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/iocontext.h>
  24 #include <linux/capability.h>
  25 #include <linux/ratelimit.h>
  26 #include <linux/kthread.h>
  27 #include <linux/raid/pq.h>
  28 #include <linux/semaphore.h>
  29 #include <linux/uuid.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
 144 DEFINE_MUTEX(uuid_mutex);
 145 static LIST_HEAD(fs_uuids);
 146 struct list_head *btrfs_get_fs_uuids(void)
 147 {
 148         return &fs_uuids;
 149 }
 150
 151 static struct btrfs_fs_devices *__alloc_fs_devices(void)
 152 {
 153         struct btrfs_fs_devices *fs_devs;
 154
 155         fs_devs = kzalloc(sizeof(*fs_devs), GFP_KERNEL);
 156         if (!fs_devs)
 157                 return ERR_PTR(-ENOMEM);
 158
 159         mutex_init(&fs_devs->device_list_mutex);
 160
 161         INIT_LIST_HEAD(&fs_devs->devices);
 162         INIT_LIST_HEAD(&fs_devs->resized_devices);
 163         INIT_LIST_HEAD(&fs_devs->alloc_list);
 164         INIT_LIST_HEAD(&fs_devs->list);
 165
 166         return fs_devs;
 167 }
 168
 169 /**
 170  * alloc_fs_devices - allocate struct btrfs_fs_devices
 171  * @fsid:       a pointer to UUID for this FS.  If NULL a new UUID is
 172  *              generated.
 173  *
 174  * Return: a pointer to a new &struct btrfs_fs_devices on success;
 175  * ERR_PTR() on error.  Returned struct is not linked onto any lists and
 176  * can be destroyed with kfree() right away.
 177  */
 178 static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid)
 179 {
 180         struct btrfs_fs_devices *fs_devs;
 181
 182         fs_devs = __alloc_fs_devices();
 183         if (IS_ERR(fs_devs))
 184                 return fs_devs;
 185
 186         if (fsid)
 187                 memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE);
 188         else
 189                 generate_random_uuid(fs_devs->fsid);
 190
 191         return fs_devs;
 192 }
 193
 194 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
 195 {
 196         struct btrfs_device *device;
 197         WARN_ON(fs_devices->opened);
 198         while (!list_empty(&fs_devices->devices)) {
 199                 device = list_entry(fs_devices->devices.next,
 200                                     struct btrfs_device, dev_list);
 201                 list_del(&device->dev_list);
 202                 rcu_string_free(device->name);
 203                 kfree(device);
 204         }
 205         kfree(fs_devices);
 206 }
 207
 208 static void btrfs_kobject_uevent(struct block_device *bdev,
 209                                  enum kobject_action action)
 210 {
 211         int ret;
 212
 213         ret = kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, action);
 214         if (ret)
 215                 pr_warn("BTRFS: Sending event '%d' to kobject: '%s' (%p): failed\n",
 216                         action,
 217                         kobject_name(&disk_to_dev(bdev->bd_disk)->kobj),
 218                         &disk_to_dev(bdev->bd_disk)->kobj);
 219 }
 220
 221 void btrfs_cleanup_fs_uuids(void)
 222 {
 223         struct btrfs_fs_devices *fs_devices;
 224
 225         while (!list_empty(&fs_uuids)) {
 226                 fs_devices = list_entry(fs_uuids.next,
 227                                         struct btrfs_fs_devices, list);
 228                 list_del(&fs_devices->list);
 229                 free_fs_devices(fs_devices);
 230         }
 231 }
 232
 233 static struct btrfs_device *__alloc_device(void)
 234 {
 235         struct btrfs_device *dev;
 236
 237         dev = kzalloc(sizeof(*dev), GFP_KERNEL);
 238         if (!dev)
 239                 return ERR_PTR(-ENOMEM);
 240
 241         INIT_LIST_HEAD(&dev->dev_list);
 242         INIT_LIST_HEAD(&dev->dev_alloc_list);
 243         INIT_LIST_HEAD(&dev->resized_list);
 244
 245         spin_lock_init(&dev->io_lock);
 246
 247         spin_lock_init(&dev->reada_lock);
 248         atomic_set(&dev->reada_in_flight, 0);
 249         atomic_set(&dev->dev_stats_ccnt, 0);
 250         btrfs_device_data_ordered_init(dev);
 251         INIT_RADIX_TREE(&dev->reada_zones, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
 252         INIT_RADIX_TREE(&dev->reada_extents, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
 253
 254         return dev;
 255 }
 256
 257 static noinline struct btrfs_device *__find_device(struct list_head *head,
 258                                                    u64 devid, u8 *uuid)
 259 {
 260         struct btrfs_device *dev;
 261
 262         list_for_each_entry(dev, head, dev_list) {
 263                 if (dev->devid == devid &&
 264                     (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
 265                         return dev;
 266                 }
 267         }
 268         return NULL;
 269 }
 270
 271 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
 272 {
 273         struct btrfs_fs_devices *fs_devices;
 274
 275         list_for_each_entry(fs_devices, &fs_uuids, list) {
 276                 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
 277                         return fs_devices;
 278         }
 279         return NULL;
 280 }
 281
 282 static int
 283 btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder,
 284                       int flush, struct block_device **bdev,
 285                       struct buffer_head **bh)
 286 {
 287         int ret;
 288
 289         *bdev = blkdev_get_by_path(device_path, flags, holder);
 290
 291         if (IS_ERR(*bdev)) {
 292                 ret = PTR_ERR(*bdev);
 293                 goto error;
 294         }
 295
 296         if (flush)
 297                 filemap_write_and_wait((*bdev)->bd_inode->i_mapping);
 298         ret = set_blocksize(*bdev, 4096);
 299         if (ret) {
 300                 blkdev_put(*bdev, flags);
 301                 goto error;
 302         }
 303         invalidate_bdev(*bdev);
 304         *bh = btrfs_read_dev_super(*bdev);
 305         if (IS_ERR(*bh)) {
 306                 ret = PTR_ERR(*bh);
 307                 blkdev_put(*bdev, flags);
 308                 goto error;
 309         }
 310
 311         return 0;
 312
 313 error:
 314         *bdev = NULL;
 315         *bh = NULL;
 316         return ret;
 317 }
 318
 319 static void requeue_list(struct btrfs_pending_bios *pending_bios,
 320                         struct bio *head, struct bio *tail)
 321 {
 322
 323         struct bio *old_head;
 324
 325         old_head = pending_bios->head;
 326         pending_bios->head = head;
 327         if (pending_bios->tail)
 328                 tail->bi_next = old_head;
 329         else
 330                 pending_bios->tail = tail;
 331 }
 332
 333 /*
 334  * we try to collect pending bios for a device so we don't get a large
 335  * number of procs sending bios down to the same device.  This greatly
 336  * improves the schedulers ability to collect and merge the bios.
 337  *
 338  * But, it also turns into a long list of bios to process and that is sure
 339  * to eventually make the worker thread block.  The solution here is to
 340  * make some progress and then put this work struct back at the end of
 341  * the list if the block device is congested.  This way, multiple devices
 342  * can make progress from a single worker thread.
 343  */
 344 static noinline void run_scheduled_bios(struct btrfs_device *device)
 345 {
 346         struct bio *pending;
 347         struct backing_dev_info *bdi;
 348         struct btrfs_fs_info *fs_info;
 349         struct btrfs_pending_bios *pending_bios;
 350         struct bio *tail;
 351         struct bio *cur;
 352         int again = 0;
 353         unsigned long num_run;
 354         unsigned long batch_run = 0;
 355         unsigned long limit;
 356         unsigned long last_waited = 0;
 357         int force_reg = 0;
 358         int sync_pending = 0;
 359         struct blk_plug plug;
 360
 361         /*
 362          * this function runs all the bios we've collected for
 363          * a particular device.  We don't want to wander off to
 364          * another device without first sending all of these down.
 365          * So, setup a plug here and finish it off before we return
 366          */
 367         blk_start_plug(&plug);
 368
 369         bdi = blk_get_backing_dev_info(device->bdev);
 370         fs_info = device->dev_root->fs_info;
 371         limit = btrfs_async_submit_limit(fs_info);
 372         limit = limit * 2 / 3;
 373
 374 loop:
 375         spin_lock(&device->io_lock);
 376
 377 loop_lock:
 378         num_run = 0;
 379
 380         /* take all the bios off the list at once and process them
 381          * later on (without the lock held).  But, remember the
 382          * tail and other pointers so the bios can be properly reinserted
 383          * into the list if we hit congestion
 384          */
 385         if (!force_reg && device->pending_sync_bios.head) {
 386                 pending_bios = &device->pending_sync_bios;
 387                 force_reg = 1;
 388         } else {
 389                 pending_bios = &device->pending_bios;
 390                 force_reg = 0;
 391         }
 392
 393         pending = pending_bios->head;
 394         tail = pending_bios->tail;
 395         WARN_ON(pending && !tail);
 396
 397         /*
 398          * if pending was null this time around, no bios need processing
 399          * at all and we can stop.  Otherwise it'll loop back up again
 400          * and do an additional check so no bios are missed.
 401          *
 402          * device->running_pending is used to synchronize with the
 403          * schedule_bio code.
 404          */
 405         if (device->pending_sync_bios.head == NULL &&
 406             device->pending_bios.head == NULL) {
 407                 again = 0;
 408                 device->running_pending = 0;
 409         } else {
 410                 again = 1;
 411                 device->running_pending = 1;
 412         }
 413
 414         pending_bios->head = NULL;
 415         pending_bios->tail = NULL;
 416
 417         spin_unlock(&device->io_lock);
 418
 419         while (pending) {
 420
 421                 rmb();
 422                 /* we want to work on both lists, but do more bios on the
 423                  * sync list than the regular list
 424                  */
 425                 if ((num_run > 32 &&
 426                     pending_bios != &device->pending_sync_bios &&
 427                     device->pending_sync_bios.head) ||
 428                    (num_run > 64 && pending_bios == &device->pending_sync_bios &&
 429                     device->pending_bios.head)) {
 430                         spin_lock(&device->io_lock);
 431                         requeue_list(pending_bios, pending, tail);
 432                         goto loop_lock;
 433                 }
 434
 435                 cur = pending;
 436                 pending = pending->bi_next;
 437                 cur->bi_next = NULL;
 438
 439                 /*
 440                  * atomic_dec_return implies a barrier for waitqueue_active
 441                  */
 442                 if (atomic_dec_return(&fs_info->nr_async_bios) < limit &&
 443                     waitqueue_active(&fs_info->async_submit_wait))
 444                         wake_up(&fs_info->async_submit_wait);
 445
 446                 BUG_ON(atomic_read(&cur->__bi_cnt) == 0);
 447
 448                 /*
 449                  * if we're doing the sync list, record that our
 450                  * plug has some sync requests on it
 451                  *
 452                  * If we're doing the regular list and there are
 453                  * sync requests sitting around, unplug before
 454                  * we add more
 455                  */
 456                 if (pending_bios == &device->pending_sync_bios) {
 457                         sync_pending = 1;
 458                 } else if (sync_pending) {
 459                         blk_finish_plug(&plug);
 460                         blk_start_plug(&plug);
 461                         sync_pending = 0;
 462                 }
 463
 464                 btrfsic_submit_bio(cur->bi_rw, cur);
 465                 num_run++;
 466                 batch_run++;
 467
 468                 cond_resched();
 469
 470                 /*
 471                  * we made progress, there is more work to do and the bdi
 472                  * is now congested.  Back off and let other work structs
 473                  * run instead
 474                  */
 475                 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
 476                     fs_info->fs_devices->open_devices > 1) {
 477                         struct io_context *ioc;
 478
 479                         ioc = current->io_context;
 480
 481                         /*
 482                          * the main goal here is that we don't want to
 483                          * block if we're going to be able to submit
 484                          * more requests without blocking.
 485                          *
 486                          * This code does two great things, it pokes into
 487                          * the elevator code from a filesystem _and_
 488                          * it makes assumptions about how batching works.
 489                          */
 490                         if (ioc && ioc->nr_batch_requests > 0 &&
 491                             time_before(jiffies, ioc->last_waited + HZ/50UL) &&
 492                             (last_waited == 0 ||
 493                              ioc->last_waited == last_waited)) {
 494                                 /*
 495                                  * we want to go through our batch of
 496                                  * requests and stop.  So, we copy out
 497                                  * the ioc->last_waited time and test
 498                                  * against it before looping
 499                                  */
 500                                 last_waited = ioc->last_waited;
 501                                 cond_resched();
 502                                 continue;
 503                         }
 504                         spin_lock(&device->io_lock);
 505                         requeue_list(pending_bios, pending, tail);
 506                         device->running_pending = 1;
 507
 508                         spin_unlock(&device->io_lock);
 509                         btrfs_queue_work(fs_info->submit_workers,
 510                                          &device->work);
 511                         goto done;
 512                 }
 513                 /* unplug every 64 requests just for good measure */
 514                 if (batch_run % 64 == 0) {
 515                         blk_finish_plug(&plug);
 516                         blk_start_plug(&plug);
 517                         sync_pending = 0;
 518                 }
 519         }
 520
 521         cond_resched();
 522         if (again)
 523                 goto loop;
 524
 525         spin_lock(&device->io_lock);
 526         if (device->pending_bios.head || device->pending_sync_bios.head)
 527                 goto loop_lock;
 528         spin_unlock(&device->io_lock);
 529
 530 done:
 531         blk_finish_plug(&plug);
 532 }
 533
 534 static void pending_bios_fn(struct btrfs_work *work)
 535 {
 536         struct btrfs_device *device;
 537
 538         device = container_of(work, struct btrfs_device, work);
 539         run_scheduled_bios(device);
 540 }
 541
 542
 543 void btrfs_free_stale_device(struct btrfs_device *cur_dev)
 544 {
 545         struct btrfs_fs_devices *fs_devs;
 546         struct btrfs_device *dev;
 547
 548         if (!cur_dev->name)
 549                 return;
 550
 551         list_for_each_entry(fs_devs, &fs_uuids, list) {
 552                 int del = 1;
 553
 554                 if (fs_devs->opened)
 555                         continue;
 556                 if (fs_devs->seeding)
 557                         continue;
 558
 559                 list_for_each_entry(dev, &fs_devs->devices, dev_list) {
 560
 561                         if (dev == cur_dev)
 562                                 continue;
 563                         if (!dev->name)
 564                                 continue;
 565
 566                         /*
 567                          * Todo: This won't be enough. What if the same device
 568                          * comes back (with new uuid and) with its mapper path?
 569                          * But for now, this does help as mostly an admin will
 570                          * either use mapper or non mapper path throughout.
 571                          */
 572                         rcu_read_lock();
 573                         del = strcmp(rcu_str_deref(dev->name),
 574                                                 rcu_str_deref(cur_dev->name));
 575                         rcu_read_unlock();
 576                         if (!del)
 577                                 break;
 578                 }
 579
 580                 if (!del) {
 581                         /* delete the stale device */
 582                         if (fs_devs->num_devices == 1) {
 583                                 btrfs_sysfs_remove_fsid(fs_devs);
 584                                 list_del(&fs_devs->list);
 585                                 free_fs_devices(fs_devs);
 586                         } else {
 587                                 fs_devs->num_devices--;
 588                                 list_del(&dev->dev_list);
 589                                 rcu_string_free(dev->name);
 590                                 kfree(dev);
 591                         }
 592                         break;
 593                 }
 594         }
 595 }
 596
 597 /*
 598  * Add new device to list of registered devices
 599  *
 600  * Returns:
 601  * 1   - first time device is seen
 602  * 0   - device already known
 603  * < 0 - error
 604  */
 605 static noinline int device_list_add(const char *path,
 606                            struct btrfs_super_block *disk_super,
 607                            u64 devid, struct btrfs_fs_devices **fs_devices_ret)
 608 {
 609         struct btrfs_device *device;
 610         struct btrfs_fs_devices *fs_devices;
 611         struct rcu_string *name;
 612         int ret = 0;
 613         u64 found_transid = btrfs_super_generation(disk_super);
 614
 615         fs_devices = find_fsid(disk_super->fsid);
 616         if (!fs_devices) {
 617                 fs_devices = alloc_fs_devices(disk_super->fsid);
 618                 if (IS_ERR(fs_devices))
 619                         return PTR_ERR(fs_devices);
 620
 621                 list_add(&fs_devices->list, &fs_uuids);
 622
 623                 device = NULL;
 624         } else {
 625                 device = __find_device(&fs_devices->devices, devid,
 626                                        disk_super->dev_item.uuid);
 627         }
 628
 629         if (!device) {
 630                 if (fs_devices->opened)
 631                         return -EBUSY;
 632
 633                 device = btrfs_alloc_device(NULL, &devid,
 634                                             disk_super->dev_item.uuid);
 635                 if (IS_ERR(device)) {
 636                         /* we can safely leave the fs_devices entry around */
 637                         return PTR_ERR(device);
 638                 }
 639
 640                 name = rcu_string_strdup(path, GFP_NOFS);
 641                 if (!name) {
 642                         kfree(device);
 643                         return -ENOMEM;
 644                 }
 645                 rcu_assign_pointer(device->name, name);
 646
 647                 mutex_lock(&fs_devices->device_list_mutex);
 648                 list_add_rcu(&device->dev_list, &fs_devices->devices);
 649                 fs_devices->num_devices++;
 650                 mutex_unlock(&fs_devices->device_list_mutex);
 651
 652                 ret = 1;
 653                 device->fs_devices = fs_devices;
 654         } else if (!device->name || strcmp(device->name->str, path)) {
 655                 /*
 656                  * When FS is already mounted.
 657                  * 1. If you are here and if the device->name is NULL that
 658                  *    means this device was missing at time of FS mount.
 659                  * 2. If you are here and if the device->name is different
 660                  *    from 'path' that means either
 661                  *      a. The same device disappeared and reappeared with
 662                  *         different name. or
 663                  *      b. The missing-disk-which-was-replaced, has
 664                  *         reappeared now.
 665                  *
 666                  * We must allow 1 and 2a above. But 2b would be a spurious
 667                  * and unintentional.
 668                  *
 669                  * Further in case of 1 and 2a above, the disk at 'path'
 670                  * would have missed some transaction when it was away and
 671                  * in case of 2a the stale bdev has to be updated as well.
 672                  * 2b must not be allowed at all time.
 673                  */
 674
 675                 /*
 676                  * For now, we do allow update to btrfs_fs_device through the
 677                  * btrfs dev scan cli after FS has been mounted.  We're still
 678                  * tracking a problem where systems fail mount by subvolume id
 679                  * when we reject replacement on a mounted FS.
 680                  */
 681                 if (!fs_devices->opened && found_transid < device->generation) {
 682                         /*
 683                          * That is if the FS is _not_ mounted and if you
 684                          * are here, that means there is more than one
 685                          * disk with same uuid and devid.We keep the one
 686                          * with larger generation number or the last-in if
 687                          * generation are equal.
 688                          */
 689                         return -EEXIST;
 690                 }
 691
 692                 name = rcu_string_strdup(path, GFP_NOFS);
 693                 if (!name)
 694                         return -ENOMEM;
 695                 rcu_string_free(device->name);
 696                 rcu_assign_pointer(device->name, name);
 697                 if (device->missing) {
 698                         fs_devices->missing_devices--;
 699                         device->missing = 0;
 700                 }
 701         }
 702
 703         /*
 704          * Unmount does not free the btrfs_device struct but would zero
 705          * generation along with most of the other members. So just update
 706          * it back. We need it to pick the disk with largest generation
 707          * (as above).
 708          */
 709         if (!fs_devices->opened)
 710                 device->generation = found_transid;
 711
 712         /*
 713          * if there is new btrfs on an already registered device,
 714          * then remove the stale device entry.
 715          */
 716         if (ret > 0)
 717                 btrfs_free_stale_device(device);
 718
 719         *fs_devices_ret = fs_devices;
 720
 721         return ret;
 722 }
 723
 724 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
 725 {
 726         struct btrfs_fs_devices *fs_devices;
 727         struct btrfs_device *device;
 728         struct btrfs_device *orig_dev;
 729
 730         fs_devices = alloc_fs_devices(orig->fsid);
 731         if (IS_ERR(fs_devices))
 732                 return fs_devices;
 733
 734         mutex_lock(&orig->device_list_mutex);
 735         fs_devices->total_devices = orig->total_devices;
 736
 737         /* We have held the volume lock, it is safe to get the devices. */
 738         list_for_each_entry(orig_dev, &orig->devices, dev_list) {
 739                 struct rcu_string *name;
 740
 741                 device = btrfs_alloc_device(NULL, &orig_dev->devid,
 742                                             orig_dev->uuid);
 743                 if (IS_ERR(device))
 744                         goto error;
 745
 746                 /*
 747                  * This is ok to do without rcu read locked because we hold the
 748                  * uuid mutex so nothing we touch in here is going to disappear.
 749                  */
 750                 if (orig_dev->name) {
 751                         name = rcu_string_strdup(orig_dev->name->str,
 752                                         GFP_KERNEL);
 753                         if (!name) {
 754                                 kfree(device);
 755                                 goto error;
 756                         }
 757                         rcu_assign_pointer(device->name, name);
 758                 }
 759
 760                 list_add(&device->dev_list, &fs_devices->devices);
 761                 device->fs_devices = fs_devices;
 762                 fs_devices->num_devices++;
 763         }
 764         mutex_unlock(&orig->device_list_mutex);
 765         return fs_devices;
 766 error:
 767         mutex_unlock(&orig->device_list_mutex);
 768         free_fs_devices(fs_devices);
 769         return ERR_PTR(-ENOMEM);
 770 }
 771
 772 void btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices, int step)
 773 {
 774         struct btrfs_device *device, *next;
 775         struct btrfs_device *latest_dev = NULL;
 776
 777         mutex_lock(&uuid_mutex);
 778 again:
 779         /* This is the initialized path, it is safe to release the devices. */
 780         list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
 781                 if (device->in_fs_metadata) {
 782                         if (!device->is_tgtdev_for_dev_replace &&
 783                             (!latest_dev ||
 784                              device->generation > latest_dev->generation)) {
 785                                 latest_dev = device;
 786                         }
 787                         continue;
 788                 }
 789
 790                 if (device->devid == BTRFS_DEV_REPLACE_DEVID) {
 791                         /*
 792                          * In the first step, keep the device which has
 793                          * the correct fsid and the devid that is used
 794                          * for the dev_replace procedure.
 795                          * In the second step, the dev_replace state is
 796                          * read from the device tree and it is known
 797                          * whether the procedure is really active or
 798                          * not, which means whether this device is
 799                          * used or whether it should be removed.
 800                          */
 801                         if (step == 0 || device->is_tgtdev_for_dev_replace) {
 802                                 continue;
 803                         }
 804                 }
 805                 if (device->bdev) {
 806                         blkdev_put(device->bdev, device->mode);
 807                         device->bdev = NULL;
 808                         fs_devices->open_devices--;
 809                 }
 810                 if (device->writeable) {
 811                         list_del_init(&device->dev_alloc_list);
 812                         device->writeable = 0;
 813                         if (!device->is_tgtdev_for_dev_replace)
 814                                 fs_devices->rw_devices--;
 815                 }
 816                 list_del_init(&device->dev_list);
 817                 fs_devices->num_devices--;
 818                 rcu_string_free(device->name);
 819                 kfree(device);
 820         }
 821
 822         if (fs_devices->seed) {
 823                 fs_devices = fs_devices->seed;
 824                 goto again;
 825         }
 826
 827         fs_devices->latest_bdev = latest_dev->bdev;
 828
 829         mutex_unlock(&uuid_mutex);
 830 }
 831
 832 static void __free_device(struct work_struct *work)
 833 {
 834         struct btrfs_device *device;
 835
 836         device = container_of(work, struct btrfs_device, rcu_work);
 837         rcu_string_free(device->name);
 838         kfree(device);
 839 }
 840
 841 static void free_device(struct rcu_head *head)
 842 {
 843         struct btrfs_device *device;
 844
 845         device = container_of(head, struct btrfs_device, rcu);
 846
 847         INIT_WORK(&device->rcu_work, __free_device);
 848         schedule_work(&device->rcu_work);
 849 }
 850
 851 static void btrfs_close_bdev(struct btrfs_device *device)
 852 {
 853         if (device->bdev && device->writeable) {
 854                 sync_blockdev(device->bdev);
 855                 invalidate_bdev(device->bdev);
 856         }
 857
 858         if (device->bdev)
 859                 blkdev_put(device->bdev, device->mode);
 860 }
 861
 862 static void btrfs_close_one_device(struct btrfs_device *device)
 863 {
 864         struct btrfs_fs_devices *fs_devices = device->fs_devices;
 865         struct btrfs_device *new_device;
 866         struct rcu_string *name;
 867
 868         if (device->bdev)
 869                 fs_devices->open_devices--;
 870
 871         if (device->writeable &&
 872             device->devid != BTRFS_DEV_REPLACE_DEVID) {
 873                 list_del_init(&device->dev_alloc_list);
 874                 fs_devices->rw_devices--;
 875         }
 876
 877         if (device->missing)
 878                 fs_devices->missing_devices--;
 879
 880         btrfs_close_bdev(device);
 881
 882         new_device = btrfs_alloc_device(NULL, &device->devid,
 883                                         device->uuid);
 884         BUG_ON(IS_ERR(new_device)); /* -ENOMEM */
 885
 886         /* Safe because we are under uuid_mutex */
 887         if (device->name) {
 888                 name = rcu_string_strdup(device->name->str, GFP_NOFS);
 889                 BUG_ON(!name); /* -ENOMEM */
 890                 rcu_assign_pointer(new_device->name, name);
 891         }
 892
 893         list_replace_rcu(&device->dev_list, &new_device->dev_list);
 894         new_device->fs_devices = device->fs_devices;
 895
 896         call_rcu(&device->rcu, free_device);
 897 }
 898
 899 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
 900 {
 901         struct btrfs_device *device, *tmp;
 902
 903         if (--fs_devices->opened > 0)
 904                 return 0;
 905
 906         mutex_lock(&fs_devices->device_list_mutex);
 907         list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list) {
 908                 btrfs_close_one_device(device);
 909         }
 910         mutex_unlock(&fs_devices->device_list_mutex);
 911
 912         WARN_ON(fs_devices->open_devices);
 913         WARN_ON(fs_devices->rw_devices);
 914         fs_devices->opened = 0;
 915         fs_devices->seeding = 0;
 916
 917         return 0;
 918 }
 919
 920 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
 921 {
 922         struct btrfs_fs_devices *seed_devices = NULL;
 923         int ret;
 924
 925         mutex_lock(&uuid_mutex);
 926         ret = __btrfs_close_devices(fs_devices);
 927         if (!fs_devices->opened) {
 928                 seed_devices = fs_devices->seed;
 929                 fs_devices->seed = NULL;
 930         }
 931         mutex_unlock(&uuid_mutex);
 932
 933         while (seed_devices) {
 934                 fs_devices = seed_devices;
 935                 seed_devices = fs_devices->seed;
 936                 __btrfs_close_devices(fs_devices);
 937                 free_fs_devices(fs_devices);
 938         }
 939         /*
 940          * Wait for rcu kworkers under __btrfs_close_devices
 941          * to finish all blkdev_puts so device is really
 942          * free when umount is done.
 943          */
 944         rcu_barrier();
 945         return ret;
 946 }
 947
 948 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
 949                                 fmode_t flags, void *holder)
 950 {
 951         struct request_queue *q;
 952         struct block_device *bdev;
 953         struct list_head *head = &fs_devices->devices;
 954         struct btrfs_device *device;
 955         struct btrfs_device *latest_dev = NULL;
 956         struct buffer_head *bh;
 957         struct btrfs_super_block *disk_super;
 958         u64 devid;
 959         int seeding = 1;
 960         int ret = 0;
 961
 962         flags |= FMODE_EXCL;
 963
 964         list_for_each_entry(device, head, dev_list) {
 965                 if (device->bdev)
 966                         continue;
 967                 if (!device->name)
 968                         continue;
 969
 970                 /* Just open everything we can; ignore failures here */
 971                 if (btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1,
 972                                             &bdev, &bh))
 973                         continue;
 974
 975                 disk_super = (struct btrfs_super_block *)bh->b_data;
 976                 devid = btrfs_stack_device_id(&disk_super->dev_item);
 977                 if (devid != device->devid)
 978                         goto error_brelse;
 979
 980                 if (memcmp(device->uuid, disk_super->dev_item.uuid,
 981                            BTRFS_UUID_SIZE))
 982                         goto error_brelse;
 983
 984                 device->generation = btrfs_super_generation(disk_super);
 985                 if (!latest_dev ||
 986                     device->generation > latest_dev->generation)
 987                         latest_dev = device;
 988
 989                 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
 990                         device->writeable = 0;
 991                 } else {
 992                         device->writeable = !bdev_read_only(bdev);
 993                         seeding = 0;
 994                 }
 995
 996                 q = bdev_get_queue(bdev);
 997                 if (blk_queue_discard(q))
 998                         device->can_discard = 1;
 999
1000                 device->bdev = bdev;
1001                 device->in_fs_metadata = 0;
1002                 device->mode = flags;
1003
1004                 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
1005                         fs_devices->rotating = 1;
1006
1007                 fs_devices->open_devices++;
1008                 if (device->writeable &&
1009                     device->devid != BTRFS_DEV_REPLACE_DEVID) {
1010                         fs_devices->rw_devices++;
1011                         list_add(&device->dev_alloc_list,
1012                                  &fs_devices->alloc_list);
1013                 }
1014                 brelse(bh);
1015                 continue;
1016
1017 error_brelse:
1018                 brelse(bh);
1019                 blkdev_put(bdev, flags);
1020                 continue;
1021         }
1022         if (fs_devices->open_devices == 0) {
1023                 ret = -EINVAL;
1024                 goto out;
1025         }
1026         fs_devices->seeding = seeding;
1027         fs_devices->opened = 1;
1028         fs_devices->latest_bdev = latest_dev->bdev;
1029         fs_devices->total_rw_bytes = 0;
1030 out:
1031         return ret;
1032 }
1033
1034 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
1035                        fmode_t flags, void *holder)
1036 {
1037         int ret;
1038
1039         mutex_lock(&uuid_mutex);
1040         if (fs_devices->opened) {
1041                 fs_devices->opened++;
1042                 ret = 0;
1043         } else {
1044                 ret = __btrfs_open_devices(fs_devices, flags, holder);
1045         }
1046         mutex_unlock(&uuid_mutex);
1047         return ret;
1048 }
1049
1050 void btrfs_release_disk_super(struct page *page)
1051 {
1052         kunmap(page);
1053         put_page(page);
1054 }
1055
1056 int btrfs_read_disk_super(struct block_device *bdev, u64 bytenr,
1057                 struct page **page, struct btrfs_super_block **disk_super)
1058 {
1059         void *p;
1060         pgoff_t index;
1061
1062         /* make sure our super fits in the device */
1063         if (bytenr + PAGE_SIZE >= i_size_read(bdev->bd_inode))
1064                 return 1;
1065
1066         /* make sure our super fits in the page */
1067         if (sizeof(**disk_super) > PAGE_SIZE)
1068                 return 1;
1069
1070         /* make sure our super doesn't straddle pages on disk */
1071         index = bytenr >> PAGE_SHIFT;
1072         if ((bytenr + sizeof(**disk_super) - 1) >> PAGE_SHIFT != index)
1073                 return 1;
1074
1075         /* pull in the page with our super */
1076         *page = read_cache_page_gfp(bdev->bd_inode->i_mapping,
1077                                    index, GFP_KERNEL);
1078
1079         if (IS_ERR_OR_NULL(*page))
1080                 return 1;
1081
1082         p = kmap(*page);
1083
1084         /* align our pointer to the offset of the super block */
1085         *disk_super = p + (bytenr & ~PAGE_MASK);
1086
1087         if (btrfs_super_bytenr(*disk_super) != bytenr ||
1088             btrfs_super_magic(*disk_super) != BTRFS_MAGIC) {
1089                 btrfs_release_disk_super(*page);
1090                 return 1;
1091         }
1092
1093         if ((*disk_super)->label[0] &&
1094                 (*disk_super)->label[BTRFS_LABEL_SIZE - 1])
1095                 (*disk_super)->label[BTRFS_LABEL_SIZE - 1] = '\0';
1096
1097         return 0;
1098 }
1099
1100 /*
1101  * Look for a btrfs signature on a device. This may be called out of the mount path
1102  * and we are not allowed to call set_blocksize during the scan. The superblock
1103  * is read via pagecache
1104  */
1105 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
1106                           struct btrfs_fs_devices **fs_devices_ret)
1107 {
1108         struct btrfs_super_block *disk_super;
1109         struct block_device *bdev;
1110         struct page *page;
1111         int ret = -EINVAL;
1112         u64 devid;
1113         u64 transid;
1114         u64 total_devices;
1115         u64 bytenr;
1116
1117         /*
1118          * we would like to check all the supers, but that would make
1119          * a btrfs mount succeed after a mkfs from a different FS.
