fs/btrfs/volumes.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18 #include <linux/sched.h>
  19 #include <linux/bio.h>
  20 #include <linux/slab.h>
  21 #include <linux/buffer_head.h>
  22 #include <linux/blkdev.h>
  23 #include <linux/random.h>
  24 #include <linux/iocontext.h>
  25 #include <linux/capability.h>
  26 #include <linux/ratelimit.h>
  27 #include <linux/kthread.h>
  28 #include <linux/raid/pq.h>
  29 #include <linux/semaphore.h>
  30 #include <asm/div64.h>
  31 #include "ctree.h"
  32 #include "extent_map.h"
  33 #include "disk-io.h"
  34 #include "transaction.h"
  35 #include "print-tree.h"
  36 #include "volumes.h"
  37 #include "raid56.h"
  38 #include "async-thread.h"
  39 #include "check-integrity.h"
  40 #include "rcu-string.h"
  41 #include "math.h"
  42 #include "dev-replace.h"
  43 #include "sysfs.h"
  44
  45 const struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = {
  46         [BTRFS_RAID_RAID10] = {
  47                 .sub_stripes    = 2,
  48                 .dev_stripes    = 1,
  49                 .devs_max       = 0,    /* 0 == as many as possible */
  50                 .devs_min       = 4,
  51                 .tolerated_failures = 1,
  52                 .devs_increment = 2,
  53                 .ncopies        = 2,
  54         },
  55         [BTRFS_RAID_RAID1] = {
  56                 .sub_stripes    = 1,
  57                 .dev_stripes    = 1,
  58                 .devs_max       = 2,
  59                 .devs_min       = 2,
  60                 .tolerated_failures = 1,
  61                 .devs_increment = 2,
  62                 .ncopies        = 2,
  63         },
  64         [BTRFS_RAID_DUP] = {
  65                 .sub_stripes    = 1,
  66                 .dev_stripes    = 2,
  67                 .devs_max       = 1,
  68                 .devs_min       = 1,
  69                 .tolerated_failures = 0,
  70                 .devs_increment = 1,
  71                 .ncopies        = 2,
  72         },
  73         [BTRFS_RAID_RAID0] = {
  74                 .sub_stripes    = 1,
  75                 .dev_stripes    = 1,
  76                 .devs_max       = 0,
  77                 .devs_min       = 2,
  78                 .tolerated_failures = 0,
  79                 .devs_increment = 1,
  80                 .ncopies        = 1,
  81         },
  82         [BTRFS_RAID_SINGLE] = {
  83                 .sub_stripes    = 1,
  84                 .dev_stripes    = 1,
  85                 .devs_max       = 1,
  86                 .devs_min       = 1,
  87                 .tolerated_failures = 0,
  88                 .devs_increment = 1,
  89                 .ncopies        = 1,
  90         },
  91         [BTRFS_RAID_RAID5] = {
  92                 .sub_stripes    = 1,
  93                 .dev_stripes    = 1,
  94                 .devs_max       = 0,
  95                 .devs_min       = 2,
  96                 .tolerated_failures = 1,
  97                 .devs_increment = 1,
  98                 .ncopies        = 2,
  99         },
 100         [BTRFS_RAID_RAID6] = {
 101                 .sub_stripes    = 1,
 102                 .dev_stripes    = 1,
 103                 .devs_max       = 0,
 104                 .devs_min       = 3,
 105                 .tolerated_failures = 2,
 106                 .devs_increment = 1,
 107                 .ncopies        = 3,
 108         },
 109 };
 110
 111 const u64 btrfs_raid_group[BTRFS_NR_RAID_TYPES] = {
 112         [BTRFS_RAID_RAID10] = BTRFS_BLOCK_GROUP_RAID10,
 113         [BTRFS_RAID_RAID1]  = BTRFS_BLOCK_GROUP_RAID1,
 114         [BTRFS_RAID_DUP]    = BTRFS_BLOCK_GROUP_DUP,
 115         [BTRFS_RAID_RAID0]  = BTRFS_BLOCK_GROUP_RAID0,
 116         [BTRFS_RAID_SINGLE] = 0,
 117         [BTRFS_RAID_RAID5]  = BTRFS_BLOCK_GROUP_RAID5,
 118         [BTRFS_RAID_RAID6]  = BTRFS_BLOCK_GROUP_RAID6,
 119 };
 120
 121 static int init_first_rw_device(struct btrfs_trans_handle *trans,
 122                                 struct btrfs_root *root,
 123                                 struct btrfs_device *device);
 124 static int btrfs_relocate_sys_chunks(struct btrfs_root *root);
 125 static void __btrfs_reset_dev_stats(struct btrfs_device *dev);
 126 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev);
 127 static void btrfs_dev_stat_print_on_load(struct btrfs_device *device);
 128
 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         btrfs_device_data_ordered_init(dev);
 236         INIT_RADIX_TREE(&dev->reada_zones, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
 237         INIT_RADIX_TREE(&dev->reada_extents, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
 238
 239         return dev;
 240 }
 241
 242 static noinline struct btrfs_device *__find_device(struct list_head *head,
 243                                                    u64 devid, u8 *uuid)
 244 {
 245         struct btrfs_device *dev;
 246
 247         list_for_each_entry(dev, head, dev_list) {
 248                 if (dev->devid == devid &&
 249                     (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
 250                         return dev;
 251                 }
 252         }
 253         return NULL;
 254 }
 255
 256 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
 257 {
 258         struct btrfs_fs_devices *fs_devices;
 259
 260         list_for_each_entry(fs_devices, &fs_uuids, list) {
 261                 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
 262                         return fs_devices;
 263         }
 264         return NULL;
 265 }
 266
 267 static int
 268 btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder,
 269                       int flush, struct block_device **bdev,
 270                       struct buffer_head **bh)
 271 {
 272         int ret;
 273
 274         *bdev = blkdev_get_by_path(device_path, flags, holder);
 275
 276         if (IS_ERR(*bdev)) {
 277                 ret = PTR_ERR(*bdev);
 278                 goto error;
 279         }
 280
 281         if (flush)
 282                 filemap_write_and_wait((*bdev)->bd_inode->i_mapping);
 283         ret = set_blocksize(*bdev, 4096);
 284         if (ret) {
 285                 blkdev_put(*bdev, flags);
 286                 goto error;
 287         }
 288         invalidate_bdev(*bdev);
 289         *bh = btrfs_read_dev_super(*bdev);
 290         if (IS_ERR(*bh)) {
 291                 ret = PTR_ERR(*bh);
 292                 blkdev_put(*bdev, flags);
 293                 goto error;
 294         }
 295
 296         return 0;
 297
 298 error:
 299         *bdev = NULL;
 300         *bh = NULL;
 301         return ret;
 302 }
 303
 304 static void requeue_list(struct btrfs_pending_bios *pending_bios,
 305                         struct bio *head, struct bio *tail)
 306 {
 307
 308         struct bio *old_head;
 309
 310         old_head = pending_bios->head;
 311         pending_bios->head = head;
 312         if (pending_bios->tail)
 313                 tail->bi_next = old_head;
 314         else
 315                 pending_bios->tail = tail;
 316 }
 317
 318 /*
 319  * we try to collect pending bios for a device so we don't get a large
 320  * number of procs sending bios down to the same device.  This greatly
 321  * improves the schedulers ability to collect and merge the bios.
 322  *
 323  * But, it also turns into a long list of bios to process and that is sure
 324  * to eventually make the worker thread block.  The solution here is to
 325  * make some progress and then put this work struct back at the end of
 326  * the list if the block device is congested.  This way, multiple devices
 327  * can make progress from a single worker thread.
 328  */
 329 static noinline void run_scheduled_bios(struct btrfs_device *device)
 330 {
 331         struct bio *pending;
 332         struct backing_dev_info *bdi;
 333         struct btrfs_fs_info *fs_info;
 334         struct btrfs_pending_bios *pending_bios;
 335         struct bio *tail;
 336         struct bio *cur;
 337         int again = 0;
 338         unsigned long num_run;
 339         unsigned long batch_run = 0;
 340         unsigned long limit;
 341         unsigned long last_waited = 0;
 342         int force_reg = 0;
 343         int sync_pending = 0;
 344         struct blk_plug plug;
 345
 346         /*
 347          * this function runs all the bios we've collected for
 348          * a particular device.  We don't want to wander off to
 349          * another device without first sending all of these down.
 350          * So, setup a plug here and finish it off before we return
 351          */
 352         blk_start_plug(&plug);
 353
 354         bdi = blk_get_backing_dev_info(device->bdev);
 355         fs_info = device->dev_root->fs_info;
 356         limit = btrfs_async_submit_limit(fs_info);
 357         limit = limit * 2 / 3;
 358
 359 loop:
 360         spin_lock(&device->io_lock);
 361
 362 loop_lock:
 363         num_run = 0;
 364
 365         /* take all the bios off the list at once and process them
 366          * later on (without the lock held).  But, remember the
 367          * tail and other pointers so the bios can be properly reinserted
 368          * into the list if we hit congestion
 369          */
 370         if (!force_reg && device->pending_sync_bios.head) {
 371                 pending_bios = &device->pending_sync_bios;
 372                 force_reg = 1;
 373         } else {
 374                 pending_bios = &device->pending_bios;
 375                 force_reg = 0;
 376         }
 377
 378         pending = pending_bios->head;
 379         tail = pending_bios->tail;
 380         WARN_ON(pending && !tail);
 381
 382         /*
 383          * if pending was null this time around, no bios need processing
 384          * at all and we can stop.  Otherwise it'll loop back up again
 385          * and do an additional check so no bios are missed.
 386          *
 387          * device->running_pending is used to synchronize with the
 388          * schedule_bio code.
 389          */
 390         if (device->pending_sync_bios.head == NULL &&
 391             device->pending_bios.head == NULL) {
 392                 again = 0;
 393                 device->running_pending = 0;
 394         } else {
 395                 again = 1;
 396                 device->running_pending = 1;
 397         }
 398
 399         pending_bios->head = NULL;
 400         pending_bios->tail = NULL;
 401
 402         spin_unlock(&device->io_lock);
 403
 404         while (pending) {
 405
 406                 rmb();
 407                 /* we want to work on both lists, but do more bios on the
 408                  * sync list than the regular list
 409                  */
 410                 if ((num_run > 32 &&
 411                     pending_bios != &device->pending_sync_bios &&
 412                     device->pending_sync_bios.head) ||
 413                    (num_run > 64 && pending_bios == &device->pending_sync_bios &&
 414                     device->pending_bios.head)) {
 415                         spin_lock(&device->io_lock);
 416                         requeue_list(pending_bios, pending, tail);
 417                         goto loop_lock;
 418                 }
 419
 420                 cur = pending;
 421                 pending = pending->bi_next;
 422                 cur->bi_next = NULL;
 423
 424                 /*
 425                  * atomic_dec_return implies a barrier for waitqueue_active
 426                  */
 427                 if (atomic_dec_return(&fs_info->nr_async_bios) < limit &&
 428                     waitqueue_active(&fs_info->async_submit_wait))
 429                         wake_up(&fs_info->async_submit_wait);
 430
 431                 BUG_ON(atomic_read(&cur->__bi_cnt) == 0);
 432
 433                 /*
 434                  * if we're doing the sync list, record that our
 435                  * plug has some sync requests on it
 436                  *
 437                  * If we're doing the regular list and there are
 438                  * sync requests sitting around, unplug before
 439                  * we add more
 440                  */
 441                 if (pending_bios == &device->pending_sync_bios) {
 442                         sync_pending = 1;
 443                 } else if (sync_pending) {
 444                         blk_finish_plug(&plug);
 445                         blk_start_plug(&plug);
 446                         sync_pending = 0;
 447                 }
 448
 449                 btrfsic_submit_bio(cur->bi_rw, cur);
 450                 num_run++;
 451                 batch_run++;
 452
 453                 cond_resched();
 454
 455                 /*
 456                  * we made progress, there is more work to do and the bdi
 457                  * is now congested.  Back off and let other work structs
 458                  * run instead
 459                  */
 460                 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
 461                     fs_info->fs_devices->open_devices > 1) {
 462                         struct io_context *ioc;
 463
 464                         ioc = current->io_context;
 465
 466                         /*
 467                          * the main goal here is that we don't want to
 468                          * block if we're going to be able to submit
 469                          * more requests without blocking.
 470                          *
 471                          * This code does two great things, it pokes into
 472                          * the elevator code from a filesystem _and_
 473                          * it makes assumptions about how batching works.
 474                          */
 475                         if (ioc && ioc->nr_batch_requests > 0 &&
 476                             time_before(jiffies, ioc->last_waited + HZ/50UL) &&
 477                             (last_waited == 0 ||
 478                              ioc->last_waited == last_waited)) {
 479                                 /*
 480                                  * we want to go through our batch of
 481                                  * requests and stop.  So, we copy out
 482                                  * the ioc->last_waited time and test
 483                                  * against it before looping
 484                                  */
 485                                 last_waited = ioc->last_waited;
 486                                 cond_resched();
 487                                 continue;
 488                         }
 489                         spin_lock(&device->io_lock);
 490                         requeue_list(pending_bios, pending, tail);
 491                         device->running_pending = 1;
 492
 493                         spin_unlock(&device->io_lock);
 494                         btrfs_queue_work(fs_info->submit_workers,
 495                                          &device->work);
 496                         goto done;
 497                 }
 498                 /* unplug every 64 requests just for good measure */
 499                 if (batch_run % 64 == 0) {
 500                         blk_finish_plug(&plug);
 501                         blk_start_plug(&plug);
 502                         sync_pending = 0;
 503                 }
 504         }
 505
 506         cond_resched();
 507         if (again)
 508                 goto loop;
 509
 510         spin_lock(&device->io_lock);
 511         if (device->pending_bios.head || device->pending_sync_bios.head)
 512                 goto loop_lock;
 513         spin_unlock(&device->io_lock);
 514
 515 done:
 516         blk_finish_plug(&plug);
 517 }
 518
 519 static void pending_bios_fn(struct btrfs_work *work)
 520 {
 521         struct btrfs_device *device;
 522
 523         device = container_of(work, struct btrfs_device, work);
 524         run_scheduled_bios(device);
 525 }
 526
 527
 528 void btrfs_free_stale_device(struct btrfs_device *cur_dev)
 529 {
 530         struct btrfs_fs_devices *fs_devs;
 531         struct btrfs_device *dev;
 532
 533         if (!cur_dev->name)
 534                 return;
 535
 536         list_for_each_entry(fs_devs, &fs_uuids, list) {
 537                 int del = 1;
 538
 539                 if (fs_devs->opened)
 540                         continue;
 541                 if (fs_devs->seeding)
 542                         continue;
 543
 544                 list_for_each_entry(dev, &fs_devs->devices, dev_list) {
 545
 546                         if (dev == cur_dev)
 547                                 continue;
 548                         if (!dev->name)
 549                                 continue;
 550
 551                         /*
 552                          * Todo: This won't be enough. What if the same device
 553                          * comes back (with new uuid and) with its mapper path?
 554                          * But for now, this does help as mostly an admin will
 555                          * either use mapper or non mapper path throughout.
