5106                                     struct btrfs_mapping_tree *map_tree,
5107                                     u64 logical)
5108 {
5109         struct extent_map *em;
5110         struct map_lookup *map;
5111         struct extent_map_tree *em_tree = &map_tree->map_tree;
5112         unsigned long len = root->sectorsize;
5113
5114         read_lock(&em_tree->lock);
5115         em = lookup_extent_mapping(em_tree, logical, len);
5116         read_unlock(&em_tree->lock);
5117         BUG_ON(!em);
5118
5119         BUG_ON(em->start > logical || em->start + em->len < logical);
5120         map = em->map_lookup;
5121         if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5122                 len = map->stripe_len * nr_data_stripes(map);
5123         free_extent_map(em);
5124         return len;
5125 }
5126
5127 int btrfs_is_parity_mirror(struct btrfs_mapping_tree *map_tree,
5128                            u64 logical, u64 len, int mirror_num)
5129 {
5130         struct extent_map *em;
5131         struct map_lookup *map;
5132         struct extent_map_tree *em_tree = &map_tree->map_tree;
5133         int ret = 0;
5134
5135         read_lock(&em_tree->lock);
5136         em = lookup_extent_mapping(em_tree, logical, len);
5137         read_unlock(&em_tree->lock);
5138         BUG_ON(!em);
5139
5140         BUG_ON(em->start > logical || em->start + em->len < logical);
5141         map = em->map_lookup;
5142         if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5143                 ret = 1;
5144         free_extent_map(em);
5145         return ret;
5146 }
5147
5148 static int find_live_mirror(struct btrfs_fs_info *fs_info,
5149                             struct map_lookup *map, int first, int num,
5150                             int optimal, int dev_replace_is_ongoing)
5151 {
5152         int i;
5153         int tolerance;
5154         struct btrfs_device *srcdev;
5155
5156         if (dev_replace_is_ongoing &&
5157             fs_info->dev_replace.cont_reading_from_srcdev_mode ==
5158              BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
5159                 srcdev = fs_info->dev_replace.srcdev;
5160         else
5161                 srcdev = NULL;
5162
5163         /*
5164          * try to avoid the drive that is the source drive for a
5165          * dev-replace procedure, only choose it if no other non-missing
5166          * mirror is available
5167          */
5168         for (tolerance = 0; tolerance < 2; tolerance++) {
5169                 if (map->stripes[optimal].dev->bdev &&
5170                     (tolerance || map->stripes[optimal].dev != srcdev))
5171                         return optimal;
5172                 for (i = first; i < first + num; i++) {
5173                         if (map->stripes[i].dev->bdev &&
5174                             (tolerance || map->stripes[i].dev != srcdev))
5175                                 return i;
5176                 }
5177         }
5178
5179         /* we couldn't find one that doesn't fail.  Just return something
5180          * and the io error handling code will clean up eventually
5181          */
5182         return optimal;
5183 }
5184
5185 static inline int parity_smaller(u64 a, u64 b)
5186 {
5187         return a > b;
5188 }
5189
5190 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
5191 static void sort_parity_stripes(struct btrfs_bio *bbio, int num_stripes)
5192 {
5193         struct btrfs_bio_stripe s;
5194         int i;
5195         u64 l;
5196         int again = 1;
5197
5198         while (again) {
5199                 again = 0;
5200                 for (i = 0; i < num_stripes - 1; i++) {
5201                         if (parity_smaller(bbio->raid_map[i],
5202                                            bbio->raid_map[i+1])) {
5203                                 s = bbio->stripes[i];
5204                                 l = bbio->raid_map[i];
5205                                 bbio->stripes[i] = bbio->stripes[i+1];
5206                                 bbio->raid_map[i] = bbio->raid_map[i+1];
5207                                 bbio->stripes[i+1] = s;
5208                                 bbio->raid_map[i+1] = l;
5209
5210                                 again = 1;
5211                         }
5212                 }
5213         }
5214 }
5215
5216 static struct btrfs_bio *alloc_btrfs_bio(int total_stripes, int real_stripes)
5217 {
5218         struct btrfs_bio *bbio = kzalloc(
5219                  /* the size of the btrfs_bio */
5220                 sizeof(struct btrfs_bio) +
5221                 /* plus the variable array for the stripes */
5222                 sizeof(struct btrfs_bio_stripe) * (total_stripes) +
5223                 /* plus the variable array for the tgt dev */
5224                 sizeof(int) * (real_stripes) +
5225                 /*
5226                  * plus the raid_map, which includes both the tgt dev
5227                  * and the stripes
5228                  */
5229                 sizeof(u64) * (total_stripes),
5230                 GFP_NOFS|__GFP_NOFAIL);
5231
5232         atomic_set(&bbio->error, 0);
5233         atomic_set(&bbio->refs, 1);
5234
5235         return bbio;
5236 }
5237
5238 void btrfs_get_bbio(struct btrfs_bio *bbio)
5239 {
5240         WARN_ON(!atomic_read(&bbio->refs));
5241         atomic_inc(&bbio->refs);
5242 }
5243
5244 void btrfs_put_bbio(struct btrfs_bio *bbio)
5245 {
5246         if (!bbio)
5247                 return;
5248         if (atomic_dec_and_test(&bbio->refs))
5249                 kfree(bbio);
5250 }
5251
5252 static int __btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
5253                              u64 logical, u64 *length,
5254                              struct btrfs_bio **bbio_ret,
5255                              int mirror_num, int need_raid_map)
5256 {
5257         struct extent_map *em;
5258         struct map_lookup *map;
5259         struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
5260         struct extent_map_tree *em_tree = &map_tree->map_tree;
5261         u64 offset;
5262         u64 stripe_offset;
5263         u64 stripe_end_offset;
5264         u64 stripe_nr;
5265         u64 stripe_nr_orig;
5266         u64 stripe_nr_end;
5267         u64 stripe_len;
5268         u32 stripe_index;
5269         int i;
5270         int ret = 0;
5271         int num_stripes;
5272         int max_errors = 0;
5273         int tgtdev_indexes = 0;
5274         struct btrfs_bio *bbio = NULL;
5275         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
5276         int dev_replace_is_ongoing = 0;
5277         int num_alloc_stripes;
5278         int patch_the_first_stripe_for_dev_replace = 0;
5279         u64 physical_to_patch_in_first_stripe = 0;
5280         u64 raid56_full_stripe_start = (u64)-1;
5281
5282         read_lock(&em_tree->lock);
5283         em = lookup_extent_mapping(em_tree, logical, *length);
5284         read_unlock(&em_tree->lock);
5285
5286         if (!em) {
5287                 btrfs_crit(fs_info, "unable to find logical %llu len %llu",
5288                         logical, *length);
5289                 return -EINVAL;
5290         }
5291
5292         if (em->start > logical || em->start + em->len < logical) {
5293                 btrfs_crit(fs_info, "found a bad mapping, wanted %Lu, "
5294                            "found %Lu-%Lu", logical, em->start,
5295                            em->start + em->len);
5296                 free_extent_map(em);
5297                 return -EINVAL;
5298         }
5299
5300         map = em->map_lookup;
5301         offset = logical - em->start;
5302
5303         stripe_len = map->stripe_len;
5304         stripe_nr = offset;
5305         /*
5306          * stripe_nr counts the total number of stripes we have to stride
5307          * to get to this block
5308          */
5309         stripe_nr = div64_u64(stripe_nr, stripe_len);
5310
5311         stripe_offset = stripe_nr * stripe_len;
5312         if (offset < stripe_offset) {
5313                 btrfs_crit(fs_info, "stripe math has gone wrong, "
5314                            "stripe_offset=%llu, offset=%llu, start=%llu, "
5315                            "logical=%llu, stripe_len=%llu",
5316                            stripe_offset, offset, em->start, logical,
5317                            stripe_len);
5318                 free_extent_map(em);
5319                 return -EINVAL;
5320         }
5321
5322         /* stripe_offset is the offset of this block in its stripe*/
5323         stripe_offset = offset - stripe_offset;
5324
5325         /* if we're here for raid56, we need to know the stripe aligned start */
5326         if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5327                 unsigned long full_stripe_len = stripe_len * nr_data_stripes(map);
5328                 raid56_full_stripe_start = offset;
5329
5330                 /* allow a write of a full stripe, but make sure we don't
5331                  * allow straddling of stripes
5332                  */
5333                 raid56_full_stripe_start = div64_u64(raid56_full_stripe_start,
5334                                 full_stripe_len);
5335                 raid56_full_stripe_start *= full_stripe_len;
5336         }
5337
5338         if (rw & REQ_DISCARD) {
5339                 /* we don't discard raid56 yet */
5340                 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5341                         ret = -EOPNOTSUPP;
5342                         goto out;
5343                 }
5344                 *length = min_t(u64, em->len - offset, *length);
5345         } else if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
5346                 u64 max_len;
5347                 /* For writes to RAID[56], allow a full stripeset across all disks.