1120          * So, we need to add a special mount option to scan for
1121          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
1122          */
1123         bytenr = btrfs_sb_offset(0);
1124         flags |= FMODE_EXCL;
1125         mutex_lock(&uuid_mutex);
1126
1127         bdev = blkdev_get_by_path(path, flags, holder);
1128         if (IS_ERR(bdev)) {
1129                 ret = PTR_ERR(bdev);
1130                 goto error;
1131         }
1132
1133         if (btrfs_read_disk_super(bdev, bytenr, &page, &disk_super))
1134                 goto error_bdev_put;
1135
1136         devid = btrfs_stack_device_id(&disk_super->dev_item);
1137         transid = btrfs_super_generation(disk_super);
1138         total_devices = btrfs_super_num_devices(disk_super);
1139
1140         ret = device_list_add(path, disk_super, devid, fs_devices_ret);
1141         if (ret > 0) {
1142                 if (disk_super->label[0]) {
1143                         printk(KERN_INFO "BTRFS: device label %s ", disk_super->label);
1144                 } else {
1145                         printk(KERN_INFO "BTRFS: device fsid %pU ", disk_super->fsid);
1146                 }
1147
1148                 printk(KERN_CONT "devid %llu transid %llu %s\n", devid, transid, path);
1149                 ret = 0;
1150         }
1151         if (!ret && fs_devices_ret)
1152                 (*fs_devices_ret)->total_devices = total_devices;
1153
1154         btrfs_release_disk_super(page);
1155
1156 error_bdev_put:
1157         blkdev_put(bdev, flags);
1158 error:
1159         mutex_unlock(&uuid_mutex);
1160         return ret;
1161 }
1162
1163 /* helper to account the used device space in the range */
1164 int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
1165                                    u64 end, u64 *length)
1166 {
1167         struct btrfs_key key;
1168         struct btrfs_root *root = device->dev_root;
1169         struct btrfs_dev_extent *dev_extent;
1170         struct btrfs_path *path;
1171         u64 extent_end;
1172         int ret;
1173         int slot;
1174         struct extent_buffer *l;
1175
1176         *length = 0;
1177
1178         if (start >= device->total_bytes || device->is_tgtdev_for_dev_replace)
1179                 return 0;
1180
1181         path = btrfs_alloc_path();
1182         if (!path)
1183                 return -ENOMEM;
1184         path->reada = READA_FORWARD;
1185
1186         key.objectid = device->devid;
1187         key.offset = start;
1188         key.type = BTRFS_DEV_EXTENT_KEY;
1189
1190         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1191         if (ret < 0)
1192                 goto out;
1193         if (ret > 0) {
1194                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1195                 if (ret < 0)
1196                         goto out;
1197         }
1198
1199         while (1) {
1200                 l = path->nodes[0];
1201                 slot = path->slots[0];
1202                 if (slot >= btrfs_header_nritems(l)) {
1203                         ret = btrfs_next_leaf(root, path);
1204                         if (ret == 0)
1205                                 continue;
1206                         if (ret < 0)
1207                                 goto out;
1208
1209                         break;
1210                 }
1211                 btrfs_item_key_to_cpu(l, &key, slot);
1212
1213                 if (key.objectid < device->devid)
1214                         goto next;
1215
1216                 if (key.objectid > device->devid)
1217                         break;
1218
1219                 if (key.type != BTRFS_DEV_EXTENT_KEY)
1220                         goto next;
1221
1222                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1223                 extent_end = key.offset + btrfs_dev_extent_length(l,
1224                                                                   dev_extent);
1225                 if (key.offset <= start && extent_end > end) {
1226                         *length = end - start + 1;
1227                         break;
1228                 } else if (key.offset <= start && extent_end > start)
1229                         *length += extent_end - start;
1230                 else if (key.offset > start && extent_end <= end)
1231                         *length += extent_end - key.offset;
1232                 else if (key.offset > start && key.offset <= end) {
1233                         *length += end - key.offset + 1;
1234                         break;
1235                 } else if (key.offset > end)
1236                         break;
1237
1238 next:
1239                 path->slots[0]++;
1240         }
1241         ret = 0;
1242 out:
1243         btrfs_free_path(path);
1244         return ret;
1245 }
1246
1247 static int contains_pending_extent(struct btrfs_transaction *transaction,
1248                                    struct btrfs_device *device,
1249                                    u64 *start, u64 len)
1250 {
1251         struct btrfs_fs_info *fs_info = device->dev_root->fs_info;
1252         struct extent_map *em;
1253         struct list_head *search_list = &fs_info->pinned_chunks;
1254         int ret = 0;
1255         u64 physical_start = *start;
1256
1257         if (transaction)
1258                 search_list = &transaction->pending_chunks;
1259 again:
1260         list_for_each_entry(em, search_list, list) {
1261                 struct map_lookup *map;
1262                 int i;
1263
1264                 map = em->map_lookup;
1265                 for (i = 0; i < map->num_stripes; i++) {
1266                         u64 end;
1267
1268                         if (map->stripes[i].dev != device)
1269                                 continue;
1270                         if (map->stripes[i].physical >= physical_start + len ||
1271                             map->stripes[i].physical + em->orig_block_len <=
1272                             physical_start)
1273                                 continue;
1274                         /*
1275                          * Make sure that while processing the pinned list we do
1276                          * not override our *start with a lower value, because
1277                          * we can have pinned chunks that fall within this
1278                          * device hole and that have lower physical addresses
1279                          * than the pending chunks we processed before. If we
1280                          * do not take this special care we can end up getting
1281                          * 2 pending chunks that start at the same physical
1282                          * device offsets because the end offset of a pinned
1283                          * chunk can be equal to the start offset of some
1284                          * pending chunk.
1285                          */
1286                         end = map->stripes[i].physical + em->orig_block_len;
1287                         if (end > *start) {
1288                                 *start = end;
1289                                 ret = 1;
1290                         }
1291                 }
1292         }
1293         if (search_list != &fs_info->pinned_chunks) {
1294                 search_list = &fs_info->pinned_chunks;
1295                 goto again;
1296         }
1297
1298         return ret;
1299 }
1300
1301
1302 /*
1303  * find_free_dev_extent_start - find free space in the specified device
1304  * @device:       the device which we search the free space in
1305  * @num_bytes:    the size of the free space that we need
1306  * @search_start: the position from which to begin the search
1307  * @start:        store the start of the free space.
1308  * @len:          the size of the free space. that we find, or the size
1309  *                of the max free space if we don't find suitable free space
1310  *
1311  * this uses a pretty simple search, the expectation is that it is
1312  * called very infrequently and that a given device has a small number
1313  * of extents
1314  *
1315  * @start is used to store the start of the free space if we find. But if we
1316  * don't find suitable free space, it will be used to store the start position
1317  * of the max free space.
1318  *
1319  * @len is used to store the size of the free space that we find.
1320  * But if we don't find suitable free space, it is used to store the size of
1321  * the max free space.
1322  */
1323 int find_free_dev_extent_start(struct btrfs_transaction *transaction,
1324                                struct btrfs_device *device, u64 num_bytes,
1325                                u64 search_start, u64 *start, u64 *len)
1326 {
1327         struct btrfs_key key;
1328         struct btrfs_root *root = device->dev_root;
1329         struct btrfs_dev_extent *dev_extent;
1330         struct btrfs_path *path;
1331         u64 hole_size;
1332         u64 max_hole_start;
1333         u64 max_hole_size;
1334         u64 extent_end;
1335         u64 search_end = device->total_bytes;
1336         int ret;
1337         int slot;
1338         struct extent_buffer *l;
1339         u64 min_search_start;
1340
1341         /*
1342          * We don't want to overwrite the superblock on the drive nor any area
1343          * used by the boot loader (grub for example), so we make sure to start
1344          * at an offset of at least 1MB.
1345          */
1346         min_search_start = max(root->fs_info->alloc_start, 1024ull * 1024);
1347         search_start = max(search_start, min_search_start);
1348
1349         path = btrfs_alloc_path();
1350         if (!path)
1351                 return -ENOMEM;
1352
1353         max_hole_start = search_start;
1354         max_hole_size = 0;
1355
1356 again:
1357         if (search_start >= search_end || device->is_tgtdev_for_dev_replace) {
1358                 ret = -ENOSPC;
1359                 goto out;
1360         }
1361
1362         path->reada = READA_FORWARD;
1363         path->search_commit_root = 1;
1364         path->skip_locking = 1;
1365
1366         key.objectid = device->devid;
1367         key.offset = search_start;
1368         key.type = BTRFS_DEV_EXTENT_KEY;
1369
1370         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1371         if (ret < 0)
1372                 goto out;
1373         if (ret > 0) {
1374                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1375                 if (ret < 0)
1376                         goto out;
1377         }
1378
1379         while (1) {
1380                 l = path->nodes[0];
1381                 slot = path->slots[0];
1382                 if (slot >= btrfs_header_nritems(l)) {
1383                         ret = btrfs_next_leaf(root, path);
1384                         if (ret == 0)
1385                                 continue;
1386                         if (ret < 0)
1387                                 goto out;
1388
1389                         break;
1390                 }
1391                 btrfs_item_key_to_cpu(l, &key, slot);
1392
1393                 if (key.objectid < device->devid)
1394                         goto next;
1395
1396                 if (key.objectid > device->devid)
1397                         break;
1398
1399                 if (key.type != BTRFS_DEV_EXTENT_KEY)
1400                         goto next;
1401
1402                 if (key.offset > search_start) {
1403                         hole_size = key.offset - search_start;
1404
1405                         /*
1406                          * Have to check before we set max_hole_start, otherwise
1407                          * we could end up sending back this offset anyway.
1408                          */
1409                         if (contains_pending_extent(transaction, device,
1410                                                     &search_start,
1411                                                     hole_size)) {
1412                                 if (key.offset >= search_start) {
1413                                         hole_size = key.offset - search_start;
1414                                 } else {
1415                                         WARN_ON_ONCE(1);
1416                                         hole_size = 0;
1417                                 }
1418                         }
1419
1420                         if (hole_size > max_hole_size) {
1421                                 max_hole_start = search_start;
1422                                 max_hole_size = hole_size;
1423                         }
1424
1425                         /*
1426                          * If this free space is greater than which we need,
1427                          * it must be the max free space that we have found
1428                          * until now, so max_hole_start must point to the start
1429                          * of this free space and the length of this free space
1430                          * is stored in max_hole_size. Thus, we return
1431                          * max_hole_start and max_hole_size and go back to the
1432                          * caller.
1433                          */
1434                         if (hole_size >= num_bytes) {
1435                                 ret = 0;
1436                                 goto out;
1437                         }
1438                 }
1439
1440                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1441                 extent_end = key.offset + btrfs_dev_extent_length(l,
1442                                                                   dev_extent);
1443                 if (extent_end > search_start)
1444                         search_start = extent_end;
1445 next:
1446                 path->slots[0]++;
1447                 cond_resched();
1448         }
1449
1450         /*
1451          * At this point, search_start should be the end of
1452          * allocated dev extents, and when shrinking the device,
1453          * search_end may be smaller than search_start.
1454          */
1455         if (search_end > search_start) {
1456                 hole_size = search_end - search_start;
1457
1458                 if (contains_pending_extent(transaction, device, &search_start,
1459                                             hole_size)) {
1460                         btrfs_release_path(path);
1461                         goto again;
1462                 }
1463
1464                 if (hole_size > max_hole_size) {
1465                         max_hole_start = search_start;
1466                         max_hole_size = hole_size;
1467                 }
1468         }
1469
1470         /* See above. */
1471         if (max_hole_size < num_bytes)
1472                 ret = -ENOSPC;
1473         else
1474                 ret = 0;
1475
1476 out:
1477         btrfs_free_path(path);
1478         *start = max_hole_start;
1479         if (len)
1480                 *len = max_hole_size;
1481         return ret;
1482 }
1483
1484 int find_free_dev_extent(struct btrfs_trans_handle *trans,
1485                          struct btrfs_device *device, u64 num_bytes,
1486                          u64 *start, u64 *len)
1487 {
1488         /* FIXME use last free of some kind */
1489         return find_free_dev_extent_start(trans->transaction, device,
1490                                           num_bytes, 0, start, len);
1491 }
1492
1493 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1494                           struct btrfs_device *device,
1495                           u64 start, u64 *dev_extent_len)
1496 {
1497         int ret;
1498         struct btrfs_path *path;
1499         struct btrfs_root *root = device->dev_root;
1500         struct btrfs_key key;
1501         struct btrfs_key found_key;
1502         struct extent_buffer *leaf = NULL;
1503         struct btrfs_dev_extent *extent = NULL;
1504
1505         path = btrfs_alloc_path();
1506         if (!path)
1507                 return -ENOMEM;
1508
1509         key.objectid = device->devid;
1510         key.offset = start;
1511         key.type = BTRFS_DEV_EXTENT_KEY;
1512 again:
1513         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1514         if (ret > 0) {
1515                 ret = btrfs_previous_item(root, path, key.objectid,
1516                                           BTRFS_DEV_EXTENT_KEY);
1517                 if (ret)
1518                         goto out;
1519                 leaf = path->nodes[0];
1520                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1521                 extent = btrfs_item_ptr(leaf, path->slots[0],
1522                                         struct btrfs_dev_extent);
1523                 BUG_ON(found_key.offset > start || found_key.offset +
1524                        btrfs_dev_extent_length(leaf, extent) < start);
1525                 key = found_key;
1526                 btrfs_release_path(path);
1527                 goto again;
1528         } else if (ret == 0) {
1529                 leaf = path->nodes[0];
1530                 extent = btrfs_item_ptr(leaf, path->slots[0],
1531                                         struct btrfs_dev_extent);
1532         } else {
1533                 btrfs_handle_fs_error(root->fs_info, ret, "Slot search failed");
1534                 goto out;
1535         }
1536
1537         *dev_extent_len = btrfs_dev_extent_length(leaf, extent);
1538
1539         ret = btrfs_del_item(trans, root, path);
1540         if (ret) {
1541                 btrfs_handle_fs_error(root->fs_info, ret,
1542                             "Failed to remove dev extent item");
1543         } else {
1544                 set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags);
1545         }
1546 out:
1547         btrfs_free_path(path);
1548         return ret;
1549 }
1550
1551 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1552                                   struct btrfs_device *device,
1553                                   u64 chunk_tree, u64 chunk_objectid,
1554                                   u64 chunk_offset, u64 start, u64 num_bytes)
1555 {
1556         int ret;
1557         struct btrfs_path *path;
1558         struct btrfs_root *root = device->dev_root;
1559         struct btrfs_dev_extent *extent;
1560         struct extent_buffer *leaf;
1561         struct btrfs_key key;
1562
1563         WARN_ON(!device->in_fs_metadata);
1564         WARN_ON(device->is_tgtdev_for_dev_replace);
1565         path = btrfs_alloc_path();
1566         if (!path)
1567                 return -ENOMEM;
1568
1569         key.objectid = device->devid;
1570         key.offset = start;
1571         key.type = BTRFS_DEV_EXTENT_KEY;
1572         ret = btrfs_insert_empty_item(trans, root, path, &key,
1573                                       sizeof(*extent));
1574         if (ret)
1575                 goto out;
1576
1577         leaf = path->nodes[0];
1578         extent = btrfs_item_ptr(leaf, path->slots[0],
1579                                 struct btrfs_dev_extent);
1580         btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
1581         btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
1582         btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1583
1584         write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
1585                     btrfs_dev_extent_chunk_tree_uuid(extent), BTRFS_UUID_SIZE);
1586
1587         btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1588         btrfs_mark_buffer_dirty(leaf);
1589 out:
1590         btrfs_free_path(path);
1591         return ret;
1592 }
1593
1594 static u64 find_next_chunk(struct btrfs_fs_info *fs_info)
1595 {
1596         struct extent_map_tree *em_tree;
1597         struct extent_map *em;
1598         struct rb_node *n;
1599         u64 ret = 0;
1600
1601         em_tree = &fs_info->mapping_tree.map_tree;
1602         read_lock(&em_tree->lock);
1603         n = rb_last(&em_tree->map);
1604         if (n) {
1605                 em = rb_entry(n, struct extent_map, rb_node);
1606                 ret = em->start + em->len;
1607         }
1608         read_unlock(&em_tree->lock);
1609
1610         return ret;
1611 }
1612
1613 static noinline int find_next_devid(struct btrfs_fs_info *fs_info,
1614                                     u64 *devid_ret)
1615 {
1616         int ret;
1617         struct btrfs_key key;
1618         struct btrfs_key found_key;
1619         struct btrfs_path *path;
1620
1621         path = btrfs_alloc_path();
1622         if (!path)
1623                 return -ENOMEM;
1624
1625         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1626         key.type = BTRFS_DEV_ITEM_KEY;
1627         key.offset = (u64)-1;
1628
1629         ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
1630         if (ret < 0)
1631                 goto error;
1632
1633         BUG_ON(ret == 0); /* Corruption */
1634
1635         ret = btrfs_previous_item(fs_info->chunk_root, path,
1636                                   BTRFS_DEV_ITEMS_OBJECTID,
1637                                   BTRFS_DEV_ITEM_KEY);
1638         if (ret) {
1639                 *devid_ret = 1;
1640         } else {
1641                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1642                                       path->slots[0]);
1643                 *devid_ret = found_key.offset + 1;
1644         }
1645         ret = 0;
1646 error:
1647         btrfs_free_path(path);
1648         return ret;
1649 }
1650
1651 /*
1652  * the device information is stored in the chunk root
1653  * the btrfs_device struct should be fully filled in
1654  */
1655 static int btrfs_add_device(struct btrfs_trans_handle *trans,
1656                             struct btrfs_root *root,
1657                             struct btrfs_device *device)
1658 {
1659         int ret;
1660         struct btrfs_path *path;
1661         struct btrfs_dev_item *dev_item;
1662         struct extent_buffer *leaf;
1663         struct btrfs_key key;
1664         unsigned long ptr;
1665
1666         root = root->fs_info->chunk_root;
1667
1668         path = btrfs_alloc_path();
1669         if (!path)
1670                 return -ENOMEM;
1671
1672         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1673         key.type = BTRFS_DEV_ITEM_KEY;
1674         key.offset = device->devid;
1675
1676         ret = btrfs_insert_empty_item(trans, root, path, &key,
1677                                       sizeof(*dev_item));
1678         if (ret)
1679                 goto out;
1680
1681         leaf = path->nodes[0];
1682         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1683
1684         btrfs_set_device_id(leaf, dev_item, device->devid);
1685         btrfs_set_device_generation(leaf, dev_item, 0);
1686         btrfs_set_device_type(leaf, dev_item, device->type);
1687         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1688         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1689         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1690         btrfs_set_device_total_bytes(leaf, dev_item,
1691                                      btrfs_device_get_disk_total_bytes(device));
1692         btrfs_set_device_bytes_used(leaf, dev_item,
1693                                     btrfs_device_get_bytes_used(device));
1694         btrfs_set_device_group(leaf, dev_item, 0);
1695         btrfs_set_device_seek_speed(leaf, dev_item, 0);
1696         btrfs_set_device_bandwidth(leaf, dev_item, 0);
1697         btrfs_set_device_start_offset(leaf, dev_item, 0);
1698
1699         ptr = btrfs_device_uuid(dev_item);
1700         write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1701         ptr = btrfs_device_fsid(dev_item);
1702         write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1703         btrfs_mark_buffer_dirty(leaf);
1704
1705         ret = 0;
1706 out:
1707         btrfs_free_path(path);
1708         return ret;
1709 }
1710
1711 /*
1712  * Function to update ctime/mtime for a given device path.
1713  * Mainly used for ctime/mtime based probe like libblkid.
1714  */
1715 static void update_dev_time(char *path_name)
1716 {
1717         struct file *filp;
1718
1719         filp = filp_open(path_name, O_RDWR, 0);
1720         if (IS_ERR(filp))
1721                 return;
1722         file_update_time(filp);
1723         filp_close(filp, NULL);
1724 }
1725
1726 static int btrfs_rm_dev_item(struct btrfs_root *root,
1727                              struct btrfs_device *device)
1728 {
1729         int ret;
1730         struct btrfs_path *path;
1731         struct btrfs_key key;
1732         struct btrfs_trans_handle *trans;
1733
1734         root = root->fs_info->chunk_root;
1735
1736         path = btrfs_alloc_path();
1737         if (!path)
1738                 return -ENOMEM;
1739
1740         trans = btrfs_start_transaction(root, 0);
1741         if (IS_ERR(trans)) {
1742                 btrfs_free_path(path);
1743                 return PTR_ERR(trans);
1744         }
1745         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1746         key.type = BTRFS_DEV_ITEM_KEY;
1747         key.offset = device->devid;
1748
1749         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1750         if (ret < 0)
1751                 goto out;
1752
1753         if (ret > 0) {
1754                 ret = -ENOENT;
1755                 goto out;
1756         }
1757
1758         ret = btrfs_del_item(trans, root, path);
1759         if (ret)
1760                 goto out;
1761 out:
1762         btrfs_free_path(path);
1763         btrfs_commit_transaction(trans, root);
1764         return ret;
1765 }
1766
1767 /*
1768  * Verify that @num_devices satisfies the RAID profile constraints in the whole
1769  * filesystem. It's up to the caller to adjust that number regarding eg. device
1770  * replace.
1771  */
1772 static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info,
1773                 u64 num_devices)
1774 {
1775         u64 all_avail;
1776         unsigned seq;
1777         int i;
1778
1779         do {
1780                 seq = read_seqbegin(&fs_info->profiles_lock);
1781
1782                 all_avail = fs_info->avail_data_alloc_bits |
1783                             fs_info->avail_system_alloc_bits |
1784                             fs_info->avail_metadata_alloc_bits;
1785         } while (read_seqretry(&fs_info->profiles_lock, seq));
1786
1787         for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
1788                 if (!(all_avail & btrfs_raid_group[i]))
1789                         continue;
1790
1791                 if (num_devices < btrfs_raid_array[i].devs_min) {
1792                         int ret = btrfs_raid_mindev_error[i];
1793
1794                         if (ret)
1795                                 return ret;
1796                 }
1797         }
1798
1799         return 0;
1800 }
1801
1802 struct btrfs_device *btrfs_find_next_active_device(struct btrfs_fs_devices *fs_devs,
1803                                         struct btrfs_device *device)
1804 {
1805         struct btrfs_device *next_device;
1806
1807         list_for_each_entry(next_device, &fs_devs->devices, dev_list) {
1808                 if (next_device != device &&
1809                         !next_device->missing && next_device->bdev)
1810                         return next_device;
1811         }
1812
1813         return NULL;
1814 }
1815
1816 /*
1817  * Helper function to check if the given device is part of s_bdev / latest_bdev
1818  * and replace it with the provided or the next active device, in the context
1819  * where this function called, there should be always be another device (or
1820  * this_dev) which is active.
1821  */
1822 void btrfs_assign_next_active_device(struct btrfs_fs_info *fs_info,
1823                 struct btrfs_device *device, struct btrfs_device *this_dev)
1824 {
1825         struct btrfs_device *next_device;
1826
1827         if (this_dev)
1828                 next_device = this_dev;
1829         else
1830                 next_device = btrfs_find_next_active_device(fs_info->fs_devices,
1831                                                                 device);
1832         ASSERT(next_device);
1833
1834         if (fs_info->sb->s_bdev &&
1835                         (fs_info->sb->s_bdev == device->bdev))
1836                 fs_info->sb->s_bdev = next_device->bdev;
1837
1838         if (fs_info->fs_devices->latest_bdev == device->bdev)
1839                 fs_info->fs_devices->latest_bdev = next_device->bdev;
1840 }
1841
1842 int btrfs_rm_device(struct btrfs_root *root, char *device_path, u64 devid)
1843 {
1844         struct btrfs_device *device;
1845         struct btrfs_fs_devices *cur_devices;
1846         u64 num_devices;
1847         int ret = 0;
1848         bool clear_super = false;
1849         char *dev_name = NULL;
1850
1851         mutex_lock(&uuid_mutex);
1852
1853         num_devices = root->fs_info->fs_devices->num_devices;
1854         btrfs_dev_replace_lock(&root->fs_info->dev_replace, 0);
1855         if (btrfs_dev_replace_is_ongoing(&root->fs_info->dev_replace)) {
1856                 WARN_ON(num_devices < 1);
1857                 num_devices--;
1858         }
1859         btrfs_dev_replace_unlock(&root->fs_info->dev_replace, 0);
1860
1861         ret = btrfs_check_raid_min_devices(root->fs_info, num_devices - 1);
1862         if (ret)
1863                 goto out;
1864
1865         ret = btrfs_find_device_by_devspec(root, devid, device_path,
1866                                 &device);
1867         if (ret)
1868                 goto out;
1869
1870         if (device->is_tgtdev_for_dev_replace) {
1871                 ret = BTRFS_ERROR_DEV_TGT_REPLACE;
1872                 goto out;
1873         }
1874
1875         if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1876                 ret = BTRFS_ERROR_DEV_ONLY_WRITABLE;
1877                 goto out;
1878         }
1879
1880         if (device->writeable) {
1881                 lock_chunks(root);
1882                 list_del_init(&device->dev_alloc_list);
1883                 device->fs_devices->rw_devices--;
1884                 unlock_chunks(root);
1885                 dev_name = kstrdup(device->name->str, GFP_KERNEL);
1886                 if (!dev_name) {
1887                         ret = -ENOMEM;
1888                         goto error_undo;
1889                 }
1890                 clear_super = true;
1891         }
1892
1893         mutex_unlock(&uuid_mutex);
1894         ret = btrfs_shrink_device(device, 0);
1895         mutex_lock(&uuid_mutex);
1896         if (ret)
1897                 goto error_undo;
1898
1899         /*
1900          * TODO: the superblock still includes this device in its num_devices
1901          * counter although write_all_supers() is not locked out. This
1902          * could give a filesystem state which requires a degraded mount.
1903          */
1904         ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1905         if (ret)
1906                 goto error_undo;
1907
1908         device->in_fs_metadata = 0;
1909         btrfs_scrub_cancel_dev(root->fs_info, device);
1910
1911         /*
1912          * the device list mutex makes sure that we don't change
1913          * the device list while someone else is writing out all
1914          * the device supers. Whoever is writing all supers, should
1915          * lock the device list mutex before getting the number of
1916          * devices in the super block (super_copy). Conversely,
1917          * whoever updates the number of devices in the super block
1918          * (super_copy) should hold the device list mutex.
