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_fs_info *fs_info);
 138 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info);
 139 static void __btrfs_reset_dev_stats(struct btrfs_device *dev);
 140 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev);
 141 static void btrfs_dev_stat_print_on_load(struct btrfs_device *device);
 142 static int __btrfs_map_block(struct btrfs_fs_info *fs_info,
 143                              enum btrfs_map_op op,
 144                              u64 logical, u64 *length,
 145                              struct btrfs_bio **bbio_ret,
 146                              int mirror_num, int need_raid_map);
 147
 148 DEFINE_MUTEX(uuid_mutex);
 149 static LIST_HEAD(fs_uuids);
 150 struct list_head *btrfs_get_fs_uuids(void)
 151 {
 152         return &fs_uuids;
 153 }
 154
 155 static struct btrfs_fs_devices *__alloc_fs_devices(void)
 156 {
 157         struct btrfs_fs_devices *fs_devs;
 158
 159         fs_devs = kzalloc(sizeof(*fs_devs), GFP_KERNEL);
 160         if (!fs_devs)
 161                 return ERR_PTR(-ENOMEM);
 162
 163         mutex_init(&fs_devs->device_list_mutex);
 164
 165         INIT_LIST_HEAD(&fs_devs->devices);
 166         INIT_LIST_HEAD(&fs_devs->resized_devices);
 167         INIT_LIST_HEAD(&fs_devs->alloc_list);
 168         INIT_LIST_HEAD(&fs_devs->list);
 169
 170         return fs_devs;
 171 }
 172
 173 /**
 174  * alloc_fs_devices - allocate struct btrfs_fs_devices
 175  * @fsid:       a pointer to UUID for this FS.  If NULL a new UUID is
 176  *              generated.
 177  *
 178  * Return: a pointer to a new &struct btrfs_fs_devices on success;
 179  * ERR_PTR() on error.  Returned struct is not linked onto any lists and
 180  * can be destroyed with kfree() right away.
 181  */
 182 static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid)
 183 {
 184         struct btrfs_fs_devices *fs_devs;
 185
 186         fs_devs = __alloc_fs_devices();
 187         if (IS_ERR(fs_devs))
 188                 return fs_devs;
 189
 190         if (fsid)
 191                 memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE);
 192         else
 193                 generate_random_uuid(fs_devs->fsid);
 194
 195         return fs_devs;
 196 }
 197
 198 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
 199 {
 200         struct btrfs_device *device;
 201         WARN_ON(fs_devices->opened);
 202         while (!list_empty(&fs_devices->devices)) {
 203                 device = list_entry(fs_devices->devices.next,
 204                                     struct btrfs_device, dev_list);
 205                 list_del(&device->dev_list);
 206                 rcu_string_free(device->name);
 207                 kfree(device);
 208         }
 209         kfree(fs_devices);
 210 }
 211
 212 static void btrfs_kobject_uevent(struct block_device *bdev,
 213                                  enum kobject_action action)
 214 {
 215         int ret;
 216
 217         ret = kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, action);
 218         if (ret)
 219                 pr_warn("BTRFS: Sending event '%d' to kobject: '%s' (%p): failed\n",
 220                         action,
 221                         kobject_name(&disk_to_dev(bdev->bd_disk)->kobj),
 222                         &disk_to_dev(bdev->bd_disk)->kobj);
 223 }
 224
 225 void btrfs_cleanup_fs_uuids(void)
 226 {
 227         struct btrfs_fs_devices *fs_devices;
 228
 229         while (!list_empty(&fs_uuids)) {
 230                 fs_devices = list_entry(fs_uuids.next,
 231                                         struct btrfs_fs_devices, list);
 232                 list_del(&fs_devices->list);
 233                 free_fs_devices(fs_devices);
 234         }
 235 }
 236
 237 static struct btrfs_device *__alloc_device(void)
 238 {
 239         struct btrfs_device *dev;
 240
 241         dev = kzalloc(sizeof(*dev), GFP_KERNEL);
 242         if (!dev)
 243                 return ERR_PTR(-ENOMEM);
 244
 245         INIT_LIST_HEAD(&dev->dev_list);
 246         INIT_LIST_HEAD(&dev->dev_alloc_list);
 247         INIT_LIST_HEAD(&dev->resized_list);
 248
 249         spin_lock_init(&dev->io_lock);
 250
 251         spin_lock_init(&dev->reada_lock);
 252         atomic_set(&dev->reada_in_flight, 0);
 253         atomic_set(&dev->dev_stats_ccnt, 0);
 254         btrfs_device_data_ordered_init(dev);
 255         INIT_RADIX_TREE(&dev->reada_zones, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
 256         INIT_RADIX_TREE(&dev->reada_extents, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
 257
 258         return dev;
 259 }
 260
 261 static noinline struct btrfs_device *__find_device(struct list_head *head,
 262                                                    u64 devid, u8 *uuid)
 263 {
 264         struct btrfs_device *dev;
 265
 266         list_for_each_entry(dev, head, dev_list) {
 267                 if (dev->devid == devid &&
 268                     (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
 269                         return dev;
 270                 }
 271         }
 272         return NULL;
 273 }
 274
 275 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
 276 {
 277         struct btrfs_fs_devices *fs_devices;
 278
 279         list_for_each_entry(fs_devices, &fs_uuids, list) {
 280                 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
 281                         return fs_devices;
 282         }
 283         return NULL;
 284 }
 285
 286 static int
 287 btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder,
 288                       int flush, struct block_device **bdev,
 289                       struct buffer_head **bh)
 290 {
 291         int ret;
 292
 293         *bdev = blkdev_get_by_path(device_path, flags, holder);
 294
 295         if (IS_ERR(*bdev)) {
 296                 ret = PTR_ERR(*bdev);
 297                 goto error;
 298         }
 299
 300         if (flush)
 301                 filemap_write_and_wait((*bdev)->bd_inode->i_mapping);
 302         ret = set_blocksize(*bdev, 4096);
 303         if (ret) {
 304                 blkdev_put(*bdev, flags);
 305                 goto error;
 306         }
 307         invalidate_bdev(*bdev);
 308         *bh = btrfs_read_dev_super(*bdev);
 309         if (IS_ERR(*bh)) {
 310                 ret = PTR_ERR(*bh);
 311                 blkdev_put(*bdev, flags);
 312                 goto error;
 313         }
 314
 315         return 0;
 316
 317 error:
 318         *bdev = NULL;
 319         *bh = NULL;
 320         return ret;
 321 }
 322
 323 static void requeue_list(struct btrfs_pending_bios *pending_bios,
 324                         struct bio *head, struct bio *tail)
 325 {
 326
 327         struct bio *old_head;
 328
 329         old_head = pending_bios->head;
 330         pending_bios->head = head;
 331         if (pending_bios->tail)
 332                 tail->bi_next = old_head;
 333         else
 334                 pending_bios->tail = tail;
 335 }
 336
 337 /*
 338  * we try to collect pending bios for a device so we don't get a large
 339  * number of procs sending bios down to the same device.  This greatly
 340  * improves the schedulers ability to collect and merge the bios.
 341  *
 342  * But, it also turns into a long list of bios to process and that is sure
 343  * to eventually make the worker thread block.  The solution here is to
 344  * make some progress and then put this work struct back at the end of
 345  * the list if the block device is congested.  This way, multiple devices
 346  * can make progress from a single worker thread.
 347  */
 348 static noinline void run_scheduled_bios(struct btrfs_device *device)
 349 {
 350         struct btrfs_fs_info *fs_info = device->fs_info;
 351         struct bio *pending;
 352         struct backing_dev_info *bdi;
 353         struct btrfs_pending_bios *pending_bios;
 354         struct bio *tail;
 355         struct bio *cur;
 356         int again = 0;
 357         unsigned long num_run;
 358         unsigned long batch_run = 0;
 359         unsigned long limit;
 360         unsigned long last_waited = 0;
 361         int force_reg = 0;
 362         int sync_pending = 0;
 363         struct blk_plug plug;
 364
 365         /*
 366          * this function runs all the bios we've collected for
 367          * a particular device.  We don't want to wander off to
 368          * another device without first sending all of these down.
 369          * So, setup a plug here and finish it off before we return
 370          */
 371         blk_start_plug(&plug);
 372
 373         bdi = device->bdev->bd_bdi;
 374         limit = btrfs_async_submit_limit(fs_info);
 375         limit = limit * 2 / 3;
 376
 377 loop:
 378         spin_lock(&device->io_lock);
 379
 380 loop_lock:
 381         num_run = 0;
 382
 383         /* take all the bios off the list at once and process them
 384          * later on (without the lock held).  But, remember the
 385          * tail and other pointers so the bios can be properly reinserted
 386          * into the list if we hit congestion
 387          */
 388         if (!force_reg && device->pending_sync_bios.head) {
 389                 pending_bios = &device->pending_sync_bios;
 390                 force_reg = 1;
 391         } else {
 392                 pending_bios = &device->pending_bios;
 393                 force_reg = 0;
 394         }
 395
 396         pending = pending_bios->head;
 397         tail = pending_bios->tail;
 398         WARN_ON(pending && !tail);
 399
 400         /*
 401          * if pending was null this time around, no bios need processing
 402          * at all and we can stop.  Otherwise it'll loop back up again
 403          * and do an additional check so no bios are missed.
 404          *
 405          * device->running_pending is used to synchronize with the
 406          * schedule_bio code.
 407          */
 408         if (device->pending_sync_bios.head == NULL &&
 409             device->pending_bios.head == NULL) {
 410                 again = 0;
 411                 device->running_pending = 0;
 412         } else {
 413                 again = 1;
 414                 device->running_pending = 1;
 415         }
 416
 417         pending_bios->head = NULL;
 418         pending_bios->tail = NULL;
 419
 420         spin_unlock(&device->io_lock);
 421
 422         while (pending) {
 423
 424                 rmb();
 425                 /* we want to work on both lists, but do more bios on the
 426                  * sync list than the regular list
 427                  */
 428                 if ((num_run > 32 &&
 429                     pending_bios != &device->pending_sync_bios &&
 430                     device->pending_sync_bios.head) ||
 431                    (num_run > 64 && pending_bios == &device->pending_sync_bios &&
 432                     device->pending_bios.head)) {
 433                         spin_lock(&device->io_lock);
 434                         requeue_list(pending_bios, pending, tail);
 435                         goto loop_lock;
 436                 }
 437
 438                 cur = pending;
 439                 pending = pending->bi_next;
 440                 cur->bi_next = NULL;
 441
 442                 /*
 443                  * atomic_dec_return implies a barrier for waitqueue_active
 444                  */
 445                 if (atomic_dec_return(&fs_info->nr_async_bios) < limit &&
 446                     waitqueue_active(&fs_info->async_submit_wait))
 447                         wake_up(&fs_info->async_submit_wait);
 448
 449                 BUG_ON(atomic_read(&cur->__bi_cnt) == 0);
 450
 451                 /*
 452                  * if we're doing the sync list, record that our
 453                  * plug has some sync requests on it
 454                  *
 455                  * If we're doing the regular list and there are
 456                  * sync requests sitting around, unplug before
 457                  * we add more
 458                  */
 459                 if (pending_bios == &device->pending_sync_bios) {
 460                         sync_pending = 1;
 461                 } else if (sync_pending) {
 462                         blk_finish_plug(&plug);
 463                         blk_start_plug(&plug);
 464                         sync_pending = 0;
 465                 }
 466
 467                 btrfsic_submit_bio(cur);
 468                 num_run++;
 469                 batch_run++;
 470
 471                 cond_resched();
 472
 473                 /*
 474                  * we made progress, there is more work to do and the bdi
 475                  * is now congested.  Back off and let other work structs
 476                  * run instead
 477                  */
 478                 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
 479                     fs_info->fs_devices->open_devices > 1) {
 480                         struct io_context *ioc;
 481
 482                         ioc = current->io_context;
 483
 484                         /*
 485                          * the main goal here is that we don't want to
 486                          * block if we're going to be able to submit
 487                          * more requests without blocking.
 488                          *
 489                          * This code does two great things, it pokes into
 490                          * the elevator code from a filesystem _and_
 491                          * it makes assumptions about how batching works.
 492                          */
 493                         if (ioc && ioc->nr_batch_requests > 0 &&
 494                             time_before(jiffies, ioc->last_waited + HZ/50UL) &&
 495                             (last_waited == 0 ||
 496                              ioc->last_waited == last_waited)) {
 497                                 /*
 498                                  * we want to go through our batch of
 499                                  * requests and stop.  So, we copy out
 500                                  * the ioc->last_waited time and test
 501                                  * against it before looping
 502                                  */
 503                                 last_waited = ioc->last_waited;
 504                                 cond_resched();
 505                                 continue;
 506                         }
 507                         spin_lock(&device->io_lock);
 508                         requeue_list(pending_bios, pending, tail);
 509                         device->running_pending = 1;
 510
 511                         spin_unlock(&device->io_lock);
 512                         btrfs_queue_work(fs_info->submit_workers,
 513                                          &device->work);
 514                         goto done;
 515                 }
 516                 /* unplug every 64 requests just for good measure */
 517                 if (batch_run % 64 == 0) {
 518                         blk_finish_plug(&plug);
 519                         blk_start_plug(&plug);
 520                         sync_pending = 0;
 521                 }
 522         }
 523
 524         cond_resched();
 525         if (again)
 526                 goto loop;
 527
 528         spin_lock(&device->io_lock);
 529         if (device->pending_bios.head || device->pending_sync_bios.head)
 530                 goto loop_lock;
 531         spin_unlock(&device->io_lock);
 532
 533 done:
 534         blk_finish_plug(&plug);
 535 }
 536
 537 static void pending_bios_fn(struct btrfs_work *work)
 538 {
 539         struct btrfs_device *device;
 540
 541         device = container_of(work, struct btrfs_device, work);
 542         run_scheduled_bios(device);
 543 }
 544
 545
 546 void btrfs_free_stale_device(struct btrfs_device *cur_dev)
 547 {
 548         struct btrfs_fs_devices *fs_devs;
 549         struct btrfs_device *dev;
 550
 551         if (!cur_dev->name)
 552                 return;
 553
 554         list_for_each_entry(fs_devs, &fs_uuids, list) {
 555                 int del = 1;
 556
 557                 if (fs_devs->opened)
 558                         continue;
 559                 if (fs_devs->seeding)
 560                         continue;
 561
 562                 list_for_each_entry(dev, &fs_devs->devices, dev_list) {
 563
 564                         if (dev == cur_dev)
 565                                 continue;
 566                         if (!dev->name)
 567                                 continue;
 568
 569                         /*
 570                          * Todo: This won't be enough. What if the same device
 571                          * comes back (with new uuid and) with its mapper path?
 572                          * But for now, this does help as mostly an admin will
 573                          * either use mapper or non mapper path throughout.
 574                          */
 575                         rcu_read_lock();
 576                         del = strcmp(rcu_str_deref(dev->name),
 577                                                 rcu_str_deref(cur_dev->name));
 578                         rcu_read_unlock();
 579                         if (!del)
 580                                 break;
 581                 }
 582
 583                 if (!del) {
 584                         /* delete the stale device */
 585                         if (fs_devs->num_devices == 1) {
 586                                 btrfs_sysfs_remove_fsid(fs_devs);
 587                                 list_del(&fs_devs->list);
 588                                 free_fs_devices(fs_devs);
 589                         } else {
 590                                 fs_devs->num_devices--;
 591                                 list_del(&dev->dev_list);
 592                                 rcu_string_free(dev->name);
 593                                 kfree(dev);
 594                         }
 595                         break;
 596                 }
 597         }
 598 }
 599
 600 /*
 601  * Add new device to list of registered devices
 602  *
 603  * Returns:
 604  * 1   - first time device is seen
 605  * 0   - device already known
 606  * < 0 - error
 607  */
 608 static noinline int device_list_add(const char *path,
 609                            struct btrfs_super_block *disk_super,
 610                            u64 devid, struct btrfs_fs_devices **fs_devices_ret)
 611 {
 612         struct btrfs_device *device;
 613         struct btrfs_fs_devices *fs_devices;
 614         struct rcu_string *name;
 615         int ret = 0;
 616         u64 found_transid = btrfs_super_generation(disk_super);
 617
 618         fs_devices = find_fsid(disk_super->fsid);
 619         if (!fs_devices) {
 620                 fs_devices = alloc_fs_devices(disk_super->fsid);
 621                 if (IS_ERR(fs_devices))
 622                         return PTR_ERR(fs_devices);
 623
 624                 list_add(&fs_devices->list, &fs_uuids);
 625
 626                 device = NULL;
 627         } else {
 628                 device = __find_device(&fs_devices->devices, devid,
 629                                        disk_super->dev_item.uuid);
 630         }
 631
 632         if (!device) {
 633                 if (fs_devices->opened)
 634                         return -EBUSY;
 635
 636                 device = btrfs_alloc_device(NULL, &devid,
 637                                             disk_super->dev_item.uuid);
 638                 if (IS_ERR(device)) {
 639                         /* we can safely leave the fs_devices entry around */
 640                         return PTR_ERR(device);
 641                 }
 642
 643                 name = rcu_string_strdup(path, GFP_NOFS);
 644                 if (!name) {
 645                         kfree(device);
 646                         return -ENOMEM;
 647                 }
 648                 rcu_assign_pointer(device->name, name);
 649
 650                 mutex_lock(&fs_devices->device_list_mutex);
 651                 list_add_rcu(&device->dev_list, &fs_devices->devices);
 652                 fs_devices->num_devices++;
 653                 mutex_unlock(&fs_devices->device_list_mutex);
 654
 655                 ret = 1;
 656                 device->fs_devices = fs_devices;
 657         } else if (!device->name || strcmp(device->name->str, path)) {
 658                 /*
 659                  * When FS is already mounted.
 660                  * 1. If you are here and if the device->name is NULL that
 661                  *    means this device was missing at time of FS mount.
 662                  * 2. If you are here and if the device->name is different
 663                  *    from 'path' that means either
 664                  *      a. The same device disappeared and reappeared with
 665                  *         different name. or
 666                  *      b. The missing-disk-which-was-replaced, has
 667                  *         reappeared now.
 668                  *
 669                  * We must allow 1 and 2a above. But 2b would be a spurious
 670                  * and unintentional.
 671                  *
 672                  * Further in case of 1 and 2a above, the disk at 'path'
 673                  * would have missed some transaction when it was away and
 674                  * in case of 2a the stale bdev has to be updated as well.
 675                  * 2b must not be allowed at all time.
 676                  */
 677
 678                 /*
 679                  * For now, we do allow update to btrfs_fs_device through the
 680                  * btrfs dev scan cli after FS has been mounted.  We're still
 681                  * tracking a problem where systems fail mount by subvolume id
 682                  * when we reject replacement on a mounted FS.
 683                  */
 684                 if (!fs_devices->opened && found_transid < device->generation) {
 685                         /*
 686                          * That is if the FS is _not_ mounted and if you
 687                          * are here, that means there is more than one
 688                          * disk with same uuid and devid.We keep the one
 689                          * with larger generation number or the last-in if
 690                          * generation are equal.
 691                          */
 692                         return -EEXIST;
 693                 }
 694
 695                 name = rcu_string_strdup(path, GFP_NOFS);
 696                 if (!name)
 697                         return -ENOMEM;
 698                 rcu_string_free(device->name);
 699                 rcu_assign_pointer(device->name, name);
 700                 if (device->missing) {
 701                         fs_devices->missing_devices--;
 702                         device->missing = 0;
 703                 }
 704         }
 705
 706         /*
 707          * Unmount does not free the btrfs_device struct but would zero
 708          * generation along with most of the other members. So just update
 709          * it back. We need it to pick the disk with largest generation
 710          * (as above).
 711          */
 712         if (!fs_devices->opened)
 713                 device->generation = found_transid;
 714
 715         /*
 716          * if there is new btrfs on an already registered device,
 717          * then remove the stale device entry.
 718          */
 719         if (ret > 0)
 720                 btrfs_free_stale_device(device);
 721
 722         *fs_devices_ret = fs_devices;
 723
 724         return ret;
 725 }
 726
 727 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
 728 {
 729         struct btrfs_fs_devices *fs_devices;
 730         struct btrfs_device *device;
 731         struct btrfs_device *orig_dev;
 732
 733         fs_devices = alloc_fs_devices(orig->fsid);
 734         if (IS_ERR(fs_devices))
 735                 return fs_devices;
 736
 737         mutex_lock(&orig->device_list_mutex);
 738         fs_devices->total_devices = orig->total_devices;
 739
 740         /* We have held the volume lock, it is safe to get the devices. */
 741         list_for_each_entry(orig_dev, &orig->devices, dev_list) {
 742                 struct rcu_string *name;
 743
 744                 device = btrfs_alloc_device(NULL, &orig_dev->devid,
 745                                             orig_dev->uuid);
 746                 if (IS_ERR(device))
 747                         goto error;
 748
 749                 /*
 750                  * This is ok to do without rcu read locked because we hold the
 751                  * uuid mutex so nothing we touch in here is going to disappear.
 752                  */
 753                 if (orig_dev->name) {
 754                         name = rcu_string_strdup(orig_dev->name->str,
 755                                         GFP_KERNEL);
 756                         if (!name) {
 757                                 kfree(device);
 758                                 goto error;
 759                         }
 760                         rcu_assign_pointer(device->name, name);
 761                 }
 762
 763                 list_add(&device->dev_list, &fs_devices->devices);
 764                 device->fs_devices = fs_devices;
 765                 fs_devices->num_devices++;
 766         }
 767         mutex_unlock(&orig->device_list_mutex);
 768         return fs_devices;
 769 error:
 770         mutex_unlock(&orig->device_list_mutex);
 771         free_fs_devices(fs_devices);
 772         return ERR_PTR(-ENOMEM);
 773 }
 774
 775 void btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices, int step)
 776 {
 777         struct btrfs_device *device, *next;
 778         struct btrfs_device *latest_dev = NULL;
 779
 780         mutex_lock(&uuid_mutex);
 781 again:
 782         /* This is the initialized path, it is safe to release the devices. */
 783         list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
 784                 if (device->in_fs_metadata) {
 785                         if (!device->is_tgtdev_for_dev_replace &&
 786                             (!latest_dev ||
 787                              device->generation > latest_dev->generation)) {
 788                                 latest_dev = device;
 789                         }
 790                         continue;
 791                 }
 792
 793                 if (device->devid == BTRFS_DEV_REPLACE_DEVID) {
 794                         /*
 795                          * In the first step, keep the device which has
 796                          * the correct fsid and the devid that is used
 797                          * for the dev_replace procedure.
 798                          * In the second step, the dev_replace state is
 799                          * read from the device tree and it is known
 800                          * whether the procedure is really active or
 801                          * not, which means whether this device is
 802                          * used or whether it should be removed.
 803                          */
 804                         if (step == 0 || device->is_tgtdev_for_dev_replace) {
 805                                 continue;
 806                         }
 807                 }
 808                 if (device->bdev) {
 809                         blkdev_put(device->bdev, device->mode);
 810                         device->bdev = NULL;
 811                         fs_devices->open_devices--;
 812                 }
 813                 if (device->writeable) {
 814                         list_del_init(&device->dev_alloc_list);
 815                         device->writeable = 0;
 816                         if (!device->is_tgtdev_for_dev_replace)
 817                                 fs_devices->rw_devices--;
 818                 }
 819                 list_del_init(&device->dev_list);
 820                 fs_devices->num_devices--;
 821                 rcu_string_free(device->name);
 822                 kfree(device);
 823         }
 824
 825         if (fs_devices->seed) {
 826                 fs_devices = fs_devices->seed;
 827                 goto again;
 828         }
 829
 830         fs_devices->latest_bdev = latest_dev->bdev;
 831
 832         mutex_unlock(&uuid_mutex);
 833 }
 834
 835 static void __free_device(struct work_struct *work)
 836 {
 837         struct btrfs_device *device;
 838
 839         device = container_of(work, struct btrfs_device, rcu_work);
 840         rcu_string_free(device->name);
 841         kfree(device);
 842 }
 843
 844 static void free_device(struct rcu_head *head)
 845 {
 846         struct btrfs_device *device;
 847
 848         device = container_of(head, struct btrfs_device, rcu);
 849
 850         INIT_WORK(&device->rcu_work, __free_device);
 851         schedule_work(&device->rcu_work);
 852 }
 853
 854 static void btrfs_close_bdev(struct btrfs_device *device)
 855 {
 856         if (device->bdev && device->writeable) {
 857                 sync_blockdev(device->bdev);
 858                 invalidate_bdev(device->bdev);
 859         }
 860
 861         if (device->bdev)
 862                 blkdev_put(device->bdev, device->mode);
 863 }
 864
 865 static void btrfs_prepare_close_one_device(struct btrfs_device *device)
 866 {
 867         struct btrfs_fs_devices *fs_devices = device->fs_devices;
 868         struct btrfs_device *new_device;
 869         struct rcu_string *name;
 870
 871         if (device->bdev)
 872                 fs_devices->open_devices--;
 873
 874         if (device->writeable &&
 875             device->devid != BTRFS_DEV_REPLACE_DEVID) {
 876                 list_del_init(&device->dev_alloc_list);
 877                 fs_devices->rw_devices--;
 878         }
 879
 880         if (device->missing)
 881                 fs_devices->missing_devices--;
 882
 883         new_device = btrfs_alloc_device(NULL, &device->devid,
 884                                         device->uuid);
 885         BUG_ON(IS_ERR(new_device)); /* -ENOMEM */
 886
 887         /* Safe because we are under uuid_mutex */
 888         if (device->name) {
 889                 name = rcu_string_strdup(device->name->str, GFP_NOFS);
 890                 BUG_ON(!name); /* -ENOMEM */
 891                 rcu_assign_pointer(new_device->name, name);
 892         }
 893
 894         list_replace_rcu(&device->dev_list, &new_device->dev_list);
 895         new_device->fs_devices = device->fs_devices;
 896 }
 897
 898 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
 899 {
 900         struct btrfs_device *device, *tmp;
 901         struct list_head pending_put;
 902
 903         INIT_LIST_HEAD(&pending_put);
 904
 905         if (--fs_devices->opened > 0)
 906                 return 0;
 907
 908         mutex_lock(&fs_devices->device_list_mutex);
 909         list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list) {
 910                 btrfs_prepare_close_one_device(device);
 911                 list_add(&device->dev_list, &pending_put);
 912         }
 913         mutex_unlock(&fs_devices->device_list_mutex);
 914
 915         /*
 916          * btrfs_show_devname() is using the device_list_mutex,
 917          * sometimes call to blkdev_put() leads vfs calling
 918          * into this func. So do put outside of device_list_mutex,
 919          * as of now.
 920          */
 921         while (!list_empty(&pending_put)) {
 922                 device = list_first_entry(&pending_put,
 923                                 struct btrfs_device, dev_list);
 924                 list_del(&device->dev_list);
 925                 btrfs_close_bdev(device);
 926                 call_rcu(&device->rcu, free_device);
 927         }
 928
 929         WARN_ON(fs_devices->open_devices);
 930         WARN_ON(fs_devices->rw_devices);
 931         fs_devices->opened = 0;
 932         fs_devices->seeding = 0;
 933
 934         return 0;
 935 }
 936
 937 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
 938 {
 939         struct btrfs_fs_devices *seed_devices = NULL;
 940         int ret;
 941
 942         mutex_lock(&uuid_mutex);
 943         ret = __btrfs_close_devices(fs_devices);
 944         if (!fs_devices->opened) {
 945                 seed_devices = fs_devices->seed;
 946                 fs_devices->seed = NULL;
 947         }
 948         mutex_unlock(&uuid_mutex);
 949
 950         while (seed_devices) {
 951                 fs_devices = seed_devices;
 952                 seed_devices = fs_devices->seed;
 953                 __btrfs_close_devices(fs_devices);
 954                 free_fs_devices(fs_devices);
 955         }
 956         /*
 957          * Wait for rcu kworkers under __btrfs_close_devices
 958          * to finish all blkdev_puts so device is really
 959          * free when umount is done.
