fs/btrfs/volumes.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18 #include <linux/sched.h>
  19 #include <linux/bio.h>
  20 #include <linux/slab.h>
  21 #include <linux/buffer_head.h>
  22 #include <linux/blkdev.h>
  23 #include <linux/random.h>
  24 #include <linux/iocontext.h>
  25 #include <linux/capability.h>
  26 #include <linux/ratelimit.h>
  27 #include <linux/kthread.h>
  28 #include <linux/raid/pq.h>
  29 #include <linux/semaphore.h>
  30 #include <asm/div64.h>
  31 #include "ctree.h"
  32 #include "extent_map.h"
  33 #include "disk-io.h"
  34 #include "transaction.h"
  35 #include "print-tree.h"
  36 #include "volumes.h"
  37 #include "raid56.h"
  38 #include "async-thread.h"
  39 #include "check-integrity.h"
  40 #include "rcu-string.h"
  41 #include "math.h"
  42 #include "dev-replace.h"
  43 #include "sysfs.h"
  44
  45 const struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = {
  46         [BTRFS_RAID_RAID10] = {
  47                 .sub_stripes    = 2,
  48                 .dev_stripes    = 1,
  49                 .devs_max       = 0,    /* 0 == as many as possible */
  50                 .devs_min       = 4,
  51                 .tolerated_failures = 1,
  52                 .devs_increment = 2,
  53                 .ncopies        = 2,
  54         },
  55         [BTRFS_RAID_RAID1] = {
  56                 .sub_stripes    = 1,
  57                 .dev_stripes    = 1,
  58                 .devs_max       = 2,
  59                 .devs_min       = 2,
  60                 .tolerated_failures = 1,
  61                 .devs_increment = 2,
  62                 .ncopies        = 2,
  63         },
  64         [BTRFS_RAID_DUP] = {
  65                 .sub_stripes    = 1,
  66                 .dev_stripes    = 2,
  67                 .devs_max       = 1,
  68                 .devs_min       = 1,
  69                 .tolerated_failures = 0,
  70                 .devs_increment = 1,
  71                 .ncopies        = 2,
  72         },
  73         [BTRFS_RAID_RAID0] = {
  74                 .sub_stripes    = 1,
  75                 .dev_stripes    = 1,
  76                 .devs_max       = 0,
  77                 .devs_min       = 2,
  78                 .tolerated_failures = 0,
  79                 .devs_increment = 1,
  80                 .ncopies        = 1,
  81         },
  82         [BTRFS_RAID_SINGLE] = {
  83                 .sub_stripes    = 1,
  84                 .dev_stripes    = 1,
  85                 .devs_max       = 1,
  86                 .devs_min       = 1,
  87                 .tolerated_failures = 0,
  88                 .devs_increment = 1,
  89                 .ncopies        = 1,
  90         },
  91         [BTRFS_RAID_RAID5] = {
  92                 .sub_stripes    = 1,
  93                 .dev_stripes    = 1,
  94                 .devs_max       = 0,
  95                 .devs_min       = 2,
  96                 .tolerated_failures = 1,
  97                 .devs_increment = 1,
  98                 .ncopies        = 2,
  99         },
 100         [BTRFS_RAID_RAID6] = {
 101                 .sub_stripes    = 1,
 102                 .dev_stripes    = 1,
 103                 .devs_max       = 0,
 104                 .devs_min       = 3,
 105                 .tolerated_failures = 2,
 106                 .devs_increment = 1,
 107                 .ncopies        = 3,
 108         },
 109 };
 110
 111 const u64 const btrfs_raid_group[BTRFS_NR_RAID_TYPES] = {
 112         [BTRFS_RAID_RAID10] = BTRFS_BLOCK_GROUP_RAID10,
 113         [BTRFS_RAID_RAID1]  = BTRFS_BLOCK_GROUP_RAID1,
 114         [BTRFS_RAID_DUP]    = BTRFS_BLOCK_GROUP_DUP,
 115         [BTRFS_RAID_RAID0]  = BTRFS_BLOCK_GROUP_RAID0,
 116         [BTRFS_RAID_SINGLE] = 0,
 117         [BTRFS_RAID_RAID5]  = BTRFS_BLOCK_GROUP_RAID5,
 118         [BTRFS_RAID_RAID6]  = BTRFS_BLOCK_GROUP_RAID6,
 119 };
 120
 121 static int init_first_rw_device(struct btrfs_trans_handle *trans,
 122                                 struct btrfs_root *root,
 123                                 struct btrfs_device *device);
 124 static int btrfs_relocate_sys_chunks(struct btrfs_root *root);
 125 static void __btrfs_reset_dev_stats(struct btrfs_device *dev);
 126 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev);
 127 static void btrfs_dev_stat_print_on_load(struct btrfs_device *device);
 128
 129 DEFINE_MUTEX(uuid_mutex);
 130 static LIST_HEAD(fs_uuids);
 131 struct list_head *btrfs_get_fs_uuids(void)
 132 {
 133         return &fs_uuids;
 134 }
 135
 136 static struct btrfs_fs_devices *__alloc_fs_devices(void)
 137 {
 138         struct btrfs_fs_devices *fs_devs;
 139
 140         fs_devs = kzalloc(sizeof(*fs_devs), GFP_NOFS);
 141         if (!fs_devs)
 142                 return ERR_PTR(-ENOMEM);
 143
 144         mutex_init(&fs_devs->device_list_mutex);
 145
 146         INIT_LIST_HEAD(&fs_devs->devices);
 147         INIT_LIST_HEAD(&fs_devs->resized_devices);
 148         INIT_LIST_HEAD(&fs_devs->alloc_list);
 149         INIT_LIST_HEAD(&fs_devs->list);
 150
 151         return fs_devs;
 152 }
 153
 154 /**
 155  * alloc_fs_devices - allocate struct btrfs_fs_devices
 156  * @fsid:       a pointer to UUID for this FS.  If NULL a new UUID is
 157  *              generated.
 158  *
 159  * Return: a pointer to a new &struct btrfs_fs_devices on success;
 160  * ERR_PTR() on error.  Returned struct is not linked onto any lists and
 161  * can be destroyed with kfree() right away.
 162  */
 163 static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid)
 164 {
 165         struct btrfs_fs_devices *fs_devs;
 166
 167         fs_devs = __alloc_fs_devices();
 168         if (IS_ERR(fs_devs))
 169                 return fs_devs;
 170
 171         if (fsid)
 172                 memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE);
 173         else
 174                 generate_random_uuid(fs_devs->fsid);
 175
 176         return fs_devs;
 177 }
 178
 179 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
 180 {
 181         struct btrfs_device *device;
 182         WARN_ON(fs_devices->opened);
 183         while (!list_empty(&fs_devices->devices)) {
 184                 device = list_entry(fs_devices->devices.next,
 185                                     struct btrfs_device, dev_list);
 186                 list_del(&device->dev_list);
 187                 rcu_string_free(device->name);
 188                 kfree(device);
 189         }
 190         kfree(fs_devices);
 191 }
 192
 193 static void btrfs_kobject_uevent(struct block_device *bdev,
 194                                  enum kobject_action action)
 195 {
 196         int ret;
 197
 198         ret = kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, action);
 199         if (ret)
 200                 pr_warn("BTRFS: Sending event '%d' to kobject: '%s' (%p): failed\n",
 201                         action,
 202                         kobject_name(&disk_to_dev(bdev->bd_disk)->kobj),
 203                         &disk_to_dev(bdev->bd_disk)->kobj);
 204 }
 205
 206 void btrfs_cleanup_fs_uuids(void)
 207 {
 208         struct btrfs_fs_devices *fs_devices;
 209
 210         while (!list_empty(&fs_uuids)) {
 211                 fs_devices = list_entry(fs_uuids.next,
 212                                         struct btrfs_fs_devices, list);
 213                 list_del(&fs_devices->list);
 214                 free_fs_devices(fs_devices);
 215         }
 216 }
 217
 218 static struct btrfs_device *__alloc_device(void)
 219 {
 220         struct btrfs_device *dev;
 221
 222         dev = kzalloc(sizeof(*dev), GFP_NOFS);
 223         if (!dev)
 224                 return ERR_PTR(-ENOMEM);
 225
 226         INIT_LIST_HEAD(&dev->dev_list);
 227         INIT_LIST_HEAD(&dev->dev_alloc_list);
 228         INIT_LIST_HEAD(&dev->resized_list);
 229
 230         spin_lock_init(&dev->io_lock);
 231
 232         spin_lock_init(&dev->reada_lock);
 233         atomic_set(&dev->reada_in_flight, 0);
 234         atomic_set(&dev->dev_stats_ccnt, 0);
 235         INIT_RADIX_TREE(&dev->reada_zones, GFP_NOFS & ~__GFP_WAIT);
 236         INIT_RADIX_TREE(&dev->reada_extents, GFP_NOFS & ~__GFP_WAIT);
 237
 238         return dev;
 239 }
 240
 241 static noinline struct btrfs_device *__find_device(struct list_head *head,
 242                                                    u64 devid, u8 *uuid)
 243 {
 244         struct btrfs_device *dev;
 245
 246         list_for_each_entry(dev, head, dev_list) {
 247                 if (dev->devid == devid &&
 248                     (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
 249                         return dev;
 250                 }
 251         }
 252         return NULL;
 253 }
 254
 255 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
 256 {
 257         struct btrfs_fs_devices *fs_devices;
 258
 259         list_for_each_entry(fs_devices, &fs_uuids, list) {
 260                 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
 261                         return fs_devices;
 262         }
 263         return NULL;
 264 }
 265
 266 static int
 267 btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder,
 268                       int flush, struct block_device **bdev,
 269                       struct buffer_head **bh)
 270 {
 271         int ret;
 272
 273         *bdev = blkdev_get_by_path(device_path, flags, holder);
 274
 275         if (IS_ERR(*bdev)) {
 276                 ret = PTR_ERR(*bdev);
 277                 goto error;
 278         }
 279
 280         if (flush)
 281                 filemap_write_and_wait((*bdev)->bd_inode->i_mapping);
 282         ret = set_blocksize(*bdev, 4096);
 283         if (ret) {
 284                 blkdev_put(*bdev, flags);
 285                 goto error;
 286         }
 287         invalidate_bdev(*bdev);
 288         *bh = btrfs_read_dev_super(*bdev);
 289         if (IS_ERR(*bh)) {
 290                 ret = PTR_ERR(*bh);
 291                 blkdev_put(*bdev, flags);
 292                 goto error;
 293         }
 294
 295         return 0;
 296
 297 error:
 298         *bdev = NULL;
 299         *bh = NULL;
 300         return ret;
 301 }
 302
 303 static void requeue_list(struct btrfs_pending_bios *pending_bios,
 304                         struct bio *head, struct bio *tail)
 305 {
 306
 307         struct bio *old_head;
 308
 309         old_head = pending_bios->head;
 310         pending_bios->head = head;
 311         if (pending_bios->tail)
 312                 tail->bi_next = old_head;
 313         else
 314                 pending_bios->tail = tail;
 315 }
 316
 317 /*
 318  * we try to collect pending bios for a device so we don't get a large
 319  * number of procs sending bios down to the same device.  This greatly
 320  * improves the schedulers ability to collect and merge the bios.
 321  *
 322  * But, it also turns into a long list of bios to process and that is sure
 323  * to eventually make the worker thread block.  The solution here is to
 324  * make some progress and then put this work struct back at the end of
 325  * the list if the block device is congested.  This way, multiple devices
 326  * can make progress from a single worker thread.
 327  */
 328 static noinline void run_scheduled_bios(struct btrfs_device *device)
 329 {
 330         struct bio *pending;
 331         struct backing_dev_info *bdi;
 332         struct btrfs_fs_info *fs_info;
 333         struct btrfs_pending_bios *pending_bios;
 334         struct bio *tail;
 335         struct bio *cur;
 336         int again = 0;
 337         unsigned long num_run;
 338         unsigned long batch_run = 0;
 339         unsigned long limit;
 340         unsigned long last_waited = 0;
 341         int force_reg = 0;
 342         int sync_pending = 0;
 343         struct blk_plug plug;
 344
 345         /*
 346          * this function runs all the bios we've collected for
 347          * a particular device.  We don't want to wander off to
 348          * another device without first sending all of these down.
 349          * So, setup a plug here and finish it off before we return
 350          */
 351         blk_start_plug(&plug);
 352
 353         bdi = blk_get_backing_dev_info(device->bdev);
 354         fs_info = device->dev_root->fs_info;
 355         limit = btrfs_async_submit_limit(fs_info);
 356         limit = limit * 2 / 3;
 357
 358 loop:
 359         spin_lock(&device->io_lock);
 360
 361 loop_lock:
 362         num_run = 0;
 363
 364         /* take all the bios off the list at once and process them
 365          * later on (without the lock held).  But, remember the
 366          * tail and other pointers so the bios can be properly reinserted
 367          * into the list if we hit congestion
 368          */
 369         if (!force_reg && device->pending_sync_bios.head) {
 370                 pending_bios = &device->pending_sync_bios;
 371                 force_reg = 1;
 372         } else {
 373                 pending_bios = &device->pending_bios;
 374                 force_reg = 0;
 375         }
 376
 377         pending = pending_bios->head;
 378         tail = pending_bios->tail;
 379         WARN_ON(pending && !tail);
 380
 381         /*
 382          * if pending was null this time around, no bios need processing
 383          * at all and we can stop.  Otherwise it'll loop back up again
 384          * and do an additional check so no bios are missed.
 385          *
 386          * device->running_pending is used to synchronize with the
 387          * schedule_bio code.
 388          */
 389         if (device->pending_sync_bios.head == NULL &&
 390             device->pending_bios.head == NULL) {
 391                 again = 0;
 392                 device->running_pending = 0;
 393         } else {
 394                 again = 1;
 395                 device->running_pending = 1;
 396         }
 397
 398         pending_bios->head = NULL;
 399         pending_bios->tail = NULL;
 400
 401         spin_unlock(&device->io_lock);
 402
 403         while (pending) {
 404
 405                 rmb();
 406                 /* we want to work on both lists, but do more bios on the
 407                  * sync list than the regular list
 408                  */
 409                 if ((num_run > 32 &&
 410                     pending_bios != &device->pending_sync_bios &&
 411                     device->pending_sync_bios.head) ||
 412                    (num_run > 64 && pending_bios == &device->pending_sync_bios &&
 413                     device->pending_bios.head)) {
 414                         spin_lock(&device->io_lock);
 415                         requeue_list(pending_bios, pending, tail);
 416                         goto loop_lock;
 417                 }
 418
 419                 cur = pending;
 420                 pending = pending->bi_next;
 421                 cur->bi_next = NULL;
 422
 423                 /*
 424                  * atomic_dec_return implies a barrier for waitqueue_active
 425                  */
 426                 if (atomic_dec_return(&fs_info->nr_async_bios) < limit &&
 427                     waitqueue_active(&fs_info->async_submit_wait))
 428                         wake_up(&fs_info->async_submit_wait);
 429
 430                 BUG_ON(atomic_read(&cur->__bi_cnt) == 0);
 431
 432                 /*
 433                  * if we're doing the sync list, record that our
 434                  * plug has some sync requests on it
 435                  *
 436                  * If we're doing the regular list and there are
 437                  * sync requests sitting around, unplug before
 438                  * we add more
 439                  */
 440                 if (pending_bios == &device->pending_sync_bios) {
 441                         sync_pending = 1;
 442                 } else if (sync_pending) {
 443                         blk_finish_plug(&plug);
 444                         blk_start_plug(&plug);
 445                         sync_pending = 0;
 446                 }
 447
 448                 btrfsic_submit_bio(cur->bi_rw, cur);
 449                 num_run++;
 450                 batch_run++;
 451
 452                 cond_resched();
 453
 454                 /*
 455                  * we made progress, there is more work to do and the bdi
 456                  * is now congested.  Back off and let other work structs
 457                  * run instead
 458                  */
 459                 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
 460                     fs_info->fs_devices->open_devices > 1) {
 461                         struct io_context *ioc;
 462
 463                         ioc = current->io_context;
 464
 465                         /*
 466                          * the main goal here is that we don't want to
 467                          * block if we're going to be able to submit
 468                          * more requests without blocking.
 469                          *
 470                          * This code does two great things, it pokes into
 471                          * the elevator code from a filesystem _and_
 472                          * it makes assumptions about how batching works.
 473                          */
 474                         if (ioc && ioc->nr_batch_requests > 0 &&
 475                             time_before(jiffies, ioc->last_waited + HZ/50UL) &&
 476                             (last_waited == 0 ||
 477                              ioc->last_waited == last_waited)) {
 478                                 /*
 479                                  * we want to go through our batch of
 480                                  * requests and stop.  So, we copy out
 481                                  * the ioc->last_waited time and test
 482                                  * against it before looping
 483                                  */
 484                                 last_waited = ioc->last_waited;
 485                                 cond_resched();
 486                                 continue;
 487                         }
 488                         spin_lock(&device->io_lock);
 489                         requeue_list(pending_bios, pending, tail);
 490                         device->running_pending = 1;
 491
 492                         spin_unlock(&device->io_lock);
 493                         btrfs_queue_work(fs_info->submit_workers,
 494                                          &device->work);
 495                         goto done;
 496                 }
 497                 /* unplug every 64 requests just for good measure */
 498                 if (batch_run % 64 == 0) {
 499                         blk_finish_plug(&plug);
 500                         blk_start_plug(&plug);
 501                         sync_pending = 0;
 502                 }
 503         }
 504
 505         cond_resched();
 506         if (again)
 507                 goto loop;
 508
 509         spin_lock(&device->io_lock);
 510         if (device->pending_bios.head || device->pending_sync_bios.head)
 511                 goto loop_lock;
 512         spin_unlock(&device->io_lock);
 513
 514 done:
 515         blk_finish_plug(&plug);
 516 }
 517
 518 static void pending_bios_fn(struct btrfs_work *work)
 519 {
 520         struct btrfs_device *device;
 521
 522         device = container_of(work, struct btrfs_device, work);
 523         run_scheduled_bios(device);
 524 }
 525
 526
 527 void btrfs_free_stale_device(struct btrfs_device *cur_dev)
 528 {
 529         struct btrfs_fs_devices *fs_devs;
 530         struct btrfs_device *dev;
 531
 532         if (!cur_dev->name)
 533                 return;
 534
 535         list_for_each_entry(fs_devs, &fs_uuids, list) {
 536                 int del = 1;
 537
 538                 if (fs_devs->opened)
 539                         continue;
 540                 if (fs_devs->seeding)
 541                         continue;
 542
 543                 list_for_each_entry(dev, &fs_devs->devices, dev_list) {
 544
 545                         if (dev == cur_dev)
 546                                 continue;
 547                         if (!dev->name)
 548                                 continue;
 549
 550                         /*
 551                          * Todo: This won't be enough. What if the same device
 552                          * comes back (with new uuid and) with its mapper path?
 553                          * But for now, this does help as mostly an admin will
 554                          * either use mapper or non mapper path throughout.
 555                          */
 556                         rcu_read_lock();
 557                         del = strcmp(rcu_str_deref(dev->name),
 558                                                 rcu_str_deref(cur_dev->name));
 559                         rcu_read_unlock();
 560                         if (!del)
 561                                 break;
 562                 }
 563
 564                 if (!del) {
 565                         /* delete the stale device */
 566                         if (fs_devs->num_devices == 1) {
 567                                 btrfs_sysfs_remove_fsid(fs_devs);
 568                                 list_del(&fs_devs->list);
 569                                 free_fs_devices(fs_devs);
 570                         } else {
 571                                 fs_devs->num_devices--;
 572                                 list_del(&dev->dev_list);
 573                                 rcu_string_free(dev->name);
 574                                 kfree(dev);
 575                         }
 576                         break;
 577                 }
 578         }
 579 }
 580
 581 /*
 582  * Add new device to list of registered devices
 583  *
 584  * Returns:
 585  * 1   - first time device is seen
 586  * 0   - device already known
 587  * < 0 - error
 588  */
 589 static noinline int device_list_add(const char *path,
 590                            struct btrfs_super_block *disk_super,
 591                            u64 devid, struct btrfs_fs_devices **fs_devices_ret)
 592 {
 593         struct btrfs_device *device;
 594         struct btrfs_fs_devices *fs_devices;
 595         struct rcu_string *name;
 596         int ret = 0;
 597         u64 found_transid = btrfs_super_generation(disk_super);
 598
 599         fs_devices = find_fsid(disk_super->fsid);
 600         if (!fs_devices) {
 601                 fs_devices = alloc_fs_devices(disk_super->fsid);
 602                 if (IS_ERR(fs_devices))
 603                         return PTR_ERR(fs_devices);
 604
 605                 list_add(&fs_devices->list, &fs_uuids);
 606
 607                 device = NULL;
 608         } else {
 609                 device = __find_device(&fs_devices->devices, devid,
 610                                        disk_super->dev_item.uuid);
 611         }
 612
 613         if (!device) {
 614                 if (fs_devices->opened)
 615                         return -EBUSY;
 616
 617                 device = btrfs_alloc_device(NULL, &devid,
 618                                             disk_super->dev_item.uuid);
 619                 if (IS_ERR(device)) {
 620                         /* we can safely leave the fs_devices entry around */
 621                         return PTR_ERR(device);
 622                 }
 623
 624                 name = rcu_string_strdup(path, GFP_NOFS);
 625                 if (!name) {
 626                         kfree(device);
 627                         return -ENOMEM;
 628                 }
 629                 rcu_assign_pointer(device->name, name);
 630
 631                 mutex_lock(&fs_devices->device_list_mutex);
 632                 list_add_rcu(&device->dev_list, &fs_devices->devices);
 633                 fs_devices->num_devices++;
 634                 mutex_unlock(&fs_devices->device_list_mutex);
 635
 636                 ret = 1;
 637                 device->fs_devices = fs_devices;
 638         } else if (!device->name || strcmp(device->name->str, path)) {
 639                 /*
 640                  * When FS is already mounted.
 641                  * 1. If you are here and if the device->name is NULL that
 642                  *    means this device was missing at time of FS mount.
 643                  * 2. If you are here and if the device->name is different
 644                  *    from 'path' that means either
 645                  *      a. The same device disappeared and reappeared with
 646                  *         different name. or
 647                  *      b. The missing-disk-which-was-replaced, has
 648                  *         reappeared now.
 649                  *
 650                  * We must allow 1 and 2a above. But 2b would be a spurious
 651                  * and unintentional.
 652                  *
 653                  * Further in case of 1 and 2a above, the disk at 'path'
 654                  * would have missed some transaction when it was away and
 655                  * in case of 2a the stale bdev has to be updated as well.
 656                  * 2b must not be allowed at all time.
 657                  */
 658
 659                 /*
 660                  * For now, we do allow update to btrfs_fs_device through the
 661                  * btrfs dev scan cli after FS has been mounted.  We're still
 662                  * tracking a problem where systems fail mount by subvolume id
 663                  * when we reject replacement on a mounted FS.
 664                  */
 665                 if (!fs_devices->opened && found_transid < device->generation) {
 666                         /*
 667                          * That is if the FS is _not_ mounted and if you
 668                          * are here, that means there is more than one
 669                          * disk with same uuid and devid.We keep the one
 670                          * with larger generation number or the last-in if
 671                          * generation are equal.
 672                          */
 673                         return -EEXIST;
 674                 }
 675
 676                 name = rcu_string_strdup(path, GFP_NOFS);
 677                 if (!name)
 678                         return -ENOMEM;
 679                 rcu_string_free(device->name);
 680                 rcu_assign_pointer(device->name, name);
 681                 if (device->missing) {
 682                         fs_devices->missing_devices--;
 683                         device->missing = 0;
 684                 }
 685         }
 686
 687         /*
 688          * Unmount does not free the btrfs_device struct but would zero
 689          * generation along with most of the other members. So just update
 690          * it back. We need it to pick the disk with largest generation
 691          * (as above).
 692          */
 693         if (!fs_devices->opened)
 694                 device->generation = found_transid;
 695
 696         /*
 697          * if there is new btrfs on an already registered device,
 698          * then remove the stale device entry.
 699          */
 700         btrfs_free_stale_device(device);
 701
 702         *fs_devices_ret = fs_devices;
 703
 704         return ret;
 705 }
 706
 707 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
 708 {
 709         struct btrfs_fs_devices *fs_devices;
 710         struct btrfs_device *device;
 711         struct btrfs_device *orig_dev;
 712
 713         fs_devices = alloc_fs_devices(orig->fsid);
 714         if (IS_ERR(fs_devices))
 715                 return fs_devices;
 716
 717         mutex_lock(&orig->device_list_mutex);
 718         fs_devices->total_devices = orig->total_devices;
 719
 720         /* We have held the volume lock, it is safe to get the devices. */
 721         list_for_each_entry(orig_dev, &orig->devices, dev_list) {
 722                 struct rcu_string *name;
 723
 724                 device = btrfs_alloc_device(NULL, &orig_dev->devid,
 725                                             orig_dev->uuid);
 726                 if (IS_ERR(device))
 727                         goto error;
 728
 729                 /*
 730                  * This is ok to do without rcu read locked because we hold the
 731                  * uuid mutex so nothing we touch in here is going to disappear.
 732                  */
 733                 if (orig_dev->name) {
 734                         name = rcu_string_strdup(orig_dev->name->str, GFP_NOFS);
 735                         if (!name) {
 736                                 kfree(device);
 737                                 goto error;
 738                         }
 739                         rcu_assign_pointer(device->name, name);
 740                 }
 741
 742                 list_add(&device->dev_list, &fs_devices->devices);
 743                 device->fs_devices = fs_devices;
 744                 fs_devices->num_devices++;
 745         }
 746         mutex_unlock(&orig->device_list_mutex);
 747         return fs_devices;
 748 error:
 749         mutex_unlock(&orig->device_list_mutex);
 750         free_fs_devices(fs_devices);
 751         return ERR_PTR(-ENOMEM);
 752 }
 753
 754 void btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices, int step)
 755 {
 756         struct btrfs_device *device, *next;
 757         struct btrfs_device *latest_dev = NULL;
 758
 759         mutex_lock(&uuid_mutex);
 760 again:
 761         /* This is the initialized path, it is safe to release the devices. */
 762         list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
 763                 if (device->in_fs_metadata) {
 764                         if (!device->is_tgtdev_for_dev_replace &&
 765                             (!latest_dev ||
 766                              device->generation > latest_dev->generation)) {
 767                                 latest_dev = device;
 768                         }
 769                         continue;
 770                 }
 771
 772                 if (device->devid == BTRFS_DEV_REPLACE_DEVID) {
 773                         /*
 774                          * In the first step, keep the device which has
 775                          * the correct fsid and the devid that is used
 776                          * for the dev_replace procedure.
 777                          * In the second step, the dev_replace state is
 778                          * read from the device tree and it is known
 779                          * whether the procedure is really active or
 780                          * not, which means whether this device is
 781                          * used or whether it should be removed.
 782                          */
 783                         if (step == 0 || device->is_tgtdev_for_dev_replace) {
 784                                 continue;
 785                         }
 786                 }
 787                 if (device->bdev) {
 788                         blkdev_put(device->bdev, device->mode);
 789                         device->bdev = NULL;
 790                         fs_devices->open_devices--;
 791                 }
 792                 if (device->writeable) {
 793                         list_del_init(&device->dev_alloc_list);
 794                         device->writeable = 0;
 795                         if (!device->is_tgtdev_for_dev_replace)
 796                                 fs_devices->rw_devices--;
 797                 }
 798                 list_del_init(&device->dev_list);
 799                 fs_devices->num_devices--;
 800                 rcu_string_free(device->name);
 801                 kfree(device);
 802         }
 803
 804         if (fs_devices->seed) {
 805                 fs_devices = fs_devices->seed;
 806                 goto again;
 807         }
 808
 809         fs_devices->latest_bdev = latest_dev->bdev;
 810
 811         mutex_unlock(&uuid_mutex);
 812 }
 813
 814 static void __free_device(struct work_struct *work)
 815 {
 816         struct btrfs_device *device;
 817
 818         device = container_of(work, struct btrfs_device, rcu_work);
 819
 820         if (device->bdev)
 821                 blkdev_put(device->bdev, device->mode);
 822
 823         rcu_string_free(device->name);
 824         kfree(device);
 825 }
 826
 827 static void free_device(struct rcu_head *head)
 828 {
 829         struct btrfs_device *device;
 830
 831         device = container_of(head, struct btrfs_device, rcu);
 832
 833         INIT_WORK(&device->rcu_work, __free_device);
 834         schedule_work(&device->rcu_work);
 835 }
 836
 837 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
 838 {
 839         struct btrfs_device *device, *tmp;
 840
 841         if (--fs_devices->opened > 0)
 842                 return 0;
 843
 844         mutex_lock(&fs_devices->device_list_mutex);
 845         list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list) {
 846                 btrfs_close_one_device(device);
 847         }
 848         mutex_unlock(&fs_devices->device_list_mutex);
 849
 850         WARN_ON(fs_devices->open_devices);
 851         WARN_ON(fs_devices->rw_devices);
 852         fs_devices->opened = 0;
 853         fs_devices->seeding = 0;
 854
 855         return 0;
 856 }
 857
 858 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
 859 {
 860         struct btrfs_fs_devices *seed_devices = NULL;
 861         int ret;
 862
 863         mutex_lock(&uuid_mutex);
 864         ret = __btrfs_close_devices(fs_devices);
 865         if (!fs_devices->opened) {
 866                 seed_devices = fs_devices->seed;
 867                 fs_devices->seed = NULL;
 868         }
 869         mutex_unlock(&uuid_mutex);
 870
 871         while (seed_devices) {
 872                 fs_devices = seed_devices;
 873                 seed_devices = fs_devices->seed;
 874                 __btrfs_close_devices(fs_devices);
 875                 free_fs_devices(fs_devices);
 876         }
 877         /*
 878          * Wait for rcu kworkers under __btrfs_close_devices
 879          * to finish all blkdev_puts so device is really
 880          * free when umount is done.