 556                          */
 557                         rcu_read_lock();
 558                         del = strcmp(rcu_str_deref(dev->name),
 559                                                 rcu_str_deref(cur_dev->name));
 560                         rcu_read_unlock();
 561                         if (!del)
 562                                 break;
 563                 }
 564
 565                 if (!del) {
 566                         /* delete the stale device */
 567                         if (fs_devs->num_devices == 1) {
 568                                 btrfs_sysfs_remove_fsid(fs_devs);
 569                                 list_del(&fs_devs->list);
 570                                 free_fs_devices(fs_devs);
 571                                 break;
 572                         } else {
 573                                 fs_devs->num_devices--;
 574                                 list_del(&dev->dev_list);
 575                                 rcu_string_free(dev->name);
 576                                 kfree(dev);
 577                         }
 578                         break;
 579                 }
 580         }
 581 }
 582
 583 /*
 584  * Add new device to list of registered devices
 585  *
 586  * Returns:
 587  * 1   - first time device is seen
 588  * 0   - device already known
 589  * < 0 - error
 590  */
 591 static noinline int device_list_add(const char *path,
 592                            struct btrfs_super_block *disk_super,
 593                            u64 devid, struct btrfs_fs_devices **fs_devices_ret)
 594 {
 595         struct btrfs_device *device;
 596         struct btrfs_fs_devices *fs_devices;
 597         struct rcu_string *name;
 598         int ret = 0;
 599         u64 found_transid = btrfs_super_generation(disk_super);
 600
 601         fs_devices = find_fsid(disk_super->fsid);
 602         if (!fs_devices) {
 603                 fs_devices = alloc_fs_devices(disk_super->fsid);
 604                 if (IS_ERR(fs_devices))
 605                         return PTR_ERR(fs_devices);
 606
 607                 list_add(&fs_devices->list, &fs_uuids);
 608
 609                 device = NULL;
 610         } else {
 611                 device = __find_device(&fs_devices->devices, devid,
 612                                        disk_super->dev_item.uuid);
 613         }
 614
 615         if (!device) {
 616                 if (fs_devices->opened)
 617                         return -EBUSY;
 618
 619                 device = btrfs_alloc_device(NULL, &devid,
 620                                             disk_super->dev_item.uuid);
 621                 if (IS_ERR(device)) {
 622                         /* we can safely leave the fs_devices entry around */
 623                         return PTR_ERR(device);
 624                 }
 625
 626                 name = rcu_string_strdup(path, GFP_NOFS);
 627                 if (!name) {
 628                         kfree(device);
 629                         return -ENOMEM;
 630                 }
 631                 rcu_assign_pointer(device->name, name);
 632
 633                 mutex_lock(&fs_devices->device_list_mutex);
 634                 list_add_rcu(&device->dev_list, &fs_devices->devices);
 635                 fs_devices->num_devices++;
 636                 mutex_unlock(&fs_devices->device_list_mutex);
 637
 638                 ret = 1;
 639                 device->fs_devices = fs_devices;
 640         } else if (!device->name || strcmp(device->name->str, path)) {
 641                 /*
 642                  * When FS is already mounted.
 643                  * 1. If you are here and if the device->name is NULL that
 644                  *    means this device was missing at time of FS mount.
 645                  * 2. If you are here and if the device->name is different
 646                  *    from 'path' that means either
 647                  *      a. The same device disappeared and reappeared with
 648                  *         different name. or
 649                  *      b. The missing-disk-which-was-replaced, has
 650                  *         reappeared now.
 651                  *
 652                  * We must allow 1 and 2a above. But 2b would be a spurious
 653                  * and unintentional.
 654                  *
 655                  * Further in case of 1 and 2a above, the disk at 'path'
 656                  * would have missed some transaction when it was away and
 657                  * in case of 2a the stale bdev has to be updated as well.
 658                  * 2b must not be allowed at all time.
 659                  */
 660
 661                 /*
 662                  * For now, we do allow update to btrfs_fs_device through the
 663                  * btrfs dev scan cli after FS has been mounted.  We're still
 664                  * tracking a problem where systems fail mount by subvolume id
 665                  * when we reject replacement on a mounted FS.
 666                  */
 667                 if (!fs_devices->opened && found_transid < device->generation) {
 668                         /*
 669                          * That is if the FS is _not_ mounted and if you
 670                          * are here, that means there is more than one
 671                          * disk with same uuid and devid.We keep the one
 672                          * with larger generation number or the last-in if
 673                          * generation are equal.
 674                          */
 675                         return -EEXIST;
 676                 }
 677
 678                 name = rcu_string_strdup(path, GFP_NOFS);
 679                 if (!name)
 680                         return -ENOMEM;
 681                 rcu_string_free(device->name);
 682                 rcu_assign_pointer(device->name, name);
 683                 if (device->missing) {
 684                         fs_devices->missing_devices--;
 685                         device->missing = 0;
 686                 }
 687         }
 688
 689         /*
 690          * Unmount does not free the btrfs_device struct but would zero
 691          * generation along with most of the other members. So just update
 692          * it back. We need it to pick the disk with largest generation
 693          * (as above).
 694          */
 695         if (!fs_devices->opened)
 696                 device->generation = found_transid;
 697
 698         /*
 699          * if there is new btrfs on an already registered device,
 700          * then remove the stale device entry.
 701          */
 702         btrfs_free_stale_device(device);
 703
 704         *fs_devices_ret = fs_devices;
 705
 706         return ret;
 707 }
 708
 709 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
 710 {
 711         struct btrfs_fs_devices *fs_devices;
 712         struct btrfs_device *device;
 713         struct btrfs_device *orig_dev;
 714
 715         fs_devices = alloc_fs_devices(orig->fsid);
 716         if (IS_ERR(fs_devices))
 717                 return fs_devices;
 718
 719         mutex_lock(&orig->device_list_mutex);
 720         fs_devices->total_devices = orig->total_devices;
 721
 722         /* We have held the volume lock, it is safe to get the devices. */
 723         list_for_each_entry(orig_dev, &orig->devices, dev_list) {
 724                 struct rcu_string *name;
 725
 726                 device = btrfs_alloc_device(NULL, &orig_dev->devid,
 727                                             orig_dev->uuid);
 728                 if (IS_ERR(device))
 729                         goto error;
 730
 731                 /*
 732                  * This is ok to do without rcu read locked because we hold the
 733                  * uuid mutex so nothing we touch in here is going to disappear.
 734                  */
 735                 if (orig_dev->name) {
 736                         name = rcu_string_strdup(orig_dev->name->str, GFP_NOFS);
 737                         if (!name) {
 738                                 kfree(device);
 739                                 goto error;
 740                         }
 741                         rcu_assign_pointer(device->name, name);
 742                 }
 743
 744                 list_add(&device->dev_list, &fs_devices->devices);
 745                 device->fs_devices = fs_devices;
 746                 fs_devices->num_devices++;
 747         }
 748         mutex_unlock(&orig->device_list_mutex);
 749         return fs_devices;
 750 error:
 751         mutex_unlock(&orig->device_list_mutex);
 752         free_fs_devices(fs_devices);
 753         return ERR_PTR(-ENOMEM);
 754 }
 755
 756 void btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices, int step)
 757 {
 758         struct btrfs_device *device, *next;
 759         struct btrfs_device *latest_dev = NULL;
 760
 761         mutex_lock(&uuid_mutex);
 762 again:
 763         /* This is the initialized path, it is safe to release the devices. */
 764         list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
 765                 if (device->in_fs_metadata) {
 766                         if (!device->is_tgtdev_for_dev_replace &&
 767                             (!latest_dev ||
 768                              device->generation > latest_dev->generation)) {
 769                                 latest_dev = device;
 770                         }
 771                         continue;
 772                 }
 773
 774                 if (device->devid == BTRFS_DEV_REPLACE_DEVID) {
 775                         /*
 776                          * In the first step, keep the device which has
 777                          * the correct fsid and the devid that is used
 778                          * for the dev_replace procedure.
 779                          * In the second step, the dev_replace state is
 780                          * read from the device tree and it is known
 781                          * whether the procedure is really active or
 782                          * not, which means whether this device is
 783                          * used or whether it should be removed.
 784                          */
 785                         if (step == 0 || device->is_tgtdev_for_dev_replace) {
 786                                 continue;
 787                         }
 788                 }
 789                 if (device->bdev) {
 790                         blkdev_put(device->bdev, device->mode);
 791                         device->bdev = NULL;
 792                         fs_devices->open_devices--;
 793                 }
 794                 if (device->writeable) {
 795                         list_del_init(&device->dev_alloc_list);
 796                         device->writeable = 0;
 797                         if (!device->is_tgtdev_for_dev_replace)
 798                                 fs_devices->rw_devices--;
 799                 }
 800                 list_del_init(&device->dev_list);
 801                 fs_devices->num_devices--;
 802                 rcu_string_free(device->name);
 803                 kfree(device);
 804         }
 805
 806         if (fs_devices->seed) {
 807                 fs_devices = fs_devices->seed;
 808                 goto again;
 809         }
 810
 811         fs_devices->latest_bdev = latest_dev->bdev;
 812
 813         mutex_unlock(&uuid_mutex);
 814 }
 815
 816 static void __free_device(struct work_struct *work)
 817 {
 818         struct btrfs_device *device;
 819
 820         device = container_of(work, struct btrfs_device, rcu_work);
 821
 822         if (device->bdev)
 823                 blkdev_put(device->bdev, device->mode);
 824
 825         rcu_string_free(device->name);
 826         kfree(device);
 827 }
 828
 829 static void free_device(struct rcu_head *head)
 830 {
 831         struct btrfs_device *device;
 832
 833         device = container_of(head, struct btrfs_device, rcu);
 834
 835         INIT_WORK(&device->rcu_work, __free_device);
 836         schedule_work(&device->rcu_work);
 837 }
 838
 839 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
 840 {
 841         struct btrfs_device *device, *tmp;
 842
 843         if (--fs_devices->opened > 0)
 844                 return 0;
 845
 846         mutex_lock(&fs_devices->device_list_mutex);
 847         list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list) {
 848                 btrfs_close_one_device(device);
 849         }
 850         mutex_unlock(&fs_devices->device_list_mutex);
 851
 852         WARN_ON(fs_devices->open_devices);
 853         WARN_ON(fs_devices->rw_devices);
 854         fs_devices->opened = 0;
 855         fs_devices->seeding = 0;
 856
 857         return 0;
 858 }
 859
 860 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
 861 {
 862         struct btrfs_fs_devices *seed_devices = NULL;
 863         int ret;
 864
 865         mutex_lock(&uuid_mutex);
 866         ret = __btrfs_close_devices(fs_devices);
 867         if (!fs_devices->opened) {
 868                 seed_devices = fs_devices->seed;
 869                 fs_devices->seed = NULL;
 870         }
 871         mutex_unlock(&uuid_mutex);
 872
 873         while (seed_devices) {
 874                 fs_devices = seed_devices;
 875                 seed_devices = fs_devices->seed;
 876                 __btrfs_close_devices(fs_devices);
 877                 free_fs_devices(fs_devices);
 878         }
 879         /*
 880          * Wait for rcu kworkers under __btrfs_close_devices
 881          * to finish all blkdev_puts so device is really
 882          * free when umount is done.
 883          */
 884         rcu_barrier();
 885         return ret;
 886 }
 887
 888 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
 889                                 fmode_t flags, void *holder)
 890 {
 891         struct request_queue *q;
 892         struct block_device *bdev;
 893         struct list_head *head = &fs_devices->devices;
 894         struct btrfs_device *device;
 895         struct btrfs_device *latest_dev = NULL;
 896         struct buffer_head *bh;
 897         struct btrfs_super_block *disk_super;
 898         u64 devid;
 899         int seeding = 1;
 900         int ret = 0;
 901
 902         flags |= FMODE_EXCL;
 903
 904         list_for_each_entry(device, head, dev_list) {
 905                 if (device->bdev)
 906                         continue;
 907                 if (!device->name)
 908                         continue;
 909
 910                 /* Just open everything we can; ignore failures here */
 911                 if (btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1,
 912                                             &bdev, &bh))
 913                         continue;
 914
 915                 disk_super = (struct btrfs_super_block *)bh->b_data;
 916                 devid = btrfs_stack_device_id(&disk_super->dev_item);
 917                 if (devid != device->devid)
 918                         goto error_brelse;
 919
 920                 if (memcmp(device->uuid, disk_super->dev_item.uuid,
 921                            BTRFS_UUID_SIZE))
 922                         goto error_brelse;
 923
 924                 device->generation = btrfs_super_generation(disk_super);
 925                 if (!latest_dev ||
 926                     device->generation > latest_dev->generation)
 927                         latest_dev = device;
 928
 929                 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
 930                         device->writeable = 0;
 931                 } else {
 932                         device->writeable = !bdev_read_only(bdev);
 933                         seeding = 0;
 934                 }
 935
 936                 q = bdev_get_queue(bdev);
 937                 if (blk_queue_discard(q))
 938                         device->can_discard = 1;
 939
 940                 device->bdev = bdev;
 941                 device->in_fs_metadata = 0;
 942                 device->mode = flags;
 943
 944                 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
 945                         fs_devices->rotating = 1;
 946
 947                 fs_devices->open_devices++;
 948                 if (device->writeable &&
 949                     device->devid != BTRFS_DEV_REPLACE_DEVID) {
 950                         fs_devices->rw_devices++;
 951                         list_add(&device->dev_alloc_list,
 952                                  &fs_devices->alloc_list);
 953                 }
 954                 brelse(bh);
 955                 continue;
 956
 957 error_brelse:
 958                 brelse(bh);
 959                 blkdev_put(bdev, flags);
 960                 continue;
 961         }
 962         if (fs_devices->open_devices == 0) {
 963                 ret = -EINVAL;
 964                 goto out;
 965         }
 966         fs_devices->seeding = seeding;
 967         fs_devices->opened = 1;
 968         fs_devices->latest_bdev = latest_dev->bdev;
 969         fs_devices->total_rw_bytes = 0;
 970 out:
 971         return ret;
 972 }
 973
 974 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
 975                        fmode_t flags, void *holder)
 976 {
 977         int ret;
 978
 979         mutex_lock(&uuid_mutex);
 980         if (fs_devices->opened) {
 981                 fs_devices->opened++;
 982                 ret = 0;
 983         } else {
 984                 ret = __btrfs_open_devices(fs_devices, flags, holder);
 985         }
 986         mutex_unlock(&uuid_mutex);
 987         return ret;
 988 }
 989
 990 /*
 991  * Look for a btrfs signature on a device. This may be called out of the mount path
 992  * and we are not allowed to call set_blocksize during the scan. The superblock
 993  * is read via pagecache
 994  */
 995 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
 996                           struct btrfs_fs_devices **fs_devices_ret)
 997 {
 998         struct btrfs_super_block *disk_super;
 999         struct block_device *bdev;
1000         struct page *page;
1001         void *p;
1002         int ret = -EINVAL;
1003         u64 devid;
1004         u64 transid;
1005         u64 total_devices;
1006         u64 bytenr;
1007         pgoff_t index;
1008
1009         /*
1010          * we would like to check all the supers, but that would make
1011          * a btrfs mount succeed after a mkfs from a different FS.