 960          */
 961         rcu_barrier();
 962         return ret;
 963 }
 964
 965 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
 966                                 fmode_t flags, void *holder)
 967 {
 968         struct request_queue *q;
 969         struct block_device *bdev;
 970         struct list_head *head = &fs_devices->devices;
 971         struct btrfs_device *device;
 972         struct btrfs_device *latest_dev = NULL;
 973         struct buffer_head *bh;
 974         struct btrfs_super_block *disk_super;
 975         u64 devid;
 976         int seeding = 1;
 977         int ret = 0;
 978
 979         flags |= FMODE_EXCL;
 980
 981         list_for_each_entry(device, head, dev_list) {
 982                 if (device->bdev)
 983                         continue;
 984                 if (!device->name)
 985                         continue;
 986
 987                 /* Just open everything we can; ignore failures here */
 988                 if (btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1,
 989                                             &bdev, &bh))
 990                         continue;
 991
 992                 disk_super = (struct btrfs_super_block *)bh->b_data;
 993                 devid = btrfs_stack_device_id(&disk_super->dev_item);
 994                 if (devid != device->devid)
 995                         goto error_brelse;
 996
 997                 if (memcmp(device->uuid, disk_super->dev_item.uuid,
 998                            BTRFS_UUID_SIZE))
 999                         goto error_brelse;
1000
1001                 device->generation = btrfs_super_generation(disk_super);
1002                 if (!latest_dev ||
1003                     device->generation > latest_dev->generation)
1004                         latest_dev = device;
1005
1006                 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
1007                         device->writeable = 0;
1008                 } else {
1009                         device->writeable = !bdev_read_only(bdev);
1010                         seeding = 0;
1011                 }
1012
1013                 q = bdev_get_queue(bdev);
1014                 if (blk_queue_discard(q))
1015                         device->can_discard = 1;
1016                 if (!blk_queue_nonrot(q))
1017                         fs_devices->rotating = 1;
1018
1019                 device->bdev = bdev;
1020                 device->in_fs_metadata = 0;
1021                 device->mode = flags;
1022
1023                 fs_devices->open_devices++;
1024                 if (device->writeable &&
1025                     device->devid != BTRFS_DEV_REPLACE_DEVID) {
1026                         fs_devices->rw_devices++;
1027                         list_add(&device->dev_alloc_list,
1028                                  &fs_devices->alloc_list);
1029                 }
1030                 brelse(bh);
1031                 continue;
1032
1033 error_brelse:
1034                 brelse(bh);
1035                 blkdev_put(bdev, flags);
1036                 continue;
1037         }
1038         if (fs_devices->open_devices == 0) {
1039                 ret = -EINVAL;
1040                 goto out;
1041         }
1042         fs_devices->seeding = seeding;
1043         fs_devices->opened = 1;
1044         fs_devices->latest_bdev = latest_dev->bdev;
1045         fs_devices->total_rw_bytes = 0;
1046 out:
1047         return ret;
1048 }
1049
1050 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
1051                        fmode_t flags, void *holder)
1052 {
1053         int ret;
1054
1055         mutex_lock(&uuid_mutex);
1056         if (fs_devices->opened) {
1057                 fs_devices->opened++;
1058                 ret = 0;
1059         } else {
1060                 ret = __btrfs_open_devices(fs_devices, flags, holder);
1061         }
1062         mutex_unlock(&uuid_mutex);
1063         return ret;
1064 }
1065
1066 void btrfs_release_disk_super(struct page *page)
1067 {
1068         kunmap(page);
1069         put_page(page);
1070 }
1071
1072 int btrfs_read_disk_super(struct block_device *bdev, u64 bytenr,
1073                 struct page **page, struct btrfs_super_block **disk_super)
1074 {
1075         void *p;
1076         pgoff_t index;
1077
1078         /* make sure our super fits in the device */
1079         if (bytenr + PAGE_SIZE >= i_size_read(bdev->bd_inode))
1080                 return 1;
1081
1082         /* make sure our super fits in the page */
1083         if (sizeof(**disk_super) > PAGE_SIZE)
1084                 return 1;
1085
1086         /* make sure our super doesn't straddle pages on disk */
1087         index = bytenr >> PAGE_SHIFT;
1088         if ((bytenr + sizeof(**disk_super) - 1) >> PAGE_SHIFT != index)
1089                 return 1;
1090
1091         /* pull in the page with our super */
1092         *page = read_cache_page_gfp(bdev->bd_inode->i_mapping,
1093                                    index, GFP_KERNEL);
1094
1095         if (IS_ERR_OR_NULL(*page))
1096                 return 1;
1097
1098         p = kmap(*page);
1099
1100         /* align our pointer to the offset of the super block */
1101         *disk_super = p + (bytenr & ~PAGE_MASK);
1102
1103         if (btrfs_super_bytenr(*disk_super) != bytenr ||
1104             btrfs_super_magic(*disk_super) != BTRFS_MAGIC) {
1105                 btrfs_release_disk_super(*page);
1106                 return 1;
1107         }
1108
1109         if ((*disk_super)->label[0] &&
1110                 (*disk_super)->label[BTRFS_LABEL_SIZE - 1])
1111                 (*disk_super)->label[BTRFS_LABEL_SIZE - 1] = '\0';
1112
1113         return 0;
1114 }
1115
1116 /*
1117  * Look for a btrfs signature on a device. This may be called out of the mount path
1118  * and we are not allowed to call set_blocksize during the scan. The superblock
1119  * is read via pagecache
1120  */
1121 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
1122                           struct btrfs_fs_devices **fs_devices_ret)
1123 {
1124         struct btrfs_super_block *disk_super;
1125         struct block_device *bdev;
1126         struct page *page;
1127         int ret = -EINVAL;
1128         u64 devid;
1129         u64 transid;
1130         u64 total_devices;
1131         u64 bytenr;
1132
1133         /*
1134          * we would like to check all the supers, but that would make
1135          * a btrfs mount succeed after a mkfs from a different FS.
1136          * So, we need to add a special mount option to scan for
1137          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
1138          */
1139         bytenr = btrfs_sb_offset(0);
1140         flags |= FMODE_EXCL;
1141         mutex_lock(&uuid_mutex);
1142
1143         bdev = blkdev_get_by_path(path, flags, holder);
1144         if (IS_ERR(bdev)) {
1145                 ret = PTR_ERR(bdev);
1146                 goto error;
1147         }
1148
1149         if (btrfs_read_disk_super(bdev, bytenr, &page, &disk_super))
1150                 goto error_bdev_put;
1151
1152         devid = btrfs_stack_device_id(&disk_super->dev_item);
1153         transid = btrfs_super_generation(disk_super);
1154         total_devices = btrfs_super_num_devices(disk_super);
1155
1156         ret = device_list_add(path, disk_super, devid, fs_devices_ret);
1157         if (ret > 0) {
1158                 if (disk_super->label[0]) {
1159                         pr_info("BTRFS: device label %s ", disk_super->label);
1160                 } else {
1161                         pr_info("BTRFS: device fsid %pU ", disk_super->fsid);
1162                 }
1163
1164                 pr_cont("devid %llu transid %llu %s\n", devid, transid, path);
1165                 ret = 0;
1166         }
1167         if (!ret && fs_devices_ret)
1168                 (*fs_devices_ret)->total_devices = total_devices;
1169
1170         btrfs_release_disk_super(page);
1171
1172 error_bdev_put:
1173         blkdev_put(bdev, flags);
1174 error:
1175         mutex_unlock(&uuid_mutex);
1176         return ret;
1177 }
1178
1179 /* helper to account the used device space in the range */
1180 int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
1181                                    u64 end, u64 *length)
1182 {
1183         struct btrfs_key key;
1184         struct btrfs_root *root = device->fs_info->dev_root;
1185         struct btrfs_dev_extent *dev_extent;
1186         struct btrfs_path *path;
1187         u64 extent_end;
1188         int ret;
1189         int slot;
1190         struct extent_buffer *l;
1191
1192         *length = 0;
1193
1194         if (start >= device->total_bytes || device->is_tgtdev_for_dev_replace)
1195                 return 0;
1196
1197         path = btrfs_alloc_path();
1198         if (!path)
1199                 return -ENOMEM;
1200         path->reada = READA_FORWARD;
1201
1202         key.objectid = device->devid;
1203         key.offset = start;
1204         key.type = BTRFS_DEV_EXTENT_KEY;
1205
1206         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1207         if (ret < 0)
1208                 goto out;
1209         if (ret > 0) {
1210                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1211                 if (ret < 0)
1212                         goto out;
1213         }
1214
1215         while (1) {
1216                 l = path->nodes[0];
1217                 slot = path->slots[0];
1218                 if (slot >= btrfs_header_nritems(l)) {
1219                         ret = btrfs_next_leaf(root, path);
1220                         if (ret == 0)
1221                                 continue;
1222                         if (ret < 0)
1223                                 goto out;
1224
1225                         break;
1226                 }
1227                 btrfs_item_key_to_cpu(l, &key, slot);
1228
1229                 if (key.objectid < device->devid)
1230                         goto next;
1231
1232                 if (key.objectid > device->devid)
1233                         break;
1234
1235                 if (key.type != BTRFS_DEV_EXTENT_KEY)
1236                         goto next;
1237
1238                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1239                 extent_end = key.offset + btrfs_dev_extent_length(l,
1240                                                                   dev_extent);
1241                 if (key.offset <= start && extent_end > end) {
1242                         *length = end - start + 1;
1243                         break;
1244                 } else if (key.offset <= start && extent_end > start)
1245                         *length += extent_end - start;
1246                 else if (key.offset > start && extent_end <= end)
1247                         *length += extent_end - key.offset;
1248                 else if (key.offset > start && key.offset <= end) {
1249                         *length += end - key.offset + 1;
1250                         break;
1251                 } else if (key.offset > end)
1252                         break;
1253
1254 next:
1255                 path->slots[0]++;
1256         }
1257         ret = 0;
1258 out:
1259         btrfs_free_path(path);
1260         return ret;
1261 }
1262
1263 static int contains_pending_extent(struct btrfs_transaction *transaction,
1264                                    struct btrfs_device *device,
1265                                    u64 *start, u64 len)
1266 {
1267         struct btrfs_fs_info *fs_info = device->fs_info;
1268         struct extent_map *em;
1269         struct list_head *search_list = &fs_info->pinned_chunks;
1270         int ret = 0;
1271         u64 physical_start = *start;
1272
1273         if (transaction)
1274                 search_list = &transaction->pending_chunks;
1275 again:
1276         list_for_each_entry(em, search_list, list) {
1277                 struct map_lookup *map;
1278                 int i;
1279
1280                 map = em->map_lookup;
1281                 for (i = 0; i < map->num_stripes; i++) {
1282                         u64 end;
1283
1284                         if (map->stripes[i].dev != device)
1285                                 continue;
1286                         if (map->stripes[i].physical >= physical_start + len ||
1287                             map->stripes[i].physical + em->orig_block_len <=
1288                             physical_start)
1289                                 continue;
1290                         /*
1291                          * Make sure that while processing the pinned list we do
1292                          * not override our *start with a lower value, because
1293                          * we can have pinned chunks that fall within this
1294                          * device hole and that have lower physical addresses
1295                          * than the pending chunks we processed before. If we
1296                          * do not take this special care we can end up getting
1297                          * 2 pending chunks that start at the same physical
1298                          * device offsets because the end offset of a pinned
1299                          * chunk can be equal to the start offset of some
1300                          * pending chunk.
1301                          */
1302                         end = map->stripes[i].physical + em->orig_block_len;
1303                         if (end > *start) {
1304                                 *start = end;
1305                                 ret = 1;
1306                         }
1307                 }
1308         }
1309         if (search_list != &fs_info->pinned_chunks) {
1310                 search_list = &fs_info->pinned_chunks;
1311                 goto again;
1312         }
1313
1314         return ret;
1315 }
1316
1317
1318 /*
1319  * find_free_dev_extent_start - find free space in the specified device
1320  * @device:       the device which we search the free space in
1321  * @num_bytes:    the size of the free space that we need
1322  * @search_start: the position from which to begin the search
1323  * @start:        store the start of the free space.
1324  * @len:          the size of the free space. that we find, or the size
1325  *                of the max free space if we don't find suitable free space
1326  *
1327  * this uses a pretty simple search, the expectation is that it is
1328  * called very infrequently and that a given device has a small number
1329  * of extents
1330  *
1331  * @start is used to store the start of the free space if we find. But if we
1332  * don't find suitable free space, it will be used to store the start position
1333  * of the max free space.
1334  *
1335  * @len is used to store the size of the free space that we find.
1336  * But if we don't find suitable free space, it is used to store the size of
1337  * the max free space.
1338  */
1339 int find_free_dev_extent_start(struct btrfs_transaction *transaction,
1340                                struct btrfs_device *device, u64 num_bytes,
1341                                u64 search_start, u64 *start, u64 *len)
1342 {
1343         struct btrfs_fs_info *fs_info = device->fs_info;
1344         struct btrfs_root *root = fs_info->dev_root;
1345         struct btrfs_key key;
1346         struct btrfs_dev_extent *dev_extent;
1347         struct btrfs_path *path;
1348         u64 hole_size;
1349         u64 max_hole_start;
1350         u64 max_hole_size;
1351         u64 extent_end;
1352         u64 search_end = device->total_bytes;
1353         int ret;
1354         int slot;
1355         struct extent_buffer *l;
1356
1357         /*
1358          * We don't want to overwrite the superblock on the drive nor any area
1359          * used by the boot loader (grub for example), so we make sure to start
1360          * at an offset of at least 1MB.
1361          */
1362         search_start = max_t(u64, search_start, SZ_1M);
1363
1364         path = btrfs_alloc_path();
1365         if (!path)
1366                 return -ENOMEM;
1367
1368         max_hole_start = search_start;
1369         max_hole_size = 0;
1370
1371 again:
1372         if (search_start >= search_end || device->is_tgtdev_for_dev_replace) {
1373                 ret = -ENOSPC;
1374                 goto out;
1375         }
1376
1377         path->reada = READA_FORWARD;
1378         path->search_commit_root = 1;
1379         path->skip_locking = 1;
1380
1381         key.objectid = device->devid;
1382         key.offset = search_start;
1383         key.type = BTRFS_DEV_EXTENT_KEY;
1384
1385         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1386         if (ret < 0)
1387                 goto out;
1388         if (ret > 0) {
1389                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1390                 if (ret < 0)
1391                         goto out;
1392         }
1393
1394         while (1) {
1395                 l = path->nodes[0];
1396                 slot = path->slots[0];
1397                 if (slot >= btrfs_header_nritems(l)) {
1398                         ret = btrfs_next_leaf(root, path);
1399                         if (ret == 0)
1400                                 continue;
1401                         if (ret < 0)
1402                                 goto out;
1403
1404                         break;
1405                 }
1406                 btrfs_item_key_to_cpu(l, &key, slot);
1407
1408                 if (key.objectid < device->devid)
1409                         goto next;
1410
1411                 if (key.objectid > device->devid)
1412                         break;
1413
1414                 if (key.type != BTRFS_DEV_EXTENT_KEY)
1415                         goto next;
1416
1417                 if (key.offset > search_start) {
1418                         hole_size = key.offset - search_start;
1419
1420                         /*
1421                          * Have to check before we set max_hole_start, otherwise
1422                          * we could end up sending back this offset anyway.
1423                          */
1424                         if (contains_pending_extent(transaction, device,
1425                                                     &search_start,
1426                                                     hole_size)) {
1427                                 if (key.offset >= search_start) {
1428                                         hole_size = key.offset - search_start;
1429                                 } else {
1430                                         WARN_ON_ONCE(1);
1431                                         hole_size = 0;
1432                                 }
1433                         }
1434
1435                         if (hole_size > max_hole_size) {
1436                                 max_hole_start = search_start;
1437                                 max_hole_size = hole_size;
1438                         }
1439
1440                         /*
1441                          * If this free space is greater than which we need,
1442                          * it must be the max free space that we have found
1443                          * until now, so max_hole_start must point to the start
1444                          * of this free space and the length of this free space
1445                          * is stored in max_hole_size. Thus, we return
1446                          * max_hole_start and max_hole_size and go back to the
1447                          * caller.
1448                          */
1449                         if (hole_size >= num_bytes) {
1450                                 ret = 0;
1451                                 goto out;
1452                         }
1453                 }
1454
1455                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1456                 extent_end = key.offset + btrfs_dev_extent_length(l,
1457                                                                   dev_extent);
1458                 if (extent_end > search_start)
1459                         search_start = extent_end;
1460 next:
1461                 path->slots[0]++;
1462                 cond_resched();
1463         }
1464
1465         /*
1466          * At this point, search_start should be the end of
1467          * allocated dev extents, and when shrinking the device,
1468          * search_end may be smaller than search_start.
1469          */
1470         if (search_end > search_start) {
1471                 hole_size = search_end - search_start;
1472
1473                 if (contains_pending_extent(transaction, device, &search_start,
1474                                             hole_size)) {
1475                         btrfs_release_path(path);
1476                         goto again;
1477                 }
1478
1479                 if (hole_size > max_hole_size) {
1480                         max_hole_start = search_start;
1481                         max_hole_size = hole_size;
1482                 }
1483         }
1484
1485         /* See above. */
1486         if (max_hole_size < num_bytes)
1487                 ret = -ENOSPC;
1488         else
1489                 ret = 0;
1490
1491 out:
1492         btrfs_free_path(path);
1493         *start = max_hole_start;
1494         if (len)
1495                 *len = max_hole_size;
1496         return ret;
1497 }
1498
1499 int find_free_dev_extent(struct btrfs_trans_handle *trans,
1500                          struct btrfs_device *device, u64 num_bytes,
1501                          u64 *start, u64 *len)
1502 {
1503         /* FIXME use last free of some kind */
1504         return find_free_dev_extent_start(trans->transaction, device,
1505                                           num_bytes, 0, start, len);
1506 }
1507
1508 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1509                           struct btrfs_device *device,
1510                           u64 start, u64 *dev_extent_len)
1511 {
1512         struct btrfs_fs_info *fs_info = device->fs_info;
1513         struct btrfs_root *root = fs_info->dev_root;
1514         int ret;
1515         struct btrfs_path *path;
1516         struct btrfs_key key;
1517         struct btrfs_key found_key;
1518         struct extent_buffer *leaf = NULL;
1519         struct btrfs_dev_extent *extent = NULL;
1520
1521         path = btrfs_alloc_path();
1522         if (!path)
1523                 return -ENOMEM;
1524
1525         key.objectid = device->devid;
1526         key.offset = start;
1527         key.type = BTRFS_DEV_EXTENT_KEY;
1528 again:
1529         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1530         if (ret > 0) {
1531                 ret = btrfs_previous_item(root, path, key.objectid,
1532                                           BTRFS_DEV_EXTENT_KEY);
1533                 if (ret)
1534                         goto out;
1535                 leaf = path->nodes[0];
1536                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1537                 extent = btrfs_item_ptr(leaf, path->slots[0],
1538                                         struct btrfs_dev_extent);
1539                 BUG_ON(found_key.offset > start || found_key.offset +
1540                        btrfs_dev_extent_length(leaf, extent) < start);
1541                 key = found_key;
1542                 btrfs_release_path(path);
1543                 goto again;
1544         } else if (ret == 0) {
1545                 leaf = path->nodes[0];
1546                 extent = btrfs_item_ptr(leaf, path->slots[0],
1547                                         struct btrfs_dev_extent);
1548         } else {
1549                 btrfs_handle_fs_error(fs_info, ret, "Slot search failed");
1550                 goto out;
1551         }
1552
1553         *dev_extent_len = btrfs_dev_extent_length(leaf, extent);
1554
1555         ret = btrfs_del_item(trans, root, path);
1556         if (ret) {
1557                 btrfs_handle_fs_error(fs_info, ret,
1558                                       "Failed to remove dev extent item");
1559         } else {
1560                 set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags);
1561         }
1562 out:
1563         btrfs_free_path(path);
1564         return ret;
1565 }
1566
1567 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1568                                   struct btrfs_device *device,
1569                                   u64 chunk_tree, u64 chunk_objectid,
1570                                   u64 chunk_offset, u64 start, u64 num_bytes)
1571 {
1572         int ret;
1573         struct btrfs_path *path;
1574         struct btrfs_fs_info *fs_info = device->fs_info;
1575         struct btrfs_root *root = fs_info->dev_root;
1576         struct btrfs_dev_extent *extent;
1577         struct extent_buffer *leaf;
1578         struct btrfs_key key;
1579
1580         WARN_ON(!device->in_fs_metadata);
1581         WARN_ON(device->is_tgtdev_for_dev_replace);
1582         path = btrfs_alloc_path();
1583         if (!path)
1584                 return -ENOMEM;
1585
1586         key.objectid = device->devid;
1587         key.offset = start;
1588         key.type = BTRFS_DEV_EXTENT_KEY;
1589         ret = btrfs_insert_empty_item(trans, root, path, &key,
1590                                       sizeof(*extent));
1591         if (ret)
1592                 goto out;
1593
1594         leaf = path->nodes[0];
1595         extent = btrfs_item_ptr(leaf, path->slots[0],
1596                                 struct btrfs_dev_extent);
1597         btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
1598         btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
1599         btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1600
1601         write_extent_buffer_chunk_tree_uuid(leaf, fs_info->chunk_tree_uuid);
1602
1603         btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1604         btrfs_mark_buffer_dirty(leaf);
1605 out:
1606         btrfs_free_path(path);
1607         return ret;
1608 }
1609
1610 static u64 find_next_chunk(struct btrfs_fs_info *fs_info)
1611 {
1612         struct extent_map_tree *em_tree;
1613         struct extent_map *em;
1614         struct rb_node *n;
1615         u64 ret = 0;
1616
1617         em_tree = &fs_info->mapping_tree.map_tree;
1618         read_lock(&em_tree->lock);
1619         n = rb_last(&em_tree->map);
1620         if (n) {
1621                 em = rb_entry(n, struct extent_map, rb_node);
1622                 ret = em->start + em->len;
1623         }
1624         read_unlock(&em_tree->lock);
1625
1626         return ret;
1627 }
1628
1629 static noinline int find_next_devid(struct btrfs_fs_info *fs_info,
1630                                     u64 *devid_ret)
1631 {
1632         int ret;
1633         struct btrfs_key key;
1634         struct btrfs_key found_key;
1635         struct btrfs_path *path;
1636
1637         path = btrfs_alloc_path();
1638         if (!path)
1639                 return -ENOMEM;
1640
1641         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1642         key.type = BTRFS_DEV_ITEM_KEY;
1643         key.offset = (u64)-1;
1644
1645         ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
1646         if (ret < 0)
1647                 goto error;
1648
1649         BUG_ON(ret == 0); /* Corruption */
1650
1651         ret = btrfs_previous_item(fs_info->chunk_root, path,
1652                                   BTRFS_DEV_ITEMS_OBJECTID,
1653                                   BTRFS_DEV_ITEM_KEY);
1654         if (ret) {
1655                 *devid_ret = 1;
1656         } else {
1657                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1658                                       path->slots[0]);
1659                 *devid_ret = found_key.offset + 1;
1660         }
1661         ret = 0;
1662 error:
1663         btrfs_free_path(path);
1664         return ret;
1665 }
1666
1667 /*
1668  * the device information is stored in the chunk root
1669  * the btrfs_device struct should be fully filled in
1670  */
1671 static int btrfs_add_device(struct btrfs_trans_handle *trans,
1672                             struct btrfs_fs_info *fs_info,
1673                             struct btrfs_device *device)
1674 {
1675         struct btrfs_root *root = fs_info->chunk_root;
1676         int ret;
1677         struct btrfs_path *path;
1678         struct btrfs_dev_item *dev_item;
1679         struct extent_buffer *leaf;
1680         struct btrfs_key key;
1681         unsigned long ptr;
1682
1683         path = btrfs_alloc_path();
1684         if (!path)
1685                 return -ENOMEM;
1686
1687         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1688         key.type = BTRFS_DEV_ITEM_KEY;
1689         key.offset = device->devid;
1690
1691         ret = btrfs_insert_empty_item(trans, root, path, &key,
1692                                       sizeof(*dev_item));
1693         if (ret)
1694                 goto out;
1695
1696         leaf = path->nodes[0];
1697         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1698
1699         btrfs_set_device_id(leaf, dev_item, device->devid);
1700         btrfs_set_device_generation(leaf, dev_item, 0);
1701         btrfs_set_device_type(leaf, dev_item, device->type);
1702         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1703         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1704         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1705         btrfs_set_device_total_bytes(leaf, dev_item,
1706                                      btrfs_device_get_disk_total_bytes(device));
1707         btrfs_set_device_bytes_used(leaf, dev_item,
1708                                     btrfs_device_get_bytes_used(device));
1709         btrfs_set_device_group(leaf, dev_item, 0);
1710         btrfs_set_device_seek_speed(leaf, dev_item, 0);
1711         btrfs_set_device_bandwidth(leaf, dev_item, 0);
1712         btrfs_set_device_start_offset(leaf, dev_item, 0);
1713
1714         ptr = btrfs_device_uuid(dev_item);
1715         write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1716         ptr = btrfs_device_fsid(dev_item);
1717         write_extent_buffer(leaf, fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1718         btrfs_mark_buffer_dirty(leaf);
1719
1720         ret = 0;
1721 out:
1722         btrfs_free_path(path);
1723         return ret;
1724 }
1725
1726 /*
1727  * Function to update ctime/mtime for a given device path.
1728  * Mainly used for ctime/mtime based probe like libblkid.
1729  */
1730 static void update_dev_time(const char *path_name)
1731 {
1732         struct file *filp;
1733
1734         filp = filp_open(path_name, O_RDWR, 0);
1735         if (IS_ERR(filp))
1736                 return;
1737         file_update_time(filp);
1738         filp_close(filp, NULL);
1739 }
1740
1741 static int btrfs_rm_dev_item(struct btrfs_fs_info *fs_info,
1742                              struct btrfs_device *device)
1743 {
1744         struct btrfs_root *root = fs_info->chunk_root;
1745         int ret;
1746         struct btrfs_path *path;
1747         struct btrfs_key key;
1748         struct btrfs_trans_handle *trans;
1749
1750         path = btrfs_alloc_path();
1751         if (!path)
1752                 return -ENOMEM;
1753
1754         trans = btrfs_start_transaction(root, 0);
1755         if (IS_ERR(trans)) {
1756                 btrfs_free_path(path);
1757                 return PTR_ERR(trans);
1758         }
1759         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1760         key.type = BTRFS_DEV_ITEM_KEY;
1761         key.offset = device->devid;
1762
1763         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1764         if (ret < 0)
1765                 goto out;
1766
1767         if (ret > 0) {
1768                 ret = -ENOENT;
1769                 goto out;
1770         }
1771
1772         ret = btrfs_del_item(trans, root, path);
1773         if (ret)
1774                 goto out;
1775 out:
1776         btrfs_free_path(path);
1777         btrfs_commit_transaction(trans);
1778         return ret;
1779 }
1780
1781 /*
1782  * Verify that @num_devices satisfies the RAID profile constraints in the whole
1783  * filesystem. It's up to the caller to adjust that number regarding eg. device
1784  * replace.
1785  */
1786 static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info,
1787                 u64 num_devices)
1788 {
1789         u64 all_avail;
1790         unsigned seq;
1791         int i;
1792
1793         do {
1794                 seq = read_seqbegin(&fs_info->profiles_lock);
1795
1796                 all_avail = fs_info->avail_data_alloc_bits |
1797                             fs_info->avail_system_alloc_bits |
1798                             fs_info->avail_metadata_alloc_bits;
1799         } while (read_seqretry(&fs_info->profiles_lock, seq));
1800
1801         for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
1802                 if (!(all_avail & btrfs_raid_group[i]))
1803                         continue;
1804
1805                 if (num_devices < btrfs_raid_array[i].devs_min) {
1806                         int ret = btrfs_raid_mindev_error[i];
1807
1808                         if (ret)
1809                                 return ret;
1810                 }
1811         }
1812
1813         return 0;
1814 }
1815
1816 struct btrfs_device *btrfs_find_next_active_device(struct btrfs_fs_devices *fs_devs,
1817                                         struct btrfs_device *device)
1818 {
1819         struct btrfs_device *next_device;
1820
1821         list_for_each_entry(next_device, &fs_devs->devices, dev_list) {
1822                 if (next_device != device &&
1823                         !next_device->missing && next_device->bdev)
1824                         return next_device;
1825         }
1826
1827         return NULL;
1828 }
1829
1830 /*
1831  * Helper function to check if the given device is part of s_bdev / latest_bdev
1832  * and replace it with the provided or the next active device, in the context
1833  * where this function called, there should be always be another device (or
1834  * this_dev) which is active.
1835  */
1836 void btrfs_assign_next_active_device(struct btrfs_fs_info *fs_info,
1837                 struct btrfs_device *device, struct btrfs_device *this_dev)
1838 {
1839         struct btrfs_device *next_device;
1840
1841         if (this_dev)
1842                 next_device = this_dev;
1843         else
1844                 next_device = btrfs_find_next_active_device(fs_info->fs_devices,
1845                                                                 device);
1846         ASSERT(next_device);
1847
1848         if (fs_info->sb->s_bdev &&
1849                         (fs_info->sb->s_bdev == device->bdev))
1850                 fs_info->sb->s_bdev = next_device->bdev;
1851
1852         if (fs_info->fs_devices->latest_bdev == device->bdev)
1853                 fs_info->fs_devices->latest_bdev = next_device->bdev;
1854 }
1855
1856 int btrfs_rm_device(struct btrfs_fs_info *fs_info, const char *device_path,
1857                 u64 devid)
1858 {
1859         struct btrfs_device *device;
1860         struct btrfs_fs_devices *cur_devices;
1861         u64 num_devices;
1862         int ret = 0;
1863         bool clear_super = false;
1864
1865         mutex_lock(&uuid_mutex);
1866
1867         num_devices = fs_info->fs_devices->num_devices;
1868         btrfs_dev_replace_lock(&fs_info->dev_replace, 0);
1869         if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
1870                 WARN_ON(num_devices < 1);
1871                 num_devices--;
1872         }
1873         btrfs_dev_replace_unlock(&fs_info->dev_replace, 0);
1874
1875         ret = btrfs_check_raid_min_devices(fs_info, num_devices - 1);
1876         if (ret)
1877                 goto out;
1878
1879         ret = btrfs_find_device_by_devspec(fs_info, devid, device_path,
1880                                            &device);
1881         if (ret)
1882                 goto out;
1883
1884         if (device->is_tgtdev_for_dev_replace) {
1885                 ret = BTRFS_ERROR_DEV_TGT_REPLACE;
1886                 goto out;
1887         }
1888
1889         if (device->writeable && fs_info->fs_devices->rw_devices == 1) {
1890                 ret = BTRFS_ERROR_DEV_ONLY_WRITABLE;
1891                 goto out;
1892         }
1893
1894         if (device->writeable) {
1895                 mutex_lock(&fs_info->chunk_mutex);
1896                 list_del_init(&device->dev_alloc_list);
1897                 device->fs_devices->rw_devices--;
1898                 mutex_unlock(&fs_info->chunk_mutex);
1899                 clear_super = true;
1900         }
1901
1902         mutex_unlock(&uuid_mutex);
1903         ret = btrfs_shrink_device(device, 0);
1904         mutex_lock(&uuid_mutex);
1905         if (ret)
1906                 goto error_undo;
1907
1908         /*
1909          * TODO: the superblock still includes this device in its num_devices
1910          * counter although write_all_supers() is not locked out. This
1911          * could give a filesystem state which requires a degraded mount.
1912          */
1913         ret = btrfs_rm_dev_item(fs_info, device);
1914         if (ret)
1915                 goto error_undo;
1916
1917         device->in_fs_metadata = 0;
1918         btrfs_scrub_cancel_dev(fs_info, device);
1919
1920         /*
1921          * the device list mutex makes sure that we don't change
1922          * the device list while someone else is writing out all
1923          * the device supers. Whoever is writing all supers, should
1924          * lock the device list mutex before getting the number of
1925          * devices in the super block (super_copy). Conversely,
1926          * whoever updates the number of devices in the super block
1927          * (super_copy) should hold the device list mutex.
1928          */
1929
1930         cur_devices = device->fs_devices;
1931         mutex_lock(&fs_info->fs_devices->device_list_mutex);
1932         list_del_rcu(&device->dev_list);
1933
1934         device->fs_devices->num_devices--;
1935         device->fs_devices->total_devices--;
1936
1937         if (device->missing)
1938                 device->fs_devices->missing_devices--;
1939
1940         btrfs_assign_next_active_device(fs_info, device, NULL);
1941
1942         if (device->bdev) {
1943                 device->fs_devices->open_devices--;
1944                 /* remove sysfs entry */
1945                 btrfs_sysfs_rm_device_link(fs_info->fs_devices, device);
1946         }
1947
1948         num_devices = btrfs_super_num_devices(fs_info->super_copy) - 1;
1949         btrfs_set_super_num_devices(fs_info->super_copy, num_devices);
1950         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
1951
1952         /*
1953          * at this point, the device is zero sized and detached from
1954          * the devices list.  All that's left is to zero out the old
1955          * supers and free the device.