 881          */
 882         rcu_barrier();
 883         return ret;
 884 }
 885
 886 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
 887                                 fmode_t flags, void *holder)
 888 {
 889         struct request_queue *q;
 890         struct block_device *bdev;
 891         struct list_head *head = &fs_devices->devices;
 892         struct btrfs_device *device;
 893         struct btrfs_device *latest_dev = NULL;
 894         struct buffer_head *bh;
 895         struct btrfs_super_block *disk_super;
 896         u64 devid;
 897         int seeding = 1;
 898         int ret = 0;
 899
 900         flags |= FMODE_EXCL;
 901
 902         list_for_each_entry(device, head, dev_list) {
 903                 if (device->bdev)
 904                         continue;
 905                 if (!device->name)
 906                         continue;
 907
 908                 /* Just open everything we can; ignore failures here */
 909                 if (btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1,
 910                                             &bdev, &bh))
 911                         continue;
 912
 913                 disk_super = (struct btrfs_super_block *)bh->b_data;
 914                 devid = btrfs_stack_device_id(&disk_super->dev_item);
 915                 if (devid != device->devid)
 916                         goto error_brelse;
 917
 918                 if (memcmp(device->uuid, disk_super->dev_item.uuid,
 919                            BTRFS_UUID_SIZE))
 920                         goto error_brelse;
 921
 922                 device->generation = btrfs_super_generation(disk_super);
 923                 if (!latest_dev ||
 924                     device->generation > latest_dev->generation)
 925                         latest_dev = device;
 926
 927                 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
 928                         device->writeable = 0;
 929                 } else {
 930                         device->writeable = !bdev_read_only(bdev);
 931                         seeding = 0;
 932                 }
 933
 934                 q = bdev_get_queue(bdev);
 935                 if (blk_queue_discard(q))
 936                         device->can_discard = 1;
 937
 938                 device->bdev = bdev;
 939                 device->in_fs_metadata = 0;
 940                 device->mode = flags;
 941
 942                 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
 943                         fs_devices->rotating = 1;
 944
 945                 fs_devices->open_devices++;
 946                 if (device->writeable &&
 947                     device->devid != BTRFS_DEV_REPLACE_DEVID) {
 948                         fs_devices->rw_devices++;
 949                         list_add(&device->dev_alloc_list,
 950                                  &fs_devices->alloc_list);
 951                 }
 952                 brelse(bh);
 953                 continue;
 954
 955 error_brelse:
 956                 brelse(bh);
 957                 blkdev_put(bdev, flags);
 958                 continue;
 959         }
 960         if (fs_devices->open_devices == 0) {
 961                 ret = -EINVAL;
 962                 goto out;
 963         }
 964         fs_devices->seeding = seeding;
 965         fs_devices->opened = 1;
 966         fs_devices->latest_bdev = latest_dev->bdev;
 967         fs_devices->total_rw_bytes = 0;
 968 out:
 969         return ret;
 970 }
 971
 972 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
 973                        fmode_t flags, void *holder)
 974 {
 975         int ret;
 976
 977         mutex_lock(&uuid_mutex);
 978         if (fs_devices->opened) {
 979                 fs_devices->opened++;
 980                 ret = 0;
 981         } else {
 982                 ret = __btrfs_open_devices(fs_devices, flags, holder);
 983         }
 984         mutex_unlock(&uuid_mutex);
 985         return ret;
 986 }
 987
 988 /*
 989  * Look for a btrfs signature on a device. This may be called out of the mount path
 990  * and we are not allowed to call set_blocksize during the scan. The superblock
 991  * is read via pagecache
 992  */
 993 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
 994                           struct btrfs_fs_devices **fs_devices_ret)
 995 {
 996         struct btrfs_super_block *disk_super;
 997         struct block_device *bdev;
 998         struct page *page;
 999         void *p;
1000         int ret = -EINVAL;
1001         u64 devid;
1002         u64 transid;
1003         u64 total_devices;
1004         u64 bytenr;
1005         pgoff_t index;
1006
1007         /*
1008          * we would like to check all the supers, but that would make
1009          * a btrfs mount succeed after a mkfs from a different FS.
1010          * So, we need to add a special mount option to scan for
1011          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
1012          */
1013         bytenr = btrfs_sb_offset(0);
1014         flags |= FMODE_EXCL;
1015         mutex_lock(&uuid_mutex);
1016
1017         bdev = blkdev_get_by_path(path, flags, holder);
1018
1019         if (IS_ERR(bdev)) {
1020                 ret = PTR_ERR(bdev);
1021                 goto error;
1022         }
1023
1024         /* make sure our super fits in the device */
1025         if (bytenr + PAGE_CACHE_SIZE >= i_size_read(bdev->bd_inode))
1026                 goto error_bdev_put;
1027
1028         /* make sure our super fits in the page */
1029         if (sizeof(*disk_super) > PAGE_CACHE_SIZE)
1030                 goto error_bdev_put;
1031
1032         /* make sure our super doesn't straddle pages on disk */
1033         index = bytenr >> PAGE_CACHE_SHIFT;
1034         if ((bytenr + sizeof(*disk_super) - 1) >> PAGE_CACHE_SHIFT != index)
1035                 goto error_bdev_put;
1036
1037         /* pull in the page with our super */
1038         page = read_cache_page_gfp(bdev->bd_inode->i_mapping,
1039                                    index, GFP_NOFS);
1040
1041         if (IS_ERR_OR_NULL(page))
1042                 goto error_bdev_put;
1043
1044         p = kmap(page);
1045
1046         /* align our pointer to the offset of the super block */
1047         disk_super = p + (bytenr & ~PAGE_CACHE_MASK);
1048
1049         if (btrfs_super_bytenr(disk_super) != bytenr ||
1050             btrfs_super_magic(disk_super) != BTRFS_MAGIC)
1051                 goto error_unmap;
1052
1053         devid = btrfs_stack_device_id(&disk_super->dev_item);
1054         transid = btrfs_super_generation(disk_super);
1055         total_devices = btrfs_super_num_devices(disk_super);
1056
1057         ret = device_list_add(path, disk_super, devid, fs_devices_ret);
1058         if (ret > 0) {
1059                 if (disk_super->label[0]) {
1060                         if (disk_super->label[BTRFS_LABEL_SIZE - 1])
1061                                 disk_super->label[BTRFS_LABEL_SIZE - 1] = '\0';
1062                         printk(KERN_INFO "BTRFS: device label %s ", disk_super->label);
1063                 } else {
1064                         printk(KERN_INFO "BTRFS: device fsid %pU ", disk_super->fsid);
1065                 }
1066
1067                 printk(KERN_CONT "devid %llu transid %llu %s\n", devid, transid, path);
1068                 ret = 0;
1069         }
1070         if (!ret && fs_devices_ret)
1071                 (*fs_devices_ret)->total_devices = total_devices;
1072
1073 error_unmap:
1074         kunmap(page);
1075         page_cache_release(page);
1076
1077 error_bdev_put:
1078         blkdev_put(bdev, flags);
1079 error:
1080         mutex_unlock(&uuid_mutex);
1081         return ret;
1082 }
1083
1084 /* helper to account the used device space in the range */
1085 int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
1086                                    u64 end, u64 *length)
1087 {
1088         struct btrfs_key key;
1089         struct btrfs_root *root = device->dev_root;
1090         struct btrfs_dev_extent *dev_extent;
1091         struct btrfs_path *path;
1092         u64 extent_end;
1093         int ret;
1094         int slot;
1095         struct extent_buffer *l;
1096
1097         *length = 0;
1098
1099         if (start >= device->total_bytes || device->is_tgtdev_for_dev_replace)
1100                 return 0;
1101
1102         path = btrfs_alloc_path();
1103         if (!path)
1104                 return -ENOMEM;
1105         path->reada = 2;
1106
1107         key.objectid = device->devid;
1108         key.offset = start;
1109         key.type = BTRFS_DEV_EXTENT_KEY;
1110
1111         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1112         if (ret < 0)
1113                 goto out;
1114         if (ret > 0) {
1115                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1116                 if (ret < 0)
1117                         goto out;
1118         }
1119
1120         while (1) {
1121                 l = path->nodes[0];
1122                 slot = path->slots[0];
1123                 if (slot >= btrfs_header_nritems(l)) {
1124                         ret = btrfs_next_leaf(root, path);
1125                         if (ret == 0)
1126                                 continue;
1127                         if (ret < 0)
1128                                 goto out;
1129
1130                         break;
1131                 }
1132                 btrfs_item_key_to_cpu(l, &key, slot);
1133
1134                 if (key.objectid < device->devid)
1135                         goto next;
1136
1137                 if (key.objectid > device->devid)
1138                         break;
1139
1140                 if (key.type != BTRFS_DEV_EXTENT_KEY)
1141                         goto next;
1142
1143                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1144                 extent_end = key.offset + btrfs_dev_extent_length(l,
1145                                                                   dev_extent);
1146                 if (key.offset <= start && extent_end > end) {
1147                         *length = end - start + 1;
1148                         break;
1149                 } else if (key.offset <= start && extent_end > start)
1150                         *length += extent_end - start;
1151                 else if (key.offset > start && extent_end <= end)
1152                         *length += extent_end - key.offset;
1153                 else if (key.offset > start && key.offset <= end) {
1154                         *length += end - key.offset + 1;
1155                         break;
1156                 } else if (key.offset > end)
1157                         break;
1158
1159 next:
1160                 path->slots[0]++;
1161         }
1162         ret = 0;
1163 out:
1164         btrfs_free_path(path);
1165         return ret;
1166 }
1167
1168 static int contains_pending_extent(struct btrfs_transaction *transaction,
1169                                    struct btrfs_device *device,
1170                                    u64 *start, u64 len)
1171 {
1172         struct btrfs_fs_info *fs_info = device->dev_root->fs_info;
1173         struct extent_map *em;
1174         struct list_head *search_list = &fs_info->pinned_chunks;
1175         int ret = 0;
1176         u64 physical_start = *start;
1177
1178         if (transaction)
1179                 search_list = &transaction->pending_chunks;
1180 again:
1181         list_for_each_entry(em, search_list, list) {
1182                 struct map_lookup *map;
1183                 int i;
1184
1185                 map = (struct map_lookup *)em->bdev;
1186                 for (i = 0; i < map->num_stripes; i++) {
1187                         u64 end;
1188
1189                         if (map->stripes[i].dev != device)
1190                                 continue;
1191                         if (map->stripes[i].physical >= physical_start + len ||
1192                             map->stripes[i].physical + em->orig_block_len <=
1193                             physical_start)
1194                                 continue;
1195                         /*
1196                          * Make sure that while processing the pinned list we do
1197                          * not override our *start with a lower value, because
1198                          * we can have pinned chunks that fall within this
1199                          * device hole and that have lower physical addresses
1200                          * than the pending chunks we processed before. If we
1201                          * do not take this special care we can end up getting
1202                          * 2 pending chunks that start at the same physical
1203                          * device offsets because the end offset of a pinned
1204                          * chunk can be equal to the start offset of some
1205                          * pending chunk.
1206                          */
1207                         end = map->stripes[i].physical + em->orig_block_len;
1208                         if (end > *start) {
1209                                 *start = end;
1210                                 ret = 1;
1211                         }
1212                 }
1213         }
1214         if (search_list != &fs_info->pinned_chunks) {
1215                 search_list = &fs_info->pinned_chunks;
1216                 goto again;
1217         }
1218
1219         return ret;
1220 }
1221
1222
1223 /*
1224  * find_free_dev_extent_start - find free space in the specified device
1225  * @device:       the device which we search the free space in
1226  * @num_bytes:    the size of the free space that we need
1227  * @search_start: the position from which to begin the search
1228  * @start:        store the start of the free space.
1229  * @len:          the size of the free space. that we find, or the size
1230  *                of the max free space if we don't find suitable free space
1231  *
1232  * this uses a pretty simple search, the expectation is that it is
1233  * called very infrequently and that a given device has a small number
1234  * of extents
1235  *
1236  * @start is used to store the start of the free space if we find. But if we
1237  * don't find suitable free space, it will be used to store the start position
1238  * of the max free space.
1239  *
1240  * @len is used to store the size of the free space that we find.
1241  * But if we don't find suitable free space, it is used to store the size of
1242  * the max free space.
1243  */
1244 int find_free_dev_extent_start(struct btrfs_transaction *transaction,
1245                                struct btrfs_device *device, u64 num_bytes,
1246                                u64 search_start, u64 *start, u64 *len)
1247 {
1248         struct btrfs_key key;
1249         struct btrfs_root *root = device->dev_root;
1250         struct btrfs_dev_extent *dev_extent;
1251         struct btrfs_path *path;
1252         u64 hole_size;
1253         u64 max_hole_start;
1254         u64 max_hole_size;
1255         u64 extent_end;
1256         u64 search_end = device->total_bytes;
1257         int ret;
1258         int slot;
1259         struct extent_buffer *l;
1260
1261         path = btrfs_alloc_path();
1262         if (!path)
1263                 return -ENOMEM;
1264
1265         max_hole_start = search_start;
1266         max_hole_size = 0;
1267
1268 again:
1269         if (search_start >= search_end || device->is_tgtdev_for_dev_replace) {
1270                 ret = -ENOSPC;
1271                 goto out;
1272         }
1273
1274         path->reada = 2;
1275         path->search_commit_root = 1;
1276         path->skip_locking = 1;
1277
1278         key.objectid = device->devid;
1279         key.offset = search_start;
1280         key.type = BTRFS_DEV_EXTENT_KEY;
1281
1282         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1283         if (ret < 0)
1284                 goto out;
1285         if (ret > 0) {
1286                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1287                 if (ret < 0)
1288                         goto out;
1289         }
1290
1291         while (1) {
1292                 l = path->nodes[0];
1293                 slot = path->slots[0];
1294                 if (slot >= btrfs_header_nritems(l)) {
1295                         ret = btrfs_next_leaf(root, path);
1296                         if (ret == 0)
1297                                 continue;
1298                         if (ret < 0)
1299                                 goto out;
1300
1301                         break;
1302                 }
1303                 btrfs_item_key_to_cpu(l, &key, slot);
1304
1305                 if (key.objectid < device->devid)
1306                         goto next;
1307
1308                 if (key.objectid > device->devid)
1309                         break;
1310
1311                 if (key.type != BTRFS_DEV_EXTENT_KEY)
1312                         goto next;
1313
1314                 if (key.offset > search_start) {
1315                         hole_size = key.offset - search_start;
1316
1317                         /*
1318                          * Have to check before we set max_hole_start, otherwise
1319                          * we could end up sending back this offset anyway.
1320                          */
1321                         if (contains_pending_extent(transaction, device,
1322                                                     &search_start,
1323                                                     hole_size)) {
1324                                 if (key.offset >= search_start) {
1325                                         hole_size = key.offset - search_start;
1326                                 } else {
1327                                         WARN_ON_ONCE(1);
1328                                         hole_size = 0;
1329                                 }
1330                         }
1331
1332                         if (hole_size > max_hole_size) {
1333                                 max_hole_start = search_start;
1334                                 max_hole_size = hole_size;
1335                         }
1336
1337                         /*
1338                          * If this free space is greater than which we need,
1339                          * it must be the max free space that we have found
1340                          * until now, so max_hole_start must point to the start
1341                          * of this free space and the length of this free space
1342                          * is stored in max_hole_size. Thus, we return
1343                          * max_hole_start and max_hole_size and go back to the
1344                          * caller.
1345                          */
1346                         if (hole_size >= num_bytes) {
1347                                 ret = 0;
1348                                 goto out;
1349                         }
1350                 }
1351
1352                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1353                 extent_end = key.offset + btrfs_dev_extent_length(l,
1354                                                                   dev_extent);
1355                 if (extent_end > search_start)
1356                         search_start = extent_end;
1357 next:
1358                 path->slots[0]++;
1359                 cond_resched();
1360         }
1361
1362         /*
1363          * At this point, search_start should be the end of
1364          * allocated dev extents, and when shrinking the device,
1365          * search_end may be smaller than search_start.
1366          */
1367         if (search_end > search_start) {
1368                 hole_size = search_end - search_start;
1369
1370                 if (contains_pending_extent(transaction, device, &search_start,
1371                                             hole_size)) {
1372                         btrfs_release_path(path);
1373                         goto again;
1374                 }
1375
1376                 if (hole_size > max_hole_size) {
1377                         max_hole_start = search_start;
1378                         max_hole_size = hole_size;
1379                 }
1380         }
1381
1382         /* See above. */
1383         if (max_hole_size < num_bytes)
1384                 ret = -ENOSPC;
1385         else
1386                 ret = 0;
1387
1388 out:
1389         btrfs_free_path(path);
1390         *start = max_hole_start;
1391         if (len)
1392                 *len = max_hole_size;
1393         return ret;
1394 }
1395
1396 int find_free_dev_extent(struct btrfs_trans_handle *trans,
1397                          struct btrfs_device *device, u64 num_bytes,
1398                          u64 *start, u64 *len)
1399 {
1400         struct btrfs_root *root = device->dev_root;
1401         u64 search_start;
1402
1403         /* FIXME use last free of some kind */
1404
1405         /*
1406          * we don't want to overwrite the superblock on the drive,
1407          * so we make sure to start at an offset of at least 1MB
1408          */
1409         search_start = max(root->fs_info->alloc_start, 1024ull * 1024);
1410         return find_free_dev_extent_start(trans->transaction, device,
1411                                           num_bytes, search_start, start, len);
1412 }
1413
1414 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1415                           struct btrfs_device *device,
1416                           u64 start, u64 *dev_extent_len)
1417 {
1418         int ret;
1419         struct btrfs_path *path;
1420         struct btrfs_root *root = device->dev_root;
1421         struct btrfs_key key;
1422         struct btrfs_key found_key;
1423         struct extent_buffer *leaf = NULL;
1424         struct btrfs_dev_extent *extent = NULL;
1425
1426         path = btrfs_alloc_path();
1427         if (!path)
1428                 return -ENOMEM;
1429
1430         key.objectid = device->devid;
1431         key.offset = start;
1432         key.type = BTRFS_DEV_EXTENT_KEY;
1433 again:
1434         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1435         if (ret > 0) {
1436                 ret = btrfs_previous_item(root, path, key.objectid,
1437                                           BTRFS_DEV_EXTENT_KEY);
1438                 if (ret)
1439                         goto out;
1440                 leaf = path->nodes[0];
1441                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1442                 extent = btrfs_item_ptr(leaf, path->slots[0],
1443                                         struct btrfs_dev_extent);
1444                 BUG_ON(found_key.offset > start || found_key.offset +
1445                        btrfs_dev_extent_length(leaf, extent) < start);
1446                 key = found_key;
1447                 btrfs_release_path(path);
1448                 goto again;
1449         } else if (ret == 0) {
1450                 leaf = path->nodes[0];
1451                 extent = btrfs_item_ptr(leaf, path->slots[0],
1452                                         struct btrfs_dev_extent);
1453         } else {
1454                 btrfs_std_error(root->fs_info, ret, "Slot search failed");
1455                 goto out;
1456         }
1457
1458         *dev_extent_len = btrfs_dev_extent_length(leaf, extent);
1459
1460         ret = btrfs_del_item(trans, root, path);
1461         if (ret) {
1462                 btrfs_std_error(root->fs_info, ret,
1463                             "Failed to remove dev extent item");
1464         } else {
1465                 trans->transaction->have_free_bgs = 1;
1466         }
1467 out:
1468         btrfs_free_path(path);
1469         return ret;
1470 }
1471
1472 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1473                                   struct btrfs_device *device,
1474                                   u64 chunk_tree, u64 chunk_objectid,
1475                                   u64 chunk_offset, u64 start, u64 num_bytes)
1476 {
1477         int ret;
1478         struct btrfs_path *path;
1479         struct btrfs_root *root = device->dev_root;
1480         struct btrfs_dev_extent *extent;
1481         struct extent_buffer *leaf;
1482         struct btrfs_key key;
1483
1484         WARN_ON(!device->in_fs_metadata);
1485         WARN_ON(device->is_tgtdev_for_dev_replace);
1486         path = btrfs_alloc_path();
1487         if (!path)
1488                 return -ENOMEM;
1489
1490         key.objectid = device->devid;
1491         key.offset = start;
1492         key.type = BTRFS_DEV_EXTENT_KEY;
1493         ret = btrfs_insert_empty_item(trans, root, path, &key,
1494                                       sizeof(*extent));
1495         if (ret)
1496                 goto out;
1497
1498         leaf = path->nodes[0];
1499         extent = btrfs_item_ptr(leaf, path->slots[0],
1500                                 struct btrfs_dev_extent);
1501         btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
1502         btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
1503         btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1504
1505         write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
1506                     btrfs_dev_extent_chunk_tree_uuid(extent), BTRFS_UUID_SIZE);
1507
1508         btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1509         btrfs_mark_buffer_dirty(leaf);
1510 out:
1511         btrfs_free_path(path);
1512         return ret;
1513 }
1514
1515 static u64 find_next_chunk(struct btrfs_fs_info *fs_info)
1516 {
1517         struct extent_map_tree *em_tree;
1518         struct extent_map *em;
1519         struct rb_node *n;
1520         u64 ret = 0;
1521
1522         em_tree = &fs_info->mapping_tree.map_tree;
1523         read_lock(&em_tree->lock);
1524         n = rb_last(&em_tree->map);
1525         if (n) {
1526                 em = rb_entry(n, struct extent_map, rb_node);
1527                 ret = em->start + em->len;
1528         }
1529         read_unlock(&em_tree->lock);
1530
1531         return ret;
1532 }
1533
1534 static noinline int find_next_devid(struct btrfs_fs_info *fs_info,
1535                                     u64 *devid_ret)
1536 {
1537         int ret;
1538         struct btrfs_key key;
1539         struct btrfs_key found_key;
1540         struct btrfs_path *path;
1541
1542         path = btrfs_alloc_path();
1543         if (!path)
1544                 return -ENOMEM;
1545
1546         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1547         key.type = BTRFS_DEV_ITEM_KEY;
1548         key.offset = (u64)-1;
1549
1550         ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
1551         if (ret < 0)
1552                 goto error;
1553
1554         BUG_ON(ret == 0); /* Corruption */
1555
1556         ret = btrfs_previous_item(fs_info->chunk_root, path,
1557                                   BTRFS_DEV_ITEMS_OBJECTID,
1558                                   BTRFS_DEV_ITEM_KEY);
1559         if (ret) {
1560                 *devid_ret = 1;
1561         } else {
1562                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1563                                       path->slots[0]);
1564                 *devid_ret = found_key.offset + 1;
1565         }
1566         ret = 0;
1567 error:
1568         btrfs_free_path(path);
1569         return ret;
1570 }
1571
1572 /*
1573  * the device information is stored in the chunk root
1574  * the btrfs_device struct should be fully filled in
1575  */
1576 static int btrfs_add_device(struct btrfs_trans_handle *trans,
1577                             struct btrfs_root *root,
1578                             struct btrfs_device *device)
1579 {
1580         int ret;
1581         struct btrfs_path *path;
1582         struct btrfs_dev_item *dev_item;
1583         struct extent_buffer *leaf;
1584         struct btrfs_key key;
1585         unsigned long ptr;
1586
1587         root = root->fs_info->chunk_root;
1588
1589         path = btrfs_alloc_path();
1590         if (!path)
1591                 return -ENOMEM;
1592
1593         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1594         key.type = BTRFS_DEV_ITEM_KEY;
1595         key.offset = device->devid;
1596
1597         ret = btrfs_insert_empty_item(trans, root, path, &key,
1598                                       sizeof(*dev_item));
1599         if (ret)
1600                 goto out;
1601
1602         leaf = path->nodes[0];
1603         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1604
1605         btrfs_set_device_id(leaf, dev_item, device->devid);
1606         btrfs_set_device_generation(leaf, dev_item, 0);
1607         btrfs_set_device_type(leaf, dev_item, device->type);
1608         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1609         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1610         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1611         btrfs_set_device_total_bytes(leaf, dev_item,
1612                                      btrfs_device_get_disk_total_bytes(device));
1613         btrfs_set_device_bytes_used(leaf, dev_item,
1614                                     btrfs_device_get_bytes_used(device));
1615         btrfs_set_device_group(leaf, dev_item, 0);
1616         btrfs_set_device_seek_speed(leaf, dev_item, 0);
1617         btrfs_set_device_bandwidth(leaf, dev_item, 0);
1618         btrfs_set_device_start_offset(leaf, dev_item, 0);
1619
1620         ptr = btrfs_device_uuid(dev_item);
1621         write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1622         ptr = btrfs_device_fsid(dev_item);
1623         write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1624         btrfs_mark_buffer_dirty(leaf);
1625
1626         ret = 0;
1627 out:
1628         btrfs_free_path(path);
1629         return ret;
1630 }
1631
1632 /*
1633  * Function to update ctime/mtime for a given device path.
1634  * Mainly used for ctime/mtime based probe like libblkid.
1635  */
1636 static void update_dev_time(char *path_name)
1637 {
1638         struct file *filp;
1639
1640         filp = filp_open(path_name, O_RDWR, 0);
1641         if (IS_ERR(filp))
1642                 return;
1643         file_update_time(filp);
1644         filp_close(filp, NULL);
1645         return;
1646 }
1647
1648 static int btrfs_rm_dev_item(struct btrfs_root *root,
1649                              struct btrfs_device *device)
1650 {
1651         int ret;
1652         struct btrfs_path *path;
1653         struct btrfs_key key;
1654         struct btrfs_trans_handle *trans;
1655
1656         root = root->fs_info->chunk_root;
1657
1658         path = btrfs_alloc_path();
1659         if (!path)
1660                 return -ENOMEM;
1661
1662         trans = btrfs_start_transaction(root, 0);
1663         if (IS_ERR(trans)) {
1664                 btrfs_free_path(path);
1665                 return PTR_ERR(trans);
1666         }
1667         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1668         key.type = BTRFS_DEV_ITEM_KEY;
1669         key.offset = device->devid;
1670
1671         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1672         if (ret < 0)
1673                 goto out;
1674
1675         if (ret > 0) {
1676                 ret = -ENOENT;
1677                 goto out;
1678         }
1679
1680         ret = btrfs_del_item(trans, root, path);
1681         if (ret)
1682                 goto out;
1683 out:
1684         btrfs_free_path(path);
1685         btrfs_commit_transaction(trans, root);
1686         return ret;
1687 }
1688
1689 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
1690 {
1691         struct btrfs_device *device;
1692         struct btrfs_device *next_device;
1693         struct block_device *bdev;
1694         struct buffer_head *bh = NULL;
1695         struct btrfs_super_block *disk_super;
1696         struct btrfs_fs_devices *cur_devices;
1697         u64 all_avail;
1698         u64 devid;
1699         u64 num_devices;
1700         u8 *dev_uuid;
1701         unsigned seq;
1702         int ret = 0;
1703         bool clear_super = false;
1704
1705         mutex_lock(&uuid_mutex);
1706
1707         do {
1708                 seq = read_seqbegin(&root->fs_info->profiles_lock);
1709
1710                 all_avail = root->fs_info->avail_data_alloc_bits |
1711                             root->fs_info->avail_system_alloc_bits |
1712                             root->fs_info->avail_metadata_alloc_bits;
1713         } while (read_seqretry(&root->fs_info->profiles_lock, seq));
1714
1715         num_devices = root->fs_info->fs_devices->num_devices;
1716         btrfs_dev_replace_lock(&root->fs_info->dev_replace);
1717         if (btrfs_dev_replace_is_ongoing(&root->fs_info->dev_replace)) {
1718                 WARN_ON(num_devices < 1);
1719                 num_devices--;
1720         }
1721         btrfs_dev_replace_unlock(&root->fs_info->dev_replace);
1722
1723         if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) && num_devices <= 4) {
1724                 ret = BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET;
1725                 goto out;
1726         }
1727
1728         if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) && num_devices <= 2) {
1729                 ret = BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET;
1730                 goto out;
1731         }
1732
1733         if ((all_avail & BTRFS_BLOCK_GROUP_RAID5) &&
1734             root->fs_info->fs_devices->rw_devices <= 2) {
1735                 ret = BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET;
1736                 goto out;
1737         }
1738         if ((all_avail & BTRFS_BLOCK_GROUP_RAID6) &&
1739             root->fs_info->fs_devices->rw_devices <= 3) {
1740                 ret = BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET;
1741                 goto out;
1742         }
1743
1744         if (strcmp(device_path, "missing") == 0) {
1745                 struct list_head *devices;
1746                 struct btrfs_device *tmp;
1747
1748                 device = NULL;
1749                 devices = &root->fs_info->fs_devices->devices;
1750                 /*
1751                  * It is safe to read the devices since the volume_mutex
1752                  * is held.