1012          * So, we need to add a special mount option to scan for
1013          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
1014          */
1015         bytenr = btrfs_sb_offset(0);
1016         flags |= FMODE_EXCL;
1017         mutex_lock(&uuid_mutex);
1018
1019         bdev = blkdev_get_by_path(path, flags, holder);
1020
1021         if (IS_ERR(bdev)) {
1022                 ret = PTR_ERR(bdev);
1023                 goto error;
1024         }
1025
1026         /* make sure our super fits in the device */
1027         if (bytenr + PAGE_CACHE_SIZE >= i_size_read(bdev->bd_inode))
1028                 goto error_bdev_put;
1029
1030         /* make sure our super fits in the page */
1031         if (sizeof(*disk_super) > PAGE_CACHE_SIZE)
1032                 goto error_bdev_put;
1033
1034         /* make sure our super doesn't straddle pages on disk */
1035         index = bytenr >> PAGE_CACHE_SHIFT;
1036         if ((bytenr + sizeof(*disk_super) - 1) >> PAGE_CACHE_SHIFT != index)
1037                 goto error_bdev_put;
1038
1039         /* pull in the page with our super */
1040         page = read_cache_page_gfp(bdev->bd_inode->i_mapping,
1041                                    index, GFP_NOFS);
1042
1043         if (IS_ERR_OR_NULL(page))
1044                 goto error_bdev_put;
1045
1046         p = kmap(page);
1047
1048         /* align our pointer to the offset of the super block */
1049         disk_super = p + (bytenr & ~PAGE_CACHE_MASK);
1050
1051         if (btrfs_super_bytenr(disk_super) != bytenr ||
1052             btrfs_super_magic(disk_super) != BTRFS_MAGIC)
1053                 goto error_unmap;
1054
1055         devid = btrfs_stack_device_id(&disk_super->dev_item);
1056         transid = btrfs_super_generation(disk_super);
1057         total_devices = btrfs_super_num_devices(disk_super);
1058
1059         ret = device_list_add(path, disk_super, devid, fs_devices_ret);
1060         if (ret > 0) {
1061                 if (disk_super->label[0]) {
1062                         if (disk_super->label[BTRFS_LABEL_SIZE - 1])
1063                                 disk_super->label[BTRFS_LABEL_SIZE - 1] = '\0';
1064                         printk(KERN_INFO "BTRFS: device label %s ", disk_super->label);
1065                 } else {
1066                         printk(KERN_INFO "BTRFS: device fsid %pU ", disk_super->fsid);
1067                 }
1068
1069                 printk(KERN_CONT "devid %llu transid %llu %s\n", devid, transid, path);
1070                 ret = 0;
1071         }
1072         if (!ret && fs_devices_ret)
1073                 (*fs_devices_ret)->total_devices = total_devices;
1074
1075 error_unmap:
1076         kunmap(page);
1077         page_cache_release(page);
1078
1079 error_bdev_put:
1080         blkdev_put(bdev, flags);
1081 error:
1082         mutex_unlock(&uuid_mutex);
1083         return ret;
1084 }
1085
1086 /* helper to account the used device space in the range */
1087 int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
1088                                    u64 end, u64 *length)
1089 {
1090         struct btrfs_key key;
1091         struct btrfs_root *root = device->dev_root;
1092         struct btrfs_dev_extent *dev_extent;
1093         struct btrfs_path *path;
1094         u64 extent_end;
1095         int ret;
1096         int slot;
1097         struct extent_buffer *l;
1098
1099         *length = 0;
1100
1101         if (start >= device->total_bytes || device->is_tgtdev_for_dev_replace)
1102                 return 0;
1103
1104         path = btrfs_alloc_path();
1105         if (!path)
1106                 return -ENOMEM;
1107         path->reada = 2;
1108
1109         key.objectid = device->devid;
1110         key.offset = start;
1111         key.type = BTRFS_DEV_EXTENT_KEY;
1112
1113         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1114         if (ret < 0)
1115                 goto out;
1116         if (ret > 0) {
1117                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1118                 if (ret < 0)
1119                         goto out;
1120         }
1121
1122         while (1) {
1123                 l = path->nodes[0];
1124                 slot = path->slots[0];
1125                 if (slot >= btrfs_header_nritems(l)) {
1126                         ret = btrfs_next_leaf(root, path);
1127                         if (ret == 0)
1128                                 continue;
1129                         if (ret < 0)
1130                                 goto out;
1131
1132                         break;
1133                 }
1134                 btrfs_item_key_to_cpu(l, &key, slot);
1135
1136                 if (key.objectid < device->devid)
1137                         goto next;
1138
1139                 if (key.objectid > device->devid)
1140                         break;
1141
1142                 if (key.type != BTRFS_DEV_EXTENT_KEY)
1143                         goto next;
1144
1145                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1146                 extent_end = key.offset + btrfs_dev_extent_length(l,
1147                                                                   dev_extent);
1148                 if (key.offset <= start && extent_end > end) {
1149                         *length = end - start + 1;
1150                         break;
1151                 } else if (key.offset <= start && extent_end > start)
1152                         *length += extent_end - start;
1153                 else if (key.offset > start && extent_end <= end)
1154                         *length += extent_end - key.offset;
1155                 else if (key.offset > start && key.offset <= end) {
1156                         *length += end - key.offset + 1;
1157                         break;
1158                 } else if (key.offset > end)
1159                         break;
1160
1161 next:
1162                 path->slots[0]++;
1163         }
1164         ret = 0;
1165 out:
1166         btrfs_free_path(path);
1167         return ret;
1168 }
1169
1170 static int contains_pending_extent(struct btrfs_transaction *transaction,
1171                                    struct btrfs_device *device,
1172                                    u64 *start, u64 len)
1173 {
1174         struct btrfs_fs_info *fs_info = device->dev_root->fs_info;
1175         struct extent_map *em;
1176         struct list_head *search_list = &fs_info->pinned_chunks;
1177         int ret = 0;
1178         u64 physical_start = *start;
1179
1180         if (transaction)
1181                 search_list = &transaction->pending_chunks;
1182 again:
1183         list_for_each_entry(em, search_list, list) {
1184                 struct map_lookup *map;
1185                 int i;
1186
1187                 map = em->map_lookup;
1188                 for (i = 0; i < map->num_stripes; i++) {
1189                         u64 end;
1190
1191                         if (map->stripes[i].dev != device)
1192                                 continue;
1193                         if (map->stripes[i].physical >= physical_start + len ||
1194                             map->stripes[i].physical + em->orig_block_len <=
1195                             physical_start)
1196                                 continue;
1197                         /*
1198                          * Make sure that while processing the pinned list we do
1199                          * not override our *start with a lower value, because
1200                          * we can have pinned chunks that fall within this
1201                          * device hole and that have lower physical addresses
1202                          * than the pending chunks we processed before. If we
1203                          * do not take this special care we can end up getting
1204                          * 2 pending chunks that start at the same physical
1205                          * device offsets because the end offset of a pinned
1206                          * chunk can be equal to the start offset of some
1207                          * pending chunk.
1208                          */
1209                         end = map->stripes[i].physical + em->orig_block_len;
1210                         if (end > *start) {
1211                                 *start = end;
1212                                 ret = 1;
1213                         }
1214                 }
1215         }
1216         if (search_list != &fs_info->pinned_chunks) {
1217                 search_list = &fs_info->pinned_chunks;
1218                 goto again;
1219         }
1220
1221         return ret;
1222 }
1223
1224
1225 /*
1226  * find_free_dev_extent_start - find free space in the specified device
1227  * @device:       the device which we search the free space in
1228  * @num_bytes:    the size of the free space that we need
1229  * @search_start: the position from which to begin the search
1230  * @start:        store the start of the free space.
1231  * @len:          the size of the free space. that we find, or the size
1232  *                of the max free space if we don't find suitable free space
1233  *
1234  * this uses a pretty simple search, the expectation is that it is
1235  * called very infrequently and that a given device has a small number
1236  * of extents
1237  *
1238  * @start is used to store the start of the free space if we find. But if we
1239  * don't find suitable free space, it will be used to store the start position
1240  * of the max free space.
1241  *
1242  * @len is used to store the size of the free space that we find.
1243  * But if we don't find suitable free space, it is used to store the size of
1244  * the max free space.
1245  */
1246 int find_free_dev_extent_start(struct btrfs_transaction *transaction,
1247                                struct btrfs_device *device, u64 num_bytes,
1248                                u64 search_start, u64 *start, u64 *len)
1249 {
1250         struct btrfs_key key;
1251         struct btrfs_root *root = device->dev_root;
1252         struct btrfs_dev_extent *dev_extent;
1253         struct btrfs_path *path;
1254         u64 hole_size;
1255         u64 max_hole_start;
1256         u64 max_hole_size;
1257         u64 extent_end;
1258         u64 search_end = device->total_bytes;
1259         int ret;
1260         int slot;
1261         struct extent_buffer *l;
1262         u64 min_search_start;
1263
1264         /*
1265          * We don't want to overwrite the superblock on the drive nor any area
1266          * used by the boot loader (grub for example), so we make sure to start
1267          * at an offset of at least 1MB.
1268          */
1269         min_search_start = max(root->fs_info->alloc_start, 1024ull * 1024);
1270         search_start = max(search_start, min_search_start);
1271
1272         path = btrfs_alloc_path();
1273         if (!path)
1274                 return -ENOMEM;
1275
1276         max_hole_start = search_start;
1277         max_hole_size = 0;
1278
1279 again:
1280         if (search_start >= search_end || device->is_tgtdev_for_dev_replace) {
1281                 ret = -ENOSPC;
1282                 goto out;
1283         }
1284
1285         path->reada = 2;
1286         path->search_commit_root = 1;
1287         path->skip_locking = 1;
1288
1289         key.objectid = device->devid;
1290         key.offset = search_start;
1291         key.type = BTRFS_DEV_EXTENT_KEY;
1292
1293         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1294         if (ret < 0)
1295                 goto out;
1296         if (ret > 0) {
1297                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1298                 if (ret < 0)
1299                         goto out;
1300         }
1301
1302         while (1) {
1303                 l = path->nodes[0];
1304                 slot = path->slots[0];
1305                 if (slot >= btrfs_header_nritems(l)) {
1306                         ret = btrfs_next_leaf(root, path);
1307                         if (ret == 0)
1308                                 continue;
1309                         if (ret < 0)
1310                                 goto out;
1311
1312                         break;
1313                 }
1314                 btrfs_item_key_to_cpu(l, &key, slot);
1315
1316                 if (key.objectid < device->devid)
1317                         goto next;
1318
1319                 if (key.objectid > device->devid)
1320                         break;
1321
1322                 if (key.type != BTRFS_DEV_EXTENT_KEY)
1323                         goto next;
1324
1325                 if (key.offset > search_start) {
1326                         hole_size = key.offset - search_start;
1327
1328                         /*
1329                          * Have to check before we set max_hole_start, otherwise
1330                          * we could end up sending back this offset anyway.
1331                          */
1332                         if (contains_pending_extent(transaction, device,
1333                                                     &search_start,
1334                                                     hole_size)) {
1335                                 if (key.offset >= search_start) {
1336                                         hole_size = key.offset - search_start;
1337                                 } else {
1338                                         WARN_ON_ONCE(1);
1339                                         hole_size = 0;
1340                                 }
1341                         }
1342
1343                         if (hole_size > max_hole_size) {
1344                                 max_hole_start = search_start;
1345                                 max_hole_size = hole_size;
1346                         }
1347
1348                         /*
1349                          * If this free space is greater than which we need,
1350                          * it must be the max free space that we have found
1351                          * until now, so max_hole_start must point to the start
1352                          * of this free space and the length of this free space
1353                          * is stored in max_hole_size. Thus, we return
1354                          * max_hole_start and max_hole_size and go back to the
1355                          * caller.
1356                          */
1357                         if (hole_size >= num_bytes) {
1358                                 ret = 0;
1359                                 goto out;
1360                         }
1361                 }
1362
1363                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1364                 extent_end = key.offset + btrfs_dev_extent_length(l,
1365                                                                   dev_extent);
1366                 if (extent_end > search_start)
1367                         search_start = extent_end;
1368 next:
1369                 path->slots[0]++;
1370                 cond_resched();
1371         }
1372
1373         /*
1374          * At this point, search_start should be the end of
1375          * allocated dev extents, and when shrinking the device,
1376          * search_end may be smaller than search_start.
1377          */
1378         if (search_end > search_start) {
1379                 hole_size = search_end - search_start;
1380
1381                 if (contains_pending_extent(transaction, device, &search_start,
1382                                             hole_size)) {
1383                         btrfs_release_path(path);
1384                         goto again;
1385                 }
1386
1387                 if (hole_size > max_hole_size) {
1388                         max_hole_start = search_start;
1389                         max_hole_size = hole_size;
1390                 }
1391         }
1392
1393         /* See above. */
1394         if (max_hole_size < num_bytes)
1395                 ret = -ENOSPC;
1396         else
1397                 ret = 0;
1398
1399 out:
1400         btrfs_free_path(path);
1401         *start = max_hole_start;
1402         if (len)
1403                 *len = max_hole_size;
1404         return ret;
1405 }
1406
1407 int find_free_dev_extent(struct btrfs_trans_handle *trans,
1408                          struct btrfs_device *device, u64 num_bytes,
1409                          u64 *start, u64 *len)
1410 {
1411         /* FIXME use last free of some kind */
1412         return find_free_dev_extent_start(trans->transaction, device,
1413                                           num_bytes, 0, start, len);
1414 }
1415
1416 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1417                           struct btrfs_device *device,
1418                           u64 start, u64 *dev_extent_len)
1419 {
1420         int ret;
1421         struct btrfs_path *path;
1422         struct btrfs_root *root = device->dev_root;
1423         struct btrfs_key key;
1424         struct btrfs_key found_key;
1425         struct extent_buffer *leaf = NULL;
1426         struct btrfs_dev_extent *extent = NULL;
1427
1428         path = btrfs_alloc_path();
1429         if (!path)
1430                 return -ENOMEM;
1431
1432         key.objectid = device->devid;
1433         key.offset = start;
1434         key.type = BTRFS_DEV_EXTENT_KEY;
1435 again:
1436         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1437         if (ret > 0) {
1438                 ret = btrfs_previous_item(root, path, key.objectid,
1439                                           BTRFS_DEV_EXTENT_KEY);
1440                 if (ret)
1441                         goto out;
1442                 leaf = path->nodes[0];
1443                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1444                 extent = btrfs_item_ptr(leaf, path->slots[0],
1445                                         struct btrfs_dev_extent);
1446                 BUG_ON(found_key.offset > start || found_key.offset +
1447                        btrfs_dev_extent_length(leaf, extent) < start);
1448                 key = found_key;
1449                 btrfs_release_path(path);
1450                 goto again;
1451         } else if (ret == 0) {
1452                 leaf = path->nodes[0];
1453                 extent = btrfs_item_ptr(leaf, path->slots[0],
1454                                         struct btrfs_dev_extent);
1455         } else {
1456                 btrfs_std_error(root->fs_info, ret, "Slot search failed");
1457                 goto out;
1458         }
1459
1460         *dev_extent_len = btrfs_dev_extent_length(leaf, extent);
1461
1462         ret = btrfs_del_item(trans, root, path);
1463         if (ret) {
1464                 btrfs_std_error(root->fs_info, ret,
1465                             "Failed to remove dev extent item");
1466         } else {
1467                 set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags);
1468         }
1469 out:
1470         btrfs_free_path(path);
1471         return ret;
1472 }
1473
1474 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1475                                   struct btrfs_device *device,
1476                                   u64 chunk_tree, u64 chunk_objectid,
1477                                   u64 chunk_offset, u64 start, u64 num_bytes)
1478 {
1479         int ret;
1480         struct btrfs_path *path;
1481         struct btrfs_root *root = device->dev_root;
1482         struct btrfs_dev_extent *extent;
1483         struct extent_buffer *leaf;
1484         struct btrfs_key key;
1485
1486         WARN_ON(!device->in_fs_metadata);
1487         WARN_ON(device->is_tgtdev_for_dev_replace);
1488         path = btrfs_alloc_path();
1489         if (!path)
1490                 return -ENOMEM;
1491
1492         key.objectid = device->devid;
1493         key.offset = start;
1494         key.type = BTRFS_DEV_EXTENT_KEY;
1495         ret = btrfs_insert_empty_item(trans, root, path, &key,
1496                                       sizeof(*extent));
1497         if (ret)
1498                 goto out;
1499
1500         leaf = path->nodes[0];
1501         extent = btrfs_item_ptr(leaf, path->slots[0],
1502                                 struct btrfs_dev_extent);
1503         btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
1504         btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
1505         btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1506
1507         write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
1508                     btrfs_dev_extent_chunk_tree_uuid(extent), BTRFS_UUID_SIZE);
1509
1510         btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1511         btrfs_mark_buffer_dirty(leaf);
1512 out:
1513         btrfs_free_path(path);
1514         return ret;
1515 }
1516
1517 static u64 find_next_chunk(struct btrfs_fs_info *fs_info)
1518 {
1519         struct extent_map_tree *em_tree;
1520         struct extent_map *em;
1521         struct rb_node *n;
1522         u64 ret = 0;
1523
1524         em_tree = &fs_info->mapping_tree.map_tree;
1525         read_lock(&em_tree->lock);
1526         n = rb_last(&em_tree->map);
1527         if (n) {
1528                 em = rb_entry(n, struct extent_map, rb_node);
1529                 ret = em->start + em->len;
1530         }
1531         read_unlock(&em_tree->lock);
1532
1533         return ret;
1534 }
1535
1536 static noinline int find_next_devid(struct btrfs_fs_info *fs_info,
1537                                     u64 *devid_ret)
1538 {
1539         int ret;
1540         struct btrfs_key key;
1541         struct btrfs_key found_key;
1542         struct btrfs_path *path;
1543
1544         path = btrfs_alloc_path();
1545         if (!path)
1546                 return -ENOMEM;
1547
1548         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1549         key.type = BTRFS_DEV_ITEM_KEY;
1550         key.offset = (u64)-1;
1551
1552         ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
1553         if (ret < 0)
1554                 goto error;
1555
1556         BUG_ON(ret == 0); /* Corruption */
1557
1558         ret = btrfs_previous_item(fs_info->chunk_root, path,
1559                                   BTRFS_DEV_ITEMS_OBJECTID,
1560                                   BTRFS_DEV_ITEM_KEY);
1561         if (ret) {
1562                 *devid_ret = 1;
1563         } else {
1564                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1565                                       path->slots[0]);
1566                 *devid_ret = found_key.offset + 1;
1567         }
1568         ret = 0;
1569 error:
1570         btrfs_free_path(path);
1571         return ret;
1572 }
1573
1574 /*
1575  * the device information is stored in the chunk root
1576  * the btrfs_device struct should be fully filled in
1577  */
1578 static int btrfs_add_device(struct btrfs_trans_handle *trans,
1579                             struct btrfs_root *root,
1580                             struct btrfs_device *device)
1581 {
1582         int ret;
1583         struct btrfs_path *path;
1584         struct btrfs_dev_item *dev_item;
1585         struct extent_buffer *leaf;
1586         struct btrfs_key key;
1587         unsigned long ptr;
1588
1589         root = root->fs_info->chunk_root;
1590
1591         path = btrfs_alloc_path();
1592         if (!path)
1593                 return -ENOMEM;
1594
1595         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1596         key.type = BTRFS_DEV_ITEM_KEY;
1597         key.offset = device->devid;
1598
1599         ret = btrfs_insert_empty_item(trans, root, path, &key,
1600                                       sizeof(*dev_item));
1601         if (ret)
1602                 goto out;
1603
1604         leaf = path->nodes[0];
1605         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1606
1607         btrfs_set_device_id(leaf, dev_item, device->devid);
1608         btrfs_set_device_generation(leaf, dev_item, 0);
1609         btrfs_set_device_type(leaf, dev_item, device->type);
1610         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1611         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1612         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1613         btrfs_set_device_total_bytes(leaf, dev_item,
1614                                      btrfs_device_get_disk_total_bytes(device));
1615         btrfs_set_device_bytes_used(leaf, dev_item,
1616                                     btrfs_device_get_bytes_used(device));
1617         btrfs_set_device_group(leaf, dev_item, 0);
1618         btrfs_set_device_seek_speed(leaf, dev_item, 0);
1619         btrfs_set_device_bandwidth(leaf, dev_item, 0);
1620         btrfs_set_device_start_offset(leaf, dev_item, 0);
1621
1622         ptr = btrfs_device_uuid(dev_item);
1623         write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1624         ptr = btrfs_device_fsid(dev_item);
1625         write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1626         btrfs_mark_buffer_dirty(leaf);
1627
1628         ret = 0;
1629 out:
1630         btrfs_free_path(path);
1631         return ret;
1632 }
1633
1634 /*
1635  * Function to update ctime/mtime for a given device path.