1919          */
1920
1921         cur_devices = device->fs_devices;
1922         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1923         list_del_rcu(&device->dev_list);
1924
1925         device->fs_devices->num_devices--;
1926         device->fs_devices->total_devices--;
1927
1928         if (device->missing)
1929                 device->fs_devices->missing_devices--;
1930
1931         btrfs_assign_next_active_device(root->fs_info, device, NULL);
1932
1933         if (device->bdev) {
1934                 device->fs_devices->open_devices--;
1935                 /* remove sysfs entry */
1936                 btrfs_sysfs_rm_device_link(root->fs_info->fs_devices, device);
1937         }
1938
1939         btrfs_close_bdev(device);
1940
1941         call_rcu(&device->rcu, free_device);
1942
1943         num_devices = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
1944         btrfs_set_super_num_devices(root->fs_info->super_copy, num_devices);
1945         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1946
1947         if (cur_devices->open_devices == 0) {
1948                 struct btrfs_fs_devices *fs_devices;
1949                 fs_devices = root->fs_info->fs_devices;
1950                 while (fs_devices) {
1951                         if (fs_devices->seed == cur_devices) {
1952                                 fs_devices->seed = cur_devices->seed;
1953                                 break;
1954                         }
1955                         fs_devices = fs_devices->seed;
1956                 }
1957                 cur_devices->seed = NULL;
1958                 __btrfs_close_devices(cur_devices);
1959                 free_fs_devices(cur_devices);
1960         }
1961
1962         root->fs_info->num_tolerated_disk_barrier_failures =
1963                 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
1964
1965         /*
1966          * at this point, the device is zero sized.  We want to
1967          * remove it from the devices list and zero out the old super
1968          */
1969         if (clear_super) {
1970                 struct block_device *bdev;
1971
1972                 bdev = blkdev_get_by_path(dev_name, FMODE_READ | FMODE_EXCL,
1973                                                 root->fs_info->bdev_holder);
1974                 if (!IS_ERR(bdev)) {
1975                         btrfs_scratch_superblocks(bdev, dev_name);
1976                         blkdev_put(bdev, FMODE_READ | FMODE_EXCL);
1977                 }
1978         }
1979
1980 out:
1981         kfree(dev_name);
1982
1983         mutex_unlock(&uuid_mutex);
1984         return ret;
1985
1986 error_undo:
1987         if (device->writeable) {
1988                 lock_chunks(root);
1989                 list_add(&device->dev_alloc_list,
1990                          &root->fs_info->fs_devices->alloc_list);
1991                 device->fs_devices->rw_devices++;
1992                 unlock_chunks(root);
1993         }
1994         goto out;
1995 }
1996
1997 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_fs_info *fs_info,
1998                                         struct btrfs_device *srcdev)
1999 {
2000         struct btrfs_fs_devices *fs_devices;
2001
2002         WARN_ON(!mutex_is_locked(&fs_info->fs_devices->device_list_mutex));
2003
2004         /*
2005          * in case of fs with no seed, srcdev->fs_devices will point
2006          * to fs_devices of fs_info. However when the dev being replaced is
2007          * a seed dev it will point to the seed's local fs_devices. In short
2008          * srcdev will have its correct fs_devices in both the cases.
2009          */
2010         fs_devices = srcdev->fs_devices;
2011
2012         list_del_rcu(&srcdev->dev_list);
2013         list_del_rcu(&srcdev->dev_alloc_list);
2014         fs_devices->num_devices--;
2015         if (srcdev->missing)
2016                 fs_devices->missing_devices--;
2017
2018         if (srcdev->writeable)
2019                 fs_devices->rw_devices--;
2020
2021         if (srcdev->bdev)
2022                 fs_devices->open_devices--;
2023 }
2024
2025 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_fs_info *fs_info,
2026                                       struct btrfs_device *srcdev)
2027 {
2028         struct btrfs_fs_devices *fs_devices = srcdev->fs_devices;
2029
2030         if (srcdev->writeable) {
2031                 /* zero out the old super if it is writable */
2032                 btrfs_scratch_superblocks(srcdev->bdev, srcdev->name->str);
2033         }
2034
2035         btrfs_close_bdev(srcdev);
2036
2037         call_rcu(&srcdev->rcu, free_device);
2038
2039         /*
2040          * unless fs_devices is seed fs, num_devices shouldn't go
2041          * zero
2042          */
2043         BUG_ON(!fs_devices->num_devices && !fs_devices->seeding);
2044
2045         /* if this is no devs we rather delete the fs_devices */
2046         if (!fs_devices->num_devices) {
2047                 struct btrfs_fs_devices *tmp_fs_devices;
2048
2049                 tmp_fs_devices = fs_info->fs_devices;
2050                 while (tmp_fs_devices) {
2051                         if (tmp_fs_devices->seed == fs_devices) {
2052                                 tmp_fs_devices->seed = fs_devices->seed;
2053                                 break;
2054                         }
2055                         tmp_fs_devices = tmp_fs_devices->seed;
2056                 }
2057                 fs_devices->seed = NULL;
2058                 __btrfs_close_devices(fs_devices);
2059                 free_fs_devices(fs_devices);
2060         }
2061 }
2062
2063 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
2064                                       struct btrfs_device *tgtdev)
2065 {
2066         mutex_lock(&uuid_mutex);
2067         WARN_ON(!tgtdev);
2068         mutex_lock(&fs_info->fs_devices->device_list_mutex);
2069
2070         btrfs_sysfs_rm_device_link(fs_info->fs_devices, tgtdev);
2071
2072         if (tgtdev->bdev)
2073                 fs_info->fs_devices->open_devices--;
2074
2075         fs_info->fs_devices->num_devices--;
2076
2077         btrfs_assign_next_active_device(fs_info, tgtdev, NULL);
2078
2079         list_del_rcu(&tgtdev->dev_list);
2080
2081         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2082         mutex_unlock(&uuid_mutex);
2083
2084         /*
2085          * The update_dev_time() with in btrfs_scratch_superblocks()
2086          * may lead to a call to btrfs_show_devname() which will try
2087          * to hold device_list_mutex. And here this device
2088          * is already out of device list, so we don't have to hold
2089          * the device_list_mutex lock.
2090          */
2091         btrfs_scratch_superblocks(tgtdev->bdev, tgtdev->name->str);
2092
2093         btrfs_close_bdev(tgtdev);
2094         call_rcu(&tgtdev->rcu, free_device);
2095 }
2096
2097 static int btrfs_find_device_by_path(struct btrfs_root *root, char *device_path,
2098                                      struct btrfs_device **device)
2099 {
2100         int ret = 0;
2101         struct btrfs_super_block *disk_super;
2102         u64 devid;
2103         u8 *dev_uuid;
2104         struct block_device *bdev;
2105         struct buffer_head *bh;
2106
2107         *device = NULL;
2108         ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
2109                                     root->fs_info->bdev_holder, 0, &bdev, &bh);
2110         if (ret)
2111                 return ret;
2112         disk_super = (struct btrfs_super_block *)bh->b_data;
2113         devid = btrfs_stack_device_id(&disk_super->dev_item);
2114         dev_uuid = disk_super->dev_item.uuid;
2115         *device = btrfs_find_device(root->fs_info, devid, dev_uuid,
2116                                     disk_super->fsid);
2117         brelse(bh);
2118         if (!*device)
2119                 ret = -ENOENT;
2120         blkdev_put(bdev, FMODE_READ);
2121         return ret;
2122 }
2123
2124 int btrfs_find_device_missing_or_by_path(struct btrfs_root *root,
2125                                          char *device_path,
2126                                          struct btrfs_device **device)
2127 {
2128         *device = NULL;
2129         if (strcmp(device_path, "missing") == 0) {
2130                 struct list_head *devices;
2131                 struct btrfs_device *tmp;
2132
2133                 devices = &root->fs_info->fs_devices->devices;
2134                 /*
2135                  * It is safe to read the devices since the volume_mutex
2136                  * is held by the caller.
2137                  */
2138                 list_for_each_entry(tmp, devices, dev_list) {
2139                         if (tmp->in_fs_metadata && !tmp->bdev) {
2140                                 *device = tmp;
2141                                 break;
2142                         }
2143                 }
2144
2145                 if (!*device)
2146                         return BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
2147
2148                 return 0;
2149         } else {
2150                 return btrfs_find_device_by_path(root, device_path, device);
2151         }
2152 }
2153
2154 /*
2155  * Lookup a device given by device id, or the path if the id is 0.
2156  */
2157 int btrfs_find_device_by_devspec(struct btrfs_root *root, u64 devid,
2158                                          char *devpath,
2159                                          struct btrfs_device **device)
2160 {
2161         int ret;
2162
2163         if (devid) {
2164                 ret = 0;
2165                 *device = btrfs_find_device(root->fs_info, devid, NULL,
2166                                             NULL);
2167                 if (!*device)
2168                         ret = -ENOENT;
2169         } else {
2170                 if (!devpath || !devpath[0])
2171                         return -EINVAL;
2172
2173                 ret = btrfs_find_device_missing_or_by_path(root, devpath,
2174                                                            device);
2175         }
2176         return ret;
2177 }
2178
2179 /*
2180  * does all the dirty work required for changing file system's UUID.
2181  */
2182 static int btrfs_prepare_sprout(struct btrfs_root *root)
2183 {
2184         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2185         struct btrfs_fs_devices *old_devices;
2186         struct btrfs_fs_devices *seed_devices;
2187         struct btrfs_super_block *disk_super = root->fs_info->super_copy;
2188         struct btrfs_device *device;
2189         u64 super_flags;
2190
2191         BUG_ON(!mutex_is_locked(&uuid_mutex));
2192         if (!fs_devices->seeding)
2193                 return -EINVAL;
2194
2195         seed_devices = __alloc_fs_devices();
2196         if (IS_ERR(seed_devices))
2197                 return PTR_ERR(seed_devices);
2198
2199         old_devices = clone_fs_devices(fs_devices);
2200         if (IS_ERR(old_devices)) {
2201                 kfree(seed_devices);
2202                 return PTR_ERR(old_devices);
2203         }
2204
2205         list_add(&old_devices->list, &fs_uuids);
2206
2207         memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
2208         seed_devices->opened = 1;
2209         INIT_LIST_HEAD(&seed_devices->devices);
2210         INIT_LIST_HEAD(&seed_devices->alloc_list);
2211         mutex_init(&seed_devices->device_list_mutex);
2212
2213         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2214         list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
2215                               synchronize_rcu);
2216         list_for_each_entry(device, &seed_devices->devices, dev_list)
2217                 device->fs_devices = seed_devices;
2218
2219         lock_chunks(root);
2220         list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
2221         unlock_chunks(root);
2222
2223         fs_devices->seeding = 0;
2224         fs_devices->num_devices = 0;
2225         fs_devices->open_devices = 0;
2226         fs_devices->missing_devices = 0;
2227         fs_devices->rotating = 0;
2228         fs_devices->seed = seed_devices;
2229
2230         generate_random_uuid(fs_devices->fsid);
2231         memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2232         memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2233         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2234
2235         super_flags = btrfs_super_flags(disk_super) &
2236                       ~BTRFS_SUPER_FLAG_SEEDING;
2237         btrfs_set_super_flags(disk_super, super_flags);
2238
2239         return 0;
2240 }
2241
2242 /*
2243  * Store the expected generation for seed devices in device items.
2244  */
2245 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
2246                                struct btrfs_root *root)
2247 {
2248         struct btrfs_path *path;
2249         struct extent_buffer *leaf;
2250         struct btrfs_dev_item *dev_item;
2251         struct btrfs_device *device;
2252         struct btrfs_key key;
2253         u8 fs_uuid[BTRFS_UUID_SIZE];
2254         u8 dev_uuid[BTRFS_UUID_SIZE];
2255         u64 devid;
2256         int ret;
2257
2258         path = btrfs_alloc_path();
2259         if (!path)
2260                 return -ENOMEM;
2261
2262         root = root->fs_info->chunk_root;
2263         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2264         key.offset = 0;
2265         key.type = BTRFS_DEV_ITEM_KEY;
2266
2267         while (1) {
2268                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2269                 if (ret < 0)
2270                         goto error;
2271
2272                 leaf = path->nodes[0];
2273 next_slot:
2274                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2275                         ret = btrfs_next_leaf(root, path);
2276                         if (ret > 0)
2277                                 break;
2278                         if (ret < 0)
2279                                 goto error;
2280                         leaf = path->nodes[0];
2281                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2282                         btrfs_release_path(path);
2283                         continue;
2284                 }
2285
2286                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2287                 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
2288                     key.type != BTRFS_DEV_ITEM_KEY)
2289                         break;
2290
2291                 dev_item = btrfs_item_ptr(leaf, path->slots[0],
2292                                           struct btrfs_dev_item);
2293                 devid = btrfs_device_id(leaf, dev_item);
2294                 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
2295                                    BTRFS_UUID_SIZE);
2296                 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
2297                                    BTRFS_UUID_SIZE);
2298                 device = btrfs_find_device(root->fs_info, devid, dev_uuid,
2299                                            fs_uuid);
2300                 BUG_ON(!device); /* Logic error */
2301
2302                 if (device->fs_devices->seeding) {
2303                         btrfs_set_device_generation(leaf, dev_item,
2304                                                     device->generation);
2305                         btrfs_mark_buffer_dirty(leaf);
2306                 }
2307
2308                 path->slots[0]++;
2309                 goto next_slot;
2310         }
2311         ret = 0;
2312 error:
2313         btrfs_free_path(path);
2314         return ret;
2315 }
2316
2317 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
2318 {
2319         struct request_queue *q;
2320         struct btrfs_trans_handle *trans;
2321         struct btrfs_device *device;
2322         struct block_device *bdev;
2323         struct list_head *devices;
2324         struct super_block *sb = root->fs_info->sb;
2325         struct rcu_string *name;
2326         u64 tmp;
2327         int seeding_dev = 0;
2328         int ret = 0;
2329
2330         if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
2331                 return -EROFS;
2332
2333         bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2334                                   root->fs_info->bdev_holder);
2335         if (IS_ERR(bdev))
2336                 return PTR_ERR(bdev);
2337
2338         if (root->fs_info->fs_devices->seeding) {
2339                 seeding_dev = 1;
2340                 down_write(&sb->s_umount);
2341                 mutex_lock(&uuid_mutex);
2342         }
2343
2344         filemap_write_and_wait(bdev->bd_inode->i_mapping);
2345
2346         devices = &root->fs_info->fs_devices->devices;
2347
2348         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2349         list_for_each_entry(device, devices, dev_list) {
2350                 if (device->bdev == bdev) {
2351                         ret = -EEXIST;
2352                         mutex_unlock(
2353                                 &root->fs_info->fs_devices->device_list_mutex);
2354                         goto error;
2355                 }
2356         }
2357         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2358
2359         device = btrfs_alloc_device(root->fs_info, NULL, NULL);
2360         if (IS_ERR(device)) {
2361                 /* we can safely leave the fs_devices entry around */
2362                 ret = PTR_ERR(device);
2363                 goto error;
2364         }
2365
2366         name = rcu_string_strdup(device_path, GFP_KERNEL);
2367         if (!name) {
2368                 kfree(device);
2369                 ret = -ENOMEM;
2370                 goto error;
2371         }
2372         rcu_assign_pointer(device->name, name);
2373
2374         trans = btrfs_start_transaction(root, 0);
2375         if (IS_ERR(trans)) {
2376                 rcu_string_free(device->name);
2377                 kfree(device);
2378                 ret = PTR_ERR(trans);
2379                 goto error;
2380         }
2381
2382         q = bdev_get_queue(bdev);
2383         if (blk_queue_discard(q))
2384                 device->can_discard = 1;
2385         device->writeable = 1;
2386         device->generation = trans->transid;
2387         device->io_width = root->sectorsize;
2388         device->io_align = root->sectorsize;
2389         device->sector_size = root->sectorsize;
2390         device->total_bytes = i_size_read(bdev->bd_inode);
2391         device->disk_total_bytes = device->total_bytes;
2392         device->commit_total_bytes = device->total_bytes;
2393         device->dev_root = root->fs_info->dev_root;
2394         device->bdev = bdev;
2395         device->in_fs_metadata = 1;
2396         device->is_tgtdev_for_dev_replace = 0;
2397         device->mode = FMODE_EXCL;
2398         device->dev_stats_valid = 1;
2399         set_blocksize(device->bdev, 4096);
2400
2401         if (seeding_dev) {
2402                 sb->s_flags &= ~MS_RDONLY;
2403                 ret = btrfs_prepare_sprout(root);
2404                 BUG_ON(ret); /* -ENOMEM */
2405         }
2406
2407         device->fs_devices = root->fs_info->fs_devices;
2408
2409         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2410         lock_chunks(root);
2411         list_add_rcu(&device->dev_list, &root->fs_info->fs_devices->devices);
2412         list_add(&device->dev_alloc_list,
2413                  &root->fs_info->fs_devices->alloc_list);
2414         root->fs_info->fs_devices->num_devices++;
2415         root->fs_info->fs_devices->open_devices++;
2416         root->fs_info->fs_devices->rw_devices++;
2417         root->fs_info->fs_devices->total_devices++;
2418         root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
2419
2420         spin_lock(&root->fs_info->free_chunk_lock);
2421         root->fs_info->free_chunk_space += device->total_bytes;
2422         spin_unlock(&root->fs_info->free_chunk_lock);
2423
2424         if (!blk_queue_nonrot(bdev_get_queue(bdev)))
2425                 root->fs_info->fs_devices->rotating = 1;
2426
2427         tmp = btrfs_super_total_bytes(root->fs_info->super_copy);
2428         btrfs_set_super_total_bytes(root->fs_info->super_copy,
2429                                     tmp + device->total_bytes);
2430
2431         tmp = btrfs_super_num_devices(root->fs_info->super_copy);
2432         btrfs_set_super_num_devices(root->fs_info->super_copy,
2433                                     tmp + 1);
2434
2435         /* add sysfs device entry */
2436         btrfs_sysfs_add_device_link(root->fs_info->fs_devices, device);
2437
2438         /*
2439          * we've got more storage, clear any full flags on the space
2440          * infos
2441          */
2442         btrfs_clear_space_info_full(root->fs_info);
2443
2444         unlock_chunks(root);
2445         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2446
2447         if (seeding_dev) {
2448                 lock_chunks(root);
2449                 ret = init_first_rw_device(trans, root, device);
2450                 unlock_chunks(root);
2451                 if (ret) {
2452                         btrfs_abort_transaction(trans, ret);
2453                         goto error_trans;
2454                 }
2455         }
2456
2457         ret = btrfs_add_device(trans, root, device);
2458         if (ret) {
2459                 btrfs_abort_transaction(trans, ret);
2460                 goto error_trans;
2461         }
2462
2463         if (seeding_dev) {
2464                 char fsid_buf[BTRFS_UUID_UNPARSED_SIZE];
2465
2466                 ret = btrfs_finish_sprout(trans, root);
2467                 if (ret) {
2468                         btrfs_abort_transaction(trans, ret);
2469                         goto error_trans;
2470                 }
2471
2472                 /* Sprouting would change fsid of the mounted root,
2473                  * so rename the fsid on the sysfs
2474                  */
2475                 snprintf(fsid_buf, BTRFS_UUID_UNPARSED_SIZE, "%pU",
2476                                                 root->fs_info->fsid);
2477                 if (kobject_rename(&root->fs_info->fs_devices->fsid_kobj,
2478                                                                 fsid_buf))
2479                         btrfs_warn(root->fs_info,
2480                                 "sysfs: failed to create fsid for sprout");
2481         }
2482
2483         root->fs_info->num_tolerated_disk_barrier_failures =
2484                 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
2485         ret = btrfs_commit_transaction(trans, root);
2486
2487         if (seeding_dev) {
2488                 mutex_unlock(&uuid_mutex);
2489                 up_write(&sb->s_umount);
2490
2491                 if (ret) /* transaction commit */
2492                         return ret;
2493
2494                 ret = btrfs_relocate_sys_chunks(root);
2495                 if (ret < 0)
2496                         btrfs_handle_fs_error(root->fs_info, ret,
2497                                     "Failed to relocate sys chunks after "
2498                                     "device initialization. This can be fixed "
2499                                     "using the \"btrfs balance\" command.");
2500                 trans = btrfs_attach_transaction(root);
2501                 if (IS_ERR(trans)) {
2502                         if (PTR_ERR(trans) == -ENOENT)
2503                                 return 0;
2504                         return PTR_ERR(trans);
2505                 }
2506                 ret = btrfs_commit_transaction(trans, root);
2507         }
2508
2509         /* Update ctime/mtime for libblkid */
2510         update_dev_time(device_path);
2511         return ret;
2512
2513 error_trans:
2514         btrfs_end_transaction(trans, root);
2515         rcu_string_free(device->name);
2516         btrfs_sysfs_rm_device_link(root->fs_info->fs_devices, device);
2517         kfree(device);
2518 error:
2519         blkdev_put(bdev, FMODE_EXCL);
2520         if (seeding_dev) {
2521                 mutex_unlock(&uuid_mutex);
2522                 up_write(&sb->s_umount);
2523         }
2524         return ret;
2525 }
2526
2527 int btrfs_init_dev_replace_tgtdev(struct btrfs_root *root, char *device_path,
2528                                   struct btrfs_device *srcdev,
2529                                   struct btrfs_device **device_out)
2530 {
2531         struct request_queue *q;
2532         struct btrfs_device *device;
2533         struct block_device *bdev;
2534         struct btrfs_fs_info *fs_info = root->fs_info;
2535         struct list_head *devices;
2536         struct rcu_string *name;
2537         u64 devid = BTRFS_DEV_REPLACE_DEVID;
2538         int ret = 0;
2539
2540         *device_out = NULL;
2541         if (fs_info->fs_devices->seeding) {
2542                 btrfs_err(fs_info, "the filesystem is a seed filesystem!");
2543                 return -EINVAL;
2544         }
2545
2546         bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2547                                   fs_info->bdev_holder);
2548         if (IS_ERR(bdev)) {
2549                 btrfs_err(fs_info, "target device %s is invalid!", device_path);
2550                 return PTR_ERR(bdev);
2551         }
2552
2553         filemap_write_and_wait(bdev->bd_inode->i_mapping);
2554
2555         devices = &fs_info->fs_devices->devices;
2556         list_for_each_entry(device, devices, dev_list) {
2557                 if (device->bdev == bdev) {
2558                         btrfs_err(fs_info, "target device is in the filesystem!");
2559                         ret = -EEXIST;
2560                         goto error;
2561                 }
2562         }
2563
2564
2565         if (i_size_read(bdev->bd_inode) <
2566             btrfs_device_get_total_bytes(srcdev)) {
2567                 btrfs_err(fs_info, "target device is smaller than source device!");
2568                 ret = -EINVAL;
2569                 goto error;
2570         }
2571
2572
2573         device = btrfs_alloc_device(NULL, &devid, NULL);
2574         if (IS_ERR(device)) {
2575                 ret = PTR_ERR(device);
2576                 goto error;
2577         }
2578
2579         name = rcu_string_strdup(device_path, GFP_NOFS);
2580         if (!name) {
2581                 kfree(device);
2582                 ret = -ENOMEM;
2583                 goto error;
2584         }
2585         rcu_assign_pointer(device->name, name);
2586
2587         q = bdev_get_queue(bdev);
2588         if (blk_queue_discard(q))
2589                 device->can_discard = 1;
2590         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2591         device->writeable = 1;
2592         device->generation = 0;
2593         device->io_width = root->sectorsize;
2594         device->io_align = root->sectorsize;
2595         device->sector_size = root->sectorsize;
2596         device->total_bytes = btrfs_device_get_total_bytes(srcdev);
2597         device->disk_total_bytes = btrfs_device_get_disk_total_bytes(srcdev);
2598         device->bytes_used = btrfs_device_get_bytes_used(srcdev);
2599         ASSERT(list_empty(&srcdev->resized_list));
2600         device->commit_total_bytes = srcdev->commit_total_bytes;
2601         device->commit_bytes_used = device->bytes_used;
2602         device->dev_root = fs_info->dev_root;
2603         device->bdev = bdev;
2604         device->in_fs_metadata = 1;
2605         device->is_tgtdev_for_dev_replace = 1;
2606         device->mode = FMODE_EXCL;
2607         device->dev_stats_valid = 1;
2608         set_blocksize(device->bdev, 4096);
2609         device->fs_devices = fs_info->fs_devices;
2610         list_add(&device->dev_list, &fs_info->fs_devices->devices);
2611         fs_info->fs_devices->num_devices++;
2612         fs_info->fs_devices->open_devices++;
2613         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2614
2615         *device_out = device;
2616         return ret;
2617
2618 error:
2619         blkdev_put(bdev, FMODE_EXCL);
2620         return ret;
2621 }
2622
2623 void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info *fs_info,
2624                                               struct btrfs_device *tgtdev)
2625 {
2626         WARN_ON(fs_info->fs_devices->rw_devices == 0);
2627         tgtdev->io_width = fs_info->dev_root->sectorsize;
2628         tgtdev->io_align = fs_info->dev_root->sectorsize;
2629         tgtdev->sector_size = fs_info->dev_root->sectorsize;
2630         tgtdev->dev_root = fs_info->dev_root;
2631         tgtdev->in_fs_metadata = 1;
2632 }
2633
2634 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2635                                         struct btrfs_device *device)
2636 {
2637         int ret;
2638         struct btrfs_path *path;
2639         struct btrfs_root *root;
2640         struct btrfs_dev_item *dev_item;
2641         struct extent_buffer *leaf;
2642         struct btrfs_key key;
2643
2644         root = device->dev_root->fs_info->chunk_root;
2645
2646         path = btrfs_alloc_path();
2647         if (!path)
2648                 return -ENOMEM;
2649
2650         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2651         key.type = BTRFS_DEV_ITEM_KEY;
2652         key.offset = device->devid;
2653
2654         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2655         if (ret < 0)
2656                 goto out;
2657
2658         if (ret > 0) {
2659                 ret = -ENOENT;
2660                 goto out;
2661         }
2662
2663         leaf = path->nodes[0];
2664         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2665
2666         btrfs_set_device_id(leaf, dev_item, device->devid);
2667         btrfs_set_device_type(leaf, dev_item, device->type);
2668         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2669         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2670         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2671         btrfs_set_device_total_bytes(leaf, dev_item,
2672                                      btrfs_device_get_disk_total_bytes(device));
2673         btrfs_set_device_bytes_used(leaf, dev_item,
2674                                     btrfs_device_get_bytes_used(device));
2675         btrfs_mark_buffer_dirty(leaf);
2676
2677 out:
2678         btrfs_free_path(path);
2679         return ret;
2680 }
2681
2682 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2683                       struct btrfs_device *device, u64 new_size)
2684 {
2685         struct btrfs_super_block *super_copy =
2686                 device->dev_root->fs_info->super_copy;
2687         struct btrfs_fs_devices *fs_devices;
2688         u64 old_total;
2689         u64 diff;
2690
2691         if (!device->writeable)
2692                 return -EACCES;
2693
2694         lock_chunks(device->dev_root);
2695         old_total = btrfs_super_total_bytes(super_copy);
2696         diff = new_size - device->total_bytes;
2697
2698         if (new_size <= device->total_bytes ||
2699             device->is_tgtdev_for_dev_replace) {
2700                 unlock_chunks(device->dev_root);
2701                 return -EINVAL;
2702         }
2703
2704         fs_devices = device->dev_root->fs_info->fs_devices;
2705
2706         btrfs_set_super_total_bytes(super_copy, old_total + diff);
2707         device->fs_devices->total_rw_bytes += diff;
2708
2709         btrfs_device_set_total_bytes(device, new_size);
2710         btrfs_device_set_disk_total_bytes(device, new_size);
2711         btrfs_clear_space_info_full(device->dev_root->fs_info);
2712         if (list_empty(&device->resized_list))
2713                 list_add_tail(&device->resized_list,
2714                               &fs_devices->resized_devices);
2715         unlock_chunks(device->dev_root);
2716
2717         return btrfs_update_device(trans, device);
2718 }
2719
2720 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
2721                             struct btrfs_root *root, u64 chunk_objectid,
2722                             u64 chunk_offset)
2723 {
2724         int ret;
2725         struct btrfs_path *path;
2726         struct btrfs_key key;
2727
2728         root = root->fs_info->chunk_root;
2729         path = btrfs_alloc_path();
2730         if (!path)
2731                 return -ENOMEM;
2732
2733         key.objectid = chunk_objectid;
2734         key.offset = chunk_offset;
2735         key.type = BTRFS_CHUNK_ITEM_KEY;
2736
2737         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2738         if (ret < 0)
2739                 goto out;
2740         else if (ret > 0) { /* Logic error or corruption */
2741                 btrfs_handle_fs_error(root->fs_info, -ENOENT,
2742                             "Failed lookup while freeing chunk.");
2743                 ret = -ENOENT;
2744                 goto out;
2745         }
2746
2747         ret = btrfs_del_item(trans, root, path);
2748         if (ret < 0)
2749                 btrfs_handle_fs_error(root->fs_info, ret,
2750                             "Failed to delete chunk item.");
2751 out:
2752         btrfs_free_path(path);
2753         return ret;
2754 }
2755
2756 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
2757                         chunk_offset)
2758 {
2759         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
2760         struct btrfs_disk_key *disk_key;
2761         struct btrfs_chunk *chunk;
2762         u8 *ptr;
2763         int ret = 0;
2764         u32 num_stripes;
2765         u32 array_size;
2766         u32 len = 0;
2767         u32 cur;
2768         struct btrfs_key key;
2769
2770         lock_chunks(root);
2771         array_size = btrfs_super_sys_array_size(super_copy);
2772
2773         ptr = super_copy->sys_chunk_array;
2774         cur = 0;
2775
2776         while (cur < array_size) {
2777                 disk_key = (struct btrfs_disk_key *)ptr;
2778                 btrfs_disk_key_to_cpu(&key, disk_key);
2779
2780                 len = sizeof(*disk_key);
2781
2782                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2783                         chunk = (struct btrfs_chunk *)(ptr + len);
2784                         num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2785                         len += btrfs_chunk_item_size(num_stripes);
2786                 } else {
2787                         ret = -EIO;
2788                         break;
2789                 }
2790                 if (key.objectid == chunk_objectid &&
2791                     key.offset == chunk_offset) {
2792                         memmove(ptr, ptr + len, array_size - (cur + len));
2793                         array_size -= len;
2794                         btrfs_set_super_sys_array_size(super_copy, array_size);
2795                 } else {
2796                         ptr += len;
2797                         cur += len;
2798                 }
2799         }
2800         unlock_chunks(root);
2801         return ret;
2802 }
2803
2804 int btrfs_remove_chunk(struct btrfs_trans_handle *trans,
2805                        struct btrfs_root *root, u64 chunk_offset)
2806 {
2807         struct extent_map_tree *em_tree;
2808         struct extent_map *em;
2809         struct btrfs_root *extent_root = root->fs_info->extent_root;
2810         struct map_lookup *map;
2811         u64 dev_extent_len = 0;
2812         u64 chunk_objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2813         int i, ret = 0;
2814         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2815
2816         /* Just in case */
2817         root = root->fs_info->chunk_root;
2818         em_tree = &root->fs_info->mapping_tree.map_tree;
2819
2820         read_lock(&em_tree->lock);
2821         em = lookup_extent_mapping(em_tree, chunk_offset, 1);
2822         read_unlock(&em_tree->lock);
2823
2824         if (!em || em->start > chunk_offset ||
2825             em->start + em->len < chunk_offset) {
2826                 /*
2827                  * This is a logic error, but we don't want to just rely on the
2828                  * user having built with ASSERT enabled, so if ASSERT doesn't
2829                  * do anything we still error out.
2830                  */
2831                 ASSERT(0);
2832                 if (em)
2833                         free_extent_map(em);
2834                 return -EINVAL;
2835         }
2836         map = em->map_lookup;
2837         lock_chunks(root->fs_info->chunk_root);
2838         check_system_chunk(trans, extent_root, map->type);
2839         unlock_chunks(root->fs_info->chunk_root);
2840
2841         /*
2842          * Take the device list mutex to prevent races with the final phase of
2843          * a device replace operation that replaces the device object associated
2844          * with map stripes (dev-replace.c:btrfs_dev_replace_finishing()).