1753                  */
1754                 list_for_each_entry(tmp, devices, dev_list) {
1755                         if (tmp->in_fs_metadata &&
1756                             !tmp->is_tgtdev_for_dev_replace &&
1757                             !tmp->bdev) {
1758                                 device = tmp;
1759                                 break;
1760                         }
1761                 }
1762                 bdev = NULL;
1763                 bh = NULL;
1764                 disk_super = NULL;
1765                 if (!device) {
1766                         ret = BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
1767                         goto out;
1768                 }
1769         } else {
1770                 ret = btrfs_get_bdev_and_sb(device_path,
1771                                             FMODE_WRITE | FMODE_EXCL,
1772                                             root->fs_info->bdev_holder, 0,
1773                                             &bdev, &bh);
1774                 if (ret)
1775                         goto out;
1776                 disk_super = (struct btrfs_super_block *)bh->b_data;
1777                 devid = btrfs_stack_device_id(&disk_super->dev_item);
1778                 dev_uuid = disk_super->dev_item.uuid;
1779                 device = btrfs_find_device(root->fs_info, devid, dev_uuid,
1780                                            disk_super->fsid);
1781                 if (!device) {
1782                         ret = -ENOENT;
1783                         goto error_brelse;
1784                 }
1785         }
1786
1787         if (device->is_tgtdev_for_dev_replace) {
1788                 ret = BTRFS_ERROR_DEV_TGT_REPLACE;
1789                 goto error_brelse;
1790         }
1791
1792         if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1793                 ret = BTRFS_ERROR_DEV_ONLY_WRITABLE;
1794                 goto error_brelse;
1795         }
1796
1797         if (device->writeable) {
1798                 lock_chunks(root);
1799                 list_del_init(&device->dev_alloc_list);
1800                 device->fs_devices->rw_devices--;
1801                 unlock_chunks(root);
1802                 clear_super = true;
1803         }
1804
1805         mutex_unlock(&uuid_mutex);
1806         ret = btrfs_shrink_device(device, 0);
1807         mutex_lock(&uuid_mutex);
1808         if (ret)
1809                 goto error_undo;
1810
1811         /*
1812          * TODO: the superblock still includes this device in its num_devices
1813          * counter although write_all_supers() is not locked out. This
1814          * could give a filesystem state which requires a degraded mount.
1815          */
1816         ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1817         if (ret)
1818                 goto error_undo;
1819
1820         device->in_fs_metadata = 0;
1821         btrfs_scrub_cancel_dev(root->fs_info, device);
1822
1823         /*
1824          * the device list mutex makes sure that we don't change
1825          * the device list while someone else is writing out all
1826          * the device supers. Whoever is writing all supers, should
1827          * lock the device list mutex before getting the number of
1828          * devices in the super block (super_copy). Conversely,
1829          * whoever updates the number of devices in the super block
1830          * (super_copy) should hold the device list mutex.
1831          */
1832
1833         cur_devices = device->fs_devices;
1834         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1835         list_del_rcu(&device->dev_list);
1836
1837         device->fs_devices->num_devices--;
1838         device->fs_devices->total_devices--;
1839
1840         if (device->missing)
1841                 device->fs_devices->missing_devices--;
1842
1843         next_device = list_entry(root->fs_info->fs_devices->devices.next,
1844                                  struct btrfs_device, dev_list);
1845         if (device->bdev == root->fs_info->sb->s_bdev)
1846                 root->fs_info->sb->s_bdev = next_device->bdev;
1847         if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1848                 root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1849
1850         if (device->bdev) {
1851                 device->fs_devices->open_devices--;
1852                 /* remove sysfs entry */
1853                 btrfs_sysfs_rm_device_link(root->fs_info->fs_devices, device);
1854         }
1855
1856         call_rcu(&device->rcu, free_device);
1857
1858         num_devices = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
1859         btrfs_set_super_num_devices(root->fs_info->super_copy, num_devices);
1860         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1861
1862         if (cur_devices->open_devices == 0) {
1863                 struct btrfs_fs_devices *fs_devices;
1864                 fs_devices = root->fs_info->fs_devices;
1865                 while (fs_devices) {
1866                         if (fs_devices->seed == cur_devices) {
1867                                 fs_devices->seed = cur_devices->seed;
1868                                 break;
1869                         }
1870                         fs_devices = fs_devices->seed;
1871                 }
1872                 cur_devices->seed = NULL;
1873                 __btrfs_close_devices(cur_devices);
1874                 free_fs_devices(cur_devices);
1875         }
1876
1877         root->fs_info->num_tolerated_disk_barrier_failures =
1878                 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
1879
1880         /*
1881          * at this point, the device is zero sized.  We want to
1882          * remove it from the devices list and zero out the old super
1883          */
1884         if (clear_super && disk_super) {
1885                 u64 bytenr;
1886                 int i;
1887
1888                 /* make sure this device isn't detected as part of
1889                  * the FS anymore
1890                  */
1891                 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1892                 set_buffer_dirty(bh);
1893                 sync_dirty_buffer(bh);
1894
1895                 /* clear the mirror copies of super block on the disk
1896                  * being removed, 0th copy is been taken care above and
1897                  * the below would take of the rest
1898                  */
1899                 for (i = 1; i < BTRFS_SUPER_MIRROR_MAX; i++) {
1900                         bytenr = btrfs_sb_offset(i);
1901                         if (bytenr + BTRFS_SUPER_INFO_SIZE >=
1902                                         i_size_read(bdev->bd_inode))
1903                                 break;
1904
1905                         brelse(bh);
1906                         bh = __bread(bdev, bytenr / 4096,
1907                                         BTRFS_SUPER_INFO_SIZE);
1908                         if (!bh)
1909                                 continue;
1910
1911                         disk_super = (struct btrfs_super_block *)bh->b_data;
1912
1913                         if (btrfs_super_bytenr(disk_super) != bytenr ||
1914                                 btrfs_super_magic(disk_super) != BTRFS_MAGIC) {
1915                                 continue;
1916                         }
1917                         memset(&disk_super->magic, 0,
1918                                                 sizeof(disk_super->magic));
1919                         set_buffer_dirty(bh);
1920                         sync_dirty_buffer(bh);
1921                 }
1922         }
1923
1924         ret = 0;
1925
1926         if (bdev) {
1927                 /* Notify udev that device has changed */
1928                 btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
1929
1930                 /* Update ctime/mtime for device path for libblkid */
1931                 update_dev_time(device_path);
1932         }
1933
1934 error_brelse:
1935         brelse(bh);
1936         if (bdev)
1937                 blkdev_put(bdev, FMODE_READ | FMODE_EXCL);
1938 out:
1939         mutex_unlock(&uuid_mutex);
1940         return ret;
1941 error_undo:
1942         if (device->writeable) {
1943                 lock_chunks(root);
1944                 list_add(&device->dev_alloc_list,
1945                          &root->fs_info->fs_devices->alloc_list);
1946                 device->fs_devices->rw_devices++;
1947                 unlock_chunks(root);
1948         }
1949         goto error_brelse;
1950 }
1951
1952 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_fs_info *fs_info,
1953                                         struct btrfs_device *srcdev)
1954 {
1955         struct btrfs_fs_devices *fs_devices;
1956
1957         WARN_ON(!mutex_is_locked(&fs_info->fs_devices->device_list_mutex));
1958
1959         /*
1960          * in case of fs with no seed, srcdev->fs_devices will point
1961          * to fs_devices of fs_info. However when the dev being replaced is
1962          * a seed dev it will point to the seed's local fs_devices. In short
1963          * srcdev will have its correct fs_devices in both the cases.
1964          */
1965         fs_devices = srcdev->fs_devices;
1966
1967         list_del_rcu(&srcdev->dev_list);
1968         list_del_rcu(&srcdev->dev_alloc_list);
1969         fs_devices->num_devices--;
1970         if (srcdev->missing)
1971                 fs_devices->missing_devices--;
1972
1973         if (srcdev->writeable) {
1974                 fs_devices->rw_devices--;
1975                 /* zero out the old super if it is writable */
1976                 btrfs_scratch_superblocks(srcdev->bdev,
1977                                         rcu_str_deref(srcdev->name));
1978         }
1979
1980         if (srcdev->bdev)
1981                 fs_devices->open_devices--;
1982 }
1983
1984 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_fs_info *fs_info,
1985                                       struct btrfs_device *srcdev)
1986 {
1987         struct btrfs_fs_devices *fs_devices = srcdev->fs_devices;
1988
1989         call_rcu(&srcdev->rcu, free_device);
1990
1991         /*
1992          * unless fs_devices is seed fs, num_devices shouldn't go
1993          * zero
1994          */
1995         BUG_ON(!fs_devices->num_devices && !fs_devices->seeding);
1996
1997         /* if this is no devs we rather delete the fs_devices */
1998         if (!fs_devices->num_devices) {
1999                 struct btrfs_fs_devices *tmp_fs_devices;
2000
2001                 tmp_fs_devices = fs_info->fs_devices;
2002                 while (tmp_fs_devices) {
2003                         if (tmp_fs_devices->seed == fs_devices) {
2004                                 tmp_fs_devices->seed = fs_devices->seed;
2005                                 break;
2006                         }
2007                         tmp_fs_devices = tmp_fs_devices->seed;
2008                 }
2009                 fs_devices->seed = NULL;
2010                 __btrfs_close_devices(fs_devices);
2011                 free_fs_devices(fs_devices);
2012         }
2013 }
2014
2015 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
2016                                       struct btrfs_device *tgtdev)
2017 {
2018         struct btrfs_device *next_device;
2019
2020         mutex_lock(&uuid_mutex);
2021         WARN_ON(!tgtdev);
2022         mutex_lock(&fs_info->fs_devices->device_list_mutex);
2023
2024         btrfs_sysfs_rm_device_link(fs_info->fs_devices, tgtdev);
2025
2026         if (tgtdev->bdev) {
2027                 btrfs_scratch_superblocks(tgtdev->bdev,
2028                                         rcu_str_deref(tgtdev->name));
2029                 fs_info->fs_devices->open_devices--;
2030         }
2031         fs_info->fs_devices->num_devices--;
2032
2033         next_device = list_entry(fs_info->fs_devices->devices.next,
2034                                  struct btrfs_device, dev_list);
2035         if (tgtdev->bdev == fs_info->sb->s_bdev)
2036                 fs_info->sb->s_bdev = next_device->bdev;
2037         if (tgtdev->bdev == fs_info->fs_devices->latest_bdev)
2038                 fs_info->fs_devices->latest_bdev = next_device->bdev;
2039         list_del_rcu(&tgtdev->dev_list);
2040
2041         call_rcu(&tgtdev->rcu, free_device);
2042
2043         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2044         mutex_unlock(&uuid_mutex);
2045 }
2046
2047 static int btrfs_find_device_by_path(struct btrfs_root *root, char *device_path,
2048                                      struct btrfs_device **device)
2049 {
2050         int ret = 0;
2051         struct btrfs_super_block *disk_super;
2052         u64 devid;
2053         u8 *dev_uuid;
2054         struct block_device *bdev;
2055         struct buffer_head *bh;
2056
2057         *device = NULL;
2058         ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
2059                                     root->fs_info->bdev_holder, 0, &bdev, &bh);
2060         if (ret)
2061                 return ret;
2062         disk_super = (struct btrfs_super_block *)bh->b_data;
2063         devid = btrfs_stack_device_id(&disk_super->dev_item);
2064         dev_uuid = disk_super->dev_item.uuid;
2065         *device = btrfs_find_device(root->fs_info, devid, dev_uuid,
2066                                     disk_super->fsid);
2067         brelse(bh);
2068         if (!*device)
2069                 ret = -ENOENT;
2070         blkdev_put(bdev, FMODE_READ);
2071         return ret;
2072 }
2073
2074 int btrfs_find_device_missing_or_by_path(struct btrfs_root *root,
2075                                          char *device_path,
2076                                          struct btrfs_device **device)
2077 {
2078         *device = NULL;
2079         if (strcmp(device_path, "missing") == 0) {
2080                 struct list_head *devices;
2081                 struct btrfs_device *tmp;
2082
2083                 devices = &root->fs_info->fs_devices->devices;
2084                 /*
2085                  * It is safe to read the devices since the volume_mutex
2086                  * is held by the caller.
2087                  */
2088                 list_for_each_entry(tmp, devices, dev_list) {
2089                         if (tmp->in_fs_metadata && !tmp->bdev) {
2090                                 *device = tmp;
2091                                 break;
2092                         }
2093                 }
2094
2095                 if (!*device)
2096                         return BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
2097
2098                 return 0;
2099         } else {
2100                 return btrfs_find_device_by_path(root, device_path, device);
2101         }
2102 }
2103
2104 /*
2105  * does all the dirty work required for changing file system's UUID.
2106  */
2107 static int btrfs_prepare_sprout(struct btrfs_root *root)
2108 {
2109         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2110         struct btrfs_fs_devices *old_devices;
2111         struct btrfs_fs_devices *seed_devices;
2112         struct btrfs_super_block *disk_super = root->fs_info->super_copy;
2113         struct btrfs_device *device;
2114         u64 super_flags;
2115
2116         BUG_ON(!mutex_is_locked(&uuid_mutex));
2117         if (!fs_devices->seeding)
2118                 return -EINVAL;
2119
2120         seed_devices = __alloc_fs_devices();
2121         if (IS_ERR(seed_devices))
2122                 return PTR_ERR(seed_devices);
2123
2124         old_devices = clone_fs_devices(fs_devices);
2125         if (IS_ERR(old_devices)) {
2126                 kfree(seed_devices);
2127                 return PTR_ERR(old_devices);
2128         }
2129
2130         list_add(&old_devices->list, &fs_uuids);
2131
2132         memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
2133         seed_devices->opened = 1;
2134         INIT_LIST_HEAD(&seed_devices->devices);
2135         INIT_LIST_HEAD(&seed_devices->alloc_list);
2136         mutex_init(&seed_devices->device_list_mutex);
2137
2138         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2139         list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
2140                               synchronize_rcu);
2141         list_for_each_entry(device, &seed_devices->devices, dev_list)
2142                 device->fs_devices = seed_devices;
2143
2144         lock_chunks(root);
2145         list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
2146         unlock_chunks(root);
2147
2148         fs_devices->seeding = 0;
2149         fs_devices->num_devices = 0;
2150         fs_devices->open_devices = 0;
2151         fs_devices->missing_devices = 0;
2152         fs_devices->rotating = 0;
2153         fs_devices->seed = seed_devices;
2154
2155         generate_random_uuid(fs_devices->fsid);
2156         memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2157         memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2158         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2159
2160         super_flags = btrfs_super_flags(disk_super) &
2161                       ~BTRFS_SUPER_FLAG_SEEDING;
2162         btrfs_set_super_flags(disk_super, super_flags);
2163
2164         return 0;
2165 }
2166
2167 /*
2168  * strore the expected generation for seed devices in device items.
2169  */
2170 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
2171                                struct btrfs_root *root)
2172 {
2173         struct btrfs_path *path;
2174         struct extent_buffer *leaf;
2175         struct btrfs_dev_item *dev_item;
2176         struct btrfs_device *device;
2177         struct btrfs_key key;
2178         u8 fs_uuid[BTRFS_UUID_SIZE];
2179         u8 dev_uuid[BTRFS_UUID_SIZE];
2180         u64 devid;
2181         int ret;
2182
2183         path = btrfs_alloc_path();
2184         if (!path)
2185                 return -ENOMEM;
2186
2187         root = root->fs_info->chunk_root;
2188         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2189         key.offset = 0;
2190         key.type = BTRFS_DEV_ITEM_KEY;
2191
2192         while (1) {
2193                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2194                 if (ret < 0)
2195                         goto error;
2196
2197                 leaf = path->nodes[0];
2198 next_slot:
2199                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2200                         ret = btrfs_next_leaf(root, path);
2201                         if (ret > 0)
2202                                 break;
2203                         if (ret < 0)
2204                                 goto error;
2205                         leaf = path->nodes[0];
2206                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2207                         btrfs_release_path(path);
2208                         continue;
2209                 }
2210
2211                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2212                 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
2213                     key.type != BTRFS_DEV_ITEM_KEY)
2214                         break;
2215
2216                 dev_item = btrfs_item_ptr(leaf, path->slots[0],
2217                                           struct btrfs_dev_item);
2218                 devid = btrfs_device_id(leaf, dev_item);
2219                 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
2220                                    BTRFS_UUID_SIZE);
2221                 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
2222                                    BTRFS_UUID_SIZE);
2223                 device = btrfs_find_device(root->fs_info, devid, dev_uuid,
2224                                            fs_uuid);
2225                 BUG_ON(!device); /* Logic error */
2226
2227                 if (device->fs_devices->seeding) {
2228                         btrfs_set_device_generation(leaf, dev_item,
2229                                                     device->generation);
2230                         btrfs_mark_buffer_dirty(leaf);
2231                 }
2232
2233                 path->slots[0]++;
2234                 goto next_slot;
2235         }
2236         ret = 0;
2237 error:
2238         btrfs_free_path(path);
2239         return ret;
2240 }
2241
2242 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
2243 {
2244         struct request_queue *q;
2245         struct btrfs_trans_handle *trans;
2246         struct btrfs_device *device;
2247         struct block_device *bdev;
2248         struct list_head *devices;
2249         struct super_block *sb = root->fs_info->sb;
2250         struct rcu_string *name;
2251         u64 tmp;
2252         int seeding_dev = 0;
2253         int ret = 0;
2254
2255         if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
2256                 return -EROFS;
2257
2258         bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2259                                   root->fs_info->bdev_holder);
2260         if (IS_ERR(bdev))
2261                 return PTR_ERR(bdev);
2262
2263         if (root->fs_info->fs_devices->seeding) {
2264                 seeding_dev = 1;
2265                 down_write(&sb->s_umount);
2266                 mutex_lock(&uuid_mutex);
2267         }
2268
2269         filemap_write_and_wait(bdev->bd_inode->i_mapping);
2270
2271         devices = &root->fs_info->fs_devices->devices;
2272
2273         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2274         list_for_each_entry(device, devices, dev_list) {
2275                 if (device->bdev == bdev) {
2276                         ret = -EEXIST;
2277                         mutex_unlock(
2278                                 &root->fs_info->fs_devices->device_list_mutex);
2279                         goto error;
2280                 }
2281         }
2282         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2283
2284         device = btrfs_alloc_device(root->fs_info, NULL, NULL);
2285         if (IS_ERR(device)) {
2286                 /* we can safely leave the fs_devices entry around */
2287                 ret = PTR_ERR(device);
2288                 goto error;
2289         }
2290
2291         name = rcu_string_strdup(device_path, GFP_NOFS);
2292         if (!name) {
2293                 kfree(device);
2294                 ret = -ENOMEM;
2295                 goto error;
2296         }
2297         rcu_assign_pointer(device->name, name);
2298
2299         trans = btrfs_start_transaction(root, 0);
2300         if (IS_ERR(trans)) {
2301                 rcu_string_free(device->name);
2302                 kfree(device);
2303                 ret = PTR_ERR(trans);
2304                 goto error;
2305         }
2306
2307         q = bdev_get_queue(bdev);
2308         if (blk_queue_discard(q))
2309                 device->can_discard = 1;
2310         device->writeable = 1;
2311         device->generation = trans->transid;
2312         device->io_width = root->sectorsize;
2313         device->io_align = root->sectorsize;
2314         device->sector_size = root->sectorsize;
2315         device->total_bytes = i_size_read(bdev->bd_inode);
2316         device->disk_total_bytes = device->total_bytes;
2317         device->commit_total_bytes = device->total_bytes;
2318         device->dev_root = root->fs_info->dev_root;
2319         device->bdev = bdev;
2320         device->in_fs_metadata = 1;
2321         device->is_tgtdev_for_dev_replace = 0;
2322         device->mode = FMODE_EXCL;
2323         device->dev_stats_valid = 1;
2324         set_blocksize(device->bdev, 4096);
2325
2326         if (seeding_dev) {
2327                 sb->s_flags &= ~MS_RDONLY;
2328                 ret = btrfs_prepare_sprout(root);
2329                 BUG_ON(ret); /* -ENOMEM */
2330         }
2331
2332         device->fs_devices = root->fs_info->fs_devices;
2333
2334         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2335         lock_chunks(root);
2336         list_add_rcu(&device->dev_list, &root->fs_info->fs_devices->devices);
2337         list_add(&device->dev_alloc_list,
2338                  &root->fs_info->fs_devices->alloc_list);
2339         root->fs_info->fs_devices->num_devices++;
2340         root->fs_info->fs_devices->open_devices++;
2341         root->fs_info->fs_devices->rw_devices++;
2342         root->fs_info->fs_devices->total_devices++;
2343         root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
2344
2345         spin_lock(&root->fs_info->free_chunk_lock);
2346         root->fs_info->free_chunk_space += device->total_bytes;
2347         spin_unlock(&root->fs_info->free_chunk_lock);
2348
2349         if (!blk_queue_nonrot(bdev_get_queue(bdev)))
2350                 root->fs_info->fs_devices->rotating = 1;
2351
2352         tmp = btrfs_super_total_bytes(root->fs_info->super_copy);
2353         btrfs_set_super_total_bytes(root->fs_info->super_copy,
2354                                     tmp + device->total_bytes);
2355
2356         tmp = btrfs_super_num_devices(root->fs_info->super_copy);
2357         btrfs_set_super_num_devices(root->fs_info->super_copy,
2358                                     tmp + 1);
2359
2360         /* add sysfs device entry */
2361         btrfs_sysfs_add_device_link(root->fs_info->fs_devices, device);
2362
2363         /*
2364          * we've got more storage, clear any full flags on the space
2365          * infos
2366          */
2367         btrfs_clear_space_info_full(root->fs_info);
2368
2369         unlock_chunks(root);
2370         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2371
2372         if (seeding_dev) {
2373                 lock_chunks(root);
2374                 ret = init_first_rw_device(trans, root, device);
2375                 unlock_chunks(root);
2376                 if (ret) {
2377                         btrfs_abort_transaction(trans, root, ret);
2378                         goto error_trans;
2379                 }
2380         }
2381
2382         ret = btrfs_add_device(trans, root, device);
2383         if (ret) {
2384                 btrfs_abort_transaction(trans, root, ret);
2385                 goto error_trans;
2386         }
2387
2388         if (seeding_dev) {
2389                 char fsid_buf[BTRFS_UUID_UNPARSED_SIZE];
2390
2391                 ret = btrfs_finish_sprout(trans, root);
2392                 if (ret) {
2393                         btrfs_abort_transaction(trans, root, ret);
2394                         goto error_trans;
2395                 }
2396
2397                 /* Sprouting would change fsid of the mounted root,
2398                  * so rename the fsid on the sysfs
2399                  */
2400                 snprintf(fsid_buf, BTRFS_UUID_UNPARSED_SIZE, "%pU",
2401                                                 root->fs_info->fsid);
2402                 if (kobject_rename(&root->fs_info->fs_devices->fsid_kobj,
2403                                                                 fsid_buf))
2404                         btrfs_warn(root->fs_info,
2405                                 "sysfs: failed to create fsid for sprout");
2406         }
2407
2408         root->fs_info->num_tolerated_disk_barrier_failures =
2409                 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
2410         ret = btrfs_commit_transaction(trans, root);
2411
2412         if (seeding_dev) {
2413                 mutex_unlock(&uuid_mutex);
2414                 up_write(&sb->s_umount);
2415
2416                 if (ret) /* transaction commit */
2417                         return ret;
2418
2419                 ret = btrfs_relocate_sys_chunks(root);
2420                 if (ret < 0)
2421                         btrfs_std_error(root->fs_info, ret,
2422                                     "Failed to relocate sys chunks after "
2423                                     "device initialization. This can be fixed "
2424                                     "using the \"btrfs balance\" command.");
2425                 trans = btrfs_attach_transaction(root);
2426                 if (IS_ERR(trans)) {
2427                         if (PTR_ERR(trans) == -ENOENT)
2428                                 return 0;
2429                         return PTR_ERR(trans);
2430                 }
2431                 ret = btrfs_commit_transaction(trans, root);
2432         }
2433
2434         /* Update ctime/mtime for libblkid */
2435         update_dev_time(device_path);
2436         return ret;
2437
2438 error_trans:
2439         btrfs_end_transaction(trans, root);
2440         rcu_string_free(device->name);
2441         btrfs_sysfs_rm_device_link(root->fs_info->fs_devices, device);
2442         kfree(device);
2443 error:
2444         blkdev_put(bdev, FMODE_EXCL);
2445         if (seeding_dev) {
2446                 mutex_unlock(&uuid_mutex);
2447                 up_write(&sb->s_umount);
2448         }
2449         return ret;
2450 }
2451
2452 int btrfs_init_dev_replace_tgtdev(struct btrfs_root *root, char *device_path,
2453                                   struct btrfs_device *srcdev,
2454                                   struct btrfs_device **device_out)
2455 {
2456         struct request_queue *q;
2457         struct btrfs_device *device;
2458         struct block_device *bdev;
2459         struct btrfs_fs_info *fs_info = root->fs_info;
2460         struct list_head *devices;
2461         struct rcu_string *name;
2462         u64 devid = BTRFS_DEV_REPLACE_DEVID;
2463         int ret = 0;
2464
2465         *device_out = NULL;
2466         if (fs_info->fs_devices->seeding) {
2467                 btrfs_err(fs_info, "the filesystem is a seed filesystem!");
2468                 return -EINVAL;
2469         }
2470
2471         bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2472                                   fs_info->bdev_holder);
2473         if (IS_ERR(bdev)) {
2474                 btrfs_err(fs_info, "target device %s is invalid!", device_path);
2475                 return PTR_ERR(bdev);
2476         }
2477
2478         filemap_write_and_wait(bdev->bd_inode->i_mapping);
2479
2480         devices = &fs_info->fs_devices->devices;
2481         list_for_each_entry(device, devices, dev_list) {
2482                 if (device->bdev == bdev) {
2483                         btrfs_err(fs_info, "target device is in the filesystem!");
2484                         ret = -EEXIST;
2485                         goto error;
2486                 }
2487         }
2488
2489
2490         if (i_size_read(bdev->bd_inode) <
2491             btrfs_device_get_total_bytes(srcdev)) {
2492                 btrfs_err(fs_info, "target device is smaller than source device!");
2493                 ret = -EINVAL;
2494                 goto error;
2495         }
2496
2497
2498         device = btrfs_alloc_device(NULL, &devid, NULL);
2499         if (IS_ERR(device)) {
2500                 ret = PTR_ERR(device);
2501                 goto error;
2502         }
2503
2504         name = rcu_string_strdup(device_path, GFP_NOFS);
2505         if (!name) {
2506                 kfree(device);
2507                 ret = -ENOMEM;
2508                 goto error;
2509         }
2510         rcu_assign_pointer(device->name, name);
2511
2512         q = bdev_get_queue(bdev);
2513         if (blk_queue_discard(q))
2514                 device->can_discard = 1;
2515         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2516         device->writeable = 1;
2517         device->generation = 0;
2518         device->io_width = root->sectorsize;
2519         device->io_align = root->sectorsize;
2520         device->sector_size = root->sectorsize;
2521         device->total_bytes = btrfs_device_get_total_bytes(srcdev);
2522         device->disk_total_bytes = btrfs_device_get_disk_total_bytes(srcdev);
2523         device->bytes_used = btrfs_device_get_bytes_used(srcdev);
2524         ASSERT(list_empty(&srcdev->resized_list));
2525         device->commit_total_bytes = srcdev->commit_total_bytes;
2526         device->commit_bytes_used = device->bytes_used;
2527         device->dev_root = fs_info->dev_root;
2528         device->bdev = bdev;
2529         device->in_fs_metadata = 1;
2530         device->is_tgtdev_for_dev_replace = 1;
2531         device->mode = FMODE_EXCL;
2532         device->dev_stats_valid = 1;
2533         set_blocksize(device->bdev, 4096);
2534         device->fs_devices = fs_info->fs_devices;
2535         list_add(&device->dev_list, &fs_info->fs_devices->devices);
2536         fs_info->fs_devices->num_devices++;
2537         fs_info->fs_devices->open_devices++;
2538         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2539
2540         *device_out = device;
2541         return ret;
2542
2543 error:
2544         blkdev_put(bdev, FMODE_EXCL);
2545         return ret;
2546 }
2547
2548 void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info *fs_info,
2549                                               struct btrfs_device *tgtdev)
2550 {
2551         WARN_ON(fs_info->fs_devices->rw_devices == 0);
2552         tgtdev->io_width = fs_info->dev_root->sectorsize;
2553         tgtdev->io_align = fs_info->dev_root->sectorsize;
2554         tgtdev->sector_size = fs_info->dev_root->sectorsize;
2555         tgtdev->dev_root = fs_info->dev_root;
2556         tgtdev->in_fs_metadata = 1;
2557 }
2558
2559 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2560                                         struct btrfs_device *device)
2561 {
2562         int ret;
2563         struct btrfs_path *path;
2564         struct btrfs_root *root;
2565         struct btrfs_dev_item *dev_item;
2566         struct extent_buffer *leaf;
2567         struct btrfs_key key;
2568
2569         root = device->dev_root->fs_info->chunk_root;
2570
2571         path = btrfs_alloc_path();
2572         if (!path)
2573                 return -ENOMEM;
2574
2575         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2576         key.type = BTRFS_DEV_ITEM_KEY;
2577         key.offset = device->devid;
2578
2579         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2580         if (ret < 0)
2581                 goto out;
2582
2583         if (ret > 0) {
2584                 ret = -ENOENT;
2585                 goto out;
2586         }
2587
2588         leaf = path->nodes[0];
2589         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2590
2591         btrfs_set_device_id(leaf, dev_item, device->devid);
2592         btrfs_set_device_type(leaf, dev_item, device->type);
2593         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2594         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2595         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2596         btrfs_set_device_total_bytes(leaf, dev_item,
2597                                      btrfs_device_get_disk_total_bytes(device));
2598         btrfs_set_device_bytes_used(leaf, dev_item,
2599                                     btrfs_device_get_bytes_used(device));
2600         btrfs_mark_buffer_dirty(leaf);
2601
2602 out:
2603         btrfs_free_path(path);
2604         return ret;
2605 }
2606
2607 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2608                       struct btrfs_device *device, u64 new_size)
2609 {
2610         struct btrfs_super_block *super_copy =
2611                 device->dev_root->fs_info->super_copy;
2612         struct btrfs_fs_devices *fs_devices;
2613         u64 old_total;
2614         u64 diff;
2615
2616         if (!device->writeable)
2617                 return -EACCES;
2618
2619         lock_chunks(device->dev_root);
2620         old_total = btrfs_super_total_bytes(super_copy);
2621         diff = new_size - device->total_bytes;
2622
2623         if (new_size <= device->total_bytes ||
2624             device->is_tgtdev_for_dev_replace) {
2625                 unlock_chunks(device->dev_root);
2626                 return -EINVAL;
2627         }
2628
2629         fs_devices = device->dev_root->fs_info->fs_devices;
2630
2631         btrfs_set_super_total_bytes(super_copy, old_total + diff);
2632         device->fs_devices->total_rw_bytes += diff;
2633
2634         btrfs_device_set_total_bytes(device, new_size);
2635         btrfs_device_set_disk_total_bytes(device, new_size);
2636         btrfs_clear_space_info_full(device->dev_root->fs_info);
2637         if (list_empty(&device->resized_list))
2638                 list_add_tail(&device->resized_list,
2639                               &fs_devices->resized_devices);
2640         unlock_chunks(device->dev_root);
2641
2642         return btrfs_update_device(trans, device);
2643 }
2644
2645 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
2646                             struct btrfs_root *root, u64 chunk_objectid,
2647                             u64 chunk_offset)
2648 {
2649         int ret;
2650         struct btrfs_path *path;
2651         struct btrfs_key key;
2652
2653         root = root->fs_info->chunk_root;
2654         path = btrfs_alloc_path();
2655         if (!path)
2656                 return -ENOMEM;
2657
2658         key.objectid = chunk_objectid;
2659         key.offset = chunk_offset;
2660         key.type = BTRFS_CHUNK_ITEM_KEY;
2661
2662         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2663         if (ret < 0)
2664                 goto out;
2665         else if (ret > 0) { /* Logic error or corruption */
2666                 btrfs_std_error(root->fs_info, -ENOENT,
2667                             "Failed lookup while freeing chunk.");
2668                 ret = -ENOENT;
2669                 goto out;
2670         }
2671
2672         ret = btrfs_del_item(trans, root, path);
2673         if (ret < 0)
2674                 btrfs_std_error(root->fs_info, ret,
2675                             "Failed to delete chunk item.");
2676 out:
2677         btrfs_free_path(path);
2678         return ret;
2679 }
2680
2681 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
2682                         chunk_offset)
2683 {
2684         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
2685         struct btrfs_disk_key *disk_key;
2686         struct btrfs_chunk *chunk;
2687         u8 *ptr;
2688         int ret = 0;
2689         u32 num_stripes;
2690         u32 array_size;
2691         u32 len = 0;
2692         u32 cur;
2693         struct btrfs_key key;
2694
2695         lock_chunks(root);
2696         array_size = btrfs_super_sys_array_size(super_copy);
2697
2698         ptr = super_copy->sys_chunk_array;
2699         cur = 0;
2700
2701         while (cur < array_size) {
2702                 disk_key = (struct btrfs_disk_key *)ptr;
2703                 btrfs_disk_key_to_cpu(&key, disk_key);
2704
2705                 len = sizeof(*disk_key);
2706
2707                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2708                         chunk = (struct btrfs_chunk *)(ptr + len);
2709                         num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2710                         len += btrfs_chunk_item_size(num_stripes);
2711                 } else {
2712                         ret = -EIO;
2713                         break;
2714                 }
2715                 if (key.objectid == chunk_objectid &&
2716                     key.offset == chunk_offset) {
2717                         memmove(ptr, ptr + len, array_size - (cur + len));
2718                         array_size -= len;
2719                         btrfs_set_super_sys_array_size(super_copy, array_size);
2720                 } else {
2721                         ptr += len;
2722                         cur += len;
2723                 }
2724         }
2725         unlock_chunks(root);
2726         return ret;
2727 }
2728
2729 int btrfs_remove_chunk(struct btrfs_trans_handle *trans,
2730                        struct btrfs_root *root, u64 chunk_offset)
2731 {
2732         struct extent_map_tree *em_tree;
2733         struct extent_map *em;
2734         struct btrfs_root *extent_root = root->fs_info->extent_root;
2735         struct map_lookup *map;
2736         u64 dev_extent_len = 0;
2737         u64 chunk_objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2738         int i, ret = 0;
2739
2740         /* Just in case */
2741         root = root->fs_info->chunk_root;
2742         em_tree = &root->fs_info->mapping_tree.map_tree;
2743
2744         read_lock(&em_tree->lock);
2745         em = lookup_extent_mapping(em_tree, chunk_offset, 1);
2746         read_unlock(&em_tree->lock);
2747
2748         if (!em || em->start > chunk_offset ||
2749             em->start + em->len < chunk_offset) {
2750                 /*
2751                  * This is a logic error, but we don't want to just rely on the
2752                  * user having built with ASSERT enabled, so if ASSERT doens't
2753                  * do anything we still error out.