1636  * Mainly used for ctime/mtime based probe like libblkid.
1637  */
1638 static void update_dev_time(char *path_name)
1639 {
1640         struct file *filp;
1641
1642         filp = filp_open(path_name, O_RDWR, 0);
1643         if (IS_ERR(filp))
1644                 return;
1645         file_update_time(filp);
1646         filp_close(filp, NULL);
1647         return;
1648 }
1649
1650 static int btrfs_rm_dev_item(struct btrfs_root *root,
1651                              struct btrfs_device *device)
1652 {
1653         int ret;
1654         struct btrfs_path *path;
1655         struct btrfs_key key;
1656         struct btrfs_trans_handle *trans;
1657
1658         root = root->fs_info->chunk_root;
1659
1660         path = btrfs_alloc_path();
1661         if (!path)
1662                 return -ENOMEM;
1663
1664         trans = btrfs_start_transaction(root, 0);
1665         if (IS_ERR(trans)) {
1666                 btrfs_free_path(path);
1667                 return PTR_ERR(trans);
1668         }
1669         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1670         key.type = BTRFS_DEV_ITEM_KEY;
1671         key.offset = device->devid;
1672
1673         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1674         if (ret < 0)
1675                 goto out;
1676
1677         if (ret > 0) {
1678                 ret = -ENOENT;
1679                 goto out;
1680         }
1681
1682         ret = btrfs_del_item(trans, root, path);
1683         if (ret)
1684                 goto out;
1685 out:
1686         btrfs_free_path(path);
1687         btrfs_commit_transaction(trans, root);
1688         return ret;
1689 }
1690
1691 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
1692 {
1693         struct btrfs_device *device;
1694         struct btrfs_device *next_device;
1695         struct block_device *bdev;
1696         struct buffer_head *bh = NULL;
1697         struct btrfs_super_block *disk_super;
1698         struct btrfs_fs_devices *cur_devices;
1699         u64 all_avail;
1700         u64 devid;
1701         u64 num_devices;
1702         u8 *dev_uuid;
1703         unsigned seq;
1704         int ret = 0;
1705         bool clear_super = false;
1706
1707         mutex_lock(&uuid_mutex);
1708
1709         do {
1710                 seq = read_seqbegin(&root->fs_info->profiles_lock);
1711
1712                 all_avail = root->fs_info->avail_data_alloc_bits |
1713                             root->fs_info->avail_system_alloc_bits |
1714                             root->fs_info->avail_metadata_alloc_bits;
1715         } while (read_seqretry(&root->fs_info->profiles_lock, seq));
1716
1717         num_devices = root->fs_info->fs_devices->num_devices;
1718         btrfs_dev_replace_lock(&root->fs_info->dev_replace);
1719         if (btrfs_dev_replace_is_ongoing(&root->fs_info->dev_replace)) {
1720                 WARN_ON(num_devices < 1);
1721                 num_devices--;
1722         }
1723         btrfs_dev_replace_unlock(&root->fs_info->dev_replace);
1724
1725         if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) && num_devices <= 4) {
1726                 ret = BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET;
1727                 goto out;
1728         }
1729
1730         if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) && num_devices <= 2) {
1731                 ret = BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET;
1732                 goto out;
1733         }
1734
1735         if ((all_avail & BTRFS_BLOCK_GROUP_RAID5) &&
1736             root->fs_info->fs_devices->rw_devices <= 2) {
1737                 ret = BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET;
1738                 goto out;
1739         }
1740         if ((all_avail & BTRFS_BLOCK_GROUP_RAID6) &&
1741             root->fs_info->fs_devices->rw_devices <= 3) {
1742                 ret = BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET;
1743                 goto out;
1744         }
1745
1746         if (strcmp(device_path, "missing") == 0) {
1747                 struct list_head *devices;
1748                 struct btrfs_device *tmp;
1749
1750                 device = NULL;
1751                 devices = &root->fs_info->fs_devices->devices;
1752                 /*
1753                  * It is safe to read the devices since the volume_mutex
1754                  * is held.
1755                  */
1756                 list_for_each_entry(tmp, devices, dev_list) {
1757                         if (tmp->in_fs_metadata &&
1758                             !tmp->is_tgtdev_for_dev_replace &&
1759                             !tmp->bdev) {
1760                                 device = tmp;
1761                                 break;
1762                         }
1763                 }
1764                 bdev = NULL;
1765                 bh = NULL;
1766                 disk_super = NULL;
1767                 if (!device) {
1768                         ret = BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
1769                         goto out;
1770                 }
1771         } else {
1772                 ret = btrfs_get_bdev_and_sb(device_path,
1773                                             FMODE_WRITE | FMODE_EXCL,
1774                                             root->fs_info->bdev_holder, 0,
1775                                             &bdev, &bh);
1776                 if (ret)
1777                         goto out;
1778                 disk_super = (struct btrfs_super_block *)bh->b_data;
1779                 devid = btrfs_stack_device_id(&disk_super->dev_item);
1780                 dev_uuid = disk_super->dev_item.uuid;
1781                 device = btrfs_find_device(root->fs_info, devid, dev_uuid,
1782                                            disk_super->fsid);
1783                 if (!device) {
1784                         ret = -ENOENT;
1785                         goto error_brelse;
1786                 }
1787         }
1788
1789         if (device->is_tgtdev_for_dev_replace) {
1790                 ret = BTRFS_ERROR_DEV_TGT_REPLACE;
1791                 goto error_brelse;
1792         }
1793
1794         if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1795                 ret = BTRFS_ERROR_DEV_ONLY_WRITABLE;
1796                 goto error_brelse;
1797         }
1798
1799         if (device->writeable) {
1800                 lock_chunks(root);
1801                 list_del_init(&device->dev_alloc_list);
1802                 device->fs_devices->rw_devices--;
1803                 unlock_chunks(root);
1804                 clear_super = true;
1805         }
1806
1807         mutex_unlock(&uuid_mutex);
1808         ret = btrfs_shrink_device(device, 0);
1809         mutex_lock(&uuid_mutex);
1810         if (ret)
1811                 goto error_undo;
1812
1813         /*
1814          * TODO: the superblock still includes this device in its num_devices
1815          * counter although write_all_supers() is not locked out. This
1816          * could give a filesystem state which requires a degraded mount.
1817          */
1818         ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1819         if (ret)
1820                 goto error_undo;
1821
1822         device->in_fs_metadata = 0;
1823         btrfs_scrub_cancel_dev(root->fs_info, device);
1824
1825         /*
1826          * the device list mutex makes sure that we don't change
1827          * the device list while someone else is writing out all
1828          * the device supers. Whoever is writing all supers, should
1829          * lock the device list mutex before getting the number of
1830          * devices in the super block (super_copy). Conversely,
1831          * whoever updates the number of devices in the super block
1832          * (super_copy) should hold the device list mutex.
1833          */
1834
1835         cur_devices = device->fs_devices;
1836         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1837         list_del_rcu(&device->dev_list);
1838
1839         device->fs_devices->num_devices--;
1840         device->fs_devices->total_devices--;
1841
1842         if (device->missing)
1843                 device->fs_devices->missing_devices--;
1844
1845         next_device = list_entry(root->fs_info->fs_devices->devices.next,
1846                                  struct btrfs_device, dev_list);
1847         if (device->bdev == root->fs_info->sb->s_bdev)
1848                 root->fs_info->sb->s_bdev = next_device->bdev;
1849         if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1850                 root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1851
1852         if (device->bdev) {
1853                 device->fs_devices->open_devices--;
1854                 /* remove sysfs entry */
1855                 btrfs_sysfs_rm_device_link(root->fs_info->fs_devices, device);
1856         }
1857
1858         call_rcu(&device->rcu, free_device);
1859
1860         num_devices = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
1861         btrfs_set_super_num_devices(root->fs_info->super_copy, num_devices);
1862         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1863
1864         if (cur_devices->open_devices == 0) {
1865                 struct btrfs_fs_devices *fs_devices;
1866                 fs_devices = root->fs_info->fs_devices;
1867                 while (fs_devices) {
1868                         if (fs_devices->seed == cur_devices) {
1869                                 fs_devices->seed = cur_devices->seed;
1870                                 break;
1871                         }
1872                         fs_devices = fs_devices->seed;
1873                 }
1874                 cur_devices->seed = NULL;
1875                 __btrfs_close_devices(cur_devices);
1876                 free_fs_devices(cur_devices);
1877         }
1878
1879         root->fs_info->num_tolerated_disk_barrier_failures =
1880                 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
1881
1882         /*
1883          * at this point, the device is zero sized.  We want to
1884          * remove it from the devices list and zero out the old super
1885          */
1886         if (clear_super && disk_super) {
1887                 u64 bytenr;
1888                 int i;
1889
1890                 /* make sure this device isn't detected as part of
1891                  * the FS anymore
1892                  */
1893                 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1894                 set_buffer_dirty(bh);
1895                 sync_dirty_buffer(bh);
1896
1897                 /* clear the mirror copies of super block on the disk
1898                  * being removed, 0th copy is been taken care above and
1899                  * the below would take of the rest
1900                  */
1901                 for (i = 1; i < BTRFS_SUPER_MIRROR_MAX; i++) {
1902                         bytenr = btrfs_sb_offset(i);
1903                         if (bytenr + BTRFS_SUPER_INFO_SIZE >=
1904                                         i_size_read(bdev->bd_inode))
1905                                 break;
1906
1907                         brelse(bh);
1908                         bh = __bread(bdev, bytenr / 4096,
1909                                         BTRFS_SUPER_INFO_SIZE);
1910                         if (!bh)
1911                                 continue;
1912
1913                         disk_super = (struct btrfs_super_block *)bh->b_data;
1914
1915                         if (btrfs_super_bytenr(disk_super) != bytenr ||
1916                                 btrfs_super_magic(disk_super) != BTRFS_MAGIC) {
1917                                 continue;
1918                         }
1919                         memset(&disk_super->magic, 0,
1920                                                 sizeof(disk_super->magic));
1921                         set_buffer_dirty(bh);
1922                         sync_dirty_buffer(bh);
1923                 }
1924         }
1925
1926         ret = 0;
1927
1928         if (bdev) {
1929                 /* Notify udev that device has changed */
1930                 btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
1931
1932                 /* Update ctime/mtime for device path for libblkid */
1933                 update_dev_time(device_path);
1934         }
1935
1936 error_brelse:
1937         brelse(bh);
1938         if (bdev)
1939                 blkdev_put(bdev, FMODE_READ | FMODE_EXCL);
1940 out:
1941         mutex_unlock(&uuid_mutex);
1942         return ret;
1943 error_undo:
1944         if (device->writeable) {
1945                 lock_chunks(root);
1946                 list_add(&device->dev_alloc_list,
1947                          &root->fs_info->fs_devices->alloc_list);
1948                 device->fs_devices->rw_devices++;
1949                 unlock_chunks(root);
1950         }
1951         goto error_brelse;
1952 }
1953
1954 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_fs_info *fs_info,
1955                                         struct btrfs_device *srcdev)
1956 {
1957         struct btrfs_fs_devices *fs_devices;
1958
1959         WARN_ON(!mutex_is_locked(&fs_info->fs_devices->device_list_mutex));
1960
1961         /*
1962          * in case of fs with no seed, srcdev->fs_devices will point
1963          * to fs_devices of fs_info. However when the dev being replaced is
1964          * a seed dev it will point to the seed's local fs_devices. In short
1965          * srcdev will have its correct fs_devices in both the cases.