5348                    For other RAID types and for RAID[56] reads, just allow a single
5349                    stripe (on a single disk). */
5350                 if ((map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
5351                     (rw & REQ_WRITE)) {
5352                         max_len = stripe_len * nr_data_stripes(map) -
5353                                 (offset - raid56_full_stripe_start);
5354                 } else {
5355                         /* we limit the length of each bio to what fits in a stripe */
5356                         max_len = stripe_len - stripe_offset;
5357                 }
5358                 *length = min_t(u64, em->len - offset, max_len);
5359         } else {
5360                 *length = em->len - offset;
5361         }
5362
5363         /* This is for when we're called from btrfs_merge_bio_hook() and all
5364            it cares about is the length */
5365         if (!bbio_ret)
5366                 goto out;
5367
5368         btrfs_dev_replace_lock(dev_replace, 0);
5369         dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
5370         if (!dev_replace_is_ongoing)
5371                 btrfs_dev_replace_unlock(dev_replace, 0);
5372         else
5373                 btrfs_dev_replace_set_lock_blocking(dev_replace);
5374
5375         if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
5376             !(rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) &&
5377             dev_replace->tgtdev != NULL) {
5378                 /*
5379                  * in dev-replace case, for repair case (that's the only
5380                  * case where the mirror is selected explicitly when
5381                  * calling btrfs_map_block), blocks left of the left cursor
5382                  * can also be read from the target drive.
5383                  * For REQ_GET_READ_MIRRORS, the target drive is added as
5384                  * the last one to the array of stripes. For READ, it also
5385                  * needs to be supported using the same mirror number.
5386                  * If the requested block is not left of the left cursor,
5387                  * EIO is returned. This can happen because btrfs_num_copies()
5388                  * returns one more in the dev-replace case.
5389                  */
5390                 u64 tmp_length = *length;
5391                 struct btrfs_bio *tmp_bbio = NULL;
5392                 int tmp_num_stripes;
5393                 u64 srcdev_devid = dev_replace->srcdev->devid;
5394                 int index_srcdev = 0;
5395                 int found = 0;
5396                 u64 physical_of_found = 0;
5397
5398                 ret = __btrfs_map_block(fs_info, REQ_GET_READ_MIRRORS,
5399                              logical, &tmp_length, &tmp_bbio, 0, 0);
5400                 if (ret) {
5401                         WARN_ON(tmp_bbio != NULL);
5402                         goto out;
5403                 }
5404
5405                 tmp_num_stripes = tmp_bbio->num_stripes;
5406                 if (mirror_num > tmp_num_stripes) {
5407                         /*
5408                          * REQ_GET_READ_MIRRORS does not contain this
5409                          * mirror, that means that the requested area
5410                          * is not left of the left cursor
5411                          */
5412                         ret = -EIO;
5413                         btrfs_put_bbio(tmp_bbio);
5414                         goto out;
5415                 }
5416
5417                 /*
5418                  * process the rest of the function using the mirror_num
5419                  * of the source drive. Therefore look it up first.
5420                  * At the end, patch the device pointer to the one of the
5421                  * target drive.
5422                  */
5423                 for (i = 0; i < tmp_num_stripes; i++) {
5424                         if (tmp_bbio->stripes[i].dev->devid != srcdev_devid)
5425                                 continue;
5426
5427                         /*
5428                          * In case of DUP, in order to keep it simple, only add
5429                          * the mirror with the lowest physical address
5430                          */
5431                         if (found &&
5432                             physical_of_found <= tmp_bbio->stripes[i].physical)
5433                                 continue;
5434
5435                         index_srcdev = i;
5436                         found = 1;
5437                         physical_of_found = tmp_bbio->stripes[i].physical;
5438                 }
5439
5440                 btrfs_put_bbio(tmp_bbio);
5441
5442                 if (!found) {
5443                         WARN_ON(1);
5444                         ret = -EIO;
5445                         goto out;
5446                 }
5447
5448                 mirror_num = index_srcdev + 1;
5449                 patch_the_first_stripe_for_dev_replace = 1;
5450                 physical_to_patch_in_first_stripe = physical_of_found;
5451         } else if (mirror_num > map->num_stripes) {
5452                 mirror_num = 0;
5453         }
5454
5455         num_stripes = 1;
5456         stripe_index = 0;
5457         stripe_nr_orig = stripe_nr;
5458         stripe_nr_end = ALIGN(offset + *length, map->stripe_len);
5459         stripe_nr_end = div_u64(stripe_nr_end, map->stripe_len);
5460         stripe_end_offset = stripe_nr_end * map->stripe_len -
5461                             (offset + *length);
5462
5463         if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5464                 if (rw & REQ_DISCARD)
5465                         num_stripes = min_t(u64, map->num_stripes,
5466                                             stripe_nr_end - stripe_nr_orig);
5467                 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5468                                 &stripe_index);
5469                 if (!(rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)))
5470                         mirror_num = 1;
5471         } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
5472                 if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS))
5473                         num_stripes = map->num_stripes;
5474                 else if (mirror_num)
5475                         stripe_index = mirror_num - 1;
5476                 else {
5477                         stripe_index = find_live_mirror(fs_info, map, 0,
5478                                             map->num_stripes,
5479                                             current->pid % map->num_stripes,
5480                                             dev_replace_is_ongoing);
5481                         mirror_num = stripe_index + 1;
5482                 }
5483
5484         } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
5485                 if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) {
5486                         num_stripes = map->num_stripes;
5487                 } else if (mirror_num) {
5488                         stripe_index = mirror_num - 1;
5489                 } else {
5490                         mirror_num = 1;
5491                 }
5492
5493         } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5494                 u32 factor = map->num_stripes / map->sub_stripes;
5495
5496                 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
5497                 stripe_index *= map->sub_stripes;
5498
5499                 if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS))
5500                         num_stripes = map->sub_stripes;
5501                 else if (rw & REQ_DISCARD)
5502                         num_stripes = min_t(u64, map->sub_stripes *
5503                                             (stripe_nr_end - stripe_nr_orig),
5504                                             map->num_stripes);
5505                 else if (mirror_num)
5506                         stripe_index += mirror_num - 1;
5507                 else {
5508                         int old_stripe_index = stripe_index;
5509                         stripe_index = find_live_mirror(fs_info, map,
5510                                               stripe_index,
5511                                               map->sub_stripes, stripe_index +
5512                                               current->pid % map->sub_stripes,
5513                                               dev_replace_is_ongoing);
5514                         mirror_num = stripe_index - old_stripe_index + 1;
5515                 }
5516
5517         } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5518                 if (need_raid_map &&
5519                     ((rw & (REQ_WRITE | REQ_GET_READ_MIRRORS)) ||
5520                      mirror_num > 1)) {
5521                         /* push stripe_nr back to the start of the full stripe */
5522                         stripe_nr = div_u64(raid56_full_stripe_start,
5523                                         stripe_len * nr_data_stripes(map));
5524
5525                         /* RAID[56] write or recovery. Return all stripes */
5526                         num_stripes = map->num_stripes;
5527                         max_errors = nr_parity_stripes(map);
5528
5529                         *length = map->stripe_len;
5530                         stripe_index = 0;
5531                         stripe_offset = 0;
5532                 } else {
5533                         /*
5534                          * Mirror #0 or #1 means the original data block.
5535                          * Mirror #2 is RAID5 parity block.
5536                          * Mirror #3 is RAID6 Q block.