2754                  */
2755                 ASSERT(0);
2756                 if (em)
2757                         free_extent_map(em);
2758                 return -EINVAL;
2759         }
2760         map = (struct map_lookup *)em->bdev;
2761         lock_chunks(root->fs_info->chunk_root);
2762         check_system_chunk(trans, extent_root, map->type);
2763         unlock_chunks(root->fs_info->chunk_root);
2764
2765         for (i = 0; i < map->num_stripes; i++) {
2766                 struct btrfs_device *device = map->stripes[i].dev;
2767                 ret = btrfs_free_dev_extent(trans, device,
2768                                             map->stripes[i].physical,
2769                                             &dev_extent_len);
2770                 if (ret) {
2771                         btrfs_abort_transaction(trans, root, ret);
2772                         goto out;
2773                 }
2774
2775                 if (device->bytes_used > 0) {
2776                         lock_chunks(root);
2777                         btrfs_device_set_bytes_used(device,
2778                                         device->bytes_used - dev_extent_len);
2779                         spin_lock(&root->fs_info->free_chunk_lock);
2780                         root->fs_info->free_chunk_space += dev_extent_len;
2781                         spin_unlock(&root->fs_info->free_chunk_lock);
2782                         btrfs_clear_space_info_full(root->fs_info);
2783                         unlock_chunks(root);
2784                 }
2785
2786                 if (map->stripes[i].dev) {
2787                         ret = btrfs_update_device(trans, map->stripes[i].dev);
2788                         if (ret) {
2789                                 btrfs_abort_transaction(trans, root, ret);
2790                                 goto out;
2791                         }
2792                 }
2793         }
2794         ret = btrfs_free_chunk(trans, root, chunk_objectid, chunk_offset);
2795         if (ret) {
2796                 btrfs_abort_transaction(trans, root, ret);
2797                 goto out;
2798         }
2799
2800         trace_btrfs_chunk_free(root, map, chunk_offset, em->len);
2801
2802         if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2803                 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
2804                 if (ret) {
2805                         btrfs_abort_transaction(trans, root, ret);
2806                         goto out;
2807                 }
2808         }
2809
2810         ret = btrfs_remove_block_group(trans, extent_root, chunk_offset, em);
2811         if (ret) {
2812                 btrfs_abort_transaction(trans, extent_root, ret);
2813                 goto out;
2814         }
2815
2816 out:
2817         /* once for us */
2818         free_extent_map(em);
2819         return ret;
2820 }
2821
2822 static int btrfs_relocate_chunk(struct btrfs_root *root, u64 chunk_offset)
2823 {
2824         struct btrfs_root *extent_root;
2825         struct btrfs_trans_handle *trans;
2826         int ret;
2827
2828         root = root->fs_info->chunk_root;
2829         extent_root = root->fs_info->extent_root;
2830
2831         /*
2832          * Prevent races with automatic removal of unused block groups.
2833          * After we relocate and before we remove the chunk with offset
2834          * chunk_offset, automatic removal of the block group can kick in,
2835          * resulting in a failure when calling btrfs_remove_chunk() below.
2836          *
2837          * Make sure to acquire this mutex before doing a tree search (dev
2838          * or chunk trees) to find chunks. Otherwise the cleaner kthread might
2839          * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after
2840          * we release the path used to search the chunk/dev tree and before
2841          * the current task acquires this mutex and calls us.
2842          */
2843         ASSERT(mutex_is_locked(&root->fs_info->delete_unused_bgs_mutex));
2844
2845         ret = btrfs_can_relocate(extent_root, chunk_offset);
2846         if (ret)
2847                 return -ENOSPC;
2848
2849         /* step one, relocate all the extents inside this chunk */
2850         btrfs_scrub_pause(root);
2851         ret = btrfs_relocate_block_group(extent_root, chunk_offset);
2852         btrfs_scrub_continue(root);
2853         if (ret)
2854                 return ret;
2855
2856         trans = btrfs_start_transaction(root, 0);
2857         if (IS_ERR(trans)) {
2858                 ret = PTR_ERR(trans);
2859                 btrfs_std_error(root->fs_info, ret, NULL);
2860                 return ret;
2861         }
2862
2863         /*
2864          * step two, delete the device extents and the
2865          * chunk tree entries
2866          */
2867         ret = btrfs_remove_chunk(trans, root, chunk_offset);
2868         btrfs_end_transaction(trans, root);
2869         return ret;
2870 }
2871
2872 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
2873 {
2874         struct btrfs_root *chunk_root = root->fs_info->chunk_root;
2875         struct btrfs_path *path;
2876         struct extent_buffer *leaf;
2877         struct btrfs_chunk *chunk;
2878         struct btrfs_key key;
2879         struct btrfs_key found_key;
2880         u64 chunk_type;
2881         bool retried = false;
2882         int failed = 0;
2883         int ret;
2884
2885         path = btrfs_alloc_path();
2886         if (!path)
2887                 return -ENOMEM;
2888
2889 again:
2890         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2891         key.offset = (u64)-1;
2892         key.type = BTRFS_CHUNK_ITEM_KEY;
2893
2894         while (1) {
2895                 mutex_lock(&root->fs_info->delete_unused_bgs_mutex);
2896                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2897                 if (ret < 0) {
2898                         mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
2899                         goto error;
2900                 }
2901                 BUG_ON(ret == 0); /* Corruption */
2902
2903                 ret = btrfs_previous_item(chunk_root, path, key.objectid,
2904                                           key.type);
2905                 if (ret)
2906                         mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
2907                 if (ret < 0)
2908                         goto error;
2909                 if (ret > 0)
2910                         break;
2911
2912                 leaf = path->nodes[0];
2913                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2914
2915                 chunk = btrfs_item_ptr(leaf, path->slots[0],
2916                                        struct btrfs_chunk);
2917                 chunk_type = btrfs_chunk_type(leaf, chunk);
2918                 btrfs_release_path(path);
2919
2920                 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
2921                         ret = btrfs_relocate_chunk(chunk_root,
2922                                                    found_key.offset);
2923                         if (ret == -ENOSPC)
2924                                 failed++;
2925                         else
2926                                 BUG_ON(ret);
2927                 }
2928                 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
2929
2930                 if (found_key.offset == 0)
2931                         break;
2932                 key.offset = found_key.offset - 1;
2933         }
2934         ret = 0;
2935         if (failed && !retried) {
2936                 failed = 0;
2937                 retried = true;
2938                 goto again;
2939         } else if (WARN_ON(failed && retried)) {
2940                 ret = -ENOSPC;
2941         }
2942 error:
2943         btrfs_free_path(path);
2944         return ret;
2945 }
2946
2947 static int insert_balance_item(struct btrfs_root *root,
2948                                struct btrfs_balance_control *bctl)
2949 {
2950         struct btrfs_trans_handle *trans;
2951         struct btrfs_balance_item *item;
2952         struct btrfs_disk_balance_args disk_bargs;
2953         struct btrfs_path *path;
2954         struct extent_buffer *leaf;
2955         struct btrfs_key key;
2956         int ret, err;
2957
2958         path = btrfs_alloc_path();
2959         if (!path)
2960                 return -ENOMEM;
2961
2962         trans = btrfs_start_transaction(root, 0);
2963         if (IS_ERR(trans)) {
2964                 btrfs_free_path(path);
2965                 return PTR_ERR(trans);
2966         }
2967
2968         key.objectid = BTRFS_BALANCE_OBJECTID;
2969         key.type = BTRFS_BALANCE_ITEM_KEY;
2970         key.offset = 0;
2971
2972         ret = btrfs_insert_empty_item(trans, root, path, &key,
2973                                       sizeof(*item));
2974         if (ret)
2975                 goto out;
2976
2977         leaf = path->nodes[0];
2978         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
2979
2980         memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
2981
2982         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
2983         btrfs_set_balance_data(leaf, item, &disk_bargs);
2984         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
2985         btrfs_set_balance_meta(leaf, item, &disk_bargs);
2986         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
2987         btrfs_set_balance_sys(leaf, item, &disk_bargs);
2988
2989         btrfs_set_balance_flags(leaf, item, bctl->flags);
2990
2991         btrfs_mark_buffer_dirty(leaf);
2992 out:
2993         btrfs_free_path(path);
2994         err = btrfs_commit_transaction(trans, root);
2995         if (err && !ret)
2996                 ret = err;
2997         return ret;
2998 }
2999
3000 static int del_balance_item(struct btrfs_root *root)
3001 {
3002         struct btrfs_trans_handle *trans;
3003         struct btrfs_path *path;
3004         struct btrfs_key key;
3005         int ret, err;
3006
3007         path = btrfs_alloc_path();
3008         if (!path)
3009                 return -ENOMEM;
3010
3011         trans = btrfs_start_transaction(root, 0);
3012         if (IS_ERR(trans)) {
3013                 btrfs_free_path(path);
3014                 return PTR_ERR(trans);
3015         }
3016
3017         key.objectid = BTRFS_BALANCE_OBJECTID;
3018         key.type = BTRFS_BALANCE_ITEM_KEY;
3019         key.offset = 0;
3020
3021         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3022         if (ret < 0)
3023                 goto out;
3024         if (ret > 0) {
3025                 ret = -ENOENT;
3026                 goto out;
3027         }
3028
3029         ret = btrfs_del_item(trans, root, path);
3030 out:
3031         btrfs_free_path(path);
3032         err = btrfs_commit_transaction(trans, root);
3033         if (err && !ret)
3034                 ret = err;
3035         return ret;
3036 }
3037
3038 /*
3039  * This is a heuristic used to reduce the number of chunks balanced on
3040  * resume after balance was interrupted.
3041  */
3042 static void update_balance_args(struct btrfs_balance_control *bctl)
3043 {
3044         /*
3045          * Turn on soft mode for chunk types that were being converted.
3046          */
3047         if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
3048                 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
3049         if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
3050                 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
3051         if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
3052                 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
3053
3054         /*
3055          * Turn on usage filter if is not already used.  The idea is
3056          * that chunks that we have already balanced should be
3057          * reasonably full.  Don't do it for chunks that are being
3058          * converted - that will keep us from relocating unconverted
3059          * (albeit full) chunks.
3060          */
3061         if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3062             !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3063                 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
3064                 bctl->data.usage = 90;
3065         }
3066         if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3067             !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3068                 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
3069                 bctl->sys.usage = 90;
3070         }
3071         if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3072             !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3073                 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
3074                 bctl->meta.usage = 90;
3075         }
3076 }
3077
3078 /*
3079  * Should be called with both balance and volume mutexes held to
3080  * serialize other volume operations (add_dev/rm_dev/resize) with
3081  * restriper.  Same goes for unset_balance_control.
3082  */
3083 static void set_balance_control(struct btrfs_balance_control *bctl)
3084 {
3085         struct btrfs_fs_info *fs_info = bctl->fs_info;
3086
3087         BUG_ON(fs_info->balance_ctl);
3088
3089         spin_lock(&fs_info->balance_lock);
3090         fs_info->balance_ctl = bctl;
3091         spin_unlock(&fs_info->balance_lock);
3092 }
3093
3094 static void unset_balance_control(struct btrfs_fs_info *fs_info)
3095 {
3096         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3097
3098         BUG_ON(!fs_info->balance_ctl);
3099
3100         spin_lock(&fs_info->balance_lock);
3101         fs_info->balance_ctl = NULL;
3102         spin_unlock(&fs_info->balance_lock);
3103
3104         kfree(bctl);
3105 }
3106
3107 /*
3108  * Balance filters.  Return 1 if chunk should be filtered out
3109  * (should not be balanced).
3110  */
3111 static int chunk_profiles_filter(u64 chunk_type,
3112                                  struct btrfs_balance_args *bargs)
3113 {
3114         chunk_type = chunk_to_extended(chunk_type) &
3115                                 BTRFS_EXTENDED_PROFILE_MASK;
3116
3117         if (bargs->profiles & chunk_type)
3118                 return 0;
3119
3120         return 1;
3121 }
3122
3123 static int chunk_usage_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
3124                               struct btrfs_balance_args *bargs)
3125 {
3126         struct btrfs_block_group_cache *cache;
3127         u64 chunk_used, user_thresh;
3128         int ret = 1;
3129
3130         cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3131         chunk_used = btrfs_block_group_used(&cache->item);
3132
3133         if (bargs->usage == 0)
3134                 user_thresh = 1;
3135         else if (bargs->usage > 100)
3136                 user_thresh = cache->key.offset;
3137         else
3138                 user_thresh = div_factor_fine(cache->key.offset,
3139                                               bargs->usage);
3140
3141         if (chunk_used < user_thresh)
3142                 ret = 0;
3143
3144         btrfs_put_block_group(cache);
3145         return ret;
3146 }
3147
3148 static int chunk_devid_filter(struct extent_buffer *leaf,
3149                               struct btrfs_chunk *chunk,
3150                               struct btrfs_balance_args *bargs)
3151 {
3152         struct btrfs_stripe *stripe;
3153         int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3154         int i;
3155
3156         for (i = 0; i < num_stripes; i++) {
3157                 stripe = btrfs_stripe_nr(chunk, i);
3158                 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
3159                         return 0;
3160         }
3161
3162         return 1;
3163 }
3164
3165 /* [pstart, pend) */
3166 static int chunk_drange_filter(struct extent_buffer *leaf,
3167                                struct btrfs_chunk *chunk,
3168                                u64 chunk_offset,
3169                                struct btrfs_balance_args *bargs)
3170 {
3171         struct btrfs_stripe *stripe;
3172         int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3173         u64 stripe_offset;
3174         u64 stripe_length;
3175         int factor;
3176         int i;
3177
3178         if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
3179                 return 0;
3180
3181         if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP |
3182              BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) {
3183                 factor = num_stripes / 2;
3184         } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) {
3185                 factor = num_stripes - 1;
3186         } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) {
3187                 factor = num_stripes - 2;
3188         } else {
3189                 factor = num_stripes;
3190         }
3191
3192         for (i = 0; i < num_stripes; i++) {
3193                 stripe = btrfs_stripe_nr(chunk, i);
3194                 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
3195                         continue;
3196
3197                 stripe_offset = btrfs_stripe_offset(leaf, stripe);
3198                 stripe_length = btrfs_chunk_length(leaf, chunk);
3199                 stripe_length = div_u64(stripe_length, factor);
3200
3201                 if (stripe_offset < bargs->pend &&
3202                     stripe_offset + stripe_length > bargs->pstart)
3203                         return 0;
3204         }
3205
3206         return 1;
3207 }
3208
3209 /* [vstart, vend) */
3210 static int chunk_vrange_filter(struct extent_buffer *leaf,
3211                                struct btrfs_chunk *chunk,
3212                                u64 chunk_offset,
3213                                struct btrfs_balance_args *bargs)
3214 {
3215         if (chunk_offset < bargs->vend &&
3216             chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
3217                 /* at least part of the chunk is inside this vrange */
3218                 return 0;
3219
3220         return 1;
3221 }
3222
3223 static int chunk_soft_convert_filter(u64 chunk_type,
3224                                      struct btrfs_balance_args *bargs)
3225 {
3226         if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
3227                 return 0;
3228
3229         chunk_type = chunk_to_extended(chunk_type) &
3230                                 BTRFS_EXTENDED_PROFILE_MASK;
3231
3232         if (bargs->target == chunk_type)
3233                 return 1;
3234
3235         return 0;
3236 }
3237
3238 static int should_balance_chunk(struct btrfs_root *root,
3239                                 struct extent_buffer *leaf,
3240                                 struct btrfs_chunk *chunk, u64 chunk_offset)
3241 {
3242         struct btrfs_balance_control *bctl = root->fs_info->balance_ctl;
3243         struct btrfs_balance_args *bargs = NULL;
3244         u64 chunk_type = btrfs_chunk_type(leaf, chunk);
3245
3246         /* type filter */
3247         if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
3248               (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
3249                 return 0;
3250         }
3251
3252         if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3253                 bargs = &bctl->data;
3254         else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3255                 bargs = &bctl->sys;
3256         else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3257                 bargs = &bctl->meta;
3258
3259         /* profiles filter */
3260         if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
3261             chunk_profiles_filter(chunk_type, bargs)) {
3262                 return 0;
3263         }
3264
3265         /* usage filter */
3266         if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
3267             chunk_usage_filter(bctl->fs_info, chunk_offset, bargs)) {
3268                 return 0;
3269         }
3270
3271         /* devid filter */
3272         if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
3273             chunk_devid_filter(leaf, chunk, bargs)) {
3274                 return 0;
3275         }
3276
3277         /* drange filter, makes sense only with devid filter */
3278         if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
3279             chunk_drange_filter(leaf, chunk, chunk_offset, bargs)) {
3280                 return 0;
3281         }
3282
3283         /* vrange filter */
3284         if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
3285             chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
3286                 return 0;
3287         }
3288
3289         /* soft profile changing mode */
3290         if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
3291             chunk_soft_convert_filter(chunk_type, bargs)) {
3292                 return 0;
3293         }
3294
3295         /*
3296          * limited by count, must be the last filter
3297          */
3298         if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) {
3299                 if (bargs->limit == 0)
3300                         return 0;
3301                 else
3302                         bargs->limit--;
3303         }
3304
3305         return 1;
3306 }
3307
3308 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
3309 {
3310         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3311         struct btrfs_root *chunk_root = fs_info->chunk_root;
3312         struct btrfs_root *dev_root = fs_info->dev_root;
3313         struct list_head *devices;
3314         struct btrfs_device *device;
3315         u64 old_size;
3316         u64 size_to_free;
3317         struct btrfs_chunk *chunk;
3318         struct btrfs_path *path;
3319         struct btrfs_key key;
3320         struct btrfs_key found_key;
3321         struct btrfs_trans_handle *trans;
3322         struct extent_buffer *leaf;
3323         int slot;
3324         int ret;
3325         int enospc_errors = 0;
3326         bool counting = true;
3327         u64 limit_data = bctl->data.limit;
3328         u64 limit_meta = bctl->meta.limit;
3329         u64 limit_sys = bctl->sys.limit;
3330
3331         /* step one make some room on all the devices */
3332         devices = &fs_info->fs_devices->devices;
3333         list_for_each_entry(device, devices, dev_list) {
3334                 old_size = btrfs_device_get_total_bytes(device);
3335                 size_to_free = div_factor(old_size, 1);
3336                 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
3337                 if (!device->writeable ||
3338                     btrfs_device_get_total_bytes(device) -
3339                     btrfs_device_get_bytes_used(device) > size_to_free ||
3340                     device->is_tgtdev_for_dev_replace)
3341                         continue;
3342
3343                 ret = btrfs_shrink_device(device, old_size - size_to_free);
3344                 if (ret == -ENOSPC)
3345                         break;
3346                 BUG_ON(ret);
3347
3348                 trans = btrfs_start_transaction(dev_root, 0);
3349                 BUG_ON(IS_ERR(trans));
3350
3351                 ret = btrfs_grow_device(trans, device, old_size);
3352                 BUG_ON(ret);
3353
3354                 btrfs_end_transaction(trans, dev_root);
3355         }
3356
3357         /* step two, relocate all the chunks */
3358         path = btrfs_alloc_path();
3359         if (!path) {
3360                 ret = -ENOMEM;
3361                 goto error;
3362         }
3363
3364         /* zero out stat counters */
3365         spin_lock(&fs_info->balance_lock);
3366         memset(&bctl->stat, 0, sizeof(bctl->stat));
3367         spin_unlock(&fs_info->balance_lock);
3368 again:
3369         if (!counting) {
3370                 bctl->data.limit = limit_data;
3371                 bctl->meta.limit = limit_meta;
3372                 bctl->sys.limit = limit_sys;
3373         }
3374         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3375         key.offset = (u64)-1;
3376         key.type = BTRFS_CHUNK_ITEM_KEY;
3377
3378         while (1) {
3379                 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
3380                     atomic_read(&fs_info->balance_cancel_req)) {
3381                         ret = -ECANCELED;
3382                         goto error;
3383                 }
3384
3385                 mutex_lock(&fs_info->delete_unused_bgs_mutex);
3386                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3387                 if (ret < 0) {
3388                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3389                         goto error;
3390                 }
3391
3392                 /*
3393                  * this shouldn't happen, it means the last relocate
3394                  * failed
3395                  */
3396                 if (ret == 0)
3397                         BUG(); /* FIXME break ? */
3398
3399                 ret = btrfs_previous_item(chunk_root, path, 0,
3400                                           BTRFS_CHUNK_ITEM_KEY);
3401                 if (ret) {
3402                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3403                         ret = 0;
3404                         break;
3405                 }
3406
3407                 leaf = path->nodes[0];
3408                 slot = path->slots[0];
3409                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3410
3411                 if (found_key.objectid != key.objectid) {
3412                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3413                         break;
3414                 }
3415
3416                 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3417
3418                 if (!counting) {
3419                         spin_lock(&fs_info->balance_lock);
3420                         bctl->stat.considered++;
3421                         spin_unlock(&fs_info->balance_lock);
3422                 }
3423
3424                 ret = should_balance_chunk(chunk_root, leaf, chunk,
3425                                            found_key.offset);
3426                 btrfs_release_path(path);
3427                 if (!ret) {
3428                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3429                         goto loop;
3430                 }
3431
3432                 if (counting) {
3433                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3434                         spin_lock(&fs_info->balance_lock);
3435                         bctl->stat.expected++;
3436                         spin_unlock(&fs_info->balance_lock);
3437                         goto loop;
3438                 }
3439
3440                 ret = btrfs_relocate_chunk(chunk_root,
3441                                            found_key.offset);
3442                 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3443                 if (ret && ret != -ENOSPC)
3444                         goto error;
3445                 if (ret == -ENOSPC) {
3446                         enospc_errors++;
3447                 } else {
3448                         spin_lock(&fs_info->balance_lock);
3449                         bctl->stat.completed++;
3450                         spin_unlock(&fs_info->balance_lock);
3451                 }
3452 loop:
3453                 if (found_key.offset == 0)
3454                         break;
3455                 key.offset = found_key.offset - 1;
3456         }
3457
3458         if (counting) {
3459                 btrfs_release_path(path);
3460                 counting = false;
3461                 goto again;
3462         }
3463 error:
3464         btrfs_free_path(path);
3465         if (enospc_errors) {
3466                 btrfs_info(fs_info, "%d enospc errors during balance",
3467                        enospc_errors);
3468                 if (!ret)
3469                         ret = -ENOSPC;
3470         }
3471
3472         return ret;
3473 }
3474
3475 /**
3476  * alloc_profile_is_valid - see if a given profile is valid and reduced
3477  * @flags: profile to validate
3478  * @extended: if true @flags is treated as an extended profile
3479  */
3480 static int alloc_profile_is_valid(u64 flags, int extended)
3481 {
3482         u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
3483                                BTRFS_BLOCK_GROUP_PROFILE_MASK);
3484
3485         flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
3486
3487         /* 1) check that all other bits are zeroed */
3488         if (flags & ~mask)
3489                 return 0;
3490
3491         /* 2) see if profile is reduced */
3492         if (flags == 0)
3493                 return !extended; /* "0" is valid for usual profiles */
3494
3495         /* true if exactly one bit set */
3496         return (flags & (flags - 1)) == 0;
3497 }
3498
3499 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
3500 {
3501         /* cancel requested || normal exit path */
3502         return atomic_read(&fs_info->balance_cancel_req) ||
3503                 (atomic_read(&fs_info->balance_pause_req) == 0 &&
3504                  atomic_read(&fs_info->balance_cancel_req) == 0);
3505 }
3506
3507 static void __cancel_balance(struct btrfs_fs_info *fs_info)
3508 {
3509         int ret;
3510
3511         unset_balance_control(fs_info);
3512         ret = del_balance_item(fs_info->tree_root);
3513         if (ret)
3514                 btrfs_std_error(fs_info, ret, NULL);
3515
3516         atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3517 }
3518
3519 /* Non-zero return value signifies invalidity */
3520 static inline int validate_convert_profile(struct btrfs_balance_args *bctl_arg,
3521                 u64 allowed)
3522 {
3523         return ((bctl_arg->flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3524                 (!alloc_profile_is_valid(bctl_arg->target, 1) ||
3525                  (bctl_arg->target & ~allowed)));
3526 }
3527
3528 /*
3529  * Should be called with both balance and volume mutexes held
3530  */
3531 int btrfs_balance(struct btrfs_balance_control *bctl,
3532                   struct btrfs_ioctl_balance_args *bargs)
3533 {
3534         struct btrfs_fs_info *fs_info = bctl->fs_info;
3535         u64 allowed;
3536         int mixed = 0;
3537         int ret;
3538         u64 num_devices;
3539         unsigned seq;
3540
3541         if (btrfs_fs_closing(fs_info) ||
3542             atomic_read(&fs_info->balance_pause_req) ||
3543             atomic_read(&fs_info->balance_cancel_req)) {
3544                 ret = -EINVAL;
3545                 goto out;
3546         }
3547
3548         allowed = btrfs_super_incompat_flags(fs_info->super_copy);
3549         if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
3550                 mixed = 1;
3551
3552         /*
3553          * In case of mixed groups both data and meta should be picked,
3554          * and identical options should be given for both of them.