1966          */
1967         fs_devices = srcdev->fs_devices;
1968
1969         list_del_rcu(&srcdev->dev_list);
1970         list_del_rcu(&srcdev->dev_alloc_list);
1971         fs_devices->num_devices--;
1972         if (srcdev->missing)
1973                 fs_devices->missing_devices--;
1974
1975         if (srcdev->writeable) {
1976                 fs_devices->rw_devices--;
1977                 /* zero out the old super if it is writable */
1978                 btrfs_scratch_superblocks(srcdev->bdev, srcdev->name->str);
1979         }
1980
1981         if (srcdev->bdev)
1982                 fs_devices->open_devices--;
1983 }
1984
1985 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_fs_info *fs_info,
1986                                       struct btrfs_device *srcdev)
1987 {
1988         struct btrfs_fs_devices *fs_devices = srcdev->fs_devices;
1989
1990         call_rcu(&srcdev->rcu, free_device);
1991
1992         /*
1993          * unless fs_devices is seed fs, num_devices shouldn't go
1994          * zero
1995          */
1996         BUG_ON(!fs_devices->num_devices && !fs_devices->seeding);
1997
1998         /* if this is no devs we rather delete the fs_devices */
1999         if (!fs_devices->num_devices) {
2000                 struct btrfs_fs_devices *tmp_fs_devices;
2001
2002                 tmp_fs_devices = fs_info->fs_devices;
2003                 while (tmp_fs_devices) {
2004                         if (tmp_fs_devices->seed == fs_devices) {
2005                                 tmp_fs_devices->seed = fs_devices->seed;
2006                                 break;
2007                         }
2008                         tmp_fs_devices = tmp_fs_devices->seed;
2009                 }
2010                 fs_devices->seed = NULL;
2011                 __btrfs_close_devices(fs_devices);
2012                 free_fs_devices(fs_devices);
2013         }
2014 }
2015
2016 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
2017                                       struct btrfs_device *tgtdev)
2018 {
2019         struct btrfs_device *next_device;
2020
2021         mutex_lock(&uuid_mutex);
2022         WARN_ON(!tgtdev);
2023         mutex_lock(&fs_info->fs_devices->device_list_mutex);
2024
2025         btrfs_sysfs_rm_device_link(fs_info->fs_devices, tgtdev);
2026
2027         if (tgtdev->bdev) {
2028                 btrfs_scratch_superblocks(tgtdev->bdev, tgtdev->name->str);
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 = em->map_lookup;
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_trans_remove_block_group(root->fs_info,
2857                                                      chunk_offset);
2858         if (IS_ERR(trans)) {
2859                 ret = PTR_ERR(trans);
2860                 btrfs_std_error(root->fs_info, ret, NULL);
2861                 return ret;
2862         }
2863
2864         /*
2865          * step two, delete the device extents and the
2866          * chunk tree entries
2867          */
2868         ret = btrfs_remove_chunk(trans, root, chunk_offset);
2869         btrfs_end_transaction(trans, root);
2870         return ret;
2871 }
2872
2873 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
2874 {
2875         struct btrfs_root *chunk_root = root->fs_info->chunk_root;
2876         struct btrfs_path *path;
2877         struct extent_buffer *leaf;
2878         struct btrfs_chunk *chunk;
2879         struct btrfs_key key;
2880         struct btrfs_key found_key;
2881         u64 chunk_type;
2882         bool retried = false;
2883         int failed = 0;
2884         int ret;
2885
2886         path = btrfs_alloc_path();
2887         if (!path)
2888                 return -ENOMEM;
2889
2890 again:
2891         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2892         key.offset = (u64)-1;
2893         key.type = BTRFS_CHUNK_ITEM_KEY;
2894
2895         while (1) {
2896                 mutex_lock(&root->fs_info->delete_unused_bgs_mutex);
2897                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2898                 if (ret < 0) {
2899                         mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
2900                         goto error;
2901                 }
2902                 BUG_ON(ret == 0); /* Corruption */
2903
2904                 ret = btrfs_previous_item(chunk_root, path, key.objectid,
2905                                           key.type);
2906                 if (ret)
2907                         mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
2908                 if (ret < 0)
2909                         goto error;
2910                 if (ret > 0)
2911                         break;
2912
2913                 leaf = path->nodes[0];
2914                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2915
2916                 chunk = btrfs_item_ptr(leaf, path->slots[0],
2917                                        struct btrfs_chunk);
2918                 chunk_type = btrfs_chunk_type(leaf, chunk);
2919                 btrfs_release_path(path);
2920
2921                 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
2922                         ret = btrfs_relocate_chunk(chunk_root,
2923                                                    found_key.offset);
2924                         if (ret == -ENOSPC)
2925                                 failed++;
2926                         else
2927                                 BUG_ON(ret);
2928                 }
2929                 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
2930
2931                 if (found_key.offset == 0)
2932                         break;
2933                 key.offset = found_key.offset - 1;
2934         }
2935         ret = 0;
2936         if (failed && !retried) {
2937                 failed = 0;
2938                 retried = true;
2939                 goto again;
2940         } else if (WARN_ON(failed && retried)) {
2941                 ret = -ENOSPC;
2942         }
2943 error:
2944         btrfs_free_path(path);
2945         return ret;
2946 }
2947
2948 static int insert_balance_item(struct btrfs_root *root,
2949                                struct btrfs_balance_control *bctl)
2950 {
2951         struct btrfs_trans_handle *trans;
2952         struct btrfs_balance_item *item;
2953         struct btrfs_disk_balance_args disk_bargs;
2954         struct btrfs_path *path;
2955         struct extent_buffer *leaf;
2956         struct btrfs_key key;
2957         int ret, err;
2958
2959         path = btrfs_alloc_path();
2960         if (!path)
2961                 return -ENOMEM;
2962
2963         trans = btrfs_start_transaction(root, 0);
2964         if (IS_ERR(trans)) {
2965                 btrfs_free_path(path);
2966                 return PTR_ERR(trans);
2967         }
2968
2969         key.objectid = BTRFS_BALANCE_OBJECTID;
2970         key.type = BTRFS_BALANCE_ITEM_KEY;
2971         key.offset = 0;
2972
2973         ret = btrfs_insert_empty_item(trans, root, path, &key,
2974                                       sizeof(*item));
2975         if (ret)
2976                 goto out;
2977
2978         leaf = path->nodes[0];
2979         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
2980
2981         memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
2982
2983         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
2984         btrfs_set_balance_data(leaf, item, &disk_bargs);
2985         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
2986         btrfs_set_balance_meta(leaf, item, &disk_bargs);
2987         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
2988         btrfs_set_balance_sys(leaf, item, &disk_bargs);
2989
2990         btrfs_set_balance_flags(leaf, item, bctl->flags);
2991
2992         btrfs_mark_buffer_dirty(leaf);
2993 out:
2994         btrfs_free_path(path);
2995         err = btrfs_commit_transaction(trans, root);
2996         if (err && !ret)
2997                 ret = err;
2998         return ret;
2999 }
3000
3001 static int del_balance_item(struct btrfs_root *root)
3002 {
3003         struct btrfs_trans_handle *trans;
3004         struct btrfs_path *path;
3005         struct btrfs_key key;
3006         int ret, err;
3007
3008         path = btrfs_alloc_path();
3009         if (!path)
3010                 return -ENOMEM;
3011
3012         trans = btrfs_start_transaction(root, 0);
3013         if (IS_ERR(trans)) {
3014                 btrfs_free_path(path);
3015                 return PTR_ERR(trans);
3016         }
3017
3018         key.objectid = BTRFS_BALANCE_OBJECTID;
3019         key.type = BTRFS_BALANCE_ITEM_KEY;
3020         key.offset = 0;
3021
3022         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3023         if (ret < 0)
3024                 goto out;
3025         if (ret > 0) {
3026                 ret = -ENOENT;
3027                 goto out;
3028         }
3029
3030         ret = btrfs_del_item(trans, root, path);
3031 out:
3032         btrfs_free_path(path);
3033         err = btrfs_commit_transaction(trans, root);
3034         if (err && !ret)
3035                 ret = err;
3036         return ret;
3037 }
3038
3039 /*
3040  * This is a heuristic used to reduce the number of chunks balanced on
3041  * resume after balance was interrupted.
3042  */
3043 static void update_balance_args(struct btrfs_balance_control *bctl)
3044 {
3045         /*
3046          * Turn on soft mode for chunk types that were being converted.
3047          */
3048         if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
3049                 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
3050         if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
3051                 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
3052         if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
3053                 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
3054
3055         /*
3056          * Turn on usage filter if is not already used.  The idea is
3057          * that chunks that we have already balanced should be
3058          * reasonably full.  Don't do it for chunks that are being
3059          * converted - that will keep us from relocating unconverted
3060          * (albeit full) chunks.
3061          */
3062         if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3063             !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3064             !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3065                 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
3066                 bctl->data.usage = 90;
3067         }
3068         if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3069             !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3070             !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3071                 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
3072                 bctl->sys.usage = 90;
3073         }
3074         if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3075             !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3076             !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3077                 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
3078                 bctl->meta.usage = 90;
3079         }
3080 }
3081
3082 /*
3083  * Should be called with both balance and volume mutexes held to
3084  * serialize other volume operations (add_dev/rm_dev/resize) with
3085  * restriper.  Same goes for unset_balance_control.
3086  */
3087 static void set_balance_control(struct btrfs_balance_control *bctl)
3088 {
3089         struct btrfs_fs_info *fs_info = bctl->fs_info;
3090
3091         BUG_ON(fs_info->balance_ctl);
3092
3093         spin_lock(&fs_info->balance_lock);
3094         fs_info->balance_ctl = bctl;
3095         spin_unlock(&fs_info->balance_lock);
3096 }
3097
3098 static void unset_balance_control(struct btrfs_fs_info *fs_info)
3099 {
3100         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3101
3102         BUG_ON(!fs_info->balance_ctl);
3103
3104         spin_lock(&fs_info->balance_lock);
3105         fs_info->balance_ctl = NULL;
3106         spin_unlock(&fs_info->balance_lock);
3107
3108         kfree(bctl);
3109 }
3110
3111 /*
3112  * Balance filters.  Return 1 if chunk should be filtered out
3113  * (should not be balanced).
3114  */
3115 static int chunk_profiles_filter(u64 chunk_type,
3116                                  struct btrfs_balance_args *bargs)
3117 {
3118         chunk_type = chunk_to_extended(chunk_type) &
3119                                 BTRFS_EXTENDED_PROFILE_MASK;
3120
3121         if (bargs->profiles & chunk_type)
3122                 return 0;
3123
3124         return 1;
3125 }
3126
3127 static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
3128                               struct btrfs_balance_args *bargs)
3129 {
3130         struct btrfs_block_group_cache *cache;
3131         u64 chunk_used;
3132         u64 user_thresh_min;
3133         u64 user_thresh_max;
3134         int ret = 1;
3135
3136         cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3137         chunk_used = btrfs_block_group_used(&cache->item);
3138
3139         if (bargs->usage_min == 0)
3140                 user_thresh_min = 0;
3141         else
3142                 user_thresh_min = div_factor_fine(cache->key.offset,
3143                                         bargs->usage_min);
3144
3145         if (bargs->usage_max == 0)
3146                 user_thresh_max = 1;
3147         else if (bargs->usage_max > 100)
3148                 user_thresh_max = cache->key.offset;
3149         else
3150                 user_thresh_max = div_factor_fine(cache->key.offset,
3151                                         bargs->usage_max);
3152
3153         if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max)
3154                 ret = 0;
3155
3156         btrfs_put_block_group(cache);
3157         return ret;
3158 }
3159
3160 static int chunk_usage_filter(struct btrfs_fs_info *fs_info,
3161                 u64 chunk_offset, struct btrfs_balance_args *bargs)
3162 {
3163         struct btrfs_block_group_cache *cache;
3164         u64 chunk_used, user_thresh;
3165         int ret = 1;
3166
3167         cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3168         chunk_used = btrfs_block_group_used(&cache->item);
3169
3170         if (bargs->usage_min == 0)
3171                 user_thresh = 1;
3172         else if (bargs->usage > 100)
3173                 user_thresh = cache->key.offset;
3174         else
3175                 user_thresh = div_factor_fine(cache->key.offset,
3176                                               bargs->usage);
3177
3178         if (chunk_used < user_thresh)
3179                 ret = 0;
3180
3181         btrfs_put_block_group(cache);
3182         return ret;
3183 }
3184
3185 static int chunk_devid_filter(struct extent_buffer *leaf,
3186                               struct btrfs_chunk *chunk,
3187                               struct btrfs_balance_args *bargs)
3188 {
3189         struct btrfs_stripe *stripe;
3190         int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3191         int i;
3192
3193         for (i = 0; i < num_stripes; i++) {
3194                 stripe = btrfs_stripe_nr(chunk, i);
3195                 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
3196                         return 0;
3197         }
3198
3199         return 1;
3200 }
3201
3202 /* [pstart, pend) */
3203 static int chunk_drange_filter(struct extent_buffer *leaf,
3204                                struct btrfs_chunk *chunk,
3205                                u64 chunk_offset,
3206                                struct btrfs_balance_args *bargs)
3207 {
3208         struct btrfs_stripe *stripe;
3209         int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3210         u64 stripe_offset;
3211         u64 stripe_length;
3212         int factor;
3213         int i;
3214
3215         if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
3216                 return 0;
3217
3218         if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP |
3219              BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) {
3220                 factor = num_stripes / 2;
3221         } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) {
3222                 factor = num_stripes - 1;
3223         } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) {
3224                 factor = num_stripes - 2;
3225         } else {
3226                 factor = num_stripes;
3227         }
3228
3229         for (i = 0; i < num_stripes; i++) {
3230                 stripe = btrfs_stripe_nr(chunk, i);
3231                 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
3232                         continue;
3233
3234                 stripe_offset = btrfs_stripe_offset(leaf, stripe);
3235                 stripe_length = btrfs_chunk_length(leaf, chunk);
3236                 stripe_length = div_u64(stripe_length, factor);
3237
3238                 if (stripe_offset < bargs->pend &&
3239                     stripe_offset + stripe_length > bargs->pstart)
3240                         return 0;
3241         }
3242
3243         return 1;
3244 }
3245
3246 /* [vstart, vend) */
3247 static int chunk_vrange_filter(struct extent_buffer *leaf,
3248                                struct btrfs_chunk *chunk,
3249                                u64 chunk_offset,
3250                                struct btrfs_balance_args *bargs)
3251 {
3252         if (chunk_offset < bargs->vend &&
3253             chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
3254                 /* at least part of the chunk is inside this vrange */
3255                 return 0;
3256
3257         return 1;
3258 }
3259
3260 static int chunk_stripes_range_filter(struct extent_buffer *leaf,
3261                                struct btrfs_chunk *chunk,
3262                                struct btrfs_balance_args *bargs)
3263 {
3264         int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3265
3266         if (bargs->stripes_min <= num_stripes
3267                         && num_stripes <= bargs->stripes_max)
3268                 return 0;
3269
3270         return 1;
3271 }
3272
3273 static int chunk_soft_convert_filter(u64 chunk_type,
3274                                      struct btrfs_balance_args *bargs)
3275 {
3276         if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
3277                 return 0;
3278
3279         chunk_type = chunk_to_extended(chunk_type) &
3280                                 BTRFS_EXTENDED_PROFILE_MASK;
3281
3282         if (bargs->target == chunk_type)
3283                 return 1;
3284
3285         return 0;
3286 }
3287
3288 static int should_balance_chunk(struct btrfs_root *root,
3289                                 struct extent_buffer *leaf,
3290                                 struct btrfs_chunk *chunk, u64 chunk_offset)
3291 {
3292         struct btrfs_balance_control *bctl = root->fs_info->balance_ctl;
3293         struct btrfs_balance_args *bargs = NULL;
3294         u64 chunk_type = btrfs_chunk_type(leaf, chunk);
3295
3296         /* type filter */
3297         if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
3298               (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
3299                 return 0;
3300         }
3301
3302         if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3303                 bargs = &bctl->data;
3304         else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3305                 bargs = &bctl->sys;
3306         else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3307                 bargs = &bctl->meta;
3308
3309         /* profiles filter */
3310         if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
3311             chunk_profiles_filter(chunk_type, bargs)) {
3312                 return 0;
3313         }
3314
3315         /* usage filter */
3316         if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
3317             chunk_usage_filter(bctl->fs_info, chunk_offset, bargs)) {
3318                 return 0;
3319         } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3320             chunk_usage_range_filter(bctl->fs_info, chunk_offset, bargs)) {
3321                 return 0;
3322         }
3323
3324         /* devid filter */
3325         if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
3326             chunk_devid_filter(leaf, chunk, bargs)) {
3327                 return 0;
3328         }
3329
3330         /* drange filter, makes sense only with devid filter */
3331         if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
3332             chunk_drange_filter(leaf, chunk, chunk_offset, bargs)) {
3333                 return 0;
3334         }
3335
3336         /* vrange filter */
3337         if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
3338             chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
3339                 return 0;
3340         }
3341
3342         /* stripes filter */
3343         if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) &&
3344             chunk_stripes_range_filter(leaf, chunk, bargs)) {
3345                 return 0;
3346         }
3347
3348         /* soft profile changing mode */
3349         if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
3350             chunk_soft_convert_filter(chunk_type, bargs)) {
3351                 return 0;
3352         }
3353
3354         /*
3355          * limited by count, must be the last filter
3356          */
3357         if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) {
3358                 if (bargs->limit == 0)
3359                         return 0;
3360                 else
3361                         bargs->limit--;
3362         } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) {
3363                 /*
3364                  * Same logic as the 'limit' filter; the minimum cannot be
3365                  * determined here because we do not have the global informatoin
3366                  * about the count of all chunks that satisfy the filters.