5537                          */
5538                         stripe_nr = div_u64_rem(stripe_nr,
5539                                         nr_data_stripes(map), &stripe_index);
5540                         if (mirror_num > 1)
5541                                 stripe_index = nr_data_stripes(map) +
5542                                                 mirror_num - 2;
5543
5544                         /* We distribute the parity blocks across stripes */
5545                         div_u64_rem(stripe_nr + stripe_index, map->num_stripes,
5546                                         &stripe_index);
5547                         if (!(rw & (REQ_WRITE | REQ_DISCARD |
5548                                     REQ_GET_READ_MIRRORS)) && mirror_num <= 1)
5549                                 mirror_num = 1;
5550                 }
5551         } else {
5552                 /*
5553                  * after this, stripe_nr is the number of stripes on this
5554                  * device we have to walk to find the data, and stripe_index is
5555                  * the number of our device in the stripe array
5556                  */
5557                 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5558                                 &stripe_index);
5559                 mirror_num = stripe_index + 1;
5560         }
5561         if (stripe_index >= map->num_stripes) {
5562                 btrfs_crit(fs_info, "stripe index math went horribly wrong, "
5563                            "got stripe_index=%u, num_stripes=%u",
5564                            stripe_index, map->num_stripes);
5565                 ret = -EINVAL;
5566                 goto out;
5567         }
5568
5569         num_alloc_stripes = num_stripes;
5570         if (dev_replace_is_ongoing) {
5571                 if (rw & (REQ_WRITE | REQ_DISCARD))
5572                         num_alloc_stripes <<= 1;
5573                 if (rw & REQ_GET_READ_MIRRORS)
5574                         num_alloc_stripes++;
5575                 tgtdev_indexes = num_stripes;
5576         }
5577
5578         bbio = alloc_btrfs_bio(num_alloc_stripes, tgtdev_indexes);
5579         if (!bbio) {
5580                 ret = -ENOMEM;
5581                 goto out;
5582         }
5583         if (dev_replace_is_ongoing)
5584                 bbio->tgtdev_map = (int *)(bbio->stripes + num_alloc_stripes);
5585
5586         /* build raid_map */
5587         if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK &&
5588             need_raid_map && ((rw & (REQ_WRITE | REQ_GET_READ_MIRRORS)) ||
5589             mirror_num > 1)) {
5590                 u64 tmp;
5591                 unsigned rot;
5592
5593                 bbio->raid_map = (u64 *)((void *)bbio->stripes +
5594                                  sizeof(struct btrfs_bio_stripe) *
5595                                  num_alloc_stripes +
5596                                  sizeof(int) * tgtdev_indexes);
5597
5598                 /* Work out the disk rotation on this stripe-set */
5599                 div_u64_rem(stripe_nr, num_stripes, &rot);
5600
5601                 /* Fill in the logical address of each stripe */
5602                 tmp = stripe_nr * nr_data_stripes(map);
5603                 for (i = 0; i < nr_data_stripes(map); i++)
5604                         bbio->raid_map[(i+rot) % num_stripes] =
5605                                 em->start + (tmp + i) * map->stripe_len;
5606
5607                 bbio->raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE;
5608                 if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5609                         bbio->raid_map[(i+rot+1) % num_stripes] =
5610                                 RAID6_Q_STRIPE;
5611         }
5612
5613         if (rw & REQ_DISCARD) {
5614                 u32 factor = 0;
5615                 u32 sub_stripes = 0;
5616                 u64 stripes_per_dev = 0;
5617                 u32 remaining_stripes = 0;
5618                 u32 last_stripe = 0;
5619
5620                 if (map->type &
5621                     (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID10)) {
5622                         if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5623                                 sub_stripes = 1;
5624                         else
5625                                 sub_stripes = map->sub_stripes;
5626
5627                         factor = map->num_stripes / sub_stripes;
5628                         stripes_per_dev = div_u64_rem(stripe_nr_end -
5629                                                       stripe_nr_orig,
5630                                                       factor,
5631                                                       &remaining_stripes);
5632                         div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
5633                         last_stripe *= sub_stripes;
5634                 }
5635
5636                 for (i = 0; i < num_stripes; i++) {
5637                         bbio->stripes[i].physical =
5638                                 map->stripes[stripe_index].physical +
5639                                 stripe_offset + stripe_nr * map->stripe_len;
5640                         bbio->stripes[i].dev = map->stripes[stripe_index].dev;
5641
5642                         if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5643                                          BTRFS_BLOCK_GROUP_RAID10)) {
5644                                 bbio->stripes[i].length = stripes_per_dev *
5645                                                           map->stripe_len;
5646
5647                                 if (i / sub_stripes < remaining_stripes)
5648                                         bbio->stripes[i].length +=
5649                                                 map->stripe_len;
5650
5651                                 /*
5652                                  * Special for the first stripe and
5653                                  * the last stripe:
5654                                  *
5655                                  * |-------|...|-------|
5656                                  *     |----------|
5657                                  *    off     end_off
5658                                  */
5659                                 if (i < sub_stripes)
5660                                         bbio->stripes[i].length -=
5661                                                 stripe_offset;
5662
5663                                 if (stripe_index >= last_stripe &&
5664                                     stripe_index <= (last_stripe +
5665                                                      sub_stripes - 1))
5666                                         bbio->stripes[i].length -=
5667                                                 stripe_end_offset;
5668
5669                                 if (i == sub_stripes - 1)
5670                                         stripe_offset = 0;
5671                         } else
5672                                 bbio->stripes[i].length = *length;
5673
5674                         stripe_index++;
5675                         if (stripe_index == map->num_stripes) {
5676                                 /* This could only happen for RAID0/10 */
5677                                 stripe_index = 0;
5678                                 stripe_nr++;
5679                         }
5680                 }
5681         } else {
5682                 for (i = 0; i < num_stripes; i++) {
5683                         bbio->stripes[i].physical =
5684                                 map->stripes[stripe_index].physical +
5685                                 stripe_offset +
5686                                 stripe_nr * map->stripe_len;
5687                         bbio->stripes[i].dev =
5688                                 map->stripes[stripe_index].dev;
5689                         stripe_index++;
5690                 }
5691         }
5692
5693         if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS))
5694                 max_errors = btrfs_chunk_max_errors(map);
5695
5696         if (bbio->raid_map)
5697                 sort_parity_stripes(bbio, num_stripes);
5698
5699         tgtdev_indexes = 0;
5700         if (dev_replace_is_ongoing && (rw & (REQ_WRITE | REQ_DISCARD)) &&
5701             dev_replace->tgtdev != NULL) {
5702                 int index_where_to_add;
5703                 u64 srcdev_devid = dev_replace->srcdev->devid;
5704
5705                 /*
5706                  * duplicate the write operations while the dev replace
5707                  * procedure is running. Since the copying of the old disk
5708                  * to the new disk takes place at run time while the
5709                  * filesystem is mounted writable, the regular write
5710                  * operations to the old disk have to be duplicated to go
5711                  * to the new disk as well.
5712                  * Note that device->missing is handled by the caller, and
5713                  * that the write to the old disk is already set up in the
5714                  * stripes array.
5715                  */
5716                 index_where_to_add = num_stripes;
5717                 for (i = 0; i < num_stripes; i++) {
5718                         if (bbio->stripes[i].dev->devid == srcdev_devid) {
5719                                 /* write to new disk, too */
5720                                 struct btrfs_bio_stripe *new =
5721                                         bbio->stripes + index_where_to_add;
5722                                 struct btrfs_bio_stripe *old =
5723                                         bbio->stripes + i;
5724
5725                                 new->physical = old->physical;
5726                                 new->length = old->length;
5727                                 new->dev = dev_replace->tgtdev;
5728                                 bbio->tgtdev_map[i] = index_where_to_add;
5729                                 index_where_to_add++;
5730                                 max_errors++;
5731                                 tgtdev_indexes++;
5732                         }
5733                 }
5734                 num_stripes = index_where_to_add;
5735         } else if (dev_replace_is_ongoing && (rw & REQ_GET_READ_MIRRORS) &&
5736                    dev_replace->tgtdev != NULL) {
5737                 u64 srcdev_devid = dev_replace->srcdev->devid;
5738                 int index_srcdev = 0;
5739                 int found = 0;
5740                 u64 physical_of_found = 0;
5741
5742                 /*
5743                  * During the dev-replace procedure, the target drive can
5744                  * also be used to read data in case it is needed to repair
5745                  * a corrupt block elsewhere. This is possible if the
5746                  * requested area is left of the left cursor. In this area,
5747                  * the target drive is a full copy of the source drive.
5748                  */
5749                 for (i = 0; i < num_stripes; i++) {
5750                         if (bbio->stripes[i].dev->devid == srcdev_devid) {
5751                                 /*
5752                                  * In case of DUP, in order to keep it
5753                                  * simple, only add the mirror with the
5754                                  * lowest physical address