3555          */
3556         allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
3557         if (mixed && (bctl->flags & allowed)) {
3558                 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
3559                     !(bctl->flags & BTRFS_BALANCE_METADATA) ||
3560                     memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
3561                         btrfs_err(fs_info, "with mixed groups data and "
3562                                    "metadata balance options must be the same");
3563                         ret = -EINVAL;
3564                         goto out;
3565                 }
3566         }
3567
3568         num_devices = fs_info->fs_devices->num_devices;
3569         btrfs_dev_replace_lock(&fs_info->dev_replace);
3570         if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
3571                 BUG_ON(num_devices < 1);
3572                 num_devices--;
3573         }
3574         btrfs_dev_replace_unlock(&fs_info->dev_replace);
3575         allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE;
3576         if (num_devices == 1)
3577                 allowed |= BTRFS_BLOCK_GROUP_DUP;
3578         else if (num_devices > 1)
3579                 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1);
3580         if (num_devices > 2)
3581                 allowed |= BTRFS_BLOCK_GROUP_RAID5;
3582         if (num_devices > 3)
3583                 allowed |= (BTRFS_BLOCK_GROUP_RAID10 |
3584                             BTRFS_BLOCK_GROUP_RAID6);
3585         if (validate_convert_profile(&bctl->data, allowed)) {
3586                 btrfs_err(fs_info, "unable to start balance with target "
3587                            "data profile %llu",
3588                        bctl->data.target);
3589                 ret = -EINVAL;
3590                 goto out;
3591         }
3592         if (validate_convert_profile(&bctl->meta, allowed)) {
3593                 btrfs_err(fs_info,
3594                            "unable to start balance with target metadata profile %llu",
3595                        bctl->meta.target);
3596                 ret = -EINVAL;
3597                 goto out;
3598         }
3599         if (validate_convert_profile(&bctl->sys, allowed)) {
3600                 btrfs_err(fs_info,
3601                            "unable to start balance with target system profile %llu",
3602                        bctl->sys.target);
3603                 ret = -EINVAL;
3604                 goto out;
3605         }
3606
3607         /* allow dup'ed data chunks only in mixed mode */
3608         if (!mixed && (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3609             (bctl->data.target & BTRFS_BLOCK_GROUP_DUP)) {
3610                 btrfs_err(fs_info, "dup for data is not allowed");
3611                 ret = -EINVAL;
3612                 goto out;
3613         }
3614
3615         /* allow to reduce meta or sys integrity only if force set */
3616         allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3617                         BTRFS_BLOCK_GROUP_RAID10 |
3618                         BTRFS_BLOCK_GROUP_RAID5 |
3619                         BTRFS_BLOCK_GROUP_RAID6;
3620         do {
3621                 seq = read_seqbegin(&fs_info->profiles_lock);
3622
3623                 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3624                      (fs_info->avail_system_alloc_bits & allowed) &&
3625                      !(bctl->sys.target & allowed)) ||
3626                     ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3627                      (fs_info->avail_metadata_alloc_bits & allowed) &&
3628                      !(bctl->meta.target & allowed))) {
3629                         if (bctl->flags & BTRFS_BALANCE_FORCE) {
3630                                 btrfs_info(fs_info, "force reducing metadata integrity");
3631                         } else {
3632                                 btrfs_err(fs_info, "balance will reduce metadata "
3633                                            "integrity, use force if you want this");
3634                                 ret = -EINVAL;
3635                                 goto out;
3636                         }
3637                 }
3638         } while (read_seqretry(&fs_info->profiles_lock, seq));
3639
3640         if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3641                 fs_info->num_tolerated_disk_barrier_failures = min(
3642                         btrfs_calc_num_tolerated_disk_barrier_failures(fs_info),
3643                         btrfs_get_num_tolerated_disk_barrier_failures(
3644                                 bctl->sys.target));
3645         }
3646
3647         ret = insert_balance_item(fs_info->tree_root, bctl);
3648         if (ret && ret != -EEXIST)
3649                 goto out;
3650
3651         if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
3652                 BUG_ON(ret == -EEXIST);
3653                 set_balance_control(bctl);
3654         } else {
3655                 BUG_ON(ret != -EEXIST);
3656                 spin_lock(&fs_info->balance_lock);
3657                 update_balance_args(bctl);
3658                 spin_unlock(&fs_info->balance_lock);
3659         }
3660
3661         atomic_inc(&fs_info->balance_running);
3662         mutex_unlock(&fs_info->balance_mutex);
3663
3664         ret = __btrfs_balance(fs_info);
3665
3666         mutex_lock(&fs_info->balance_mutex);
3667         atomic_dec(&fs_info->balance_running);
3668
3669         if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3670                 fs_info->num_tolerated_disk_barrier_failures =
3671                         btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3672         }
3673
3674         if (bargs) {
3675                 memset(bargs, 0, sizeof(*bargs));
3676                 update_ioctl_balance_args(fs_info, 0, bargs);
3677         }
3678
3679         if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
3680             balance_need_close(fs_info)) {
3681                 __cancel_balance(fs_info);
3682         }
3683
3684         wake_up(&fs_info->balance_wait_q);
3685
3686         return ret;
3687 out:
3688         if (bctl->flags & BTRFS_BALANCE_RESUME)
3689                 __cancel_balance(fs_info);
3690         else {
3691                 kfree(bctl);
3692                 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3693         }
3694         return ret;
3695 }
3696
3697 static int balance_kthread(void *data)
3698 {
3699         struct btrfs_fs_info *fs_info = data;
3700         int ret = 0;
3701
3702         mutex_lock(&fs_info->volume_mutex);
3703         mutex_lock(&fs_info->balance_mutex);
3704
3705         if (fs_info->balance_ctl) {
3706                 btrfs_info(fs_info, "continuing balance");
3707                 ret = btrfs_balance(fs_info->balance_ctl, NULL);
3708         }
3709
3710         mutex_unlock(&fs_info->balance_mutex);
3711         mutex_unlock(&fs_info->volume_mutex);
3712
3713         return ret;
3714 }
3715
3716 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
3717 {
3718         struct task_struct *tsk;
3719
3720         spin_lock(&fs_info->balance_lock);
3721         if (!fs_info->balance_ctl) {
3722                 spin_unlock(&fs_info->balance_lock);
3723                 return 0;
3724         }
3725         spin_unlock(&fs_info->balance_lock);
3726
3727         if (btrfs_test_opt(fs_info->tree_root, SKIP_BALANCE)) {
3728                 btrfs_info(fs_info, "force skipping balance");
3729                 return 0;
3730         }
3731
3732         tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
3733         return PTR_ERR_OR_ZERO(tsk);
3734 }
3735
3736 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
3737 {
3738         struct btrfs_balance_control *bctl;
3739         struct btrfs_balance_item *item;
3740         struct btrfs_disk_balance_args disk_bargs;
3741         struct btrfs_path *path;
3742         struct extent_buffer *leaf;
3743         struct btrfs_key key;
3744         int ret;
3745
3746         path = btrfs_alloc_path();
3747         if (!path)
3748                 return -ENOMEM;
3749
3750         key.objectid = BTRFS_BALANCE_OBJECTID;
3751         key.type = BTRFS_BALANCE_ITEM_KEY;
3752         key.offset = 0;
3753
3754         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
3755         if (ret < 0)
3756                 goto out;
3757         if (ret > 0) { /* ret = -ENOENT; */
3758                 ret = 0;
3759                 goto out;
3760         }
3761
3762         bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
3763         if (!bctl) {
3764                 ret = -ENOMEM;
3765                 goto out;
3766         }
3767
3768         leaf = path->nodes[0];
3769         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3770
3771         bctl->fs_info = fs_info;
3772         bctl->flags = btrfs_balance_flags(leaf, item);
3773         bctl->flags |= BTRFS_BALANCE_RESUME;
3774
3775         btrfs_balance_data(leaf, item, &disk_bargs);
3776         btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
3777         btrfs_balance_meta(leaf, item, &disk_bargs);
3778         btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
3779         btrfs_balance_sys(leaf, item, &disk_bargs);
3780         btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
3781
3782         WARN_ON(atomic_xchg(&fs_info->mutually_exclusive_operation_running, 1));
3783
3784         mutex_lock(&fs_info->volume_mutex);
3785         mutex_lock(&fs_info->balance_mutex);
3786
3787         set_balance_control(bctl);
3788
3789         mutex_unlock(&fs_info->balance_mutex);
3790         mutex_unlock(&fs_info->volume_mutex);
3791 out:
3792         btrfs_free_path(path);
3793         return ret;
3794 }
3795
3796 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
3797 {
3798         int ret = 0;
3799
3800         mutex_lock(&fs_info->balance_mutex);
3801         if (!fs_info->balance_ctl) {
3802                 mutex_unlock(&fs_info->balance_mutex);
3803                 return -ENOTCONN;
3804         }
3805
3806         if (atomic_read(&fs_info->balance_running)) {
3807                 atomic_inc(&fs_info->balance_pause_req);
3808                 mutex_unlock(&fs_info->balance_mutex);
3809
3810                 wait_event(fs_info->balance_wait_q,
3811                            atomic_read(&fs_info->balance_running) == 0);
3812
3813                 mutex_lock(&fs_info->balance_mutex);
3814                 /* we are good with balance_ctl ripped off from under us */
3815                 BUG_ON(atomic_read(&fs_info->balance_running));
3816                 atomic_dec(&fs_info->balance_pause_req);
3817         } else {
3818                 ret = -ENOTCONN;
3819         }
3820
3821         mutex_unlock(&fs_info->balance_mutex);
3822         return ret;
3823 }
3824
3825 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
3826 {
3827         if (fs_info->sb->s_flags & MS_RDONLY)
3828                 return -EROFS;
3829
3830         mutex_lock(&fs_info->balance_mutex);
3831         if (!fs_info->balance_ctl) {
3832                 mutex_unlock(&fs_info->balance_mutex);
3833                 return -ENOTCONN;
3834         }
3835
3836         atomic_inc(&fs_info->balance_cancel_req);
3837         /*
3838          * if we are running just wait and return, balance item is
3839          * deleted in btrfs_balance in this case
3840          */
3841         if (atomic_read(&fs_info->balance_running)) {
3842                 mutex_unlock(&fs_info->balance_mutex);
3843                 wait_event(fs_info->balance_wait_q,
3844                            atomic_read(&fs_info->balance_running) == 0);
3845                 mutex_lock(&fs_info->balance_mutex);
3846         } else {
3847                 /* __cancel_balance needs volume_mutex */
3848                 mutex_unlock(&fs_info->balance_mutex);
3849                 mutex_lock(&fs_info->volume_mutex);
3850                 mutex_lock(&fs_info->balance_mutex);
3851
3852                 if (fs_info->balance_ctl)
3853                         __cancel_balance(fs_info);
3854
3855                 mutex_unlock(&fs_info->volume_mutex);
3856         }
3857
3858         BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running));
3859         atomic_dec(&fs_info->balance_cancel_req);
3860         mutex_unlock(&fs_info->balance_mutex);
3861         return 0;
3862 }
3863
3864 static int btrfs_uuid_scan_kthread(void *data)
3865 {
3866         struct btrfs_fs_info *fs_info = data;
3867         struct btrfs_root *root = fs_info->tree_root;
3868         struct btrfs_key key;
3869         struct btrfs_key max_key;
3870         struct btrfs_path *path = NULL;
3871         int ret = 0;
3872         struct extent_buffer *eb;
3873         int slot;
3874         struct btrfs_root_item root_item;
3875         u32 item_size;
3876         struct btrfs_trans_handle *trans = NULL;
3877
3878         path = btrfs_alloc_path();
3879         if (!path) {
3880                 ret = -ENOMEM;
3881                 goto out;
3882         }
3883
3884         key.objectid = 0;
3885         key.type = BTRFS_ROOT_ITEM_KEY;
3886         key.offset = 0;
3887
3888         max_key.objectid = (u64)-1;
3889         max_key.type = BTRFS_ROOT_ITEM_KEY;
3890         max_key.offset = (u64)-1;
3891
3892         while (1) {
3893                 ret = btrfs_search_forward(root, &key, path, 0);
3894                 if (ret) {
3895                         if (ret > 0)
3896                                 ret = 0;
3897                         break;
3898                 }
3899
3900                 if (key.type != BTRFS_ROOT_ITEM_KEY ||
3901                     (key.objectid < BTRFS_FIRST_FREE_OBJECTID &&
3902                      key.objectid != BTRFS_FS_TREE_OBJECTID) ||
3903                     key.objectid > BTRFS_LAST_FREE_OBJECTID)
3904                         goto skip;
3905
3906                 eb = path->nodes[0];
3907                 slot = path->slots[0];
3908                 item_size = btrfs_item_size_nr(eb, slot);
3909                 if (item_size < sizeof(root_item))
3910                         goto skip;
3911
3912                 read_extent_buffer(eb, &root_item,
3913                                    btrfs_item_ptr_offset(eb, slot),
3914                                    (int)sizeof(root_item));
3915                 if (btrfs_root_refs(&root_item) == 0)
3916                         goto skip;
3917
3918                 if (!btrfs_is_empty_uuid(root_item.uuid) ||
3919                     !btrfs_is_empty_uuid(root_item.received_uuid)) {
3920                         if (trans)
3921                                 goto update_tree;
3922
3923                         btrfs_release_path(path);
3924                         /*
3925                          * 1 - subvol uuid item
3926                          * 1 - received_subvol uuid item
3927                          */
3928                         trans = btrfs_start_transaction(fs_info->uuid_root, 2);
3929                         if (IS_ERR(trans)) {
3930                                 ret = PTR_ERR(trans);
3931                                 break;
3932                         }
3933                         continue;
3934                 } else {
3935                         goto skip;
3936                 }
3937 update_tree:
3938                 if (!btrfs_is_empty_uuid(root_item.uuid)) {
3939                         ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
3940                                                   root_item.uuid,
3941                                                   BTRFS_UUID_KEY_SUBVOL,
3942                                                   key.objectid);
3943                         if (ret < 0) {
3944                                 btrfs_warn(fs_info, "uuid_tree_add failed %d",
3945                                         ret);
3946                                 break;
3947                         }
3948                 }
3949
3950                 if (!btrfs_is_empty_uuid(root_item.received_uuid)) {
3951                         ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
3952                                                   root_item.received_uuid,
3953                                                  BTRFS_UUID_KEY_RECEIVED_SUBVOL,
3954                                                   key.objectid);
3955                         if (ret < 0) {
3956                                 btrfs_warn(fs_info, "uuid_tree_add failed %d",
3957                                         ret);
3958                                 break;
3959                         }
3960                 }
3961
3962 skip:
3963                 if (trans) {
3964                         ret = btrfs_end_transaction(trans, fs_info->uuid_root);
3965                         trans = NULL;
3966                         if (ret)
3967                                 break;
3968                 }
3969
3970                 btrfs_release_path(path);
3971                 if (key.offset < (u64)-1) {
3972                         key.offset++;
3973                 } else if (key.type < BTRFS_ROOT_ITEM_KEY) {
3974                         key.offset = 0;
3975                         key.type = BTRFS_ROOT_ITEM_KEY;
3976                 } else if (key.objectid < (u64)-1) {
3977                         key.offset = 0;
3978                         key.type = BTRFS_ROOT_ITEM_KEY;
3979                         key.objectid++;
3980                 } else {
3981                         break;
3982                 }
3983                 cond_resched();
3984         }
3985
3986 out:
3987         btrfs_free_path(path);
3988         if (trans && !IS_ERR(trans))
3989                 btrfs_end_transaction(trans, fs_info->uuid_root);
3990         if (ret)
3991                 btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret);
3992         else
3993                 fs_info->update_uuid_tree_gen = 1;
3994         up(&fs_info->uuid_tree_rescan_sem);
3995         return 0;
3996 }
3997
3998 /*
3999  * Callback for btrfs_uuid_tree_iterate().
4000  * returns:
4001  * 0    check succeeded, the entry is not outdated.
4002  * < 0  if an error occured.
4003  * > 0  if the check failed, which means the caller shall remove the entry.
4004  */
4005 static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info *fs_info,
4006                                        u8 *uuid, u8 type, u64 subid)
4007 {
4008         struct btrfs_key key;
4009         int ret = 0;
4010         struct btrfs_root *subvol_root;
4011
4012         if (type != BTRFS_UUID_KEY_SUBVOL &&
4013             type != BTRFS_UUID_KEY_RECEIVED_SUBVOL)
4014                 goto out;
4015
4016         key.objectid = subid;
4017         key.type = BTRFS_ROOT_ITEM_KEY;
4018         key.offset = (u64)-1;
4019         subvol_root = btrfs_read_fs_root_no_name(fs_info, &key);
4020         if (IS_ERR(subvol_root)) {
4021                 ret = PTR_ERR(subvol_root);
4022                 if (ret == -ENOENT)
4023                         ret = 1;
4024                 goto out;
4025         }
4026
4027         switch (type) {
4028         case BTRFS_UUID_KEY_SUBVOL:
4029                 if (memcmp(uuid, subvol_root->root_item.uuid, BTRFS_UUID_SIZE))
4030                         ret = 1;
4031                 break;
4032         case BTRFS_UUID_KEY_RECEIVED_SUBVOL:
4033                 if (memcmp(uuid, subvol_root->root_item.received_uuid,
4034                            BTRFS_UUID_SIZE))
4035                         ret = 1;
4036                 break;
4037         }
4038
4039 out:
4040         return ret;
4041 }
4042
4043 static int btrfs_uuid_rescan_kthread(void *data)
4044 {
4045         struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data;
4046         int ret;
4047
4048         /*
4049          * 1st step is to iterate through the existing UUID tree and
4050          * to delete all entries that contain outdated data.
4051          * 2nd step is to add all missing entries to the UUID tree.
4052          */
4053         ret = btrfs_uuid_tree_iterate(fs_info, btrfs_check_uuid_tree_entry);
4054         if (ret < 0) {
4055                 btrfs_warn(fs_info, "iterating uuid_tree failed %d", ret);
4056                 up(&fs_info->uuid_tree_rescan_sem);
4057                 return ret;
4058         }
4059         return btrfs_uuid_scan_kthread(data);
4060 }
4061
4062 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info)
4063 {
4064         struct btrfs_trans_handle *trans;
4065         struct btrfs_root *tree_root = fs_info->tree_root;
4066         struct btrfs_root *uuid_root;
4067         struct task_struct *task;
4068         int ret;
4069
4070         /*
4071          * 1 - root node
4072          * 1 - root item
4073          */
4074         trans = btrfs_start_transaction(tree_root, 2);
4075         if (IS_ERR(trans))
4076                 return PTR_ERR(trans);
4077
4078         uuid_root = btrfs_create_tree(trans, fs_info,
4079                                       BTRFS_UUID_TREE_OBJECTID);
4080         if (IS_ERR(uuid_root)) {
4081                 ret = PTR_ERR(uuid_root);
4082                 btrfs_abort_transaction(trans, tree_root, ret);
4083                 return ret;
4084         }
4085
4086         fs_info->uuid_root = uuid_root;
4087
4088         ret = btrfs_commit_transaction(trans, tree_root);
4089         if (ret)
4090                 return ret;
4091
4092         down(&fs_info->uuid_tree_rescan_sem);
4093         task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid");
4094         if (IS_ERR(task)) {
4095                 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4096                 btrfs_warn(fs_info, "failed to start uuid_scan task");
4097                 up(&fs_info->uuid_tree_rescan_sem);
4098                 return PTR_ERR(task);
4099         }
4100
4101         return 0;
4102 }
4103
4104 int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
4105 {
4106         struct task_struct *task;
4107
4108         down(&fs_info->uuid_tree_rescan_sem);
4109         task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
4110         if (IS_ERR(task)) {
4111                 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4112                 btrfs_warn(fs_info, "failed to start uuid_rescan task");
4113                 up(&fs_info->uuid_tree_rescan_sem);
4114                 return PTR_ERR(task);
4115         }
4116
4117         return 0;
4118 }
4119
4120 /*
4121  * shrinking a device means finding all of the device extents past
4122  * the new size, and then following the back refs to the chunks.
4123  * The chunk relocation code actually frees the device extent
4124  */
4125 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
4126 {
4127         struct btrfs_trans_handle *trans;
4128         struct btrfs_root *root = device->dev_root;
4129         struct btrfs_dev_extent *dev_extent = NULL;
4130         struct btrfs_path *path;
4131         u64 length;
4132         u64 chunk_offset;
4133         int ret;
4134         int slot;
4135         int failed = 0;
4136         bool retried = false;
4137         bool checked_pending_chunks = false;
4138         struct extent_buffer *l;
4139         struct btrfs_key key;
4140         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
4141         u64 old_total = btrfs_super_total_bytes(super_copy);
4142         u64 old_size = btrfs_device_get_total_bytes(device);
4143         u64 diff = old_size - new_size;
4144
4145         if (device->is_tgtdev_for_dev_replace)
4146                 return -EINVAL;
4147
4148         path = btrfs_alloc_path();
4149         if (!path)
4150                 return -ENOMEM;
4151
4152         path->reada = 2;
4153
4154         lock_chunks(root);
4155
4156         btrfs_device_set_total_bytes(device, new_size);
4157         if (device->writeable) {
4158                 device->fs_devices->total_rw_bytes -= diff;
4159                 spin_lock(&root->fs_info->free_chunk_lock);
4160                 root->fs_info->free_chunk_space -= diff;
4161                 spin_unlock(&root->fs_info->free_chunk_lock);
4162         }
4163         unlock_chunks(root);
4164
4165 again:
4166         key.objectid = device->devid;
4167         key.offset = (u64)-1;
4168         key.type = BTRFS_DEV_EXTENT_KEY;
4169
4170         do {
4171                 mutex_lock(&root->fs_info->delete_unused_bgs_mutex);
4172                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4173                 if (ret < 0) {
4174                         mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4175                         goto done;
4176                 }
4177
4178                 ret = btrfs_previous_item(root, path, 0, key.type);
4179                 if (ret)
4180                         mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4181                 if (ret < 0)
4182                         goto done;
4183                 if (ret) {
4184                         ret = 0;
4185                         btrfs_release_path(path);
4186                         break;
4187                 }
4188
4189                 l = path->nodes[0];
4190                 slot = path->slots[0];
4191                 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
4192
4193                 if (key.objectid != device->devid) {
4194                         mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4195                         btrfs_release_path(path);
4196                         break;
4197                 }
4198
4199                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
4200                 length = btrfs_dev_extent_length(l, dev_extent);
4201
4202                 if (key.offset + length <= new_size) {
4203                         mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4204                         btrfs_release_path(path);
4205                         break;
4206                 }
4207
4208                 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
4209                 btrfs_release_path(path);
4210
4211                 ret = btrfs_relocate_chunk(root, chunk_offset);
4212                 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4213                 if (ret && ret != -ENOSPC)
4214                         goto done;
4215                 if (ret == -ENOSPC)
4216                         failed++;
4217         } while (key.offset-- > 0);
4218
4219         if (failed && !retried) {
4220                 failed = 0;
4221                 retried = true;
4222                 goto again;
4223         } else if (failed && retried) {
4224                 ret = -ENOSPC;
4225                 goto done;
4226         }
4227
4228         /* Shrinking succeeded, else we would be at "done". */
4229         trans = btrfs_start_transaction(root, 0);
4230         if (IS_ERR(trans)) {
4231                 ret = PTR_ERR(trans);
4232                 goto done;
4233         }
4234
4235         lock_chunks(root);
4236
4237         /*
4238          * We checked in the above loop all device extents that were already in
4239          * the device tree. However before we have updated the device's
4240          * total_bytes to the new size, we might have had chunk allocations that
4241          * have not complete yet (new block groups attached to transaction
4242          * handles), and therefore their device extents were not yet in the
4243          * device tree and we missed them in the loop above. So if we have any
4244          * pending chunk using a device extent that overlaps the device range
4245          * that we can not use anymore, commit the current transaction and
4246          * repeat the search on the device tree - this way we guarantee we will
4247          * not have chunks using device extents that end beyond 'new_size'.
4248          */
4249         if (!checked_pending_chunks) {
4250                 u64 start = new_size;
4251                 u64 len = old_size - new_size;
4252
4253                 if (contains_pending_extent(trans->transaction, device,
4254                                             &start, len)) {
4255                         unlock_chunks(root);
4256                         checked_pending_chunks = true;
4257                         failed = 0;
4258                         retried = false;
4259                         ret = btrfs_commit_transaction(trans, root);
4260                         if (ret)
4261                                 goto done;
4262                         goto again;
4263                 }
4264         }
4265
4266         btrfs_device_set_disk_total_bytes(device, new_size);
4267         if (list_empty(&device->resized_list))
4268                 list_add_tail(&device->resized_list,
4269                               &root->fs_info->fs_devices->resized_devices);
4270
4271         WARN_ON(diff > old_total);
4272         btrfs_set_super_total_bytes(super_copy, old_total - diff);
4273         unlock_chunks(root);
4274
4275         /* Now btrfs_update_device() will change the on-disk size. */
4276         ret = btrfs_update_device(trans, device);
4277         btrfs_end_transaction(trans, root);
4278 done:
4279         btrfs_free_path(path);
4280         if (ret) {
4281                 lock_chunks(root);
4282                 btrfs_device_set_total_bytes(device, old_size);
4283                 if (device->writeable)
4284                         device->fs_devices->total_rw_bytes += diff;
4285                 spin_lock(&root->fs_info->free_chunk_lock);
4286                 root->fs_info->free_chunk_space += diff;
4287                 spin_unlock(&root->fs_info->free_chunk_lock);
4288                 unlock_chunks(root);
4289         }
4290         return ret;
4291 }
4292
4293 static int btrfs_add_system_chunk(struct btrfs_root *root,
4294                            struct btrfs_key *key,
4295                            struct btrfs_chunk *chunk, int item_size)
4296 {
4297         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
4298         struct btrfs_disk_key disk_key;
4299         u32 array_size;
4300         u8 *ptr;
4301
4302         lock_chunks(root);
4303         array_size = btrfs_super_sys_array_size(super_copy);
4304         if (array_size + item_size + sizeof(disk_key)
4305                         > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
4306                 unlock_chunks(root);
4307                 return -EFBIG;
4308         }
4309
4310         ptr = super_copy->sys_chunk_array + array_size;
4311         btrfs_cpu_key_to_disk(&disk_key, key);
4312         memcpy(ptr, &disk_key, sizeof(disk_key));
4313         ptr += sizeof(disk_key);
4314         memcpy(ptr, chunk, item_size);
4315         item_size += sizeof(disk_key);
4316         btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
4317         unlock_chunks(root);
4318
4319         return 0;
4320 }
4321
4322 /*
4323  * sort the devices in descending order by max_avail, total_avail
4324  */
4325 static int btrfs_cmp_device_info(const void *a, const void *b)
4326 {
4327         const struct btrfs_device_info *di_a = a;
4328         const struct btrfs_device_info *di_b = b;
4329
4330         if (di_a->max_avail > di_b->max_avail)
4331                 return -1;
4332         if (di_a->max_avail < di_b->max_avail)
4333                 return 1;
4334         if (di_a->total_avail > di_b->total_avail)
4335                 return -1;
4336         if (di_a->total_avail < di_b->total_avail)
4337                 return 1;
4338         return 0;
4339 }
4340
4341 static u32 find_raid56_stripe_len(u32 data_devices, u32 dev_stripe_target)
4342 {
4343         /* TODO allow them to set a preferred stripe size */
4344         return 64 * 1024;
4345 }
4346
4347 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
4348 {
4349         if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK))
4350                 return;
4351
4352         btrfs_set_fs_incompat(info, RAID56);
4353 }
4354
4355 #define BTRFS_MAX_DEVS(r) ((BTRFS_LEAF_DATA_SIZE(r)             \
4356                         - sizeof(struct btrfs_item)             \
4357                         - sizeof(struct btrfs_chunk))           \
4358                         / sizeof(struct btrfs_stripe) + 1)
4359
4360 #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE        \
4361                                 - 2 * sizeof(struct btrfs_disk_key)     \
4362                                 - 2 * sizeof(struct btrfs_chunk))       \
4363                                 / sizeof(struct btrfs_stripe) + 1)
4364
4365 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4366                                struct btrfs_root *extent_root, u64 start,
4367                                u64 type)
4368 {
4369         struct btrfs_fs_info *info = extent_root->fs_info;
4370         struct btrfs_fs_devices *fs_devices = info->fs_devices;
4371         struct list_head *cur;
4372         struct map_lookup *map = NULL;
4373         struct extent_map_tree *em_tree;
4374         struct extent_map *em;
4375         struct btrfs_device_info *devices_info = NULL;
4376         u64 total_avail;
4377         int num_stripes;        /* total number of stripes to allocate */
4378         int data_stripes;       /* number of stripes that count for
4379                                    block group size */
4380         int sub_stripes;        /* sub_stripes info for map */
4381         int dev_stripes;        /* stripes per dev */
4382         int devs_max;           /* max devs to use */
4383         int devs_min;           /* min devs needed */
4384         int devs_increment;     /* ndevs has to be a multiple of this */
4385         int ncopies;            /* how many copies to data has */
4386         int ret;
4387         u64 max_stripe_size;
4388         u64 max_chunk_size;
4389         u64 stripe_size;
4390         u64 num_bytes;
4391         u64 raid_stripe_len = BTRFS_STRIPE_LEN;
4392         int ndevs;
4393         int i;
4394         int j;
4395         int index;
4396
4397         BUG_ON(!alloc_profile_is_valid(type, 0));
4398
4399         if (list_empty(&fs_devices->alloc_list))
4400                 return -ENOSPC;
4401
4402         index = __get_raid_index(type);
4403
4404         sub_stripes = btrfs_raid_array[index].sub_stripes;
4405         dev_stripes = btrfs_raid_array[index].dev_stripes;
4406         devs_max = btrfs_raid_array[index].devs_max;
4407         devs_min = btrfs_raid_array[index].devs_min;
4408         devs_increment = btrfs_raid_array[index].devs_increment;
4409         ncopies = btrfs_raid_array[index].ncopies;
4410
4411         if (type & BTRFS_BLOCK_GROUP_DATA) {
4412                 max_stripe_size = 1024 * 1024 * 1024;
4413                 max_chunk_size = 10 * max_stripe_size;
4414                 if (!devs_max)
4415                         devs_max = BTRFS_MAX_DEVS(info->chunk_root);
4416         } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
4417                 /* for larger filesystems, use larger metadata chunks */
4418                 if (fs_devices->total_rw_bytes > 50ULL * 1024 * 1024 * 1024)
4419                         max_stripe_size = 1024 * 1024 * 1024;
4420                 else
4421                         max_stripe_size = 256 * 1024 * 1024;
4422                 max_chunk_size = max_stripe_size;
4423                 if (!devs_max)
4424                         devs_max = BTRFS_MAX_DEVS(info->chunk_root);
4425         } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
4426                 max_stripe_size = 32 * 1024 * 1024;
4427                 max_chunk_size = 2 * max_stripe_size;
4428                 if (!devs_max)
4429                         devs_max = BTRFS_MAX_DEVS_SYS_CHUNK;
4430         } else {
4431                 btrfs_err(info, "invalid chunk type 0x%llx requested",
4432                        type);
4433                 BUG_ON(1);
4434         }
4435
4436         /* we don't want a chunk larger than 10% of writeable space */
4437         max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
4438                              max_chunk_size);
4439
4440         devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info),
4441                                GFP_NOFS);
4442         if (!devices_info)
4443                 return -ENOMEM;
4444
4445         cur = fs_devices->alloc_list.next;
4446
4447         /*
4448          * in the first pass through the devices list, we gather information
4449          * about the available holes on each device.