1956          */
1957         if (device->writeable)
1958                 btrfs_scratch_superblocks(device->bdev, device->name->str);
1959
1960         btrfs_close_bdev(device);
1961         call_rcu(&device->rcu, free_device);
1962
1963         if (cur_devices->open_devices == 0) {
1964                 struct btrfs_fs_devices *fs_devices;
1965                 fs_devices = fs_info->fs_devices;
1966                 while (fs_devices) {
1967                         if (fs_devices->seed == cur_devices) {
1968                                 fs_devices->seed = cur_devices->seed;
1969                                 break;
1970                         }
1971                         fs_devices = fs_devices->seed;
1972                 }
1973                 cur_devices->seed = NULL;
1974                 __btrfs_close_devices(cur_devices);
1975                 free_fs_devices(cur_devices);
1976         }
1977
1978         fs_info->num_tolerated_disk_barrier_failures =
1979                 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
1980
1981 out:
1982         mutex_unlock(&uuid_mutex);
1983         return ret;
1984
1985 error_undo:
1986         if (device->writeable) {
1987                 mutex_lock(&fs_info->chunk_mutex);
1988                 list_add(&device->dev_alloc_list,
1989                          &fs_info->fs_devices->alloc_list);
1990                 device->fs_devices->rw_devices++;
1991                 mutex_unlock(&fs_info->chunk_mutex);
1992         }
1993         goto out;
1994 }
1995
1996 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_fs_info *fs_info,
1997                                         struct btrfs_device *srcdev)
1998 {
1999         struct btrfs_fs_devices *fs_devices;
2000
2001         WARN_ON(!mutex_is_locked(&fs_info->fs_devices->device_list_mutex));
2002
2003         /*
2004          * in case of fs with no seed, srcdev->fs_devices will point
2005          * to fs_devices of fs_info. However when the dev being replaced is
2006          * a seed dev it will point to the seed's local fs_devices. In short
2007          * srcdev will have its correct fs_devices in both the cases.
2008          */
2009         fs_devices = srcdev->fs_devices;
2010
2011         list_del_rcu(&srcdev->dev_list);
2012         list_del_rcu(&srcdev->dev_alloc_list);
2013         fs_devices->num_devices--;
2014         if (srcdev->missing)
2015                 fs_devices->missing_devices--;
2016
2017         if (srcdev->writeable)
2018                 fs_devices->rw_devices--;
2019
2020         if (srcdev->bdev)
2021                 fs_devices->open_devices--;
2022 }
2023
2024 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_fs_info *fs_info,
2025                                       struct btrfs_device *srcdev)
2026 {
2027         struct btrfs_fs_devices *fs_devices = srcdev->fs_devices;
2028
2029         if (srcdev->writeable) {
2030                 /* zero out the old super if it is writable */
2031                 btrfs_scratch_superblocks(srcdev->bdev, srcdev->name->str);
2032         }
2033
2034         btrfs_close_bdev(srcdev);
2035
2036         call_rcu(&srcdev->rcu, free_device);
2037
2038         /*
2039          * unless fs_devices is seed fs, num_devices shouldn't go
2040          * zero
2041          */
2042         BUG_ON(!fs_devices->num_devices && !fs_devices->seeding);
2043
2044         /* if this is no devs we rather delete the fs_devices */
2045         if (!fs_devices->num_devices) {
2046                 struct btrfs_fs_devices *tmp_fs_devices;
2047
2048                 tmp_fs_devices = fs_info->fs_devices;
2049                 while (tmp_fs_devices) {
2050                         if (tmp_fs_devices->seed == fs_devices) {
2051                                 tmp_fs_devices->seed = fs_devices->seed;
2052                                 break;
2053                         }
2054                         tmp_fs_devices = tmp_fs_devices->seed;
2055                 }
2056                 fs_devices->seed = NULL;
2057                 __btrfs_close_devices(fs_devices);
2058                 free_fs_devices(fs_devices);
2059         }
2060 }
2061
2062 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
2063                                       struct btrfs_device *tgtdev)
2064 {
2065         mutex_lock(&uuid_mutex);
2066         WARN_ON(!tgtdev);
2067         mutex_lock(&fs_info->fs_devices->device_list_mutex);
2068
2069         btrfs_sysfs_rm_device_link(fs_info->fs_devices, tgtdev);
2070
2071         if (tgtdev->bdev)
2072                 fs_info->fs_devices->open_devices--;
2073
2074         fs_info->fs_devices->num_devices--;
2075
2076         btrfs_assign_next_active_device(fs_info, tgtdev, NULL);
2077
2078         list_del_rcu(&tgtdev->dev_list);
2079
2080         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2081         mutex_unlock(&uuid_mutex);
2082
2083         /*
2084          * The update_dev_time() with in btrfs_scratch_superblocks()
2085          * may lead to a call to btrfs_show_devname() which will try
2086          * to hold device_list_mutex. And here this device
2087          * is already out of device list, so we don't have to hold
2088          * the device_list_mutex lock.
2089          */
2090         btrfs_scratch_superblocks(tgtdev->bdev, tgtdev->name->str);
2091
2092         btrfs_close_bdev(tgtdev);
2093         call_rcu(&tgtdev->rcu, free_device);
2094 }
2095
2096 static int btrfs_find_device_by_path(struct btrfs_fs_info *fs_info,
2097                                      const 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                                     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(fs_info, devid, dev_uuid, disk_super->fsid);
2116         brelse(bh);
2117         if (!*device)
2118                 ret = -ENOENT;
2119         blkdev_put(bdev, FMODE_READ);
2120         return ret;
2121 }
2122
2123 int btrfs_find_device_missing_or_by_path(struct btrfs_fs_info *fs_info,
2124                                          const char *device_path,
2125                                          struct btrfs_device **device)
2126 {
2127         *device = NULL;
2128         if (strcmp(device_path, "missing") == 0) {
2129                 struct list_head *devices;
2130                 struct btrfs_device *tmp;
2131
2132                 devices = &fs_info->fs_devices->devices;
2133                 /*
2134                  * It is safe to read the devices since the volume_mutex
2135                  * is held by the caller.
2136                  */
2137                 list_for_each_entry(tmp, devices, dev_list) {
2138                         if (tmp->in_fs_metadata && !tmp->bdev) {
2139                                 *device = tmp;
2140                                 break;
2141                         }
2142                 }
2143
2144                 if (!*device)
2145                         return BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
2146
2147                 return 0;
2148         } else {
2149                 return btrfs_find_device_by_path(fs_info, device_path, device);
2150         }
2151 }
2152
2153 /*
2154  * Lookup a device given by device id, or the path if the id is 0.
2155  */
2156 int btrfs_find_device_by_devspec(struct btrfs_fs_info *fs_info, u64 devid,
2157                                  const char *devpath,
2158                                  struct btrfs_device **device)
2159 {
2160         int ret;
2161
2162         if (devid) {
2163                 ret = 0;
2164                 *device = btrfs_find_device(fs_info, devid, NULL, NULL);
2165                 if (!*device)
2166                         ret = -ENOENT;
2167         } else {
2168                 if (!devpath || !devpath[0])
2169                         return -EINVAL;
2170
2171                 ret = btrfs_find_device_missing_or_by_path(fs_info, devpath,
2172                                                            device);
2173         }
2174         return ret;
2175 }
2176
2177 /*
2178  * does all the dirty work required for changing file system's UUID.
2179  */
2180 static int btrfs_prepare_sprout(struct btrfs_fs_info *fs_info)
2181 {
2182         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2183         struct btrfs_fs_devices *old_devices;
2184         struct btrfs_fs_devices *seed_devices;
2185         struct btrfs_super_block *disk_super = fs_info->super_copy;
2186         struct btrfs_device *device;
2187         u64 super_flags;
2188
2189         BUG_ON(!mutex_is_locked(&uuid_mutex));
2190         if (!fs_devices->seeding)
2191                 return -EINVAL;
2192
2193         seed_devices = __alloc_fs_devices();
2194         if (IS_ERR(seed_devices))
2195                 return PTR_ERR(seed_devices);
2196
2197         old_devices = clone_fs_devices(fs_devices);
2198         if (IS_ERR(old_devices)) {
2199                 kfree(seed_devices);
2200                 return PTR_ERR(old_devices);
2201         }
2202
2203         list_add(&old_devices->list, &fs_uuids);
2204
2205         memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
2206         seed_devices->opened = 1;
2207         INIT_LIST_HEAD(&seed_devices->devices);
2208         INIT_LIST_HEAD(&seed_devices->alloc_list);
2209         mutex_init(&seed_devices->device_list_mutex);
2210
2211         mutex_lock(&fs_info->fs_devices->device_list_mutex);
2212         list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
2213                               synchronize_rcu);
2214         list_for_each_entry(device, &seed_devices->devices, dev_list)
2215                 device->fs_devices = seed_devices;
2216
2217         mutex_lock(&fs_info->chunk_mutex);
2218         list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
2219         mutex_unlock(&fs_info->chunk_mutex);
2220
2221         fs_devices->seeding = 0;
2222         fs_devices->num_devices = 0;
2223         fs_devices->open_devices = 0;
2224         fs_devices->missing_devices = 0;
2225         fs_devices->rotating = 0;
2226         fs_devices->seed = seed_devices;
2227
2228         generate_random_uuid(fs_devices->fsid);
2229         memcpy(fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2230         memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2231         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2232
2233         super_flags = btrfs_super_flags(disk_super) &
2234                       ~BTRFS_SUPER_FLAG_SEEDING;
2235         btrfs_set_super_flags(disk_super, super_flags);
2236
2237         return 0;
2238 }
2239
2240 /*
2241  * Store the expected generation for seed devices in device items.
2242  */
2243 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
2244                                struct btrfs_fs_info *fs_info)
2245 {
2246         struct btrfs_root *root = fs_info->chunk_root;
2247         struct btrfs_path *path;
2248         struct extent_buffer *leaf;
2249         struct btrfs_dev_item *dev_item;
2250         struct btrfs_device *device;
2251         struct btrfs_key key;
2252         u8 fs_uuid[BTRFS_UUID_SIZE];
2253         u8 dev_uuid[BTRFS_UUID_SIZE];
2254         u64 devid;
2255         int ret;
2256
2257         path = btrfs_alloc_path();
2258         if (!path)
2259                 return -ENOMEM;
2260
2261         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2262         key.offset = 0;
2263         key.type = BTRFS_DEV_ITEM_KEY;
2264
2265         while (1) {
2266                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2267                 if (ret < 0)
2268                         goto error;
2269
2270                 leaf = path->nodes[0];
2271 next_slot:
2272                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2273                         ret = btrfs_next_leaf(root, path);
2274                         if (ret > 0)
2275                                 break;
2276                         if (ret < 0)
2277                                 goto error;
2278                         leaf = path->nodes[0];
2279                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2280                         btrfs_release_path(path);
2281                         continue;
2282                 }
2283
2284                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2285                 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
2286                     key.type != BTRFS_DEV_ITEM_KEY)
2287                         break;
2288
2289                 dev_item = btrfs_item_ptr(leaf, path->slots[0],
2290                                           struct btrfs_dev_item);
2291                 devid = btrfs_device_id(leaf, dev_item);
2292                 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
2293                                    BTRFS_UUID_SIZE);
2294                 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
2295                                    BTRFS_UUID_SIZE);
2296                 device = btrfs_find_device(fs_info, devid, dev_uuid, fs_uuid);
2297                 BUG_ON(!device); /* Logic error */
2298
2299                 if (device->fs_devices->seeding) {
2300                         btrfs_set_device_generation(leaf, dev_item,
2301                                                     device->generation);
2302                         btrfs_mark_buffer_dirty(leaf);
2303                 }
2304
2305                 path->slots[0]++;
2306                 goto next_slot;
2307         }
2308         ret = 0;
2309 error:
2310         btrfs_free_path(path);
2311         return ret;
2312 }
2313
2314 int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path)
2315 {
2316         struct btrfs_root *root = fs_info->dev_root;
2317         struct request_queue *q;
2318         struct btrfs_trans_handle *trans;
2319         struct btrfs_device *device;
2320         struct block_device *bdev;
2321         struct list_head *devices;
2322         struct super_block *sb = fs_info->sb;
2323         struct rcu_string *name;
2324         u64 tmp;
2325         int seeding_dev = 0;
2326         int ret = 0;
2327
2328         if ((sb->s_flags & MS_RDONLY) && !fs_info->fs_devices->seeding)
2329                 return -EROFS;
2330
2331         bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2332                                   fs_info->bdev_holder);
2333         if (IS_ERR(bdev))
2334                 return PTR_ERR(bdev);
2335
2336         if (fs_info->fs_devices->seeding) {
2337                 seeding_dev = 1;
2338                 down_write(&sb->s_umount);
2339                 mutex_lock(&uuid_mutex);
2340         }
2341
2342         filemap_write_and_wait(bdev->bd_inode->i_mapping);
2343
2344         devices = &fs_info->fs_devices->devices;
2345
2346         mutex_lock(&fs_info->fs_devices->device_list_mutex);
2347         list_for_each_entry(device, devices, dev_list) {
2348                 if (device->bdev == bdev) {
2349                         ret = -EEXIST;
2350                         mutex_unlock(
2351                                 &fs_info->fs_devices->device_list_mutex);
2352                         goto error;
2353                 }
2354         }
2355         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2356
2357         device = btrfs_alloc_device(fs_info, NULL, NULL);
2358         if (IS_ERR(device)) {
2359                 /* we can safely leave the fs_devices entry around */
2360                 ret = PTR_ERR(device);
2361                 goto error;
2362         }
2363
2364         name = rcu_string_strdup(device_path, GFP_KERNEL);
2365         if (!name) {
2366                 kfree(device);
2367                 ret = -ENOMEM;
2368                 goto error;
2369         }
2370         rcu_assign_pointer(device->name, name);
2371
2372         trans = btrfs_start_transaction(root, 0);
2373         if (IS_ERR(trans)) {
2374                 rcu_string_free(device->name);
2375                 kfree(device);
2376                 ret = PTR_ERR(trans);
2377                 goto error;
2378         }
2379
2380         q = bdev_get_queue(bdev);
2381         if (blk_queue_discard(q))
2382                 device->can_discard = 1;
2383         device->writeable = 1;
2384         device->generation = trans->transid;
2385         device->io_width = fs_info->sectorsize;
2386         device->io_align = fs_info->sectorsize;
2387         device->sector_size = fs_info->sectorsize;
2388         device->total_bytes = i_size_read(bdev->bd_inode);
2389         device->disk_total_bytes = device->total_bytes;
2390         device->commit_total_bytes = device->total_bytes;
2391         device->fs_info = fs_info;
2392         device->bdev = bdev;
2393         device->in_fs_metadata = 1;
2394         device->is_tgtdev_for_dev_replace = 0;
2395         device->mode = FMODE_EXCL;
2396         device->dev_stats_valid = 1;
2397         set_blocksize(device->bdev, 4096);
2398
2399         if (seeding_dev) {
2400                 sb->s_flags &= ~MS_RDONLY;
2401                 ret = btrfs_prepare_sprout(fs_info);
2402                 BUG_ON(ret); /* -ENOMEM */
2403         }
2404
2405         device->fs_devices = fs_info->fs_devices;
2406
2407         mutex_lock(&fs_info->fs_devices->device_list_mutex);
2408         mutex_lock(&fs_info->chunk_mutex);
2409         list_add_rcu(&device->dev_list, &fs_info->fs_devices->devices);
2410         list_add(&device->dev_alloc_list,
2411                  &fs_info->fs_devices->alloc_list);
2412         fs_info->fs_devices->num_devices++;
2413         fs_info->fs_devices->open_devices++;
2414         fs_info->fs_devices->rw_devices++;
2415         fs_info->fs_devices->total_devices++;
2416         fs_info->fs_devices->total_rw_bytes += device->total_bytes;
2417
2418         atomic64_add(device->total_bytes, &fs_info->free_chunk_space);
2419
2420         if (!blk_queue_nonrot(q))
2421                 fs_info->fs_devices->rotating = 1;
2422
2423         tmp = btrfs_super_total_bytes(fs_info->super_copy);
2424         btrfs_set_super_total_bytes(fs_info->super_copy,
2425                                     tmp + device->total_bytes);
2426
2427         tmp = btrfs_super_num_devices(fs_info->super_copy);
2428         btrfs_set_super_num_devices(fs_info->super_copy, tmp + 1);
2429
2430         /* add sysfs device entry */
2431         btrfs_sysfs_add_device_link(fs_info->fs_devices, device);
2432
2433         /*
2434          * we've got more storage, clear any full flags on the space
2435          * infos
2436          */
2437         btrfs_clear_space_info_full(fs_info);
2438
2439         mutex_unlock(&fs_info->chunk_mutex);
2440         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2441
2442         if (seeding_dev) {
2443                 mutex_lock(&fs_info->chunk_mutex);
2444                 ret = init_first_rw_device(trans, fs_info);
2445                 mutex_unlock(&fs_info->chunk_mutex);
2446                 if (ret) {
2447                         btrfs_abort_transaction(trans, ret);
2448                         goto error_trans;
2449                 }
2450         }
2451
2452         ret = btrfs_add_device(trans, fs_info, device);
2453         if (ret) {
2454                 btrfs_abort_transaction(trans, ret);
2455                 goto error_trans;
2456         }
2457
2458         if (seeding_dev) {
2459                 char fsid_buf[BTRFS_UUID_UNPARSED_SIZE];
2460
2461                 ret = btrfs_finish_sprout(trans, fs_info);
2462                 if (ret) {
2463                         btrfs_abort_transaction(trans, ret);
2464                         goto error_trans;
2465                 }
2466
2467                 /* Sprouting would change fsid of the mounted root,
2468                  * so rename the fsid on the sysfs
2469                  */
2470                 snprintf(fsid_buf, BTRFS_UUID_UNPARSED_SIZE, "%pU",
2471                                                 fs_info->fsid);
2472                 if (kobject_rename(&fs_info->fs_devices->fsid_kobj, fsid_buf))
2473                         btrfs_warn(fs_info,
2474                                    "sysfs: failed to create fsid for sprout");
2475         }
2476
2477         fs_info->num_tolerated_disk_barrier_failures =
2478                 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
2479         ret = btrfs_commit_transaction(trans);
2480
2481         if (seeding_dev) {
2482                 mutex_unlock(&uuid_mutex);
2483                 up_write(&sb->s_umount);
2484
2485                 if (ret) /* transaction commit */
2486                         return ret;
2487
2488                 ret = btrfs_relocate_sys_chunks(fs_info);
2489                 if (ret < 0)
2490                         btrfs_handle_fs_error(fs_info, ret,
2491                                     "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command.");
2492                 trans = btrfs_attach_transaction(root);
2493                 if (IS_ERR(trans)) {
2494                         if (PTR_ERR(trans) == -ENOENT)
2495                                 return 0;
2496                         return PTR_ERR(trans);
2497                 }
2498                 ret = btrfs_commit_transaction(trans);
2499         }
2500
2501         /* Update ctime/mtime for libblkid */
2502         update_dev_time(device_path);
2503         return ret;
2504
2505 error_trans:
2506         btrfs_end_transaction(trans);
2507         rcu_string_free(device->name);
2508         btrfs_sysfs_rm_device_link(fs_info->fs_devices, device);
2509         kfree(device);
2510 error:
2511         blkdev_put(bdev, FMODE_EXCL);
2512         if (seeding_dev) {
2513                 mutex_unlock(&uuid_mutex);
2514                 up_write(&sb->s_umount);
2515         }
2516         return ret;
2517 }
2518
2519 int btrfs_init_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
2520                                   const char *device_path,
2521                                   struct btrfs_device *srcdev,
2522                                   struct btrfs_device **device_out)
2523 {
2524         struct request_queue *q;
2525         struct btrfs_device *device;
2526         struct block_device *bdev;
2527         struct list_head *devices;
2528         struct rcu_string *name;
2529         u64 devid = BTRFS_DEV_REPLACE_DEVID;
2530         int ret = 0;
2531
2532         *device_out = NULL;
2533         if (fs_info->fs_devices->seeding) {
2534                 btrfs_err(fs_info, "the filesystem is a seed filesystem!");
2535                 return -EINVAL;
2536         }
2537
2538         bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2539                                   fs_info->bdev_holder);
2540         if (IS_ERR(bdev)) {
2541                 btrfs_err(fs_info, "target device %s is invalid!", device_path);
2542                 return PTR_ERR(bdev);
2543         }
2544
2545         filemap_write_and_wait(bdev->bd_inode->i_mapping);
2546
2547         devices = &fs_info->fs_devices->devices;
2548         list_for_each_entry(device, devices, dev_list) {
2549                 if (device->bdev == bdev) {
2550                         btrfs_err(fs_info,
2551                                   "target device is in the filesystem!");
2552                         ret = -EEXIST;
2553                         goto error;
2554                 }
2555         }
2556
2557
2558         if (i_size_read(bdev->bd_inode) <
2559             btrfs_device_get_total_bytes(srcdev)) {
2560                 btrfs_err(fs_info,
2561                           "target device is smaller than source device!");
2562                 ret = -EINVAL;
2563                 goto error;
2564         }
2565
2566
2567         device = btrfs_alloc_device(NULL, &devid, NULL);
2568         if (IS_ERR(device)) {
2569                 ret = PTR_ERR(device);
2570                 goto error;
2571         }
2572
2573         name = rcu_string_strdup(device_path, GFP_KERNEL);
2574         if (!name) {
2575                 kfree(device);
2576                 ret = -ENOMEM;
2577                 goto error;
2578         }
2579         rcu_assign_pointer(device->name, name);
2580
2581         q = bdev_get_queue(bdev);
2582         if (blk_queue_discard(q))
2583                 device->can_discard = 1;
2584         mutex_lock(&fs_info->fs_devices->device_list_mutex);
2585         device->writeable = 1;
2586         device->generation = 0;
2587         device->io_width = fs_info->sectorsize;
2588         device->io_align = fs_info->sectorsize;
2589         device->sector_size = fs_info->sectorsize;
2590         device->total_bytes = btrfs_device_get_total_bytes(srcdev);
2591         device->disk_total_bytes = btrfs_device_get_disk_total_bytes(srcdev);
2592         device->bytes_used = btrfs_device_get_bytes_used(srcdev);
2593         ASSERT(list_empty(&srcdev->resized_list));
2594         device->commit_total_bytes = srcdev->commit_total_bytes;
2595         device->commit_bytes_used = device->bytes_used;
2596         device->fs_info = fs_info;
2597         device->bdev = bdev;
2598         device->in_fs_metadata = 1;
2599         device->is_tgtdev_for_dev_replace = 1;
2600         device->mode = FMODE_EXCL;
2601         device->dev_stats_valid = 1;
2602         set_blocksize(device->bdev, 4096);
2603         device->fs_devices = fs_info->fs_devices;
2604         list_add(&device->dev_list, &fs_info->fs_devices->devices);
2605         fs_info->fs_devices->num_devices++;
2606         fs_info->fs_devices->open_devices++;
2607         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2608
2609         *device_out = device;
2610         return ret;
2611
2612 error:
2613         blkdev_put(bdev, FMODE_EXCL);
2614         return ret;
2615 }
2616
2617 void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info *fs_info,
2618                                               struct btrfs_device *tgtdev)
2619 {
2620         u32 sectorsize = fs_info->sectorsize;
2621
2622         WARN_ON(fs_info->fs_devices->rw_devices == 0);
2623         tgtdev->io_width = sectorsize;
2624         tgtdev->io_align = sectorsize;
2625         tgtdev->sector_size = sectorsize;
2626         tgtdev->fs_info = fs_info;
2627         tgtdev->in_fs_metadata = 1;
2628 }
2629
2630 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2631                                         struct btrfs_device *device)
2632 {
2633         int ret;
2634         struct btrfs_path *path;
2635         struct btrfs_root *root = device->fs_info->chunk_root;
2636         struct btrfs_dev_item *dev_item;
2637         struct extent_buffer *leaf;
2638         struct btrfs_key key;
2639
2640         path = btrfs_alloc_path();
2641         if (!path)
2642                 return -ENOMEM;
2643
2644         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2645         key.type = BTRFS_DEV_ITEM_KEY;
2646         key.offset = device->devid;
2647
2648         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2649         if (ret < 0)
2650                 goto out;
2651
2652         if (ret > 0) {
2653                 ret = -ENOENT;
2654                 goto out;
2655         }
2656
2657         leaf = path->nodes[0];
2658         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2659
2660         btrfs_set_device_id(leaf, dev_item, device->devid);
2661         btrfs_set_device_type(leaf, dev_item, device->type);
2662         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2663         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2664         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2665         btrfs_set_device_total_bytes(leaf, dev_item,
2666                                      btrfs_device_get_disk_total_bytes(device));
2667         btrfs_set_device_bytes_used(leaf, dev_item,
2668                                     btrfs_device_get_bytes_used(device));
2669         btrfs_mark_buffer_dirty(leaf);
2670
2671 out:
2672         btrfs_free_path(path);
2673         return ret;
2674 }
2675
2676 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2677                       struct btrfs_device *device, u64 new_size)
2678 {
2679         struct btrfs_fs_info *fs_info = device->fs_info;
2680         struct btrfs_super_block *super_copy = fs_info->super_copy;
2681         struct btrfs_fs_devices *fs_devices;
2682         u64 old_total;
2683         u64 diff;
2684
2685         if (!device->writeable)
2686                 return -EACCES;
2687
2688         mutex_lock(&fs_info->chunk_mutex);
2689         old_total = btrfs_super_total_bytes(super_copy);
2690         diff = new_size - device->total_bytes;
2691
2692         if (new_size <= device->total_bytes ||
2693             device->is_tgtdev_for_dev_replace) {
2694                 mutex_unlock(&fs_info->chunk_mutex);
2695                 return -EINVAL;
2696         }
2697
2698         fs_devices = fs_info->fs_devices;
2699
2700         btrfs_set_super_total_bytes(super_copy, old_total + diff);
2701         device->fs_devices->total_rw_bytes += diff;
2702
2703         btrfs_device_set_total_bytes(device, new_size);
2704         btrfs_device_set_disk_total_bytes(device, new_size);
2705         btrfs_clear_space_info_full(device->fs_info);
2706         if (list_empty(&device->resized_list))
2707                 list_add_tail(&device->resized_list,
2708                               &fs_devices->resized_devices);
2709         mutex_unlock(&fs_info->chunk_mutex);
2710
2711         return btrfs_update_device(trans, device);
2712 }
2713
2714 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
2715                             struct btrfs_fs_info *fs_info, u64 chunk_objectid,
2716                             u64 chunk_offset)
2717 {
2718         struct btrfs_root *root = fs_info->chunk_root;
2719         int ret;
2720         struct btrfs_path *path;
2721         struct btrfs_key key;
2722
2723         path = btrfs_alloc_path();
2724         if (!path)
2725                 return -ENOMEM;
2726
2727         key.objectid = chunk_objectid;
2728         key.offset = chunk_offset;
2729         key.type = BTRFS_CHUNK_ITEM_KEY;
2730
2731         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2732         if (ret < 0)
2733                 goto out;
2734         else if (ret > 0) { /* Logic error or corruption */
2735                 btrfs_handle_fs_error(fs_info, -ENOENT,
2736                                       "Failed lookup while freeing chunk.");
2737                 ret = -ENOENT;
2738                 goto out;
2739         }
2740
2741         ret = btrfs_del_item(trans, root, path);
2742         if (ret < 0)
2743                 btrfs_handle_fs_error(fs_info, ret,
2744                                       "Failed to delete chunk item.");
2745 out:
2746         btrfs_free_path(path);
2747         return ret;
2748 }
2749
2750 static int btrfs_del_sys_chunk(struct btrfs_fs_info *fs_info,
2751                                u64 chunk_objectid, u64 chunk_offset)
2752 {
2753         struct btrfs_super_block *super_copy = fs_info->super_copy;
2754         struct btrfs_disk_key *disk_key;
2755         struct btrfs_chunk *chunk;
2756         u8 *ptr;
2757         int ret = 0;
2758         u32 num_stripes;
2759         u32 array_size;
2760         u32 len = 0;
2761         u32 cur;
2762         struct btrfs_key key;
2763
2764         mutex_lock(&fs_info->chunk_mutex);
2765         array_size = btrfs_super_sys_array_size(super_copy);
2766
2767         ptr = super_copy->sys_chunk_array;
2768         cur = 0;
2769
2770         while (cur < array_size) {
2771                 disk_key = (struct btrfs_disk_key *)ptr;
2772                 btrfs_disk_key_to_cpu(&key, disk_key);
2773
2774                 len = sizeof(*disk_key);
2775
2776                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2777                         chunk = (struct btrfs_chunk *)(ptr + len);
2778                         num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2779                         len += btrfs_chunk_item_size(num_stripes);
2780                 } else {
2781                         ret = -EIO;
2782                         break;
2783                 }
2784                 if (key.objectid == chunk_objectid &&
2785                     key.offset == chunk_offset) {
2786                         memmove(ptr, ptr + len, array_size - (cur + len));
2787                         array_size -= len;
2788                         btrfs_set_super_sys_array_size(super_copy, array_size);
2789                 } else {
2790                         ptr += len;
2791                         cur += len;
2792                 }
2793         }
2794         mutex_unlock(&fs_info->chunk_mutex);
2795         return ret;
2796 }
2797
2798 static struct extent_map *get_chunk_map(struct btrfs_fs_info *fs_info,
2799                                         u64 logical, u64 length)
2800 {
2801         struct extent_map_tree *em_tree;
2802         struct extent_map *em;
2803
2804         em_tree = &fs_info->mapping_tree.map_tree;
2805         read_lock(&em_tree->lock);
2806         em = lookup_extent_mapping(em_tree, logical, length);
2807         read_unlock(&em_tree->lock);
2808
2809         if (!em) {
2810                 btrfs_crit(fs_info, "unable to find logical %llu length %llu",
2811                            logical, length);
2812                 return ERR_PTR(-EINVAL);
2813         }
2814
2815         if (em->start > logical || em->start + em->len < logical) {
2816                 btrfs_crit(fs_info,
2817                            "found a bad mapping, wanted %llu-%llu, found %llu-%llu",
2818                            logical, length, em->start, em->start + em->len);
2819                 free_extent_map(em);
2820                 return ERR_PTR(-EINVAL);
2821         }
2822
2823         /* callers are responsible for dropping em's ref. */
2824         return em;
2825 }
2826
2827 int btrfs_remove_chunk(struct btrfs_trans_handle *trans,
2828                        struct btrfs_fs_info *fs_info, u64 chunk_offset)
2829 {
2830         struct extent_map *em;
2831         struct map_lookup *map;
2832         u64 dev_extent_len = 0;
2833         u64 chunk_objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2834         int i, ret = 0;
2835         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2836
2837         em = get_chunk_map(fs_info, chunk_offset, 1);
2838         if (IS_ERR(em)) {
2839                 /*
2840                  * This is a logic error, but we don't want to just rely on the
2841                  * user having built with ASSERT enabled, so if ASSERT doesn't
2842                  * do anything we still error out.