5755                                  */
5756                                 if (found &&
5757                                     physical_of_found <=
5758                                      bbio->stripes[i].physical)
5759                                         continue;
5760                                 index_srcdev = i;
5761                                 found = 1;
5762                                 physical_of_found = bbio->stripes[i].physical;
5763                         }
5764                 }
5765                 if (found) {
5766                         if (physical_of_found + map->stripe_len <=
5767                             dev_replace->cursor_left) {
5768                                 struct btrfs_bio_stripe *tgtdev_stripe =
5769                                         bbio->stripes + num_stripes;
5770
5771                                 tgtdev_stripe->physical = physical_of_found;
5772                                 tgtdev_stripe->length =
5773                                         bbio->stripes[index_srcdev].length;
5774                                 tgtdev_stripe->dev = dev_replace->tgtdev;
5775                                 bbio->tgtdev_map[index_srcdev] = num_stripes;
5776
5777                                 tgtdev_indexes++;
5778                                 num_stripes++;
5779                         }
5780                 }
5781         }
5782
5783         *bbio_ret = bbio;
5784         bbio->map_type = map->type;
5785         bbio->num_stripes = num_stripes;
5786         bbio->max_errors = max_errors;
5787         bbio->mirror_num = mirror_num;
5788         bbio->num_tgtdevs = tgtdev_indexes;
5789
5790         /*
5791          * this is the case that REQ_READ && dev_replace_is_ongoing &&
5792          * mirror_num == num_stripes + 1 && dev_replace target drive is
5793          * available as a mirror
5794          */
5795         if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) {
5796                 WARN_ON(num_stripes > 1);
5797                 bbio->stripes[0].dev = dev_replace->tgtdev;
5798                 bbio->stripes[0].physical = physical_to_patch_in_first_stripe;
5799                 bbio->mirror_num = map->num_stripes + 1;
5800         }
5801 out:
5802         if (dev_replace_is_ongoing) {
5803                 btrfs_dev_replace_clear_lock_blocking(dev_replace);
5804                 btrfs_dev_replace_unlock(dev_replace, 0);
5805         }
5806         free_extent_map(em);
5807         return ret;
5808 }
5809
5810 int btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
5811                       u64 logical, u64 *length,
5812                       struct btrfs_bio **bbio_ret, int mirror_num)
5813 {
5814         return __btrfs_map_block(fs_info, rw, logical, length, bbio_ret,
5815                                  mirror_num, 0);
5816 }
5817
5818 /* For Scrub/replace */
5819 int btrfs_map_sblock(struct btrfs_fs_info *fs_info, int rw,
5820                      u64 logical, u64 *length,
5821                      struct btrfs_bio **bbio_ret, int mirror_num,
5822                      int need_raid_map)
5823 {
5824         return __btrfs_map_block(fs_info, rw, logical, length, bbio_ret,
5825                                  mirror_num, need_raid_map);
5826 }
5827
5828 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
5829                      u64 chunk_start, u64 physical, u64 devid,
5830                      u64 **logical, int *naddrs, int *stripe_len)
5831 {
5832         struct extent_map_tree *em_tree = &map_tree->map_tree;
5833         struct extent_map *em;
5834         struct map_lookup *map;
5835         u64 *buf;
5836         u64 bytenr;
5837         u64 length;
5838         u64 stripe_nr;
5839         u64 rmap_len;
5840         int i, j, nr = 0;
5841
5842         read_lock(&em_tree->lock);
5843         em = lookup_extent_mapping(em_tree, chunk_start, 1);
5844         read_unlock(&em_tree->lock);
5845
5846         if (!em) {
5847                 printk(KERN_ERR "BTRFS: couldn't find em for chunk %Lu\n",
5848                        chunk_start);
5849                 return -EIO;
5850         }
5851
5852         if (em->start != chunk_start) {
5853                 printk(KERN_ERR "BTRFS: bad chunk start, em=%Lu, wanted=%Lu\n",
5854                        em->start, chunk_start);
5855                 free_extent_map(em);
5856                 return -EIO;
5857         }
5858         map = em->map_lookup;
5859
5860         length = em->len;
5861         rmap_len = map->stripe_len;
5862
5863         if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5864                 length = div_u64(length, map->num_stripes / map->sub_stripes);
5865         else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5866                 length = div_u64(length, map->num_stripes);
5867         else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5868                 length = div_u64(length, nr_data_stripes(map));
5869                 rmap_len = map->stripe_len * nr_data_stripes(map);
5870         }
5871
5872         buf = kcalloc(map->num_stripes, sizeof(u64), GFP_NOFS);
5873         BUG_ON(!buf); /* -ENOMEM */
5874
5875         for (i = 0; i < map->num_stripes; i++) {
5876                 if (devid && map->stripes[i].dev->devid != devid)
5877                         continue;
5878                 if (map->stripes[i].physical > physical ||
5879                     map->stripes[i].physical + length <= physical)
5880                         continue;
5881
5882                 stripe_nr = physical - map->stripes[i].physical;
5883                 stripe_nr = div_u64(stripe_nr, map->stripe_len);
5884
5885                 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5886                         stripe_nr = stripe_nr * map->num_stripes + i;
5887                         stripe_nr = div_u64(stripe_nr, map->sub_stripes);
5888                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5889                         stripe_nr = stripe_nr * map->num_stripes + i;
5890                 } /* else if RAID[56], multiply by nr_data_stripes().
5891                    * Alternatively, just use rmap_len below instead of
5892                    * map->stripe_len */
5893
5894                 bytenr = chunk_start + stripe_nr * rmap_len;
5895                 WARN_ON(nr >= map->num_stripes);
5896                 for (j = 0; j < nr; j++) {
5897                         if (buf[j] == bytenr)
5898                                 break;
5899                 }
5900                 if (j == nr) {
5901                         WARN_ON(nr >= map->num_stripes);
5902                         buf[nr++] = bytenr;
5903                 }
5904         }
5905
5906         *logical = buf;
5907         *naddrs = nr;
5908         *stripe_len = rmap_len;
5909
5910         free_extent_map(em);
5911         return 0;
5912 }
5913
5914 static inline void btrfs_end_bbio(struct btrfs_bio *bbio, struct bio *bio)
5915 {
5916         bio->bi_private = bbio->private;
5917         bio->bi_end_io = bbio->end_io;
5918         bio_endio(bio);
5919
5920         btrfs_put_bbio(bbio);
5921 }
5922
5923 static void btrfs_end_bio(struct bio *bio)
5924 {
5925         struct btrfs_bio *bbio = bio->bi_private;
5926         int is_orig_bio = 0;
5927
5928         if (bio->bi_error) {
5929                 atomic_inc(&bbio->error);
5930                 if (bio->bi_error == -EIO || bio->bi_error == -EREMOTEIO) {
5931                         unsigned int stripe_index =
5932                                 btrfs_io_bio(bio)->stripe_index;
5933                         struct btrfs_device *dev;
5934
5935                         BUG_ON(stripe_index >= bbio->num_stripes);
5936                         dev = bbio->stripes[stripe_index].dev;
5937                         if (dev->bdev) {
5938                                 if (bio->bi_rw & WRITE)
5939                                         btrfs_dev_stat_inc(dev,
5940                                                 BTRFS_DEV_STAT_WRITE_ERRS);
5941                                 else
5942                                         btrfs_dev_stat_inc(dev,
5943                                                 BTRFS_DEV_STAT_READ_ERRS);
5944                                 if ((bio->bi_rw & WRITE_FLUSH) == WRITE_FLUSH)
5945                                         btrfs_dev_stat_inc(dev,
5946                                                 BTRFS_DEV_STAT_FLUSH_ERRS);
5947                                 btrfs_dev_stat_print_on_error(dev);
5948                         }
5949                 }
5950         }
5951
5952         if (bio == bbio->orig_bio)
5953                 is_orig_bio = 1;
5954
5955         btrfs_bio_counter_dec(bbio->fs_info);
5956
5957         if (atomic_dec_and_test(&bbio->stripes_pending)) {
5958                 if (!is_orig_bio) {
5959                         bio_put(bio);
5960                         bio = bbio->orig_bio;
5961                 }
5962
5963                 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
5964                 /* only send an error to the higher layers if it is
5965                  * beyond the tolerance of the btrfs bio
5966                  */
5967                 if (atomic_read(&bbio->error) > bbio->max_errors) {
5968                         bio->bi_error = -EIO;
5969                 } else {
5970                         /*
5971                          * this bio is actually up to date, we didn't
5972                          * go over the max number of errors
5973                          */
5974                         bio->bi_error = 0;
5975                 }
5976
5977                 btrfs_end_bbio(bbio, bio);
5978         } else if (!is_orig_bio) {
5979                 bio_put(bio);
5980         }
5981 }
5982
5983 /*
5984  * see run_scheduled_bios for a description of why bios are collected for
5985  * async submit.
5986  *
5987  * This will add one bio to the pending list for a device and make sure
5988  * the work struct is scheduled.
5989  */
5990 static noinline void btrfs_schedule_bio(struct btrfs_root *root,
5991                                         struct btrfs_device *device,
5992                                         int rw, struct bio *bio)
5993 {
5994         int should_queue = 1;
5995         struct btrfs_pending_bios *pending_bios;
5996
5997         if (device->missing || !device->bdev) {
5998                 bio_io_error(bio);
5999                 return;
6000         }
6001
6002         /* don't bother with additional async steps for reads, right now */
6003         if (!(rw & REQ_WRITE)) {
6004                 bio_get(bio);
6005                 btrfsic_submit_bio(rw, bio);
6006                 bio_put(bio);
6007                 return;
6008         }
6009
6010         /*
6011          * nr_async_bios allows us to reliably return congestion to the
6012          * higher layers.  Otherwise, the async bio makes it appear we have
6013          * made progress against dirty pages when we've really just put it
6014          * on a queue for later
6015          */
6016         atomic_inc(&root->fs_info->nr_async_bios);
6017         WARN_ON(bio->bi_next);
6018         bio->bi_next = NULL;
6019         bio->bi_rw |= rw;
6020
6021         spin_lock(&device->io_lock);
6022         if (bio->bi_rw & REQ_SYNC)
6023                 pending_bios = &device->pending_sync_bios;
6024         else