2845          */
2846         mutex_lock(&fs_devices->device_list_mutex);
2847         for (i = 0; i < map->num_stripes; i++) {
2848                 struct btrfs_device *device = map->stripes[i].dev;
2849                 ret = btrfs_free_dev_extent(trans, device,
2850                                             map->stripes[i].physical,
2851                                             &dev_extent_len);
2852                 if (ret) {
2853                         mutex_unlock(&fs_devices->device_list_mutex);
2854                         btrfs_abort_transaction(trans, ret);
2855                         goto out;
2856                 }
2857
2858                 if (device->bytes_used > 0) {
2859                         lock_chunks(root);
2860                         btrfs_device_set_bytes_used(device,
2861                                         device->bytes_used - dev_extent_len);
2862                         spin_lock(&root->fs_info->free_chunk_lock);
2863                         root->fs_info->free_chunk_space += dev_extent_len;
2864                         spin_unlock(&root->fs_info->free_chunk_lock);
2865                         btrfs_clear_space_info_full(root->fs_info);
2866                         unlock_chunks(root);
2867                 }
2868
2869                 if (map->stripes[i].dev) {
2870                         ret = btrfs_update_device(trans, map->stripes[i].dev);
2871                         if (ret) {
2872                                 mutex_unlock(&fs_devices->device_list_mutex);
2873                                 btrfs_abort_transaction(trans, ret);
2874                                 goto out;
2875                         }
2876                 }
2877         }
2878         mutex_unlock(&fs_devices->device_list_mutex);
2879
2880         ret = btrfs_free_chunk(trans, root, chunk_objectid, chunk_offset);
2881         if (ret) {
2882                 btrfs_abort_transaction(trans, ret);
2883                 goto out;
2884         }
2885
2886         trace_btrfs_chunk_free(root, map, chunk_offset, em->len);
2887
2888         if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2889                 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
2890                 if (ret) {
2891                         btrfs_abort_transaction(trans, ret);
2892                         goto out;
2893                 }
2894         }
2895
2896         ret = btrfs_remove_block_group(trans, extent_root, chunk_offset, em);
2897         if (ret) {
2898                 btrfs_abort_transaction(trans, ret);
2899                 goto out;
2900         }
2901
2902 out:
2903         /* once for us */
2904         free_extent_map(em);
2905         return ret;
2906 }
2907
2908 static int btrfs_relocate_chunk(struct btrfs_root *root, u64 chunk_offset)
2909 {
2910         struct btrfs_root *extent_root;
2911         struct btrfs_trans_handle *trans;
2912         int ret;
2913
2914         root = root->fs_info->chunk_root;
2915         extent_root = root->fs_info->extent_root;
2916
2917         /*
2918          * Prevent races with automatic removal of unused block groups.
2919          * After we relocate and before we remove the chunk with offset
2920          * chunk_offset, automatic removal of the block group can kick in,
2921          * resulting in a failure when calling btrfs_remove_chunk() below.
2922          *
2923          * Make sure to acquire this mutex before doing a tree search (dev
2924          * or chunk trees) to find chunks. Otherwise the cleaner kthread might
2925          * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after
2926          * we release the path used to search the chunk/dev tree and before
2927          * the current task acquires this mutex and calls us.
2928          */
2929         ASSERT(mutex_is_locked(&root->fs_info->delete_unused_bgs_mutex));
2930
2931         ret = btrfs_can_relocate(extent_root, chunk_offset);
2932         if (ret)
2933                 return -ENOSPC;
2934
2935         /* step one, relocate all the extents inside this chunk */
2936         btrfs_scrub_pause(root);
2937         ret = btrfs_relocate_block_group(extent_root, chunk_offset);
2938         btrfs_scrub_continue(root);
2939         if (ret)
2940                 return ret;
2941
2942         trans = btrfs_start_trans_remove_block_group(root->fs_info,
2943                                                      chunk_offset);
2944         if (IS_ERR(trans)) {
2945                 ret = PTR_ERR(trans);
2946                 btrfs_handle_fs_error(root->fs_info, ret, NULL);
2947                 return ret;
2948         }
2949
2950         /*
2951          * step two, delete the device extents and the
2952          * chunk tree entries
2953          */
2954         ret = btrfs_remove_chunk(trans, root, chunk_offset);
2955         btrfs_end_transaction(trans, extent_root);
2956         return ret;
2957 }
2958
2959 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
2960 {
2961         struct btrfs_root *chunk_root = root->fs_info->chunk_root;
2962         struct btrfs_path *path;
2963         struct extent_buffer *leaf;
2964         struct btrfs_chunk *chunk;
2965         struct btrfs_key key;
2966         struct btrfs_key found_key;
2967         u64 chunk_type;
2968         bool retried = false;
2969         int failed = 0;
2970         int ret;
2971
2972         path = btrfs_alloc_path();
2973         if (!path)
2974                 return -ENOMEM;
2975
2976 again:
2977         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2978         key.offset = (u64)-1;
2979         key.type = BTRFS_CHUNK_ITEM_KEY;
2980
2981         while (1) {
2982                 mutex_lock(&root->fs_info->delete_unused_bgs_mutex);
2983                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2984                 if (ret < 0) {
2985                         mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
2986                         goto error;
2987                 }
2988                 BUG_ON(ret == 0); /* Corruption */
2989
2990                 ret = btrfs_previous_item(chunk_root, path, key.objectid,
2991                                           key.type);
2992                 if (ret)
2993                         mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
2994                 if (ret < 0)
2995                         goto error;
2996                 if (ret > 0)
2997                         break;
2998
2999                 leaf = path->nodes[0];
3000                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3001
3002                 chunk = btrfs_item_ptr(leaf, path->slots[0],
3003                                        struct btrfs_chunk);
3004                 chunk_type = btrfs_chunk_type(leaf, chunk);
3005                 btrfs_release_path(path);
3006
3007                 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
3008                         ret = btrfs_relocate_chunk(chunk_root,
3009                                                    found_key.offset);
3010                         if (ret == -ENOSPC)
3011                                 failed++;
3012                         else
3013                                 BUG_ON(ret);
3014                 }
3015                 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
3016
3017                 if (found_key.offset == 0)
3018                         break;
3019                 key.offset = found_key.offset - 1;
3020         }
3021         ret = 0;
3022         if (failed && !retried) {
3023                 failed = 0;
3024                 retried = true;
3025                 goto again;
3026         } else if (WARN_ON(failed && retried)) {
3027                 ret = -ENOSPC;
3028         }
3029 error:
3030         btrfs_free_path(path);
3031         return ret;
3032 }
3033
3034 static int insert_balance_item(struct btrfs_root *root,
3035                                struct btrfs_balance_control *bctl)
3036 {
3037         struct btrfs_trans_handle *trans;
3038         struct btrfs_balance_item *item;
3039         struct btrfs_disk_balance_args disk_bargs;
3040         struct btrfs_path *path;
3041         struct extent_buffer *leaf;
3042         struct btrfs_key key;
3043         int ret, err;
3044
3045         path = btrfs_alloc_path();
3046         if (!path)
3047                 return -ENOMEM;
3048
3049         trans = btrfs_start_transaction(root, 0);
3050         if (IS_ERR(trans)) {
3051                 btrfs_free_path(path);
3052                 return PTR_ERR(trans);
3053         }
3054
3055         key.objectid = BTRFS_BALANCE_OBJECTID;
3056         key.type = BTRFS_TEMPORARY_ITEM_KEY;
3057         key.offset = 0;
3058
3059         ret = btrfs_insert_empty_item(trans, root, path, &key,
3060                                       sizeof(*item));
3061         if (ret)
3062                 goto out;
3063
3064         leaf = path->nodes[0];
3065         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3066
3067         memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
3068
3069         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
3070         btrfs_set_balance_data(leaf, item, &disk_bargs);
3071         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
3072         btrfs_set_balance_meta(leaf, item, &disk_bargs);
3073         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
3074         btrfs_set_balance_sys(leaf, item, &disk_bargs);
3075
3076         btrfs_set_balance_flags(leaf, item, bctl->flags);
3077
3078         btrfs_mark_buffer_dirty(leaf);
3079 out:
3080         btrfs_free_path(path);
3081         err = btrfs_commit_transaction(trans, root);
3082         if (err && !ret)
3083                 ret = err;
3084         return ret;
3085 }
3086
3087 static int del_balance_item(struct btrfs_root *root)
3088 {
3089         struct btrfs_trans_handle *trans;
3090         struct btrfs_path *path;
3091         struct btrfs_key key;
3092         int ret, err;
3093
3094         path = btrfs_alloc_path();
3095         if (!path)
3096                 return -ENOMEM;
3097
3098         trans = btrfs_start_transaction(root, 0);
3099         if (IS_ERR(trans)) {
3100                 btrfs_free_path(path);
3101                 return PTR_ERR(trans);
3102         }
3103
3104         key.objectid = BTRFS_BALANCE_OBJECTID;
3105         key.type = BTRFS_TEMPORARY_ITEM_KEY;
3106         key.offset = 0;
3107
3108         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3109         if (ret < 0)
3110                 goto out;
3111         if (ret > 0) {
3112                 ret = -ENOENT;
3113                 goto out;
3114         }
3115
3116         ret = btrfs_del_item(trans, root, path);
3117 out:
3118         btrfs_free_path(path);
3119         err = btrfs_commit_transaction(trans, root);
3120         if (err && !ret)
3121                 ret = err;
3122         return ret;
3123 }
3124
3125 /*
3126  * This is a heuristic used to reduce the number of chunks balanced on
3127  * resume after balance was interrupted.
3128  */
3129 static void update_balance_args(struct btrfs_balance_control *bctl)
3130 {
3131         /*
3132          * Turn on soft mode for chunk types that were being converted.
3133          */
3134         if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
3135                 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
3136         if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
3137                 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
3138         if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
3139                 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
3140
3141         /*
3142          * Turn on usage filter if is not already used.  The idea is
3143          * that chunks that we have already balanced should be
3144          * reasonably full.  Don't do it for chunks that are being
3145          * converted - that will keep us from relocating unconverted
3146          * (albeit full) chunks.
3147          */
3148         if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3149             !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3150             !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3151                 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
3152                 bctl->data.usage = 90;
3153         }
3154         if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3155             !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3156             !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3157                 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
3158                 bctl->sys.usage = 90;
3159         }
3160         if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3161             !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3162             !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3163                 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
3164                 bctl->meta.usage = 90;
3165         }
3166 }
3167
3168 /*
3169  * Should be called with both balance and volume mutexes held to
3170  * serialize other volume operations (add_dev/rm_dev/resize) with
3171  * restriper.  Same goes for unset_balance_control.
3172  */
3173 static void set_balance_control(struct btrfs_balance_control *bctl)
3174 {
3175         struct btrfs_fs_info *fs_info = bctl->fs_info;
3176
3177         BUG_ON(fs_info->balance_ctl);
3178
3179         spin_lock(&fs_info->balance_lock);
3180         fs_info->balance_ctl = bctl;
3181         spin_unlock(&fs_info->balance_lock);
3182 }
3183
3184 static void unset_balance_control(struct btrfs_fs_info *fs_info)
3185 {
3186         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3187
3188         BUG_ON(!fs_info->balance_ctl);
3189
3190         spin_lock(&fs_info->balance_lock);
3191         fs_info->balance_ctl = NULL;
3192         spin_unlock(&fs_info->balance_lock);
3193
3194         kfree(bctl);
3195 }
3196
3197 /*
3198  * Balance filters.  Return 1 if chunk should be filtered out
3199  * (should not be balanced).
3200  */
3201 static int chunk_profiles_filter(u64 chunk_type,
3202                                  struct btrfs_balance_args *bargs)
3203 {
3204         chunk_type = chunk_to_extended(chunk_type) &
3205                                 BTRFS_EXTENDED_PROFILE_MASK;
3206
3207         if (bargs->profiles & chunk_type)
3208                 return 0;
3209
3210         return 1;
3211 }
3212
3213 static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
3214                               struct btrfs_balance_args *bargs)
3215 {
3216         struct btrfs_block_group_cache *cache;
3217         u64 chunk_used;
3218         u64 user_thresh_min;
3219         u64 user_thresh_max;
3220         int ret = 1;
3221
3222         cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3223         chunk_used = btrfs_block_group_used(&cache->item);
3224
3225         if (bargs->usage_min == 0)
3226                 user_thresh_min = 0;
3227         else
3228                 user_thresh_min = div_factor_fine(cache->key.offset,
3229                                         bargs->usage_min);
3230
3231         if (bargs->usage_max == 0)
3232                 user_thresh_max = 1;
3233         else if (bargs->usage_max > 100)
3234                 user_thresh_max = cache->key.offset;
3235         else
3236                 user_thresh_max = div_factor_fine(cache->key.offset,
3237                                         bargs->usage_max);
3238
3239         if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max)
3240                 ret = 0;
3241
3242         btrfs_put_block_group(cache);
3243         return ret;
3244 }
3245
3246 static int chunk_usage_filter(struct btrfs_fs_info *fs_info,
3247                 u64 chunk_offset, struct btrfs_balance_args *bargs)
3248 {
3249         struct btrfs_block_group_cache *cache;
3250         u64 chunk_used, user_thresh;
3251         int ret = 1;
3252
3253         cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3254         chunk_used = btrfs_block_group_used(&cache->item);
3255
3256         if (bargs->usage_min == 0)
3257                 user_thresh = 1;
3258         else if (bargs->usage > 100)
3259                 user_thresh = cache->key.offset;
3260         else
3261                 user_thresh = div_factor_fine(cache->key.offset,
3262                                               bargs->usage);
3263
3264         if (chunk_used < user_thresh)
3265                 ret = 0;
3266
3267         btrfs_put_block_group(cache);
3268         return ret;
3269 }
3270
3271 static int chunk_devid_filter(struct extent_buffer *leaf,
3272                               struct btrfs_chunk *chunk,
3273                               struct btrfs_balance_args *bargs)
3274 {
3275         struct btrfs_stripe *stripe;
3276         int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3277         int i;
3278
3279         for (i = 0; i < num_stripes; i++) {
3280                 stripe = btrfs_stripe_nr(chunk, i);
3281                 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
3282                         return 0;
3283         }
3284
3285         return 1;
3286 }
3287
3288 /* [pstart, pend) */
3289 static int chunk_drange_filter(struct extent_buffer *leaf,
3290                                struct btrfs_chunk *chunk,
3291                                u64 chunk_offset,
3292                                struct btrfs_balance_args *bargs)
3293 {
3294         struct btrfs_stripe *stripe;
3295         int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3296         u64 stripe_offset;
3297         u64 stripe_length;
3298         int factor;
3299         int i;
3300
3301         if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
3302                 return 0;
3303
3304         if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP |
3305              BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) {
3306                 factor = num_stripes / 2;
3307         } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) {
3308                 factor = num_stripes - 1;
3309         } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) {
3310                 factor = num_stripes - 2;
3311         } else {
3312                 factor = num_stripes;
3313         }
3314
3315         for (i = 0; i < num_stripes; i++) {
3316                 stripe = btrfs_stripe_nr(chunk, i);
3317                 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
3318                         continue;
3319
3320                 stripe_offset = btrfs_stripe_offset(leaf, stripe);
3321                 stripe_length = btrfs_chunk_length(leaf, chunk);
3322                 stripe_length = div_u64(stripe_length, factor);
3323
3324                 if (stripe_offset < bargs->pend &&
3325                     stripe_offset + stripe_length > bargs->pstart)
3326                         return 0;
3327         }
3328
3329         return 1;
3330 }
3331
3332 /* [vstart, vend) */
3333 static int chunk_vrange_filter(struct extent_buffer *leaf,
3334                                struct btrfs_chunk *chunk,
3335                                u64 chunk_offset,
3336                                struct btrfs_balance_args *bargs)
3337 {
3338         if (chunk_offset < bargs->vend &&
3339             chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
3340                 /* at least part of the chunk is inside this vrange */
3341                 return 0;
3342
3343         return 1;
3344 }
3345
3346 static int chunk_stripes_range_filter(struct extent_buffer *leaf,
3347                                struct btrfs_chunk *chunk,
3348                                struct btrfs_balance_args *bargs)
3349 {
3350         int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3351
3352         if (bargs->stripes_min <= num_stripes
3353                         && num_stripes <= bargs->stripes_max)
3354                 return 0;
3355
3356         return 1;
3357 }
3358
3359 static int chunk_soft_convert_filter(u64 chunk_type,
3360                                      struct btrfs_balance_args *bargs)
3361 {
3362         if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
3363                 return 0;
3364
3365         chunk_type = chunk_to_extended(chunk_type) &
3366                                 BTRFS_EXTENDED_PROFILE_MASK;
3367
3368         if (bargs->target == chunk_type)
3369                 return 1;
3370
3371         return 0;
3372 }
3373
3374 static int should_balance_chunk(struct btrfs_root *root,
3375                                 struct extent_buffer *leaf,
3376                                 struct btrfs_chunk *chunk, u64 chunk_offset)
3377 {
3378         struct btrfs_balance_control *bctl = root->fs_info->balance_ctl;
3379         struct btrfs_balance_args *bargs = NULL;
3380         u64 chunk_type = btrfs_chunk_type(leaf, chunk);
3381
3382         /* type filter */
3383         if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
3384               (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
3385                 return 0;
3386         }
3387
3388         if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3389                 bargs = &bctl->data;
3390         else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3391                 bargs = &bctl->sys;
3392         else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3393                 bargs = &bctl->meta;
3394
3395         /* profiles filter */
3396         if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
3397             chunk_profiles_filter(chunk_type, bargs)) {
3398                 return 0;
3399         }
3400
3401         /* usage filter */
3402         if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
3403             chunk_usage_filter(bctl->fs_info, chunk_offset, bargs)) {
3404                 return 0;
3405         } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3406             chunk_usage_range_filter(bctl->fs_info, chunk_offset, bargs)) {
3407                 return 0;
3408         }
3409
3410         /* devid filter */
3411         if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
3412             chunk_devid_filter(leaf, chunk, bargs)) {
3413                 return 0;
3414         }
3415
3416         /* drange filter, makes sense only with devid filter */
3417         if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
3418             chunk_drange_filter(leaf, chunk, chunk_offset, bargs)) {
3419                 return 0;
3420         }
3421
3422         /* vrange filter */
3423         if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
3424             chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
3425                 return 0;
3426         }
3427
3428         /* stripes filter */
3429         if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) &&
3430             chunk_stripes_range_filter(leaf, chunk, bargs)) {
3431                 return 0;
3432         }
3433
3434         /* soft profile changing mode */
3435         if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
3436             chunk_soft_convert_filter(chunk_type, bargs)) {
3437                 return 0;
3438         }
3439
3440         /*
3441          * limited by count, must be the last filter
3442          */
3443         if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) {
3444                 if (bargs->limit == 0)
3445                         return 0;
3446                 else
3447                         bargs->limit--;
3448         } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) {
3449                 /*
3450                  * Same logic as the 'limit' filter; the minimum cannot be
3451                  * determined here because we do not have the global information
3452                  * about the count of all chunks that satisfy the filters.
3453                  */
3454                 if (bargs->limit_max == 0)
3455                         return 0;
3456                 else
3457                         bargs->limit_max--;
3458         }
3459
3460         return 1;
3461 }
3462
3463 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
3464 {
3465         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3466         struct btrfs_root *chunk_root = fs_info->chunk_root;
3467         struct btrfs_root *dev_root = fs_info->dev_root;
3468         struct list_head *devices;
3469         struct btrfs_device *device;
3470         u64 old_size;
3471         u64 size_to_free;
3472         u64 chunk_type;
3473         struct btrfs_chunk *chunk;
3474         struct btrfs_path *path = NULL;
3475         struct btrfs_key key;
3476         struct btrfs_key found_key;
3477         struct btrfs_trans_handle *trans;
3478         struct extent_buffer *leaf;
3479         int slot;
3480         int ret;
3481         int enospc_errors = 0;
3482         bool counting = true;
3483         /* The single value limit and min/max limits use the same bytes in the */
3484         u64 limit_data = bctl->data.limit;
3485         u64 limit_meta = bctl->meta.limit;
3486         u64 limit_sys = bctl->sys.limit;
3487         u32 count_data = 0;
3488         u32 count_meta = 0;
3489         u32 count_sys = 0;
3490         int chunk_reserved = 0;
3491         u64 bytes_used = 0;
3492
3493         /* step one make some room on all the devices */
3494         devices = &fs_info->fs_devices->devices;
3495         list_for_each_entry(device, devices, dev_list) {
3496                 old_size = btrfs_device_get_total_bytes(device);
3497                 size_to_free = div_factor(old_size, 1);
3498                 size_to_free = min_t(u64, size_to_free, SZ_1M);
3499                 if (!device->writeable ||
3500                     btrfs_device_get_total_bytes(device) -
3501                     btrfs_device_get_bytes_used(device) > size_to_free ||
3502                     device->is_tgtdev_for_dev_replace)
3503                         continue;
3504
3505                 ret = btrfs_shrink_device(device, old_size - size_to_free);
3506                 if (ret == -ENOSPC)
3507                         break;
3508                 if (ret) {
3509                         /* btrfs_shrink_device never returns ret > 0 */
3510                         WARN_ON(ret > 0);
3511                         goto error;
3512                 }
3513
3514                 trans = btrfs_start_transaction(dev_root, 0);
3515                 if (IS_ERR(trans)) {
3516                         ret = PTR_ERR(trans);
3517                         btrfs_info_in_rcu(fs_info,
3518                  "resize: unable to start transaction after shrinking device %s (error %d), old size %llu, new size %llu",
3519                                           rcu_str_deref(device->name), ret,
3520                                           old_size, old_size - size_to_free);
3521                         goto error;
3522                 }
3523
3524                 ret = btrfs_grow_device(trans, device, old_size);
3525                 if (ret) {
3526                         btrfs_end_transaction(trans, dev_root);
3527                         /* btrfs_grow_device never returns ret > 0 */
3528                         WARN_ON(ret > 0);
3529                         btrfs_info_in_rcu(fs_info,
3530                  "resize: unable to grow device after shrinking device %s (error %d), old size %llu, new size %llu",
3531                                           rcu_str_deref(device->name), ret,
3532                                           old_size, old_size - size_to_free);
3533                         goto error;
3534                 }
3535
3536                 btrfs_end_transaction(trans, dev_root);
3537         }
3538
3539         /* step two, relocate all the chunks */
3540         path = btrfs_alloc_path();
3541         if (!path) {
3542                 ret = -ENOMEM;
3543                 goto error;
3544         }
3545
3546         /* zero out stat counters */
3547         spin_lock(&fs_info->balance_lock);
3548         memset(&bctl->stat, 0, sizeof(bctl->stat));
3549         spin_unlock(&fs_info->balance_lock);
3550 again:
3551         if (!counting) {
3552                 /*
3553                  * The single value limit and min/max limits use the same bytes
3554                  * in the
3555                  */
3556                 bctl->data.limit = limit_data;
3557                 bctl->meta.limit = limit_meta;
3558                 bctl->sys.limit = limit_sys;
3559         }
3560         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3561         key.offset = (u64)-1;
3562         key.type = BTRFS_CHUNK_ITEM_KEY;
3563
3564         while (1) {
3565                 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
3566                     atomic_read(&fs_info->balance_cancel_req)) {
3567                         ret = -ECANCELED;
3568                         goto error;
3569                 }
3570
3571                 mutex_lock(&fs_info->delete_unused_bgs_mutex);
3572                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3573                 if (ret < 0) {
3574                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3575                         goto error;
3576                 }
3577
3578                 /*
3579                  * this shouldn't happen, it means the last relocate
3580                  * failed
3581                  */
3582                 if (ret == 0)
3583                         BUG(); /* FIXME break ? */
3584
3585                 ret = btrfs_previous_item(chunk_root, path, 0,
3586                                           BTRFS_CHUNK_ITEM_KEY);
3587                 if (ret) {
3588                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3589                         ret = 0;
3590                         break;
3591                 }
3592
3593                 leaf = path->nodes[0];
3594                 slot = path->slots[0];
3595                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3596
3597                 if (found_key.objectid != key.objectid) {
3598                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3599                         break;
3600                 }
3601
3602                 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3603                 chunk_type = btrfs_chunk_type(leaf, chunk);
3604
3605                 if (!counting) {
3606                         spin_lock(&fs_info->balance_lock);
3607                         bctl->stat.considered++;
3608                         spin_unlock(&fs_info->balance_lock);
3609                 }
3610
3611                 ret = should_balance_chunk(chunk_root, leaf, chunk,
3612                                            found_key.offset);
3613
3614                 btrfs_release_path(path);
3615                 if (!ret) {
3616                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3617                         goto loop;
3618                 }
3619
3620                 if (counting) {
3621                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3622                         spin_lock(&fs_info->balance_lock);
3623                         bctl->stat.expected++;
3624                         spin_unlock(&fs_info->balance_lock);
3625
3626                         if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3627                                 count_data++;
3628                         else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3629                                 count_sys++;
3630                         else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3631                                 count_meta++;
3632
3633                         goto loop;
3634                 }
3635
3636                 /*
3637                  * Apply limit_min filter, no need to check if the LIMITS
3638                  * filter is used, limit_min is 0 by default
3639                  */
3640                 if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
3641                                         count_data < bctl->data.limit_min)
3642                                 || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) &&
3643                                         count_meta < bctl->meta.limit_min)
3644                                 || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) &&
3645                                         count_sys < bctl->sys.limit_min)) {
3646                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3647                         goto loop;
3648                 }
3649
3650                 ASSERT(fs_info->data_sinfo);
3651                 spin_lock(&fs_info->data_sinfo->lock);
3652                 bytes_used = fs_info->data_sinfo->bytes_used;
3653                 spin_unlock(&fs_info->data_sinfo->lock);
3654
3655                 if ((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
3656                     !chunk_reserved && !bytes_used) {
3657                         trans = btrfs_start_transaction(chunk_root, 0);
3658                         if (IS_ERR(trans)) {
3659                                 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3660                                 ret = PTR_ERR(trans);
3661                                 goto error;
3662                         }
3663
3664                         ret = btrfs_force_chunk_alloc(trans, chunk_root,
3665                                                       BTRFS_BLOCK_GROUP_DATA);
3666                         btrfs_end_transaction(trans, chunk_root);
3667                         if (ret < 0) {
3668                                 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3669                                 goto error;
3670                         }
3671                         chunk_reserved = 1;
3672                 }
3673
3674                 ret = btrfs_relocate_chunk(chunk_root,
3675                                            found_key.offset);
3676                 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3677                 if (ret && ret != -ENOSPC)
3678                         goto error;
3679                 if (ret == -ENOSPC) {
3680                         enospc_errors++;
3681                 } else {
3682                         spin_lock(&fs_info->balance_lock);
3683                         bctl->stat.completed++;
3684                         spin_unlock(&fs_info->balance_lock);
3685                 }
3686 loop:
3687                 if (found_key.offset == 0)
3688                         break;
3689                 key.offset = found_key.offset - 1;
3690         }
3691
3692         if (counting) {
3693                 btrfs_release_path(path);
3694                 counting = false;
3695                 goto again;
3696         }
3697 error:
3698         btrfs_free_path(path);
3699         if (enospc_errors) {
3700                 btrfs_info(fs_info, "%d enospc errors during balance",
3701                        enospc_errors);
3702                 if (!ret)
3703                         ret = -ENOSPC;
3704         }
3705
3706         return ret;
3707 }
3708
3709 /**
3710  * alloc_profile_is_valid - see if a given profile is valid and reduced
3711  * @flags: profile to validate
3712  * @extended: if true @flags is treated as an extended profile
3713  */
3714 static int alloc_profile_is_valid(u64 flags, int extended)
3715 {
3716         u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
3717                                BTRFS_BLOCK_GROUP_PROFILE_MASK);
3718
3719         flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
3720
3721         /* 1) check that all other bits are zeroed */
3722         if (flags & ~mask)
3723                 return 0;
3724
3725         /* 2) see if profile is reduced */
3726         if (flags == 0)
3727                 return !extended; /* "0" is valid for usual profiles */
3728
3729         /* true if exactly one bit set */
3730         return (flags & (flags - 1)) == 0;
3731 }
3732
3733 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
3734 {
3735         /* cancel requested || normal exit path */
3736         return atomic_read(&fs_info->balance_cancel_req) ||
3737                 (atomic_read(&fs_info->balance_pause_req) == 0 &&
3738                  atomic_read(&fs_info->balance_cancel_req) == 0);
3739 }
3740
3741 static void __cancel_balance(struct btrfs_fs_info *fs_info)
3742 {
3743         int ret;
3744
3745         unset_balance_control(fs_info);
3746         ret = del_balance_item(fs_info->tree_root);
3747         if (ret)
3748                 btrfs_handle_fs_error(fs_info, ret, NULL);
3749
3750         atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3751 }
3752
3753 /* Non-zero return value signifies invalidity */
3754 static inline int validate_convert_profile(struct btrfs_balance_args *bctl_arg,
3755                 u64 allowed)
3756 {
3757         return ((bctl_arg->flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3758                 (!alloc_profile_is_valid(bctl_arg->target, 1) ||
3759                  (bctl_arg->target & ~allowed)));
3760 }
3761
3762 /*
3763  * Should be called with both balance and volume mutexes held
3764  */
3765 int btrfs_balance(struct btrfs_balance_control *bctl,
3766                   struct btrfs_ioctl_balance_args *bargs)
3767 {
3768         struct btrfs_fs_info *fs_info = bctl->fs_info;
3769         u64 allowed;
3770         int mixed = 0;
3771         int ret;
3772         u64 num_devices;
3773         unsigned seq;
3774
3775         if (btrfs_fs_closing(fs_info) ||
3776             atomic_read(&fs_info->balance_pause_req) ||
3777             atomic_read(&fs_info->balance_cancel_req)) {
3778                 ret = -EINVAL;
3779                 goto out;
3780         }
3781
3782         allowed = btrfs_super_incompat_flags(fs_info->super_copy);
3783         if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
3784                 mixed = 1;
3785
3786         /*
3787          * In case of mixed groups both data and meta should be picked,
3788          * and identical options should be given for both of them.