3367                  */
3368                 if (bargs->limit_max == 0)
3369                         return 0;
3370                 else
3371                         bargs->limit_max--;
3372         }
3373
3374         return 1;
3375 }
3376
3377 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
3378 {
3379         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3380         struct btrfs_root *chunk_root = fs_info->chunk_root;
3381         struct btrfs_root *dev_root = fs_info->dev_root;
3382         struct list_head *devices;
3383         struct btrfs_device *device;
3384         u64 old_size;
3385         u64 size_to_free;
3386         u64 chunk_type;
3387         struct btrfs_chunk *chunk;
3388         struct btrfs_path *path;
3389         struct btrfs_key key;
3390         struct btrfs_key found_key;
3391         struct btrfs_trans_handle *trans;
3392         struct extent_buffer *leaf;
3393         int slot;
3394         int ret;
3395         int enospc_errors = 0;
3396         bool counting = true;
3397         /* The single value limit and min/max limits use the same bytes in the */
3398         u64 limit_data = bctl->data.limit;
3399         u64 limit_meta = bctl->meta.limit;
3400         u64 limit_sys = bctl->sys.limit;
3401         u32 count_data = 0;
3402         u32 count_meta = 0;
3403         u32 count_sys = 0;
3404         int chunk_reserved = 0;
3405
3406         /* step one make some room on all the devices */
3407         devices = &fs_info->fs_devices->devices;
3408         list_for_each_entry(device, devices, dev_list) {
3409                 old_size = btrfs_device_get_total_bytes(device);
3410                 size_to_free = div_factor(old_size, 1);
3411                 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
3412                 if (!device->writeable ||
3413                     btrfs_device_get_total_bytes(device) -
3414                     btrfs_device_get_bytes_used(device) > size_to_free ||
3415                     device->is_tgtdev_for_dev_replace)
3416                         continue;
3417
3418                 ret = btrfs_shrink_device(device, old_size - size_to_free);
3419                 if (ret == -ENOSPC)
3420                         break;
3421                 BUG_ON(ret);
3422
3423                 trans = btrfs_start_transaction(dev_root, 0);
3424                 BUG_ON(IS_ERR(trans));
3425
3426                 ret = btrfs_grow_device(trans, device, old_size);
3427                 BUG_ON(ret);
3428
3429                 btrfs_end_transaction(trans, dev_root);
3430         }
3431
3432         /* step two, relocate all the chunks */
3433         path = btrfs_alloc_path();
3434         if (!path) {
3435                 ret = -ENOMEM;
3436                 goto error;
3437         }
3438
3439         /* zero out stat counters */
3440         spin_lock(&fs_info->balance_lock);
3441         memset(&bctl->stat, 0, sizeof(bctl->stat));
3442         spin_unlock(&fs_info->balance_lock);
3443 again:
3444         if (!counting) {
3445                 /*
3446                  * The single value limit and min/max limits use the same bytes
3447                  * in the
3448                  */
3449                 bctl->data.limit = limit_data;
3450                 bctl->meta.limit = limit_meta;
3451                 bctl->sys.limit = limit_sys;
3452         }
3453         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3454         key.offset = (u64)-1;
3455         key.type = BTRFS_CHUNK_ITEM_KEY;
3456
3457         while (1) {
3458                 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
3459                     atomic_read(&fs_info->balance_cancel_req)) {
3460                         ret = -ECANCELED;
3461                         goto error;
3462                 }
3463
3464                 mutex_lock(&fs_info->delete_unused_bgs_mutex);
3465                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3466                 if (ret < 0) {
3467                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3468                         goto error;
3469                 }
3470
3471                 /*
3472                  * this shouldn't happen, it means the last relocate
3473                  * failed
3474                  */
3475                 if (ret == 0)
3476                         BUG(); /* FIXME break ? */
3477
3478                 ret = btrfs_previous_item(chunk_root, path, 0,
3479                                           BTRFS_CHUNK_ITEM_KEY);
3480                 if (ret) {
3481                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3482                         ret = 0;
3483                         break;
3484                 }
3485
3486                 leaf = path->nodes[0];
3487                 slot = path->slots[0];
3488                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3489
3490                 if (found_key.objectid != key.objectid) {
3491                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3492                         break;
3493                 }
3494
3495                 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3496                 chunk_type = btrfs_chunk_type(leaf, chunk);
3497
3498                 if (!counting) {
3499                         spin_lock(&fs_info->balance_lock);
3500                         bctl->stat.considered++;
3501                         spin_unlock(&fs_info->balance_lock);
3502                 }
3503
3504                 ret = should_balance_chunk(chunk_root, leaf, chunk,
3505                                            found_key.offset);
3506
3507                 btrfs_release_path(path);
3508                 if (!ret) {
3509                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3510                         goto loop;
3511                 }
3512
3513                 if (counting) {
3514                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3515                         spin_lock(&fs_info->balance_lock);
3516                         bctl->stat.expected++;
3517                         spin_unlock(&fs_info->balance_lock);
3518
3519                         if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3520                                 count_data++;
3521                         else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3522                                 count_sys++;
3523                         else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3524                                 count_meta++;
3525
3526                         goto loop;
3527                 }
3528
3529                 /*
3530                  * Apply limit_min filter, no need to check if the LIMITS
3531                  * filter is used, limit_min is 0 by default
3532                  */
3533                 if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
3534                                         count_data < bctl->data.limit_min)
3535                                 || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) &&
3536                                         count_meta < bctl->meta.limit_min)
3537                                 || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) &&
3538                                         count_sys < bctl->sys.limit_min)) {
3539                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3540                         goto loop;
3541                 }
3542
3543                 if ((chunk_type & BTRFS_BLOCK_GROUP_DATA) && !chunk_reserved) {
3544                         trans = btrfs_start_transaction(chunk_root, 0);
3545                         if (IS_ERR(trans)) {
3546                                 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3547                                 ret = PTR_ERR(trans);
3548                                 goto error;
3549                         }
3550
3551                         ret = btrfs_force_chunk_alloc(trans, chunk_root,
3552                                                       BTRFS_BLOCK_GROUP_DATA);
3553                         btrfs_end_transaction(trans, chunk_root);
3554                         if (ret < 0) {
3555                                 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3556                                 goto error;
3557                         }
3558                         chunk_reserved = 1;
3559                 }
3560
3561                 ret = btrfs_relocate_chunk(chunk_root,
3562                                            found_key.offset);
3563                 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3564                 if (ret && ret != -ENOSPC)
3565                         goto error;
3566                 if (ret == -ENOSPC) {
3567                         enospc_errors++;
3568                 } else {
3569                         spin_lock(&fs_info->balance_lock);
3570                         bctl->stat.completed++;
3571                         spin_unlock(&fs_info->balance_lock);
3572                 }
3573 loop:
3574                 if (found_key.offset == 0)
3575                         break;
3576                 key.offset = found_key.offset - 1;
3577         }
3578
3579         if (counting) {
3580                 btrfs_release_path(path);
3581                 counting = false;
3582                 goto again;
3583         }
3584 error:
3585         btrfs_free_path(path);
3586         if (enospc_errors) {
3587                 btrfs_info(fs_info, "%d enospc errors during balance",
3588                        enospc_errors);
3589                 if (!ret)
3590                         ret = -ENOSPC;
3591         }
3592
3593         return ret;
3594 }
3595
3596 /**
3597  * alloc_profile_is_valid - see if a given profile is valid and reduced
3598  * @flags: profile to validate
3599  * @extended: if true @flags is treated as an extended profile
3600  */
3601 static int alloc_profile_is_valid(u64 flags, int extended)
3602 {
3603         u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
3604                                BTRFS_BLOCK_GROUP_PROFILE_MASK);
3605
3606         flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
3607
3608         /* 1) check that all other bits are zeroed */
3609         if (flags & ~mask)
3610                 return 0;
3611
3612         /* 2) see if profile is reduced */
3613         if (flags == 0)
3614                 return !extended; /* "0" is valid for usual profiles */
3615
3616         /* true if exactly one bit set */
3617         return (flags & (flags - 1)) == 0;
3618 }
3619
3620 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
3621 {
3622         /* cancel requested || normal exit path */
3623         return atomic_read(&fs_info->balance_cancel_req) ||
3624                 (atomic_read(&fs_info->balance_pause_req) == 0 &&
3625                  atomic_read(&fs_info->balance_cancel_req) == 0);
3626 }
3627
3628 static void __cancel_balance(struct btrfs_fs_info *fs_info)
3629 {
3630         int ret;
3631
3632         unset_balance_control(fs_info);
3633         ret = del_balance_item(fs_info->tree_root);
3634         if (ret)
3635                 btrfs_std_error(fs_info, ret, NULL);
3636
3637         atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3638 }
3639
3640 /* Non-zero return value signifies invalidity */
3641 static inline int validate_convert_profile(struct btrfs_balance_args *bctl_arg,
3642                 u64 allowed)
3643 {
3644         return ((bctl_arg->flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3645                 (!alloc_profile_is_valid(bctl_arg->target, 1) ||
3646                  (bctl_arg->target & ~allowed)));
3647 }
3648
3649 /*
3650  * Should be called with both balance and volume mutexes held
3651  */
3652 int btrfs_balance(struct btrfs_balance_control *bctl,
3653                   struct btrfs_ioctl_balance_args *bargs)
3654 {
3655         struct btrfs_fs_info *fs_info = bctl->fs_info;
3656         u64 allowed;
3657         int mixed = 0;
3658         int ret;
3659         u64 num_devices;
3660         unsigned seq;
3661
3662         if (btrfs_fs_closing(fs_info) ||
3663             atomic_read(&fs_info->balance_pause_req) ||
3664             atomic_read(&fs_info->balance_cancel_req)) {
3665                 ret = -EINVAL;
3666                 goto out;
3667         }
3668
3669         allowed = btrfs_super_incompat_flags(fs_info->super_copy);
3670         if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
3671                 mixed = 1;
3672
3673         /*
3674          * In case of mixed groups both data and meta should be picked,
3675          * and identical options should be given for both of them.
3676          */
3677         allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
3678         if (mixed && (bctl->flags & allowed)) {
3679                 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
3680                     !(bctl->flags & BTRFS_BALANCE_METADATA) ||
3681                     memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
3682                         btrfs_err(fs_info, "with mixed groups data and "
3683                                    "metadata balance options must be the same");
3684                         ret = -EINVAL;
3685                         goto out;
3686                 }
3687         }
3688
3689         num_devices = fs_info->fs_devices->num_devices;
3690         btrfs_dev_replace_lock(&fs_info->dev_replace);
3691         if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
3692                 BUG_ON(num_devices < 1);
3693                 num_devices--;
3694         }
3695         btrfs_dev_replace_unlock(&fs_info->dev_replace);
3696         allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE;
3697         if (num_devices == 1)
3698                 allowed |= BTRFS_BLOCK_GROUP_DUP;
3699         else if (num_devices > 1)
3700                 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1);
3701         if (num_devices > 2)
3702                 allowed |= BTRFS_BLOCK_GROUP_RAID5;
3703         if (num_devices > 3)
3704                 allowed |= (BTRFS_BLOCK_GROUP_RAID10 |
3705                             BTRFS_BLOCK_GROUP_RAID6);
3706         if (validate_convert_profile(&bctl->data, allowed)) {
3707                 btrfs_err(fs_info, "unable to start balance with target "
3708                            "data profile %llu",
3709                        bctl->data.target);
3710                 ret = -EINVAL;
3711                 goto out;
3712         }
3713         if (validate_convert_profile(&bctl->meta, allowed)) {
3714                 btrfs_err(fs_info,
3715                            "unable to start balance with target metadata profile %llu",
3716                        bctl->meta.target);
3717                 ret = -EINVAL;
3718                 goto out;
3719         }
3720         if (validate_convert_profile(&bctl->sys, allowed)) {
3721                 btrfs_err(fs_info,
3722                            "unable to start balance with target system profile %llu",
3723                        bctl->sys.target);
3724                 ret = -EINVAL;
3725                 goto out;
3726         }
3727
3728         /* allow dup'ed data chunks only in mixed mode */
3729         if (!mixed && (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3730             (bctl->data.target & BTRFS_BLOCK_GROUP_DUP)) {
3731                 btrfs_err(fs_info, "dup for data is not allowed");
3732                 ret = -EINVAL;
3733                 goto out;
3734         }
3735
3736         /* allow to reduce meta or sys integrity only if force set */
3737         allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3738                         BTRFS_BLOCK_GROUP_RAID10 |
3739                         BTRFS_BLOCK_GROUP_RAID5 |
3740                         BTRFS_BLOCK_GROUP_RAID6;
3741         do {
3742                 seq = read_seqbegin(&fs_info->profiles_lock);
3743
3744                 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3745                      (fs_info->avail_system_alloc_bits & allowed) &&
3746                      !(bctl->sys.target & allowed)) ||
3747                     ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3748                      (fs_info->avail_metadata_alloc_bits & allowed) &&
3749                      !(bctl->meta.target & allowed))) {
3750                         if (bctl->flags & BTRFS_BALANCE_FORCE) {
3751                                 btrfs_info(fs_info, "force reducing metadata integrity");
3752                         } else {
3753                                 btrfs_err(fs_info, "balance will reduce metadata "
3754                                            "integrity, use force if you want this");
3755                                 ret = -EINVAL;
3756                                 goto out;
3757                         }
3758                 }
3759         } while (read_seqretry(&fs_info->profiles_lock, seq));
3760
3761         if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3762                 fs_info->num_tolerated_disk_barrier_failures = min(
3763                         btrfs_calc_num_tolerated_disk_barrier_failures(fs_info),
3764                         btrfs_get_num_tolerated_disk_barrier_failures(
3765                                 bctl->sys.target));
3766         }
3767
3768         ret = insert_balance_item(fs_info->tree_root, bctl);
3769         if (ret && ret != -EEXIST)
3770                 goto out;
3771
3772         if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
3773                 BUG_ON(ret == -EEXIST);
3774                 set_balance_control(bctl);
3775         } else {
3776                 BUG_ON(ret != -EEXIST);
3777                 spin_lock(&fs_info->balance_lock);
3778                 update_balance_args(bctl);
3779                 spin_unlock(&fs_info->balance_lock);
3780         }
3781
3782         atomic_inc(&fs_info->balance_running);
3783         mutex_unlock(&fs_info->balance_mutex);
3784
3785         ret = __btrfs_balance(fs_info);
3786
3787         mutex_lock(&fs_info->balance_mutex);
3788         atomic_dec(&fs_info->balance_running);
3789
3790         if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3791                 fs_info->num_tolerated_disk_barrier_failures =
3792                         btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3793         }
3794
3795         if (bargs) {
3796                 memset(bargs, 0, sizeof(*bargs));
3797                 update_ioctl_balance_args(fs_info, 0, bargs);
3798         }
3799
3800         if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
3801             balance_need_close(fs_info)) {
3802                 __cancel_balance(fs_info);
3803         }
3804
3805         wake_up(&fs_info->balance_wait_q);
3806
3807         return ret;
3808 out:
3809         if (bctl->flags & BTRFS_BALANCE_RESUME)
3810                 __cancel_balance(fs_info);
3811         else {
3812                 kfree(bctl);
3813                 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3814         }
3815         return ret;
3816 }
3817
3818 static int balance_kthread(void *data)
3819 {
3820         struct btrfs_fs_info *fs_info = data;
3821         int ret = 0;
3822
3823         mutex_lock(&fs_info->volume_mutex);
3824         mutex_lock(&fs_info->balance_mutex);
3825
3826         if (fs_info->balance_ctl) {
3827                 btrfs_info(fs_info, "continuing balance");
3828                 ret = btrfs_balance(fs_info->balance_ctl, NULL);
3829         }
3830
3831         mutex_unlock(&fs_info->balance_mutex);
3832         mutex_unlock(&fs_info->volume_mutex);
3833
3834         return ret;
3835 }
3836
3837 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
3838 {
3839         struct task_struct *tsk;
3840
3841         spin_lock(&fs_info->balance_lock);
3842         if (!fs_info->balance_ctl) {
3843                 spin_unlock(&fs_info->balance_lock);
3844                 return 0;
3845         }
3846         spin_unlock(&fs_info->balance_lock);
3847
3848         if (btrfs_test_opt(fs_info->tree_root, SKIP_BALANCE)) {
3849                 btrfs_info(fs_info, "force skipping balance");
3850                 return 0;
3851         }
3852
3853         /*
3854          * A ro->rw remount sequence should continue with the paused balance
3855          * regardless of who pauses it, system or the user as of now, so set
3856          * the resume flag.