6025                 pending_bios = &device->pending_bios;
6026
6027         if (pending_bios->tail)
6028                 pending_bios->tail->bi_next = bio;
6029
6030         pending_bios->tail = bio;
6031         if (!pending_bios->head)
6032                 pending_bios->head = bio;
6033         if (device->running_pending)
6034                 should_queue = 0;
6035
6036         spin_unlock(&device->io_lock);
6037
6038         if (should_queue)
6039                 btrfs_queue_work(root->fs_info->submit_workers,
6040                                  &device->work);
6041 }
6042
6043 static void submit_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio,
6044                               struct bio *bio, u64 physical, int dev_nr,
6045                               int rw, int async)
6046 {
6047         struct btrfs_device *dev = bbio->stripes[dev_nr].dev;
6048
6049         bio->bi_private = bbio;
6050         btrfs_io_bio(bio)->stripe_index = dev_nr;
6051         bio->bi_end_io = btrfs_end_bio;
6052         bio->bi_iter.bi_sector = physical >> 9;
6053 #ifdef DEBUG
6054         {
6055                 struct rcu_string *name;
6056
6057                 rcu_read_lock();
6058                 name = rcu_dereference(dev->name);
6059                 pr_debug("btrfs_map_bio: rw %d, sector=%llu, dev=%lu "
6060                          "(%s id %llu), size=%u\n", rw,
6061                          (u64)bio->bi_iter.bi_sector, (u_long)dev->bdev->bd_dev,
6062                          name->str, dev->devid, bio->bi_iter.bi_size);
6063                 rcu_read_unlock();
6064         }
6065 #endif
6066         bio->bi_bdev = dev->bdev;
6067
6068         btrfs_bio_counter_inc_noblocked(root->fs_info);
6069
6070         if (async)
6071                 btrfs_schedule_bio(root, dev, rw, bio);
6072         else
6073                 btrfsic_submit_bio(rw, bio);
6074 }
6075
6076 static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical)
6077 {
6078         atomic_inc(&bbio->error);
6079         if (atomic_dec_and_test(&bbio->stripes_pending)) {
6080                 /* Should be the original bio. */
6081                 WARN_ON(bio != bbio->orig_bio);
6082
6083                 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
6084                 bio->bi_iter.bi_sector = logical >> 9;
6085                 bio->bi_error = -EIO;
6086                 btrfs_end_bbio(bbio, bio);
6087         }
6088 }
6089
6090 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
6091                   int mirror_num, int async_submit)
6092 {
6093         struct btrfs_device *dev;
6094         struct bio *first_bio = bio;
6095         u64 logical = (u64)bio->bi_iter.bi_sector << 9;
6096         u64 length = 0;
6097         u64 map_length;
6098         int ret;
6099         int dev_nr;
6100         int total_devs;
6101         struct btrfs_bio *bbio = NULL;
6102
6103         length = bio->bi_iter.bi_size;
6104         map_length = length;
6105
6106         btrfs_bio_counter_inc_blocked(root->fs_info);
6107         ret = __btrfs_map_block(root->fs_info, rw, logical, &map_length, &bbio,
6108                               mirror_num, 1);
6109         if (ret) {
6110                 btrfs_bio_counter_dec(root->fs_info);
6111                 return ret;
6112         }
6113
6114         total_devs = bbio->num_stripes;
6115         bbio->orig_bio = first_bio;
6116         bbio->private = first_bio->bi_private;
6117         bbio->end_io = first_bio->bi_end_io;
6118         bbio->fs_info = root->fs_info;
6119         atomic_set(&bbio->stripes_pending, bbio->num_stripes);
6120
6121         if ((bbio->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
6122             ((rw & WRITE) || (mirror_num > 1))) {
6123                 /* In this case, map_length has been set to the length of
6124                    a single stripe; not the whole write */
6125                 if (rw & WRITE) {
6126                         ret = raid56_parity_write(root, bio, bbio, map_length);
6127                 } else {
6128                         ret = raid56_parity_recover(root, bio, bbio, map_length,
6129                                                     mirror_num, 1);
6130                 }
6131
6132                 btrfs_bio_counter_dec(root->fs_info);
6133                 return ret;
6134         }
6135
6136         if (map_length < length) {
6137                 btrfs_crit(root->fs_info, "mapping failed logical %llu bio len %llu len %llu",
6138                         logical, length, map_length);
6139                 BUG();
6140         }
6141
6142         for (dev_nr = 0; dev_nr < total_devs; dev_nr++) {
6143                 dev = bbio->stripes[dev_nr].dev;
6144                 if (!dev || !dev->bdev || (rw & WRITE && !dev->writeable)) {
6145                         bbio_error(bbio, first_bio, logical);
6146                         continue;
6147                 }
6148
6149                 if (dev_nr < total_devs - 1) {
6150                         bio = btrfs_bio_clone(first_bio, GFP_NOFS);
6151                         BUG_ON(!bio); /* -ENOMEM */
6152                 } else
6153                         bio = first_bio;
6154
6155                 submit_stripe_bio(root, bbio, bio,
6156                                   bbio->stripes[dev_nr].physical, dev_nr, rw,
6157                                   async_submit);
6158         }
6159         btrfs_bio_counter_dec(root->fs_info);
6160         return 0;
6161 }
6162
6163 struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid,
6164                                        u8 *uuid, u8 *fsid)
6165 {
6166         struct btrfs_device *device;
6167         struct btrfs_fs_devices *cur_devices;
6168
6169         cur_devices = fs_info->fs_devices;
6170         while (cur_devices) {
6171                 if (!fsid ||
6172                     !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
6173                         device = __find_device(&cur_devices->devices,
6174                                                devid, uuid);
6175                         if (device)
6176                                 return device;
6177                 }
6178                 cur_devices = cur_devices->seed;
6179         }
6180         return NULL;
6181 }
6182
6183 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
6184                                             struct btrfs_fs_devices *fs_devices,
6185                                             u64 devid, u8 *dev_uuid)
6186 {
6187         struct btrfs_device *device;
6188
6189         device = btrfs_alloc_device(NULL, &devid, dev_uuid);
6190         if (IS_ERR(device))
6191                 return NULL;
6192
6193         list_add(&device->dev_list, &fs_devices->devices);
6194         device->fs_devices = fs_devices;
6195         fs_devices->num_devices++;
6196
6197         device->missing = 1;
6198         fs_devices->missing_devices++;
6199
6200         return device;
6201 }
6202
6203 /**
6204  * btrfs_alloc_device - allocate struct btrfs_device
6205  * @fs_info:    used only for generating a new devid, can be NULL if
6206  *              devid is provided (i.e. @devid != NULL).
6207  * @devid:      a pointer to devid for this device.  If NULL a new devid
6208  *              is generated.
6209  * @uuid:       a pointer to UUID for this device.  If NULL a new UUID
6210  *              is generated.
6211  *
6212  * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
6213  * on error.  Returned struct is not linked onto any lists and can be
6214  * destroyed with kfree() right away.
6215  */
6216 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
6217                                         const u64 *devid,
6218                                         const u8 *uuid)
6219 {
6220         struct btrfs_device *dev;
6221         u64 tmp;
6222
6223         if (WARN_ON(!devid && !fs_info))
6224                 return ERR_PTR(-EINVAL);
6225
6226         dev = __alloc_device();
6227         if (IS_ERR(dev))
6228                 return dev;
6229
6230         if (devid)
6231                 tmp = *devid;
6232         else {
6233                 int ret;
6234
6235                 ret = find_next_devid(fs_info, &tmp);
6236                 if (ret) {
6237                         kfree(dev);
6238                         return ERR_PTR(ret);
6239                 }
6240         }
6241         dev->devid = tmp;
6242
6243         if (uuid)
6244                 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE);
6245         else
6246                 generate_random_uuid(dev->uuid);
6247
6248         btrfs_init_work(&dev->work, btrfs_submit_helper,
6249                         pending_bios_fn, NULL, NULL);
6250
6251         return dev;
6252 }
6253
6254 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
6255                           struct extent_buffer *leaf,
6256                           struct btrfs_chunk *chunk)
6257 {
6258         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
6259         struct map_lookup *map;
6260         struct extent_map *em;
6261         u64 logical;
6262         u64 length;
6263         u64 stripe_len;
6264         u64 devid;
6265         u8 uuid[BTRFS_UUID_SIZE];
6266         int num_stripes;
6267         int ret;
6268         int i;
6269
6270         logical = key->offset;
6271         length = btrfs_chunk_length(leaf, chunk);
6272         stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6273         num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6274         /* Validation check */
6275         if (!num_stripes) {
6276                 btrfs_err(root->fs_info, "invalid chunk num_stripes: %u",
6277                           num_stripes);
6278                 return -EIO;
6279         }
6280         if (!IS_ALIGNED(logical, root->sectorsize)) {
6281                 btrfs_err(root->fs_info,
6282                           "invalid chunk logical %llu", logical);
6283                 return -EIO;
6284         }
6285         if (!length || !IS_ALIGNED(length, root->sectorsize)) {
6286                 btrfs_err(root->fs_info,
6287                         "invalid chunk length %llu", length);
6288                 return -EIO;
6289         }
6290         if (!is_power_of_2(stripe_len) || stripe_len != BTRFS_STRIPE_LEN) {
6291                 btrfs_err(root->fs_info, "invalid chunk stripe length: %llu",
6292                           stripe_len);
6293                 return -EIO;
6294         }
6295         if (~(BTRFS_BLOCK_GROUP_TYPE_MASK | BTRFS_BLOCK_GROUP_PROFILE_MASK) &
6296             btrfs_chunk_type(leaf, chunk)) {
6297                 btrfs_err(root->fs_info, "unrecognized chunk type: %llu",
6298                           ~(BTRFS_BLOCK_GROUP_TYPE_MASK |
6299                             BTRFS_BLOCK_GROUP_PROFILE_MASK) &
6300                           btrfs_chunk_type(leaf, chunk));
6301                 return -EIO;
6302         }
6303
6304         read_lock(&map_tree->map_tree.lock);
6305         em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
6306         read_unlock(&map_tree->map_tree.lock);
6307
6308         /* already mapped? */
6309         if (em && em->start <= logical && em->start + em->len > logical) {
6310                 free_extent_map(em);
6311                 return 0;
6312         } else if (em) {
6313                 free_extent_map(em);
6314         }
6315
6316         em = alloc_extent_map();