3789          */
3790         allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
3791         if (mixed && (bctl->flags & allowed)) {
3792                 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
3793                     !(bctl->flags & BTRFS_BALANCE_METADATA) ||
3794                     memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
3795                         btrfs_err(fs_info, "with mixed groups data and "
3796                                    "metadata balance options must be the same");
3797                         ret = -EINVAL;
3798                         goto out;
3799                 }
3800         }
3801
3802         num_devices = fs_info->fs_devices->num_devices;
3803         btrfs_dev_replace_lock(&fs_info->dev_replace, 0);
3804         if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
3805                 BUG_ON(num_devices < 1);
3806                 num_devices--;
3807         }
3808         btrfs_dev_replace_unlock(&fs_info->dev_replace, 0);
3809         allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE | BTRFS_BLOCK_GROUP_DUP;
3810         if (num_devices > 1)
3811                 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1);
3812         if (num_devices > 2)
3813                 allowed |= BTRFS_BLOCK_GROUP_RAID5;
3814         if (num_devices > 3)
3815                 allowed |= (BTRFS_BLOCK_GROUP_RAID10 |
3816                             BTRFS_BLOCK_GROUP_RAID6);
3817         if (validate_convert_profile(&bctl->data, allowed)) {
3818                 btrfs_err(fs_info, "unable to start balance with target "
3819                            "data profile %llu",
3820                        bctl->data.target);
3821                 ret = -EINVAL;
3822                 goto out;
3823         }
3824         if (validate_convert_profile(&bctl->meta, allowed)) {
3825                 btrfs_err(fs_info,
3826                            "unable to start balance with target metadata profile %llu",
3827                        bctl->meta.target);
3828                 ret = -EINVAL;
3829                 goto out;
3830         }
3831         if (validate_convert_profile(&bctl->sys, allowed)) {
3832                 btrfs_err(fs_info,
3833                            "unable to start balance with target system profile %llu",
3834                        bctl->sys.target);
3835                 ret = -EINVAL;
3836                 goto out;
3837         }
3838
3839         /* allow to reduce meta or sys integrity only if force set */
3840         allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3841                         BTRFS_BLOCK_GROUP_RAID10 |
3842                         BTRFS_BLOCK_GROUP_RAID5 |
3843                         BTRFS_BLOCK_GROUP_RAID6;
3844         do {
3845                 seq = read_seqbegin(&fs_info->profiles_lock);
3846
3847                 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3848                      (fs_info->avail_system_alloc_bits & allowed) &&
3849                      !(bctl->sys.target & allowed)) ||
3850                     ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3851                      (fs_info->avail_metadata_alloc_bits & allowed) &&
3852                      !(bctl->meta.target & allowed))) {
3853                         if (bctl->flags & BTRFS_BALANCE_FORCE) {
3854                                 btrfs_info(fs_info, "force reducing metadata integrity");
3855                         } else {
3856                                 btrfs_err(fs_info, "balance will reduce metadata "
3857                                            "integrity, use force if you want this");
3858                                 ret = -EINVAL;
3859                                 goto out;
3860                         }
3861                 }
3862         } while (read_seqretry(&fs_info->profiles_lock, seq));
3863
3864         if (btrfs_get_num_tolerated_disk_barrier_failures(bctl->meta.target) <
3865                 btrfs_get_num_tolerated_disk_barrier_failures(bctl->data.target)) {
3866                 btrfs_warn(fs_info,
3867         "metadata profile 0x%llx has lower redundancy than data profile 0x%llx",
3868                         bctl->meta.target, bctl->data.target);
3869         }
3870
3871         if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3872                 fs_info->num_tolerated_disk_barrier_failures = min(
3873                         btrfs_calc_num_tolerated_disk_barrier_failures(fs_info),
3874                         btrfs_get_num_tolerated_disk_barrier_failures(
3875                                 bctl->sys.target));
3876         }
3877
3878         ret = insert_balance_item(fs_info->tree_root, bctl);
3879         if (ret && ret != -EEXIST)
3880                 goto out;
3881
3882         if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
3883                 BUG_ON(ret == -EEXIST);
3884                 set_balance_control(bctl);
3885         } else {
3886                 BUG_ON(ret != -EEXIST);
3887                 spin_lock(&fs_info->balance_lock);
3888                 update_balance_args(bctl);
3889                 spin_unlock(&fs_info->balance_lock);
3890         }
3891
3892         atomic_inc(&fs_info->balance_running);
3893         mutex_unlock(&fs_info->balance_mutex);
3894
3895         ret = __btrfs_balance(fs_info);
3896
3897         mutex_lock(&fs_info->balance_mutex);
3898         atomic_dec(&fs_info->balance_running);
3899
3900         if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3901                 fs_info->num_tolerated_disk_barrier_failures =
3902                         btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3903         }
3904
3905         if (bargs) {
3906                 memset(bargs, 0, sizeof(*bargs));
3907                 update_ioctl_balance_args(fs_info, 0, bargs);
3908         }
3909
3910         if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
3911             balance_need_close(fs_info)) {
3912                 __cancel_balance(fs_info);
3913         }
3914
3915         wake_up(&fs_info->balance_wait_q);
3916
3917         return ret;
3918 out:
3919         if (bctl->flags & BTRFS_BALANCE_RESUME)
3920                 __cancel_balance(fs_info);
3921         else {
3922                 kfree(bctl);
3923                 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3924         }
3925         return ret;
3926 }
3927
3928 static int balance_kthread(void *data)
3929 {
3930         struct btrfs_fs_info *fs_info = data;
3931         int ret = 0;
3932
3933         mutex_lock(&fs_info->volume_mutex);
3934         mutex_lock(&fs_info->balance_mutex);
3935
3936         if (fs_info->balance_ctl) {
3937                 btrfs_info(fs_info, "continuing balance");
3938                 ret = btrfs_balance(fs_info->balance_ctl, NULL);
3939         }
3940
3941         mutex_unlock(&fs_info->balance_mutex);
3942         mutex_unlock(&fs_info->volume_mutex);
3943
3944         return ret;
3945 }
3946
3947 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
3948 {
3949         struct task_struct *tsk;
3950
3951         spin_lock(&fs_info->balance_lock);
3952         if (!fs_info->balance_ctl) {
3953                 spin_unlock(&fs_info->balance_lock);
3954                 return 0;
3955         }
3956         spin_unlock(&fs_info->balance_lock);
3957
3958         if (btrfs_test_opt(fs_info, SKIP_BALANCE)) {
3959                 btrfs_info(fs_info, "force skipping balance");
3960                 return 0;
3961         }
3962
3963         tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
3964         return PTR_ERR_OR_ZERO(tsk);
3965 }
3966
3967 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
3968 {
3969         struct btrfs_balance_control *bctl;
3970         struct btrfs_balance_item *item;
3971         struct btrfs_disk_balance_args disk_bargs;
3972         struct btrfs_path *path;
3973         struct extent_buffer *leaf;
3974         struct btrfs_key key;
3975         int ret;
3976
3977         path = btrfs_alloc_path();
3978         if (!path)
3979                 return -ENOMEM;
3980
3981         key.objectid = BTRFS_BALANCE_OBJECTID;
3982         key.type = BTRFS_TEMPORARY_ITEM_KEY;
3983         key.offset = 0;
3984
3985         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
3986         if (ret < 0)
3987                 goto out;
3988         if (ret > 0) { /* ret = -ENOENT; */
3989                 ret = 0;
3990                 goto out;
3991         }
3992
3993         bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
3994         if (!bctl) {
3995                 ret = -ENOMEM;
3996                 goto out;
3997         }
3998
3999         leaf = path->nodes[0];
4000         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
4001
4002         bctl->fs_info = fs_info;
4003         bctl->flags = btrfs_balance_flags(leaf, item);
4004         bctl->flags |= BTRFS_BALANCE_RESUME;
4005
4006         btrfs_balance_data(leaf, item, &disk_bargs);
4007         btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
4008         btrfs_balance_meta(leaf, item, &disk_bargs);
4009         btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
4010         btrfs_balance_sys(leaf, item, &disk_bargs);
4011         btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
4012
4013         WARN_ON(atomic_xchg(&fs_info->mutually_exclusive_operation_running, 1));
4014
4015         mutex_lock(&fs_info->volume_mutex);
4016         mutex_lock(&fs_info->balance_mutex);
4017
4018         set_balance_control(bctl);
4019
4020         mutex_unlock(&fs_info->balance_mutex);
4021         mutex_unlock(&fs_info->volume_mutex);
4022 out:
4023         btrfs_free_path(path);
4024         return ret;
4025 }
4026
4027 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
4028 {
4029         int ret = 0;
4030
4031         mutex_lock(&fs_info->balance_mutex);
4032         if (!fs_info->balance_ctl) {
4033                 mutex_unlock(&fs_info->balance_mutex);
4034                 return -ENOTCONN;
4035         }
4036
4037         if (atomic_read(&fs_info->balance_running)) {
4038                 atomic_inc(&fs_info->balance_pause_req);
4039                 mutex_unlock(&fs_info->balance_mutex);
4040
4041                 wait_event(fs_info->balance_wait_q,
4042                            atomic_read(&fs_info->balance_running) == 0);
4043
4044                 mutex_lock(&fs_info->balance_mutex);
4045                 /* we are good with balance_ctl ripped off from under us */
4046                 BUG_ON(atomic_read(&fs_info->balance_running));
4047                 atomic_dec(&fs_info->balance_pause_req);
4048         } else {
4049                 ret = -ENOTCONN;
4050         }
4051
4052         mutex_unlock(&fs_info->balance_mutex);
4053         return ret;
4054 }
4055
4056 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
4057 {
4058         if (fs_info->sb->s_flags & MS_RDONLY)
4059                 return -EROFS;
4060
4061         mutex_lock(&fs_info->balance_mutex);
4062         if (!fs_info->balance_ctl) {
4063                 mutex_unlock(&fs_info->balance_mutex);
4064                 return -ENOTCONN;
4065         }
4066
4067         atomic_inc(&fs_info->balance_cancel_req);
4068         /*
4069          * if we are running just wait and return, balance item is
4070          * deleted in btrfs_balance in this case
4071          */
4072         if (atomic_read(&fs_info->balance_running)) {
4073                 mutex_unlock(&fs_info->balance_mutex);
4074                 wait_event(fs_info->balance_wait_q,
4075                            atomic_read(&fs_info->balance_running) == 0);
4076                 mutex_lock(&fs_info->balance_mutex);
4077         } else {
4078                 /* __cancel_balance needs volume_mutex */
4079                 mutex_unlock(&fs_info->balance_mutex);
4080                 mutex_lock(&fs_info->volume_mutex);
4081                 mutex_lock(&fs_info->balance_mutex);
4082
4083                 if (fs_info->balance_ctl)
4084                         __cancel_balance(fs_info);
4085
4086                 mutex_unlock(&fs_info->volume_mutex);
4087         }
4088
4089         BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running));
4090         atomic_dec(&fs_info->balance_cancel_req);
4091         mutex_unlock(&fs_info->balance_mutex);
4092         return 0;
4093 }
4094
4095 static int btrfs_uuid_scan_kthread(void *data)
4096 {
4097         struct btrfs_fs_info *fs_info = data;
4098         struct btrfs_root *root = fs_info->tree_root;
4099         struct btrfs_key key;
4100         struct btrfs_key max_key;
4101         struct btrfs_path *path = NULL;
4102         int ret = 0;
4103         struct extent_buffer *eb;
4104         int slot;
4105         struct btrfs_root_item root_item;
4106         u32 item_size;
4107         struct btrfs_trans_handle *trans = NULL;
4108
4109         path = btrfs_alloc_path();
4110         if (!path) {
4111                 ret = -ENOMEM;
4112                 goto out;
4113         }
4114
4115         key.objectid = 0;
4116         key.type = BTRFS_ROOT_ITEM_KEY;
4117         key.offset = 0;
4118
4119         max_key.objectid = (u64)-1;
4120         max_key.type = BTRFS_ROOT_ITEM_KEY;
4121         max_key.offset = (u64)-1;
4122
4123         while (1) {
4124                 ret = btrfs_search_forward(root, &key, path, 0);
4125                 if (ret) {
4126                         if (ret > 0)
4127                                 ret = 0;
4128                         break;
4129                 }
4130
4131                 if (key.type != BTRFS_ROOT_ITEM_KEY ||
4132                     (key.objectid < BTRFS_FIRST_FREE_OBJECTID &&
4133                      key.objectid != BTRFS_FS_TREE_OBJECTID) ||
4134                     key.objectid > BTRFS_LAST_FREE_OBJECTID)
4135                         goto skip;
4136
4137                 eb = path->nodes[0];
4138                 slot = path->slots[0];
4139                 item_size = btrfs_item_size_nr(eb, slot);
4140                 if (item_size < sizeof(root_item))
4141                         goto skip;
4142
4143                 read_extent_buffer(eb, &root_item,
4144                                    btrfs_item_ptr_offset(eb, slot),
4145                                    (int)sizeof(root_item));
4146                 if (btrfs_root_refs(&root_item) == 0)
4147                         goto skip;
4148
4149                 if (!btrfs_is_empty_uuid(root_item.uuid) ||
4150                     !btrfs_is_empty_uuid(root_item.received_uuid)) {
4151                         if (trans)
4152                                 goto update_tree;
4153
4154                         btrfs_release_path(path);
4155                         /*
4156                          * 1 - subvol uuid item
4157                          * 1 - received_subvol uuid item
4158                          */
4159                         trans = btrfs_start_transaction(fs_info->uuid_root, 2);
4160                         if (IS_ERR(trans)) {
4161                                 ret = PTR_ERR(trans);
4162                                 break;
4163                         }
4164                         continue;
4165                 } else {
4166                         goto skip;
4167                 }
4168 update_tree:
4169                 if (!btrfs_is_empty_uuid(root_item.uuid)) {
4170                         ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
4171                                                   root_item.uuid,
4172                                                   BTRFS_UUID_KEY_SUBVOL,
4173                                                   key.objectid);
4174                         if (ret < 0) {
4175                                 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4176                                         ret);
4177                                 break;
4178                         }
4179                 }
4180
4181                 if (!btrfs_is_empty_uuid(root_item.received_uuid)) {
4182                         ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
4183                                                   root_item.received_uuid,
4184                                                  BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4185                                                   key.objectid);
4186                         if (ret < 0) {
4187                                 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4188                                         ret);
4189                                 break;
4190                         }
4191                 }
4192
4193 skip:
4194                 if (trans) {
4195                         ret = btrfs_end_transaction(trans, fs_info->uuid_root);
4196                         trans = NULL;
4197                         if (ret)
4198                                 break;
4199                 }
4200
4201                 btrfs_release_path(path);
4202                 if (key.offset < (u64)-1) {
4203                         key.offset++;
4204                 } else if (key.type < BTRFS_ROOT_ITEM_KEY) {
4205                         key.offset = 0;
4206                         key.type = BTRFS_ROOT_ITEM_KEY;
4207                 } else if (key.objectid < (u64)-1) {
4208                         key.offset = 0;
4209                         key.type = BTRFS_ROOT_ITEM_KEY;
4210                         key.objectid++;
4211                 } else {
4212                         break;
4213                 }
4214                 cond_resched();
4215         }
4216
4217 out:
4218         btrfs_free_path(path);
4219         if (trans && !IS_ERR(trans))
4220                 btrfs_end_transaction(trans, fs_info->uuid_root);
4221         if (ret)
4222                 btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret);
4223         else
4224                 fs_info->update_uuid_tree_gen = 1;
4225         up(&fs_info->uuid_tree_rescan_sem);
4226         return 0;
4227 }
4228
4229 /*
4230  * Callback for btrfs_uuid_tree_iterate().
4231  * returns:
4232  * 0    check succeeded, the entry is not outdated.
4233  * < 0  if an error occurred.
4234  * > 0  if the check failed, which means the caller shall remove the entry.
4235  */
4236 static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info *fs_info,
4237                                        u8 *uuid, u8 type, u64 subid)
4238 {
4239         struct btrfs_key key;
4240         int ret = 0;
4241         struct btrfs_root *subvol_root;
4242
4243         if (type != BTRFS_UUID_KEY_SUBVOL &&
4244             type != BTRFS_UUID_KEY_RECEIVED_SUBVOL)
4245                 goto out;
4246
4247         key.objectid = subid;
4248         key.type = BTRFS_ROOT_ITEM_KEY;
4249         key.offset = (u64)-1;
4250         subvol_root = btrfs_read_fs_root_no_name(fs_info, &key);
4251         if (IS_ERR(subvol_root)) {
4252                 ret = PTR_ERR(subvol_root);
4253                 if (ret == -ENOENT)
4254                         ret = 1;
4255                 goto out;
4256         }
4257
4258         switch (type) {
4259         case BTRFS_UUID_KEY_SUBVOL:
4260                 if (memcmp(uuid, subvol_root->root_item.uuid, BTRFS_UUID_SIZE))
4261                         ret = 1;
4262                 break;
4263         case BTRFS_UUID_KEY_RECEIVED_SUBVOL:
4264                 if (memcmp(uuid, subvol_root->root_item.received_uuid,
4265                            BTRFS_UUID_SIZE))
4266                         ret = 1;
4267                 break;
4268         }
4269
4270 out:
4271         return ret;
4272 }
4273
4274 static int btrfs_uuid_rescan_kthread(void *data)
4275 {
4276         struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data;
4277         int ret;
4278
4279         /*
4280          * 1st step is to iterate through the existing UUID tree and
4281          * to delete all entries that contain outdated data.
4282          * 2nd step is to add all missing entries to the UUID tree.
4283          */
4284         ret = btrfs_uuid_tree_iterate(fs_info, btrfs_check_uuid_tree_entry);
4285         if (ret < 0) {
4286                 btrfs_warn(fs_info, "iterating uuid_tree failed %d", ret);
4287                 up(&fs_info->uuid_tree_rescan_sem);
4288                 return ret;
4289         }
4290         return btrfs_uuid_scan_kthread(data);
4291 }
4292
4293 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info)
4294 {
4295         struct btrfs_trans_handle *trans;
4296         struct btrfs_root *tree_root = fs_info->tree_root;
4297         struct btrfs_root *uuid_root;
4298         struct task_struct *task;
4299         int ret;
4300
4301         /*
4302          * 1 - root node
4303          * 1 - root item
4304          */
4305         trans = btrfs_start_transaction(tree_root, 2);
4306         if (IS_ERR(trans))
4307                 return PTR_ERR(trans);
4308
4309         uuid_root = btrfs_create_tree(trans, fs_info,
4310                                       BTRFS_UUID_TREE_OBJECTID);
4311         if (IS_ERR(uuid_root)) {
4312                 ret = PTR_ERR(uuid_root);
4313                 btrfs_abort_transaction(trans, ret);
4314                 btrfs_end_transaction(trans, tree_root);
4315                 return ret;
4316         }
4317
4318         fs_info->uuid_root = uuid_root;
4319
4320         ret = btrfs_commit_transaction(trans, tree_root);
4321         if (ret)
4322                 return ret;
4323
4324         down(&fs_info->uuid_tree_rescan_sem);
4325         task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid");
4326         if (IS_ERR(task)) {
4327                 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4328                 btrfs_warn(fs_info, "failed to start uuid_scan task");
4329                 up(&fs_info->uuid_tree_rescan_sem);
4330                 return PTR_ERR(task);
4331         }
4332
4333         return 0;
4334 }
4335
4336 int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
4337 {
4338         struct task_struct *task;
4339
4340         down(&fs_info->uuid_tree_rescan_sem);
4341         task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
4342         if (IS_ERR(task)) {
4343                 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4344                 btrfs_warn(fs_info, "failed to start uuid_rescan task");
4345                 up(&fs_info->uuid_tree_rescan_sem);
4346                 return PTR_ERR(task);
4347         }
4348
4349         return 0;
4350 }
4351
4352 /*
4353  * shrinking a device means finding all of the device extents past
4354  * the new size, and then following the back refs to the chunks.
4355  * The chunk relocation code actually frees the device extent
4356  */
4357 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
4358 {
4359         struct btrfs_trans_handle *trans;
4360         struct btrfs_root *root = device->dev_root;
4361         struct btrfs_dev_extent *dev_extent = NULL;
4362         struct btrfs_path *path;
4363         u64 length;
4364         u64 chunk_offset;
4365         int ret;
4366         int slot;
4367         int failed = 0;
4368         bool retried = false;
4369         bool checked_pending_chunks = false;
4370         struct extent_buffer *l;
4371         struct btrfs_key key;
4372         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
4373         u64 old_total = btrfs_super_total_bytes(super_copy);
4374         u64 old_size = btrfs_device_get_total_bytes(device);
4375         u64 diff = old_size - new_size;
4376
4377         if (device->is_tgtdev_for_dev_replace)
4378                 return -EINVAL;
4379
4380         path = btrfs_alloc_path();
4381         if (!path)
4382                 return -ENOMEM;
4383
4384         path->reada = READA_FORWARD;
4385
4386         lock_chunks(root);
4387
4388         btrfs_device_set_total_bytes(device, new_size);
4389         if (device->writeable) {
4390                 device->fs_devices->total_rw_bytes -= diff;
4391                 spin_lock(&root->fs_info->free_chunk_lock);
4392                 root->fs_info->free_chunk_space -= diff;
4393                 spin_unlock(&root->fs_info->free_chunk_lock);
4394         }
4395         unlock_chunks(root);
4396
4397 again:
4398         key.objectid = device->devid;
4399         key.offset = (u64)-1;
4400         key.type = BTRFS_DEV_EXTENT_KEY;
4401
4402         do {
4403                 mutex_lock(&root->fs_info->delete_unused_bgs_mutex);
4404                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4405                 if (ret < 0) {
4406                         mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4407                         goto done;
4408                 }
4409
4410                 ret = btrfs_previous_item(root, path, 0, key.type);
4411                 if (ret)
4412                         mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4413                 if (ret < 0)
4414                         goto done;
4415                 if (ret) {
4416                         ret = 0;
4417                         btrfs_release_path(path);
4418                         break;
4419                 }
4420
4421                 l = path->nodes[0];
4422                 slot = path->slots[0];
4423                 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
4424
4425                 if (key.objectid != device->devid) {
4426                         mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4427                         btrfs_release_path(path);
4428                         break;
4429                 }
4430
4431                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
4432                 length = btrfs_dev_extent_length(l, dev_extent);
4433
4434                 if (key.offset + length <= new_size) {
4435                         mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4436                         btrfs_release_path(path);
4437                         break;
4438                 }
4439
4440                 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
4441                 btrfs_release_path(path);
4442
4443                 ret = btrfs_relocate_chunk(root, chunk_offset);
4444                 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4445                 if (ret && ret != -ENOSPC)
4446                         goto done;
4447                 if (ret == -ENOSPC)
4448                         failed++;
4449         } while (key.offset-- > 0);
4450
4451         if (failed && !retried) {
4452                 failed = 0;
4453                 retried = true;
4454                 goto again;
4455         } else if (failed && retried) {
4456                 ret = -ENOSPC;
4457                 goto done;
4458         }
4459
4460         /* Shrinking succeeded, else we would be at "done". */
4461         trans = btrfs_start_transaction(root, 0);
4462         if (IS_ERR(trans)) {
4463                 ret = PTR_ERR(trans);
4464                 goto done;
4465         }
4466
4467         lock_chunks(root);
4468
4469         /*
4470          * We checked in the above loop all device extents that were already in
4471          * the device tree. However before we have updated the device's
4472          * total_bytes to the new size, we might have had chunk allocations that
4473          * have not complete yet (new block groups attached to transaction
4474          * handles), and therefore their device extents were not yet in the
4475          * device tree and we missed them in the loop above. So if we have any
4476          * pending chunk using a device extent that overlaps the device range
4477          * that we can not use anymore, commit the current transaction and
4478          * repeat the search on the device tree - this way we guarantee we will
4479          * not have chunks using device extents that end beyond 'new_size'.
4480          */
4481         if (!checked_pending_chunks) {
4482                 u64 start = new_size;
4483                 u64 len = old_size - new_size;
4484
4485                 if (contains_pending_extent(trans->transaction, device,
4486                                             &start, len)) {
4487                         unlock_chunks(root);
4488                         checked_pending_chunks = true;
4489                         failed = 0;
4490                         retried = false;
4491                         ret = btrfs_commit_transaction(trans, root);
4492                         if (ret)
4493                                 goto done;
4494                         goto again;
4495                 }
4496         }
4497
4498         btrfs_device_set_disk_total_bytes(device, new_size);
4499         if (list_empty(&device->resized_list))
4500                 list_add_tail(&device->resized_list,
4501                               &root->fs_info->fs_devices->resized_devices);
4502
4503         WARN_ON(diff > old_total);
4504         btrfs_set_super_total_bytes(super_copy, old_total - diff);
4505         unlock_chunks(root);
4506
4507         /* Now btrfs_update_device() will change the on-disk size. */
4508         ret = btrfs_update_device(trans, device);
4509         btrfs_end_transaction(trans, root);
4510 done:
4511         btrfs_free_path(path);
4512         if (ret) {
4513                 lock_chunks(root);
4514                 btrfs_device_set_total_bytes(device, old_size);
4515                 if (device->writeable)
4516                         device->fs_devices->total_rw_bytes += diff;
4517                 spin_lock(&root->fs_info->free_chunk_lock);
4518                 root->fs_info->free_chunk_space += diff;
4519                 spin_unlock(&root->fs_info->free_chunk_lock);
4520                 unlock_chunks(root);
4521         }
4522         return ret;
4523 }
4524
4525 static int btrfs_add_system_chunk(struct btrfs_root *root,
4526                            struct btrfs_key *key,
4527                            struct btrfs_chunk *chunk, int item_size)
4528 {
4529         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
4530         struct btrfs_disk_key disk_key;
4531         u32 array_size;
4532         u8 *ptr;
4533
4534         lock_chunks(root);
4535         array_size = btrfs_super_sys_array_size(super_copy);
4536         if (array_size + item_size + sizeof(disk_key)
4537                         > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
4538                 unlock_chunks(root);
4539                 return -EFBIG;
4540         }
4541
4542         ptr = super_copy->sys_chunk_array + array_size;
4543         btrfs_cpu_key_to_disk(&disk_key, key);
4544         memcpy(ptr, &disk_key, sizeof(disk_key));
4545         ptr += sizeof(disk_key);
4546         memcpy(ptr, chunk, item_size);
4547         item_size += sizeof(disk_key);
4548         btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
4549         unlock_chunks(root);
4550
4551         return 0;
4552 }
4553
4554 /*
4555  * sort the devices in descending order by max_avail, total_avail
4556  */
4557 static int btrfs_cmp_device_info(const void *a, const void *b)
4558 {
4559         const struct btrfs_device_info *di_a = a;
4560         const struct btrfs_device_info *di_b = b;
4561
4562         if (di_a->max_avail > di_b->max_avail)
4563                 return -1;
4564         if (di_a->max_avail < di_b->max_avail)
4565                 return 1;
4566         if (di_a->total_avail > di_b->total_avail)
4567                 return -1;
4568         if (di_a->total_avail < di_b->total_avail)
4569                 return 1;
4570         return 0;
4571 }
4572
4573 static u32 find_raid56_stripe_len(u32 data_devices, u32 dev_stripe_target)
4574 {
4575         /* TODO allow them to set a preferred stripe size */
4576         return SZ_64K;
4577 }
4578
4579 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
4580 {
4581         if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK))
4582                 return;
4583
4584         btrfs_set_fs_incompat(info, RAID56);
4585 }
4586
4587 #define BTRFS_MAX_DEVS(r) ((BTRFS_MAX_ITEM_SIZE(r)              \
4588                         - sizeof(struct btrfs_chunk))           \
4589                         / sizeof(struct btrfs_stripe) + 1)
4590
4591 #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE        \
4592                                 - 2 * sizeof(struct btrfs_disk_key)     \
4593                                 - 2 * sizeof(struct btrfs_chunk))       \
4594                                 / sizeof(struct btrfs_stripe) + 1)
4595
4596 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4597                                struct btrfs_root *extent_root, u64 start,
4598                                u64 type)
4599 {
4600         struct btrfs_fs_info *info = extent_root->fs_info;
4601         struct btrfs_fs_devices *fs_devices = info->fs_devices;
4602         struct list_head *cur;
4603         struct map_lookup *map = NULL;
4604         struct extent_map_tree *em_tree;
4605         struct extent_map *em;
4606         struct btrfs_device_info *devices_info = NULL;
4607         u64 total_avail;
4608         int num_stripes;        /* total number of stripes to allocate */
4609         int data_stripes;       /* number of stripes that count for
4610                                    block group size */
4611         int sub_stripes;        /* sub_stripes info for map */
4612         int dev_stripes;        /* stripes per dev */
4613         int devs_max;           /* max devs to use */
4614         int devs_min;           /* min devs needed */
4615         int devs_increment;     /* ndevs has to be a multiple of this */
4616         int ncopies;            /* how many copies to data has */
4617         int ret;
4618         u64 max_stripe_size;
4619         u64 max_chunk_size;
4620         u64 stripe_size;
4621         u64 num_bytes;
4622         u64 raid_stripe_len = BTRFS_STRIPE_LEN;
4623         int ndevs;
4624         int i;
4625         int j;
4626         int index;
4627
4628         BUG_ON(!alloc_profile_is_valid(type, 0));
4629
4630         if (list_empty(&fs_devices->alloc_list))
4631                 return -ENOSPC;
4632
4633         index = __get_raid_index(type);
4634
4635         sub_stripes = btrfs_raid_array[index].sub_stripes;
4636         dev_stripes = btrfs_raid_array[index].dev_stripes;
4637         devs_max = btrfs_raid_array[index].devs_max;
4638         devs_min = btrfs_raid_array[index].devs_min;
4639         devs_increment = btrfs_raid_array[index].devs_increment;
4640         ncopies = btrfs_raid_array[index].ncopies;
4641
4642         if (type & BTRFS_BLOCK_GROUP_DATA) {
4643                 max_stripe_size = SZ_1G;
4644                 max_chunk_size = 10 * max_stripe_size;
4645                 if (!devs_max)
4646                         devs_max = BTRFS_MAX_DEVS(info->chunk_root);
4647         } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
4648                 /* for larger filesystems, use larger metadata chunks */
4649                 if (fs_devices->total_rw_bytes > 50ULL * SZ_1G)
4650                         max_stripe_size = SZ_1G;
4651                 else
4652                         max_stripe_size = SZ_256M;
4653                 max_chunk_size = max_stripe_size;
4654                 if (!devs_max)
4655                         devs_max = BTRFS_MAX_DEVS(info->chunk_root);
4656         } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
4657                 max_stripe_size = SZ_32M;
4658                 max_chunk_size = 2 * max_stripe_size;
4659                 if (!devs_max)
4660                         devs_max = BTRFS_MAX_DEVS_SYS_CHUNK;
4661         } else {
4662                 btrfs_err(info, "invalid chunk type 0x%llx requested",
4663                        type);
4664                 BUG_ON(1);
4665         }
4666
4667         /* we don't want a chunk larger than 10% of writeable space */
4668         max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
4669                              max_chunk_size);
4670
4671         devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info),
4672                                GFP_NOFS);
4673         if (!devices_info)
4674                 return -ENOMEM;
4675
4676         cur = fs_devices->alloc_list.next;
4677
4678         /*
4679          * in the first pass through the devices list, we gather information
4680          * about the available holes on each device.