4450          */
4451         ndevs = 0;
4452         while (cur != &fs_devices->alloc_list) {
4453                 struct btrfs_device *device;
4454                 u64 max_avail;
4455                 u64 dev_offset;
4456
4457                 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
4458
4459                 cur = cur->next;
4460
4461                 if (!device->writeable) {
4462                         WARN(1, KERN_ERR
4463                                "BTRFS: read-only device in alloc_list\n");
4464                         continue;
4465                 }
4466
4467                 if (!device->in_fs_metadata ||
4468                     device->is_tgtdev_for_dev_replace)
4469                         continue;
4470
4471                 if (device->total_bytes > device->bytes_used)
4472                         total_avail = device->total_bytes - device->bytes_used;
4473                 else
4474                         total_avail = 0;
4475
4476                 /* If there is no space on this device, skip it. */
4477                 if (total_avail == 0)
4478                         continue;
4479
4480                 ret = find_free_dev_extent(trans, device,
4481                                            max_stripe_size * dev_stripes,
4482                                            &dev_offset, &max_avail);
4483                 if (ret && ret != -ENOSPC)
4484                         goto error;
4485
4486                 if (ret == 0)
4487                         max_avail = max_stripe_size * dev_stripes;
4488
4489                 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes)
4490                         continue;
4491
4492                 if (ndevs == fs_devices->rw_devices) {
4493                         WARN(1, "%s: found more than %llu devices\n",
4494                              __func__, fs_devices->rw_devices);
4495                         break;
4496                 }
4497                 devices_info[ndevs].dev_offset = dev_offset;
4498                 devices_info[ndevs].max_avail = max_avail;
4499                 devices_info[ndevs].total_avail = total_avail;
4500                 devices_info[ndevs].dev = device;
4501                 ++ndevs;
4502         }
4503
4504         /*
4505          * now sort the devices by hole size / available space
4506          */
4507         sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
4508              btrfs_cmp_device_info, NULL);
4509
4510         /* round down to number of usable stripes */
4511         ndevs -= ndevs % devs_increment;
4512
4513         if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) {
4514                 ret = -ENOSPC;
4515                 goto error;
4516         }
4517
4518         if (devs_max && ndevs > devs_max)
4519                 ndevs = devs_max;
4520         /*
4521          * the primary goal is to maximize the number of stripes, so use as many
4522          * devices as possible, even if the stripes are not maximum sized.
4523          */
4524         stripe_size = devices_info[ndevs-1].max_avail;
4525         num_stripes = ndevs * dev_stripes;
4526
4527         /*
4528          * this will have to be fixed for RAID1 and RAID10 over
4529          * more drives
4530          */
4531         data_stripes = num_stripes / ncopies;
4532
4533         if (type & BTRFS_BLOCK_GROUP_RAID5) {
4534                 raid_stripe_len = find_raid56_stripe_len(ndevs - 1,
4535                                  btrfs_super_stripesize(info->super_copy));
4536                 data_stripes = num_stripes - 1;
4537         }
4538         if (type & BTRFS_BLOCK_GROUP_RAID6) {
4539                 raid_stripe_len = find_raid56_stripe_len(ndevs - 2,
4540                                  btrfs_super_stripesize(info->super_copy));
4541                 data_stripes = num_stripes - 2;
4542         }
4543
4544         /*
4545          * Use the number of data stripes to figure out how big this chunk
4546          * is really going to be in terms of logical address space,
4547          * and compare that answer with the max chunk size
4548          */
4549         if (stripe_size * data_stripes > max_chunk_size) {
4550                 u64 mask = (1ULL << 24) - 1;
4551
4552                 stripe_size = div_u64(max_chunk_size, data_stripes);
4553
4554                 /* bump the answer up to a 16MB boundary */
4555                 stripe_size = (stripe_size + mask) & ~mask;
4556
4557                 /* but don't go higher than the limits we found
4558                  * while searching for free extents
4559                  */
4560                 if (stripe_size > devices_info[ndevs-1].max_avail)
4561                         stripe_size = devices_info[ndevs-1].max_avail;
4562         }
4563
4564         stripe_size = div_u64(stripe_size, dev_stripes);
4565
4566         /* align to BTRFS_STRIPE_LEN */
4567         stripe_size = div_u64(stripe_size, raid_stripe_len);
4568         stripe_size *= raid_stripe_len;
4569
4570         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
4571         if (!map) {
4572                 ret = -ENOMEM;
4573                 goto error;
4574         }
4575         map->num_stripes = num_stripes;
4576
4577         for (i = 0; i < ndevs; ++i) {
4578                 for (j = 0; j < dev_stripes; ++j) {
4579                         int s = i * dev_stripes + j;
4580                         map->stripes[s].dev = devices_info[i].dev;
4581                         map->stripes[s].physical = devices_info[i].dev_offset +
4582                                                    j * stripe_size;
4583                 }
4584         }
4585         map->sector_size = extent_root->sectorsize;
4586         map->stripe_len = raid_stripe_len;
4587         map->io_align = raid_stripe_len;
4588         map->io_width = raid_stripe_len;
4589         map->type = type;
4590         map->sub_stripes = sub_stripes;
4591
4592         num_bytes = stripe_size * data_stripes;
4593
4594         trace_btrfs_chunk_alloc(info->chunk_root, map, start, num_bytes);
4595
4596         em = alloc_extent_map();
4597         if (!em) {
4598                 kfree(map);
4599                 ret = -ENOMEM;
4600                 goto error;
4601         }
4602         set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
4603         em->bdev = (struct block_device *)map;
4604         em->start = start;
4605         em->len = num_bytes;
4606         em->block_start = 0;
4607         em->block_len = em->len;
4608         em->orig_block_len = stripe_size;
4609
4610         em_tree = &extent_root->fs_info->mapping_tree.map_tree;
4611         write_lock(&em_tree->lock);
4612         ret = add_extent_mapping(em_tree, em, 0);
4613         if (!ret) {
4614                 list_add_tail(&em->list, &trans->transaction->pending_chunks);
4615                 atomic_inc(&em->refs);
4616         }
4617         write_unlock(&em_tree->lock);
4618         if (ret) {
4619                 free_extent_map(em);
4620                 goto error;
4621         }
4622
4623         ret = btrfs_make_block_group(trans, extent_root, 0, type,
4624                                      BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4625                                      start, num_bytes);
4626         if (ret)
4627                 goto error_del_extent;
4628
4629         for (i = 0; i < map->num_stripes; i++) {
4630                 num_bytes = map->stripes[i].dev->bytes_used + stripe_size;
4631                 btrfs_device_set_bytes_used(map->stripes[i].dev, num_bytes);
4632         }
4633
4634         spin_lock(&extent_root->fs_info->free_chunk_lock);
4635         extent_root->fs_info->free_chunk_space -= (stripe_size *
4636                                                    map->num_stripes);
4637         spin_unlock(&extent_root->fs_info->free_chunk_lock);
4638
4639         free_extent_map(em);
4640         check_raid56_incompat_flag(extent_root->fs_info, type);
4641
4642         kfree(devices_info);
4643         return 0;
4644
4645 error_del_extent:
4646         write_lock(&em_tree->lock);
4647         remove_extent_mapping(em_tree, em);
4648         write_unlock(&em_tree->lock);
4649
4650         /* One for our allocation */
4651         free_extent_map(em);
4652         /* One for the tree reference */
4653         free_extent_map(em);
4654         /* One for the pending_chunks list reference */
4655         free_extent_map(em);
4656 error:
4657         kfree(devices_info);
4658         return ret;
4659 }
4660
4661 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans,
4662                                 struct btrfs_root *extent_root,
4663                                 u64 chunk_offset, u64 chunk_size)
4664 {
4665         struct btrfs_key key;
4666         struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
4667         struct btrfs_device *device;
4668         struct btrfs_chunk *chunk;
4669         struct btrfs_stripe *stripe;
4670         struct extent_map_tree *em_tree;
4671         struct extent_map *em;
4672         struct map_lookup *map;
4673         size_t item_size;
4674         u64 dev_offset;
4675         u64 stripe_size;
4676         int i = 0;
4677         int ret;
4678
4679         em_tree = &extent_root->fs_info->mapping_tree.map_tree;
4680         read_lock(&em_tree->lock);
4681         em = lookup_extent_mapping(em_tree, chunk_offset, chunk_size);
4682         read_unlock(&em_tree->lock);
4683
4684         if (!em) {
4685                 btrfs_crit(extent_root->fs_info, "unable to find logical "
4686                            "%Lu len %Lu", chunk_offset, chunk_size);
4687                 return -EINVAL;
4688         }
4689
4690         if (em->start != chunk_offset || em->len != chunk_size) {
4691                 btrfs_crit(extent_root->fs_info, "found a bad mapping, wanted"
4692                           " %Lu-%Lu, found %Lu-%Lu", chunk_offset,
4693                           chunk_size, em->start, em->len);
4694                 free_extent_map(em);
4695                 return -EINVAL;
4696         }
4697
4698         map = (struct map_lookup *)em->bdev;
4699         item_size = btrfs_chunk_item_size(map->num_stripes);
4700         stripe_size = em->orig_block_len;
4701
4702         chunk = kzalloc(item_size, GFP_NOFS);
4703         if (!chunk) {
4704                 ret = -ENOMEM;
4705                 goto out;
4706         }
4707
4708         for (i = 0; i < map->num_stripes; i++) {
4709                 device = map->stripes[i].dev;
4710                 dev_offset = map->stripes[i].physical;
4711
4712                 ret = btrfs_update_device(trans, device);
4713                 if (ret)
4714                         goto out;
4715                 ret = btrfs_alloc_dev_extent(trans, device,
4716                                              chunk_root->root_key.objectid,
4717                                              BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4718                                              chunk_offset, dev_offset,
4719                                              stripe_size);
4720                 if (ret)
4721                         goto out;
4722         }
4723
4724         stripe = &chunk->stripe;
4725         for (i = 0; i < map->num_stripes; i++) {
4726                 device = map->stripes[i].dev;
4727                 dev_offset = map->stripes[i].physical;
4728
4729                 btrfs_set_stack_stripe_devid(stripe, device->devid);
4730                 btrfs_set_stack_stripe_offset(stripe, dev_offset);
4731                 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
4732                 stripe++;
4733         }
4734
4735         btrfs_set_stack_chunk_length(chunk, chunk_size);
4736         btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
4737         btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
4738         btrfs_set_stack_chunk_type(chunk, map->type);
4739         btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
4740         btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
4741         btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
4742         btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
4743         btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
4744
4745         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
4746         key.type = BTRFS_CHUNK_ITEM_KEY;
4747         key.offset = chunk_offset;
4748
4749         ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
4750         if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
4751                 /*
4752                  * TODO: Cleanup of inserted chunk root in case of
4753                  * failure.
4754                  */
4755                 ret = btrfs_add_system_chunk(chunk_root, &key, chunk,
4756                                              item_size);
4757         }
4758
4759 out:
4760         kfree(chunk);
4761         free_extent_map(em);
4762         return ret;
4763 }
4764
4765 /*
4766  * Chunk allocation falls into two parts. The first part does works
4767  * that make the new allocated chunk useable, but not do any operation
4768  * that modifies the chunk tree. The second part does the works that
4769  * require modifying the chunk tree. This division is important for the
4770  * bootstrap process of adding storage to a seed btrfs.
4771  */
4772 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4773                       struct btrfs_root *extent_root, u64 type)
4774 {
4775         u64 chunk_offset;
4776
4777         ASSERT(mutex_is_locked(&extent_root->fs_info->chunk_mutex));
4778         chunk_offset = find_next_chunk(extent_root->fs_info);
4779         return __btrfs_alloc_chunk(trans, extent_root, chunk_offset, type);
4780 }
4781
4782 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
4783                                          struct btrfs_root *root,
4784                                          struct btrfs_device *device)
4785 {
4786         u64 chunk_offset;
4787         u64 sys_chunk_offset;
4788         u64 alloc_profile;
4789         struct btrfs_fs_info *fs_info = root->fs_info;
4790         struct btrfs_root *extent_root = fs_info->extent_root;
4791         int ret;
4792
4793         chunk_offset = find_next_chunk(fs_info);
4794         alloc_profile = btrfs_get_alloc_profile(extent_root, 0);
4795         ret = __btrfs_alloc_chunk(trans, extent_root, chunk_offset,
4796                                   alloc_profile);
4797         if (ret)
4798                 return ret;
4799
4800         sys_chunk_offset = find_next_chunk(root->fs_info);
4801         alloc_profile = btrfs_get_alloc_profile(fs_info->chunk_root, 0);
4802         ret = __btrfs_alloc_chunk(trans, extent_root, sys_chunk_offset,
4803                                   alloc_profile);
4804         return ret;
4805 }
4806
4807 static inline int btrfs_chunk_max_errors(struct map_lookup *map)
4808 {
4809         int max_errors;
4810
4811         if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
4812                          BTRFS_BLOCK_GROUP_RAID10 |
4813                          BTRFS_BLOCK_GROUP_RAID5 |
4814                          BTRFS_BLOCK_GROUP_DUP)) {
4815                 max_errors = 1;
4816         } else if (map->type & BTRFS_BLOCK_GROUP_RAID6) {
4817                 max_errors = 2;
4818         } else {
4819                 max_errors = 0;
4820         }
4821
4822         return max_errors;
4823 }
4824
4825 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
4826 {
4827         struct extent_map *em;
4828         struct map_lookup *map;
4829         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
4830         int readonly = 0;
4831         int miss_ndevs = 0;
4832         int i;
4833
4834         read_lock(&map_tree->map_tree.lock);
4835         em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
4836         read_unlock(&map_tree->map_tree.lock);
4837         if (!em)
4838                 return 1;
4839
4840         map = (struct map_lookup *)em->bdev;
4841         for (i = 0; i < map->num_stripes; i++) {
4842                 if (map->stripes[i].dev->missing) {
4843                         miss_ndevs++;
4844                         continue;
4845                 }
4846
4847                 if (!map->stripes[i].dev->writeable) {
4848                         readonly = 1;
4849                         goto end;
4850                 }
4851         }
4852
4853         /*
4854          * If the number of missing devices is larger than max errors,
4855          * we can not write the data into that chunk successfully, so
4856          * set it readonly.
4857          */
4858         if (miss_ndevs > btrfs_chunk_max_errors(map))
4859                 readonly = 1;
4860 end:
4861         free_extent_map(em);
4862         return readonly;
4863 }
4864
4865 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
4866 {
4867         extent_map_tree_init(&tree->map_tree);
4868 }
4869
4870 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
4871 {
4872         struct extent_map *em;
4873
4874         while (1) {
4875                 write_lock(&tree->map_tree.lock);
4876                 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
4877                 if (em)
4878                         remove_extent_mapping(&tree->map_tree, em);
4879                 write_unlock(&tree->map_tree.lock);
4880                 if (!em)
4881                         break;
4882                 /* once for us */
4883                 free_extent_map(em);
4884                 /* once for the tree */
4885                 free_extent_map(em);
4886         }
4887 }
4888
4889 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
4890 {
4891         struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
4892         struct extent_map *em;
4893         struct map_lookup *map;
4894         struct extent_map_tree *em_tree = &map_tree->map_tree;
4895         int ret;
4896
4897         read_lock(&em_tree->lock);
4898         em = lookup_extent_mapping(em_tree, logical, len);
4899         read_unlock(&em_tree->lock);
4900
4901         /*
4902          * We could return errors for these cases, but that could get ugly and
4903          * we'd probably do the same thing which is just not do anything else
4904          * and exit, so return 1 so the callers don't try to use other copies.
4905          */
4906         if (!em) {
4907                 btrfs_crit(fs_info, "No mapping for %Lu-%Lu", logical,
4908                             logical+len);
4909                 return 1;
4910         }
4911
4912         if (em->start > logical || em->start + em->len < logical) {
4913                 btrfs_crit(fs_info, "Invalid mapping for %Lu-%Lu, got "
4914                             "%Lu-%Lu", logical, logical+len, em->start,
4915                             em->start + em->len);
4916                 free_extent_map(em);
4917                 return 1;
4918         }
4919
4920         map = (struct map_lookup *)em->bdev;
4921         if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
4922                 ret = map->num_stripes;
4923         else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
4924                 ret = map->sub_stripes;
4925         else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
4926                 ret = 2;
4927         else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
4928                 ret = 3;
4929         else
4930                 ret = 1;
4931         free_extent_map(em);
4932
4933         btrfs_dev_replace_lock(&fs_info->dev_replace);
4934         if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace))
4935                 ret++;
4936         btrfs_dev_replace_unlock(&fs_info->dev_replace);
4937
4938         return ret;
4939 }
4940
4941 unsigned long btrfs_full_stripe_len(struct btrfs_root *root,
4942                                     struct btrfs_mapping_tree *map_tree,
4943                                     u64 logical)
4944 {
4945         struct extent_map *em;
4946         struct map_lookup *map;
4947         struct extent_map_tree *em_tree = &map_tree->map_tree;
4948         unsigned long len = root->sectorsize;
4949
4950         read_lock(&em_tree->lock);
4951         em = lookup_extent_mapping(em_tree, logical, len);
4952         read_unlock(&em_tree->lock);
4953         BUG_ON(!em);
4954
4955         BUG_ON(em->start > logical || em->start + em->len < logical);
4956         map = (struct map_lookup *)em->bdev;
4957         if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
4958                 len = map->stripe_len * nr_data_stripes(map);
4959         free_extent_map(em);
4960         return len;
4961 }
4962
4963 int btrfs_is_parity_mirror(struct btrfs_mapping_tree *map_tree,
4964                            u64 logical, u64 len, int mirror_num)
4965 {
4966         struct extent_map *em;
4967         struct map_lookup *map;
4968         struct extent_map_tree *em_tree = &map_tree->map_tree;
4969         int ret = 0;
4970
4971         read_lock(&em_tree->lock);
4972         em = lookup_extent_mapping(em_tree, logical, len);
4973         read_unlock(&em_tree->lock);
4974         BUG_ON(!em);
4975
4976         BUG_ON(em->start > logical || em->start + em->len < logical);
4977         map = (struct map_lookup *)em->bdev;
4978         if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
4979                 ret = 1;
4980         free_extent_map(em);
4981         return ret;
4982 }
4983
4984 static int find_live_mirror(struct btrfs_fs_info *fs_info,
4985                             struct map_lookup *map, int first, int num,
4986                             int optimal, int dev_replace_is_ongoing)
4987 {
4988         int i;
4989         int tolerance;
4990         struct btrfs_device *srcdev;
4991
4992         if (dev_replace_is_ongoing &&
4993             fs_info->dev_replace.cont_reading_from_srcdev_mode ==
4994              BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
4995                 srcdev = fs_info->dev_replace.srcdev;
4996         else
4997                 srcdev = NULL;
4998
4999         /*
5000          * try to avoid the drive that is the source drive for a
5001          * dev-replace procedure, only choose it if no other non-missing
5002          * mirror is available
5003          */
5004         for (tolerance = 0; tolerance < 2; tolerance++) {
5005                 if (map->stripes[optimal].dev->bdev &&
5006                     (tolerance || map->stripes[optimal].dev != srcdev))
5007                         return optimal;
5008                 for (i = first; i < first + num; i++) {
5009                         if (map->stripes[i].dev->bdev &&
5010                             (tolerance || map->stripes[i].dev != srcdev))
5011                                 return i;
5012                 }
5013         }
5014
5015         /* we couldn't find one that doesn't fail.  Just return something
5016          * and the io error handling code will clean up eventually
5017          */
5018         return optimal;
5019 }
5020
5021 static inline int parity_smaller(u64 a, u64 b)
5022 {
5023         return a > b;
5024 }
5025
5026 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
5027 static void sort_parity_stripes(struct btrfs_bio *bbio, int num_stripes)
5028 {
5029         struct btrfs_bio_stripe s;
5030         int i;
5031         u64 l;
5032         int again = 1;
5033
5034         while (again) {
5035                 again = 0;
5036                 for (i = 0; i < num_stripes - 1; i++) {
5037                         if (parity_smaller(bbio->raid_map[i],
5038                                            bbio->raid_map[i+1])) {
5039                                 s = bbio->stripes[i];
5040                                 l = bbio->raid_map[i];
5041                                 bbio->stripes[i] = bbio->stripes[i+1];
5042                                 bbio->raid_map[i] = bbio->raid_map[i+1];
5043                                 bbio->stripes[i+1] = s;
5044                                 bbio->raid_map[i+1] = l;
5045
5046                                 again = 1;
5047                         }
5048                 }
5049         }
5050 }
5051
5052 static struct btrfs_bio *alloc_btrfs_bio(int total_stripes, int real_stripes)
5053 {
5054         struct btrfs_bio *bbio = kzalloc(
5055                  /* the size of the btrfs_bio */
5056                 sizeof(struct btrfs_bio) +
5057                 /* plus the variable array for the stripes */
5058                 sizeof(struct btrfs_bio_stripe) * (total_stripes) +
5059                 /* plus the variable array for the tgt dev */
5060                 sizeof(int) * (real_stripes) +
5061                 /*
5062                  * plus the raid_map, which includes both the tgt dev
5063                  * and the stripes
5064                  */
5065                 sizeof(u64) * (total_stripes),
5066                 GFP_NOFS|__GFP_NOFAIL);
5067
5068         atomic_set(&bbio->error, 0);
5069         atomic_set(&bbio->refs, 1);
5070
5071         return bbio;
5072 }
5073
5074 void btrfs_get_bbio(struct btrfs_bio *bbio)
5075 {
5076         WARN_ON(!atomic_read(&bbio->refs));
5077         atomic_inc(&bbio->refs);
5078 }
5079
5080 void btrfs_put_bbio(struct btrfs_bio *bbio)
5081 {
5082         if (!bbio)
5083                 return;
5084         if (atomic_dec_and_test(&bbio->refs))
5085                 kfree(bbio);
5086 }
5087
5088 static int __btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
5089                              u64 logical, u64 *length,
5090                              struct btrfs_bio **bbio_ret,
5091                              int mirror_num, int need_raid_map)
5092 {
5093         struct extent_map *em;
5094         struct map_lookup *map;
5095         struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
5096         struct extent_map_tree *em_tree = &map_tree->map_tree;
5097         u64 offset;
5098         u64 stripe_offset;
5099         u64 stripe_end_offset;
5100         u64 stripe_nr;
5101         u64 stripe_nr_orig;
5102         u64 stripe_nr_end;
5103         u64 stripe_len;
5104         u32 stripe_index;
5105         int i;
5106         int ret = 0;
5107         int num_stripes;
5108         int max_errors = 0;
5109         int tgtdev_indexes = 0;
5110         struct btrfs_bio *bbio = NULL;
5111         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
5112         int dev_replace_is_ongoing = 0;
5113         int num_alloc_stripes;
5114         int patch_the_first_stripe_for_dev_replace = 0;
5115         u64 physical_to_patch_in_first_stripe = 0;
5116         u64 raid56_full_stripe_start = (u64)-1;
5117
5118         read_lock(&em_tree->lock);
5119         em = lookup_extent_mapping(em_tree, logical, *length);
5120         read_unlock(&em_tree->lock);
5121
5122         if (!em) {
5123                 btrfs_crit(fs_info, "unable to find logical %llu len %llu",
5124                         logical, *length);
5125                 return -EINVAL;
5126         }
5127
5128         if (em->start > logical || em->start + em->len < logical) {
5129                 btrfs_crit(fs_info, "found a bad mapping, wanted %Lu, "
5130                            "found %Lu-%Lu", logical, em->start,
5131                            em->start + em->len);
5132                 free_extent_map(em);
5133                 return -EINVAL;
5134         }
5135
5136         map = (struct map_lookup *)em->bdev;
5137         offset = logical - em->start;
5138
5139         stripe_len = map->stripe_len;
5140         stripe_nr = offset;
5141         /*
5142          * stripe_nr counts the total number of stripes we have to stride
5143          * to get to this block
5144          */
5145         stripe_nr = div64_u64(stripe_nr, stripe_len);
5146
5147         stripe_offset = stripe_nr * stripe_len;
5148         BUG_ON(offset < stripe_offset);
5149
5150         /* stripe_offset is the offset of this block in its stripe*/
5151         stripe_offset = offset - stripe_offset;
5152
5153         /* if we're here for raid56, we need to know the stripe aligned start */
5154         if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5155                 unsigned long full_stripe_len = stripe_len * nr_data_stripes(map);
5156                 raid56_full_stripe_start = offset;
5157
5158                 /* allow a write of a full stripe, but make sure we don't
5159                  * allow straddling of stripes
5160                  */
5161                 raid56_full_stripe_start = div64_u64(raid56_full_stripe_start,
5162                                 full_stripe_len);
5163                 raid56_full_stripe_start *= full_stripe_len;
5164         }
5165
5166         if (rw & REQ_DISCARD) {
5167                 /* we don't discard raid56 yet */
5168                 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5169                         ret = -EOPNOTSUPP;
5170                         goto out;
5171                 }
5172                 *length = min_t(u64, em->len - offset, *length);
5173         } else if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
5174                 u64 max_len;
5175                 /* For writes to RAID[56], allow a full stripeset across all disks.
5176                    For other RAID types and for RAID[56] reads, just allow a single
5177                    stripe (on a single disk). */
5178                 if ((map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
5179                     (rw & REQ_WRITE)) {
5180                         max_len = stripe_len * nr_data_stripes(map) -
5181                                 (offset - raid56_full_stripe_start);
5182                 } else {
5183                         /* we limit the length of each bio to what fits in a stripe */
5184                         max_len = stripe_len - stripe_offset;
5185                 }
5186                 *length = min_t(u64, em->len - offset, max_len);
5187         } else {
5188                 *length = em->len - offset;
5189         }
5190
5191         /* This is for when we're called from btrfs_merge_bio_hook() and all
5192            it cares about is the length */
5193         if (!bbio_ret)
5194                 goto out;
5195
5196         btrfs_dev_replace_lock(dev_replace);
5197         dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
5198         if (!dev_replace_is_ongoing)
5199                 btrfs_dev_replace_unlock(dev_replace);
5200
5201         if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
5202             !(rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) &&
5203             dev_replace->tgtdev != NULL) {
5204                 /*
5205                  * in dev-replace case, for repair case (that's the only
5206                  * case where the mirror is selected explicitly when
5207                  * calling btrfs_map_block), blocks left of the left cursor
5208                  * can also be read from the target drive.
5209                  * For REQ_GET_READ_MIRRORS, the target drive is added as
5210                  * the last one to the array of stripes. For READ, it also
5211                  * needs to be supported using the same mirror number.