6317         if (!em)
6318                 return -ENOMEM;
6319         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
6320         if (!map) {
6321                 free_extent_map(em);
6322                 return -ENOMEM;
6323         }
6324
6325         set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
6326         em->map_lookup = map;
6327         em->start = logical;
6328         em->len = length;
6329         em->orig_start = 0;
6330         em->block_start = 0;
6331         em->block_len = em->len;
6332
6333         map->num_stripes = num_stripes;
6334         map->io_width = btrfs_chunk_io_width(leaf, chunk);
6335         map->io_align = btrfs_chunk_io_align(leaf, chunk);
6336         map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
6337         map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6338         map->type = btrfs_chunk_type(leaf, chunk);
6339         map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
6340         for (i = 0; i < num_stripes; i++) {
6341                 map->stripes[i].physical =
6342                         btrfs_stripe_offset_nr(leaf, chunk, i);
6343                 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
6344                 read_extent_buffer(leaf, uuid, (unsigned long)
6345                                    btrfs_stripe_dev_uuid_nr(chunk, i),
6346                                    BTRFS_UUID_SIZE);
6347                 map->stripes[i].dev = btrfs_find_device(root->fs_info, devid,
6348                                                         uuid, NULL);
6349                 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
6350                         free_extent_map(em);
6351                         return -EIO;
6352                 }
6353                 if (!map->stripes[i].dev) {
6354                         map->stripes[i].dev =
6355                                 add_missing_dev(root, root->fs_info->fs_devices,
6356                                                 devid, uuid);
6357                         if (!map->stripes[i].dev) {
6358                                 free_extent_map(em);
6359                                 return -EIO;
6360                         }
6361                         btrfs_warn(root->fs_info, "devid %llu uuid %pU is missing",
6362                                                 devid, uuid);
6363                 }
6364                 map->stripes[i].dev->in_fs_metadata = 1;
6365         }
6366
6367         write_lock(&map_tree->map_tree.lock);
6368         ret = add_extent_mapping(&map_tree->map_tree, em, 0);
6369         write_unlock(&map_tree->map_tree.lock);
6370         BUG_ON(ret); /* Tree corruption */
6371         free_extent_map(em);
6372
6373         return 0;
6374 }
6375
6376 static void fill_device_from_item(struct extent_buffer *leaf,
6377                                  struct btrfs_dev_item *dev_item,
6378                                  struct btrfs_device *device)
6379 {
6380         unsigned long ptr;
6381
6382         device->devid = btrfs_device_id(leaf, dev_item);
6383         device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
6384         device->total_bytes = device->disk_total_bytes;
6385         device->commit_total_bytes = device->disk_total_bytes;
6386         device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
6387         device->commit_bytes_used = device->bytes_used;
6388         device->type = btrfs_device_type(leaf, dev_item);
6389         device->io_align = btrfs_device_io_align(leaf, dev_item);
6390         device->io_width = btrfs_device_io_width(leaf, dev_item);
6391         device->sector_size = btrfs_device_sector_size(leaf, dev_item);
6392         WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
6393         device->is_tgtdev_for_dev_replace = 0;
6394
6395         ptr = btrfs_device_uuid(dev_item);
6396         read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
6397 }
6398
6399 static struct btrfs_fs_devices *open_seed_devices(struct btrfs_root *root,
6400                                                   u8 *fsid)
6401 {
6402         struct btrfs_fs_devices *fs_devices;
6403         int ret;
6404
6405         BUG_ON(!mutex_is_locked(&uuid_mutex));
6406
6407         fs_devices = root->fs_info->fs_devices->seed;
6408         while (fs_devices) {
6409                 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE))
6410                         return fs_devices;
6411
6412                 fs_devices = fs_devices->seed;
6413         }
6414
6415         fs_devices = find_fsid(fsid);
6416         if (!fs_devices) {
6417                 if (!btrfs_test_opt(root, DEGRADED))
6418                         return ERR_PTR(-ENOENT);
6419
6420                 fs_devices = alloc_fs_devices(fsid);
6421                 if (IS_ERR(fs_devices))
6422                         return fs_devices;
6423
6424                 fs_devices->seeding = 1;
6425                 fs_devices->opened = 1;
6426                 return fs_devices;
6427         }
6428
6429         fs_devices = clone_fs_devices(fs_devices);
6430         if (IS_ERR(fs_devices))
6431                 return fs_devices;
6432
6433         ret = __btrfs_open_devices(fs_devices, FMODE_READ,
6434                                    root->fs_info->bdev_holder);
6435         if (ret) {
6436                 free_fs_devices(fs_devices);
6437                 fs_devices = ERR_PTR(ret);
6438                 goto out;
6439         }
6440
6441         if (!fs_devices->seeding) {
6442                 __btrfs_close_devices(fs_devices);
6443                 free_fs_devices(fs_devices);
6444                 fs_devices = ERR_PTR(-EINVAL);
6445                 goto out;
6446         }
6447
6448         fs_devices->seed = root->fs_info->fs_devices->seed;
6449         root->fs_info->fs_devices->seed = fs_devices;
6450 out:
6451         return fs_devices;
6452 }
6453
6454 static int read_one_dev(struct btrfs_root *root,
6455                         struct extent_buffer *leaf,
6456                         struct btrfs_dev_item *dev_item)
6457 {
6458         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
6459         struct btrfs_device *device;
6460         u64 devid;
6461         int ret;
6462         u8 fs_uuid[BTRFS_UUID_SIZE];
6463         u8 dev_uuid[BTRFS_UUID_SIZE];
6464
6465         devid = btrfs_device_id(leaf, dev_item);
6466         read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
6467                            BTRFS_UUID_SIZE);
6468         read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
6469                            BTRFS_UUID_SIZE);
6470
6471         if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
6472                 fs_devices = open_seed_devices(root, fs_uuid);
6473                 if (IS_ERR(fs_devices))
6474                         return PTR_ERR(fs_devices);
6475         }
6476
6477         device = btrfs_find_device(root->fs_info, devid, dev_uuid, fs_uuid);
6478         if (!device) {
6479                 if (!btrfs_test_opt(root, DEGRADED))
6480                         return -EIO;
6481
6482                 device = add_missing_dev(root, fs_devices, devid, dev_uuid);
6483                 if (!device)
6484                         return -ENOMEM;
6485                 btrfs_warn(root->fs_info, "devid %llu uuid %pU missing",
6486                                 devid, dev_uuid);
6487         } else {
6488                 if (!device->bdev && !btrfs_test_opt(root, DEGRADED))
6489                         return -EIO;
6490
6491                 if(!device->bdev && !device->missing) {
6492                         /*
6493                          * this happens when a device that was properly setup
6494                          * in the device info lists suddenly goes bad.
6495                          * device->bdev is NULL, and so we have to set
6496                          * device->missing to one here
6497                          */
6498                         device->fs_devices->missing_devices++;
6499                         device->missing = 1;
6500                 }
6501
6502                 /* Move the device to its own fs_devices */
6503                 if (device->fs_devices != fs_devices) {
6504                         ASSERT(device->missing);
6505
6506                         list_move(&device->dev_list, &fs_devices->devices);
6507                         device->fs_devices->num_devices--;
6508                         fs_devices->num_devices++;
6509
6510                         device->fs_devices->missing_devices--;
6511                         fs_devices->missing_devices++;
6512
6513                         device->fs_devices = fs_devices;
6514                 }
6515         }
6516
6517         if (device->fs_devices != root->fs_info->fs_devices) {
6518                 BUG_ON(device->writeable);
6519                 if (device->generation !=
6520                     btrfs_device_generation(leaf, dev_item))
6521                         return -EINVAL;
6522         }
6523
6524         fill_device_from_item(leaf, dev_item, device);
6525         device->in_fs_metadata = 1;
6526         if (device->writeable && !device->is_tgtdev_for_dev_replace) {
6527                 device->fs_devices->total_rw_bytes += device->total_bytes;
6528                 spin_lock(&root->fs_info->free_chunk_lock);
6529                 root->fs_info->free_chunk_space += device->total_bytes -
6530                         device->bytes_used;
6531                 spin_unlock(&root->fs_info->free_chunk_lock);
6532         }
6533         ret = 0;
6534         return ret;
6535 }
6536
6537 int btrfs_read_sys_array(struct btrfs_root *root)
6538 {
6539         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
6540         struct extent_buffer *sb;
6541         struct btrfs_disk_key *disk_key;
6542         struct btrfs_chunk *chunk;
6543         u8 *array_ptr;
6544         unsigned long sb_array_offset;
6545         int ret = 0;
6546         u32 num_stripes;
6547         u32 array_size;
6548         u32 len = 0;
6549         u32 cur_offset;
6550         struct btrfs_key key;
6551
6552         ASSERT(BTRFS_SUPER_INFO_SIZE <= root->nodesize);
6553         /*
6554          * This will create extent buffer of nodesize, superblock size is
6555          * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will
6556          * overallocate but we can keep it as-is, only the first page is used.
6557          */
6558         sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET);
6559         if (!sb)
6560                 return -ENOMEM;
6561         set_extent_buffer_uptodate(sb);
6562         btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
6563         /*
6564          * The sb extent buffer is artificial and just used to read the system array.
6565          * set_extent_buffer_uptodate() call does not properly mark all it's
6566          * pages up-to-date when the page is larger: extent does not cover the
6567          * whole page and consequently check_page_uptodate does not find all
6568          * the page's extents up-to-date (the hole beyond sb),
6569          * write_extent_buffer then triggers a WARN_ON.
6570          *
6571          * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
6572          * but sb spans only this function. Add an explicit SetPageUptodate call
6573          * to silence the warning eg. on PowerPC 64.