3857          */
3858         spin_lock(&fs_info->balance_lock);
3859         fs_info->balance_ctl->flags |= BTRFS_BALANCE_RESUME;
3860         spin_unlock(&fs_info->balance_lock);
3861
3862         tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
3863         return PTR_ERR_OR_ZERO(tsk);
3864 }
3865
3866 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
3867 {
3868         struct btrfs_balance_control *bctl;
3869         struct btrfs_balance_item *item;
3870         struct btrfs_disk_balance_args disk_bargs;
3871         struct btrfs_path *path;
3872         struct extent_buffer *leaf;
3873         struct btrfs_key key;
3874         int ret;
3875
3876         path = btrfs_alloc_path();
3877         if (!path)
3878                 return -ENOMEM;
3879
3880         key.objectid = BTRFS_BALANCE_OBJECTID;
3881         key.type = BTRFS_BALANCE_ITEM_KEY;
3882         key.offset = 0;
3883
3884         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
3885         if (ret < 0)
3886                 goto out;
3887         if (ret > 0) { /* ret = -ENOENT; */
3888                 ret = 0;
3889                 goto out;
3890         }
3891
3892         bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
3893         if (!bctl) {
3894                 ret = -ENOMEM;
3895                 goto out;
3896         }
3897
3898         leaf = path->nodes[0];
3899         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3900
3901         bctl->fs_info = fs_info;
3902         bctl->flags = btrfs_balance_flags(leaf, item);
3903         bctl->flags |= BTRFS_BALANCE_RESUME;
3904
3905         btrfs_balance_data(leaf, item, &disk_bargs);
3906         btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
3907         btrfs_balance_meta(leaf, item, &disk_bargs);
3908         btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
3909         btrfs_balance_sys(leaf, item, &disk_bargs);
3910         btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
3911
3912         WARN_ON(atomic_xchg(&fs_info->mutually_exclusive_operation_running, 1));
3913
3914         mutex_lock(&fs_info->volume_mutex);
3915         mutex_lock(&fs_info->balance_mutex);
3916
3917         set_balance_control(bctl);
3918
3919         mutex_unlock(&fs_info->balance_mutex);
3920         mutex_unlock(&fs_info->volume_mutex);
3921 out:
3922         btrfs_free_path(path);
3923         return ret;
3924 }
3925
3926 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
3927 {
3928         int ret = 0;
3929
3930         mutex_lock(&fs_info->balance_mutex);
3931         if (!fs_info->balance_ctl) {
3932                 mutex_unlock(&fs_info->balance_mutex);
3933                 return -ENOTCONN;
3934         }
3935
3936         if (atomic_read(&fs_info->balance_running)) {
3937                 atomic_inc(&fs_info->balance_pause_req);
3938                 mutex_unlock(&fs_info->balance_mutex);
3939
3940                 wait_event(fs_info->balance_wait_q,
3941                            atomic_read(&fs_info->balance_running) == 0);
3942
3943                 mutex_lock(&fs_info->balance_mutex);
3944                 /* we are good with balance_ctl ripped off from under us */
3945                 BUG_ON(atomic_read(&fs_info->balance_running));
3946                 atomic_dec(&fs_info->balance_pause_req);
3947         } else {
3948                 ret = -ENOTCONN;
3949         }
3950
3951         mutex_unlock(&fs_info->balance_mutex);
3952         return ret;
3953 }
3954
3955 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
3956 {
3957         if (fs_info->sb->s_flags & MS_RDONLY)
3958                 return -EROFS;
3959
3960         mutex_lock(&fs_info->balance_mutex);
3961         if (!fs_info->balance_ctl) {
3962                 mutex_unlock(&fs_info->balance_mutex);
3963                 return -ENOTCONN;
3964         }
3965
3966         atomic_inc(&fs_info->balance_cancel_req);
3967         /*
3968          * if we are running just wait and return, balance item is
3969          * deleted in btrfs_balance in this case
3970          */
3971         if (atomic_read(&fs_info->balance_running)) {
3972                 mutex_unlock(&fs_info->balance_mutex);
3973                 wait_event(fs_info->balance_wait_q,
3974                            atomic_read(&fs_info->balance_running) == 0);
3975                 mutex_lock(&fs_info->balance_mutex);
3976         } else {
3977                 /* __cancel_balance needs volume_mutex */
3978                 mutex_unlock(&fs_info->balance_mutex);
3979                 mutex_lock(&fs_info->volume_mutex);
3980                 mutex_lock(&fs_info->balance_mutex);
3981
3982                 if (fs_info->balance_ctl)
3983                         __cancel_balance(fs_info);
3984
3985                 mutex_unlock(&fs_info->volume_mutex);
3986         }
3987
3988         BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running));
3989         atomic_dec(&fs_info->balance_cancel_req);
3990         mutex_unlock(&fs_info->balance_mutex);
3991         return 0;
3992 }
3993
3994 static int btrfs_uuid_scan_kthread(void *data)
3995 {
3996         struct btrfs_fs_info *fs_info = data;
3997         struct btrfs_root *root = fs_info->tree_root;
3998         struct btrfs_key key;
3999         struct btrfs_key max_key;
4000         struct btrfs_path *path = NULL;
4001         int ret = 0;
4002         struct extent_buffer *eb;
4003         int slot;
4004         struct btrfs_root_item root_item;
4005         u32 item_size;
4006         struct btrfs_trans_handle *trans = NULL;
4007
4008         path = btrfs_alloc_path();
4009         if (!path) {
4010                 ret = -ENOMEM;
4011                 goto out;
4012         }
4013
4014         key.objectid = 0;
4015         key.type = BTRFS_ROOT_ITEM_KEY;
4016         key.offset = 0;
4017
4018         max_key.objectid = (u64)-1;
4019         max_key.type = BTRFS_ROOT_ITEM_KEY;
4020         max_key.offset = (u64)-1;
4021
4022         while (1) {
4023                 ret = btrfs_search_forward(root, &key, path, 0);
4024                 if (ret) {
4025                         if (ret > 0)
4026                                 ret = 0;
4027                         break;
4028                 }
4029
4030                 if (key.type != BTRFS_ROOT_ITEM_KEY ||
4031                     (key.objectid < BTRFS_FIRST_FREE_OBJECTID &&
4032                      key.objectid != BTRFS_FS_TREE_OBJECTID) ||
4033                     key.objectid > BTRFS_LAST_FREE_OBJECTID)
4034                         goto skip;
4035
4036                 eb = path->nodes[0];
4037                 slot = path->slots[0];
4038                 item_size = btrfs_item_size_nr(eb, slot);
4039                 if (item_size < sizeof(root_item))
4040                         goto skip;
4041
4042                 read_extent_buffer(eb, &root_item,
4043                                    btrfs_item_ptr_offset(eb, slot),
4044                                    (int)sizeof(root_item));
4045                 if (btrfs_root_refs(&root_item) == 0)
4046                         goto skip;
4047
4048                 if (!btrfs_is_empty_uuid(root_item.uuid) ||
4049                     !btrfs_is_empty_uuid(root_item.received_uuid)) {
4050                         if (trans)
4051                                 goto update_tree;
4052
4053                         btrfs_release_path(path);
4054                         /*
4055                          * 1 - subvol uuid item
4056                          * 1 - received_subvol uuid item
4057                          */
4058                         trans = btrfs_start_transaction(fs_info->uuid_root, 2);
4059                         if (IS_ERR(trans)) {
4060                                 ret = PTR_ERR(trans);
4061                                 break;
4062                         }
4063                         continue;
4064                 } else {
4065                         goto skip;
4066                 }
4067 update_tree:
4068                 if (!btrfs_is_empty_uuid(root_item.uuid)) {
4069                         ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
4070                                                   root_item.uuid,
4071                                                   BTRFS_UUID_KEY_SUBVOL,
4072                                                   key.objectid);
4073                         if (ret < 0) {
4074                                 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4075                                         ret);
4076                                 break;
4077                         }
4078                 }
4079
4080                 if (!btrfs_is_empty_uuid(root_item.received_uuid)) {
4081                         ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
4082                                                   root_item.received_uuid,
4083                                                  BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4084                                                   key.objectid);
4085                         if (ret < 0) {
4086                                 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4087                                         ret);
4088                                 break;
4089                         }
4090                 }
4091
4092 skip:
4093                 if (trans) {
4094                         ret = btrfs_end_transaction(trans, fs_info->uuid_root);
4095                         trans = NULL;
4096                         if (ret)
4097                                 break;
4098                 }
4099
4100                 btrfs_release_path(path);
4101                 if (key.offset < (u64)-1) {
4102                         key.offset++;
4103                 } else if (key.type < BTRFS_ROOT_ITEM_KEY) {
4104                         key.offset = 0;
4105                         key.type = BTRFS_ROOT_ITEM_KEY;
4106                 } else if (key.objectid < (u64)-1) {
4107                         key.offset = 0;
4108                         key.type = BTRFS_ROOT_ITEM_KEY;
4109                         key.objectid++;
4110                 } else {
4111                         break;
4112                 }
4113                 cond_resched();
4114         }
4115
4116 out:
4117         btrfs_free_path(path);
4118         if (trans && !IS_ERR(trans))
4119                 btrfs_end_transaction(trans, fs_info->uuid_root);
4120         if (ret)
4121                 btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret);
4122         else
4123                 fs_info->update_uuid_tree_gen = 1;
4124         up(&fs_info->uuid_tree_rescan_sem);
4125         return 0;
4126 }
4127
4128 /*
4129  * Callback for btrfs_uuid_tree_iterate().
4130  * returns:
4131  * 0    check succeeded, the entry is not outdated.
4132  * < 0  if an error occured.
4133  * > 0  if the check failed, which means the caller shall remove the entry.
4134  */
4135 static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info *fs_info,
4136                                        u8 *uuid, u8 type, u64 subid)
4137 {
4138         struct btrfs_key key;
4139         int ret = 0;
4140         struct btrfs_root *subvol_root;
4141
4142         if (type != BTRFS_UUID_KEY_SUBVOL &&
4143             type != BTRFS_UUID_KEY_RECEIVED_SUBVOL)
4144                 goto out;
4145
4146         key.objectid = subid;
4147         key.type = BTRFS_ROOT_ITEM_KEY;
4148         key.offset = (u64)-1;
4149         subvol_root = btrfs_read_fs_root_no_name(fs_info, &key);
4150         if (IS_ERR(subvol_root)) {
4151                 ret = PTR_ERR(subvol_root);
4152                 if (ret == -ENOENT)
4153                         ret = 1;
4154                 goto out;
4155         }
4156
4157         switch (type) {
4158         case BTRFS_UUID_KEY_SUBVOL:
4159                 if (memcmp(uuid, subvol_root->root_item.uuid, BTRFS_UUID_SIZE))
4160                         ret = 1;
4161                 break;
4162         case BTRFS_UUID_KEY_RECEIVED_SUBVOL:
4163                 if (memcmp(uuid, subvol_root->root_item.received_uuid,
4164                            BTRFS_UUID_SIZE))
4165                         ret = 1;
4166                 break;
4167         }
4168
4169 out:
4170         return ret;
4171 }
4172
4173 static int btrfs_uuid_rescan_kthread(void *data)
4174 {
4175         struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data;
4176         int ret;
4177
4178         /*
4179          * 1st step is to iterate through the existing UUID tree and
4180          * to delete all entries that contain outdated data.
4181          * 2nd step is to add all missing entries to the UUID tree.
4182          */
4183         ret = btrfs_uuid_tree_iterate(fs_info, btrfs_check_uuid_tree_entry);
4184         if (ret < 0) {
4185                 btrfs_warn(fs_info, "iterating uuid_tree failed %d", ret);
4186                 up(&fs_info->uuid_tree_rescan_sem);
4187                 return ret;
4188         }
4189         return btrfs_uuid_scan_kthread(data);
4190 }
4191
4192 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info)
4193 {
4194         struct btrfs_trans_handle *trans;
4195         struct btrfs_root *tree_root = fs_info->tree_root;
4196         struct btrfs_root *uuid_root;
4197         struct task_struct *task;
4198         int ret;
4199
4200         /*
4201          * 1 - root node
4202          * 1 - root item
4203          */
4204         trans = btrfs_start_transaction(tree_root, 2);
4205         if (IS_ERR(trans))
4206                 return PTR_ERR(trans);
4207
4208         uuid_root = btrfs_create_tree(trans, fs_info,
4209                                       BTRFS_UUID_TREE_OBJECTID);
4210         if (IS_ERR(uuid_root)) {
4211                 ret = PTR_ERR(uuid_root);
4212                 btrfs_abort_transaction(trans, tree_root, ret);
4213                 return ret;
4214         }
4215
4216         fs_info->uuid_root = uuid_root;
4217
4218         ret = btrfs_commit_transaction(trans, tree_root);
4219         if (ret)
4220                 return ret;
4221
4222         down(&fs_info->uuid_tree_rescan_sem);
4223         task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid");
4224         if (IS_ERR(task)) {
4225                 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4226                 btrfs_warn(fs_info, "failed to start uuid_scan task");
4227                 up(&fs_info->uuid_tree_rescan_sem);
4228                 return PTR_ERR(task);
4229         }
4230
4231         return 0;
4232 }
4233
4234 int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
4235 {
4236         struct task_struct *task;
4237
4238         down(&fs_info->uuid_tree_rescan_sem);
4239         task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
4240         if (IS_ERR(task)) {
4241                 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4242                 btrfs_warn(fs_info, "failed to start uuid_rescan task");
4243                 up(&fs_info->uuid_tree_rescan_sem);
4244                 return PTR_ERR(task);
4245         }
4246
4247         return 0;
4248 }
4249
4250 /*
4251  * shrinking a device means finding all of the device extents past
4252  * the new size, and then following the back refs to the chunks.