5212                  * If the requested block is not left of the left cursor,
5213                  * EIO is returned. This can happen because btrfs_num_copies()
5214                  * returns one more in the dev-replace case.
5215                  */
5216                 u64 tmp_length = *length;
5217                 struct btrfs_bio *tmp_bbio = NULL;
5218                 int tmp_num_stripes;
5219                 u64 srcdev_devid = dev_replace->srcdev->devid;
5220                 int index_srcdev = 0;
5221                 int found = 0;
5222                 u64 physical_of_found = 0;
5223
5224                 ret = __btrfs_map_block(fs_info, REQ_GET_READ_MIRRORS,
5225                              logical, &tmp_length, &tmp_bbio, 0, 0);
5226                 if (ret) {
5227                         WARN_ON(tmp_bbio != NULL);
5228                         goto out;
5229                 }
5230
5231                 tmp_num_stripes = tmp_bbio->num_stripes;
5232                 if (mirror_num > tmp_num_stripes) {
5233                         /*
5234                          * REQ_GET_READ_MIRRORS does not contain this
5235                          * mirror, that means that the requested area
5236                          * is not left of the left cursor
5237                          */
5238                         ret = -EIO;
5239                         btrfs_put_bbio(tmp_bbio);
5240                         goto out;
5241                 }
5242
5243                 /*
5244                  * process the rest of the function using the mirror_num
5245                  * of the source drive. Therefore look it up first.
5246                  * At the end, patch the device pointer to the one of the
5247                  * target drive.
5248                  */
5249                 for (i = 0; i < tmp_num_stripes; i++) {
5250                         if (tmp_bbio->stripes[i].dev->devid == srcdev_devid) {
5251                                 /*
5252                                  * In case of DUP, in order to keep it
5253                                  * simple, only add the mirror with the
5254                                  * lowest physical address
5255                                  */
5256                                 if (found &&
5257                                     physical_of_found <=
5258                                      tmp_bbio->stripes[i].physical)
5259                                         continue;
5260                                 index_srcdev = i;
5261                                 found = 1;
5262                                 physical_of_found =
5263                                         tmp_bbio->stripes[i].physical;
5264                         }
5265                 }
5266
5267                 if (found) {
5268                         mirror_num = index_srcdev + 1;
5269                         patch_the_first_stripe_for_dev_replace = 1;
5270                         physical_to_patch_in_first_stripe = physical_of_found;
5271                 } else {
5272                         WARN_ON(1);
5273                         ret = -EIO;
5274                         btrfs_put_bbio(tmp_bbio);
5275                         goto out;
5276                 }
5277
5278                 btrfs_put_bbio(tmp_bbio);
5279         } else if (mirror_num > map->num_stripes) {
5280                 mirror_num = 0;
5281         }
5282
5283         num_stripes = 1;
5284         stripe_index = 0;
5285         stripe_nr_orig = stripe_nr;
5286         stripe_nr_end = ALIGN(offset + *length, map->stripe_len);
5287         stripe_nr_end = div_u64(stripe_nr_end, map->stripe_len);
5288         stripe_end_offset = stripe_nr_end * map->stripe_len -
5289                             (offset + *length);
5290
5291         if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5292                 if (rw & REQ_DISCARD)
5293                         num_stripes = min_t(u64, map->num_stripes,
5294                                             stripe_nr_end - stripe_nr_orig);
5295                 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5296                                 &stripe_index);
5297                 if (!(rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)))
5298                         mirror_num = 1;
5299         } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
5300                 if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS))
5301                         num_stripes = map->num_stripes;
5302                 else if (mirror_num)
5303                         stripe_index = mirror_num - 1;
5304                 else {
5305                         stripe_index = find_live_mirror(fs_info, map, 0,
5306                                             map->num_stripes,
5307                                             current->pid % map->num_stripes,
5308                                             dev_replace_is_ongoing);
5309                         mirror_num = stripe_index + 1;
5310                 }
5311
5312         } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
5313                 if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) {
5314                         num_stripes = map->num_stripes;
5315                 } else if (mirror_num) {
5316                         stripe_index = mirror_num - 1;
5317                 } else {
5318                         mirror_num = 1;
5319                 }
5320
5321         } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5322                 u32 factor = map->num_stripes / map->sub_stripes;
5323
5324                 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
5325                 stripe_index *= map->sub_stripes;
5326
5327                 if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS))
5328                         num_stripes = map->sub_stripes;
5329                 else if (rw & REQ_DISCARD)
5330                         num_stripes = min_t(u64, map->sub_stripes *
5331                                             (stripe_nr_end - stripe_nr_orig),
5332                                             map->num_stripes);
5333                 else if (mirror_num)
5334                         stripe_index += mirror_num - 1;
5335                 else {
5336                         int old_stripe_index = stripe_index;
5337                         stripe_index = find_live_mirror(fs_info, map,
5338                                               stripe_index,
5339                                               map->sub_stripes, stripe_index +
5340                                               current->pid % map->sub_stripes,
5341                                               dev_replace_is_ongoing);
5342                         mirror_num = stripe_index - old_stripe_index + 1;
5343                 }
5344
5345         } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5346                 if (need_raid_map &&
5347                     ((rw & (REQ_WRITE | REQ_GET_READ_MIRRORS)) ||
5348                      mirror_num > 1)) {
5349                         /* push stripe_nr back to the start of the full stripe */
5350                         stripe_nr = div_u64(raid56_full_stripe_start,
5351                                         stripe_len * nr_data_stripes(map));
5352
5353                         /* RAID[56] write or recovery. Return all stripes */
5354                         num_stripes = map->num_stripes;
5355                         max_errors = nr_parity_stripes(map);
5356
5357                         *length = map->stripe_len;
5358                         stripe_index = 0;
5359                         stripe_offset = 0;
5360                 } else {
5361                         /*
5362                          * Mirror #0 or #1 means the original data block.
5363                          * Mirror #2 is RAID5 parity block.
5364                          * Mirror #3 is RAID6 Q block.
5365                          */
5366                         stripe_nr = div_u64_rem(stripe_nr,
5367                                         nr_data_stripes(map), &stripe_index);
5368                         if (mirror_num > 1)
5369                                 stripe_index = nr_data_stripes(map) +
5370                                                 mirror_num - 2;
5371
5372                         /* We distribute the parity blocks across stripes */
5373                         div_u64_rem(stripe_nr + stripe_index, map->num_stripes,
5374                                         &stripe_index);
5375                         if (!(rw & (REQ_WRITE | REQ_DISCARD |
5376                                     REQ_GET_READ_MIRRORS)) && mirror_num <= 1)
5377                                 mirror_num = 1;
5378                 }
5379         } else {
5380                 /*
5381                  * after this, stripe_nr is the number of stripes on this
5382                  * device we have to walk to find the data, and stripe_index is
5383                  * the number of our device in the stripe array
5384                  */
5385                 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5386                                 &stripe_index);
5387                 mirror_num = stripe_index + 1;
5388         }
5389         BUG_ON(stripe_index >= map->num_stripes);
5390
5391         num_alloc_stripes = num_stripes;
5392         if (dev_replace_is_ongoing) {
5393                 if (rw & (REQ_WRITE | REQ_DISCARD))
5394                         num_alloc_stripes <<= 1;
5395                 if (rw & REQ_GET_READ_MIRRORS)
5396                         num_alloc_stripes++;
5397                 tgtdev_indexes = num_stripes;
5398         }
5399
5400         bbio = alloc_btrfs_bio(num_alloc_stripes, tgtdev_indexes);
5401         if (!bbio) {
5402                 ret = -ENOMEM;
5403                 goto out;
5404         }
5405         if (dev_replace_is_ongoing)
5406                 bbio->tgtdev_map = (int *)(bbio->stripes + num_alloc_stripes);
5407
5408         /* build raid_map */
5409         if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK &&
5410             need_raid_map && ((rw & (REQ_WRITE | REQ_GET_READ_MIRRORS)) ||
5411             mirror_num > 1)) {
5412                 u64 tmp;
5413                 unsigned rot;
5414
5415                 bbio->raid_map = (u64 *)((void *)bbio->stripes +
5416                                  sizeof(struct btrfs_bio_stripe) *
5417                                  num_alloc_stripes +
5418                                  sizeof(int) * tgtdev_indexes);
5419
5420                 /* Work out the disk rotation on this stripe-set */
5421                 div_u64_rem(stripe_nr, num_stripes, &rot);
5422
5423                 /* Fill in the logical address of each stripe */
5424                 tmp = stripe_nr * nr_data_stripes(map);
5425                 for (i = 0; i < nr_data_stripes(map); i++)
5426                         bbio->raid_map[(i+rot) % num_stripes] =
5427                                 em->start + (tmp + i) * map->stripe_len;
5428
5429                 bbio->raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE;
5430                 if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5431                         bbio->raid_map[(i+rot+1) % num_stripes] =
5432                                 RAID6_Q_STRIPE;
5433         }
5434
5435         if (rw & REQ_DISCARD) {
5436                 u32 factor = 0;
5437                 u32 sub_stripes = 0;
5438                 u64 stripes_per_dev = 0;
5439                 u32 remaining_stripes = 0;
5440                 u32 last_stripe = 0;
5441
5442                 if (map->type &
5443                     (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID10)) {
5444                         if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5445                                 sub_stripes = 1;
5446                         else
5447                                 sub_stripes = map->sub_stripes;
5448
5449                         factor = map->num_stripes / sub_stripes;
5450                         stripes_per_dev = div_u64_rem(stripe_nr_end -
5451                                                       stripe_nr_orig,
5452                                                       factor,
5453                                                       &remaining_stripes);
5454                         div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
5455                         last_stripe *= sub_stripes;
5456                 }
5457
5458                 for (i = 0; i < num_stripes; i++) {
5459                         bbio->stripes[i].physical =
5460                                 map->stripes[stripe_index].physical +
5461                                 stripe_offset + stripe_nr * map->stripe_len;
5462                         bbio->stripes[i].dev = map->stripes[stripe_index].dev;
5463
5464                         if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5465                                          BTRFS_BLOCK_GROUP_RAID10)) {
5466                                 bbio->stripes[i].length = stripes_per_dev *
5467                                                           map->stripe_len;
5468
5469                                 if (i / sub_stripes < remaining_stripes)
5470                                         bbio->stripes[i].length +=
5471                                                 map->stripe_len;
5472
5473                                 /*
5474                                  * Special for the first stripe and
5475                                  * the last stripe:
5476                                  *
5477                                  * |-------|...|-------|
5478                                  *     |----------|
5479                                  *    off     end_off
5480                                  */
5481                                 if (i < sub_stripes)
5482                                         bbio->stripes[i].length -=
5483                                                 stripe_offset;
5484
5485                                 if (stripe_index >= last_stripe &&
5486                                     stripe_index <= (last_stripe +
5487                                                      sub_stripes - 1))
5488                                         bbio->stripes[i].length -=
5489                                                 stripe_end_offset;
5490
5491                                 if (i == sub_stripes - 1)
5492                                         stripe_offset = 0;
5493                         } else
5494                                 bbio->stripes[i].length = *length;
5495
5496                         stripe_index++;
5497                         if (stripe_index == map->num_stripes) {
5498                                 /* This could only happen for RAID0/10 */
5499                                 stripe_index = 0;
5500                                 stripe_nr++;
5501                         }
5502                 }
5503         } else {
5504                 for (i = 0; i < num_stripes; i++) {
5505                         bbio->stripes[i].physical =
5506                                 map->stripes[stripe_index].physical +
5507                                 stripe_offset +
5508                                 stripe_nr * map->stripe_len;
5509                         bbio->stripes[i].dev =
5510                                 map->stripes[stripe_index].dev;
5511                         stripe_index++;
5512                 }
5513         }
5514
5515         if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS))
5516                 max_errors = btrfs_chunk_max_errors(map);
5517
5518         if (bbio->raid_map)
5519                 sort_parity_stripes(bbio, num_stripes);
5520
5521         tgtdev_indexes = 0;
5522         if (dev_replace_is_ongoing && (rw & (REQ_WRITE | REQ_DISCARD)) &&
5523             dev_replace->tgtdev != NULL) {
5524                 int index_where_to_add;
5525                 u64 srcdev_devid = dev_replace->srcdev->devid;
5526
5527                 /*
5528                  * duplicate the write operations while the dev replace
5529                  * procedure is running. Since the copying of the old disk
5530                  * to the new disk takes place at run time while the
5531                  * filesystem is mounted writable, the regular write
5532                  * operations to the old disk have to be duplicated to go
5533                  * to the new disk as well.
5534                  * Note that device->missing is handled by the caller, and
5535                  * that the write to the old disk is already set up in the
5536                  * stripes array.
5537                  */
5538                 index_where_to_add = num_stripes;
5539                 for (i = 0; i < num_stripes; i++) {
5540                         if (bbio->stripes[i].dev->devid == srcdev_devid) {
5541                                 /* write to new disk, too */
5542                                 struct btrfs_bio_stripe *new =
5543                                         bbio->stripes + index_where_to_add;
5544                                 struct btrfs_bio_stripe *old =
5545                                         bbio->stripes + i;
5546
5547                                 new->physical = old->physical;
5548                                 new->length = old->length;
5549                                 new->dev = dev_replace->tgtdev;
5550                                 bbio->tgtdev_map[i] = index_where_to_add;
5551                                 index_where_to_add++;
5552                                 max_errors++;
5553                                 tgtdev_indexes++;
5554                         }
5555                 }
5556                 num_stripes = index_where_to_add;
5557         } else if (dev_replace_is_ongoing && (rw & REQ_GET_READ_MIRRORS) &&
5558                    dev_replace->tgtdev != NULL) {
5559                 u64 srcdev_devid = dev_replace->srcdev->devid;
5560                 int index_srcdev = 0;
5561                 int found = 0;
5562                 u64 physical_of_found = 0;
5563
5564                 /*
5565                  * During the dev-replace procedure, the target drive can
5566                  * also be used to read data in case it is needed to repair
5567                  * a corrupt block elsewhere. This is possible if the
5568                  * requested area is left of the left cursor. In this area,
5569                  * the target drive is a full copy of the source drive.
5570                  */
5571                 for (i = 0; i < num_stripes; i++) {
5572                         if (bbio->stripes[i].dev->devid == srcdev_devid) {
5573                                 /*
5574                                  * In case of DUP, in order to keep it
5575                                  * simple, only add the mirror with the
5576                                  * lowest physical address
5577                                  */
5578                                 if (found &&
5579                                     physical_of_found <=
5580                                      bbio->stripes[i].physical)
5581                                         continue;
5582                                 index_srcdev = i;
5583                                 found = 1;
5584                                 physical_of_found = bbio->stripes[i].physical;
5585                         }
5586                 }
5587                 if (found) {
5588                         if (physical_of_found + map->stripe_len <=
5589                             dev_replace->cursor_left) {
5590                                 struct btrfs_bio_stripe *tgtdev_stripe =
5591                                         bbio->stripes + num_stripes;
5592
5593                                 tgtdev_stripe->physical = physical_of_found;
5594                                 tgtdev_stripe->length =
5595                                         bbio->stripes[index_srcdev].length;
5596                                 tgtdev_stripe->dev = dev_replace->tgtdev;
5597                                 bbio->tgtdev_map[index_srcdev] = num_stripes;
5598
5599                                 tgtdev_indexes++;
5600                                 num_stripes++;
5601                         }
5602                 }
5603         }
5604
5605         *bbio_ret = bbio;
5606         bbio->map_type = map->type;
5607         bbio->num_stripes = num_stripes;
5608         bbio->max_errors = max_errors;
5609         bbio->mirror_num = mirror_num;
5610         bbio->num_tgtdevs = tgtdev_indexes;
5611
5612         /*
5613          * this is the case that REQ_READ && dev_replace_is_ongoing &&
5614          * mirror_num == num_stripes + 1 && dev_replace target drive is
5615          * available as a mirror
5616          */
5617         if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) {
5618                 WARN_ON(num_stripes > 1);
5619                 bbio->stripes[0].dev = dev_replace->tgtdev;
5620                 bbio->stripes[0].physical = physical_to_patch_in_first_stripe;
5621                 bbio->mirror_num = map->num_stripes + 1;
5622         }
5623 out:
5624         if (dev_replace_is_ongoing)
5625                 btrfs_dev_replace_unlock(dev_replace);
5626         free_extent_map(em);
5627         return ret;
5628 }
5629
5630 int btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
5631                       u64 logical, u64 *length,
5632                       struct btrfs_bio **bbio_ret, int mirror_num)
5633 {
5634         return __btrfs_map_block(fs_info, rw, logical, length, bbio_ret,
5635                                  mirror_num, 0);
5636 }
5637
5638 /* For Scrub/replace */
5639 int btrfs_map_sblock(struct btrfs_fs_info *fs_info, int rw,
5640                      u64 logical, u64 *length,
5641                      struct btrfs_bio **bbio_ret, int mirror_num,
5642                      int need_raid_map)
5643 {
5644         return __btrfs_map_block(fs_info, rw, logical, length, bbio_ret,
5645                                  mirror_num, need_raid_map);
5646 }
5647
5648 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
5649                      u64 chunk_start, u64 physical, u64 devid,
5650                      u64 **logical, int *naddrs, int *stripe_len)
5651 {
5652         struct extent_map_tree *em_tree = &map_tree->map_tree;
5653         struct extent_map *em;
5654         struct map_lookup *map;
5655         u64 *buf;
5656         u64 bytenr;
5657         u64 length;
5658         u64 stripe_nr;
5659         u64 rmap_len;
5660         int i, j, nr = 0;
5661
5662         read_lock(&em_tree->lock);
5663         em = lookup_extent_mapping(em_tree, chunk_start, 1);
5664         read_unlock(&em_tree->lock);
5665
5666         if (!em) {
5667                 printk(KERN_ERR "BTRFS: couldn't find em for chunk %Lu\n",
5668                        chunk_start);
5669                 return -EIO;
5670         }
5671
5672         if (em->start != chunk_start) {
5673                 printk(KERN_ERR "BTRFS: bad chunk start, em=%Lu, wanted=%Lu\n",
5674                        em->start, chunk_start);
5675                 free_extent_map(em);
5676                 return -EIO;
5677         }
5678         map = (struct map_lookup *)em->bdev;
5679
5680         length = em->len;
5681         rmap_len = map->stripe_len;
5682
5683         if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5684                 length = div_u64(length, map->num_stripes / map->sub_stripes);
5685         else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5686                 length = div_u64(length, map->num_stripes);
5687         else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5688                 length = div_u64(length, nr_data_stripes(map));
5689                 rmap_len = map->stripe_len * nr_data_stripes(map);
5690         }
5691
5692         buf = kcalloc(map->num_stripes, sizeof(u64), GFP_NOFS);
5693         BUG_ON(!buf); /* -ENOMEM */
5694
5695         for (i = 0; i < map->num_stripes; i++) {
5696                 if (devid && map->stripes[i].dev->devid != devid)
5697                         continue;
5698                 if (map->stripes[i].physical > physical ||
5699                     map->stripes[i].physical + length <= physical)
5700                         continue;
5701
5702                 stripe_nr = physical - map->stripes[i].physical;
5703                 stripe_nr = div_u64(stripe_nr, map->stripe_len);
5704
5705                 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5706                         stripe_nr = stripe_nr * map->num_stripes + i;
5707                         stripe_nr = div_u64(stripe_nr, map->sub_stripes);
5708                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5709                         stripe_nr = stripe_nr * map->num_stripes + i;
5710                 } /* else if RAID[56], multiply by nr_data_stripes().
5711                    * Alternatively, just use rmap_len below instead of
5712                    * map->stripe_len */
5713
5714                 bytenr = chunk_start + stripe_nr * rmap_len;
5715                 WARN_ON(nr >= map->num_stripes);
5716                 for (j = 0; j < nr; j++) {
5717                         if (buf[j] == bytenr)
5718                                 break;
5719                 }
5720                 if (j == nr) {
5721                         WARN_ON(nr >= map->num_stripes);
5722                         buf[nr++] = bytenr;
5723                 }
5724         }
5725
5726         *logical = buf;
5727         *naddrs = nr;
5728         *stripe_len = rmap_len;
5729
5730         free_extent_map(em);
5731         return 0;
5732 }
5733
5734 static inline void btrfs_end_bbio(struct btrfs_bio *bbio, struct bio *bio)
5735 {
5736         bio->bi_private = bbio->private;
5737         bio->bi_end_io = bbio->end_io;
5738         bio_endio(bio);
5739
5740         btrfs_put_bbio(bbio);
5741 }
5742
5743 static void btrfs_end_bio(struct bio *bio)
5744 {
5745         struct btrfs_bio *bbio = bio->bi_private;
5746         int is_orig_bio = 0;
5747
5748         if (bio->bi_error) {
5749                 atomic_inc(&bbio->error);
5750                 if (bio->bi_error == -EIO || bio->bi_error == -EREMOTEIO) {
5751                         unsigned int stripe_index =
5752                                 btrfs_io_bio(bio)->stripe_index;
5753                         struct btrfs_device *dev;
5754
5755                         BUG_ON(stripe_index >= bbio->num_stripes);
5756                         dev = bbio->stripes[stripe_index].dev;
5757                         if (dev->bdev) {
5758                                 if (bio->bi_rw & WRITE)
5759                                         btrfs_dev_stat_inc(dev,
5760                                                 BTRFS_DEV_STAT_WRITE_ERRS);
5761                                 else
5762                                         btrfs_dev_stat_inc(dev,
5763                                                 BTRFS_DEV_STAT_READ_ERRS);
5764                                 if ((bio->bi_rw & WRITE_FLUSH) == WRITE_FLUSH)
5765                                         btrfs_dev_stat_inc(dev,
5766                                                 BTRFS_DEV_STAT_FLUSH_ERRS);
5767                                 btrfs_dev_stat_print_on_error(dev);
5768                         }
5769                 }
5770         }
5771
5772         if (bio == bbio->orig_bio)
5773                 is_orig_bio = 1;
5774
5775         btrfs_bio_counter_dec(bbio->fs_info);
5776
5777         if (atomic_dec_and_test(&bbio->stripes_pending)) {
5778                 if (!is_orig_bio) {
5779                         bio_put(bio);
5780                         bio = bbio->orig_bio;
5781                 }
5782
5783                 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
5784                 /* only send an error to the higher layers if it is
5785                  * beyond the tolerance of the btrfs bio
5786                  */
5787                 if (atomic_read(&bbio->error) > bbio->max_errors) {
5788                         bio->bi_error = -EIO;
5789                 } else {
5790                         /*
5791                          * this bio is actually up to date, we didn't
5792                          * go over the max number of errors
5793                          */
5794                         bio->bi_error = 0;
5795                 }
5796
5797                 btrfs_end_bbio(bbio, bio);
5798         } else if (!is_orig_bio) {
5799                 bio_put(bio);
5800         }
5801 }
5802
5803 /*
5804  * see run_scheduled_bios for a description of why bios are collected for
5805  * async submit.
5806  *
5807  * This will add one bio to the pending list for a device and make sure
5808  * the work struct is scheduled.
5809  */
5810 static noinline void btrfs_schedule_bio(struct btrfs_root *root,
5811                                         struct btrfs_device *device,
5812                                         int rw, struct bio *bio)
5813 {
5814         int should_queue = 1;
5815         struct btrfs_pending_bios *pending_bios;
5816
5817         if (device->missing || !device->bdev) {
5818                 bio_io_error(bio);
5819                 return;
5820         }
5821
5822         /* don't bother with additional async steps for reads, right now */
5823         if (!(rw & REQ_WRITE)) {
5824                 bio_get(bio);
5825                 btrfsic_submit_bio(rw, bio);
5826                 bio_put(bio);
5827                 return;
5828         }
5829
5830         /*
5831          * nr_async_bios allows us to reliably return congestion to the
5832          * higher layers.  Otherwise, the async bio makes it appear we have
5833          * made progress against dirty pages when we've really just put it
5834          * on a queue for later
5835          */
5836         atomic_inc(&root->fs_info->nr_async_bios);
5837         WARN_ON(bio->bi_next);
5838         bio->bi_next = NULL;
5839         bio->bi_rw |= rw;
5840
5841         spin_lock(&device->io_lock);
5842         if (bio->bi_rw & REQ_SYNC)
5843                 pending_bios = &device->pending_sync_bios;
5844         else
5845                 pending_bios = &device->pending_bios;
5846
5847         if (pending_bios->tail)
5848                 pending_bios->tail->bi_next = bio;
5849
5850         pending_bios->tail = bio;
5851         if (!pending_bios->head)
5852                 pending_bios->head = bio;
5853         if (device->running_pending)
5854                 should_queue = 0;
5855
5856         spin_unlock(&device->io_lock);
5857
5858         if (should_queue)
5859                 btrfs_queue_work(root->fs_info->submit_workers,
5860                                  &device->work);
5861 }
5862
5863 static void submit_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio,
5864                               struct bio *bio, u64 physical, int dev_nr,
5865                               int rw, int async)
5866 {
5867         struct btrfs_device *dev = bbio->stripes[dev_nr].dev;
5868
5869         bio->bi_private = bbio;
5870         btrfs_io_bio(bio)->stripe_index = dev_nr;
5871         bio->bi_end_io = btrfs_end_bio;
5872         bio->bi_iter.bi_sector = physical >> 9;
5873 #ifdef DEBUG
5874         {
5875                 struct rcu_string *name;
5876
5877                 rcu_read_lock();
5878                 name = rcu_dereference(dev->name);
5879                 pr_debug("btrfs_map_bio: rw %d, sector=%llu, dev=%lu "
5880                          "(%s id %llu), size=%u\n", rw,
5881                          (u64)bio->bi_iter.bi_sector, (u_long)dev->bdev->bd_dev,
5882                          name->str, dev->devid, bio->bi_iter.bi_size);
5883                 rcu_read_unlock();
5884         }
5885 #endif
5886         bio->bi_bdev = dev->bdev;
5887
5888         btrfs_bio_counter_inc_noblocked(root->fs_info);
5889
5890         if (async)
5891                 btrfs_schedule_bio(root, dev, rw, bio);
5892         else
5893                 btrfsic_submit_bio(rw, bio);
5894 }
5895
5896 static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical)
5897 {
5898         atomic_inc(&bbio->error);
5899         if (atomic_dec_and_test(&bbio->stripes_pending)) {
5900                 /* Shoud be the original bio. */
5901                 WARN_ON(bio != bbio->orig_bio);
5902
5903                 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
5904                 bio->bi_iter.bi_sector = logical >> 9;
5905                 bio->bi_error = -EIO;
5906                 btrfs_end_bbio(bbio, bio);
5907         }
5908 }
5909
5910 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
5911                   int mirror_num, int async_submit)
5912 {
5913         struct btrfs_device *dev;
5914         struct bio *first_bio = bio;
5915         u64 logical = (u64)bio->bi_iter.bi_sector << 9;
5916         u64 length = 0;
5917         u64 map_length;
5918         int ret;
5919         int dev_nr;
5920         int total_devs;
5921         struct btrfs_bio *bbio = NULL;
5922
5923         length = bio->bi_iter.bi_size;
5924         map_length = length;
5925
5926         btrfs_bio_counter_inc_blocked(root->fs_info);
5927         ret = __btrfs_map_block(root->fs_info, rw, logical, &map_length, &bbio,
5928                               mirror_num, 1);
5929         if (ret) {
5930                 btrfs_bio_counter_dec(root->fs_info);
5931                 return ret;
5932         }
5933
5934         total_devs = bbio->num_stripes;
5935         bbio->orig_bio = first_bio;
5936         bbio->private = first_bio->bi_private;
5937         bbio->end_io = first_bio->bi_end_io;
5938         bbio->fs_info = root->fs_info;
5939         atomic_set(&bbio->stripes_pending, bbio->num_stripes);
5940
5941         if (bbio->raid_map) {
5942                 /* In this case, map_length has been set to the length of
5943                    a single stripe; not the whole write */
5944                 if (rw & WRITE) {
5945                         ret = raid56_parity_write(root, bio, bbio, map_length);
5946                 } else {
5947                         ret = raid56_parity_recover(root, bio, bbio, map_length,
5948                                                     mirror_num, 1);
5949                 }
5950
5951                 btrfs_bio_counter_dec(root->fs_info);
5952                 return ret;
5953         }
5954
5955         if (map_length < length) {
5956                 btrfs_crit(root->fs_info, "mapping failed logical %llu bio len %llu len %llu",
5957                         logical, length, map_length);
5958                 BUG();
5959         }
5960
5961         for (dev_nr = 0; dev_nr < total_devs; dev_nr++) {
5962                 dev = bbio->stripes[dev_nr].dev;
5963                 if (!dev || !dev->bdev || (rw & WRITE && !dev->writeable)) {
5964                         bbio_error(bbio, first_bio, logical);
5965                         continue;
5966                 }
5967
5968                 if (dev_nr < total_devs - 1) {
5969                         bio = btrfs_bio_clone(first_bio, GFP_NOFS);
5970                         BUG_ON(!bio); /* -ENOMEM */
5971                 } else
5972                         bio = first_bio;
5973
5974                 submit_stripe_bio(root, bbio, bio,
5975                                   bbio->stripes[dev_nr].physical, dev_nr, rw,
5976                                   async_submit);
5977         }
5978         btrfs_bio_counter_dec(root->fs_info);
5979         return 0;
5980 }
5981
5982 struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid,
5983                                        u8 *uuid, u8 *fsid)
5984 {
5985         struct btrfs_device *device;
5986         struct btrfs_fs_devices *cur_devices;
5987
5988         cur_devices = fs_info->fs_devices;
5989         while (cur_devices) {
5990                 if (!fsid ||
5991                     !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
5992                         device = __find_device(&cur_devices->devices,
5993                                                devid, uuid);
5994                         if (device)
5995                                 return device;
5996                 }
5997                 cur_devices = cur_devices->seed;
5998         }
5999         return NULL;
6000 }
6001
6002 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
6003                                             struct btrfs_fs_devices *fs_devices,
6004                                             u64 devid, u8 *dev_uuid)
6005 {
6006         struct btrfs_device *device;
6007
6008         device = btrfs_alloc_device(NULL, &devid, dev_uuid);
6009         if (IS_ERR(device))
6010                 return NULL;
6011
6012         list_add(&device->dev_list, &fs_devices->devices);
6013         device->fs_devices = fs_devices;
6014         fs_devices->num_devices++;
6015
6016         device->missing = 1;
6017         fs_devices->missing_devices++;
6018
6019         return device;
6020 }
6021
6022 /**
6023  * btrfs_alloc_device - allocate struct btrfs_device
6024  * @fs_info:    used only for generating a new devid, can be NULL if
6025  *              devid is provided (i.e. @devid != NULL).