4681          */
4682         ndevs = 0;
4683         while (cur != &fs_devices->alloc_list) {
4684                 struct btrfs_device *device;
4685                 u64 max_avail;
4686                 u64 dev_offset;
4687
4688                 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
4689
4690                 cur = cur->next;
4691
4692                 if (!device->writeable) {
4693                         WARN(1, KERN_ERR
4694                                "BTRFS: read-only device in alloc_list\n");
4695                         continue;
4696                 }
4697
4698                 if (!device->in_fs_metadata ||
4699                     device->is_tgtdev_for_dev_replace)
4700                         continue;
4701
4702                 if (device->total_bytes > device->bytes_used)
4703                         total_avail = device->total_bytes - device->bytes_used;
4704                 else
4705                         total_avail = 0;
4706
4707                 /* If there is no space on this device, skip it. */
4708                 if (total_avail == 0)
4709                         continue;
4710
4711                 ret = find_free_dev_extent(trans, device,
4712                                            max_stripe_size * dev_stripes,
4713                                            &dev_offset, &max_avail);
4714                 if (ret && ret != -ENOSPC)
4715                         goto error;
4716
4717                 if (ret == 0)
4718                         max_avail = max_stripe_size * dev_stripes;
4719
4720                 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes)
4721                         continue;
4722
4723                 if (ndevs == fs_devices->rw_devices) {
4724                         WARN(1, "%s: found more than %llu devices\n",
4725                              __func__, fs_devices->rw_devices);
4726                         break;
4727                 }
4728                 devices_info[ndevs].dev_offset = dev_offset;
4729                 devices_info[ndevs].max_avail = max_avail;
4730                 devices_info[ndevs].total_avail = total_avail;
4731                 devices_info[ndevs].dev = device;
4732                 ++ndevs;
4733         }
4734
4735         /*
4736          * now sort the devices by hole size / available space
4737          */
4738         sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
4739              btrfs_cmp_device_info, NULL);
4740
4741         /* round down to number of usable stripes */
4742         ndevs -= ndevs % devs_increment;
4743
4744         if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) {
4745                 ret = -ENOSPC;
4746                 goto error;
4747         }
4748
4749         if (devs_max && ndevs > devs_max)
4750                 ndevs = devs_max;
4751         /*
4752          * the primary goal is to maximize the number of stripes, so use as many
4753          * devices as possible, even if the stripes are not maximum sized.
4754          */
4755         stripe_size = devices_info[ndevs-1].max_avail;
4756         num_stripes = ndevs * dev_stripes;
4757
4758         /*
4759          * this will have to be fixed for RAID1 and RAID10 over
4760          * more drives
4761          */
4762         data_stripes = num_stripes / ncopies;
4763
4764         if (type & BTRFS_BLOCK_GROUP_RAID5) {
4765                 raid_stripe_len = find_raid56_stripe_len(ndevs - 1,
4766                                                 extent_root->stripesize);
4767                 data_stripes = num_stripes - 1;
4768         }
4769         if (type & BTRFS_BLOCK_GROUP_RAID6) {
4770                 raid_stripe_len = find_raid56_stripe_len(ndevs - 2,
4771                                                 extent_root->stripesize);
4772                 data_stripes = num_stripes - 2;
4773         }
4774
4775         /*
4776          * Use the number of data stripes to figure out how big this chunk
4777          * is really going to be in terms of logical address space,
4778          * and compare that answer with the max chunk size
4779          */
4780         if (stripe_size * data_stripes > max_chunk_size) {
4781                 u64 mask = (1ULL << 24) - 1;
4782
4783                 stripe_size = div_u64(max_chunk_size, data_stripes);
4784
4785                 /* bump the answer up to a 16MB boundary */
4786                 stripe_size = (stripe_size + mask) & ~mask;
4787
4788                 /* but don't go higher than the limits we found
4789                  * while searching for free extents
4790                  */
4791                 if (stripe_size > devices_info[ndevs-1].max_avail)
4792                         stripe_size = devices_info[ndevs-1].max_avail;
4793         }
4794
4795         stripe_size = div_u64(stripe_size, dev_stripes);
4796
4797         /* align to BTRFS_STRIPE_LEN */
4798         stripe_size = div_u64(stripe_size, raid_stripe_len);
4799         stripe_size *= raid_stripe_len;
4800
4801         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
4802         if (!map) {
4803                 ret = -ENOMEM;
4804                 goto error;
4805         }
4806         map->num_stripes = num_stripes;
4807
4808         for (i = 0; i < ndevs; ++i) {
4809                 for (j = 0; j < dev_stripes; ++j) {
4810                         int s = i * dev_stripes + j;
4811                         map->stripes[s].dev = devices_info[i].dev;
4812                         map->stripes[s].physical = devices_info[i].dev_offset +
4813                                                    j * stripe_size;
4814                 }
4815         }
4816         map->sector_size = extent_root->sectorsize;
4817         map->stripe_len = raid_stripe_len;
4818         map->io_align = raid_stripe_len;
4819         map->io_width = raid_stripe_len;
4820         map->type = type;
4821         map->sub_stripes = sub_stripes;
4822
4823         num_bytes = stripe_size * data_stripes;
4824
4825         trace_btrfs_chunk_alloc(info->chunk_root, map, start, num_bytes);
4826
4827         em = alloc_extent_map();
4828         if (!em) {
4829                 kfree(map);
4830                 ret = -ENOMEM;
4831                 goto error;
4832         }
4833         set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
4834         em->map_lookup = map;
4835         em->start = start;
4836         em->len = num_bytes;
4837         em->block_start = 0;
4838         em->block_len = em->len;
4839         em->orig_block_len = stripe_size;
4840
4841         em_tree = &extent_root->fs_info->mapping_tree.map_tree;
4842         write_lock(&em_tree->lock);
4843         ret = add_extent_mapping(em_tree, em, 0);
4844         if (!ret) {
4845                 list_add_tail(&em->list, &trans->transaction->pending_chunks);
4846                 atomic_inc(&em->refs);
4847         }
4848         write_unlock(&em_tree->lock);
4849         if (ret) {
4850                 free_extent_map(em);
4851                 goto error;
4852         }
4853
4854         ret = btrfs_make_block_group(trans, extent_root, 0, type,
4855                                      BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4856                                      start, num_bytes);
4857         if (ret)
4858                 goto error_del_extent;
4859
4860         for (i = 0; i < map->num_stripes; i++) {
4861                 num_bytes = map->stripes[i].dev->bytes_used + stripe_size;
4862                 btrfs_device_set_bytes_used(map->stripes[i].dev, num_bytes);
4863         }
4864
4865         spin_lock(&extent_root->fs_info->free_chunk_lock);
4866         extent_root->fs_info->free_chunk_space -= (stripe_size *
4867                                                    map->num_stripes);
4868         spin_unlock(&extent_root->fs_info->free_chunk_lock);
4869
4870         free_extent_map(em);
4871         check_raid56_incompat_flag(extent_root->fs_info, type);
4872
4873         kfree(devices_info);
4874         return 0;
4875
4876 error_del_extent:
4877         write_lock(&em_tree->lock);
4878         remove_extent_mapping(em_tree, em);
4879         write_unlock(&em_tree->lock);
4880
4881         /* One for our allocation */
4882         free_extent_map(em);
4883         /* One for the tree reference */
4884         free_extent_map(em);
4885         /* One for the pending_chunks list reference */
4886         free_extent_map(em);
4887 error:
4888         kfree(devices_info);
4889         return ret;
4890 }
4891
4892 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans,
4893                                 struct btrfs_root *extent_root,
4894                                 u64 chunk_offset, u64 chunk_size)
4895 {
4896         struct btrfs_key key;
4897         struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
4898         struct btrfs_device *device;
4899         struct btrfs_chunk *chunk;
4900         struct btrfs_stripe *stripe;
4901         struct extent_map_tree *em_tree;
4902         struct extent_map *em;
4903         struct map_lookup *map;
4904         size_t item_size;
4905         u64 dev_offset;
4906         u64 stripe_size;
4907         int i = 0;
4908         int ret = 0;
4909
4910         em_tree = &extent_root->fs_info->mapping_tree.map_tree;
4911         read_lock(&em_tree->lock);
4912         em = lookup_extent_mapping(em_tree, chunk_offset, chunk_size);
4913         read_unlock(&em_tree->lock);
4914
4915         if (!em) {
4916                 btrfs_crit(extent_root->fs_info, "unable to find logical "
4917                            "%Lu len %Lu", chunk_offset, chunk_size);
4918                 return -EINVAL;
4919         }
4920
4921         if (em->start != chunk_offset || em->len != chunk_size) {
4922                 btrfs_crit(extent_root->fs_info, "found a bad mapping, wanted"
4923                           " %Lu-%Lu, found %Lu-%Lu", chunk_offset,
4924                           chunk_size, em->start, em->len);
4925                 free_extent_map(em);
4926                 return -EINVAL;
4927         }
4928
4929         map = em->map_lookup;
4930         item_size = btrfs_chunk_item_size(map->num_stripes);
4931         stripe_size = em->orig_block_len;
4932
4933         chunk = kzalloc(item_size, GFP_NOFS);
4934         if (!chunk) {
4935                 ret = -ENOMEM;
4936                 goto out;
4937         }
4938
4939         /*
4940          * Take the device list mutex to prevent races with the final phase of
4941          * a device replace operation that replaces the device object associated
4942          * with the map's stripes, because the device object's id can change
4943          * at any time during that final phase of the device replace operation
4944          * (dev-replace.c:btrfs_dev_replace_finishing()).
4945          */
4946         mutex_lock(&chunk_root->fs_info->fs_devices->device_list_mutex);
4947         for (i = 0; i < map->num_stripes; i++) {
4948                 device = map->stripes[i].dev;
4949                 dev_offset = map->stripes[i].physical;
4950
4951                 ret = btrfs_update_device(trans, device);
4952                 if (ret)
4953                         break;
4954                 ret = btrfs_alloc_dev_extent(trans, device,
4955                                              chunk_root->root_key.objectid,
4956                                              BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4957                                              chunk_offset, dev_offset,
4958                                              stripe_size);
4959                 if (ret)
4960                         break;
4961         }
4962         if (ret) {
4963                 mutex_unlock(&chunk_root->fs_info->fs_devices->device_list_mutex);
4964                 goto out;
4965         }
4966
4967         stripe = &chunk->stripe;
4968         for (i = 0; i < map->num_stripes; i++) {
4969                 device = map->stripes[i].dev;
4970                 dev_offset = map->stripes[i].physical;
4971
4972                 btrfs_set_stack_stripe_devid(stripe, device->devid);
4973                 btrfs_set_stack_stripe_offset(stripe, dev_offset);
4974                 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
4975                 stripe++;
4976         }
4977         mutex_unlock(&chunk_root->fs_info->fs_devices->device_list_mutex);
4978
4979         btrfs_set_stack_chunk_length(chunk, chunk_size);
4980         btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
4981         btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
4982         btrfs_set_stack_chunk_type(chunk, map->type);
4983         btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
4984         btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
4985         btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
4986         btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
4987         btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
4988
4989         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
4990         key.type = BTRFS_CHUNK_ITEM_KEY;
4991         key.offset = chunk_offset;
4992
4993         ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
4994         if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
4995                 /*
4996                  * TODO: Cleanup of inserted chunk root in case of
4997                  * failure.
4998                  */
4999                 ret = btrfs_add_system_chunk(chunk_root, &key, chunk,
5000                                              item_size);
5001         }
5002
5003 out:
5004         kfree(chunk);
5005         free_extent_map(em);
5006         return ret;
5007 }
5008
5009 /*
5010  * Chunk allocation falls into two parts. The first part does works
5011  * that make the new allocated chunk useable, but not do any operation
5012  * that modifies the chunk tree. The second part does the works that
5013  * require modifying the chunk tree. This division is important for the
5014  * bootstrap process of adding storage to a seed btrfs.
5015  */
5016 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
5017                       struct btrfs_root *extent_root, u64 type)
5018 {
5019         u64 chunk_offset;
5020
5021         ASSERT(mutex_is_locked(&extent_root->fs_info->chunk_mutex));
5022         chunk_offset = find_next_chunk(extent_root->fs_info);
5023         return __btrfs_alloc_chunk(trans, extent_root, chunk_offset, type);
5024 }
5025
5026 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
5027                                          struct btrfs_root *root,
5028                                          struct btrfs_device *device)
5029 {
5030         u64 chunk_offset;
5031         u64 sys_chunk_offset;
5032         u64 alloc_profile;
5033         struct btrfs_fs_info *fs_info = root->fs_info;
5034         struct btrfs_root *extent_root = fs_info->extent_root;
5035         int ret;
5036
5037         chunk_offset = find_next_chunk(fs_info);
5038         alloc_profile = btrfs_get_alloc_profile(extent_root, 0);
5039         ret = __btrfs_alloc_chunk(trans, extent_root, chunk_offset,
5040                                   alloc_profile);
5041         if (ret)
5042                 return ret;
5043
5044         sys_chunk_offset = find_next_chunk(root->fs_info);
5045         alloc_profile = btrfs_get_alloc_profile(fs_info->chunk_root, 0);
5046         ret = __btrfs_alloc_chunk(trans, extent_root, sys_chunk_offset,
5047                                   alloc_profile);
5048         return ret;
5049 }
5050
5051 static inline int btrfs_chunk_max_errors(struct map_lookup *map)
5052 {
5053         int max_errors;
5054
5055         if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
5056                          BTRFS_BLOCK_GROUP_RAID10 |
5057                          BTRFS_BLOCK_GROUP_RAID5 |
5058                          BTRFS_BLOCK_GROUP_DUP)) {
5059                 max_errors = 1;
5060         } else if (map->type & BTRFS_BLOCK_GROUP_RAID6) {
5061                 max_errors = 2;
5062         } else {
5063                 max_errors = 0;
5064         }
5065
5066         return max_errors;
5067 }
5068
5069 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
5070 {
5071         struct extent_map *em;
5072         struct map_lookup *map;
5073         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
5074         int readonly = 0;
5075         int miss_ndevs = 0;
5076         int i;
5077
5078         read_lock(&map_tree->map_tree.lock);
5079         em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
5080         read_unlock(&map_tree->map_tree.lock);
5081         if (!em)
5082                 return 1;
5083
5084         map = em->map_lookup;
5085         for (i = 0; i < map->num_stripes; i++) {
5086                 if (map->stripes[i].dev->missing) {
5087                         miss_ndevs++;
5088                         continue;
5089                 }
5090
5091                 if (!map->stripes[i].dev->writeable) {
5092                         readonly = 1;
5093                         goto end;
5094                 }
5095         }
5096
5097         /*
5098          * If the number of missing devices is larger than max errors,
5099          * we can not write the data into that chunk successfully, so
5100          * set it readonly.
5101          */
5102         if (miss_ndevs > btrfs_chunk_max_errors(map))
5103                 readonly = 1;
5104 end:
5105         free_extent_map(em);
5106         return readonly;
5107 }
5108
5109 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
5110 {
5111         extent_map_tree_init(&tree->map_tree);
5112 }
5113
5114 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
5115 {
5116         struct extent_map *em;
5117
5118         while (1) {
5119                 write_lock(&tree->map_tree.lock);
5120                 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
5121                 if (em)
5122                         remove_extent_mapping(&tree->map_tree, em);
5123                 write_unlock(&tree->map_tree.lock);
5124                 if (!em)
5125                         break;
5126                 /* once for us */
5127                 free_extent_map(em);
5128                 /* once for the tree */
5129                 free_extent_map(em);
5130         }
5131 }
5132
5133 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5134 {
5135         struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
5136         struct extent_map *em;
5137         struct map_lookup *map;
5138         struct extent_map_tree *em_tree = &map_tree->map_tree;
5139         int ret;
5140
5141         read_lock(&em_tree->lock);
5142         em = lookup_extent_mapping(em_tree, logical, len);
5143         read_unlock(&em_tree->lock);
5144
5145         /*
5146          * We could return errors for these cases, but that could get ugly and
5147          * we'd probably do the same thing which is just not do anything else
5148          * and exit, so return 1 so the callers don't try to use other copies.
5149          */
5150         if (!em) {
5151                 btrfs_crit(fs_info, "No mapping for %Lu-%Lu", logical,
5152                             logical+len);
5153                 return 1;
5154         }
5155
5156         if (em->start > logical || em->start + em->len < logical) {
5157                 btrfs_crit(fs_info, "Invalid mapping for %Lu-%Lu, got "
5158                             "%Lu-%Lu", logical, logical+len, em->start,
5159                             em->start + em->len);
5160                 free_extent_map(em);
5161                 return 1;
5162         }
5163
5164         map = em->map_lookup;
5165         if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
5166                 ret = map->num_stripes;
5167         else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5168                 ret = map->sub_stripes;
5169         else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
5170                 ret = 2;
5171         else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5172                 ret = 3;
5173         else
5174                 ret = 1;
5175         free_extent_map(em);
5176
5177         btrfs_dev_replace_lock(&fs_info->dev_replace, 0);
5178         if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace))
5179                 ret++;
5180         btrfs_dev_replace_unlock(&fs_info->dev_replace, 0);
5181
5182         return ret;
5183 }
5184
5185 unsigned long btrfs_full_stripe_len(struct btrfs_root *root,
5186                                     struct btrfs_mapping_tree *map_tree,
5187                                     u64 logical)
5188 {
5189         struct extent_map *em;
5190         struct map_lookup *map;
5191         struct extent_map_tree *em_tree = &map_tree->map_tree;
5192         unsigned long len = root->sectorsize;
5193
5194         read_lock(&em_tree->lock);
5195         em = lookup_extent_mapping(em_tree, logical, len);
5196         read_unlock(&em_tree->lock);
5197         BUG_ON(!em);
5198
5199         BUG_ON(em->start > logical || em->start + em->len < logical);
5200         map = em->map_lookup;
5201         if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5202                 len = map->stripe_len * nr_data_stripes(map);
5203         free_extent_map(em);
5204         return len;
5205 }
5206
5207 int btrfs_is_parity_mirror(struct btrfs_mapping_tree *map_tree,
5208                            u64 logical, u64 len, int mirror_num)
5209 {
5210         struct extent_map *em;
5211         struct map_lookup *map;
5212         struct extent_map_tree *em_tree = &map_tree->map_tree;
5213         int ret = 0;
5214
5215         read_lock(&em_tree->lock);
5216         em = lookup_extent_mapping(em_tree, logical, len);
5217         read_unlock(&em_tree->lock);
5218         BUG_ON(!em);
5219
5220         BUG_ON(em->start > logical || em->start + em->len < logical);
5221         map = em->map_lookup;
5222         if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5223                 ret = 1;
5224         free_extent_map(em);
5225         return ret;
5226 }
5227
5228 static int find_live_mirror(struct btrfs_fs_info *fs_info,
5229                             struct map_lookup *map, int first, int num,
5230                             int optimal, int dev_replace_is_ongoing)
5231 {
5232         int i;
5233         int tolerance;
5234         struct btrfs_device *srcdev;
5235
5236         if (dev_replace_is_ongoing &&
5237             fs_info->dev_replace.cont_reading_from_srcdev_mode ==
5238              BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
5239                 srcdev = fs_info->dev_replace.srcdev;
5240         else
5241                 srcdev = NULL;
5242
5243         /*
5244          * try to avoid the drive that is the source drive for a
5245          * dev-replace procedure, only choose it if no other non-missing
5246          * mirror is available
5247          */
5248         for (tolerance = 0; tolerance < 2; tolerance++) {
5249                 if (map->stripes[optimal].dev->bdev &&
5250                     (tolerance || map->stripes[optimal].dev != srcdev))
5251                         return optimal;
5252                 for (i = first; i < first + num; i++) {
5253                         if (map->stripes[i].dev->bdev &&
5254                             (tolerance || map->stripes[i].dev != srcdev))
5255                                 return i;
5256                 }
5257         }
5258
5259         /* we couldn't find one that doesn't fail.  Just return something
5260          * and the io error handling code will clean up eventually
5261          */
5262         return optimal;
5263 }
5264
5265 static inline int parity_smaller(u64 a, u64 b)
5266 {
5267         return a > b;
5268 }
5269
5270 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
5271 static void sort_parity_stripes(struct btrfs_bio *bbio, int num_stripes)
5272 {
5273         struct btrfs_bio_stripe s;
5274         int i;
5275         u64 l;
5276         int again = 1;
5277
5278         while (again) {
5279                 again = 0;
5280                 for (i = 0; i < num_stripes - 1; i++) {
5281                         if (parity_smaller(bbio->raid_map[i],
5282                                            bbio->raid_map[i+1])) {
5283                                 s = bbio->stripes[i];
5284                                 l = bbio->raid_map[i];
5285                                 bbio->stripes[i] = bbio->stripes[i+1];
5286                                 bbio->raid_map[i] = bbio->raid_map[i+1];
5287                                 bbio->stripes[i+1] = s;
5288                                 bbio->raid_map[i+1] = l;
5289
5290                                 again = 1;
5291                         }
5292                 }
5293         }
5294 }
5295
5296 static struct btrfs_bio *alloc_btrfs_bio(int total_stripes, int real_stripes)
5297 {
5298         struct btrfs_bio *bbio = kzalloc(
5299                  /* the size of the btrfs_bio */
5300                 sizeof(struct btrfs_bio) +
5301                 /* plus the variable array for the stripes */
5302                 sizeof(struct btrfs_bio_stripe) * (total_stripes) +
5303                 /* plus the variable array for the tgt dev */
5304                 sizeof(int) * (real_stripes) +
5305                 /*
5306                  * plus the raid_map, which includes both the tgt dev
5307                  * and the stripes
5308                  */
5309                 sizeof(u64) * (total_stripes),
5310                 GFP_NOFS|__GFP_NOFAIL);
5311
5312         atomic_set(&bbio->error, 0);
5313         atomic_set(&bbio->refs, 1);
5314
5315         return bbio;
5316 }
5317
5318 void btrfs_get_bbio(struct btrfs_bio *bbio)
5319 {
5320         WARN_ON(!atomic_read(&bbio->refs));
5321         atomic_inc(&bbio->refs);
5322 }
5323
5324 void btrfs_put_bbio(struct btrfs_bio *bbio)
5325 {
5326         if (!bbio)
5327                 return;
5328         if (atomic_dec_and_test(&bbio->refs))
5329                 kfree(bbio);
5330 }
5331
5332 static int __btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
5333                              u64 logical, u64 *length,
5334                              struct btrfs_bio **bbio_ret,
5335                              int mirror_num, int need_raid_map)
5336 {
5337         struct extent_map *em;
5338         struct map_lookup *map;
5339         struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
5340         struct extent_map_tree *em_tree = &map_tree->map_tree;
5341         u64 offset;
5342         u64 stripe_offset;
5343         u64 stripe_end_offset;
5344         u64 stripe_nr;
5345         u64 stripe_nr_orig;
5346         u64 stripe_nr_end;
5347         u64 stripe_len;
5348         u32 stripe_index;
5349         int i;
5350         int ret = 0;
5351         int num_stripes;
5352         int max_errors = 0;
5353         int tgtdev_indexes = 0;
5354         struct btrfs_bio *bbio = NULL;
5355         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
5356         int dev_replace_is_ongoing = 0;
5357         int num_alloc_stripes;
5358         int patch_the_first_stripe_for_dev_replace = 0;
5359         u64 physical_to_patch_in_first_stripe = 0;
5360         u64 raid56_full_stripe_start = (u64)-1;
5361
5362         read_lock(&em_tree->lock);
5363         em = lookup_extent_mapping(em_tree, logical, *length);
5364         read_unlock(&em_tree->lock);
5365
5366         if (!em) {
5367                 btrfs_crit(fs_info, "unable to find logical %llu len %llu",
5368                         logical, *length);
5369                 return -EINVAL;
5370         }
5371
5372         if (em->start > logical || em->start + em->len < logical) {
5373                 btrfs_crit(fs_info, "found a bad mapping, wanted %Lu, "
5374                            "found %Lu-%Lu", logical, em->start,
5375                            em->start + em->len);
5376                 free_extent_map(em);
5377                 return -EINVAL;
5378         }
5379
5380         map = em->map_lookup;
5381         offset = logical - em->start;
5382
5383         stripe_len = map->stripe_len;
5384         stripe_nr = offset;
5385         /*
5386          * stripe_nr counts the total number of stripes we have to stride
5387          * to get to this block
5388          */
5389         stripe_nr = div64_u64(stripe_nr, stripe_len);
5390
5391         stripe_offset = stripe_nr * stripe_len;
5392         if (offset < stripe_offset) {
5393                 btrfs_crit(fs_info, "stripe math has gone wrong, "
5394                            "stripe_offset=%llu, offset=%llu, start=%llu, "
5395                            "logical=%llu, stripe_len=%llu",
5396                            stripe_offset, offset, em->start, logical,
5397                            stripe_len);
5398                 free_extent_map(em);
5399                 return -EINVAL;
5400         }
5401
5402         /* stripe_offset is the offset of this block in its stripe*/
5403         stripe_offset = offset - stripe_offset;
5404
5405         /* if we're here for raid56, we need to know the stripe aligned start */
5406         if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5407                 unsigned long full_stripe_len = stripe_len * nr_data_stripes(map);
5408                 raid56_full_stripe_start = offset;
5409
5410                 /* allow a write of a full stripe, but make sure we don't
5411                  * allow straddling of stripes
5412                  */
5413                 raid56_full_stripe_start = div64_u64(raid56_full_stripe_start,
5414                                 full_stripe_len);
5415                 raid56_full_stripe_start *= full_stripe_len;
5416         }
5417
5418         if (rw & REQ_DISCARD) {
5419                 /* we don't discard raid56 yet */
5420                 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5421                         ret = -EOPNOTSUPP;
5422                         goto out;
5423                 }
5424                 *length = min_t(u64, em->len - offset, *length);
5425         } else if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
5426                 u64 max_len;
5427                 /* For writes to RAID[56], allow a full stripeset across all disks.
5428                    For other RAID types and for RAID[56] reads, just allow a single
5429                    stripe (on a single disk). */
5430                 if ((map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
5431                     (rw & REQ_WRITE)) {
5432                         max_len = stripe_len * nr_data_stripes(map) -
5433                                 (offset - raid56_full_stripe_start);
5434                 } else {
5435                         /* we limit the length of each bio to what fits in a stripe */
5436                         max_len = stripe_len - stripe_offset;
5437                 }
5438                 *length = min_t(u64, em->len - offset, max_len);
5439         } else {
5440                 *length = em->len - offset;
5441         }
5442
5443         /* This is for when we're called from btrfs_merge_bio_hook() and all
5444            it cares about is the length */
5445         if (!bbio_ret)
5446                 goto out;
5447
5448         btrfs_dev_replace_lock(dev_replace, 0);
5449         dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
5450         if (!dev_replace_is_ongoing)
5451                 btrfs_dev_replace_unlock(dev_replace, 0);
5452         else
5453                 btrfs_dev_replace_set_lock_blocking(dev_replace);
5454
5455         if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
5456             !(rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) &&
5457             dev_replace->tgtdev != NULL) {
5458                 /*
5459                  * in dev-replace case, for repair case (that's the only
5460                  * case where the mirror is selected explicitly when
5461                  * calling btrfs_map_block), blocks left of the left cursor
5462                  * can also be read from the target drive.
5463                  * For REQ_GET_READ_MIRRORS, the target drive is added as
5464                  * the last one to the array of stripes. For READ, it also
5465                  * needs to be supported using the same mirror number.
5466                  * If the requested block is not left of the left cursor,
5467                  * EIO is returned. This can happen because btrfs_num_copies()
5468                  * returns one more in the dev-replace case.
5469                  */
5470                 u64 tmp_length = *length;
5471                 struct btrfs_bio *tmp_bbio = NULL;
5472                 int tmp_num_stripes;
5473                 u64 srcdev_devid = dev_replace->srcdev->devid;
5474                 int index_srcdev = 0;
5475                 int found = 0;
5476                 u64 physical_of_found = 0;
5477
5478                 ret = __btrfs_map_block(fs_info, REQ_GET_READ_MIRRORS,
5479                              logical, &tmp_length, &tmp_bbio, 0, 0);
5480                 if (ret) {
5481                         WARN_ON(tmp_bbio != NULL);
5482                         goto out;
5483                 }
5484
5485                 tmp_num_stripes = tmp_bbio->num_stripes;
5486                 if (mirror_num > tmp_num_stripes) {
5487                         /*
5488                          * REQ_GET_READ_MIRRORS does not contain this
5489                          * mirror, that means that the requested area
5490                          * is not left of the left cursor
5491                          */
5492                         ret = -EIO;
5493                         btrfs_put_bbio(tmp_bbio);
5494                         goto out;
5495                 }
5496
5497                 /*
5498                  * process the rest of the function using the mirror_num
5499                  * of the source drive. Therefore look it up first.
5500                  * At the end, patch the device pointer to the one of the
5501                  * target drive.