2843                  */
2844                 ASSERT(0);
2845                 return PTR_ERR(em);
2846         }
2847         map = em->map_lookup;
2848         mutex_lock(&fs_info->chunk_mutex);
2849         check_system_chunk(trans, fs_info, map->type);
2850         mutex_unlock(&fs_info->chunk_mutex);
2851
2852         /*
2853          * Take the device list mutex to prevent races with the final phase of
2854          * a device replace operation that replaces the device object associated
2855          * with map stripes (dev-replace.c:btrfs_dev_replace_finishing()).
2856          */
2857         mutex_lock(&fs_devices->device_list_mutex);
2858         for (i = 0; i < map->num_stripes; i++) {
2859                 struct btrfs_device *device = map->stripes[i].dev;
2860                 ret = btrfs_free_dev_extent(trans, device,
2861                                             map->stripes[i].physical,
2862                                             &dev_extent_len);
2863                 if (ret) {
2864                         mutex_unlock(&fs_devices->device_list_mutex);
2865                         btrfs_abort_transaction(trans, ret);
2866                         goto out;
2867                 }
2868
2869                 if (device->bytes_used > 0) {
2870                         mutex_lock(&fs_info->chunk_mutex);
2871                         btrfs_device_set_bytes_used(device,
2872                                         device->bytes_used - dev_extent_len);
2873                         atomic64_add(dev_extent_len, &fs_info->free_chunk_space);
2874                         btrfs_clear_space_info_full(fs_info);
2875                         mutex_unlock(&fs_info->chunk_mutex);
2876                 }
2877
2878                 if (map->stripes[i].dev) {
2879                         ret = btrfs_update_device(trans, map->stripes[i].dev);
2880                         if (ret) {
2881                                 mutex_unlock(&fs_devices->device_list_mutex);
2882                                 btrfs_abort_transaction(trans, ret);
2883                                 goto out;
2884                         }
2885                 }
2886         }
2887         mutex_unlock(&fs_devices->device_list_mutex);
2888
2889         ret = btrfs_free_chunk(trans, fs_info, chunk_objectid, chunk_offset);
2890         if (ret) {
2891                 btrfs_abort_transaction(trans, ret);
2892                 goto out;
2893         }
2894
2895         trace_btrfs_chunk_free(fs_info, map, chunk_offset, em->len);
2896
2897         if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2898                 ret = btrfs_del_sys_chunk(fs_info, chunk_objectid,
2899                                           chunk_offset);
2900                 if (ret) {
2901                         btrfs_abort_transaction(trans, ret);
2902                         goto out;
2903                 }
2904         }
2905
2906         ret = btrfs_remove_block_group(trans, fs_info, chunk_offset, em);
2907         if (ret) {
2908                 btrfs_abort_transaction(trans, ret);
2909                 goto out;
2910         }
2911
2912 out:
2913         /* once for us */
2914         free_extent_map(em);
2915         return ret;
2916 }
2917
2918 static int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
2919 {
2920         struct btrfs_root *root = fs_info->chunk_root;
2921         struct btrfs_trans_handle *trans;
2922         int ret;
2923
2924         /*
2925          * Prevent races with automatic removal of unused block groups.
2926          * After we relocate and before we remove the chunk with offset
2927          * chunk_offset, automatic removal of the block group can kick in,
2928          * resulting in a failure when calling btrfs_remove_chunk() below.
2929          *
2930          * Make sure to acquire this mutex before doing a tree search (dev
2931          * or chunk trees) to find chunks. Otherwise the cleaner kthread might
2932          * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after
2933          * we release the path used to search the chunk/dev tree and before
2934          * the current task acquires this mutex and calls us.
2935          */
2936         ASSERT(mutex_is_locked(&fs_info->delete_unused_bgs_mutex));
2937
2938         ret = btrfs_can_relocate(fs_info, chunk_offset);
2939         if (ret)
2940                 return -ENOSPC;
2941
2942         /* step one, relocate all the extents inside this chunk */
2943         btrfs_scrub_pause(fs_info);
2944         ret = btrfs_relocate_block_group(fs_info, chunk_offset);
2945         btrfs_scrub_continue(fs_info);
2946         if (ret)
2947                 return ret;
2948
2949         trans = btrfs_start_trans_remove_block_group(root->fs_info,
2950                                                      chunk_offset);
2951         if (IS_ERR(trans)) {
2952                 ret = PTR_ERR(trans);
2953                 btrfs_handle_fs_error(root->fs_info, ret, NULL);
2954                 return ret;
2955         }
2956
2957         /*
2958          * step two, delete the device extents and the
2959          * chunk tree entries
2960          */
2961         ret = btrfs_remove_chunk(trans, fs_info, chunk_offset);
2962         btrfs_end_transaction(trans);
2963         return ret;
2964 }
2965
2966 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info)
2967 {
2968         struct btrfs_root *chunk_root = fs_info->chunk_root;
2969         struct btrfs_path *path;
2970         struct extent_buffer *leaf;
2971         struct btrfs_chunk *chunk;
2972         struct btrfs_key key;
2973         struct btrfs_key found_key;
2974         u64 chunk_type;
2975         bool retried = false;
2976         int failed = 0;
2977         int ret;
2978
2979         path = btrfs_alloc_path();
2980         if (!path)
2981                 return -ENOMEM;
2982
2983 again:
2984         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2985         key.offset = (u64)-1;
2986         key.type = BTRFS_CHUNK_ITEM_KEY;
2987
2988         while (1) {
2989                 mutex_lock(&fs_info->delete_unused_bgs_mutex);
2990                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2991                 if (ret < 0) {
2992                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
2993                         goto error;
2994                 }
2995                 BUG_ON(ret == 0); /* Corruption */
2996
2997                 ret = btrfs_previous_item(chunk_root, path, key.objectid,
2998                                           key.type);
2999                 if (ret)
3000                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3001                 if (ret < 0)
3002                         goto error;
3003                 if (ret > 0)
3004                         break;
3005
3006                 leaf = path->nodes[0];
3007                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3008
3009                 chunk = btrfs_item_ptr(leaf, path->slots[0],
3010                                        struct btrfs_chunk);
3011                 chunk_type = btrfs_chunk_type(leaf, chunk);
3012                 btrfs_release_path(path);
3013
3014                 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
3015                         ret = btrfs_relocate_chunk(fs_info, found_key.offset);
3016                         if (ret == -ENOSPC)
3017                                 failed++;
3018                         else
3019                                 BUG_ON(ret);
3020                 }
3021                 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3022
3023                 if (found_key.offset == 0)
3024                         break;
3025                 key.offset = found_key.offset - 1;
3026         }
3027         ret = 0;
3028         if (failed && !retried) {
3029                 failed = 0;
3030                 retried = true;
3031                 goto again;
3032         } else if (WARN_ON(failed && retried)) {
3033                 ret = -ENOSPC;
3034         }
3035 error:
3036         btrfs_free_path(path);
3037         return ret;
3038 }
3039
3040 static int insert_balance_item(struct btrfs_fs_info *fs_info,
3041                                struct btrfs_balance_control *bctl)
3042 {
3043         struct btrfs_root *root = fs_info->tree_root;
3044         struct btrfs_trans_handle *trans;
3045         struct btrfs_balance_item *item;
3046         struct btrfs_disk_balance_args disk_bargs;
3047         struct btrfs_path *path;
3048         struct extent_buffer *leaf;
3049         struct btrfs_key key;
3050         int ret, err;
3051
3052         path = btrfs_alloc_path();
3053         if (!path)
3054                 return -ENOMEM;
3055
3056         trans = btrfs_start_transaction(root, 0);
3057         if (IS_ERR(trans)) {
3058                 btrfs_free_path(path);
3059                 return PTR_ERR(trans);
3060         }
3061
3062         key.objectid = BTRFS_BALANCE_OBJECTID;
3063         key.type = BTRFS_TEMPORARY_ITEM_KEY;
3064         key.offset = 0;
3065
3066         ret = btrfs_insert_empty_item(trans, root, path, &key,
3067                                       sizeof(*item));
3068         if (ret)
3069                 goto out;
3070
3071         leaf = path->nodes[0];
3072         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3073
3074         memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item));
3075
3076         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
3077         btrfs_set_balance_data(leaf, item, &disk_bargs);
3078         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
3079         btrfs_set_balance_meta(leaf, item, &disk_bargs);
3080         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
3081         btrfs_set_balance_sys(leaf, item, &disk_bargs);
3082
3083         btrfs_set_balance_flags(leaf, item, bctl->flags);
3084
3085         btrfs_mark_buffer_dirty(leaf);
3086 out:
3087         btrfs_free_path(path);
3088         err = btrfs_commit_transaction(trans);
3089         if (err && !ret)
3090                 ret = err;
3091         return ret;
3092 }
3093
3094 static int del_balance_item(struct btrfs_fs_info *fs_info)
3095 {
3096         struct btrfs_root *root = fs_info->tree_root;
3097         struct btrfs_trans_handle *trans;
3098         struct btrfs_path *path;
3099         struct btrfs_key key;
3100         int ret, err;
3101
3102         path = btrfs_alloc_path();
3103         if (!path)
3104                 return -ENOMEM;
3105
3106         trans = btrfs_start_transaction(root, 0);
3107         if (IS_ERR(trans)) {
3108                 btrfs_free_path(path);
3109                 return PTR_ERR(trans);
3110         }
3111
3112         key.objectid = BTRFS_BALANCE_OBJECTID;
3113         key.type = BTRFS_TEMPORARY_ITEM_KEY;
3114         key.offset = 0;
3115
3116         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3117         if (ret < 0)
3118                 goto out;
3119         if (ret > 0) {
3120                 ret = -ENOENT;
3121                 goto out;
3122         }
3123
3124         ret = btrfs_del_item(trans, root, path);
3125 out:
3126         btrfs_free_path(path);
3127         err = btrfs_commit_transaction(trans);
3128         if (err && !ret)
3129                 ret = err;
3130         return ret;
3131 }
3132
3133 /*
3134  * This is a heuristic used to reduce the number of chunks balanced on
3135  * resume after balance was interrupted.
3136  */
3137 static void update_balance_args(struct btrfs_balance_control *bctl)
3138 {
3139         /*
3140          * Turn on soft mode for chunk types that were being converted.
3141          */
3142         if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
3143                 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
3144         if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
3145                 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
3146         if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
3147                 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
3148
3149         /*
3150          * Turn on usage filter if is not already used.  The idea is
3151          * that chunks that we have already balanced should be
3152          * reasonably full.  Don't do it for chunks that are being
3153          * converted - that will keep us from relocating unconverted
3154          * (albeit full) chunks.
3155          */
3156         if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3157             !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3158             !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3159                 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
3160                 bctl->data.usage = 90;
3161         }
3162         if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3163             !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3164             !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3165                 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
3166                 bctl->sys.usage = 90;
3167         }
3168         if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3169             !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3170             !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3171                 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
3172                 bctl->meta.usage = 90;
3173         }
3174 }
3175
3176 /*
3177  * Should be called with both balance and volume mutexes held to
3178  * serialize other volume operations (add_dev/rm_dev/resize) with
3179  * restriper.  Same goes for unset_balance_control.
3180  */
3181 static void set_balance_control(struct btrfs_balance_control *bctl)
3182 {
3183         struct btrfs_fs_info *fs_info = bctl->fs_info;
3184
3185         BUG_ON(fs_info->balance_ctl);
3186
3187         spin_lock(&fs_info->balance_lock);
3188         fs_info->balance_ctl = bctl;
3189         spin_unlock(&fs_info->balance_lock);
3190 }
3191
3192 static void unset_balance_control(struct btrfs_fs_info *fs_info)
3193 {
3194         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3195
3196         BUG_ON(!fs_info->balance_ctl);
3197
3198         spin_lock(&fs_info->balance_lock);
3199         fs_info->balance_ctl = NULL;
3200         spin_unlock(&fs_info->balance_lock);
3201
3202         kfree(bctl);
3203 }
3204
3205 /*
3206  * Balance filters.  Return 1 if chunk should be filtered out
3207  * (should not be balanced).
3208  */
3209 static int chunk_profiles_filter(u64 chunk_type,
3210                                  struct btrfs_balance_args *bargs)
3211 {
3212         chunk_type = chunk_to_extended(chunk_type) &
3213                                 BTRFS_EXTENDED_PROFILE_MASK;
3214
3215         if (bargs->profiles & chunk_type)
3216                 return 0;
3217
3218         return 1;
3219 }
3220
3221 static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
3222                               struct btrfs_balance_args *bargs)
3223 {
3224         struct btrfs_block_group_cache *cache;
3225         u64 chunk_used;
3226         u64 user_thresh_min;
3227         u64 user_thresh_max;
3228         int ret = 1;
3229
3230         cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3231         chunk_used = btrfs_block_group_used(&cache->item);
3232
3233         if (bargs->usage_min == 0)
3234                 user_thresh_min = 0;
3235         else
3236                 user_thresh_min = div_factor_fine(cache->key.offset,
3237                                         bargs->usage_min);
3238
3239         if (bargs->usage_max == 0)
3240                 user_thresh_max = 1;
3241         else if (bargs->usage_max > 100)
3242                 user_thresh_max = cache->key.offset;
3243         else
3244                 user_thresh_max = div_factor_fine(cache->key.offset,
3245                                         bargs->usage_max);
3246
3247         if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max)
3248                 ret = 0;
3249
3250         btrfs_put_block_group(cache);
3251         return ret;
3252 }
3253
3254 static int chunk_usage_filter(struct btrfs_fs_info *fs_info,
3255                 u64 chunk_offset, struct btrfs_balance_args *bargs)
3256 {
3257         struct btrfs_block_group_cache *cache;
3258         u64 chunk_used, user_thresh;
3259         int ret = 1;
3260
3261         cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3262         chunk_used = btrfs_block_group_used(&cache->item);
3263
3264         if (bargs->usage_min == 0)
3265                 user_thresh = 1;
3266         else if (bargs->usage > 100)
3267                 user_thresh = cache->key.offset;
3268         else
3269                 user_thresh = div_factor_fine(cache->key.offset,
3270                                               bargs->usage);
3271
3272         if (chunk_used < user_thresh)
3273                 ret = 0;
3274
3275         btrfs_put_block_group(cache);
3276         return ret;
3277 }
3278
3279 static int chunk_devid_filter(struct extent_buffer *leaf,
3280                               struct btrfs_chunk *chunk,
3281                               struct btrfs_balance_args *bargs)
3282 {
3283         struct btrfs_stripe *stripe;
3284         int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3285         int i;
3286
3287         for (i = 0; i < num_stripes; i++) {
3288                 stripe = btrfs_stripe_nr(chunk, i);
3289                 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
3290                         return 0;
3291         }
3292
3293         return 1;
3294 }
3295
3296 /* [pstart, pend) */
3297 static int chunk_drange_filter(struct extent_buffer *leaf,
3298                                struct btrfs_chunk *chunk,
3299                                u64 chunk_offset,
3300                                struct btrfs_balance_args *bargs)
3301 {
3302         struct btrfs_stripe *stripe;
3303         int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3304         u64 stripe_offset;
3305         u64 stripe_length;
3306         int factor;
3307         int i;
3308
3309         if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
3310                 return 0;
3311
3312         if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP |
3313              BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) {
3314                 factor = num_stripes / 2;
3315         } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) {
3316                 factor = num_stripes - 1;
3317         } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) {
3318                 factor = num_stripes - 2;
3319         } else {
3320                 factor = num_stripes;
3321         }
3322
3323         for (i = 0; i < num_stripes; i++) {
3324                 stripe = btrfs_stripe_nr(chunk, i);
3325                 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
3326                         continue;
3327
3328                 stripe_offset = btrfs_stripe_offset(leaf, stripe);
3329                 stripe_length = btrfs_chunk_length(leaf, chunk);
3330                 stripe_length = div_u64(stripe_length, factor);
3331
3332                 if (stripe_offset < bargs->pend &&
3333                     stripe_offset + stripe_length > bargs->pstart)
3334                         return 0;
3335         }
3336
3337         return 1;
3338 }
3339
3340 /* [vstart, vend) */
3341 static int chunk_vrange_filter(struct extent_buffer *leaf,
3342                                struct btrfs_chunk *chunk,
3343                                u64 chunk_offset,
3344                                struct btrfs_balance_args *bargs)
3345 {
3346         if (chunk_offset < bargs->vend &&
3347             chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
3348                 /* at least part of the chunk is inside this vrange */
3349                 return 0;
3350
3351         return 1;
3352 }
3353
3354 static int chunk_stripes_range_filter(struct extent_buffer *leaf,
3355                                struct btrfs_chunk *chunk,
3356                                struct btrfs_balance_args *bargs)
3357 {
3358         int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3359
3360         if (bargs->stripes_min <= num_stripes
3361                         && num_stripes <= bargs->stripes_max)
3362                 return 0;
3363
3364         return 1;
3365 }
3366
3367 static int chunk_soft_convert_filter(u64 chunk_type,
3368                                      struct btrfs_balance_args *bargs)
3369 {
3370         if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
3371                 return 0;
3372
3373         chunk_type = chunk_to_extended(chunk_type) &
3374                                 BTRFS_EXTENDED_PROFILE_MASK;
3375
3376         if (bargs->target == chunk_type)
3377                 return 1;
3378
3379         return 0;
3380 }
3381
3382 static int should_balance_chunk(struct btrfs_fs_info *fs_info,
3383                                 struct extent_buffer *leaf,
3384                                 struct btrfs_chunk *chunk, u64 chunk_offset)
3385 {
3386         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3387         struct btrfs_balance_args *bargs = NULL;
3388         u64 chunk_type = btrfs_chunk_type(leaf, chunk);
3389
3390         /* type filter */
3391         if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
3392               (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
3393                 return 0;
3394         }
3395
3396         if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3397                 bargs = &bctl->data;
3398         else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3399                 bargs = &bctl->sys;
3400         else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3401                 bargs = &bctl->meta;
3402
3403         /* profiles filter */
3404         if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
3405             chunk_profiles_filter(chunk_type, bargs)) {
3406                 return 0;
3407         }
3408
3409         /* usage filter */
3410         if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
3411             chunk_usage_filter(fs_info, chunk_offset, bargs)) {
3412                 return 0;
3413         } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3414             chunk_usage_range_filter(fs_info, chunk_offset, bargs)) {
3415                 return 0;
3416         }
3417
3418         /* devid filter */
3419         if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
3420             chunk_devid_filter(leaf, chunk, bargs)) {
3421                 return 0;
3422         }
3423
3424         /* drange filter, makes sense only with devid filter */
3425         if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
3426             chunk_drange_filter(leaf, chunk, chunk_offset, bargs)) {
3427                 return 0;
3428         }
3429
3430         /* vrange filter */
3431         if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
3432             chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
3433                 return 0;
3434         }
3435
3436         /* stripes filter */
3437         if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) &&
3438             chunk_stripes_range_filter(leaf, chunk, bargs)) {
3439                 return 0;
3440         }
3441
3442         /* soft profile changing mode */
3443         if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
3444             chunk_soft_convert_filter(chunk_type, bargs)) {
3445                 return 0;
3446         }
3447
3448         /*
3449          * limited by count, must be the last filter
3450          */
3451         if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) {
3452                 if (bargs->limit == 0)
3453                         return 0;
3454                 else
3455                         bargs->limit--;
3456         } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) {
3457                 /*
3458                  * Same logic as the 'limit' filter; the minimum cannot be
3459                  * determined here because we do not have the global information
3460                  * about the count of all chunks that satisfy the filters.
3461                  */
3462                 if (bargs->limit_max == 0)
3463                         return 0;
3464                 else
3465                         bargs->limit_max--;
3466         }
3467
3468         return 1;
3469 }
3470
3471 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
3472 {
3473         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3474         struct btrfs_root *chunk_root = fs_info->chunk_root;
3475         struct btrfs_root *dev_root = fs_info->dev_root;
3476         struct list_head *devices;
3477         struct btrfs_device *device;
3478         u64 old_size;
3479         u64 size_to_free;
3480         u64 chunk_type;
3481         struct btrfs_chunk *chunk;
3482         struct btrfs_path *path = NULL;
3483         struct btrfs_key key;
3484         struct btrfs_key found_key;
3485         struct btrfs_trans_handle *trans;
3486         struct extent_buffer *leaf;
3487         int slot;
3488         int ret;
3489         int enospc_errors = 0;
3490         bool counting = true;
3491         /* The single value limit and min/max limits use the same bytes in the */
3492         u64 limit_data = bctl->data.limit;
3493         u64 limit_meta = bctl->meta.limit;
3494         u64 limit_sys = bctl->sys.limit;
3495         u32 count_data = 0;
3496         u32 count_meta = 0;
3497         u32 count_sys = 0;
3498         int chunk_reserved = 0;
3499         u64 bytes_used = 0;
3500
3501         /* step one make some room on all the devices */
3502         devices = &fs_info->fs_devices->devices;
3503         list_for_each_entry(device, devices, dev_list) {
3504                 old_size = btrfs_device_get_total_bytes(device);
3505                 size_to_free = div_factor(old_size, 1);
3506                 size_to_free = min_t(u64, size_to_free, SZ_1M);
3507                 if (!device->writeable ||
3508                     btrfs_device_get_total_bytes(device) -
3509                     btrfs_device_get_bytes_used(device) > size_to_free ||
3510                     device->is_tgtdev_for_dev_replace)
3511                         continue;
3512
3513                 ret = btrfs_shrink_device(device, old_size - size_to_free);
3514                 if (ret == -ENOSPC)
3515                         break;
3516                 if (ret) {
3517                         /* btrfs_shrink_device never returns ret > 0 */
3518                         WARN_ON(ret > 0);
3519                         goto error;
3520                 }
3521
3522                 trans = btrfs_start_transaction(dev_root, 0);
3523                 if (IS_ERR(trans)) {
3524                         ret = PTR_ERR(trans);
3525                         btrfs_info_in_rcu(fs_info,
3526                  "resize: unable to start transaction after shrinking device %s (error %d), old size %llu, new size %llu",
3527                                           rcu_str_deref(device->name), ret,
3528                                           old_size, old_size - size_to_free);
3529                         goto error;
3530                 }
3531
3532                 ret = btrfs_grow_device(trans, device, old_size);
3533                 if (ret) {
3534                         btrfs_end_transaction(trans);
3535                         /* btrfs_grow_device never returns ret > 0 */
3536                         WARN_ON(ret > 0);
3537                         btrfs_info_in_rcu(fs_info,
3538                  "resize: unable to grow device after shrinking device %s (error %d), old size %llu, new size %llu",
3539                                           rcu_str_deref(device->name), ret,
3540                                           old_size, old_size - size_to_free);
3541                         goto error;
3542                 }
3543
3544                 btrfs_end_transaction(trans);
3545         }
3546
3547         /* step two, relocate all the chunks */
3548         path = btrfs_alloc_path();
3549         if (!path) {
3550                 ret = -ENOMEM;
3551                 goto error;
3552         }
3553
3554         /* zero out stat counters */
3555         spin_lock(&fs_info->balance_lock);
3556         memset(&bctl->stat, 0, sizeof(bctl->stat));
3557         spin_unlock(&fs_info->balance_lock);
3558 again:
3559         if (!counting) {
3560                 /*
3561                  * The single value limit and min/max limits use the same bytes
3562                  * in the
3563                  */
3564                 bctl->data.limit = limit_data;
3565                 bctl->meta.limit = limit_meta;
3566                 bctl->sys.limit = limit_sys;
3567         }
3568         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3569         key.offset = (u64)-1;
3570         key.type = BTRFS_CHUNK_ITEM_KEY;
3571
3572         while (1) {
3573                 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
3574                     atomic_read(&fs_info->balance_cancel_req)) {
3575                         ret = -ECANCELED;
3576                         goto error;
3577                 }
3578
3579                 mutex_lock(&fs_info->delete_unused_bgs_mutex);
3580                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3581                 if (ret < 0) {
3582                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3583                         goto error;
3584                 }
3585
3586                 /*
3587                  * this shouldn't happen, it means the last relocate
3588                  * failed
3589                  */
3590                 if (ret == 0)
3591                         BUG(); /* FIXME break ? */
3592
3593                 ret = btrfs_previous_item(chunk_root, path, 0,
3594                                           BTRFS_CHUNK_ITEM_KEY);
3595                 if (ret) {
3596                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3597                         ret = 0;
3598                         break;
3599                 }
3600
3601                 leaf = path->nodes[0];
3602                 slot = path->slots[0];
3603                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3604
3605                 if (found_key.objectid != key.objectid) {
3606                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3607                         break;
3608                 }
3609
3610                 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3611                 chunk_type = btrfs_chunk_type(leaf, chunk);
3612
3613                 if (!counting) {
3614                         spin_lock(&fs_info->balance_lock);
3615                         bctl->stat.considered++;
3616                         spin_unlock(&fs_info->balance_lock);
3617                 }
3618
3619                 ret = should_balance_chunk(fs_info, leaf, chunk,
3620                                            found_key.offset);
3621
3622                 btrfs_release_path(path);
3623                 if (!ret) {
3624                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3625                         goto loop;
3626                 }
3627
3628                 if (counting) {
3629                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3630                         spin_lock(&fs_info->balance_lock);
3631                         bctl->stat.expected++;
3632                         spin_unlock(&fs_info->balance_lock);
3633
3634                         if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3635                                 count_data++;
3636                         else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3637                                 count_sys++;
3638                         else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3639                                 count_meta++;
3640
3641                         goto loop;
3642                 }
3643
3644                 /*
3645                  * Apply limit_min filter, no need to check if the LIMITS
3646                  * filter is used, limit_min is 0 by default
3647                  */
3648                 if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
3649                                         count_data < bctl->data.limit_min)
3650                                 || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) &&
3651                                         count_meta < bctl->meta.limit_min)
3652                                 || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) &&
3653                                         count_sys < bctl->sys.limit_min)) {
3654                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3655                         goto loop;
3656                 }
3657
3658                 ASSERT(fs_info->data_sinfo);
3659                 spin_lock(&fs_info->data_sinfo->lock);
3660                 bytes_used = fs_info->data_sinfo->bytes_used;
3661                 spin_unlock(&fs_info->data_sinfo->lock);
3662
3663                 if ((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
3664                     !chunk_reserved && !bytes_used) {
3665                         trans = btrfs_start_transaction(chunk_root, 0);
3666                         if (IS_ERR(trans)) {
3667                                 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3668                                 ret = PTR_ERR(trans);
3669                                 goto error;
3670                         }
3671
3672                         ret = btrfs_force_chunk_alloc(trans, fs_info,
3673                                                       BTRFS_BLOCK_GROUP_DATA);
3674                         btrfs_end_transaction(trans);
3675                         if (ret < 0) {
3676                                 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3677                                 goto error;
3678                         }
3679                         chunk_reserved = 1;
3680                 }
3681
3682                 ret = btrfs_relocate_chunk(fs_info, found_key.offset);
3683                 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3684                 if (ret && ret != -ENOSPC)
3685                         goto error;
3686                 if (ret == -ENOSPC) {
3687                         enospc_errors++;
3688                 } else {
3689                         spin_lock(&fs_info->balance_lock);
3690                         bctl->stat.completed++;
3691                         spin_unlock(&fs_info->balance_lock);
3692                 }
3693 loop:
3694                 if (found_key.offset == 0)
3695                         break;
3696                 key.offset = found_key.offset - 1;
3697         }
3698
3699         if (counting) {
3700                 btrfs_release_path(path);
3701                 counting = false;
3702                 goto again;
3703         }
3704 error:
3705         btrfs_free_path(path);
3706         if (enospc_errors) {
3707                 btrfs_info(fs_info, "%d enospc errors during balance",
3708                            enospc_errors);
3709                 if (!ret)
3710                         ret = -ENOSPC;
3711         }
3712
3713         return ret;
3714 }
3715
3716 /**
3717  * alloc_profile_is_valid - see if a given profile is valid and reduced
3718  * @flags: profile to validate
3719  * @extended: if true @flags is treated as an extended profile
3720  */
3721 static int alloc_profile_is_valid(u64 flags, int extended)
3722 {
3723         u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
3724                                BTRFS_BLOCK_GROUP_PROFILE_MASK);
3725
3726         flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
3727
3728         /* 1) check that all other bits are zeroed */
3729         if (flags & ~mask)
3730                 return 0;
3731
3732         /* 2) see if profile is reduced */
3733         if (flags == 0)
3734                 return !extended; /* "0" is valid for usual profiles */
3735
3736         /* true if exactly one bit set */
3737         return (flags & (flags - 1)) == 0;
3738 }
3739
3740 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
3741 {
3742         /* cancel requested || normal exit path */
3743         return atomic_read(&fs_info->balance_cancel_req) ||
3744                 (atomic_read(&fs_info->balance_pause_req) == 0 &&
3745                  atomic_read(&fs_info->balance_cancel_req) == 0);
3746 }
3747
3748 static void __cancel_balance(struct btrfs_fs_info *fs_info)
3749 {
3750         int ret;
3751
3752         unset_balance_control(fs_info);
3753         ret = del_balance_item(fs_info);
3754         if (ret)
3755                 btrfs_handle_fs_error(fs_info, ret, NULL);
3756
3757         clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
3758 }
3759
3760 /* Non-zero return value signifies invalidity */
3761 static inline int validate_convert_profile(struct btrfs_balance_args *bctl_arg,
3762                 u64 allowed)
3763 {
3764         return ((bctl_arg->flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3765                 (!alloc_profile_is_valid(bctl_arg->target, 1) ||
3766                  (bctl_arg->target & ~allowed)));
3767 }
3768
3769 /*
3770  * Should be called with both balance and volume mutexes held
3771  */
3772 int btrfs_balance(struct btrfs_balance_control *bctl,
3773                   struct btrfs_ioctl_balance_args *bargs)
3774 {
3775         struct btrfs_fs_info *fs_info = bctl->fs_info;
3776         u64 meta_target, data_target;
3777         u64 allowed;
3778         int mixed = 0;
3779         int ret;
3780         u64 num_devices;
3781         unsigned seq;
3782
3783         if (btrfs_fs_closing(fs_info) ||
3784             atomic_read(&fs_info->balance_pause_req) ||
3785             atomic_read(&fs_info->balance_cancel_req)) {
3786                 ret = -EINVAL;
3787                 goto out;
3788         }
3789
3790         allowed = btrfs_super_incompat_flags(fs_info->super_copy);
3791         if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
3792                 mixed = 1;
3793
3794         /*
3795          * In case of mixed groups both data and meta should be picked,
3796          * and identical options should be given for both of them.