6026  * @devid:      a pointer to devid for this device.  If NULL a new devid
6027  *              is generated.
6028  * @uuid:       a pointer to UUID for this device.  If NULL a new UUID
6029  *              is generated.
6030  *
6031  * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
6032  * on error.  Returned struct is not linked onto any lists and can be
6033  * destroyed with kfree() right away.
6034  */
6035 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
6036                                         const u64 *devid,
6037                                         const u8 *uuid)
6038 {
6039         struct btrfs_device *dev;
6040         u64 tmp;
6041
6042         if (WARN_ON(!devid && !fs_info))
6043                 return ERR_PTR(-EINVAL);
6044
6045         dev = __alloc_device();
6046         if (IS_ERR(dev))
6047                 return dev;
6048
6049         if (devid)
6050                 tmp = *devid;
6051         else {
6052                 int ret;
6053
6054                 ret = find_next_devid(fs_info, &tmp);
6055                 if (ret) {
6056                         kfree(dev);
6057                         return ERR_PTR(ret);
6058                 }
6059         }
6060         dev->devid = tmp;
6061
6062         if (uuid)
6063                 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE);
6064         else
6065                 generate_random_uuid(dev->uuid);
6066
6067         btrfs_init_work(&dev->work, btrfs_submit_helper,
6068                         pending_bios_fn, NULL, NULL);
6069
6070         return dev;
6071 }
6072
6073 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
6074                           struct extent_buffer *leaf,
6075                           struct btrfs_chunk *chunk)
6076 {
6077         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
6078         struct map_lookup *map;
6079         struct extent_map *em;
6080         u64 logical;
6081         u64 length;
6082         u64 devid;
6083         u8 uuid[BTRFS_UUID_SIZE];
6084         int num_stripes;
6085         int ret;
6086         int i;
6087
6088         logical = key->offset;
6089         length = btrfs_chunk_length(leaf, chunk);
6090
6091         read_lock(&map_tree->map_tree.lock);
6092         em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
6093         read_unlock(&map_tree->map_tree.lock);
6094
6095         /* already mapped? */
6096         if (em && em->start <= logical && em->start + em->len > logical) {
6097                 free_extent_map(em);
6098                 return 0;
6099         } else if (em) {
6100                 free_extent_map(em);
6101         }
6102
6103         em = alloc_extent_map();
6104         if (!em)
6105                 return -ENOMEM;
6106         num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6107         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
6108         if (!map) {
6109                 free_extent_map(em);
6110                 return -ENOMEM;
6111         }
6112
6113         set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
6114         em->bdev = (struct block_device *)map;
6115         em->start = logical;
6116         em->len = length;
6117         em->orig_start = 0;
6118         em->block_start = 0;
6119         em->block_len = em->len;
6120
6121         map->num_stripes = num_stripes;
6122         map->io_width = btrfs_chunk_io_width(leaf, chunk);
6123         map->io_align = btrfs_chunk_io_align(leaf, chunk);
6124         map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
6125         map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6126         map->type = btrfs_chunk_type(leaf, chunk);
6127         map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
6128         for (i = 0; i < num_stripes; i++) {
6129                 map->stripes[i].physical =
6130                         btrfs_stripe_offset_nr(leaf, chunk, i);
6131                 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
6132                 read_extent_buffer(leaf, uuid, (unsigned long)
6133                                    btrfs_stripe_dev_uuid_nr(chunk, i),
6134                                    BTRFS_UUID_SIZE);
6135                 map->stripes[i].dev = btrfs_find_device(root->fs_info, devid,
6136                                                         uuid, NULL);
6137                 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
6138                         free_extent_map(em);
6139                         return -EIO;
6140                 }
6141                 if (!map->stripes[i].dev) {
6142                         map->stripes[i].dev =
6143                                 add_missing_dev(root, root->fs_info->fs_devices,
6144                                                 devid, uuid);
6145                         if (!map->stripes[i].dev) {
6146                                 free_extent_map(em);
6147                                 return -EIO;
6148                         }
6149                         btrfs_warn(root->fs_info, "devid %llu uuid %pU is missing",
6150                                                 devid, uuid);
6151                 }
6152                 map->stripes[i].dev->in_fs_metadata = 1;
6153         }
6154
6155         write_lock(&map_tree->map_tree.lock);
6156         ret = add_extent_mapping(&map_tree->map_tree, em, 0);
6157         write_unlock(&map_tree->map_tree.lock);
6158         BUG_ON(ret); /* Tree corruption */
6159         free_extent_map(em);
6160
6161         return 0;
6162 }
6163
6164 static void fill_device_from_item(struct extent_buffer *leaf,
6165                                  struct btrfs_dev_item *dev_item,
6166                                  struct btrfs_device *device)
6167 {
6168         unsigned long ptr;
6169
6170         device->devid = btrfs_device_id(leaf, dev_item);
6171         device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
6172         device->total_bytes = device->disk_total_bytes;
6173         device->commit_total_bytes = device->disk_total_bytes;
6174         device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
6175         device->commit_bytes_used = device->bytes_used;
6176         device->type = btrfs_device_type(leaf, dev_item);
6177         device->io_align = btrfs_device_io_align(leaf, dev_item);
6178         device->io_width = btrfs_device_io_width(leaf, dev_item);
6179         device->sector_size = btrfs_device_sector_size(leaf, dev_item);
6180         WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
6181         device->is_tgtdev_for_dev_replace = 0;
6182
6183         ptr = btrfs_device_uuid(dev_item);
6184         read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
6185 }
6186
6187 static struct btrfs_fs_devices *open_seed_devices(struct btrfs_root *root,
6188                                                   u8 *fsid)
6189 {
6190         struct btrfs_fs_devices *fs_devices;
6191         int ret;
6192
6193         BUG_ON(!mutex_is_locked(&uuid_mutex));
6194
6195         fs_devices = root->fs_info->fs_devices->seed;
6196         while (fs_devices) {
6197                 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE))
6198                         return fs_devices;
6199
6200                 fs_devices = fs_devices->seed;
6201         }
6202
6203         fs_devices = find_fsid(fsid);
6204         if (!fs_devices) {
6205                 if (!btrfs_test_opt(root, DEGRADED))
6206                         return ERR_PTR(-ENOENT);
6207
6208                 fs_devices = alloc_fs_devices(fsid);
6209                 if (IS_ERR(fs_devices))
6210                         return fs_devices;
6211
6212                 fs_devices->seeding = 1;
6213                 fs_devices->opened = 1;
6214                 return fs_devices;
6215         }
6216
6217         fs_devices = clone_fs_devices(fs_devices);
6218         if (IS_ERR(fs_devices))
6219                 return fs_devices;
6220
6221         ret = __btrfs_open_devices(fs_devices, FMODE_READ,
6222                                    root->fs_info->bdev_holder);
6223         if (ret) {
6224                 free_fs_devices(fs_devices);
6225                 fs_devices = ERR_PTR(ret);
6226                 goto out;
6227         }
6228
6229         if (!fs_devices->seeding) {
6230                 __btrfs_close_devices(fs_devices);
6231                 free_fs_devices(fs_devices);
6232                 fs_devices = ERR_PTR(-EINVAL);
6233                 goto out;
6234         }
6235
6236         fs_devices->seed = root->fs_info->fs_devices->seed;
6237         root->fs_info->fs_devices->seed = fs_devices;
6238 out:
6239         return fs_devices;
6240 }
6241
6242 static int read_one_dev(struct btrfs_root *root,
6243                         struct extent_buffer *leaf,
6244                         struct btrfs_dev_item *dev_item)
6245 {
6246         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
6247         struct btrfs_device *device;
6248         u64 devid;
6249         int ret;
6250         u8 fs_uuid[BTRFS_UUID_SIZE];
6251         u8 dev_uuid[BTRFS_UUID_SIZE];
6252
6253         devid = btrfs_device_id(leaf, dev_item);
6254         read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
6255                            BTRFS_UUID_SIZE);
6256         read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
6257                            BTRFS_UUID_SIZE);
6258
6259         if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
6260                 fs_devices = open_seed_devices(root, fs_uuid);
6261                 if (IS_ERR(fs_devices))
6262                         return PTR_ERR(fs_devices);
6263         }
6264
6265         device = btrfs_find_device(root->fs_info, devid, dev_uuid, fs_uuid);
6266         if (!device) {
6267                 if (!btrfs_test_opt(root, DEGRADED))
6268                         return -EIO;
6269
6270                 device = add_missing_dev(root, fs_devices, devid, dev_uuid);
6271                 if (!device)
6272                         return -ENOMEM;
6273                 btrfs_warn(root->fs_info, "devid %llu uuid %pU missing",
6274                                 devid, dev_uuid);
6275         } else {
6276                 if (!device->bdev && !btrfs_test_opt(root, DEGRADED))
6277                         return -EIO;
6278
6279                 if(!device->bdev && !device->missing) {
6280                         /*
6281                          * this happens when a device that was properly setup
6282                          * in the device info lists suddenly goes bad.
6283                          * device->bdev is NULL, and so we have to set
6284                          * device->missing to one here
6285                          */
6286                         device->fs_devices->missing_devices++;
6287                         device->missing = 1;
6288                 }
6289
6290                 /* Move the device to its own fs_devices */
6291                 if (device->fs_devices != fs_devices) {
6292                         ASSERT(device->missing);
6293
6294                         list_move(&device->dev_list, &fs_devices->devices);
6295                         device->fs_devices->num_devices--;
6296                         fs_devices->num_devices++;
6297
6298                         device->fs_devices->missing_devices--;
6299                         fs_devices->missing_devices++;
6300
6301                         device->fs_devices = fs_devices;
6302                 }
6303         }
6304
6305         if (device->fs_devices != root->fs_info->fs_devices) {
6306                 BUG_ON(device->writeable);
6307                 if (device->generation !=
6308                     btrfs_device_generation(leaf, dev_item))
6309                         return -EINVAL;
6310         }
6311
6312         fill_device_from_item(leaf, dev_item, device);
6313         device->in_fs_metadata = 1;
6314         if (device->writeable && !device->is_tgtdev_for_dev_replace) {
6315                 device->fs_devices->total_rw_bytes += device->total_bytes;
6316                 spin_lock(&root->fs_info->free_chunk_lock);
6317                 root->fs_info->free_chunk_space += device->total_bytes -
6318                         device->bytes_used;
6319                 spin_unlock(&root->fs_info->free_chunk_lock);
6320         }
6321         ret = 0;
6322         return ret;
6323 }
6324
6325 int btrfs_read_sys_array(struct btrfs_root *root)
6326 {
6327         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
6328         struct extent_buffer *sb;
6329         struct btrfs_disk_key *disk_key;
6330         struct btrfs_chunk *chunk;
6331         u8 *array_ptr;
6332         unsigned long sb_array_offset;
6333         int ret = 0;
6334         u32 num_stripes;
6335         u32 array_size;
6336         u32 len = 0;
6337         u32 cur_offset;
6338         struct btrfs_key key;
6339
6340         ASSERT(BTRFS_SUPER_INFO_SIZE <= root->nodesize);
6341         /*
6342          * This will create extent buffer of nodesize, superblock size is
6343          * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will
6344          * overallocate but we can keep it as-is, only the first page is used.
6345          */
6346         sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET);
6347         if (!sb)
6348                 return -ENOMEM;
6349         btrfs_set_buffer_uptodate(sb);
6350         btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
6351         /*
6352          * The sb extent buffer is artifical and just used to read the system array.
6353          * btrfs_set_buffer_uptodate() call does not properly mark all it's
6354          * pages up-to-date when the page is larger: extent does not cover the
6355          * whole page and consequently check_page_uptodate does not find all
6356          * the page's extents up-to-date (the hole beyond sb),
6357          * write_extent_buffer then triggers a WARN_ON.
6358          *
6359          * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
6360          * but sb spans only this function. Add an explicit SetPageUptodate call
6361          * to silence the warning eg. on PowerPC 64.
6362          */
6363         if (PAGE_CACHE_SIZE > BTRFS_SUPER_INFO_SIZE)
6364                 SetPageUptodate(sb->pages[0]);
6365
6366         write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
6367         array_size = btrfs_super_sys_array_size(super_copy);
6368
6369         array_ptr = super_copy->sys_chunk_array;
6370         sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array);
6371         cur_offset = 0;
6372
6373         while (cur_offset < array_size) {
6374                 disk_key = (struct btrfs_disk_key *)array_ptr;
6375                 len = sizeof(*disk_key);
6376                 if (cur_offset + len > array_size)
6377                         goto out_short_read;
6378
6379                 btrfs_disk_key_to_cpu(&key, disk_key);
6380
6381                 array_ptr += len;
6382                 sb_array_offset += len;
6383                 cur_offset += len;
6384
6385                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
6386                         chunk = (struct btrfs_chunk *)sb_array_offset;
6387                         /*
6388                          * At least one btrfs_chunk with one stripe must be
6389                          * present, exact stripe count check comes afterwards
6390                          */
6391                         len = btrfs_chunk_item_size(1);
6392                         if (cur_offset + len > array_size)
6393                                 goto out_short_read;
6394
6395                         num_stripes = btrfs_chunk_num_stripes(sb, chunk);
6396                         len = btrfs_chunk_item_size(num_stripes);
6397                         if (cur_offset + len > array_size)
6398                                 goto out_short_read;
6399
6400                         ret = read_one_chunk(root, &key, sb, chunk);
6401                         if (ret)
6402                                 break;
6403                 } else {
6404                         ret = -EIO;
6405                         break;
6406                 }
6407                 array_ptr += len;
6408                 sb_array_offset += len;
6409                 cur_offset += len;
6410         }
6411         free_extent_buffer(sb);
6412         return ret;
6413
6414 out_short_read:
6415         printk(KERN_ERR "BTRFS: sys_array too short to read %u bytes at offset %u\n",
6416                         len, cur_offset);
6417         free_extent_buffer(sb);
6418         return -EIO;
6419 }
6420
6421 int btrfs_read_chunk_tree(struct btrfs_root *root)
6422 {
6423         struct btrfs_path *path;
6424         struct extent_buffer *leaf;
6425         struct btrfs_key key;
6426         struct btrfs_key found_key;
6427         int ret;
6428         int slot;
6429
6430         root = root->fs_info->chunk_root;
6431
6432         path = btrfs_alloc_path();
6433         if (!path)
6434                 return -ENOMEM;
6435
6436         mutex_lock(&uuid_mutex);
6437         lock_chunks(root);
6438
6439         /*
6440          * Read all device items, and then all the chunk items. All
6441          * device items are found before any chunk item (their object id
6442          * is smaller than the lowest possible object id for a chunk
6443          * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
6444          */
6445         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
6446         key.offset = 0;
6447         key.type = 0;
6448         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6449         if (ret < 0)
6450                 goto error;
6451         while (1) {
6452                 leaf = path->nodes[0];
6453                 slot = path->slots[0];
6454                 if (slot >= btrfs_header_nritems(leaf)) {
6455                         ret = btrfs_next_leaf(root, path);
6456                         if (ret == 0)
6457                                 continue;
6458                         if (ret < 0)
6459                                 goto error;
6460                         break;
6461                 }
6462                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6463                 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
6464                         struct btrfs_dev_item *dev_item;
6465                         dev_item = btrfs_item_ptr(leaf, slot,
6466                                                   struct btrfs_dev_item);
6467                         ret = read_one_dev(root, leaf, dev_item);
6468                         if (ret)
6469                                 goto error;
6470                 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
6471                         struct btrfs_chunk *chunk;
6472                         chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
6473                         ret = read_one_chunk(root, &found_key, leaf, chunk);
6474                         if (ret)
6475                                 goto error;
6476                 }
6477                 path->slots[0]++;
6478         }
6479         ret = 0;
6480 error:
6481         unlock_chunks(root);
6482         mutex_unlock(&uuid_mutex);
6483
6484         btrfs_free_path(path);
6485         return ret;
6486 }
6487
6488 void btrfs_init_devices_late(struct btrfs_fs_info *fs_info)
6489 {
6490         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6491         struct btrfs_device *device;
6492
6493         while (fs_devices) {
6494                 mutex_lock(&fs_devices->device_list_mutex);
6495                 list_for_each_entry(device, &fs_devices->devices, dev_list)
6496                         device->dev_root = fs_info->dev_root;
6497                 mutex_unlock(&fs_devices->device_list_mutex);
6498
6499                 fs_devices = fs_devices->seed;
6500         }
6501 }
6502
6503 static void __btrfs_reset_dev_stats(struct btrfs_device *dev)
6504 {
6505         int i;
6506
6507         for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6508                 btrfs_dev_stat_reset(dev, i);
6509 }
6510
6511 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
6512 {
6513         struct btrfs_key key;
6514         struct btrfs_key found_key;
6515         struct btrfs_root *dev_root = fs_info->dev_root;
6516         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6517         struct extent_buffer *eb;
6518         int slot;
6519         int ret = 0;
6520         struct btrfs_device *device;
6521         struct btrfs_path *path = NULL;
6522         int i;
6523
6524         path = btrfs_alloc_path();
6525         if (!path) {
6526                 ret = -ENOMEM;
6527                 goto out;
6528         }
6529
6530         mutex_lock(&fs_devices->device_list_mutex);
6531         list_for_each_entry(device, &fs_devices->devices, dev_list) {
6532                 int item_size;
6533                 struct btrfs_dev_stats_item *ptr;
6534
6535                 key.objectid = 0;
6536                 key.type = BTRFS_DEV_STATS_KEY;
6537                 key.offset = device->devid;
6538                 ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
6539                 if (ret) {
6540                         __btrfs_reset_dev_stats(device);
6541                         device->dev_stats_valid = 1;
6542                         btrfs_release_path(path);
6543                         continue;
6544                 }
6545                 slot = path->slots[0];
6546                 eb = path->nodes[0];
6547                 btrfs_item_key_to_cpu(eb, &found_key, slot);
6548                 item_size = btrfs_item_size_nr(eb, slot);
6549
6550                 ptr = btrfs_item_ptr(eb, slot,
6551                                      struct btrfs_dev_stats_item);
6552
6553                 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
6554                         if (item_size >= (1 + i) * sizeof(__le64))
6555                                 btrfs_dev_stat_set(device, i,
6556                                         btrfs_dev_stats_value(eb, ptr, i));
6557                         else
6558                                 btrfs_dev_stat_reset(device, i);
6559                 }
6560
6561                 device->dev_stats_valid = 1;
6562                 btrfs_dev_stat_print_on_load(device);
6563                 btrfs_release_path(path);
6564         }
6565         mutex_unlock(&fs_devices->device_list_mutex);
6566
6567 out:
6568         btrfs_free_path(path);
6569         return ret < 0 ? ret : 0;
6570 }
6571
6572 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
6573                                 struct btrfs_root *dev_root,
6574                                 struct btrfs_device *device)
6575 {
6576         struct btrfs_path *path;
6577         struct btrfs_key key;
6578         struct extent_buffer *eb;
6579         struct btrfs_dev_stats_item *ptr;
6580         int ret;
6581         int i;
6582
6583         key.objectid = 0;
6584         key.type = BTRFS_DEV_STATS_KEY;
6585         key.offset = device->devid;
6586
6587         path = btrfs_alloc_path();
6588         BUG_ON(!path);
6589         ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
6590         if (ret < 0) {
6591                 btrfs_warn_in_rcu(dev_root->fs_info,
6592                         "error %d while searching for dev_stats item for device %s",
6593                               ret, rcu_str_deref(device->name));
6594                 goto out;
6595         }
6596
6597         if (ret == 0 &&
6598             btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
6599                 /* need to delete old one and insert a new one */
6600                 ret = btrfs_del_item(trans, dev_root, path);
6601                 if (ret != 0) {
6602                         btrfs_warn_in_rcu(dev_root->fs_info,
6603                                 "delete too small dev_stats item for device %s failed %d",
6604                                       rcu_str_deref(device->name), ret);
6605                         goto out;
6606                 }
6607                 ret = 1;
6608         }
6609
6610         if (ret == 1) {
6611                 /* need to insert a new item */
6612                 btrfs_release_path(path);
6613                 ret = btrfs_insert_empty_item(trans, dev_root, path,
6614                                               &key, sizeof(*ptr));
6615                 if (ret < 0) {
6616                         btrfs_warn_in_rcu(dev_root->fs_info,
6617                                 "insert dev_stats item for device %s failed %d",
6618                                 rcu_str_deref(device->name), ret);
6619                         goto out;
6620                 }
6621         }
6622
6623         eb = path->nodes[0];
6624         ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
6625         for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6626                 btrfs_set_dev_stats_value(eb, ptr, i,
6627                                           btrfs_dev_stat_read(device, i));
6628         btrfs_mark_buffer_dirty(eb);
6629
6630 out:
6631         btrfs_free_path(path);
6632         return ret;
6633 }
6634
6635 /*
6636  * called from commit_transaction. Writes all changed device stats to disk.
6637  */
6638 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans,
6639                         struct btrfs_fs_info *fs_info)
6640 {
6641         struct btrfs_root *dev_root = fs_info->dev_root;
6642         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6643         struct btrfs_device *device;
6644         int stats_cnt;
6645         int ret = 0;
6646
6647         mutex_lock(&fs_devices->device_list_mutex);
6648         list_for_each_entry(device, &fs_devices->devices, dev_list) {
6649                 if (!device->dev_stats_valid || !btrfs_dev_stats_dirty(device))
6650                         continue;
6651
6652                 stats_cnt = atomic_read(&device->dev_stats_ccnt);
6653                 ret = update_dev_stat_item(trans, dev_root, device);
6654                 if (!ret)
6655                         atomic_sub(stats_cnt, &device->dev_stats_ccnt);
6656         }
6657         mutex_unlock(&fs_devices->device_list_mutex);
6658
6659         return ret;
6660 }
6661
6662 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
6663 {
6664         btrfs_dev_stat_inc(dev, index);
6665         btrfs_dev_stat_print_on_error(dev);
6666 }
6667
6668 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
6669 {
6670         if (!dev->dev_stats_valid)
6671                 return;
6672         btrfs_err_rl_in_rcu(dev->dev_root->fs_info,
6673                 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
6674                            rcu_str_deref(dev->name),
6675                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
6676                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
6677                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
6678                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
6679                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
6680 }
6681
6682 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
6683 {
6684         int i;
6685
6686         for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6687                 if (btrfs_dev_stat_read(dev, i) != 0)
6688                         break;
6689         if (i == BTRFS_DEV_STAT_VALUES_MAX)
6690                 return; /* all values == 0, suppress message */
6691
6692         btrfs_info_in_rcu(dev->dev_root->fs_info,
6693                 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
6694                rcu_str_deref(dev->name),
6695                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
6696                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
6697                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
6698                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
6699                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
6700 }
6701
6702 int btrfs_get_dev_stats(struct btrfs_root *root,
6703                         struct btrfs_ioctl_get_dev_stats *stats)
6704 {
6705         struct btrfs_device *dev;
6706         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
6707         int i;
6708
6709         mutex_lock(&fs_devices->device_list_mutex);
6710         dev = btrfs_find_device(root->fs_info, stats->devid, NULL, NULL);
6711         mutex_unlock(&fs_devices->device_list_mutex);
6712
6713         if (!dev) {
6714                 btrfs_warn(root->fs_info, "get dev_stats failed, device not found");
6715                 return -ENODEV;
6716         } else if (!dev->dev_stats_valid) {
6717                 btrfs_warn(root->fs_info, "get dev_stats failed, not yet valid");
6718                 return -ENODEV;
6719         } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
6720                 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
6721                         if (stats->nr_items > i)
6722                                 stats->values[i] =
6723                                         btrfs_dev_stat_read_and_reset(dev, i);
6724                         else
6725                                 btrfs_dev_stat_reset(dev, i);
6726                 }
6727         } else {
6728                 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6729                         if (stats->nr_items > i)
6730                                 stats->values[i] = btrfs_dev_stat_read(dev, i);
6731         }
6732         if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
6733                 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
6734         return 0;
6735 }
6736
6737 void btrfs_scratch_superblocks(struct block_device *bdev, char *device_path)
6738 {
6739         struct buffer_head *bh;
6740         struct btrfs_super_block *disk_super;
6741         int copy_num;
6742
6743         if (!bdev)
6744                 return;
6745
6746         for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX;
6747                 copy_num++) {
6748
6749                 if (btrfs_read_dev_one_super(bdev, copy_num, &bh))
6750                         continue;
6751
6752                 disk_super = (struct btrfs_super_block *)bh->b_data;
6753
6754                 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
6755                 set_buffer_dirty(bh);
6756                 sync_dirty_buffer(bh);
6757                 brelse(bh);
6758         }
6759
6760         /* Notify udev that device has changed */
6761         btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
6762
6763         /* Update ctime/mtime for device path for libblkid */
6764         update_dev_time(device_path);
6765 }
6766
6767 /*
6768  * Update the size of all devices, which is used for writing out the
6769  * super blocks.
6770  */
6771 void btrfs_update_commit_device_size(struct btrfs_fs_info *fs_info)
6772 {
6773         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6774         struct btrfs_device *curr, *next;
6775
6776         if (list_empty(&fs_devices->resized_devices))
6777                 return;
6778
6779         mutex_lock(&fs_devices->device_list_mutex);
6780         lock_chunks(fs_info->dev_root);
6781         list_for_each_entry_safe(curr, next, &fs_devices->resized_devices,
6782                                  resized_list) {
6783                 list_del_init(&curr->resized_list);
6784                 curr->commit_total_bytes = curr->disk_total_bytes;
6785         }
6786         unlock_chunks(fs_info->dev_root);
6787         mutex_unlock(&fs_devices->device_list_mutex);
6788 }
6789
6790 /* Must be invoked during the transaction commit */
6791 void btrfs_update_commit_device_bytes_used(struct btrfs_root *root,
6792                                         struct btrfs_transaction *transaction)
6793 {
6794         struct extent_map *em;
6795         struct map_lookup *map;
6796         struct btrfs_device *dev;
6797         int i;
6798
6799         if (list_empty(&transaction->pending_chunks))
6800                 return;
6801
6802         /* In order to kick the device replace finish process */
6803         lock_chunks(root);
6804         list_for_each_entry(em, &transaction->pending_chunks, list) {
6805                 map = (struct map_lookup *)em->bdev;
6806
6807                 for (i = 0; i < map->num_stripes; i++) {
6808                         dev = map->stripes[i].dev;
6809                         dev->commit_bytes_used = dev->bytes_used;
6810                 }
6811         }
6812         unlock_chunks(root);
6813 }
6814
6815 void btrfs_set_fs_info_ptr(struct btrfs_fs_info *fs_info)
6816 {
6817         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6818         while (fs_devices) {
6819                 fs_devices->fs_info = fs_info;
6820                 fs_devices = fs_devices->seed;
6821         }
6822 }
6823
6824 void btrfs_reset_fs_info_ptr(struct btrfs_fs_info *fs_info)
6825 {
6826         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6827         while (fs_devices) {
6828                 fs_devices->fs_info = NULL;
6829                 fs_devices = fs_devices->seed;
6830         }
6831 }
6832
6833 void btrfs_close_one_device(struct btrfs_device *device)
6834 {
6835         struct btrfs_fs_devices *fs_devices = device->fs_devices;
6836         struct btrfs_device *new_device;
6837         struct rcu_string *name;
6838
6839         if (device->bdev)
6840                 fs_devices->open_devices--;
6841
6842         if (device->writeable &&
6843             device->devid != BTRFS_DEV_REPLACE_DEVID) {
6844                 list_del_init(&device->dev_alloc_list);
6845                 fs_devices->rw_devices--;
6846         }
6847
6848         if (device->missing)
6849                 fs_devices->missing_devices--;
6850
6851         new_device = btrfs_alloc_device(NULL, &device->devid,
6852                                         device->uuid);
6853         BUG_ON(IS_ERR(new_device)); /* -ENOMEM */
6854
6855         /* Safe because we are under uuid_mutex */
6856         if (device->name) {
6857                 name = rcu_string_strdup(device->name->str, GFP_NOFS);
6858                 BUG_ON(!name); /* -ENOMEM */
6859                 rcu_assign_pointer(new_device->name, name);
6860         }
6861
6862         list_replace_rcu(&device->dev_list, &new_device->dev_list);
6863         new_device->fs_devices = device->fs_devices;
6864
6865         call_rcu(&device->rcu, free_device);
6866 }