5502                  */
5503                 for (i = 0; i < tmp_num_stripes; i++) {
5504                         if (tmp_bbio->stripes[i].dev->devid != srcdev_devid)
5505                                 continue;
5506
5507                         /*
5508                          * In case of DUP, in order to keep it simple, only add
5509                          * the mirror with the lowest physical address
5510                          */
5511                         if (found &&
5512                             physical_of_found <= tmp_bbio->stripes[i].physical)
5513                                 continue;
5514
5515                         index_srcdev = i;
5516                         found = 1;
5517                         physical_of_found = tmp_bbio->stripes[i].physical;
5518                 }
5519
5520                 btrfs_put_bbio(tmp_bbio);
5521
5522                 if (!found) {
5523                         WARN_ON(1);
5524                         ret = -EIO;
5525                         goto out;
5526                 }
5527
5528                 mirror_num = index_srcdev + 1;
5529                 patch_the_first_stripe_for_dev_replace = 1;
5530                 physical_to_patch_in_first_stripe = physical_of_found;
5531         } else if (mirror_num > map->num_stripes) {
5532                 mirror_num = 0;
5533         }
5534
5535         num_stripes = 1;
5536         stripe_index = 0;
5537         stripe_nr_orig = stripe_nr;
5538         stripe_nr_end = ALIGN(offset + *length, map->stripe_len);
5539         stripe_nr_end = div_u64(stripe_nr_end, map->stripe_len);
5540         stripe_end_offset = stripe_nr_end * map->stripe_len -
5541                             (offset + *length);
5542
5543         if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5544                 if (rw & REQ_DISCARD)
5545                         num_stripes = min_t(u64, map->num_stripes,
5546                                             stripe_nr_end - stripe_nr_orig);
5547                 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5548                                 &stripe_index);
5549                 if (!(rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)))
5550                         mirror_num = 1;
5551         } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
5552                 if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS))
5553                         num_stripes = map->num_stripes;
5554                 else if (mirror_num)
5555                         stripe_index = mirror_num - 1;
5556                 else {
5557                         stripe_index = find_live_mirror(fs_info, map, 0,
5558                                             map->num_stripes,
5559                                             current->pid % map->num_stripes,
5560                                             dev_replace_is_ongoing);
5561                         mirror_num = stripe_index + 1;
5562                 }
5563
5564         } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
5565                 if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) {
5566                         num_stripes = map->num_stripes;
5567                 } else if (mirror_num) {
5568                         stripe_index = mirror_num - 1;
5569                 } else {
5570                         mirror_num = 1;
5571                 }
5572
5573         } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5574                 u32 factor = map->num_stripes / map->sub_stripes;
5575
5576                 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
5577                 stripe_index *= map->sub_stripes;
5578
5579                 if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS))
5580                         num_stripes = map->sub_stripes;
5581                 else if (rw & REQ_DISCARD)
5582                         num_stripes = min_t(u64, map->sub_stripes *
5583                                             (stripe_nr_end - stripe_nr_orig),
5584                                             map->num_stripes);
5585                 else if (mirror_num)
5586                         stripe_index += mirror_num - 1;
5587                 else {
5588                         int old_stripe_index = stripe_index;
5589                         stripe_index = find_live_mirror(fs_info, map,
5590                                               stripe_index,
5591                                               map->sub_stripes, stripe_index +
5592                                               current->pid % map->sub_stripes,
5593                                               dev_replace_is_ongoing);
5594                         mirror_num = stripe_index - old_stripe_index + 1;
5595                 }
5596
5597         } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5598                 if (need_raid_map &&
5599                     ((rw & (REQ_WRITE | REQ_GET_READ_MIRRORS)) ||
5600                      mirror_num > 1)) {
5601                         /* push stripe_nr back to the start of the full stripe */
5602                         stripe_nr = div_u64(raid56_full_stripe_start,
5603                                         stripe_len * nr_data_stripes(map));
5604
5605                         /* RAID[56] write or recovery. Return all stripes */
5606                         num_stripes = map->num_stripes;
5607                         max_errors = nr_parity_stripes(map);
5608
5609                         *length = map->stripe_len;
5610                         stripe_index = 0;
5611                         stripe_offset = 0;
5612                 } else {
5613                         /*
5614                          * Mirror #0 or #1 means the original data block.
5615                          * Mirror #2 is RAID5 parity block.
5616                          * Mirror #3 is RAID6 Q block.
5617                          */
5618                         stripe_nr = div_u64_rem(stripe_nr,
5619                                         nr_data_stripes(map), &stripe_index);
5620                         if (mirror_num > 1)
5621                                 stripe_index = nr_data_stripes(map) +
5622                                                 mirror_num - 2;
5623
5624                         /* We distribute the parity blocks across stripes */
5625                         div_u64_rem(stripe_nr + stripe_index, map->num_stripes,
5626                                         &stripe_index);
5627                         if (!(rw & (REQ_WRITE | REQ_DISCARD |
5628                                     REQ_GET_READ_MIRRORS)) && mirror_num <= 1)
5629                                 mirror_num = 1;
5630                 }
5631         } else {
5632                 /*
5633                  * after this, stripe_nr is the number of stripes on this
5634                  * device we have to walk to find the data, and stripe_index is
5635                  * the number of our device in the stripe array
5636                  */
5637                 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5638                                 &stripe_index);
5639                 mirror_num = stripe_index + 1;
5640         }
5641         if (stripe_index >= map->num_stripes) {
5642                 btrfs_crit(fs_info, "stripe index math went horribly wrong, "
5643                            "got stripe_index=%u, num_stripes=%u",
5644                            stripe_index, map->num_stripes);
5645                 ret = -EINVAL;
5646                 goto out;
5647         }
5648
5649         num_alloc_stripes = num_stripes;
5650         if (dev_replace_is_ongoing) {
5651                 if (rw & (REQ_WRITE | REQ_DISCARD))
5652                         num_alloc_stripes <<= 1;
5653                 if (rw & REQ_GET_READ_MIRRORS)
5654                         num_alloc_stripes++;
5655                 tgtdev_indexes = num_stripes;
5656         }
5657
5658         bbio = alloc_btrfs_bio(num_alloc_stripes, tgtdev_indexes);
5659         if (!bbio) {
5660                 ret = -ENOMEM;
5661                 goto out;
5662         }
5663         if (dev_replace_is_ongoing)
5664                 bbio->tgtdev_map = (int *)(bbio->stripes + num_alloc_stripes);
5665
5666         /* build raid_map */
5667         if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK &&
5668             need_raid_map && ((rw & (REQ_WRITE | REQ_GET_READ_MIRRORS)) ||
5669             mirror_num > 1)) {
5670                 u64 tmp;
5671                 unsigned rot;
5672
5673                 bbio->raid_map = (u64 *)((void *)bbio->stripes +
5674                                  sizeof(struct btrfs_bio_stripe) *
5675                                  num_alloc_stripes +
5676                                  sizeof(int) * tgtdev_indexes);
5677
5678                 /* Work out the disk rotation on this stripe-set */
5679                 div_u64_rem(stripe_nr, num_stripes, &rot);
5680
5681                 /* Fill in the logical address of each stripe */
5682                 tmp = stripe_nr * nr_data_stripes(map);
5683                 for (i = 0; i < nr_data_stripes(map); i++)
5684                         bbio->raid_map[(i+rot) % num_stripes] =
5685                                 em->start + (tmp + i) * map->stripe_len;
5686
5687                 bbio->raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE;
5688                 if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5689                         bbio->raid_map[(i+rot+1) % num_stripes] =
5690                                 RAID6_Q_STRIPE;
5691         }
5692
5693         if (rw & REQ_DISCARD) {
5694                 u32 factor = 0;
5695                 u32 sub_stripes = 0;
5696                 u64 stripes_per_dev = 0;
5697                 u32 remaining_stripes = 0;
5698                 u32 last_stripe = 0;
5699
5700                 if (map->type &
5701                     (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID10)) {
5702                         if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5703                                 sub_stripes = 1;
5704                         else
5705                                 sub_stripes = map->sub_stripes;
5706
5707                         factor = map->num_stripes / sub_stripes;
5708                         stripes_per_dev = div_u64_rem(stripe_nr_end -
5709                                                       stripe_nr_orig,
5710                                                       factor,
5711                                                       &remaining_stripes);
5712                         div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
5713                         last_stripe *= sub_stripes;
5714                 }
5715
5716                 for (i = 0; i < num_stripes; i++) {
5717                         bbio->stripes[i].physical =
5718                                 map->stripes[stripe_index].physical +
5719                                 stripe_offset + stripe_nr * map->stripe_len;
5720                         bbio->stripes[i].dev = map->stripes[stripe_index].dev;
5721
5722                         if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5723                                          BTRFS_BLOCK_GROUP_RAID10)) {
5724                                 bbio->stripes[i].length = stripes_per_dev *
5725                                                           map->stripe_len;
5726
5727                                 if (i / sub_stripes < remaining_stripes)
5728                                         bbio->stripes[i].length +=
5729                                                 map->stripe_len;
5730
5731                                 /*
5732                                  * Special for the first stripe and
5733                                  * the last stripe:
5734                                  *
5735                                  * |-------|...|-------|
5736                                  *     |----------|
5737                                  *    off     end_off
5738                                  */
5739                                 if (i < sub_stripes)
5740                                         bbio->stripes[i].length -=
5741                                                 stripe_offset;
5742
5743                                 if (stripe_index >= last_stripe &&
5744                                     stripe_index <= (last_stripe +
5745                                                      sub_stripes - 1))
5746                                         bbio->stripes[i].length -=
5747                                                 stripe_end_offset;
5748
5749                                 if (i == sub_stripes - 1)
5750                                         stripe_offset = 0;
5751                         } else
5752                                 bbio->stripes[i].length = *length;
5753
5754                         stripe_index++;
5755                         if (stripe_index == map->num_stripes) {
5756                                 /* This could only happen for RAID0/10 */
5757                                 stripe_index = 0;
5758                                 stripe_nr++;
5759                         }
5760                 }
5761         } else {
5762                 for (i = 0; i < num_stripes; i++) {
5763                         bbio->stripes[i].physical =
5764                                 map->stripes[stripe_index].physical +
5765                                 stripe_offset +
5766                                 stripe_nr * map->stripe_len;
5767                         bbio->stripes[i].dev =
5768                                 map->stripes[stripe_index].dev;
5769                         stripe_index++;
5770                 }
5771         }
5772
5773         if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS))
5774                 max_errors = btrfs_chunk_max_errors(map);
5775
5776         if (bbio->raid_map)
5777                 sort_parity_stripes(bbio, num_stripes);
5778
5779         tgtdev_indexes = 0;
5780         if (dev_replace_is_ongoing && (rw & (REQ_WRITE | REQ_DISCARD)) &&
5781             dev_replace->tgtdev != NULL) {
5782                 int index_where_to_add;
5783                 u64 srcdev_devid = dev_replace->srcdev->devid;
5784
5785                 /*
5786                  * duplicate the write operations while the dev replace
5787                  * procedure is running. Since the copying of the old disk
5788                  * to the new disk takes place at run time while the
5789                  * filesystem is mounted writable, the regular write
5790                  * operations to the old disk have to be duplicated to go
5791                  * to the new disk as well.
5792                  * Note that device->missing is handled by the caller, and
5793                  * that the write to the old disk is already set up in the
5794                  * stripes array.
5795                  */
5796                 index_where_to_add = num_stripes;
5797                 for (i = 0; i < num_stripes; i++) {
5798                         if (bbio->stripes[i].dev->devid == srcdev_devid) {
5799                                 /* write to new disk, too */
5800                                 struct btrfs_bio_stripe *new =
5801                                         bbio->stripes + index_where_to_add;
5802                                 struct btrfs_bio_stripe *old =
5803                                         bbio->stripes + i;
5804
5805                                 new->physical = old->physical;
5806                                 new->length = old->length;
5807                                 new->dev = dev_replace->tgtdev;
5808                                 bbio->tgtdev_map[i] = index_where_to_add;
5809                                 index_where_to_add++;
5810                                 max_errors++;
5811                                 tgtdev_indexes++;
5812                         }
5813                 }
5814                 num_stripes = index_where_to_add;
5815         } else if (dev_replace_is_ongoing && (rw & REQ_GET_READ_MIRRORS) &&
5816                    dev_replace->tgtdev != NULL) {
5817                 u64 srcdev_devid = dev_replace->srcdev->devid;
5818                 int index_srcdev = 0;
5819                 int found = 0;
5820                 u64 physical_of_found = 0;
5821
5822                 /*
5823                  * During the dev-replace procedure, the target drive can
5824                  * also be used to read data in case it is needed to repair
5825                  * a corrupt block elsewhere. This is possible if the
5826                  * requested area is left of the left cursor. In this area,
5827                  * the target drive is a full copy of the source drive.
5828                  */
5829                 for (i = 0; i < num_stripes; i++) {
5830                         if (bbio->stripes[i].dev->devid == srcdev_devid) {
5831                                 /*
5832                                  * In case of DUP, in order to keep it
5833                                  * simple, only add the mirror with the
5834                                  * lowest physical address
5835                                  */
5836                                 if (found &&
5837                                     physical_of_found <=
5838                                      bbio->stripes[i].physical)
5839                                         continue;
5840                                 index_srcdev = i;
5841                                 found = 1;
5842                                 physical_of_found = bbio->stripes[i].physical;
5843                         }
5844                 }
5845                 if (found) {
5846                         struct btrfs_bio_stripe *tgtdev_stripe =
5847                                 bbio->stripes + num_stripes;
5848
5849                         tgtdev_stripe->physical = physical_of_found;
5850                         tgtdev_stripe->length =
5851                                 bbio->stripes[index_srcdev].length;
5852                         tgtdev_stripe->dev = dev_replace->tgtdev;
5853                         bbio->tgtdev_map[index_srcdev] = num_stripes;
5854
5855                         tgtdev_indexes++;
5856                         num_stripes++;
5857                 }
5858         }
5859
5860         *bbio_ret = bbio;
5861         bbio->map_type = map->type;
5862         bbio->num_stripes = num_stripes;
5863         bbio->max_errors = max_errors;
5864         bbio->mirror_num = mirror_num;
5865         bbio->num_tgtdevs = tgtdev_indexes;
5866
5867         /*
5868          * this is the case that REQ_READ && dev_replace_is_ongoing &&
5869          * mirror_num == num_stripes + 1 && dev_replace target drive is
5870          * available as a mirror
5871          */
5872         if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) {
5873                 WARN_ON(num_stripes > 1);
5874                 bbio->stripes[0].dev = dev_replace->tgtdev;
5875                 bbio->stripes[0].physical = physical_to_patch_in_first_stripe;
5876                 bbio->mirror_num = map->num_stripes + 1;
5877         }
5878 out:
5879         if (dev_replace_is_ongoing) {
5880                 btrfs_dev_replace_clear_lock_blocking(dev_replace);
5881                 btrfs_dev_replace_unlock(dev_replace, 0);
5882         }
5883         free_extent_map(em);
5884         return ret;
5885 }
5886
5887 int btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
5888                       u64 logical, u64 *length,
5889                       struct btrfs_bio **bbio_ret, int mirror_num)
5890 {
5891         return __btrfs_map_block(fs_info, rw, logical, length, bbio_ret,
5892                                  mirror_num, 0);
5893 }
5894
5895 /* For Scrub/replace */
5896 int btrfs_map_sblock(struct btrfs_fs_info *fs_info, int rw,
5897                      u64 logical, u64 *length,
5898                      struct btrfs_bio **bbio_ret, int mirror_num,
5899                      int need_raid_map)
5900 {
5901         return __btrfs_map_block(fs_info, rw, logical, length, bbio_ret,
5902                                  mirror_num, need_raid_map);
5903 }
5904
5905 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
5906                      u64 chunk_start, u64 physical, u64 devid,
5907                      u64 **logical, int *naddrs, int *stripe_len)
5908 {
5909         struct extent_map_tree *em_tree = &map_tree->map_tree;
5910         struct extent_map *em;
5911         struct map_lookup *map;
5912         u64 *buf;
5913         u64 bytenr;
5914         u64 length;
5915         u64 stripe_nr;
5916         u64 rmap_len;
5917         int i, j, nr = 0;
5918
5919         read_lock(&em_tree->lock);
5920         em = lookup_extent_mapping(em_tree, chunk_start, 1);
5921         read_unlock(&em_tree->lock);
5922
5923         if (!em) {
5924                 printk(KERN_ERR "BTRFS: couldn't find em for chunk %Lu\n",
5925                        chunk_start);
5926                 return -EIO;
5927         }
5928
5929         if (em->start != chunk_start) {
5930                 printk(KERN_ERR "BTRFS: bad chunk start, em=%Lu, wanted=%Lu\n",
5931                        em->start, chunk_start);
5932                 free_extent_map(em);
5933                 return -EIO;
5934         }
5935         map = em->map_lookup;
5936
5937         length = em->len;
5938         rmap_len = map->stripe_len;
5939
5940         if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5941                 length = div_u64(length, map->num_stripes / map->sub_stripes);
5942         else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5943                 length = div_u64(length, map->num_stripes);
5944         else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5945                 length = div_u64(length, nr_data_stripes(map));
5946                 rmap_len = map->stripe_len * nr_data_stripes(map);
5947         }
5948
5949         buf = kcalloc(map->num_stripes, sizeof(u64), GFP_NOFS);
5950         BUG_ON(!buf); /* -ENOMEM */
5951
5952         for (i = 0; i < map->num_stripes; i++) {
5953                 if (devid && map->stripes[i].dev->devid != devid)
5954                         continue;
5955                 if (map->stripes[i].physical > physical ||
5956                     map->stripes[i].physical + length <= physical)
5957                         continue;
5958
5959                 stripe_nr = physical - map->stripes[i].physical;
5960                 stripe_nr = div_u64(stripe_nr, map->stripe_len);
5961
5962                 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5963                         stripe_nr = stripe_nr * map->num_stripes + i;
5964                         stripe_nr = div_u64(stripe_nr, map->sub_stripes);
5965                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5966                         stripe_nr = stripe_nr * map->num_stripes + i;
5967                 } /* else if RAID[56], multiply by nr_data_stripes().
5968                    * Alternatively, just use rmap_len below instead of
5969                    * map->stripe_len */
5970
5971                 bytenr = chunk_start + stripe_nr * rmap_len;
5972                 WARN_ON(nr >= map->num_stripes);
5973                 for (j = 0; j < nr; j++) {
5974                         if (buf[j] == bytenr)
5975                                 break;
5976                 }
5977                 if (j == nr) {
5978                         WARN_ON(nr >= map->num_stripes);
5979                         buf[nr++] = bytenr;
5980                 }
5981         }
5982
5983         *logical = buf;
5984         *naddrs = nr;
5985         *stripe_len = rmap_len;
5986
5987         free_extent_map(em);
5988         return 0;
5989 }
5990
5991 static inline void btrfs_end_bbio(struct btrfs_bio *bbio, struct bio *bio)
5992 {
5993         bio->bi_private = bbio->private;
5994         bio->bi_end_io = bbio->end_io;
5995         bio_endio(bio);
5996
5997         btrfs_put_bbio(bbio);
5998 }
5999
6000 static void btrfs_end_bio(struct bio *bio)
6001 {
6002         struct btrfs_bio *bbio = bio->bi_private;
6003         int is_orig_bio = 0;
6004
6005         if (bio->bi_error) {
6006                 atomic_inc(&bbio->error);
6007                 if (bio->bi_error == -EIO || bio->bi_error == -EREMOTEIO) {
6008                         unsigned int stripe_index =
6009                                 btrfs_io_bio(bio)->stripe_index;
6010                         struct btrfs_device *dev;
6011
6012                         BUG_ON(stripe_index >= bbio->num_stripes);
6013                         dev = bbio->stripes[stripe_index].dev;
6014                         if (dev->bdev) {
6015                                 if (bio->bi_rw & WRITE)
6016                                         btrfs_dev_stat_inc(dev,
6017                                                 BTRFS_DEV_STAT_WRITE_ERRS);
6018                                 else
6019                                         btrfs_dev_stat_inc(dev,
6020                                                 BTRFS_DEV_STAT_READ_ERRS);
6021                                 if ((bio->bi_rw & WRITE_FLUSH) == WRITE_FLUSH)
6022                                         btrfs_dev_stat_inc(dev,
6023                                                 BTRFS_DEV_STAT_FLUSH_ERRS);
6024                                 btrfs_dev_stat_print_on_error(dev);
6025                         }
6026                 }
6027         }
6028
6029         if (bio == bbio->orig_bio)
6030                 is_orig_bio = 1;
6031
6032         btrfs_bio_counter_dec(bbio->fs_info);
6033
6034         if (atomic_dec_and_test(&bbio->stripes_pending)) {
6035                 if (!is_orig_bio) {
6036                         bio_put(bio);
6037                         bio = bbio->orig_bio;
6038                 }
6039
6040                 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
6041                 /* only send an error to the higher layers if it is
6042                  * beyond the tolerance of the btrfs bio
6043                  */
6044                 if (atomic_read(&bbio->error) > bbio->max_errors) {
6045                         bio->bi_error = -EIO;
6046                 } else {
6047                         /*
6048                          * this bio is actually up to date, we didn't
6049                          * go over the max number of errors
6050                          */
6051                         bio->bi_error = 0;
6052                 }
6053
6054                 btrfs_end_bbio(bbio, bio);
6055         } else if (!is_orig_bio) {
6056                 bio_put(bio);
6057         }
6058 }
6059
6060 /*
6061  * see run_scheduled_bios for a description of why bios are collected for
6062  * async submit.
6063  *
6064  * This will add one bio to the pending list for a device and make sure
6065  * the work struct is scheduled.
6066  */
6067 static noinline void btrfs_schedule_bio(struct btrfs_root *root,
6068                                         struct btrfs_device *device,
6069                                         int rw, struct bio *bio)
6070 {
6071         int should_queue = 1;
6072         struct btrfs_pending_bios *pending_bios;
6073
6074         if (device->missing || !device->bdev) {
6075                 bio_io_error(bio);
6076                 return;
6077         }
6078
6079         /* don't bother with additional async steps for reads, right now */
6080         if (!(rw & REQ_WRITE)) {
6081                 bio_get(bio);
6082                 btrfsic_submit_bio(rw, bio);
6083                 bio_put(bio);
6084                 return;
6085         }
6086
6087         /*
6088          * nr_async_bios allows us to reliably return congestion to the
6089          * higher layers.  Otherwise, the async bio makes it appear we have
6090          * made progress against dirty pages when we've really just put it
6091          * on a queue for later
6092          */
6093         atomic_inc(&root->fs_info->nr_async_bios);
6094         WARN_ON(bio->bi_next);
6095         bio->bi_next = NULL;
6096         bio->bi_rw |= rw;
6097
6098         spin_lock(&device->io_lock);
6099         if (bio->bi_rw & REQ_SYNC)
6100                 pending_bios = &device->pending_sync_bios;
6101         else
6102                 pending_bios = &device->pending_bios;
6103
6104         if (pending_bios->tail)
6105                 pending_bios->tail->bi_next = bio;
6106
6107         pending_bios->tail = bio;
6108         if (!pending_bios->head)
6109                 pending_bios->head = bio;
6110         if (device->running_pending)
6111                 should_queue = 0;
6112
6113         spin_unlock(&device->io_lock);
6114
6115         if (should_queue)
6116                 btrfs_queue_work(root->fs_info->submit_workers,
6117                                  &device->work);
6118 }
6119
6120 static void submit_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio,
6121                               struct bio *bio, u64 physical, int dev_nr,
6122                               int rw, int async)
6123 {
6124         struct btrfs_device *dev = bbio->stripes[dev_nr].dev;
6125
6126         bio->bi_private = bbio;
6127         btrfs_io_bio(bio)->stripe_index = dev_nr;
6128         bio->bi_end_io = btrfs_end_bio;
6129         bio->bi_iter.bi_sector = physical >> 9;
6130 #ifdef DEBUG
6131         {
6132                 struct rcu_string *name;
6133
6134                 rcu_read_lock();
6135                 name = rcu_dereference(dev->name);
6136                 pr_debug("btrfs_map_bio: rw %d, sector=%llu, dev=%lu "
6137                          "(%s id %llu), size=%u\n", rw,
6138                          (u64)bio->bi_iter.bi_sector, (u_long)dev->bdev->bd_dev,
6139                          name->str, dev->devid, bio->bi_iter.bi_size);
6140                 rcu_read_unlock();
6141         }
6142 #endif
6143         bio->bi_bdev = dev->bdev;
6144
6145         btrfs_bio_counter_inc_noblocked(root->fs_info);
6146
6147         if (async)
6148                 btrfs_schedule_bio(root, dev, rw, bio);
6149         else
6150                 btrfsic_submit_bio(rw, bio);
6151 }
6152
6153 static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical)
6154 {
6155         atomic_inc(&bbio->error);
6156         if (atomic_dec_and_test(&bbio->stripes_pending)) {
6157                 /* Should be the original bio. */
6158                 WARN_ON(bio != bbio->orig_bio);
6159
6160                 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
6161                 bio->bi_iter.bi_sector = logical >> 9;
6162                 bio->bi_error = -EIO;
6163                 btrfs_end_bbio(bbio, bio);
6164         }
6165 }
6166
6167 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
6168                   int mirror_num, int async_submit)
6169 {
6170         struct btrfs_device *dev;
6171         struct bio *first_bio = bio;
6172         u64 logical = (u64)bio->bi_iter.bi_sector << 9;
6173         u64 length = 0;
6174         u64 map_length;
6175         int ret;
6176         int dev_nr;
6177         int total_devs;
6178         struct btrfs_bio *bbio = NULL;
6179
6180         length = bio->bi_iter.bi_size;
6181         map_length = length;
6182
6183         btrfs_bio_counter_inc_blocked(root->fs_info);
6184         ret = __btrfs_map_block(root->fs_info, rw, logical, &map_length, &bbio,
6185                               mirror_num, 1);
6186         if (ret) {
6187                 btrfs_bio_counter_dec(root->fs_info);
6188                 return ret;
6189         }
6190
6191         total_devs = bbio->num_stripes;
6192         bbio->orig_bio = first_bio;
6193         bbio->private = first_bio->bi_private;
6194         bbio->end_io = first_bio->bi_end_io;
6195         bbio->fs_info = root->fs_info;
6196         atomic_set(&bbio->stripes_pending, bbio->num_stripes);
6197
6198         if ((bbio->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
6199             ((rw & WRITE) || (mirror_num > 1))) {
6200                 /* In this case, map_length has been set to the length of
6201                    a single stripe; not the whole write */
6202                 if (rw & WRITE) {
6203                         ret = raid56_parity_write(root, bio, bbio, map_length);
6204                 } else {
6205                         ret = raid56_parity_recover(root, bio, bbio, map_length,
6206                                                     mirror_num, 1);
6207                 }
6208
6209                 btrfs_bio_counter_dec(root->fs_info);
6210                 return ret;
6211         }
6212
6213         if (map_length < length) {
6214                 btrfs_crit(root->fs_info, "mapping failed logical %llu bio len %llu len %llu",
6215                         logical, length, map_length);
6216                 BUG();
6217         }
6218
6219         for (dev_nr = 0; dev_nr < total_devs; dev_nr++) {
6220                 dev = bbio->stripes[dev_nr].dev;
6221                 if (!dev || !dev->bdev || (rw & WRITE && !dev->writeable)) {
6222                         bbio_error(bbio, first_bio, logical);
6223                         continue;
6224                 }
6225
6226                 if (dev_nr < total_devs - 1) {
6227                         bio = btrfs_bio_clone(first_bio, GFP_NOFS);
6228                         BUG_ON(!bio); /* -ENOMEM */
6229                 } else
6230                         bio = first_bio;
6231
6232                 submit_stripe_bio(root, bbio, bio,
6233                                   bbio->stripes[dev_nr].physical, dev_nr, rw,
6234                                   async_submit);
6235         }
6236         btrfs_bio_counter_dec(root->fs_info);
6237         return 0;
6238 }
6239
6240 struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid,
6241                                        u8 *uuid, u8 *fsid)
6242 {
6243         struct btrfs_device *device;
6244         struct btrfs_fs_devices *cur_devices;
6245
6246         cur_devices = fs_info->fs_devices;
6247         while (cur_devices) {
6248                 if (!fsid ||
6249                     !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
6250                         device = __find_device(&cur_devices->devices,
6251                                                devid, uuid);
6252                         if (device)
6253                                 return device;
6254                 }
6255                 cur_devices = cur_devices->seed;
6256         }
6257         return NULL;
6258 }
6259
6260 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
6261                                             struct btrfs_fs_devices *fs_devices,
6262                                             u64 devid, u8 *dev_uuid)
6263 {
6264         struct btrfs_device *device;
6265
6266         device = btrfs_alloc_device(NULL, &devid, dev_uuid);
6267         if (IS_ERR(device))
6268                 return NULL;
6269
6270         list_add(&device->dev_list, &fs_devices->devices);
6271         device->fs_devices = fs_devices;
6272         fs_devices->num_devices++;
6273
6274         device->missing = 1;
6275         fs_devices->missing_devices++;
6276
6277         return device;
6278 }
6279
6280 /**
6281  * btrfs_alloc_device - allocate struct btrfs_device
6282  * @fs_info:    used only for generating a new devid, can be NULL if
6283  *              devid is provided (i.e. @devid != NULL).
6284  * @devid:      a pointer to devid for this device.  If NULL a new devid
6285  *              is generated.
6286  * @uuid:       a pointer to UUID for this device.  If NULL a new UUID
6287  *              is generated.
6288  *
6289  * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
6290  * on error.  Returned struct is not linked onto any lists and can be
6291  * destroyed with kfree() right away.