6574          */
6575         if (PAGE_SIZE > BTRFS_SUPER_INFO_SIZE)
6576                 SetPageUptodate(sb->pages[0]);
6577
6578         write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
6579         array_size = btrfs_super_sys_array_size(super_copy);
6580
6581         array_ptr = super_copy->sys_chunk_array;
6582         sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array);
6583         cur_offset = 0;
6584
6585         while (cur_offset < array_size) {
6586                 disk_key = (struct btrfs_disk_key *)array_ptr;
6587                 len = sizeof(*disk_key);
6588                 if (cur_offset + len > array_size)
6589                         goto out_short_read;
6590
6591                 btrfs_disk_key_to_cpu(&key, disk_key);
6592
6593                 array_ptr += len;
6594                 sb_array_offset += len;
6595                 cur_offset += len;
6596
6597                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
6598                         chunk = (struct btrfs_chunk *)sb_array_offset;
6599                         /*
6600                          * At least one btrfs_chunk with one stripe must be
6601                          * present, exact stripe count check comes afterwards
6602                          */
6603                         len = btrfs_chunk_item_size(1);
6604                         if (cur_offset + len > array_size)
6605                                 goto out_short_read;
6606
6607                         num_stripes = btrfs_chunk_num_stripes(sb, chunk);
6608                         if (!num_stripes) {
6609                                 printk(KERN_ERR
6610             "BTRFS: invalid number of stripes %u in sys_array at offset %u\n",
6611                                         num_stripes, cur_offset);
6612                                 ret = -EIO;
6613                                 break;
6614                         }
6615
6616                         len = btrfs_chunk_item_size(num_stripes);
6617                         if (cur_offset + len > array_size)
6618                                 goto out_short_read;
6619
6620                         ret = read_one_chunk(root, &key, sb, chunk);
6621                         if (ret)
6622                                 break;
6623                 } else {
6624                         printk(KERN_ERR
6625                 "BTRFS: unexpected item type %u in sys_array at offset %u\n",
6626                                 (u32)key.type, cur_offset);
6627                         ret = -EIO;
6628                         break;
6629                 }
6630                 array_ptr += len;
6631                 sb_array_offset += len;
6632                 cur_offset += len;
6633         }
6634         clear_extent_buffer_uptodate(sb);
6635         free_extent_buffer_stale(sb);
6636         return ret;
6637
6638 out_short_read:
6639         printk(KERN_ERR "BTRFS: sys_array too short to read %u bytes at offset %u\n",
6640                         len, cur_offset);
6641         clear_extent_buffer_uptodate(sb);
6642         free_extent_buffer_stale(sb);
6643         return -EIO;
6644 }
6645
6646 int btrfs_read_chunk_tree(struct btrfs_root *root)
6647 {
6648         struct btrfs_path *path;
6649         struct extent_buffer *leaf;
6650         struct btrfs_key key;
6651         struct btrfs_key found_key;
6652         int ret;
6653         int slot;
6654         u64 total_dev = 0;
6655
6656         root = root->fs_info->chunk_root;
6657
6658         path = btrfs_alloc_path();
6659         if (!path)
6660                 return -ENOMEM;
6661
6662         mutex_lock(&uuid_mutex);
6663         lock_chunks(root);
6664
6665         /*
6666          * Read all device items, and then all the chunk items. All
6667          * device items are found before any chunk item (their object id
6668          * is smaller than the lowest possible object id for a chunk
6669          * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
6670          */
6671         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
6672         key.offset = 0;
6673         key.type = 0;
6674         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6675         if (ret < 0)
6676                 goto error;
6677         while (1) {
6678                 leaf = path->nodes[0];
6679                 slot = path->slots[0];
6680                 if (slot >= btrfs_header_nritems(leaf)) {
6681                         ret = btrfs_next_leaf(root, path);
6682                         if (ret == 0)
6683                                 continue;
6684                         if (ret < 0)
6685                                 goto error;
6686                         break;
6687                 }
6688                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6689                 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
6690                         struct btrfs_dev_item *dev_item;
6691                         dev_item = btrfs_item_ptr(leaf, slot,
6692                                                   struct btrfs_dev_item);
6693                         ret = read_one_dev(root, leaf, dev_item);
6694                         if (ret)
6695                                 goto error;
6696                         total_dev++;
6697                 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
6698                         struct btrfs_chunk *chunk;
6699                         chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
6700                         ret = read_one_chunk(root, &found_key, leaf, chunk);
6701                         if (ret)
6702                                 goto error;
6703                 }
6704                 path->slots[0]++;
6705         }
6706
6707         /*
6708          * After loading chunk tree, we've got all device information,
6709          * do another round of validation checks.
6710          */
6711         if (total_dev != root->fs_info->fs_devices->total_devices) {
6712                 btrfs_err(root->fs_info,
6713            "super_num_devices %llu mismatch with num_devices %llu found here",
6714                           btrfs_super_num_devices(root->fs_info->super_copy),
6715                           total_dev);
6716                 ret = -EINVAL;
6717                 goto error;
6718         }
6719         if (btrfs_super_total_bytes(root->fs_info->super_copy) <
6720             root->fs_info->fs_devices->total_rw_bytes) {
6721                 btrfs_err(root->fs_info,
6722         "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu",
6723                           btrfs_super_total_bytes(root->fs_info->super_copy),
6724                           root->fs_info->fs_devices->total_rw_bytes);
6725                 ret = -EINVAL;
6726                 goto error;
6727         }
6728         ret = 0;
6729 error:
6730         unlock_chunks(root);
6731         mutex_unlock(&uuid_mutex);
6732
6733         btrfs_free_path(path);
6734         return ret;
6735 }
6736
6737 void btrfs_init_devices_late(struct btrfs_fs_info *fs_info)
6738 {
6739         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6740         struct btrfs_device *device;
6741
6742         while (fs_devices) {
6743                 mutex_lock(&fs_devices->device_list_mutex);
6744                 list_for_each_entry(device, &fs_devices->devices, dev_list)
6745                         device->dev_root = fs_info->dev_root;
6746                 mutex_unlock(&fs_devices->device_list_mutex);
6747
6748                 fs_devices = fs_devices->seed;
6749         }
6750 }
6751
6752 static void __btrfs_reset_dev_stats(struct btrfs_device *dev)
6753 {
6754         int i;
6755
6756         for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6757                 btrfs_dev_stat_reset(dev, i);
6758 }
6759
6760 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
6761 {
6762         struct btrfs_key key;
6763         struct btrfs_key found_key;
6764         struct btrfs_root *dev_root = fs_info->dev_root;
6765         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6766         struct extent_buffer *eb;
6767         int slot;
6768         int ret = 0;
6769         struct btrfs_device *device;
6770         struct btrfs_path *path = NULL;
6771         int i;
6772
6773         path = btrfs_alloc_path();
6774         if (!path) {
6775                 ret = -ENOMEM;
6776                 goto out;
6777         }
6778
6779         mutex_lock(&fs_devices->device_list_mutex);
6780         list_for_each_entry(device, &fs_devices->devices, dev_list) {
6781                 int item_size;
6782                 struct btrfs_dev_stats_item *ptr;
6783
6784                 key.objectid = BTRFS_DEV_STATS_OBJECTID;
6785                 key.type = BTRFS_PERSISTENT_ITEM_KEY;
6786                 key.offset = device->devid;
6787                 ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
6788                 if (ret) {
6789                         __btrfs_reset_dev_stats(device);
6790                         device->dev_stats_valid = 1;
6791                         btrfs_release_path(path);
6792                         continue;
6793                 }
6794                 slot = path->slots[0];
6795                 eb = path->nodes[0];
6796                 btrfs_item_key_to_cpu(eb, &found_key, slot);
6797                 item_size = btrfs_item_size_nr(eb, slot);
6798
6799                 ptr = btrfs_item_ptr(eb, slot,
6800                                      struct btrfs_dev_stats_item);
6801
6802                 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
6803                         if (item_size >= (1 + i) * sizeof(__le64))
6804                                 btrfs_dev_stat_set(device, i,
6805                                         btrfs_dev_stats_value(eb, ptr, i));
6806                         else
6807                                 btrfs_dev_stat_reset(device, i);
6808                 }
6809
6810                 device->dev_stats_valid = 1;
6811                 btrfs_dev_stat_print_on_load(device);
6812                 btrfs_release_path(path);
6813         }
6814         mutex_unlock(&fs_devices->device_list_mutex);
6815
6816 out:
6817         btrfs_free_path(path);
6818         return ret < 0 ? ret : 0;
6819 }
6820
6821 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
6822                                 struct btrfs_root *dev_root,
6823                                 struct btrfs_device *device)
6824 {
6825         struct btrfs_path *path;
6826         struct btrfs_key key;
6827         struct extent_buffer *eb;
6828         struct btrfs_dev_stats_item *ptr;
6829         int ret;
6830         int i;
6831
6832         key.objectid = BTRFS_DEV_STATS_OBJECTID;
6833         key.type = BTRFS_PERSISTENT_ITEM_KEY;
6834         key.offset = device->devid;
6835
6836         path = btrfs_alloc_path();
6837         BUG_ON(!path);
6838         ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
6839         if (ret < 0) {
6840                 btrfs_warn_in_rcu(dev_root->fs_info,
6841                         "error %d while searching for dev_stats item for device %s",
6842                               ret, rcu_str_deref(device->name));
6843                 goto out;
6844         }
6845
6846         if (ret == 0 &&
6847             btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
6848                 /* need to delete old one and insert a new one */
6849                 ret = btrfs_del_item(trans, dev_root, path);
6850                 if (ret != 0) {
6851                         btrfs_warn_in_rcu(dev_root->fs_info,
6852                                 "delete too small dev_stats item for device %s failed %d",
6853                                       rcu_str_deref(device->name), ret);
6854                         goto out;
6855                 }
6856                 ret = 1;
6857         }
6858
6859         if (ret == 1) {
6860                 /* need to insert a new item */
6861                 btrfs_release_path(path);
6862                 ret = btrfs_insert_empty_item(trans, dev_root, path,
6863                                               &key, sizeof(*ptr));
6864                 if (ret < 0) {
6865                         btrfs_warn_in_rcu(dev_root->fs_info,
6866                                 "insert dev_stats item for device %s failed %d",
6867                                 rcu_str_deref(device->name), ret);
6868                         goto out;
6869                 }
6870         }
6871
6872         eb = path->nodes[0];
6873         ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
6874         for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6875                 btrfs_set_dev_stats_value(eb, ptr, i,
6876                                           btrfs_dev_stat_read(device, i));
6877         btrfs_mark_buffer_dirty(eb);
6878
6879 out:
6880         btrfs_free_path(path);
6881         return ret;
6882 }
6883
6884 /*
6885  * called from commit_transaction. Writes all changed device stats to disk.