3797          */
3798         allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
3799         if (mixed && (bctl->flags & allowed)) {
3800                 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
3801                     !(bctl->flags & BTRFS_BALANCE_METADATA) ||
3802                     memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
3803                         btrfs_err(fs_info,
3804                                   "with mixed groups data and metadata balance options must be the same");
3805                         ret = -EINVAL;
3806                         goto out;
3807                 }
3808         }
3809
3810         num_devices = fs_info->fs_devices->num_devices;
3811         btrfs_dev_replace_lock(&fs_info->dev_replace, 0);
3812         if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
3813                 BUG_ON(num_devices < 1);
3814                 num_devices--;
3815         }
3816         btrfs_dev_replace_unlock(&fs_info->dev_replace, 0);
3817         allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE | BTRFS_BLOCK_GROUP_DUP;
3818         if (num_devices > 1)
3819                 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1);
3820         if (num_devices > 2)
3821                 allowed |= BTRFS_BLOCK_GROUP_RAID5;
3822         if (num_devices > 3)
3823                 allowed |= (BTRFS_BLOCK_GROUP_RAID10 |
3824                             BTRFS_BLOCK_GROUP_RAID6);
3825         if (validate_convert_profile(&bctl->data, allowed)) {
3826                 btrfs_err(fs_info,
3827                           "unable to start balance with target data profile %llu",
3828                           bctl->data.target);
3829                 ret = -EINVAL;
3830                 goto out;
3831         }
3832         if (validate_convert_profile(&bctl->meta, allowed)) {
3833                 btrfs_err(fs_info,
3834                           "unable to start balance with target metadata profile %llu",
3835                           bctl->meta.target);
3836                 ret = -EINVAL;
3837                 goto out;
3838         }
3839         if (validate_convert_profile(&bctl->sys, allowed)) {
3840                 btrfs_err(fs_info,
3841                           "unable to start balance with target system profile %llu",
3842                           bctl->sys.target);
3843                 ret = -EINVAL;
3844                 goto out;
3845         }
3846
3847         /* allow to reduce meta or sys integrity only if force set */
3848         allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3849                         BTRFS_BLOCK_GROUP_RAID10 |
3850                         BTRFS_BLOCK_GROUP_RAID5 |
3851                         BTRFS_BLOCK_GROUP_RAID6;
3852         do {
3853                 seq = read_seqbegin(&fs_info->profiles_lock);
3854
3855                 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3856                      (fs_info->avail_system_alloc_bits & allowed) &&
3857                      !(bctl->sys.target & allowed)) ||
3858                     ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3859                      (fs_info->avail_metadata_alloc_bits & allowed) &&
3860                      !(bctl->meta.target & allowed))) {
3861                         if (bctl->flags & BTRFS_BALANCE_FORCE) {
3862                                 btrfs_info(fs_info,
3863                                            "force reducing metadata integrity");
3864                         } else {
3865                                 btrfs_err(fs_info,
3866                                           "balance will reduce metadata integrity, use force if you want this");
3867                                 ret = -EINVAL;
3868                                 goto out;
3869                         }
3870                 }
3871         } while (read_seqretry(&fs_info->profiles_lock, seq));
3872
3873         /* if we're not converting, the target field is uninitialized */
3874         meta_target = (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
3875                 bctl->meta.target : fs_info->avail_metadata_alloc_bits;
3876         data_target = (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
3877                 bctl->data.target : fs_info->avail_data_alloc_bits;
3878         if (btrfs_get_num_tolerated_disk_barrier_failures(meta_target) <
3879                 btrfs_get_num_tolerated_disk_barrier_failures(data_target)) {
3880                 btrfs_warn(fs_info,
3881                            "metadata profile 0x%llx has lower redundancy than data profile 0x%llx",
3882                            meta_target, data_target);
3883         }
3884
3885         if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3886                 fs_info->num_tolerated_disk_barrier_failures = min(
3887                         btrfs_calc_num_tolerated_disk_barrier_failures(fs_info),
3888                         btrfs_get_num_tolerated_disk_barrier_failures(
3889                                 bctl->sys.target));
3890         }
3891
3892         ret = insert_balance_item(fs_info, bctl);
3893         if (ret && ret != -EEXIST)
3894                 goto out;
3895
3896         if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
3897                 BUG_ON(ret == -EEXIST);
3898                 set_balance_control(bctl);
3899         } else {
3900                 BUG_ON(ret != -EEXIST);
3901                 spin_lock(&fs_info->balance_lock);
3902                 update_balance_args(bctl);
3903                 spin_unlock(&fs_info->balance_lock);
3904         }
3905
3906         atomic_inc(&fs_info->balance_running);
3907         mutex_unlock(&fs_info->balance_mutex);
3908
3909         ret = __btrfs_balance(fs_info);
3910
3911         mutex_lock(&fs_info->balance_mutex);
3912         atomic_dec(&fs_info->balance_running);
3913
3914         if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3915                 fs_info->num_tolerated_disk_barrier_failures =
3916                         btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3917         }
3918
3919         if (bargs) {
3920                 memset(bargs, 0, sizeof(*bargs));
3921                 update_ioctl_balance_args(fs_info, 0, bargs);
3922         }
3923
3924         if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
3925             balance_need_close(fs_info)) {
3926                 __cancel_balance(fs_info);
3927         }
3928
3929         wake_up(&fs_info->balance_wait_q);
3930
3931         return ret;
3932 out:
3933         if (bctl->flags & BTRFS_BALANCE_RESUME)
3934                 __cancel_balance(fs_info);
3935         else {
3936                 kfree(bctl);
3937                 clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
3938         }
3939         return ret;
3940 }
3941
3942 static int balance_kthread(void *data)
3943 {
3944         struct btrfs_fs_info *fs_info = data;
3945         int ret = 0;
3946
3947         mutex_lock(&fs_info->volume_mutex);
3948         mutex_lock(&fs_info->balance_mutex);
3949
3950         if (fs_info->balance_ctl) {
3951                 btrfs_info(fs_info, "continuing balance");
3952                 ret = btrfs_balance(fs_info->balance_ctl, NULL);
3953         }
3954
3955         mutex_unlock(&fs_info->balance_mutex);
3956         mutex_unlock(&fs_info->volume_mutex);
3957
3958         return ret;
3959 }
3960
3961 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
3962 {
3963         struct task_struct *tsk;
3964
3965         spin_lock(&fs_info->balance_lock);
3966         if (!fs_info->balance_ctl) {
3967                 spin_unlock(&fs_info->balance_lock);
3968                 return 0;
3969         }
3970         spin_unlock(&fs_info->balance_lock);
3971
3972         if (btrfs_test_opt(fs_info, SKIP_BALANCE)) {
3973                 btrfs_info(fs_info, "force skipping balance");
3974                 return 0;
3975         }
3976
3977         tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
3978         return PTR_ERR_OR_ZERO(tsk);
3979 }
3980
3981 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
3982 {
3983         struct btrfs_balance_control *bctl;
3984         struct btrfs_balance_item *item;
3985         struct btrfs_disk_balance_args disk_bargs;
3986         struct btrfs_path *path;
3987         struct extent_buffer *leaf;
3988         struct btrfs_key key;
3989         int ret;
3990
3991         path = btrfs_alloc_path();
3992         if (!path)
3993                 return -ENOMEM;
3994
3995         key.objectid = BTRFS_BALANCE_OBJECTID;
3996         key.type = BTRFS_TEMPORARY_ITEM_KEY;
3997         key.offset = 0;
3998
3999         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
4000         if (ret < 0)
4001                 goto out;
4002         if (ret > 0) { /* ret = -ENOENT; */
4003                 ret = 0;
4004                 goto out;
4005         }
4006
4007         bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
4008         if (!bctl) {
4009                 ret = -ENOMEM;
4010                 goto out;
4011         }
4012
4013         leaf = path->nodes[0];
4014         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
4015
4016         bctl->fs_info = fs_info;
4017         bctl->flags = btrfs_balance_flags(leaf, item);
4018         bctl->flags |= BTRFS_BALANCE_RESUME;
4019
4020         btrfs_balance_data(leaf, item, &disk_bargs);
4021         btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
4022         btrfs_balance_meta(leaf, item, &disk_bargs);
4023         btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
4024         btrfs_balance_sys(leaf, item, &disk_bargs);
4025         btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
4026
4027         WARN_ON(test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags));
4028
4029         mutex_lock(&fs_info->volume_mutex);
4030         mutex_lock(&fs_info->balance_mutex);
4031
4032         set_balance_control(bctl);
4033
4034         mutex_unlock(&fs_info->balance_mutex);
4035         mutex_unlock(&fs_info->volume_mutex);
4036 out:
4037         btrfs_free_path(path);
4038         return ret;
4039 }
4040
4041 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
4042 {
4043         int ret = 0;
4044
4045         mutex_lock(&fs_info->balance_mutex);
4046         if (!fs_info->balance_ctl) {
4047                 mutex_unlock(&fs_info->balance_mutex);
4048                 return -ENOTCONN;
4049         }
4050
4051         if (atomic_read(&fs_info->balance_running)) {
4052                 atomic_inc(&fs_info->balance_pause_req);
4053                 mutex_unlock(&fs_info->balance_mutex);
4054
4055                 wait_event(fs_info->balance_wait_q,
4056                            atomic_read(&fs_info->balance_running) == 0);
4057
4058                 mutex_lock(&fs_info->balance_mutex);
4059                 /* we are good with balance_ctl ripped off from under us */
4060                 BUG_ON(atomic_read(&fs_info->balance_running));
4061                 atomic_dec(&fs_info->balance_pause_req);
4062         } else {
4063                 ret = -ENOTCONN;
4064         }
4065
4066         mutex_unlock(&fs_info->balance_mutex);
4067         return ret;
4068 }
4069
4070 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
4071 {
4072         if (fs_info->sb->s_flags & MS_RDONLY)
4073                 return -EROFS;
4074
4075         mutex_lock(&fs_info->balance_mutex);
4076         if (!fs_info->balance_ctl) {
4077                 mutex_unlock(&fs_info->balance_mutex);
4078                 return -ENOTCONN;
4079         }
4080
4081         atomic_inc(&fs_info->balance_cancel_req);
4082         /*
4083          * if we are running just wait and return, balance item is
4084          * deleted in btrfs_balance in this case
4085          */
4086         if (atomic_read(&fs_info->balance_running)) {
4087                 mutex_unlock(&fs_info->balance_mutex);
4088                 wait_event(fs_info->balance_wait_q,
4089                            atomic_read(&fs_info->balance_running) == 0);
4090                 mutex_lock(&fs_info->balance_mutex);
4091         } else {
4092                 /* __cancel_balance needs volume_mutex */
4093                 mutex_unlock(&fs_info->balance_mutex);
4094                 mutex_lock(&fs_info->volume_mutex);
4095                 mutex_lock(&fs_info->balance_mutex);
4096
4097                 if (fs_info->balance_ctl)
4098                         __cancel_balance(fs_info);
4099
4100                 mutex_unlock(&fs_info->volume_mutex);
4101         }
4102
4103         BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running));
4104         atomic_dec(&fs_info->balance_cancel_req);
4105         mutex_unlock(&fs_info->balance_mutex);
4106         return 0;
4107 }
4108
4109 static int btrfs_uuid_scan_kthread(void *data)
4110 {
4111         struct btrfs_fs_info *fs_info = data;
4112         struct btrfs_root *root = fs_info->tree_root;
4113         struct btrfs_key key;
4114         struct btrfs_key max_key;
4115         struct btrfs_path *path = NULL;
4116         int ret = 0;
4117         struct extent_buffer *eb;
4118         int slot;
4119         struct btrfs_root_item root_item;
4120         u32 item_size;
4121         struct btrfs_trans_handle *trans = NULL;
4122
4123         path = btrfs_alloc_path();
4124         if (!path) {
4125                 ret = -ENOMEM;
4126                 goto out;
4127         }
4128
4129         key.objectid = 0;
4130         key.type = BTRFS_ROOT_ITEM_KEY;
4131         key.offset = 0;
4132
4133         max_key.objectid = (u64)-1;
4134         max_key.type = BTRFS_ROOT_ITEM_KEY;
4135         max_key.offset = (u64)-1;
4136
4137         while (1) {
4138                 ret = btrfs_search_forward(root, &key, path, 0);
4139                 if (ret) {
4140                         if (ret > 0)
4141                                 ret = 0;
4142                         break;
4143                 }
4144
4145                 if (key.type != BTRFS_ROOT_ITEM_KEY ||
4146                     (key.objectid < BTRFS_FIRST_FREE_OBJECTID &&
4147                      key.objectid != BTRFS_FS_TREE_OBJECTID) ||
4148                     key.objectid > BTRFS_LAST_FREE_OBJECTID)
4149                         goto skip;
4150
4151                 eb = path->nodes[0];
4152                 slot = path->slots[0];
4153                 item_size = btrfs_item_size_nr(eb, slot);
4154                 if (item_size < sizeof(root_item))
4155                         goto skip;
4156
4157                 read_extent_buffer(eb, &root_item,
4158                                    btrfs_item_ptr_offset(eb, slot),
4159                                    (int)sizeof(root_item));
4160                 if (btrfs_root_refs(&root_item) == 0)
4161                         goto skip;
4162
4163                 if (!btrfs_is_empty_uuid(root_item.uuid) ||
4164                     !btrfs_is_empty_uuid(root_item.received_uuid)) {
4165                         if (trans)
4166                                 goto update_tree;
4167
4168                         btrfs_release_path(path);
4169                         /*
4170                          * 1 - subvol uuid item
4171                          * 1 - received_subvol uuid item
4172                          */
4173                         trans = btrfs_start_transaction(fs_info->uuid_root, 2);
4174                         if (IS_ERR(trans)) {
4175                                 ret = PTR_ERR(trans);
4176                                 break;
4177                         }
4178                         continue;
4179                 } else {
4180                         goto skip;
4181                 }
4182 update_tree:
4183                 if (!btrfs_is_empty_uuid(root_item.uuid)) {
4184                         ret = btrfs_uuid_tree_add(trans, fs_info,
4185                                                   root_item.uuid,
4186                                                   BTRFS_UUID_KEY_SUBVOL,
4187                                                   key.objectid);
4188                         if (ret < 0) {
4189                                 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4190                                         ret);
4191                                 break;
4192                         }
4193                 }
4194
4195                 if (!btrfs_is_empty_uuid(root_item.received_uuid)) {
4196                         ret = btrfs_uuid_tree_add(trans, fs_info,
4197                                                   root_item.received_uuid,
4198                                                  BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4199                                                   key.objectid);
4200                         if (ret < 0) {
4201                                 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4202                                         ret);
4203                                 break;
4204                         }
4205                 }
4206
4207 skip:
4208                 if (trans) {
4209                         ret = btrfs_end_transaction(trans);
4210                         trans = NULL;
4211                         if (ret)
4212                                 break;
4213                 }
4214
4215                 btrfs_release_path(path);
4216                 if (key.offset < (u64)-1) {
4217                         key.offset++;
4218                 } else if (key.type < BTRFS_ROOT_ITEM_KEY) {
4219                         key.offset = 0;
4220                         key.type = BTRFS_ROOT_ITEM_KEY;
4221                 } else if (key.objectid < (u64)-1) {
4222                         key.offset = 0;
4223                         key.type = BTRFS_ROOT_ITEM_KEY;
4224                         key.objectid++;
4225                 } else {
4226                         break;
4227                 }
4228                 cond_resched();
4229         }
4230
4231 out:
4232         btrfs_free_path(path);
4233         if (trans && !IS_ERR(trans))
4234                 btrfs_end_transaction(trans);
4235         if (ret)
4236                 btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret);
4237         else
4238                 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
4239         up(&fs_info->uuid_tree_rescan_sem);
4240         return 0;
4241 }
4242
4243 /*
4244  * Callback for btrfs_uuid_tree_iterate().
4245  * returns:
4246  * 0    check succeeded, the entry is not outdated.
4247  * < 0  if an error occurred.
4248  * > 0  if the check failed, which means the caller shall remove the entry.
4249  */
4250 static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info *fs_info,
4251                                        u8 *uuid, u8 type, u64 subid)
4252 {
4253         struct btrfs_key key;
4254         int ret = 0;
4255         struct btrfs_root *subvol_root;
4256
4257         if (type != BTRFS_UUID_KEY_SUBVOL &&
4258             type != BTRFS_UUID_KEY_RECEIVED_SUBVOL)
4259                 goto out;
4260
4261         key.objectid = subid;
4262         key.type = BTRFS_ROOT_ITEM_KEY;
4263         key.offset = (u64)-1;
4264         subvol_root = btrfs_read_fs_root_no_name(fs_info, &key);
4265         if (IS_ERR(subvol_root)) {
4266                 ret = PTR_ERR(subvol_root);
4267                 if (ret == -ENOENT)
4268                         ret = 1;
4269                 goto out;
4270         }
4271
4272         switch (type) {
4273         case BTRFS_UUID_KEY_SUBVOL:
4274                 if (memcmp(uuid, subvol_root->root_item.uuid, BTRFS_UUID_SIZE))
4275                         ret = 1;
4276                 break;
4277         case BTRFS_UUID_KEY_RECEIVED_SUBVOL:
4278                 if (memcmp(uuid, subvol_root->root_item.received_uuid,
4279                            BTRFS_UUID_SIZE))
4280                         ret = 1;
4281                 break;
4282         }
4283
4284 out:
4285         return ret;
4286 }
4287
4288 static int btrfs_uuid_rescan_kthread(void *data)
4289 {
4290         struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data;
4291         int ret;
4292
4293         /*
4294          * 1st step is to iterate through the existing UUID tree and
4295          * to delete all entries that contain outdated data.
4296          * 2nd step is to add all missing entries to the UUID tree.
4297          */
4298         ret = btrfs_uuid_tree_iterate(fs_info, btrfs_check_uuid_tree_entry);
4299         if (ret < 0) {
4300                 btrfs_warn(fs_info, "iterating uuid_tree failed %d", ret);
4301                 up(&fs_info->uuid_tree_rescan_sem);
4302                 return ret;
4303         }
4304         return btrfs_uuid_scan_kthread(data);
4305 }
4306
4307 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info)
4308 {
4309         struct btrfs_trans_handle *trans;
4310         struct btrfs_root *tree_root = fs_info->tree_root;
4311         struct btrfs_root *uuid_root;
4312         struct task_struct *task;
4313         int ret;
4314
4315         /*
4316          * 1 - root node
4317          * 1 - root item
4318          */
4319         trans = btrfs_start_transaction(tree_root, 2);
4320         if (IS_ERR(trans))
4321                 return PTR_ERR(trans);
4322
4323         uuid_root = btrfs_create_tree(trans, fs_info,
4324                                       BTRFS_UUID_TREE_OBJECTID);
4325         if (IS_ERR(uuid_root)) {
4326                 ret = PTR_ERR(uuid_root);
4327                 btrfs_abort_transaction(trans, ret);
4328                 btrfs_end_transaction(trans);
4329                 return ret;
4330         }
4331
4332         fs_info->uuid_root = uuid_root;
4333
4334         ret = btrfs_commit_transaction(trans);
4335         if (ret)
4336                 return ret;
4337
4338         down(&fs_info->uuid_tree_rescan_sem);
4339         task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid");
4340         if (IS_ERR(task)) {
4341                 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4342                 btrfs_warn(fs_info, "failed to start uuid_scan task");
4343                 up(&fs_info->uuid_tree_rescan_sem);
4344                 return PTR_ERR(task);
4345         }
4346
4347         return 0;
4348 }
4349
4350 int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
4351 {
4352         struct task_struct *task;
4353
4354         down(&fs_info->uuid_tree_rescan_sem);
4355         task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
4356         if (IS_ERR(task)) {
4357                 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4358                 btrfs_warn(fs_info, "failed to start uuid_rescan task");
4359                 up(&fs_info->uuid_tree_rescan_sem);
4360                 return PTR_ERR(task);
4361         }
4362
4363         return 0;
4364 }
4365
4366 /*
4367  * shrinking a device means finding all of the device extents past
4368  * the new size, and then following the back refs to the chunks.
4369  * The chunk relocation code actually frees the device extent
4370  */
4371 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
4372 {
4373         struct btrfs_fs_info *fs_info = device->fs_info;
4374         struct btrfs_root *root = fs_info->dev_root;
4375         struct btrfs_trans_handle *trans;
4376         struct btrfs_dev_extent *dev_extent = NULL;
4377         struct btrfs_path *path;
4378         u64 length;
4379         u64 chunk_offset;
4380         int ret;
4381         int slot;
4382         int failed = 0;
4383         bool retried = false;
4384         bool checked_pending_chunks = false;
4385         struct extent_buffer *l;
4386         struct btrfs_key key;
4387         struct btrfs_super_block *super_copy = fs_info->super_copy;
4388         u64 old_total = btrfs_super_total_bytes(super_copy);
4389         u64 old_size = btrfs_device_get_total_bytes(device);
4390         u64 diff = old_size - new_size;
4391
4392         if (device->is_tgtdev_for_dev_replace)
4393                 return -EINVAL;
4394
4395         path = btrfs_alloc_path();
4396         if (!path)
4397                 return -ENOMEM;
4398
4399         path->reada = READA_FORWARD;
4400
4401         mutex_lock(&fs_info->chunk_mutex);
4402
4403         btrfs_device_set_total_bytes(device, new_size);
4404         if (device->writeable) {
4405                 device->fs_devices->total_rw_bytes -= diff;
4406                 atomic64_sub(diff, &fs_info->free_chunk_space);
4407         }
4408         mutex_unlock(&fs_info->chunk_mutex);
4409
4410 again:
4411         key.objectid = device->devid;
4412         key.offset = (u64)-1;
4413         key.type = BTRFS_DEV_EXTENT_KEY;
4414
4415         do {
4416                 mutex_lock(&fs_info->delete_unused_bgs_mutex);
4417                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4418                 if (ret < 0) {
4419                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4420                         goto done;
4421                 }
4422
4423                 ret = btrfs_previous_item(root, path, 0, key.type);
4424                 if (ret)
4425                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4426                 if (ret < 0)
4427                         goto done;
4428                 if (ret) {
4429                         ret = 0;
4430                         btrfs_release_path(path);
4431                         break;
4432                 }
4433
4434                 l = path->nodes[0];
4435                 slot = path->slots[0];
4436                 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
4437
4438                 if (key.objectid != device->devid) {
4439                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4440                         btrfs_release_path(path);
4441                         break;
4442                 }
4443
4444                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
4445                 length = btrfs_dev_extent_length(l, dev_extent);
4446
4447                 if (key.offset + length <= new_size) {
4448                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4449                         btrfs_release_path(path);
4450                         break;
4451                 }
4452
4453                 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
4454                 btrfs_release_path(path);
4455
4456                 ret = btrfs_relocate_chunk(fs_info, chunk_offset);
4457                 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4458                 if (ret && ret != -ENOSPC)
4459                         goto done;
4460                 if (ret == -ENOSPC)
4461                         failed++;
4462         } while (key.offset-- > 0);
4463
4464         if (failed && !retried) {
4465                 failed = 0;
4466                 retried = true;
4467                 goto again;
4468         } else if (failed && retried) {
4469                 ret = -ENOSPC;
4470                 goto done;
4471         }
4472
4473         /* Shrinking succeeded, else we would be at "done". */
4474         trans = btrfs_start_transaction(root, 0);
4475         if (IS_ERR(trans)) {
4476                 ret = PTR_ERR(trans);
4477                 goto done;
4478         }
4479
4480         mutex_lock(&fs_info->chunk_mutex);
4481
4482         /*
4483          * We checked in the above loop all device extents that were already in
4484          * the device tree. However before we have updated the device's
4485          * total_bytes to the new size, we might have had chunk allocations that
4486          * have not complete yet (new block groups attached to transaction
4487          * handles), and therefore their device extents were not yet in the
4488          * device tree and we missed them in the loop above. So if we have any
4489          * pending chunk using a device extent that overlaps the device range
4490          * that we can not use anymore, commit the current transaction and
4491          * repeat the search on the device tree - this way we guarantee we will
4492          * not have chunks using device extents that end beyond 'new_size'.
4493          */
4494         if (!checked_pending_chunks) {
4495                 u64 start = new_size;
4496                 u64 len = old_size - new_size;
4497
4498                 if (contains_pending_extent(trans->transaction, device,
4499                                             &start, len)) {
4500                         mutex_unlock(&fs_info->chunk_mutex);
4501                         checked_pending_chunks = true;
4502                         failed = 0;
4503                         retried = false;
4504                         ret = btrfs_commit_transaction(trans);
4505                         if (ret)
4506                                 goto done;
4507                         goto again;
4508                 }
4509         }
4510
4511         btrfs_device_set_disk_total_bytes(device, new_size);
4512         if (list_empty(&device->resized_list))
4513                 list_add_tail(&device->resized_list,
4514                               &fs_info->fs_devices->resized_devices);
4515
4516         WARN_ON(diff > old_total);
4517         btrfs_set_super_total_bytes(super_copy, old_total - diff);
4518         mutex_unlock(&fs_info->chunk_mutex);
4519
4520         /* Now btrfs_update_device() will change the on-disk size. */
4521         ret = btrfs_update_device(trans, device);
4522         btrfs_end_transaction(trans);
4523 done:
4524         btrfs_free_path(path);
4525         if (ret) {
4526                 mutex_lock(&fs_info->chunk_mutex);
4527                 btrfs_device_set_total_bytes(device, old_size);
4528                 if (device->writeable)
4529                         device->fs_devices->total_rw_bytes += diff;
4530                 atomic64_add(diff, &fs_info->free_chunk_space);
4531                 mutex_unlock(&fs_info->chunk_mutex);
4532         }
4533         return ret;
4534 }
4535
4536 static int btrfs_add_system_chunk(struct btrfs_fs_info *fs_info,
4537                            struct btrfs_key *key,
4538                            struct btrfs_chunk *chunk, int item_size)
4539 {
4540         struct btrfs_super_block *super_copy = fs_info->super_copy;
4541         struct btrfs_disk_key disk_key;
4542         u32 array_size;
4543         u8 *ptr;
4544
4545         mutex_lock(&fs_info->chunk_mutex);
4546         array_size = btrfs_super_sys_array_size(super_copy);
4547         if (array_size + item_size + sizeof(disk_key)
4548                         > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
4549                 mutex_unlock(&fs_info->chunk_mutex);
4550                 return -EFBIG;
4551         }
4552
4553         ptr = super_copy->sys_chunk_array + array_size;
4554         btrfs_cpu_key_to_disk(&disk_key, key);
4555         memcpy(ptr, &disk_key, sizeof(disk_key));
4556         ptr += sizeof(disk_key);
4557         memcpy(ptr, chunk, item_size);
4558         item_size += sizeof(disk_key);
4559         btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
4560         mutex_unlock(&fs_info->chunk_mutex);
4561
4562         return 0;
4563 }
4564
4565 /*
4566  * sort the devices in descending order by max_avail, total_avail
4567  */
4568 static int btrfs_cmp_device_info(const void *a, const void *b)
4569 {
4570         const struct btrfs_device_info *di_a = a;
4571         const struct btrfs_device_info *di_b = b;
4572
4573         if (di_a->max_avail > di_b->max_avail)
4574                 return -1;
4575         if (di_a->max_avail < di_b->max_avail)
4576                 return 1;
4577         if (di_a->total_avail > di_b->total_avail)
4578                 return -1;
4579         if (di_a->total_avail < di_b->total_avail)
4580                 return 1;
4581         return 0;
4582 }
4583
4584 static u32 find_raid56_stripe_len(u32 data_devices, u32 dev_stripe_target)
4585 {
4586         /* TODO allow them to set a preferred stripe size */
4587         return SZ_64K;
4588 }
4589
4590 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
4591 {
4592         if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK))
4593                 return;
4594
4595         btrfs_set_fs_incompat(info, RAID56);
4596 }
4597
4598 #define BTRFS_MAX_DEVS(r) ((BTRFS_MAX_ITEM_SIZE(r->fs_info)             \
4599                         - sizeof(struct btrfs_chunk))           \
4600                         / sizeof(struct btrfs_stripe) + 1)
4601
4602 #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE        \
4603                                 - 2 * sizeof(struct btrfs_disk_key)     \
4604                                 - 2 * sizeof(struct btrfs_chunk))       \
4605                                 / sizeof(struct btrfs_stripe) + 1)
4606
4607 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4608                                u64 start, u64 type)
4609 {
4610         struct btrfs_fs_info *info = trans->fs_info;
4611         struct btrfs_fs_devices *fs_devices = info->fs_devices;
4612         struct list_head *cur;
4613         struct map_lookup *map = NULL;
4614         struct extent_map_tree *em_tree;
4615         struct extent_map *em;
4616         struct btrfs_device_info *devices_info = NULL;
4617         u64 total_avail;
4618         int num_stripes;        /* total number of stripes to allocate */
4619         int data_stripes;       /* number of stripes that count for
4620                                    block group size */
4621         int sub_stripes;        /* sub_stripes info for map */
4622         int dev_stripes;        /* stripes per dev */
4623         int devs_max;           /* max devs to use */
4624         int devs_min;           /* min devs needed */
4625         int devs_increment;     /* ndevs has to be a multiple of this */
4626         int ncopies;            /* how many copies to data has */
4627         int ret;
4628         u64 max_stripe_size;
4629         u64 max_chunk_size;
4630         u64 stripe_size;
4631         u64 num_bytes;
4632         u64 raid_stripe_len = BTRFS_STRIPE_LEN;
4633         int ndevs;
4634         int i;
4635         int j;
4636         int index;
4637
4638         BUG_ON(!alloc_profile_is_valid(type, 0));
4639
4640         if (list_empty(&fs_devices->alloc_list))
4641                 return -ENOSPC;
4642
4643         index = __get_raid_index(type);
4644
4645         sub_stripes = btrfs_raid_array[index].sub_stripes;
4646         dev_stripes = btrfs_raid_array[index].dev_stripes;
4647         devs_max = btrfs_raid_array[index].devs_max;
4648         devs_min = btrfs_raid_array[index].devs_min;
4649         devs_increment = btrfs_raid_array[index].devs_increment;
4650         ncopies = btrfs_raid_array[index].ncopies;
4651
4652         if (type & BTRFS_BLOCK_GROUP_DATA) {
4653                 max_stripe_size = SZ_1G;
4654                 max_chunk_size = 10 * max_stripe_size;
4655                 if (!devs_max)
4656                         devs_max = BTRFS_MAX_DEVS(info->chunk_root);
4657         } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
4658                 /* for larger filesystems, use larger metadata chunks */
4659                 if (fs_devices->total_rw_bytes > 50ULL * SZ_1G)
4660                         max_stripe_size = SZ_1G;
4661                 else
4662                         max_stripe_size = SZ_256M;
4663                 max_chunk_size = max_stripe_size;
4664                 if (!devs_max)
4665                         devs_max = BTRFS_MAX_DEVS(info->chunk_root);
4666         } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
4667                 max_stripe_size = SZ_32M;
4668                 max_chunk_size = 2 * max_stripe_size;
4669                 if (!devs_max)
4670                         devs_max = BTRFS_MAX_DEVS_SYS_CHUNK;
4671         } else {
4672                 btrfs_err(info, "invalid chunk type 0x%llx requested",
4673                        type);
4674                 BUG_ON(1);
4675         }
4676
4677         /* we don't want a chunk larger than 10% of writeable space */
4678         max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
4679                              max_chunk_size);
4680
4681         devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info),
4682                                GFP_NOFS);
4683         if (!devices_info)
4684                 return -ENOMEM;
4685
4686         cur = fs_devices->alloc_list.next;
4687
4688         /*
4689          * in the first pass through the devices list, we gather information
4690          * about the available holes on each device.