6292  */
6293 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
6294                                         const u64 *devid,
6295                                         const u8 *uuid)
6296 {
6297         struct btrfs_device *dev;
6298         u64 tmp;
6299
6300         if (WARN_ON(!devid && !fs_info))
6301                 return ERR_PTR(-EINVAL);
6302
6303         dev = __alloc_device();
6304         if (IS_ERR(dev))
6305                 return dev;
6306
6307         if (devid)
6308                 tmp = *devid;
6309         else {
6310                 int ret;
6311
6312                 ret = find_next_devid(fs_info, &tmp);
6313                 if (ret) {
6314                         kfree(dev);
6315                         return ERR_PTR(ret);
6316                 }
6317         }
6318         dev->devid = tmp;
6319
6320         if (uuid)
6321                 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE);
6322         else
6323                 generate_random_uuid(dev->uuid);
6324
6325         btrfs_init_work(&dev->work, btrfs_submit_helper,
6326                         pending_bios_fn, NULL, NULL);
6327
6328         return dev;
6329 }
6330
6331 /* Return -EIO if any error, otherwise return 0. */
6332 static int btrfs_check_chunk_valid(struct btrfs_root *root,
6333                                    struct extent_buffer *leaf,
6334                                    struct btrfs_chunk *chunk, u64 logical)
6335 {
6336         u64 length;
6337         u64 stripe_len;
6338         u16 num_stripes;
6339         u16 sub_stripes;
6340         u64 type;
6341
6342         length = btrfs_chunk_length(leaf, chunk);
6343         stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6344         num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6345         sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
6346         type = btrfs_chunk_type(leaf, chunk);
6347
6348         if (!num_stripes) {
6349                 btrfs_err(root->fs_info, "invalid chunk num_stripes: %u",
6350                           num_stripes);
6351                 return -EIO;
6352         }
6353         if (!IS_ALIGNED(logical, root->sectorsize)) {
6354                 btrfs_err(root->fs_info,
6355                           "invalid chunk logical %llu", logical);
6356                 return -EIO;
6357         }
6358         if (btrfs_chunk_sector_size(leaf, chunk) != root->sectorsize) {
6359                 btrfs_err(root->fs_info, "invalid chunk sectorsize %u",
6360                           btrfs_chunk_sector_size(leaf, chunk));
6361                 return -EIO;
6362         }
6363         if (!length || !IS_ALIGNED(length, root->sectorsize)) {
6364                 btrfs_err(root->fs_info,
6365                         "invalid chunk length %llu", length);
6366                 return -EIO;
6367         }
6368         if (!is_power_of_2(stripe_len) || stripe_len != BTRFS_STRIPE_LEN) {
6369                 btrfs_err(root->fs_info, "invalid chunk stripe length: %llu",
6370                           stripe_len);
6371                 return -EIO;
6372         }
6373         if (~(BTRFS_BLOCK_GROUP_TYPE_MASK | BTRFS_BLOCK_GROUP_PROFILE_MASK) &
6374             type) {
6375                 btrfs_err(root->fs_info, "unrecognized chunk type: %llu",
6376                           ~(BTRFS_BLOCK_GROUP_TYPE_MASK |
6377                             BTRFS_BLOCK_GROUP_PROFILE_MASK) &
6378                           btrfs_chunk_type(leaf, chunk));
6379                 return -EIO;
6380         }
6381         if ((type & BTRFS_BLOCK_GROUP_RAID10 && sub_stripes != 2) ||
6382             (type & BTRFS_BLOCK_GROUP_RAID1 && num_stripes < 1) ||
6383             (type & BTRFS_BLOCK_GROUP_RAID5 && num_stripes < 2) ||
6384             (type & BTRFS_BLOCK_GROUP_RAID6 && num_stripes < 3) ||
6385             (type & BTRFS_BLOCK_GROUP_DUP && num_stripes > 2) ||
6386             ((type & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 &&
6387              num_stripes != 1)) {
6388                 btrfs_err(root->fs_info,
6389                         "invalid num_stripes:sub_stripes %u:%u for profile %llu",
6390                         num_stripes, sub_stripes,
6391                         type & BTRFS_BLOCK_GROUP_PROFILE_MASK);
6392                 return -EIO;
6393         }
6394
6395         return 0;
6396 }
6397
6398 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
6399                           struct extent_buffer *leaf,
6400                           struct btrfs_chunk *chunk)
6401 {
6402         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
6403         struct map_lookup *map;
6404         struct extent_map *em;
6405         u64 logical;
6406         u64 length;
6407         u64 stripe_len;
6408         u64 devid;
6409         u8 uuid[BTRFS_UUID_SIZE];
6410         int num_stripes;
6411         int ret;
6412         int i;
6413
6414         logical = key->offset;
6415         length = btrfs_chunk_length(leaf, chunk);
6416         stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6417         num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6418
6419         ret = btrfs_check_chunk_valid(root, leaf, chunk, logical);
6420         if (ret)
6421                 return ret;
6422
6423         read_lock(&map_tree->map_tree.lock);
6424         em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
6425         read_unlock(&map_tree->map_tree.lock);
6426
6427         /* already mapped? */
6428         if (em && em->start <= logical && em->start + em->len > logical) {
6429                 free_extent_map(em);
6430                 return 0;
6431         } else if (em) {
6432                 free_extent_map(em);
6433         }
6434
6435         em = alloc_extent_map();
6436         if (!em)
6437                 return -ENOMEM;
6438         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
6439         if (!map) {
6440                 free_extent_map(em);
6441                 return -ENOMEM;
6442         }
6443
6444         set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
6445         em->map_lookup = map;
6446         em->start = logical;
6447         em->len = length;
6448         em->orig_start = 0;
6449         em->block_start = 0;
6450         em->block_len = em->len;
6451
6452         map->num_stripes = num_stripes;
6453         map->io_width = btrfs_chunk_io_width(leaf, chunk);
6454         map->io_align = btrfs_chunk_io_align(leaf, chunk);
6455         map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
6456         map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6457         map->type = btrfs_chunk_type(leaf, chunk);
6458         map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
6459         for (i = 0; i < num_stripes; i++) {
6460                 map->stripes[i].physical =
6461                         btrfs_stripe_offset_nr(leaf, chunk, i);
6462                 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
6463                 read_extent_buffer(leaf, uuid, (unsigned long)
6464                                    btrfs_stripe_dev_uuid_nr(chunk, i),
6465                                    BTRFS_UUID_SIZE);
6466                 map->stripes[i].dev = btrfs_find_device(root->fs_info, devid,
6467                                                         uuid, NULL);
6468                 if (!map->stripes[i].dev &&
6469                     !btrfs_test_opt(root->fs_info, DEGRADED)) {
6470                         free_extent_map(em);
6471                         return -EIO;
6472                 }
6473                 if (!map->stripes[i].dev) {
6474                         map->stripes[i].dev =
6475                                 add_missing_dev(root, root->fs_info->fs_devices,
6476                                                 devid, uuid);
6477                         if (!map->stripes[i].dev) {
6478                                 free_extent_map(em);
6479                                 return -EIO;
6480                         }
6481                         btrfs_warn(root->fs_info, "devid %llu uuid %pU is missing",
6482                                                 devid, uuid);
6483                 }
6484                 map->stripes[i].dev->in_fs_metadata = 1;
6485         }
6486
6487         write_lock(&map_tree->map_tree.lock);
6488         ret = add_extent_mapping(&map_tree->map_tree, em, 0);
6489         write_unlock(&map_tree->map_tree.lock);
6490         BUG_ON(ret); /* Tree corruption */
6491         free_extent_map(em);
6492
6493         return 0;
6494 }
6495
6496 static void fill_device_from_item(struct extent_buffer *leaf,
6497                                  struct btrfs_dev_item *dev_item,
6498                                  struct btrfs_device *device)
6499 {
6500         unsigned long ptr;
6501
6502         device->devid = btrfs_device_id(leaf, dev_item);
6503         device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
6504         device->total_bytes = device->disk_total_bytes;
6505         device->commit_total_bytes = device->disk_total_bytes;
6506         device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
6507         device->commit_bytes_used = device->bytes_used;
6508         device->type = btrfs_device_type(leaf, dev_item);
6509         device->io_align = btrfs_device_io_align(leaf, dev_item);
6510         device->io_width = btrfs_device_io_width(leaf, dev_item);
6511         device->sector_size = btrfs_device_sector_size(leaf, dev_item);
6512         WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
6513         device->is_tgtdev_for_dev_replace = 0;
6514
6515         ptr = btrfs_device_uuid(dev_item);
6516         read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
6517 }
6518
6519 static struct btrfs_fs_devices *open_seed_devices(struct btrfs_root *root,
6520                                                   u8 *fsid)
6521 {
6522         struct btrfs_fs_devices *fs_devices;
6523         int ret;
6524
6525         BUG_ON(!mutex_is_locked(&uuid_mutex));
6526
6527         fs_devices = root->fs_info->fs_devices->seed;
6528         while (fs_devices) {
6529                 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE))
6530                         return fs_devices;
6531
6532                 fs_devices = fs_devices->seed;
6533         }
6534
6535         fs_devices = find_fsid(fsid);
6536         if (!fs_devices) {
6537                 if (!btrfs_test_opt(root->fs_info, DEGRADED))
6538                         return ERR_PTR(-ENOENT);
6539
6540                 fs_devices = alloc_fs_devices(fsid);
6541                 if (IS_ERR(fs_devices))
6542                         return fs_devices;
6543
6544                 fs_devices->seeding = 1;
6545                 fs_devices->opened = 1;
6546                 return fs_devices;
6547         }
6548
6549         fs_devices = clone_fs_devices(fs_devices);
6550         if (IS_ERR(fs_devices))
6551                 return fs_devices;
6552
6553         ret = __btrfs_open_devices(fs_devices, FMODE_READ,
6554                                    root->fs_info->bdev_holder);
6555         if (ret) {
6556                 free_fs_devices(fs_devices);
6557                 fs_devices = ERR_PTR(ret);
6558                 goto out;
6559         }
6560
6561         if (!fs_devices->seeding) {
6562                 __btrfs_close_devices(fs_devices);
6563                 free_fs_devices(fs_devices);
6564                 fs_devices = ERR_PTR(-EINVAL);
6565                 goto out;
6566         }
6567
6568         fs_devices->seed = root->fs_info->fs_devices->seed;
6569         root->fs_info->fs_devices->seed = fs_devices;
6570 out:
6571         return fs_devices;
6572 }
6573
6574 static int read_one_dev(struct btrfs_root *root,
6575                         struct extent_buffer *leaf,
6576                         struct btrfs_dev_item *dev_item)
6577 {
6578         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
6579         struct btrfs_device *device;
6580         u64 devid;
6581         int ret;
6582         u8 fs_uuid[BTRFS_UUID_SIZE];
6583         u8 dev_uuid[BTRFS_UUID_SIZE];
6584
6585         devid = btrfs_device_id(leaf, dev_item);
6586         read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
6587                            BTRFS_UUID_SIZE);
6588         read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
6589                            BTRFS_UUID_SIZE);
6590
6591         if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
6592                 fs_devices = open_seed_devices(root, fs_uuid);
6593                 if (IS_ERR(fs_devices))
6594                         return PTR_ERR(fs_devices);
6595         }
6596
6597         device = btrfs_find_device(root->fs_info, devid, dev_uuid, fs_uuid);
6598         if (!device) {
6599                 if (!btrfs_test_opt(root->fs_info, DEGRADED))
6600                         return -EIO;
6601
6602                 device = add_missing_dev(root, fs_devices, devid, dev_uuid);
6603                 if (!device)
6604                         return -ENOMEM;
6605                 btrfs_warn(root->fs_info, "devid %llu uuid %pU missing",
6606                                 devid, dev_uuid);
6607         } else {
6608                 if (!device->bdev && !btrfs_test_opt(root->fs_info, DEGRADED))
6609                         return -EIO;
6610
6611                 if(!device->bdev && !device->missing) {
6612                         /*
6613                          * this happens when a device that was properly setup
6614                          * in the device info lists suddenly goes bad.
6615                          * device->bdev is NULL, and so we have to set
6616                          * device->missing to one here
6617                          */
6618                         device->fs_devices->missing_devices++;
6619                         device->missing = 1;
6620                 }
6621
6622                 /* Move the device to its own fs_devices */
6623                 if (device->fs_devices != fs_devices) {
6624                         ASSERT(device->missing);
6625
6626                         list_move(&device->dev_list, &fs_devices->devices);
6627                         device->fs_devices->num_devices--;
6628                         fs_devices->num_devices++;
6629
6630                         device->fs_devices->missing_devices--;
6631                         fs_devices->missing_devices++;
6632
6633                         device->fs_devices = fs_devices;
6634                 }
6635         }
6636
6637         if (device->fs_devices != root->fs_info->fs_devices) {
6638                 BUG_ON(device->writeable);
6639                 if (device->generation !=
6640                     btrfs_device_generation(leaf, dev_item))
6641                         return -EINVAL;
6642         }
6643
6644         fill_device_from_item(leaf, dev_item, device);
6645         device->in_fs_metadata = 1;
6646         if (device->writeable && !device->is_tgtdev_for_dev_replace) {
6647                 device->fs_devices->total_rw_bytes += device->total_bytes;
6648                 spin_lock(&root->fs_info->free_chunk_lock);
6649                 root->fs_info->free_chunk_space += device->total_bytes -
6650                         device->bytes_used;
6651                 spin_unlock(&root->fs_info->free_chunk_lock);
6652         }
6653         ret = 0;
6654         return ret;
6655 }
6656
6657 int btrfs_read_sys_array(struct btrfs_root *root)
6658 {
6659         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
6660         struct extent_buffer *sb;
6661         struct btrfs_disk_key *disk_key;
6662         struct btrfs_chunk *chunk;
6663         u8 *array_ptr;
6664         unsigned long sb_array_offset;
6665         int ret = 0;
6666         u32 num_stripes;
6667         u32 array_size;
6668         u32 len = 0;
6669         u32 cur_offset;
6670         u64 type;
6671         struct btrfs_key key;
6672
6673         ASSERT(BTRFS_SUPER_INFO_SIZE <= root->nodesize);
6674         /*
6675          * This will create extent buffer of nodesize, superblock size is
6676          * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will
6677          * overallocate but we can keep it as-is, only the first page is used.
6678          */
6679         sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET);
6680         if (IS_ERR(sb))
6681                 return PTR_ERR(sb);
6682         set_extent_buffer_uptodate(sb);
6683         btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
6684         /*
6685          * The sb extent buffer is artificial and just used to read the system array.
6686          * set_extent_buffer_uptodate() call does not properly mark all it's
6687          * pages up-to-date when the page is larger: extent does not cover the
6688          * whole page and consequently check_page_uptodate does not find all
6689          * the page's extents up-to-date (the hole beyond sb),
6690          * write_extent_buffer then triggers a WARN_ON.
6691          *
6692          * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
6693          * but sb spans only this function. Add an explicit SetPageUptodate call
6694          * to silence the warning eg. on PowerPC 64.
6695          */
6696         if (PAGE_SIZE > BTRFS_SUPER_INFO_SIZE)
6697                 SetPageUptodate(sb->pages[0]);
6698
6699         write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
6700         array_size = btrfs_super_sys_array_size(super_copy);
6701
6702         array_ptr = super_copy->sys_chunk_array;
6703         sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array);
6704         cur_offset = 0;
6705
6706         while (cur_offset < array_size) {
6707                 disk_key = (struct btrfs_disk_key *)array_ptr;
6708                 len = sizeof(*disk_key);
6709                 if (cur_offset + len > array_size)
6710                         goto out_short_read;
6711
6712                 btrfs_disk_key_to_cpu(&key, disk_key);
6713
6714                 array_ptr += len;
6715                 sb_array_offset += len;
6716                 cur_offset += len;
6717
6718                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
6719                         chunk = (struct btrfs_chunk *)sb_array_offset;
6720                         /*
6721                          * At least one btrfs_chunk with one stripe must be
6722                          * present, exact stripe count check comes afterwards
6723                          */
6724                         len = btrfs_chunk_item_size(1);
6725                         if (cur_offset + len > array_size)
6726                                 goto out_short_read;
6727
6728                         num_stripes = btrfs_chunk_num_stripes(sb, chunk);
6729                         if (!num_stripes) {
6730                                 printk(KERN_ERR
6731             "BTRFS: invalid number of stripes %u in sys_array at offset %u\n",
6732                                         num_stripes, cur_offset);
6733                                 ret = -EIO;
6734                                 break;
6735                         }
6736
6737                         type = btrfs_chunk_type(sb, chunk);
6738                         if ((type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) {
6739                                 btrfs_err(root->fs_info,
6740                             "invalid chunk type %llu in sys_array at offset %u",
6741                                         type, cur_offset);
6742                                 ret = -EIO;
6743                                 break;
6744                         }
6745
6746                         len = btrfs_chunk_item_size(num_stripes);
6747                         if (cur_offset + len > array_size)
6748                                 goto out_short_read;
6749
6750                         ret = read_one_chunk(root, &key, sb, chunk);
6751                         if (ret)
6752                                 break;
6753                 } else {
6754                         printk(KERN_ERR
6755                 "BTRFS: unexpected item type %u in sys_array at offset %u\n",
6756                                 (u32)key.type, cur_offset);
6757                         ret = -EIO;
6758                         break;
6759                 }
6760                 array_ptr += len;
6761                 sb_array_offset += len;
6762                 cur_offset += len;
6763         }
6764         clear_extent_buffer_uptodate(sb);
6765         free_extent_buffer_stale(sb);
6766         return ret;
6767
6768 out_short_read:
6769         printk(KERN_ERR "BTRFS: sys_array too short to read %u bytes at offset %u\n",
6770                         len, cur_offset);
6771         clear_extent_buffer_uptodate(sb);
6772         free_extent_buffer_stale(sb);
6773         return -EIO;
6774 }
6775
6776 int btrfs_read_chunk_tree(struct btrfs_root *root)
6777 {
6778         struct btrfs_path *path;
6779         struct extent_buffer *leaf;
6780         struct btrfs_key key;
6781         struct btrfs_key found_key;
6782         int ret;
6783         int slot;
6784         u64 total_dev = 0;
6785
6786         root = root->fs_info->chunk_root;
6787
6788         path = btrfs_alloc_path();
6789         if (!path)
6790                 return -ENOMEM;
6791
6792         mutex_lock(&uuid_mutex);
6793         lock_chunks(root);
6794
6795         /*
6796          * Read all device items, and then all the chunk items. All
6797          * device items are found before any chunk item (their object id
6798          * is smaller than the lowest possible object id for a chunk
6799          * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
6800          */
6801         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
6802         key.offset = 0;
6803         key.type = 0;
6804         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6805         if (ret < 0)
6806                 goto error;
6807         while (1) {
6808                 leaf = path->nodes[0];
6809                 slot = path->slots[0];
6810                 if (slot >= btrfs_header_nritems(leaf)) {
6811                         ret = btrfs_next_leaf(root, path);
6812                         if (ret == 0)
6813                                 continue;
6814                         if (ret < 0)
6815                                 goto error;
6816                         break;
6817                 }
6818                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6819                 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
6820                         struct btrfs_dev_item *dev_item;
6821                         dev_item = btrfs_item_ptr(leaf, slot,
6822                                                   struct btrfs_dev_item);
6823                         ret = read_one_dev(root, leaf, dev_item);
6824                         if (ret)
6825                                 goto error;
6826                         total_dev++;
6827                 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
6828                         struct btrfs_chunk *chunk;
6829                         chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
6830                         ret = read_one_chunk(root, &found_key, leaf, chunk);
6831                         if (ret)
6832                                 goto error;
6833                 }
6834                 path->slots[0]++;
6835         }
6836
6837         /*
6838          * After loading chunk tree, we've got all device information,
6839          * do another round of validation checks.
6840          */
6841         if (total_dev != root->fs_info->fs_devices->total_devices) {
6842                 btrfs_err(root->fs_info,
6843            "super_num_devices %llu mismatch with num_devices %llu found here",
6844                           btrfs_super_num_devices(root->fs_info->super_copy),
6845                           total_dev);
6846                 ret = -EINVAL;
6847                 goto error;
6848         }
6849         if (btrfs_super_total_bytes(root->fs_info->super_copy) <
6850             root->fs_info->fs_devices->total_rw_bytes) {
6851                 btrfs_err(root->fs_info,
6852         "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu",
6853                           btrfs_super_total_bytes(root->fs_info->super_copy),
6854                           root->fs_info->fs_devices->total_rw_bytes);
6855                 ret = -EINVAL;
6856                 goto error;
6857         }
6858         ret = 0;
6859 error:
6860         unlock_chunks(root);
6861         mutex_unlock(&uuid_mutex);
6862
6863         btrfs_free_path(path);
6864         return ret;
6865 }
6866
6867 void btrfs_init_devices_late(struct btrfs_fs_info *fs_info)
6868 {
6869         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6870         struct btrfs_device *device;
6871
6872         while (fs_devices) {
6873                 mutex_lock(&fs_devices->device_list_mutex);
6874                 list_for_each_entry(device, &fs_devices->devices, dev_list)
6875                         device->dev_root = fs_info->dev_root;
6876                 mutex_unlock(&fs_devices->device_list_mutex);
6877
6878                 fs_devices = fs_devices->seed;
6879         }
6880 }
6881
6882 static void __btrfs_reset_dev_stats(struct btrfs_device *dev)
6883 {
6884         int i;
6885
6886         for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6887                 btrfs_dev_stat_reset(dev, i);
6888 }
6889
6890 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
6891 {
6892         struct btrfs_key key;
6893         struct btrfs_key found_key;
6894         struct btrfs_root *dev_root = fs_info->dev_root;
6895         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6896         struct extent_buffer *eb;
6897         int slot;
6898         int ret = 0;
6899         struct btrfs_device *device;
6900         struct btrfs_path *path = NULL;
6901         int i;
6902
6903         path = btrfs_alloc_path();
6904         if (!path) {
6905                 ret = -ENOMEM;
6906                 goto out;
6907         }
6908
6909         mutex_lock(&fs_devices->device_list_mutex);
6910         list_for_each_entry(device, &fs_devices->devices, dev_list) {
6911                 int item_size;
6912                 struct btrfs_dev_stats_item *ptr;
6913
6914                 key.objectid = BTRFS_DEV_STATS_OBJECTID;
6915                 key.type = BTRFS_PERSISTENT_ITEM_KEY;
6916                 key.offset = device->devid;
6917                 ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
6918                 if (ret) {
6919                         __btrfs_reset_dev_stats(device);
6920                         device->dev_stats_valid = 1;
6921                         btrfs_release_path(path);
6922                         continue;
6923                 }
6924                 slot = path->slots[0];
6925                 eb = path->nodes[0];
6926                 btrfs_item_key_to_cpu(eb, &found_key, slot);
6927                 item_size = btrfs_item_size_nr(eb, slot);
6928
6929                 ptr = btrfs_item_ptr(eb, slot,
6930                                      struct btrfs_dev_stats_item);
6931
6932                 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
6933                         if (item_size >= (1 + i) * sizeof(__le64))
6934                                 btrfs_dev_stat_set(device, i,
6935                                         btrfs_dev_stats_value(eb, ptr, i));
6936                         else
6937                                 btrfs_dev_stat_reset(device, i);
6938                 }
6939
6940                 device->dev_stats_valid = 1;
6941                 btrfs_dev_stat_print_on_load(device);
6942                 btrfs_release_path(path);
6943         }
6944         mutex_unlock(&fs_devices->device_list_mutex);
6945
6946 out:
6947         btrfs_free_path(path);
6948         return ret < 0 ? ret : 0;
6949 }
6950
6951 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
6952                                 struct btrfs_root *dev_root,
6953                                 struct btrfs_device *device)
6954 {
6955         struct btrfs_path *path;
6956         struct btrfs_key key;
6957         struct extent_buffer *eb;
6958         struct btrfs_dev_stats_item *ptr;
6959         int ret;
6960         int i;
6961
6962         key.objectid = BTRFS_DEV_STATS_OBJECTID;
6963         key.type = BTRFS_PERSISTENT_ITEM_KEY;
6964         key.offset = device->devid;
6965
6966         path = btrfs_alloc_path();
6967         BUG_ON(!path);
6968         ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
6969         if (ret < 0) {
6970                 btrfs_warn_in_rcu(dev_root->fs_info,
6971                         "error %d while searching for dev_stats item for device %s",
6972                               ret, rcu_str_deref(device->name));
6973                 goto out;
6974         }
6975
6976         if (ret == 0 &&
6977             btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
6978                 /* need to delete old one and insert a new one */
6979                 ret = btrfs_del_item(trans, dev_root, path);
6980                 if (ret != 0) {
6981                         btrfs_warn_in_rcu(dev_root->fs_info,
6982                                 "delete too small dev_stats item for device %s failed %d",
6983                                       rcu_str_deref(device->name), ret);
6984                         goto out;
6985                 }
6986                 ret = 1;
6987         }
6988
6989         if (ret == 1) {
6990                 /* need to insert a new item */
6991                 btrfs_release_path(path);
6992                 ret = btrfs_insert_empty_item(trans, dev_root, path,
6993                                               &key, sizeof(*ptr));
6994                 if (ret < 0) {
6995                         btrfs_warn_in_rcu(dev_root->fs_info,
6996                                 "insert dev_stats item for device %s failed %d",
6997                                 rcu_str_deref(device->name), ret);
6998                         goto out;
6999                 }
7000         }
7001
7002         eb = path->nodes[0];
7003         ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
7004         for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7005                 btrfs_set_dev_stats_value(eb, ptr, i,
7006                                           btrfs_dev_stat_read(device, i));
7007         btrfs_mark_buffer_dirty(eb);
7008
7009 out:
7010         btrfs_free_path(path);
7011         return ret;
7012 }
7013
7014 /*
7015  * called from commit_transaction. Writes all changed device stats to disk.
7016  */
7017 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans,
7018                         struct btrfs_fs_info *fs_info)
7019 {
7020         struct btrfs_root *dev_root = fs_info->dev_root;
7021         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7022         struct btrfs_device *device;
7023         int stats_cnt;
7024         int ret = 0;
7025
7026         mutex_lock(&fs_devices->device_list_mutex);
7027         list_for_each_entry(device, &fs_devices->devices, dev_list) {
7028                 if (!device->dev_stats_valid || !btrfs_dev_stats_dirty(device))
7029                         continue;
7030
7031                 stats_cnt = atomic_read(&device->dev_stats_ccnt);
7032                 ret = update_dev_stat_item(trans, dev_root, device);
7033                 if (!ret)
7034                         atomic_sub(stats_cnt, &device->dev_stats_ccnt);
7035         }
7036         mutex_unlock(&fs_devices->device_list_mutex);
7037
7038         return ret;
7039 }
7040
7041 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
7042 {
7043         btrfs_dev_stat_inc(dev, index);
7044         btrfs_dev_stat_print_on_error(dev);
7045 }
7046
7047 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
7048 {
7049         if (!dev->dev_stats_valid)
7050                 return;
7051         btrfs_err_rl_in_rcu(dev->dev_root->fs_info,
7052                 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7053                            rcu_str_deref(dev->name),
7054                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7055                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7056                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7057                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7058                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7059 }
7060
7061 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
7062 {
7063         int i;
7064
7065         for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7066                 if (btrfs_dev_stat_read(dev, i) != 0)
7067                         break;
7068         if (i == BTRFS_DEV_STAT_VALUES_MAX)
7069                 return; /* all values == 0, suppress message */
7070
7071         btrfs_info_in_rcu(dev->dev_root->fs_info,
7072                 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7073                rcu_str_deref(dev->name),
7074                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7075                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7076                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7077                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7078                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7079 }
7080
7081 int btrfs_get_dev_stats(struct btrfs_root *root,
7082                         struct btrfs_ioctl_get_dev_stats *stats)
7083 {
7084         struct btrfs_device *dev;
7085         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
7086         int i;
7087
7088         mutex_lock(&fs_devices->device_list_mutex);
7089         dev = btrfs_find_device(root->fs_info, stats->devid, NULL, NULL);
7090         mutex_unlock(&fs_devices->device_list_mutex);
7091
7092         if (!dev) {
7093                 btrfs_warn(root->fs_info, "get dev_stats failed, device not found");
7094                 return -ENODEV;
7095         } else if (!dev->dev_stats_valid) {
7096                 btrfs_warn(root->fs_info, "get dev_stats failed, not yet valid");
7097                 return -ENODEV;
7098         } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
7099                 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7100                         if (stats->nr_items > i)
7101                                 stats->values[i] =
7102                                         btrfs_dev_stat_read_and_reset(dev, i);
7103                         else
7104                                 btrfs_dev_stat_reset(dev, i);
7105                 }
7106         } else {
7107                 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7108                         if (stats->nr_items > i)
7109                                 stats->values[i] = btrfs_dev_stat_read(dev, i);
7110         }
7111         if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
7112                 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
7113         return 0;
7114 }
7115
7116 void btrfs_scratch_superblocks(struct block_device *bdev, char *device_path)
7117 {
7118         struct buffer_head *bh;
7119         struct btrfs_super_block *disk_super;
7120         int copy_num;
7121
7122         if (!bdev)
7123                 return;
7124
7125         for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX;
7126                 copy_num++) {
7127
7128                 if (btrfs_read_dev_one_super(bdev, copy_num, &bh))
7129                         continue;
7130
7131                 disk_super = (struct btrfs_super_block *)bh->b_data;
7132
7133                 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
7134                 set_buffer_dirty(bh);
7135                 sync_dirty_buffer(bh);
7136                 brelse(bh);
7137         }
7138
7139         /* Notify udev that device has changed */
7140         btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
7141
7142         /* Update ctime/mtime for device path for libblkid */
7143         update_dev_time(device_path);
7144 }
7145
7146 /*
7147  * Update the size of all devices, which is used for writing out the
7148  * super blocks.
7149  */
7150 void btrfs_update_commit_device_size(struct btrfs_fs_info *fs_info)
7151 {
7152         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7153         struct btrfs_device *curr, *next;
7154
7155         if (list_empty(&fs_devices->resized_devices))
7156                 return;
7157
7158         mutex_lock(&fs_devices->device_list_mutex);
7159         lock_chunks(fs_info->dev_root);
7160         list_for_each_entry_safe(curr, next, &fs_devices->resized_devices,
7161                                  resized_list) {
7162                 list_del_init(&curr->resized_list);
7163                 curr->commit_total_bytes = curr->disk_total_bytes;
7164         }
7165         unlock_chunks(fs_info->dev_root);
7166         mutex_unlock(&fs_devices->device_list_mutex);
7167 }
7168
7169 /* Must be invoked during the transaction commit */
7170 void btrfs_update_commit_device_bytes_used(struct btrfs_root *root,
7171                                         struct btrfs_transaction *transaction)
7172 {
7173         struct extent_map *em;
7174         struct map_lookup *map;
7175         struct btrfs_device *dev;
7176         int i;
7177
7178         if (list_empty(&transaction->pending_chunks))
7179                 return;
7180
7181         /* In order to kick the device replace finish process */
7182         lock_chunks(root);
7183         list_for_each_entry(em, &transaction->pending_chunks, list) {
7184                 map = em->map_lookup;
7185
7186                 for (i = 0; i < map->num_stripes; i++) {
7187                         dev = map->stripes[i].dev;
7188                         dev->commit_bytes_used = dev->bytes_used;
7189                 }
7190         }
7191         unlock_chunks(root);
7192 }
7193
7194 void btrfs_set_fs_info_ptr(struct btrfs_fs_info *fs_info)
7195 {
7196         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7197         while (fs_devices) {
7198                 fs_devices->fs_info = fs_info;
7199                 fs_devices = fs_devices->seed;
7200         }
7201 }
7202
7203 void btrfs_reset_fs_info_ptr(struct btrfs_fs_info *fs_info)
7204 {
7205         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7206         while (fs_devices) {
7207                 fs_devices->fs_info = NULL;
7208                 fs_devices = fs_devices->seed;
7209         }
7210 }