6886  */
6887 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans,
6888                         struct btrfs_fs_info *fs_info)
6889 {
6890         struct btrfs_root *dev_root = fs_info->dev_root;
6891         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6892         struct btrfs_device *device;
6893         int stats_cnt;
6894         int ret = 0;
6895
6896         mutex_lock(&fs_devices->device_list_mutex);
6897         list_for_each_entry(device, &fs_devices->devices, dev_list) {
6898                 if (!device->dev_stats_valid || !btrfs_dev_stats_dirty(device))
6899                         continue;
6900
6901                 stats_cnt = atomic_read(&device->dev_stats_ccnt);
6902                 ret = update_dev_stat_item(trans, dev_root, device);
6903                 if (!ret)
6904                         atomic_sub(stats_cnt, &device->dev_stats_ccnt);
6905         }
6906         mutex_unlock(&fs_devices->device_list_mutex);
6907
6908         return ret;
6909 }
6910
6911 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
6912 {
6913         btrfs_dev_stat_inc(dev, index);
6914         btrfs_dev_stat_print_on_error(dev);
6915 }
6916
6917 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
6918 {
6919         if (!dev->dev_stats_valid)
6920                 return;
6921         btrfs_err_rl_in_rcu(dev->dev_root->fs_info,
6922                 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
6923                            rcu_str_deref(dev->name),
6924                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
6925                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
6926                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
6927                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
6928                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
6929 }
6930
6931 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
6932 {
6933         int i;
6934
6935         for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6936                 if (btrfs_dev_stat_read(dev, i) != 0)
6937                         break;
6938         if (i == BTRFS_DEV_STAT_VALUES_MAX)
6939                 return; /* all values == 0, suppress message */
6940
6941         btrfs_info_in_rcu(dev->dev_root->fs_info,
6942                 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
6943                rcu_str_deref(dev->name),
6944                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
6945                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
6946                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
6947                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
6948                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
6949 }
6950
6951 int btrfs_get_dev_stats(struct btrfs_root *root,
6952                         struct btrfs_ioctl_get_dev_stats *stats)
6953 {
6954         struct btrfs_device *dev;
6955         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
6956         int i;
6957
6958         mutex_lock(&fs_devices->device_list_mutex);
6959         dev = btrfs_find_device(root->fs_info, stats->devid, NULL, NULL);
6960         mutex_unlock(&fs_devices->device_list_mutex);
6961
6962         if (!dev) {
6963                 btrfs_warn(root->fs_info, "get dev_stats failed, device not found");
6964                 return -ENODEV;
6965         } else if (!dev->dev_stats_valid) {
6966                 btrfs_warn(root->fs_info, "get dev_stats failed, not yet valid");
6967                 return -ENODEV;
6968         } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
6969                 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
6970                         if (stats->nr_items > i)
6971                                 stats->values[i] =
6972                                         btrfs_dev_stat_read_and_reset(dev, i);
6973                         else
6974                                 btrfs_dev_stat_reset(dev, i);
6975                 }
6976         } else {
6977                 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6978                         if (stats->nr_items > i)
6979                                 stats->values[i] = btrfs_dev_stat_read(dev, i);
6980         }
6981         if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
6982                 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
6983         return 0;
6984 }
6985
6986 void btrfs_scratch_superblocks(struct block_device *bdev, char *device_path)
6987 {
6988         struct buffer_head *bh;
6989         struct btrfs_super_block *disk_super;
6990         int copy_num;
6991
6992         if (!bdev)
6993                 return;
6994
6995         for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX;
6996                 copy_num++) {
6997
6998                 if (btrfs_read_dev_one_super(bdev, copy_num, &bh))
6999                         continue;
7000
7001                 disk_super = (struct btrfs_super_block *)bh->b_data;
7002
7003                 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
7004                 set_buffer_dirty(bh);
7005                 sync_dirty_buffer(bh);
7006                 brelse(bh);
7007         }
7008
7009         /* Notify udev that device has changed */
7010         btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
7011
7012         /* Update ctime/mtime for device path for libblkid */
7013         update_dev_time(device_path);
7014 }
7015
7016 /*
7017  * Update the size of all devices, which is used for writing out the
7018  * super blocks.
7019  */
7020 void btrfs_update_commit_device_size(struct btrfs_fs_info *fs_info)
7021 {
7022         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7023         struct btrfs_device *curr, *next;
7024
7025         if (list_empty(&fs_devices->resized_devices))
7026                 return;
7027
7028         mutex_lock(&fs_devices->device_list_mutex);
7029         lock_chunks(fs_info->dev_root);
7030         list_for_each_entry_safe(curr, next, &fs_devices->resized_devices,
7031                                  resized_list) {
7032                 list_del_init(&curr->resized_list);
7033                 curr->commit_total_bytes = curr->disk_total_bytes;
7034         }
7035         unlock_chunks(fs_info->dev_root);
7036         mutex_unlock(&fs_devices->device_list_mutex);
7037 }
7038
7039 /* Must be invoked during the transaction commit */
7040 void btrfs_update_commit_device_bytes_used(struct btrfs_root *root,
7041                                         struct btrfs_transaction *transaction)
7042 {
7043         struct extent_map *em;
7044         struct map_lookup *map;
7045         struct btrfs_device *dev;
7046         int i;
7047
7048         if (list_empty(&transaction->pending_chunks))
7049                 return;
7050
7051         /* In order to kick the device replace finish process */
7052         lock_chunks(root);
7053         list_for_each_entry(em, &transaction->pending_chunks, list) {
7054                 map = em->map_lookup;
7055
7056                 for (i = 0; i < map->num_stripes; i++) {
7057                         dev = map->stripes[i].dev;
7058                         dev->commit_bytes_used = dev->bytes_used;
7059                 }
7060         }
7061         unlock_chunks(root);
7062 }
7063
7064 void btrfs_set_fs_info_ptr(struct btrfs_fs_info *fs_info)
7065 {
7066         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7067         while (fs_devices) {
7068                 fs_devices->fs_info = fs_info;
7069                 fs_devices = fs_devices->seed;
7070         }
7071 }
7072
7073 void btrfs_reset_fs_info_ptr(struct btrfs_fs_info *fs_info)
7074 {
7075         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7076         while (fs_devices) {
7077                 fs_devices->fs_info = NULL;
7078                 fs_devices = fs_devices->seed;
7079         }
7080 }
7081
7082 static void btrfs_close_one_device(struct btrfs_device *device)
7083 {
7084         struct btrfs_fs_devices *fs_devices = device->fs_devices;
7085         struct btrfs_device *new_device;
7086         struct rcu_string *name;
7087
7088         if (device->bdev)
7089                 fs_devices->open_devices--;
7090
7091         if (device->writeable &&
7092             device->devid != BTRFS_DEV_REPLACE_DEVID) {
7093                 list_del_init(&device->dev_alloc_list);
7094                 fs_devices->rw_devices--;
7095         }
7096
7097         if (device->missing)
7098                 fs_devices->missing_devices--;
7099
7100         new_device = btrfs_alloc_device(NULL, &device->devid,
7101                                         device->uuid);
7102         BUG_ON(IS_ERR(new_device)); /* -ENOMEM */
7103
7104         /* Safe because we are under uuid_mutex */
7105         if (device->name) {
7106                 name = rcu_string_strdup(device->name->str, GFP_NOFS);
7107                 BUG_ON(!name); /* -ENOMEM */
7108                 rcu_assign_pointer(new_device->name, name);
7109         }
7110
7111         list_replace_rcu(&device->dev_list, &new_device->dev_list);
7112         new_device->fs_devices = device->fs_devices;
7113
7114         call_rcu(&device->rcu, free_device);
7115 }