4691          */
4692         ndevs = 0;
4693         while (cur != &fs_devices->alloc_list) {
4694                 struct btrfs_device *device;
4695                 u64 max_avail;
4696                 u64 dev_offset;
4697
4698                 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
4699
4700                 cur = cur->next;
4701
4702                 if (!device->writeable) {
4703                         WARN(1, KERN_ERR
4704                                "BTRFS: read-only device in alloc_list\n");
4705                         continue;
4706                 }
4707
4708                 if (!device->in_fs_metadata ||
4709                     device->is_tgtdev_for_dev_replace)
4710                         continue;
4711
4712                 if (device->total_bytes > device->bytes_used)
4713                         total_avail = device->total_bytes - device->bytes_used;
4714                 else
4715                         total_avail = 0;
4716
4717                 /* If there is no space on this device, skip it. */
4718                 if (total_avail == 0)
4719                         continue;
4720
4721                 ret = find_free_dev_extent(trans, device,
4722                                            max_stripe_size * dev_stripes,
4723                                            &dev_offset, &max_avail);
4724                 if (ret && ret != -ENOSPC)
4725                         goto error;
4726
4727                 if (ret == 0)
4728                         max_avail = max_stripe_size * dev_stripes;
4729
4730                 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes)
4731                         continue;
4732
4733                 if (ndevs == fs_devices->rw_devices) {
4734                         WARN(1, "%s: found more than %llu devices\n",
4735                              __func__, fs_devices->rw_devices);
4736                         break;
4737                 }
4738                 devices_info[ndevs].dev_offset = dev_offset;
4739                 devices_info[ndevs].max_avail = max_avail;
4740                 devices_info[ndevs].total_avail = total_avail;
4741                 devices_info[ndevs].dev = device;
4742                 ++ndevs;
4743         }
4744
4745         /*
4746          * now sort the devices by hole size / available space
4747          */
4748         sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
4749              btrfs_cmp_device_info, NULL);
4750
4751         /* round down to number of usable stripes */
4752         ndevs -= ndevs % devs_increment;
4753
4754         if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) {
4755                 ret = -ENOSPC;
4756                 goto error;
4757         }
4758
4759         if (devs_max && ndevs > devs_max)
4760                 ndevs = devs_max;
4761         /*
4762          * the primary goal is to maximize the number of stripes, so use as many
4763          * devices as possible, even if the stripes are not maximum sized.
4764          */
4765         stripe_size = devices_info[ndevs-1].max_avail;
4766         num_stripes = ndevs * dev_stripes;
4767
4768         /*
4769          * this will have to be fixed for RAID1 and RAID10 over
4770          * more drives
4771          */
4772         data_stripes = num_stripes / ncopies;
4773
4774         if (type & BTRFS_BLOCK_GROUP_RAID5) {
4775                 raid_stripe_len = find_raid56_stripe_len(ndevs - 1,
4776                                                          info->stripesize);
4777                 data_stripes = num_stripes - 1;
4778         }
4779         if (type & BTRFS_BLOCK_GROUP_RAID6) {
4780                 raid_stripe_len = find_raid56_stripe_len(ndevs - 2,
4781                                                          info->stripesize);
4782                 data_stripes = num_stripes - 2;
4783         }
4784
4785         /*
4786          * Use the number of data stripes to figure out how big this chunk
4787          * is really going to be in terms of logical address space,
4788          * and compare that answer with the max chunk size
4789          */
4790         if (stripe_size * data_stripes > max_chunk_size) {
4791                 u64 mask = (1ULL << 24) - 1;
4792
4793                 stripe_size = div_u64(max_chunk_size, data_stripes);
4794
4795                 /* bump the answer up to a 16MB boundary */
4796                 stripe_size = (stripe_size + mask) & ~mask;
4797
4798                 /* but don't go higher than the limits we found
4799                  * while searching for free extents
4800                  */
4801                 if (stripe_size > devices_info[ndevs-1].max_avail)
4802                         stripe_size = devices_info[ndevs-1].max_avail;
4803         }
4804
4805         stripe_size = div_u64(stripe_size, dev_stripes);
4806
4807         /* align to BTRFS_STRIPE_LEN */
4808         stripe_size = div64_u64(stripe_size, raid_stripe_len);
4809         stripe_size *= raid_stripe_len;
4810
4811         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
4812         if (!map) {
4813                 ret = -ENOMEM;
4814                 goto error;
4815         }
4816         map->num_stripes = num_stripes;
4817
4818         for (i = 0; i < ndevs; ++i) {
4819                 for (j = 0; j < dev_stripes; ++j) {
4820                         int s = i * dev_stripes + j;
4821                         map->stripes[s].dev = devices_info[i].dev;
4822                         map->stripes[s].physical = devices_info[i].dev_offset +
4823                                                    j * stripe_size;
4824                 }
4825         }
4826         map->sector_size = info->sectorsize;
4827         map->stripe_len = raid_stripe_len;
4828         map->io_align = raid_stripe_len;
4829         map->io_width = raid_stripe_len;
4830         map->type = type;
4831         map->sub_stripes = sub_stripes;
4832
4833         num_bytes = stripe_size * data_stripes;
4834
4835         trace_btrfs_chunk_alloc(info, map, start, num_bytes);
4836
4837         em = alloc_extent_map();
4838         if (!em) {
4839                 kfree(map);
4840                 ret = -ENOMEM;
4841                 goto error;
4842         }
4843         set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
4844         em->map_lookup = map;
4845         em->start = start;
4846         em->len = num_bytes;
4847         em->block_start = 0;
4848         em->block_len = em->len;
4849         em->orig_block_len = stripe_size;
4850
4851         em_tree = &info->mapping_tree.map_tree;
4852         write_lock(&em_tree->lock);
4853         ret = add_extent_mapping(em_tree, em, 0);
4854         if (!ret) {
4855                 list_add_tail(&em->list, &trans->transaction->pending_chunks);
4856                 refcount_inc(&em->refs);
4857         }
4858         write_unlock(&em_tree->lock);
4859         if (ret) {
4860                 free_extent_map(em);
4861                 goto error;
4862         }
4863
4864         ret = btrfs_make_block_group(trans, info, 0, type,
4865                                      BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4866                                      start, num_bytes);
4867         if (ret)
4868                 goto error_del_extent;
4869
4870         for (i = 0; i < map->num_stripes; i++) {
4871                 num_bytes = map->stripes[i].dev->bytes_used + stripe_size;
4872                 btrfs_device_set_bytes_used(map->stripes[i].dev, num_bytes);
4873         }
4874
4875         atomic64_sub(stripe_size * map->num_stripes, &info->free_chunk_space);
4876
4877         free_extent_map(em);
4878         check_raid56_incompat_flag(info, type);
4879
4880         kfree(devices_info);
4881         return 0;
4882
4883 error_del_extent:
4884         write_lock(&em_tree->lock);
4885         remove_extent_mapping(em_tree, em);
4886         write_unlock(&em_tree->lock);
4887
4888         /* One for our allocation */
4889         free_extent_map(em);
4890         /* One for the tree reference */
4891         free_extent_map(em);
4892         /* One for the pending_chunks list reference */
4893         free_extent_map(em);
4894 error:
4895         kfree(devices_info);
4896         return ret;
4897 }
4898
4899 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans,
4900                                 struct btrfs_fs_info *fs_info,
4901                                 u64 chunk_offset, u64 chunk_size)
4902 {
4903         struct btrfs_root *extent_root = fs_info->extent_root;
4904         struct btrfs_root *chunk_root = fs_info->chunk_root;
4905         struct btrfs_key key;
4906         struct btrfs_device *device;
4907         struct btrfs_chunk *chunk;
4908         struct btrfs_stripe *stripe;
4909         struct extent_map *em;
4910         struct map_lookup *map;
4911         size_t item_size;
4912         u64 dev_offset;
4913         u64 stripe_size;
4914         int i = 0;
4915         int ret = 0;
4916
4917         em = get_chunk_map(fs_info, chunk_offset, chunk_size);
4918         if (IS_ERR(em))
4919                 return PTR_ERR(em);
4920
4921         map = em->map_lookup;
4922         item_size = btrfs_chunk_item_size(map->num_stripes);
4923         stripe_size = em->orig_block_len;
4924
4925         chunk = kzalloc(item_size, GFP_NOFS);
4926         if (!chunk) {
4927                 ret = -ENOMEM;
4928                 goto out;
4929         }
4930
4931         /*
4932          * Take the device list mutex to prevent races with the final phase of
4933          * a device replace operation that replaces the device object associated
4934          * with the map's stripes, because the device object's id can change
4935          * at any time during that final phase of the device replace operation
4936          * (dev-replace.c:btrfs_dev_replace_finishing()).
4937          */
4938         mutex_lock(&fs_info->fs_devices->device_list_mutex);
4939         for (i = 0; i < map->num_stripes; i++) {
4940                 device = map->stripes[i].dev;
4941                 dev_offset = map->stripes[i].physical;
4942
4943                 ret = btrfs_update_device(trans, device);
4944                 if (ret)
4945                         break;
4946                 ret = btrfs_alloc_dev_extent(trans, device,
4947                                              chunk_root->root_key.objectid,
4948                                              BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4949                                              chunk_offset, dev_offset,
4950                                              stripe_size);
4951                 if (ret)
4952                         break;
4953         }
4954         if (ret) {
4955                 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4956                 goto out;
4957         }
4958
4959         stripe = &chunk->stripe;
4960         for (i = 0; i < map->num_stripes; i++) {
4961                 device = map->stripes[i].dev;
4962                 dev_offset = map->stripes[i].physical;
4963
4964                 btrfs_set_stack_stripe_devid(stripe, device->devid);
4965                 btrfs_set_stack_stripe_offset(stripe, dev_offset);
4966                 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
4967                 stripe++;
4968         }
4969         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4970
4971         btrfs_set_stack_chunk_length(chunk, chunk_size);
4972         btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
4973         btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
4974         btrfs_set_stack_chunk_type(chunk, map->type);
4975         btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
4976         btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
4977         btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
4978         btrfs_set_stack_chunk_sector_size(chunk, fs_info->sectorsize);
4979         btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
4980
4981         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
4982         key.type = BTRFS_CHUNK_ITEM_KEY;
4983         key.offset = chunk_offset;
4984
4985         ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
4986         if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
4987                 /*
4988                  * TODO: Cleanup of inserted chunk root in case of
4989                  * failure.
4990                  */
4991                 ret = btrfs_add_system_chunk(fs_info, &key, chunk, item_size);
4992         }
4993
4994 out:
4995         kfree(chunk);
4996         free_extent_map(em);
4997         return ret;
4998 }
4999
5000 /*
5001  * Chunk allocation falls into two parts. The first part does works
5002  * that make the new allocated chunk useable, but not do any operation
5003  * that modifies the chunk tree. The second part does the works that
5004  * require modifying the chunk tree. This division is important for the
5005  * bootstrap process of adding storage to a seed btrfs.
5006  */
5007 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
5008                       struct btrfs_fs_info *fs_info, u64 type)
5009 {
5010         u64 chunk_offset;
5011
5012         ASSERT(mutex_is_locked(&fs_info->chunk_mutex));
5013         chunk_offset = find_next_chunk(fs_info);
5014         return __btrfs_alloc_chunk(trans, chunk_offset, type);
5015 }
5016
5017 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
5018                                          struct btrfs_fs_info *fs_info)
5019 {
5020         u64 chunk_offset;
5021         u64 sys_chunk_offset;
5022         u64 alloc_profile;
5023         int ret;
5024
5025         chunk_offset = find_next_chunk(fs_info);
5026         alloc_profile = btrfs_metadata_alloc_profile(fs_info);
5027         ret = __btrfs_alloc_chunk(trans, chunk_offset, alloc_profile);
5028         if (ret)
5029                 return ret;
5030
5031         sys_chunk_offset = find_next_chunk(fs_info);
5032         alloc_profile = btrfs_system_alloc_profile(fs_info);
5033         ret = __btrfs_alloc_chunk(trans, sys_chunk_offset, alloc_profile);
5034         return ret;
5035 }
5036
5037 static inline int btrfs_chunk_max_errors(struct map_lookup *map)
5038 {
5039         int max_errors;
5040
5041         if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
5042                          BTRFS_BLOCK_GROUP_RAID10 |
5043                          BTRFS_BLOCK_GROUP_RAID5 |
5044                          BTRFS_BLOCK_GROUP_DUP)) {
5045                 max_errors = 1;
5046         } else if (map->type & BTRFS_BLOCK_GROUP_RAID6) {
5047                 max_errors = 2;
5048         } else {
5049                 max_errors = 0;
5050         }
5051
5052         return max_errors;
5053 }
5054
5055 int btrfs_chunk_readonly(struct btrfs_fs_info *fs_info, u64 chunk_offset)
5056 {
5057         struct extent_map *em;
5058         struct map_lookup *map;
5059         int readonly = 0;
5060         int miss_ndevs = 0;
5061         int i;
5062
5063         em = get_chunk_map(fs_info, chunk_offset, 1);
5064         if (IS_ERR(em))
5065                 return 1;
5066
5067         map = em->map_lookup;
5068         for (i = 0; i < map->num_stripes; i++) {
5069                 if (map->stripes[i].dev->missing) {
5070                         miss_ndevs++;
5071                         continue;
5072                 }
5073
5074                 if (!map->stripes[i].dev->writeable) {
5075                         readonly = 1;
5076                         goto end;
5077                 }
5078         }
5079
5080         /*
5081          * If the number of missing devices is larger than max errors,
5082          * we can not write the data into that chunk successfully, so
5083          * set it readonly.
5084          */
5085         if (miss_ndevs > btrfs_chunk_max_errors(map))
5086                 readonly = 1;
5087 end:
5088         free_extent_map(em);
5089         return readonly;
5090 }
5091
5092 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
5093 {
5094         extent_map_tree_init(&tree->map_tree);
5095 }
5096
5097 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
5098 {
5099         struct extent_map *em;
5100
5101         while (1) {
5102                 write_lock(&tree->map_tree.lock);
5103                 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
5104                 if (em)
5105                         remove_extent_mapping(&tree->map_tree, em);
5106                 write_unlock(&tree->map_tree.lock);
5107                 if (!em)
5108                         break;
5109                 /* once for us */
5110                 free_extent_map(em);
5111                 /* once for the tree */
5112                 free_extent_map(em);
5113         }
5114 }
5115
5116 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5117 {
5118         struct extent_map *em;
5119         struct map_lookup *map;
5120         int ret;
5121
5122         em = get_chunk_map(fs_info, logical, len);
5123         if (IS_ERR(em))
5124                 /*
5125                  * We could return errors for these cases, but that could get
5126                  * ugly and we'd probably do the same thing which is just not do
5127                  * anything else and exit, so return 1 so the callers don't try
5128                  * to use other copies.
5129                  */
5130                 return 1;
5131
5132         map = em->map_lookup;
5133         if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
5134                 ret = map->num_stripes;
5135         else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5136                 ret = map->sub_stripes;
5137         else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
5138                 ret = 2;
5139         else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5140                 ret = 3;
5141         else
5142                 ret = 1;
5143         free_extent_map(em);
5144
5145         btrfs_dev_replace_lock(&fs_info->dev_replace, 0);
5146         if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace) &&
5147             fs_info->dev_replace.tgtdev)
5148                 ret++;
5149         btrfs_dev_replace_unlock(&fs_info->dev_replace, 0);
5150
5151         return ret;
5152 }
5153
5154 unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info,
5155                                     struct btrfs_mapping_tree *map_tree,
5156                                     u64 logical)
5157 {
5158         struct extent_map *em;
5159         struct map_lookup *map;
5160         unsigned long len = fs_info->sectorsize;
5161
5162         em = get_chunk_map(fs_info, logical, len);
5163         WARN_ON(IS_ERR(em));
5164
5165         map = em->map_lookup;
5166         if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5167                 len = map->stripe_len * nr_data_stripes(map);
5168         free_extent_map(em);
5169         return len;
5170 }
5171
5172 int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info,
5173                            u64 logical, u64 len, int mirror_num)
5174 {
5175         struct extent_map *em;
5176         struct map_lookup *map;
5177         int ret = 0;
5178
5179         em = get_chunk_map(fs_info, logical, len);
5180         WARN_ON(IS_ERR(em));
5181
5182         map = em->map_lookup;
5183         if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5184                 ret = 1;
5185         free_extent_map(em);
5186         return ret;
5187 }
5188
5189 static int find_live_mirror(struct btrfs_fs_info *fs_info,
5190                             struct map_lookup *map, int first, int num,
5191                             int optimal, int dev_replace_is_ongoing)
5192 {
5193         int i;
5194         int tolerance;
5195         struct btrfs_device *srcdev;
5196
5197         if (dev_replace_is_ongoing &&
5198             fs_info->dev_replace.cont_reading_from_srcdev_mode ==
5199              BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
5200                 srcdev = fs_info->dev_replace.srcdev;
5201         else
5202                 srcdev = NULL;
5203
5204         /*
5205          * try to avoid the drive that is the source drive for a
5206          * dev-replace procedure, only choose it if no other non-missing
5207          * mirror is available
5208          */
5209         for (tolerance = 0; tolerance < 2; tolerance++) {
5210                 if (map->stripes[optimal].dev->bdev &&
5211                     (tolerance || map->stripes[optimal].dev != srcdev))
5212                         return optimal;
5213                 for (i = first; i < first + num; i++) {
5214                         if (map->stripes[i].dev->bdev &&
5215                             (tolerance || map->stripes[i].dev != srcdev))
5216                                 return i;
5217                 }
5218         }
5219
5220         /* we couldn't find one that doesn't fail.  Just return something
5221          * and the io error handling code will clean up eventually
5222          */
5223         return optimal;
5224 }
5225
5226 static inline int parity_smaller(u64 a, u64 b)
5227 {
5228         return a > b;
5229 }
5230
5231 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
5232 static void sort_parity_stripes(struct btrfs_bio *bbio, int num_stripes)
5233 {
5234         struct btrfs_bio_stripe s;
5235         int i;
5236         u64 l;
5237         int again = 1;
5238
5239         while (again) {
5240                 again = 0;
5241                 for (i = 0; i < num_stripes - 1; i++) {
5242                         if (parity_smaller(bbio->raid_map[i],
5243                                            bbio->raid_map[i+1])) {
5244                                 s = bbio->stripes[i];
5245                                 l = bbio->raid_map[i];
5246                                 bbio->stripes[i] = bbio->stripes[i+1];
5247                                 bbio->raid_map[i] = bbio->raid_map[i+1];
5248                                 bbio->stripes[i+1] = s;
5249                                 bbio->raid_map[i+1] = l;
5250
5251                                 again = 1;
5252                         }
5253                 }
5254         }
5255 }
5256
5257 static struct btrfs_bio *alloc_btrfs_bio(int total_stripes, int real_stripes)
5258 {
5259         struct btrfs_bio *bbio = kzalloc(
5260                  /* the size of the btrfs_bio */
5261                 sizeof(struct btrfs_bio) +
5262                 /* plus the variable array for the stripes */
5263                 sizeof(struct btrfs_bio_stripe) * (total_stripes) +
5264                 /* plus the variable array for the tgt dev */
5265                 sizeof(int) * (real_stripes) +
5266                 /*
5267                  * plus the raid_map, which includes both the tgt dev
5268                  * and the stripes
5269                  */
5270                 sizeof(u64) * (total_stripes),
5271                 GFP_NOFS|__GFP_NOFAIL);
5272
5273         atomic_set(&bbio->error, 0);
5274         refcount_set(&bbio->refs, 1);
5275
5276         return bbio;
5277 }
5278
5279 void btrfs_get_bbio(struct btrfs_bio *bbio)
5280 {
5281         WARN_ON(!refcount_read(&bbio->refs));
5282         refcount_inc(&bbio->refs);
5283 }
5284
5285 void btrfs_put_bbio(struct btrfs_bio *bbio)
5286 {
5287         if (!bbio)
5288                 return;
5289         if (refcount_dec_and_test(&bbio->refs))
5290                 kfree(bbio);
5291 }
5292
5293 /* can REQ_OP_DISCARD be sent with other REQ like REQ_OP_WRITE? */
5294 /*
5295  * Please note that, discard won't be sent to target device of device
5296  * replace.
5297  */
5298 static int __btrfs_map_block_for_discard(struct btrfs_fs_info *fs_info,
5299                                          u64 logical, u64 length,
5300                                          struct btrfs_bio **bbio_ret)
5301 {
5302         struct extent_map *em;
5303         struct map_lookup *map;
5304         struct btrfs_bio *bbio;
5305         u64 offset;
5306         u64 stripe_nr;
5307         u64 stripe_nr_end;
5308         u64 stripe_end_offset;
5309         u64 stripe_cnt;
5310         u64 stripe_len;
5311         u64 stripe_offset;
5312         u64 num_stripes;
5313         u32 stripe_index;
5314         u32 factor = 0;
5315         u32 sub_stripes = 0;
5316         u64 stripes_per_dev = 0;
5317         u32 remaining_stripes = 0;
5318         u32 last_stripe = 0;
5319         int ret = 0;
5320         int i;
5321
5322         /* discard always return a bbio */
5323         ASSERT(bbio_ret);
5324
5325         em = get_chunk_map(fs_info, logical, length);
5326         if (IS_ERR(em))
5327                 return PTR_ERR(em);
5328
5329         map = em->map_lookup;
5330         /* we don't discard raid56 yet */
5331         if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5332                 ret = -EOPNOTSUPP;
5333                 goto out;
5334         }
5335
5336         offset = logical - em->start;
5337         length = min_t(u64, em->len - offset, length);
5338
5339         stripe_len = map->stripe_len;
5340         /*
5341          * stripe_nr counts the total number of stripes we have to stride
5342          * to get to this block
5343          */
5344         stripe_nr = div64_u64(offset, stripe_len);
5345
5346         /* stripe_offset is the offset of this block in its stripe */
5347         stripe_offset = offset - stripe_nr * stripe_len;
5348
5349         stripe_nr_end = round_up(offset + length, map->stripe_len);
5350         stripe_nr_end = div64_u64(stripe_nr_end, map->stripe_len);
5351         stripe_cnt = stripe_nr_end - stripe_nr;
5352         stripe_end_offset = stripe_nr_end * map->stripe_len -
5353                             (offset + length);
5354         /*
5355          * after this, stripe_nr is the number of stripes on this
5356          * device we have to walk to find the data, and stripe_index is
5357          * the number of our device in the stripe array
5358          */
5359         num_stripes = 1;
5360         stripe_index = 0;
5361         if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5362                          BTRFS_BLOCK_GROUP_RAID10)) {
5363                 if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5364                         sub_stripes = 1;
5365                 else
5366                         sub_stripes = map->sub_stripes;
5367
5368                 factor = map->num_stripes / sub_stripes;
5369                 num_stripes = min_t(u64, map->num_stripes,
5370                                     sub_stripes * stripe_cnt);
5371                 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
5372                 stripe_index *= sub_stripes;
5373                 stripes_per_dev = div_u64_rem(stripe_cnt, factor,
5374                                               &remaining_stripes);
5375                 div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
5376                 last_stripe *= sub_stripes;
5377         } else if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
5378                                 BTRFS_BLOCK_GROUP_DUP)) {
5379                 num_stripes = map->num_stripes;
5380         } else {
5381                 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5382                                         &stripe_index);
5383         }
5384
5385         bbio = alloc_btrfs_bio(num_stripes, 0);
5386         if (!bbio) {
5387                 ret = -ENOMEM;
5388                 goto out;
5389         }
5390
5391         for (i = 0; i < num_stripes; i++) {
5392                 bbio->stripes[i].physical =
5393                         map->stripes[stripe_index].physical +
5394                         stripe_offset + stripe_nr * map->stripe_len;
5395                 bbio->stripes[i].dev = map->stripes[stripe_index].dev;
5396
5397                 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5398                                  BTRFS_BLOCK_GROUP_RAID10)) {
5399                         bbio->stripes[i].length = stripes_per_dev *
5400                                 map->stripe_len;
5401
5402                         if (i / sub_stripes < remaining_stripes)
5403                                 bbio->stripes[i].length +=
5404                                         map->stripe_len;
5405
5406                         /*
5407                          * Special for the first stripe and
5408                          * the last stripe:
5409                          *
5410                          * |-------|...|-------|
5411                          *     |----------|
5412                          *    off     end_off
5413                          */
5414                         if (i < sub_stripes)
5415                                 bbio->stripes[i].length -=
5416                                         stripe_offset;
5417
5418                         if (stripe_index >= last_stripe &&
5419                             stripe_index <= (last_stripe +
5420                                              sub_stripes - 1))
5421                                 bbio->stripes[i].length -=
5422                                         stripe_end_offset;
5423
5424                         if (i == sub_stripes - 1)
5425                                 stripe_offset = 0;
5426                 } else {
5427                         bbio->stripes[i].length = length;
5428                 }
5429
5430                 stripe_index++;
5431                 if (stripe_index == map->num_stripes) {
5432                         stripe_index = 0;
5433                         stripe_nr++;
5434                 }
5435         }
5436
5437         *bbio_ret = bbio;
5438         bbio->map_type = map->type;
5439         bbio->num_stripes = num_stripes;
5440 out:
5441         free_extent_map(em);
5442         return ret;
5443 }
5444
5445 /*
5446  * In dev-replace case, for repair case (that's the only case where the mirror
5447  * is selected explicitly when calling btrfs_map_block), blocks left of the
5448  * left cursor can also be read from the target drive.
5449  *
5450  * For REQ_GET_READ_MIRRORS, the target drive is added as the last one to the
5451  * array of stripes.
5452  * For READ, it also needs to be supported using the same mirror number.
5453  *
5454  * If the requested block is not left of the left cursor, EIO is returned. This
5455  * can happen because btrfs_num_copies() returns one more in the dev-replace
5456  * case.
5457  */
5458 static int get_extra_mirror_from_replace(struct btrfs_fs_info *fs_info,
5459                                          u64 logical, u64 length,
5460                                          u64 srcdev_devid, int *mirror_num,
5461                                          u64 *physical)
5462 {
5463         struct btrfs_bio *bbio = NULL;
5464         int num_stripes;
5465         int index_srcdev = 0;
5466         int found = 0;
5467         u64 physical_of_found = 0;
5468         int i;
5469         int ret = 0;
5470
5471         ret = __btrfs_map_block(fs_info, BTRFS_MAP_GET_READ_MIRRORS,
5472                                 logical, &length, &bbio, 0, 0);
5473         if (ret) {
5474                 ASSERT(bbio == NULL);
5475                 return ret;
5476         }
5477
5478         num_stripes = bbio->num_stripes;
5479         if (*mirror_num > num_stripes) {
5480                 /*
5481                  * BTRFS_MAP_GET_READ_MIRRORS does not contain this mirror,
5482                  * that means that the requested area is not left of the left
5483                  * cursor
5484                  */
5485                 btrfs_put_bbio(bbio);
5486                 return -EIO;
5487         }
5488
5489         /*
5490          * process the rest of the function using the mirror_num of the source
5491          * drive. Therefore look it up first.  At the end, patch the device
5492          * pointer to the one of the target drive.
5493          */
5494         for (i = 0; i < num_stripes; i++) {
5495                 if (bbio->stripes[i].dev->devid != srcdev_devid)
5496                         continue;
5497
5498                 /*
5499                  * In case of DUP, in order to keep it simple, only add the
5500                  * mirror with the lowest physical address
5501                  */
5502                 if (found &&
5503                     physical_of_found <= bbio->stripes[i].physical)
5504                         continue;
5505
5506                 index_srcdev = i;
5507                 found = 1;
5508                 physical_of_found = bbio->stripes[i].physical;
5509         }
5510
5511         btrfs_put_bbio(bbio);
5512
5513         ASSERT(found);
5514         if (!found)
5515                 return -EIO;
5516
5517         *mirror_num = index_srcdev + 1;
5518         *physical = physical_of_found;
5519         return ret;
5520 }
5521
5522 static void handle_ops_on_dev_replace(enum btrfs_map_op op,
5523                                       struct btrfs_bio **bbio_ret,
5524                                       struct btrfs_dev_replace *dev_replace,
5525                                       int *num_stripes_ret, int *max_errors_ret)
5526 {
5527         struct btrfs_bio *bbio = *bbio_ret;
5528         u64 srcdev_devid = dev_replace->srcdev->devid;
5529         int tgtdev_indexes = 0;
5530         int num_stripes = *num_stripes_ret;
5531         int max_errors = *max_errors_ret;
5532         int i;
5533
5534         if (op == BTRFS_MAP_WRITE) {
5535                 int index_where_to_add;
5536
5537                 /*
5538                  * duplicate the write operations while the dev replace
5539                  * procedure is running. Since the copying of the old disk to
5540                  * the new disk takes place at run time while the filesystem is
5541                  * mounted writable, the regular write operations to the old
5542                  * disk have to be duplicated to go to the new disk as well.
5543                  *
5544                  * Note that device->missing is handled by the caller, and that
5545                  * the write to the old disk is already set up in the stripes
5546                  * array.
5547                  */
5548                 index_where_to_add = num_stripes;
5549                 for (i = 0; i < num_stripes; i++) {
5550                         if (bbio->stripes[i].dev->devid == srcdev_devid) {
5551                                 /* write to new disk, too */
5552                                 struct btrfs_bio_stripe *new =
5553                                         bbio->stripes + index_where_to_add;
5554                                 struct btrfs_bio_stripe *old =
5555                                         bbio->stripes + i;
5556
5557                                 new->physical = old->physical;
5558                                 new->length = old->length;
5559                                 new->dev = dev_replace->tgtdev;
5560                                 bbio->tgtdev_map[i] = index_where_to_add;
5561                                 index_where_to_add++;
5562                                 max_errors++;
5563                                 tgtdev_indexes++;
5564                         }
5565                 }
5566                 num_stripes = index_where_to_add;
5567         } else if (op == BTRFS_MAP_GET_READ_MIRRORS) {
5568                 int index_srcdev = 0;
5569                 int found = 0;
5570                 u64 physical_of_found = 0;
5571
5572                 /*
5573                  * During the dev-replace procedure, the target drive can also
5574                  * be used to read data in case it is needed to repair a corrupt
5575                  * block elsewhere. This is possible if the requested area is
5576                  * left of the left cursor. In this area, the target drive is a
5577                  * full copy of the source drive.
5578                  */
5579                 for (i = 0; i < num_stripes; i++) {
5580                         if (bbio->stripes[i].dev->devid == srcdev_devid) {
5581                                 /*
5582                                  * In case of DUP, in order to keep it simple,
5583                                  * only add the mirror with the lowest physical
5584                                  * address
5585                                  */
5586                                 if (found &&
5587                                     physical_of_found <=
5588                                      bbio->stripes[i].physical)
5589                                         continue;
5590                                 index_srcdev = i;
5591                                 found = 1;
5592                                 physical_of_found = bbio->stripes[i].physical;
5593                         }
5594                 }
5595                 if (found) {
5596                         struct btrfs_bio_stripe *tgtdev_stripe =
5597                                 bbio->stripes + num_stripes;
5598
5599                         tgtdev_stripe->physical = physical_of_found;
5600                         tgtdev_stripe->length =
5601                                 bbio->stripes[index_srcdev].length;
5602                         tgtdev_stripe->dev = dev_replace->tgtdev;
5603                         bbio->tgtdev_map[index_srcdev] = num_stripes;
5604
5605                         tgtdev_indexes++;
5606                         num_stripes++;
5607                 }
5608         }
5609
5610         *num_stripes_ret = num_stripes;
5611         *max_errors_ret = max_errors;
5612         bbio->num_tgtdevs = tgtdev_indexes;
5613         *bbio_ret = bbio;
5614 }
5615
5616 static bool need_full_stripe(enum btrfs_map_op op)
5617 {
5618         return (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS);
5619 }
5620
5621 static int __btrfs_map_block(struct btrfs_fs_info *fs_info,
5622                              enum btrfs_map_op op,
5623                              u64 logical, u64 *length,
5624                              struct btrfs_bio **bbio_ret,
5625                              int mirror_num, int need_raid_map)
5626 {
5627         struct extent_map *em;
5628         struct map_lookup *map;
5629         u64 offset;
5630         u64 stripe_offset;
5631         u64 stripe_nr;
5632         u64 stripe_len;
5633         u32 stripe_index;
5634         int i;
5635         int ret = 0;
5636         int num_stripes;
5637         int max_errors = 0;
5638         int tgtdev_indexes = 0;
5639         struct btrfs_bio *bbio = NULL;
5640         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
5641         int dev_replace_is_ongoing = 0;
5642         int num_alloc_stripes;
5643         int patch_the_first_stripe_for_dev_replace = 0;
5644         u64 physical_to_patch_in_first_stripe = 0;
5645         u64 raid56_full_stripe_start = (u64)-1;
5646
5647         if (op == BTRFS_MAP_DISCARD)
5648                 return __btrfs_map_block_for_discard(fs_info, logical,
5649                                                      *length, bbio_ret);
5650
5651         em = get_chunk_map(fs_info, logical, *length);
5652         if (IS_ERR(em))
5653                 return PTR_ERR(em);
5654
5655         map = em->map_lookup;
5656         offset = logical - em->start;
5657
5658         stripe_len = map->stripe_len;
5659         stripe_nr = offset;
5660         /*
5661          * stripe_nr counts the total number of stripes we have to stride
5662          * to get to this block
5663          */
5664         stripe_nr = div64_u64(stripe_nr, stripe_len);
5665
5666         stripe_offset = stripe_nr * stripe_len;
5667         if (offset < stripe_offset) {
5668                 btrfs_crit(fs_info,
5669                            "stripe math has gone wrong, stripe_offset=%llu, offset=%llu, start=%llu, logical=%llu, stripe_len=%llu",
5670                            stripe_offset, offset, em->start, logical,
5671                            stripe_len);
5672                 free_extent_map(em);
5673                 return -EINVAL;
5674         }
5675
5676         /* stripe_offset is the offset of this block in its stripe*/
5677         stripe_offset = offset - stripe_offset;
5678
5679         /* if we're here for raid56, we need to know the stripe aligned start */
5680         if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5681                 unsigned long full_stripe_len = stripe_len * nr_data_stripes(map);
5682                 raid56_full_stripe_start = offset;
5683
5684                 /* allow a write of a full stripe, but make sure we don't
5685                  * allow straddling of stripes
5686                  */
5687                 raid56_full_stripe_start = div64_u64(raid56_full_stripe_start,
5688                                 full_stripe_len);
5689                 raid56_full_stripe_start *= full_stripe_len;
5690         }
5691
5692         if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
5693                 u64 max_len;
5694                 /* For writes to RAID[56], allow a full stripeset across all disks.
5695                    For other RAID types and for RAID[56] reads, just allow a single
5696                    stripe (on a single disk). */
5697                 if ((map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
5698                     (op == BTRFS_MAP_WRITE)) {
5699                         max_len = stripe_len * nr_data_stripes(map) -
5700                                 (offset - raid56_full_stripe_start);
5701                 } else {
5702                         /* we limit the length of each bio to what fits in a stripe */
5703                         max_len = stripe_len - stripe_offset;
5704                 }
5705                 *length = min_t(u64, em->len - offset, max_len);
5706         } else {
5707                 *length = em->len - offset;
5708         }
5709
5710         /* This is for when we're called from btrfs_merge_bio_hook() and all
5711            it cares about is the length */
5712         if (!bbio_ret)
5713                 goto out;
5714
5715         btrfs_dev_replace_lock(dev_replace, 0);
5716         dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
5717         if (!dev_replace_is_ongoing)
5718                 btrfs_dev_replace_unlock(dev_replace, 0);
5719         else
5720                 btrfs_dev_replace_set_lock_blocking(dev_replace);
5721
5722         if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
5723             !need_full_stripe(op) && dev_replace->tgtdev != NULL) {
5724                 ret = get_extra_mirror_from_replace(fs_info, logical, *length,
5725                                                     dev_replace->srcdev->devid,
5726                                                     &mirror_num,
5727                                             &physical_to_patch_in_first_stripe);
5728                 if (ret)
5729                         goto out;
5730                 else
5731                         patch_the_first_stripe_for_dev_replace = 1;
5732         } else if (mirror_num > map->num_stripes) {
5733                 mirror_num = 0;
5734         }
5735
5736         num_stripes = 1;
5737         stripe_index = 0;
5738         if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5739                 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5740                                 &stripe_index);
5741                 if (op != BTRFS_MAP_WRITE && op != BTRFS_MAP_GET_READ_MIRRORS)
5742                         mirror_num = 1;
5743         } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
5744                 if (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS)
5745                         num_stripes = map->num_stripes;
5746                 else if (mirror_num)
5747                         stripe_index = mirror_num - 1;
5748                 else {
5749                         stripe_index = find_live_mirror(fs_info, map, 0,
5750                                             map->num_stripes,
5751                                             current->pid % map->num_stripes,
5752                                             dev_replace_is_ongoing);
5753                         mirror_num = stripe_index + 1;
5754                 }
5755
5756         } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
5757                 if (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS) {
5758                         num_stripes = map->num_stripes;
5759                 } else if (mirror_num) {
5760                         stripe_index = mirror_num - 1;
5761                 } else {
5762                         mirror_num = 1;
5763                 }
5764
5765         } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5766                 u32 factor = map->num_stripes / map->sub_stripes;
5767
5768                 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
5769                 stripe_index *= map->sub_stripes;
5770
5771                 if (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS)
5772                         num_stripes = map->sub_stripes;
5773                 else if (mirror_num)
5774                         stripe_index += mirror_num - 1;
5775                 else {
5776                         int old_stripe_index = stripe_index;
5777                         stripe_index = find_live_mirror(fs_info, map,
5778                                               stripe_index,
5779                                               map->sub_stripes, stripe_index +
5780                                               current->pid % map->sub_stripes,
5781                                               dev_replace_is_ongoing);
5782                         mirror_num = stripe_index - old_stripe_index + 1;
5783                 }
5784
5785         } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5786                 if (need_raid_map &&
5787                     (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS ||
5788                      mirror_num > 1)) {
5789                         /* push stripe_nr back to the start of the full stripe */
5790                         stripe_nr = div64_u64(raid56_full_stripe_start,
5791                                         stripe_len * nr_data_stripes(map));
5792
5793                         /* RAID[56] write or recovery. Return all stripes */
5794                         num_stripes = map->num_stripes;
5795                         max_errors = nr_parity_stripes(map);
5796
5797                         *length = map->stripe_len;
5798                         stripe_index = 0;
5799                         stripe_offset = 0;
5800                 } else {
5801                         /*
5802                          * Mirror #0 or #1 means the original data block.
5803                          * Mirror #2 is RAID5 parity block.
5804                          * Mirror #3 is RAID6 Q block.
5805                          */
5806                         stripe_nr = div_u64_rem(stripe_nr,
5807                                         nr_data_stripes(map), &stripe_index);
5808                         if (mirror_num > 1)
5809                                 stripe_index = nr_data_stripes(map) +
5810                                                 mirror_num - 2;
5811
5812                         /* We distribute the parity blocks across stripes */
5813                         div_u64_rem(stripe_nr + stripe_index, map->num_stripes,
5814                                         &stripe_index);
5815                         if ((op != BTRFS_MAP_WRITE &&
5816                              op != BTRFS_MAP_GET_READ_MIRRORS) &&
5817                             mirror_num <= 1)
5818                                 mirror_num = 1;
5819                 }
5820         } else {
5821                 /*
5822                  * after this, stripe_nr is the number of stripes on this
5823                  * device we have to walk to find the data, and stripe_index is
5824                  * the number of our device in the stripe array
5825                  */
5826                 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5827                                 &stripe_index);
5828                 mirror_num = stripe_index + 1;
5829         }
5830         if (stripe_index >= map->num_stripes) {
5831                 btrfs_crit(fs_info,
5832                            "stripe index math went horribly wrong, got stripe_index=%u, num_stripes=%u",
5833                            stripe_index, map->num_stripes);
5834                 ret = -EINVAL;
5835                 goto out;
5836         }
5837
5838         num_alloc_stripes = num_stripes;
5839         if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL) {
5840                 if (op == BTRFS_MAP_WRITE)
5841                         num_alloc_stripes <<= 1;
5842                 if (op == BTRFS_MAP_GET_READ_MIRRORS)
5843                         num_alloc_stripes++;
5844                 tgtdev_indexes = num_stripes;
5845         }
5846
5847         bbio = alloc_btrfs_bio(num_alloc_stripes, tgtdev_indexes);
5848         if (!bbio) {
5849                 ret = -ENOMEM;
5850                 goto out;
5851         }
5852         if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL)
5853                 bbio->tgtdev_map = (int *)(bbio->stripes + num_alloc_stripes);
5854
5855         /* build raid_map */
5856         if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK && need_raid_map &&
5857             (need_full_stripe(op) || mirror_num > 1)) {
5858                 u64 tmp;
5859                 unsigned rot;
5860
5861                 bbio->raid_map = (u64 *)((void *)bbio->stripes +
5862                                  sizeof(struct btrfs_bio_stripe) *
5863                                  num_alloc_stripes +
5864                                  sizeof(int) * tgtdev_indexes);
5865
5866                 /* Work out the disk rotation on this stripe-set */
5867                 div_u64_rem(stripe_nr, num_stripes, &rot);
5868
5869                 /* Fill in the logical address of each stripe */
5870                 tmp = stripe_nr * nr_data_stripes(map);
5871                 for (i = 0; i < nr_data_stripes(map); i++)
5872                         bbio->raid_map[(i+rot) % num_stripes] =
5873                                 em->start + (tmp + i) * map->stripe_len;
5874
5875                 bbio->raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE;
5876                 if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5877                         bbio->raid_map[(i+rot+1) % num_stripes] =
5878                                 RAID6_Q_STRIPE;
5879         }
5880
5881
5882         for (i = 0; i < num_stripes; i++) {
5883                 bbio->stripes[i].physical =
5884                         map->stripes[stripe_index].physical +
5885                         stripe_offset +
5886                         stripe_nr * map->stripe_len;
5887                 bbio->stripes[i].dev =
5888                         map->stripes[stripe_index].dev;
5889                 stripe_index++;
5890         }
5891
5892         if (need_full_stripe(op))
5893                 max_errors = btrfs_chunk_max_errors(map);
5894
5895         if (bbio->raid_map)
5896                 sort_parity_stripes(bbio, num_stripes);
5897
5898         if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL &&
5899             need_full_stripe(op)) {
5900                 handle_ops_on_dev_replace(op, &bbio, dev_replace, &num_stripes,
5901                                           &max_errors);
5902         }
5903
5904         *bbio_ret = bbio;
5905         bbio->map_type = map->type;
5906         bbio->num_stripes = num_stripes;
5907         bbio->max_errors = max_errors;
5908         bbio->mirror_num = mirror_num;
5909
5910         /*
5911          * this is the case that REQ_READ && dev_replace_is_ongoing &&
5912          * mirror_num == num_stripes + 1 && dev_replace target drive is
5913          * available as a mirror
5914          */
5915         if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) {
5916                 WARN_ON(num_stripes > 1);
5917                 bbio->stripes[0].dev = dev_replace->tgtdev;
5918                 bbio->stripes[0].physical = physical_to_patch_in_first_stripe;
5919                 bbio->mirror_num = map->num_stripes + 1;
5920         }
5921 out:
5922         if (dev_replace_is_ongoing) {
5923                 btrfs_dev_replace_clear_lock_blocking(dev_replace);
5924                 btrfs_dev_replace_unlock(dev_replace, 0);
5925         }
5926         free_extent_map(em);
5927         return ret;
5928 }
5929
5930 int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
5931                       u64 logical, u64 *length,
5932                       struct btrfs_bio **bbio_ret, int mirror_num)
5933 {
5934         return __btrfs_map_block(fs_info, op, logical, length, bbio_ret,
5935                                  mirror_num, 0);
5936 }
5937
5938 /* For Scrub/replace */
5939 int btrfs_map_sblock(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
5940                      u64 logical, u64 *length,
5941                      struct btrfs_bio **bbio_ret)
5942 {
5943         return __btrfs_map_block(fs_info, op, logical, length, bbio_ret, 0, 1);
5944 }
5945
5946 int btrfs_rmap_block(struct btrfs_fs_info *fs_info,
5947                      u64 chunk_start, u64 physical, u64 devid,
5948                      u64 **logical, int *naddrs, int *stripe_len)
5949 {
5950         struct extent_map *em;
5951         struct map_lookup *map;
5952         u64 *buf;
5953         u64 bytenr;
5954         u64 length;
5955         u64 stripe_nr;
5956         u64 rmap_len;
5957         int i, j, nr = 0;
5958
5959         em = get_chunk_map(fs_info, chunk_start, 1);
5960         if (IS_ERR(em))
5961                 return -EIO;
5962
5963         map = em->map_lookup;
5964         length = em->len;
5965         rmap_len = map->stripe_len;
5966
5967         if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5968                 length = div_u64(length, map->num_stripes / map->sub_stripes);
5969         else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5970                 length = div_u64(length, map->num_stripes);
5971         else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5972                 length = div_u64(length, nr_data_stripes(map));
5973                 rmap_len = map->stripe_len * nr_data_stripes(map);
5974         }
5975
5976         buf = kcalloc(map->num_stripes, sizeof(u64), GFP_NOFS);
5977         BUG_ON(!buf); /* -ENOMEM */
5978
5979         for (i = 0; i < map->num_stripes; i++) {
5980                 if (devid && map->stripes[i].dev->devid != devid)
5981                         continue;
5982                 if (map->stripes[i].physical > physical ||
5983                     map->stripes[i].physical + length <= physical)
5984                         continue;
5985
5986                 stripe_nr = physical - map->stripes[i].physical;
5987                 stripe_nr = div64_u64(stripe_nr, map->stripe_len);
5988
5989                 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5990                         stripe_nr = stripe_nr * map->num_stripes + i;
5991                         stripe_nr = div_u64(stripe_nr, map->sub_stripes);
5992                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5993                         stripe_nr = stripe_nr * map->num_stripes + i;
5994                 } /* else if RAID[56], multiply by nr_data_stripes().
5995                    * Alternatively, just use rmap_len below instead of
5996                    * map->stripe_len */
5997
5998                 bytenr = chunk_start + stripe_nr * rmap_len;
5999                 WARN_ON(nr >= map->num_stripes);
6000                 for (j = 0; j < nr; j++) {
6001                         if (buf[j] == bytenr)
6002                                 break;
6003                 }
6004                 if (j == nr) {
6005                         WARN_ON(nr >= map->num_stripes);
6006                         buf[nr++] = bytenr;
6007                 }
6008         }
6009
6010         *logical = buf;
6011         *naddrs = nr;
6012         *stripe_len = rmap_len;
6013
6014         free_extent_map(em);
6015         return 0;
6016 }
6017
6018 static inline void btrfs_end_bbio(struct btrfs_bio *bbio, struct bio *bio)
6019 {
6020         bio->bi_private = bbio->private;
6021         bio->bi_end_io = bbio->end_io;
6022         bio_endio(bio);
6023
6024         btrfs_put_bbio(bbio);
6025 }
6026
6027 static void btrfs_end_bio(struct bio *bio)
6028 {
6029         struct btrfs_bio *bbio = bio->bi_private;
6030         int is_orig_bio = 0;
6031
6032         if (bio->bi_error) {
6033                 atomic_inc(&bbio->error);
6034                 if (bio->bi_error == -EIO || bio->bi_error == -EREMOTEIO) {
6035                         unsigned int stripe_index =
6036                                 btrfs_io_bio(bio)->stripe_index;
6037                         struct btrfs_device *dev;
6038
6039                         BUG_ON(stripe_index >= bbio->num_stripes);
6040                         dev = bbio->stripes[stripe_index].dev;
6041                         if (dev->bdev) {
6042                                 if (bio_op(bio) == REQ_OP_WRITE)
6043                                         btrfs_dev_stat_inc(dev,
6044                                                 BTRFS_DEV_STAT_WRITE_ERRS);
6045                                 else
6046                                         btrfs_dev_stat_inc(dev,
6047                                                 BTRFS_DEV_STAT_READ_ERRS);
6048                                 if (bio->bi_opf & REQ_PREFLUSH)
6049                                         btrfs_dev_stat_inc(dev,
6050                                                 BTRFS_DEV_STAT_FLUSH_ERRS);
6051                                 btrfs_dev_stat_print_on_error(dev);
6052                         }
6053                 }
6054         }
6055
6056         if (bio == bbio->orig_bio)
6057                 is_orig_bio = 1;
6058
6059         btrfs_bio_counter_dec(bbio->fs_info);
6060
6061         if (atomic_dec_and_test(&bbio->stripes_pending)) {
6062                 if (!is_orig_bio) {
6063                         bio_put(bio);
6064                         bio = bbio->orig_bio;
6065                 }
6066
6067                 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
6068                 /* only send an error to the higher layers if it is
6069                  * beyond the tolerance of the btrfs bio
6070                  */
6071                 if (atomic_read(&bbio->error) > bbio->max_errors) {
6072                         bio->bi_error = -EIO;
6073                 } else {
6074                         /*
6075                          * this bio is actually up to date, we didn't
6076                          * go over the max number of errors
6077                          */
6078                         bio->bi_error = 0;
6079                 }
6080
6081                 btrfs_end_bbio(bbio, bio);
6082         } else if (!is_orig_bio) {
6083                 bio_put(bio);
6084         }
6085 }
6086
6087 /*
6088  * see run_scheduled_bios for a description of why bios are collected for
6089  * async submit.
6090  *
6091  * This will add one bio to the pending list for a device and make sure
6092  * the work struct is scheduled.
6093  */
6094 static noinline void btrfs_schedule_bio(struct btrfs_device *device,
6095                                         struct bio *bio)
6096 {
6097         struct btrfs_fs_info *fs_info = device->fs_info;
6098         int should_queue = 1;
6099         struct btrfs_pending_bios *pending_bios;
6100
6101         if (device->missing || !device->bdev) {
6102                 bio_io_error(bio);
6103                 return;
6104         }
6105
6106         /* don't bother with additional async steps for reads, right now */
6107         if (bio_op(bio) == REQ_OP_READ) {
6108                 bio_get(bio);
6109                 btrfsic_submit_bio(bio);
6110                 bio_put(bio);
6111                 return;
6112         }
6113
6114         /*
6115          * nr_async_bios allows us to reliably return congestion to the
6116          * higher layers.  Otherwise, the async bio makes it appear we have
6117          * made progress against dirty pages when we've really just put it
6118          * on a queue for later
6119          */
6120         atomic_inc(&fs_info->nr_async_bios);
6121         WARN_ON(bio->bi_next);
6122         bio->bi_next = NULL;
6123
6124         spin_lock(&device->io_lock);
6125         if (op_is_sync(bio->bi_opf))
6126                 pending_bios = &device->pending_sync_bios;
6127         else
6128                 pending_bios = &device->pending_bios;
6129
6130         if (pending_bios->tail)
6131                 pending_bios->tail->bi_next = bio;
6132
6133         pending_bios->tail = bio;
6134         if (!pending_bios->head)
6135                 pending_bios->head = bio;
6136         if (device->running_pending)
6137                 should_queue = 0;
6138
6139         spin_unlock(&device->io_lock);
6140
6141         if (should_queue)
6142                 btrfs_queue_work(fs_info->submit_workers, &device->work);
6143 }
6144
6145 static void submit_stripe_bio(struct btrfs_bio *bbio, struct bio *bio,
6146                               u64 physical, int dev_nr, int async)
6147 {
6148         struct btrfs_device *dev = bbio->stripes[dev_nr].dev;
6149         struct btrfs_fs_info *fs_info = bbio->fs_info;
6150
6151         bio->bi_private = bbio;
6152         btrfs_io_bio(bio)->stripe_index = dev_nr;
6153         bio->bi_end_io = btrfs_end_bio;
6154         bio->bi_iter.bi_sector = physical >> 9;
6155 #ifdef DEBUG
6156         {
6157                 struct rcu_string *name;
6158
6159                 rcu_read_lock();
6160                 name = rcu_dereference(dev->name);
6161                 btrfs_debug(fs_info,
6162                         "btrfs_map_bio: rw %d 0x%x, sector=%llu, dev=%lu (%s id %llu), size=%u",
6163                         bio_op(bio), bio->bi_opf,
6164                         (u64)bio->bi_iter.bi_sector,
6165                         (u_long)dev->bdev->bd_dev, name->str, dev->devid,
6166                         bio->bi_iter.bi_size);
6167                 rcu_read_unlock();
6168         }
6169 #endif
6170         bio->bi_bdev = dev->bdev;
6171
6172         btrfs_bio_counter_inc_noblocked(fs_info);
6173
6174         if (async)
6175                 btrfs_schedule_bio(dev, bio);
6176         else
6177                 btrfsic_submit_bio(bio);
6178 }
6179
6180 static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical)
6181 {
6182         atomic_inc(&bbio->error);
6183         if (atomic_dec_and_test(&bbio->stripes_pending)) {
6184                 /* Should be the original bio. */
6185                 WARN_ON(bio != bbio->orig_bio);
6186
6187                 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
6188                 bio->bi_iter.bi_sector = logical >> 9;
6189                 bio->bi_error = -EIO;
6190                 btrfs_end_bbio(bbio, bio);
6191         }
6192 }
6193
6194 int btrfs_map_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
6195                   int mirror_num, int async_submit)
6196 {
6197         struct btrfs_device *dev;
6198         struct bio *first_bio = bio;
6199         u64 logical = (u64)bio->bi_iter.bi_sector << 9;
6200         u64 length = 0;
6201         u64 map_length;
6202         int ret;
6203         int dev_nr;
6204         int total_devs;
6205         struct btrfs_bio *bbio = NULL;
6206
6207         length = bio->bi_iter.bi_size;
6208         map_length = length;
6209
6210         btrfs_bio_counter_inc_blocked(fs_info);
6211         ret = __btrfs_map_block(fs_info, bio_op(bio), logical,
6212                                 &map_length, &bbio, mirror_num, 1);
6213         if (ret) {
6214                 btrfs_bio_counter_dec(fs_info);
6215                 return ret;
6216         }
6217
6218         total_devs = bbio->num_stripes;
6219         bbio->orig_bio = first_bio;
6220         bbio->private = first_bio->bi_private;
6221         bbio->end_io = first_bio->bi_end_io;
6222         bbio->fs_info = fs_info;
6223         atomic_set(&bbio->stripes_pending, bbio->num_stripes);
6224
6225         if ((bbio->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
6226             ((bio_op(bio) == REQ_OP_WRITE) || (mirror_num > 1))) {
6227                 /* In this case, map_length has been set to the length of
6228                    a single stripe; not the whole write */
6229                 if (bio_op(bio) == REQ_OP_WRITE) {
6230                         ret = raid56_parity_write(fs_info, bio, bbio,
6231                                                   map_length);
6232                 } else {
6233                         ret = raid56_parity_recover(fs_info, bio, bbio,
6234                                                     map_length, mirror_num, 1);
6235                 }
6236
6237                 btrfs_bio_counter_dec(fs_info);
6238                 return ret;
6239         }
6240
6241         if (map_length < length) {
6242                 btrfs_crit(fs_info,
6243                            "mapping failed logical %llu bio len %llu len %llu",
6244                            logical, length, map_length);
6245                 BUG();
6246         }
6247
6248         for (dev_nr = 0; dev_nr < total_devs; dev_nr++) {
6249                 dev = bbio->stripes[dev_nr].dev;
6250                 if (!dev || !dev->bdev ||
6251                     (bio_op(first_bio) == REQ_OP_WRITE && !dev->writeable)) {
6252                         bbio_error(bbio, first_bio, logical);
6253                         continue;
6254                 }
6255
6256                 if (dev_nr < total_devs - 1)
6257                         bio = btrfs_bio_clone(first_bio);
6258                 else
6259                         bio = first_bio;
6260
6261                 submit_stripe_bio(bbio, bio, bbio->stripes[dev_nr].physical,
6262                                   dev_nr, async_submit);
6263         }
6264         btrfs_bio_counter_dec(fs_info);
6265         return 0;
6266 }
6267
6268 struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid,
6269                                        u8 *uuid, u8 *fsid)
6270 {
6271         struct btrfs_device *device;
6272         struct btrfs_fs_devices *cur_devices;
6273
6274         cur_devices = fs_info->fs_devices;
6275         while (cur_devices) {
6276                 if (!fsid ||
6277                     !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
6278                         device = __find_device(&cur_devices->devices,
6279                                                devid, uuid);
6280                         if (device)
6281                                 return device;
6282                 }
6283                 cur_devices = cur_devices->seed;
6284         }
6285         return NULL;
6286 }
6287
6288 static struct btrfs_device *add_missing_dev(struct btrfs_fs_devices *fs_devices,
6289                                             u64 devid, u8 *dev_uuid)
6290 {
6291         struct btrfs_device *device;
6292
6293         device = btrfs_alloc_device(NULL, &devid, dev_uuid);
6294         if (IS_ERR(device))
6295                 return NULL;
6296
6297         list_add(&device->dev_list, &fs_devices->devices);
6298         device->fs_devices = fs_devices;
6299         fs_devices->num_devices++;
6300
6301         device->missing = 1;
6302         fs_devices->missing_devices++;
6303
6304         return device;
6305 }
6306
6307 /**
6308  * btrfs_alloc_device - allocate struct btrfs_device
6309  * @fs_info:    used only for generating a new devid, can be NULL if
6310  *              devid is provided (i.e. @devid != NULL).
6311  * @devid:      a pointer to devid for this device.  If NULL a new devid
6312  *              is generated.
6313  * @uuid:       a pointer to UUID for this device.  If NULL a new UUID
6314  *              is generated.
6315  *
6316  * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
6317  * on error.  Returned struct is not linked onto any lists and can be
6318  * destroyed with kfree() right away.
6319  */
6320 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
6321                                         const u64 *devid,
6322                                         const u8 *uuid)
6323 {
6324         struct btrfs_device *dev;
6325         u64 tmp;
6326
6327         if (WARN_ON(!devid && !fs_info))
6328                 return ERR_PTR(-EINVAL);
6329
6330         dev = __alloc_device();
6331         if (IS_ERR(dev))
6332                 return dev;
6333
6334         if (devid)
6335                 tmp = *devid;
6336         else {
6337                 int ret;
6338
6339                 ret = find_next_devid(fs_info, &tmp);
6340                 if (ret) {
6341                         kfree(dev);
6342                         return ERR_PTR(ret);
6343                 }
6344         }
6345         dev->devid = tmp;
6346
6347         if (uuid)
6348                 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE);
6349         else
6350                 generate_random_uuid(dev->uuid);
6351
6352         btrfs_init_work(&dev->work, btrfs_submit_helper,
6353                         pending_bios_fn, NULL, NULL);
6354
6355         return dev;
6356 }
6357
6358 /* Return -EIO if any error, otherwise return 0. */
6359 static int btrfs_check_chunk_valid(struct btrfs_fs_info *fs_info,
6360                                    struct extent_buffer *leaf,
6361                                    struct btrfs_chunk *chunk, u64 logical)
6362 {
6363         u64 length;
6364         u64 stripe_len;
6365         u16 num_stripes;
6366         u16 sub_stripes;
6367         u64 type;
6368
6369         length = btrfs_chunk_length(leaf, chunk);
6370         stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6371         num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6372         sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
6373         type = btrfs_chunk_type(leaf, chunk);
6374
6375         if (!num_stripes) {
6376                 btrfs_err(fs_info, "invalid chunk num_stripes: %u",
6377                           num_stripes);
6378                 return -EIO;
6379         }
6380         if (!IS_ALIGNED(logical, fs_info->sectorsize)) {
6381                 btrfs_err(fs_info, "invalid chunk logical %llu", logical);
6382                 return -EIO;
6383         }
6384         if (btrfs_chunk_sector_size(leaf, chunk) != fs_info->sectorsize) {
6385                 btrfs_err(fs_info, "invalid chunk sectorsize %u",
6386                           btrfs_chunk_sector_size(leaf, chunk));
6387                 return -EIO;
6388         }
6389         if (!length || !IS_ALIGNED(length, fs_info->sectorsize)) {
6390                 btrfs_err(fs_info, "invalid chunk length %llu", length);
6391                 return -EIO;
6392         }
6393         if (!is_power_of_2(stripe_len) || stripe_len != BTRFS_STRIPE_LEN) {
6394                 btrfs_err(fs_info, "invalid chunk stripe length: %llu",
6395                           stripe_len);
6396                 return -EIO;
6397         }
6398         if (~(BTRFS_BLOCK_GROUP_TYPE_MASK | BTRFS_BLOCK_GROUP_PROFILE_MASK) &
6399             type) {
6400                 btrfs_err(fs_info, "unrecognized chunk type: %llu",
6401                           ~(BTRFS_BLOCK_GROUP_TYPE_MASK |
6402                             BTRFS_BLOCK_GROUP_PROFILE_MASK) &
6403                           btrfs_chunk_type(leaf, chunk));
6404                 return -EIO;
6405         }
6406         if ((type & BTRFS_BLOCK_GROUP_RAID10 && sub_stripes != 2) ||
6407             (type & BTRFS_BLOCK_GROUP_RAID1 && num_stripes < 1) ||
6408             (type & BTRFS_BLOCK_GROUP_RAID5 && num_stripes < 2) ||
6409             (type & BTRFS_BLOCK_GROUP_RAID6 && num_stripes < 3) ||
6410             (type & BTRFS_BLOCK_GROUP_DUP && num_stripes > 2) ||
6411             ((type & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 &&
6412              num_stripes != 1)) {
6413                 btrfs_err(fs_info,
6414                         "invalid num_stripes:sub_stripes %u:%u for profile %llu",
6415                         num_stripes, sub_stripes,
6416                         type & BTRFS_BLOCK_GROUP_PROFILE_MASK);
6417                 return -EIO;
6418         }
6419
6420         return 0;
6421 }
6422
6423 static int read_one_chunk(struct btrfs_fs_info *fs_info, struct btrfs_key *key,
6424                           struct extent_buffer *leaf,
6425                           struct btrfs_chunk *chunk)
6426 {
6427         struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
6428         struct map_lookup *map;
6429         struct extent_map *em;
6430         u64 logical;
6431         u64 length;
6432         u64 stripe_len;
6433         u64 devid;
6434         u8 uuid[BTRFS_UUID_SIZE];
6435         int num_stripes;
6436         int ret;
6437         int i;
6438
6439         logical = key->offset;
6440         length = btrfs_chunk_length(leaf, chunk);
6441         stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6442         num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6443
6444         ret = btrfs_check_chunk_valid(fs_info, leaf, chunk, logical);
6445         if (ret)
6446                 return ret;
6447
6448         read_lock(&map_tree->map_tree.lock);
6449         em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
6450         read_unlock(&map_tree->map_tree.lock);
6451
6452         /* already mapped? */
6453         if (em && em->start <= logical && em->start + em->len > logical) {
6454                 free_extent_map(em);
6455                 return 0;
6456         } else if (em) {
6457                 free_extent_map(em);
6458         }
6459
6460         em = alloc_extent_map();
6461         if (!em)
6462                 return -ENOMEM;
6463         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
6464         if (!map) {
6465                 free_extent_map(em);
6466                 return -ENOMEM;
6467         }
6468
6469         set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
6470         em->map_lookup = map;
6471         em->start = logical;
6472         em->len = length;
6473         em->orig_start = 0;
6474         em->block_start = 0;
6475         em->block_len = em->len;
6476
6477         map->num_stripes = num_stripes;
6478         map->io_width = btrfs_chunk_io_width(leaf, chunk);
6479         map->io_align = btrfs_chunk_io_align(leaf, chunk);
6480         map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
6481         map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6482         map->type = btrfs_chunk_type(leaf, chunk);
6483         map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
6484         for (i = 0; i < num_stripes; i++) {
6485                 map->stripes[i].physical =
6486                         btrfs_stripe_offset_nr(leaf, chunk, i);
6487                 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
6488                 read_extent_buffer(leaf, uuid, (unsigned long)
6489                                    btrfs_stripe_dev_uuid_nr(chunk, i),
6490                                    BTRFS_UUID_SIZE);
6491                 map->stripes[i].dev = btrfs_find_device(fs_info, devid,
6492                                                         uuid, NULL);
6493                 if (!map->stripes[i].dev &&
6494                     !btrfs_test_opt(fs_info, DEGRADED)) {
6495                         free_extent_map(em);
6496                         return -EIO;
6497                 }
6498                 if (!map->stripes[i].dev) {
6499                         map->stripes[i].dev =
6500                                 add_missing_dev(fs_info->fs_devices, devid,
6501                                                 uuid);
6502                         if (!map->stripes[i].dev) {
6503                                 free_extent_map(em);
6504                                 return -EIO;
6505                         }
6506                         btrfs_warn(fs_info, "devid %llu uuid %pU is missing",
6507                                    devid, uuid);
6508                 }
6509                 map->stripes[i].dev->in_fs_metadata = 1;
6510         }
6511
6512         write_lock(&map_tree->map_tree.lock);
6513         ret = add_extent_mapping(&map_tree->map_tree, em, 0);
6514         write_unlock(&map_tree->map_tree.lock);
6515         BUG_ON(ret); /* Tree corruption */
6516         free_extent_map(em);
6517
6518         return 0;
6519 }
6520
6521 static void fill_device_from_item(struct extent_buffer *leaf,
6522                                  struct btrfs_dev_item *dev_item,
6523                                  struct btrfs_device *device)
6524 {
6525         unsigned long ptr;
6526
6527         device->devid = btrfs_device_id(leaf, dev_item);
6528         device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
6529         device->total_bytes = device->disk_total_bytes;
6530         device->commit_total_bytes = device->disk_total_bytes;
6531         device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
6532         device->commit_bytes_used = device->bytes_used;
6533         device->type = btrfs_device_type(leaf, dev_item);
6534         device->io_align = btrfs_device_io_align(leaf, dev_item);
6535         device->io_width = btrfs_device_io_width(leaf, dev_item);
6536         device->sector_size = btrfs_device_sector_size(leaf, dev_item);
6537         WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
6538         device->is_tgtdev_for_dev_replace = 0;
6539
6540         ptr = btrfs_device_uuid(dev_item);
6541         read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
6542 }
6543
6544 static struct btrfs_fs_devices *open_seed_devices(struct btrfs_fs_info *fs_info,
6545                                                   u8 *fsid)
6546 {
6547         struct btrfs_fs_devices *fs_devices;
6548         int ret;
6549
6550         BUG_ON(!mutex_is_locked(&uuid_mutex));
6551
6552         fs_devices = fs_info->fs_devices->seed;
6553         while (fs_devices) {
6554                 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE))
6555                         return fs_devices;
6556
6557                 fs_devices = fs_devices->seed;
6558         }
6559
6560         fs_devices = find_fsid(fsid);
6561         if (!fs_devices) {
6562                 if (!btrfs_test_opt(fs_info, DEGRADED))
6563                         return ERR_PTR(-ENOENT);
6564
6565                 fs_devices = alloc_fs_devices(fsid);
6566                 if (IS_ERR(fs_devices))
6567                         return fs_devices;
6568
6569                 fs_devices->seeding = 1;
6570                 fs_devices->opened = 1;
6571                 return fs_devices;
6572         }
6573
6574         fs_devices = clone_fs_devices(fs_devices);
6575         if (IS_ERR(fs_devices))
6576                 return fs_devices;
6577
6578         ret = __btrfs_open_devices(fs_devices, FMODE_READ,
6579                                    fs_info->bdev_holder);
6580         if (ret) {
6581                 free_fs_devices(fs_devices);
6582                 fs_devices = ERR_PTR(ret);
6583                 goto out;
6584         }
6585
6586         if (!fs_devices->seeding) {
6587                 __btrfs_close_devices(fs_devices);
6588                 free_fs_devices(fs_devices);
6589                 fs_devices = ERR_PTR(-EINVAL);
6590                 goto out;
6591         }
6592
6593         fs_devices->seed = fs_info->fs_devices->seed;
6594         fs_info->fs_devices->seed = fs_devices;
6595 out:
6596         return fs_devices;
6597 }
6598
6599 static int read_one_dev(struct btrfs_fs_info *fs_info,
6600                         struct extent_buffer *leaf,
6601                         struct btrfs_dev_item *dev_item)
6602 {
6603         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6604         struct btrfs_device *device;
6605         u64 devid;
6606         int ret;
6607         u8 fs_uuid[BTRFS_UUID_SIZE];
6608         u8 dev_uuid[BTRFS_UUID_SIZE];
6609
6610         devid = btrfs_device_id(leaf, dev_item);
6611         read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
6612                            BTRFS_UUID_SIZE);
6613         read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
6614                            BTRFS_UUID_SIZE);
6615
6616         if (memcmp(fs_uuid, fs_info->fsid, BTRFS_UUID_SIZE)) {
6617                 fs_devices = open_seed_devices(fs_info, fs_uuid);
6618                 if (IS_ERR(fs_devices))
6619                         return PTR_ERR(fs_devices);
6620         }
6621
6622         device = btrfs_find_device(fs_info, devid, dev_uuid, fs_uuid);
6623         if (!device) {
6624                 if (!btrfs_test_opt(fs_info, DEGRADED))
6625                         return -EIO;
6626
6627                 device = add_missing_dev(fs_devices, devid, dev_uuid);
6628                 if (!device)
6629                         return -ENOMEM;
6630                 btrfs_warn(fs_info, "devid %llu uuid %pU missing",
6631                                 devid, dev_uuid);
6632         } else {
6633                 if (!device->bdev && !btrfs_test_opt(fs_info, DEGRADED))
6634                         return -EIO;
6635
6636                 if(!device->bdev && !device->missing) {
6637                         /*
6638                          * this happens when a device that was properly setup
6639                          * in the device info lists suddenly goes bad.
6640                          * device->bdev is NULL, and so we have to set
6641                          * device->missing to one here
6642                          */
6643                         device->fs_devices->missing_devices++;
6644                         device->missing = 1;
6645                 }
6646
6647                 /* Move the device to its own fs_devices */
6648                 if (device->fs_devices != fs_devices) {
6649                         ASSERT(device->missing);
6650
6651                         list_move(&device->dev_list, &fs_devices->devices);
6652                         device->fs_devices->num_devices--;
6653                         fs_devices->num_devices++;
6654
6655                         device->fs_devices->missing_devices--;
6656                         fs_devices->missing_devices++;
6657
6658                         device->fs_devices = fs_devices;
6659                 }
6660         }
6661
6662         if (device->fs_devices != fs_info->fs_devices) {
6663                 BUG_ON(device->writeable);
6664                 if (device->generation !=
6665                     btrfs_device_generation(leaf, dev_item))
6666                         return -EINVAL;
6667         }
6668
6669         fill_device_from_item(leaf, dev_item, device);
6670         device->in_fs_metadata = 1;
6671         if (device->writeable && !device->is_tgtdev_for_dev_replace) {
6672                 device->fs_devices->total_rw_bytes += device->total_bytes;
6673                 atomic64_add(device->total_bytes - device->bytes_used,
6674                                 &fs_info->free_chunk_space);
6675         }
6676         ret = 0;
6677         return ret;
6678 }
6679
6680 int btrfs_read_sys_array(struct btrfs_fs_info *fs_info)
6681 {
6682         struct btrfs_root *root = fs_info->tree_root;
6683         struct btrfs_super_block *super_copy = fs_info->super_copy;
6684         struct extent_buffer *sb;
6685         struct btrfs_disk_key *disk_key;
6686         struct btrfs_chunk *chunk;
6687         u8 *array_ptr;
6688         unsigned long sb_array_offset;
6689         int ret = 0;
6690         u32 num_stripes;
6691         u32 array_size;
6692         u32 len = 0;
6693         u32 cur_offset;
6694         u64 type;
6695         struct btrfs_key key;
6696
6697         ASSERT(BTRFS_SUPER_INFO_SIZE <= fs_info->nodesize);
6698         /*
6699          * This will create extent buffer of nodesize, superblock size is
6700          * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will
6701          * overallocate but we can keep it as-is, only the first page is used.
6702          */
6703         sb = btrfs_find_create_tree_block(fs_info, BTRFS_SUPER_INFO_OFFSET);
6704         if (IS_ERR(sb))
6705                 return PTR_ERR(sb);
6706         set_extent_buffer_uptodate(sb);
6707         btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
6708         /*
6709          * The sb extent buffer is artificial and just used to read the system array.
6710          * set_extent_buffer_uptodate() call does not properly mark all it's
6711          * pages up-to-date when the page is larger: extent does not cover the
6712          * whole page and consequently check_page_uptodate does not find all
6713          * the page's extents up-to-date (the hole beyond sb),
6714          * write_extent_buffer then triggers a WARN_ON.
6715          *
6716          * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
6717          * but sb spans only this function. Add an explicit SetPageUptodate call
6718          * to silence the warning eg. on PowerPC 64.
6719          */
6720         if (PAGE_SIZE > BTRFS_SUPER_INFO_SIZE)
6721                 SetPageUptodate(sb->pages[0]);
6722
6723         write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
6724         array_size = btrfs_super_sys_array_size(super_copy);
6725
6726         array_ptr = super_copy->sys_chunk_array;
6727         sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array);
6728         cur_offset = 0;
6729
6730         while (cur_offset < array_size) {
6731                 disk_key = (struct btrfs_disk_key *)array_ptr;
6732                 len = sizeof(*disk_key);
6733                 if (cur_offset + len > array_size)
6734                         goto out_short_read;
6735
6736                 btrfs_disk_key_to_cpu(&key, disk_key);
6737
6738                 array_ptr += len;
6739                 sb_array_offset += len;
6740                 cur_offset += len;
6741
6742                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
6743                         chunk = (struct btrfs_chunk *)sb_array_offset;
6744                         /*
6745                          * At least one btrfs_chunk with one stripe must be
6746                          * present, exact stripe count check comes afterwards
6747                          */
6748                         len = btrfs_chunk_item_size(1);
6749                         if (cur_offset + len > array_size)
6750                                 goto out_short_read;
6751
6752                         num_stripes = btrfs_chunk_num_stripes(sb, chunk);
6753                         if (!num_stripes) {
6754                                 btrfs_err(fs_info,
6755                                         "invalid number of stripes %u in sys_array at offset %u",
6756                                         num_stripes, cur_offset);
6757                                 ret = -EIO;
6758                                 break;
6759                         }
6760
6761                         type = btrfs_chunk_type(sb, chunk);
6762                         if ((type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) {
6763                                 btrfs_err(fs_info,
6764                             "invalid chunk type %llu in sys_array at offset %u",
6765                                         type, cur_offset);
6766                                 ret = -EIO;
6767                                 break;
6768                         }
6769
6770                         len = btrfs_chunk_item_size(num_stripes);
6771                         if (cur_offset + len > array_size)
6772                                 goto out_short_read;
6773
6774                         ret = read_one_chunk(fs_info, &key, sb, chunk);
6775                         if (ret)
6776                                 break;
6777                 } else {
6778                         btrfs_err(fs_info,
6779                             "unexpected item type %u in sys_array at offset %u",
6780                                   (u32)key.type, cur_offset);
6781                         ret = -EIO;
6782                         break;
6783                 }
6784                 array_ptr += len;
6785                 sb_array_offset += len;
6786                 cur_offset += len;
6787         }
6788         clear_extent_buffer_uptodate(sb);
6789         free_extent_buffer_stale(sb);
6790         return ret;
6791
6792 out_short_read:
6793         btrfs_err(fs_info, "sys_array too short to read %u bytes at offset %u",
6794                         len, cur_offset);
6795         clear_extent_buffer_uptodate(sb);
6796         free_extent_buffer_stale(sb);
6797         return -EIO;
6798 }
6799
6800 int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info)
6801 {
6802         struct btrfs_root *root = fs_info->chunk_root;
6803         struct btrfs_path *path;
6804         struct extent_buffer *leaf;
6805         struct btrfs_key key;
6806         struct btrfs_key found_key;
6807         int ret;
6808         int slot;
6809         u64 total_dev = 0;
6810
6811         path = btrfs_alloc_path();
6812         if (!path)
6813                 return -ENOMEM;
6814
6815         mutex_lock(&uuid_mutex);
6816         mutex_lock(&fs_info->chunk_mutex);
6817
6818         /*
6819          * Read all device items, and then all the chunk items. All
6820          * device items are found before any chunk item (their object id
6821          * is smaller than the lowest possible object id for a chunk
6822          * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
6823          */
6824         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
6825         key.offset = 0;
6826         key.type = 0;
6827         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6828         if (ret < 0)
6829                 goto error;
6830         while (1) {
6831                 leaf = path->nodes[0];
6832                 slot = path->slots[0];
6833                 if (slot >= btrfs_header_nritems(leaf)) {
6834                         ret = btrfs_next_leaf(root, path);
6835                         if (ret == 0)
6836                                 continue;
6837                         if (ret < 0)
6838                                 goto error;
6839                         break;
6840                 }
6841                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6842                 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
6843                         struct btrfs_dev_item *dev_item;
6844                         dev_item = btrfs_item_ptr(leaf, slot,
6845                                                   struct btrfs_dev_item);
6846                         ret = read_one_dev(fs_info, leaf, dev_item);
6847                         if (ret)
6848                                 goto error;
6849                         total_dev++;
6850                 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
6851                         struct btrfs_chunk *chunk;
6852                         chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
6853                         ret = read_one_chunk(fs_info, &found_key, leaf, chunk);
6854                         if (ret)
6855                                 goto error;
6856                 }
6857                 path->slots[0]++;
6858         }
6859
6860         /*
6861          * After loading chunk tree, we've got all device information,
6862          * do another round of validation checks.
6863          */
6864         if (total_dev != fs_info->fs_devices->total_devices) {
6865                 btrfs_err(fs_info,
6866            "super_num_devices %llu mismatch with num_devices %llu found here",
6867                           btrfs_super_num_devices(fs_info->super_copy),
6868                           total_dev);
6869                 ret = -EINVAL;
6870                 goto error;
6871         }
6872         if (btrfs_super_total_bytes(fs_info->super_copy) <
6873             fs_info->fs_devices->total_rw_bytes) {
6874                 btrfs_err(fs_info,
6875         "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu",
6876                           btrfs_super_total_bytes(fs_info->super_copy),
6877                           fs_info->fs_devices->total_rw_bytes);
6878                 ret = -EINVAL;
6879                 goto error;
6880         }
6881         ret = 0;
6882 error:
6883         mutex_unlock(&fs_info->chunk_mutex);
6884         mutex_unlock(&uuid_mutex);
6885
6886         btrfs_free_path(path);
6887         return ret;
6888 }
6889
6890 void btrfs_init_devices_late(struct btrfs_fs_info *fs_info)
6891 {
6892         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6893         struct btrfs_device *device;
6894
6895         while (fs_devices) {
6896                 mutex_lock(&fs_devices->device_list_mutex);
6897                 list_for_each_entry(device, &fs_devices->devices, dev_list)
6898                         device->fs_info = fs_info;
6899                 mutex_unlock(&fs_devices->device_list_mutex);
6900
6901                 fs_devices = fs_devices->seed;
6902         }
6903 }
6904
6905 static void __btrfs_reset_dev_stats(struct btrfs_device *dev)
6906 {
6907         int i;
6908
6909         for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6910                 btrfs_dev_stat_reset(dev, i);
6911 }
6912
6913 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
6914 {
6915         struct btrfs_key key;
6916         struct btrfs_key found_key;
6917         struct btrfs_root *dev_root = fs_info->dev_root;
6918         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6919         struct extent_buffer *eb;
6920         int slot;
6921         int ret = 0;
6922         struct btrfs_device *device;
6923         struct btrfs_path *path = NULL;
6924         int i;
6925
6926         path = btrfs_alloc_path();
6927         if (!path) {
6928                 ret = -ENOMEM;
6929                 goto out;
6930         }
6931
6932         mutex_lock(&fs_devices->device_list_mutex);
6933         list_for_each_entry(device, &fs_devices->devices, dev_list) {
6934                 int item_size;
6935                 struct btrfs_dev_stats_item *ptr;
6936
6937                 key.objectid = BTRFS_DEV_STATS_OBJECTID;
6938                 key.type = BTRFS_PERSISTENT_ITEM_KEY;
6939                 key.offset = device->devid;
6940                 ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
6941                 if (ret) {
6942                         __btrfs_reset_dev_stats(device);
6943                         device->dev_stats_valid = 1;
6944                         btrfs_release_path(path);
6945                         continue;
6946                 }
6947                 slot = path->slots[0];
6948                 eb = path->nodes[0];
6949                 btrfs_item_key_to_cpu(eb, &found_key, slot);
6950                 item_size = btrfs_item_size_nr(eb, slot);
6951
6952                 ptr = btrfs_item_ptr(eb, slot,
6953                                      struct btrfs_dev_stats_item);
6954
6955                 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
6956                         if (item_size >= (1 + i) * sizeof(__le64))
6957                                 btrfs_dev_stat_set(device, i,
6958                                         btrfs_dev_stats_value(eb, ptr, i));
6959                         else
6960                                 btrfs_dev_stat_reset(device, i);
6961                 }
6962
6963                 device->dev_stats_valid = 1;
6964                 btrfs_dev_stat_print_on_load(device);
6965                 btrfs_release_path(path);
6966         }
6967         mutex_unlock(&fs_devices->device_list_mutex);
6968
6969 out:
6970         btrfs_free_path(path);
6971         return ret < 0 ? ret : 0;
6972 }
6973
6974 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
6975                                 struct btrfs_fs_info *fs_info,
6976                                 struct btrfs_device *device)
6977 {
6978         struct btrfs_root *dev_root = fs_info->dev_root;
6979         struct btrfs_path *path;
6980         struct btrfs_key key;
6981         struct extent_buffer *eb;
6982         struct btrfs_dev_stats_item *ptr;
6983         int ret;
6984         int i;
6985
6986         key.objectid = BTRFS_DEV_STATS_OBJECTID;
6987         key.type = BTRFS_PERSISTENT_ITEM_KEY;
6988         key.offset = device->devid;
6989
6990         path = btrfs_alloc_path();
6991         if (!path)
6992                 return -ENOMEM;
6993         ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
6994         if (ret < 0) {
6995                 btrfs_warn_in_rcu(fs_info,
6996                         "error %d while searching for dev_stats item for device %s",
6997                               ret, rcu_str_deref(device->name));
6998                 goto out;
6999         }
7000
7001         if (ret == 0 &&
7002             btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
7003                 /* need to delete old one and insert a new one */
7004                 ret = btrfs_del_item(trans, dev_root, path);
7005                 if (ret != 0) {
7006                         btrfs_warn_in_rcu(fs_info,
7007                                 "delete too small dev_stats item for device %s failed %d",
7008                                       rcu_str_deref(device->name), ret);
7009                         goto out;
7010                 }
7011                 ret = 1;
7012         }
7013
7014         if (ret == 1) {
7015                 /* need to insert a new item */
7016                 btrfs_release_path(path);
7017                 ret = btrfs_insert_empty_item(trans, dev_root, path,
7018                                               &key, sizeof(*ptr));
7019                 if (ret < 0) {
7020                         btrfs_warn_in_rcu(fs_info,
7021                                 "insert dev_stats item for device %s failed %d",
7022                                 rcu_str_deref(device->name), ret);
7023                         goto out;
7024                 }
7025         }
7026
7027         eb = path->nodes[0];
7028         ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
7029         for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7030                 btrfs_set_dev_stats_value(eb, ptr, i,
7031                                           btrfs_dev_stat_read(device, i));
7032         btrfs_mark_buffer_dirty(eb);
7033
7034 out:
7035         btrfs_free_path(path);
7036         return ret;
7037 }
7038
7039 /*
7040  * called from commit_transaction. Writes all changed device stats to disk.
7041  */
7042 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans,
7043                         struct btrfs_fs_info *fs_info)
7044 {
7045         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7046         struct btrfs_device *device;
7047         int stats_cnt;
7048         int ret = 0;
7049
7050         mutex_lock(&fs_devices->device_list_mutex);
7051         list_for_each_entry(device, &fs_devices->devices, dev_list) {
7052                 if (!device->dev_stats_valid || !btrfs_dev_stats_dirty(device))
7053                         continue;
7054
7055                 stats_cnt = atomic_read(&device->dev_stats_ccnt);
7056                 ret = update_dev_stat_item(trans, fs_info, device);
7057                 if (!ret)
7058                         atomic_sub(stats_cnt, &device->dev_stats_ccnt);
7059         }
7060         mutex_unlock(&fs_devices->device_list_mutex);
7061
7062         return ret;
7063 }
7064
7065 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
7066 {
7067         btrfs_dev_stat_inc(dev, index);
7068         btrfs_dev_stat_print_on_error(dev);
7069 }
7070
7071 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
7072 {
7073         if (!dev->dev_stats_valid)
7074                 return;
7075         btrfs_err_rl_in_rcu(dev->fs_info,
7076                 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7077                            rcu_str_deref(dev->name),
7078                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7079                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7080                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7081                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7082                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7083 }
7084
7085 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
7086 {
7087         int i;
7088
7089         for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7090                 if (btrfs_dev_stat_read(dev, i) != 0)
7091                         break;
7092         if (i == BTRFS_DEV_STAT_VALUES_MAX)
7093                 return; /* all values == 0, suppress message */
7094
7095         btrfs_info_in_rcu(dev->fs_info,
7096                 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7097                rcu_str_deref(dev->name),
7098                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7099                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7100                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7101                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7102                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7103 }
7104
7105 int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info,
7106                         struct btrfs_ioctl_get_dev_stats *stats)
7107 {
7108         struct btrfs_device *dev;
7109         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7110         int i;
7111
7112         mutex_lock(&fs_devices->device_list_mutex);
7113         dev = btrfs_find_device(fs_info, stats->devid, NULL, NULL);
7114         mutex_unlock(&fs_devices->device_list_mutex);
7115
7116         if (!dev) {
7117                 btrfs_warn(fs_info, "get dev_stats failed, device not found");
7118                 return -ENODEV;
7119         } else if (!dev->dev_stats_valid) {
7120                 btrfs_warn(fs_info, "get dev_stats failed, not yet valid");
7121                 return -ENODEV;
7122         } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
7123                 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7124                         if (stats->nr_items > i)
7125                                 stats->values[i] =
7126                                         btrfs_dev_stat_read_and_reset(dev, i);
7127                         else
7128                                 btrfs_dev_stat_reset(dev, i);
7129                 }
7130         } else {
7131                 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7132                         if (stats->nr_items > i)
7133                                 stats->values[i] = btrfs_dev_stat_read(dev, i);
7134         }
7135         if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
7136                 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
7137         return 0;
7138 }
7139
7140 void btrfs_scratch_superblocks(struct block_device *bdev, const char *device_path)
7141 {
7142         struct buffer_head *bh;
7143         struct btrfs_super_block *disk_super;
7144         int copy_num;
7145
7146         if (!bdev)
7147                 return;
7148
7149         for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX;
7150                 copy_num++) {
7151
7152                 if (btrfs_read_dev_one_super(bdev, copy_num, &bh))
7153                         continue;
7154
7155                 disk_super = (struct btrfs_super_block *)bh->b_data;
7156
7157                 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
7158                 set_buffer_dirty(bh);
7159                 sync_dirty_buffer(bh);
7160                 brelse(bh);
7161         }
7162
7163         /* Notify udev that device has changed */
7164         btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
7165
7166         /* Update ctime/mtime for device path for libblkid */
7167         update_dev_time(device_path);
7168 }
7169
7170 /*
7171  * Update the size of all devices, which is used for writing out the
7172  * super blocks.
7173  */
7174 void btrfs_update_commit_device_size(struct btrfs_fs_info *fs_info)
7175 {
7176         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7177         struct btrfs_device *curr, *next;
7178
7179         if (list_empty(&fs_devices->resized_devices))
7180                 return;
7181
7182         mutex_lock(&fs_devices->device_list_mutex);
7183         mutex_lock(&fs_info->chunk_mutex);
7184         list_for_each_entry_safe(curr, next, &fs_devices->resized_devices,
7185                                  resized_list) {
7186                 list_del_init(&curr->resized_list);
7187                 curr->commit_total_bytes = curr->disk_total_bytes;
7188         }
7189         mutex_unlock(&fs_info->chunk_mutex);
7190         mutex_unlock(&fs_devices->device_list_mutex);
7191 }
7192
7193 /* Must be invoked during the transaction commit */
7194 void btrfs_update_commit_device_bytes_used(struct btrfs_fs_info *fs_info,
7195                                         struct btrfs_transaction *transaction)
7196 {
7197         struct extent_map *em;
7198         struct map_lookup *map;
7199         struct btrfs_device *dev;
7200         int i;
7201
7202         if (list_empty(&transaction->pending_chunks))
7203                 return;
7204
7205         /* In order to kick the device replace finish process */
7206         mutex_lock(&fs_info->chunk_mutex);
7207         list_for_each_entry(em, &transaction->pending_chunks, list) {
7208                 map = em->map_lookup;
7209
7210                 for (i = 0; i < map->num_stripes; i++) {
7211                         dev = map->stripes[i].dev;
7212                         dev->commit_bytes_used = dev->bytes_used;
7213                 }
7214         }
7215         mutex_unlock(&fs_info->chunk_mutex);
7216 }
7217
7218 void btrfs_set_fs_info_ptr(struct btrfs_fs_info *fs_info)
7219 {
7220         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7221         while (fs_devices) {
7222                 fs_devices->fs_info = fs_info;
7223                 fs_devices = fs_devices->seed;
7224         }
7225 }
7226
7227 void btrfs_reset_fs_info_ptr(struct btrfs_fs_info *fs_info)
7228 {
7229         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7230         while (fs_devices) {
7231                 fs_devices->fs_info = NULL;
7232                 fs_devices = fs_devices->seed;
7233         }
7234 }