4253  * The chunk relocation code actually frees the device extent
4254  */
4255 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
4256 {
4257         struct btrfs_trans_handle *trans;
4258         struct btrfs_root *root = device->dev_root;
4259         struct btrfs_dev_extent *dev_extent = NULL;
4260         struct btrfs_path *path;
4261         u64 length;
4262         u64 chunk_offset;
4263         int ret;
4264         int slot;
4265         int failed = 0;
4266         bool retried = false;
4267         bool checked_pending_chunks = false;
4268         struct extent_buffer *l;
4269         struct btrfs_key key;
4270         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
4271         u64 old_total = btrfs_super_total_bytes(super_copy);
4272         u64 old_size = btrfs_device_get_total_bytes(device);
4273         u64 diff = old_size - new_size;
4274
4275         if (device->is_tgtdev_for_dev_replace)
4276                 return -EINVAL;
4277
4278         path = btrfs_alloc_path();
4279         if (!path)
4280                 return -ENOMEM;
4281
4282         path->reada = 2;
4283
4284         lock_chunks(root);
4285
4286         btrfs_device_set_total_bytes(device, new_size);
4287         if (device->writeable) {
4288                 device->fs_devices->total_rw_bytes -= diff;
4289                 spin_lock(&root->fs_info->free_chunk_lock);
4290                 root->fs_info->free_chunk_space -= diff;
4291                 spin_unlock(&root->fs_info->free_chunk_lock);
4292         }
4293         unlock_chunks(root);
4294
4295 again:
4296         key.objectid = device->devid;
4297         key.offset = (u64)-1;
4298         key.type = BTRFS_DEV_EXTENT_KEY;
4299
4300         do {
4301                 mutex_lock(&root->fs_info->delete_unused_bgs_mutex);
4302                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4303                 if (ret < 0) {
4304                         mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4305                         goto done;
4306                 }
4307
4308                 ret = btrfs_previous_item(root, path, 0, key.type);
4309                 if (ret)
4310                         mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4311                 if (ret < 0)
4312                         goto done;
4313                 if (ret) {
4314                         ret = 0;
4315                         btrfs_release_path(path);
4316                         break;
4317                 }
4318
4319                 l = path->nodes[0];
4320                 slot = path->slots[0];
4321                 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
4322
4323                 if (key.objectid != device->devid) {
4324                         mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4325                         btrfs_release_path(path);
4326                         break;
4327                 }
4328
4329                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
4330                 length = btrfs_dev_extent_length(l, dev_extent);
4331
4332                 if (key.offset + length <= new_size) {
4333                         mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4334                         btrfs_release_path(path);
4335                         break;
4336                 }
4337
4338                 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
4339                 btrfs_release_path(path);
4340
4341                 ret = btrfs_relocate_chunk(root, chunk_offset);
4342                 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4343                 if (ret && ret != -ENOSPC)
4344                         goto done;
4345                 if (ret == -ENOSPC)
4346                         failed++;
4347         } while (key.offset-- > 0);
4348
4349         if (failed && !retried) {
4350                 failed = 0;
4351                 retried = true;
4352                 goto again;
4353         } else if (failed && retried) {
4354                 ret = -ENOSPC;
4355                 goto done;
4356         }
4357
4358         /* Shrinking succeeded, else we would be at "done". */
4359         trans = btrfs_start_transaction(root, 0);
4360         if (IS_ERR(trans)) {
4361                 ret = PTR_ERR(trans);
4362                 goto done;
4363         }
4364
4365         lock_chunks(root);
4366
4367         /*
4368          * We checked in the above loop all device extents that were already in
4369          * the device tree. However before we have updated the device's
4370          * total_bytes to the new size, we might have had chunk allocations that
4371          * have not complete yet (new block groups attached to transaction
4372          * handles), and therefore their device extents were not yet in the
4373          * device tree and we missed them in the loop above. So if we have any
4374          * pending chunk using a device extent that overlaps the device range
4375          * that we can not use anymore, commit the current transaction and
4376          * repeat the search on the device tree - this way we guarantee we will
4377          * not have chunks using device extents that end beyond 'new_size'.
4378          */
4379         if (!checked_pending_chunks) {
4380                 u64 start = new_size;
4381                 u64 len = old_size - new_size;
4382
4383                 if (contains_pending_extent(trans->transaction, device,
4384                                             &start, len)) {
4385                         unlock_chunks(root);
4386                         checked_pending_chunks = true;
4387                         failed = 0;
4388                         retried = false;
4389                         ret = btrfs_commit_transaction(trans, root);
4390                         if (ret)
4391                                 goto done;
4392                         goto again;
4393                 }
4394         }
4395
4396         btrfs_device_set_disk_total_bytes(device, new_size);
4397         if (list_empty(&device->resized_list))
4398                 list_add_tail(&device->resized_list,
4399                               &root->fs_info->fs_devices->resized_devices);
4400
4401         WARN_ON(diff > old_total);
4402         btrfs_set_super_total_bytes(super_copy, old_total - diff);
4403         unlock_chunks(root);
4404
4405         /* Now btrfs_update_device() will change the on-disk size. */
4406         ret = btrfs_update_device(trans, device);
4407         btrfs_end_transaction(trans, root);
4408 done:
4409         btrfs_free_path(path);
4410         if (ret) {
4411                 lock_chunks(root);
4412                 btrfs_device_set_total_bytes(device, old_size);
4413                 if (device->writeable)
4414                         device->fs_devices->total_rw_bytes += diff;
4415                 spin_lock(&root->fs_info->free_chunk_lock);
4416                 root->fs_info->free_chunk_space += diff;
4417                 spin_unlock(&root->fs_info->free_chunk_lock);
4418                 unlock_chunks(root);
4419         }
4420         return ret;
4421 }
4422
4423 static int btrfs_add_system_chunk(struct btrfs_root *root,
4424                            struct btrfs_key *key,
4425                            struct btrfs_chunk *chunk, int item_size)
4426 {
4427         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
4428         struct btrfs_disk_key disk_key;
4429         u32 array_size;
4430         u8 *ptr;
4431
4432         lock_chunks(root);
4433         array_size = btrfs_super_sys_array_size(super_copy);
4434         if (array_size + item_size + sizeof(disk_key)
4435                         > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
4436                 unlock_chunks(root);
4437                 return -EFBIG;
4438         }
4439
4440         ptr = super_copy->sys_chunk_array + array_size;
4441         btrfs_cpu_key_to_disk(&disk_key, key);
4442         memcpy(ptr, &disk_key, sizeof(disk_key));
4443         ptr += sizeof(disk_key);
4444         memcpy(ptr, chunk, item_size);
4445         item_size += sizeof(disk_key);
4446         btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
4447         unlock_chunks(root);
4448
4449         return 0;
4450 }
4451
4452 /*
4453  * sort the devices in descending order by max_avail, total_avail
4454  */
4455 static int btrfs_cmp_device_info(const void *a, const void *b)
4456 {
4457         const struct btrfs_device_info *di_a = a;
4458         const struct btrfs_device_info *di_b = b;
4459
4460         if (di_a->max_avail > di_b->max_avail)
4461                 return -1;
4462         if (di_a->max_avail < di_b->max_avail)
4463                 return 1;
4464         if (di_a->total_avail > di_b->total_avail)
4465                 return -1;
4466         if (di_a->total_avail < di_b->total_avail)
4467                 return 1;
4468         return 0;
4469 }
4470
4471 static u32 find_raid56_stripe_len(u32 data_devices, u32 dev_stripe_target)
4472 {
4473         /* TODO allow them to set a preferred stripe size */
4474         return 64 * 1024;
4475 }
4476
4477 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
4478 {
4479         if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK))
4480                 return;
4481
4482         btrfs_set_fs_incompat(info, RAID56);
4483 }
4484
4485 #define BTRFS_MAX_DEVS(r) ((BTRFS_LEAF_DATA_SIZE(r)             \
4486                         - sizeof(struct btrfs_item)             \
4487                         - sizeof(struct btrfs_chunk))           \
4488                         / sizeof(struct btrfs_stripe) + 1)
4489
4490 #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE        \
4491                                 - 2 * sizeof(struct btrfs_disk_key)     \
4492                                 - 2 * sizeof(struct btrfs_chunk))       \
4493                                 / sizeof(struct btrfs_stripe) + 1)
4494
4495 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4496                                struct btrfs_root *extent_root, u64 start,
4497                                u64 type)
4498 {
4499         struct btrfs_fs_info *info = extent_root->fs_info;
4500         struct btrfs_fs_devices *fs_devices = info->fs_devices;
4501         struct list_head *cur;
4502         struct map_lookup *map = NULL;
4503         struct extent_map_tree *em_tree;
4504         struct extent_map *em;
4505         struct btrfs_device_info *devices_info = NULL;
4506         u64 total_avail;
4507         int num_stripes;        /* total number of stripes to allocate */
4508         int data_stripes;       /* number of stripes that count for
4509                                    block group size */
4510         int sub_stripes;        /* sub_stripes info for map */
4511         int dev_stripes;        /* stripes per dev */
4512         int devs_max;           /* max devs to use */
4513         int devs_min;           /* min devs needed */
4514         int devs_increment;     /* ndevs has to be a multiple of this */
4515         int ncopies;            /* how many copies to data has */
4516         int ret;
4517         u64 max_stripe_size;
4518         u64 max_chunk_size;
4519         u64 stripe_size;
4520         u64 num_bytes;
4521         u64 raid_stripe_len = BTRFS_STRIPE_LEN;
4522         int ndevs;
4523         int i;
4524         int j;
4525         int index;
4526
4527         BUG_ON(!alloc_profile_is_valid(type, 0));
4528
4529         if (list_empty(&fs_devices->alloc_list))
4530                 return -ENOSPC;
4531
4532         index = __get_raid_index(type);
4533
4534         sub_stripes = btrfs_raid_array[index].sub_stripes;
4535         dev_stripes = btrfs_raid_array[index].dev_stripes;
4536         devs_max = btrfs_raid_array[index].devs_max;
4537         devs_min = btrfs_raid_array[index].devs_min;
4538         devs_increment = btrfs_raid_array[index].devs_increment;
4539         ncopies = btrfs_raid_array[index].ncopies;
4540
4541         if (type & BTRFS_BLOCK_GROUP_DATA) {
4542                 max_stripe_size = 1024 * 1024 * 1024;
4543                 max_chunk_size = BTRFS_MAX_DATA_CHUNK_SIZE;
4544                 if (!devs_max)
4545                         devs_max = BTRFS_MAX_DEVS(info->chunk_root);
4546         } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
4547                 /* for larger filesystems, use larger metadata chunks */
4548                 if (fs_devices->total_rw_bytes > 50ULL * 1024 * 1024 * 1024)
4549                         max_stripe_size = 1024 * 1024 * 1024;
4550                 else
4551                         max_stripe_size = 256 * 1024 * 1024;
4552                 max_chunk_size = max_stripe_size;
4553                 if (!devs_max)
4554                         devs_max = BTRFS_MAX_DEVS(info->chunk_root);
4555         } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
4556                 max_stripe_size = 32 * 1024 * 1024;
4557                 max_chunk_size = 2 * max_stripe_size;
4558                 if (!devs_max)
4559                         devs_max = BTRFS_MAX_DEVS_SYS_CHUNK;
4560         } else {
4561                 btrfs_err(info, "invalid chunk type 0x%llx requested",
4562                        type);
4563                 BUG_ON(1);
4564         }
4565
4566         /* we don't want a chunk larger than 10% of writeable space */
4567         max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
4568                              max_chunk_size);
4569
4570         devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info),
4571                                GFP_NOFS);
4572         if (!devices_info)
4573                 return -ENOMEM;
4574
4575         cur = fs_devices->alloc_list.next;
4576
4577         /*
4578          * in the first pass through the devices list, we gather information
4579          * about the available holes on each device.
4580          */
4581         ndevs = 0;
4582         while (cur != &fs_devices->alloc_list) {
4583                 struct btrfs_device *device;
4584                 u64 max_avail;
4585                 u64 dev_offset;
4586
4587                 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
4588
4589                 cur = cur->next;
4590
4591                 if (!device->writeable) {
4592                         WARN(1, KERN_ERR
4593                                "BTRFS: read-only device in alloc_list\n");
4594                         continue;
4595                 }
4596
4597                 if (!device->in_fs_metadata ||
4598                     device->is_tgtdev_for_dev_replace)
4599                         continue;
4600
4601                 if (device->total_bytes > device->bytes_used)
4602                         total_avail = device->total_bytes - device->bytes_used;
4603                 else
4604                         total_avail = 0;
4605
4606                 /* If there is no space on this device, skip it. */
4607                 if (total_avail == 0)
4608                         continue;
4609
4610                 ret = find_free_dev_extent(trans, device,
4611                                            max_stripe_size * dev_stripes,
4612                                            &dev_offset, &max_avail);
4613                 if (ret && ret != -ENOSPC)
4614                         goto error;
4615
4616                 if (ret == 0)
4617                         max_avail = max_stripe_size * dev_stripes;
4618
4619                 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes)
4620                         continue;
4621
4622                 if (ndevs == fs_devices->rw_devices) {
4623                         WARN(1, "%s: found more than %llu devices\n",
4624                              __func__, fs_devices->rw_devices);
4625                         break;
4626                 }
4627                 devices_info[ndevs].dev_offset = dev_offset;
4628                 devices_info[ndevs].max_avail = max_avail;
4629                 devices_info[ndevs].total_avail = total_avail;
4630                 devices_info[ndevs].dev = device;
4631                 ++ndevs;
4632         }
4633
4634         /*
4635          * now sort the devices by hole size / available space
4636          */
4637         sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
4638              btrfs_cmp_device_info, NULL);
4639
4640         /* round down to number of usable stripes */
4641         ndevs -= ndevs % devs_increment;
4642
4643         if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) {
4644                 ret = -ENOSPC;
4645                 goto error;
4646         }
4647
4648         if (devs_max && ndevs > devs_max)
4649                 ndevs = devs_max;
4650         /*
4651          * The primary goal is to maximize the number of stripes, so use as
4652          * many devices as possible, even if the stripes are not maximum sized.
4653          *
4654          * The DUP profile stores more than one stripe per device, the
4655          * max_avail is the total size so we have to adjust.
4656          */
4657         stripe_size = div_u64(devices_info[ndevs - 1].max_avail, dev_stripes);
4658         num_stripes = ndevs * dev_stripes;
4659
4660         /*
4661          * this will have to be fixed for RAID1 and RAID10 over
4662          * more drives
4663          */
4664         data_stripes = num_stripes / ncopies;
4665
4666         if (type & BTRFS_BLOCK_GROUP_RAID5) {
4667                 raid_stripe_len = find_raid56_stripe_len(ndevs - 1,
4668                                  btrfs_super_stripesize(info->super_copy));
4669                 data_stripes = num_stripes - 1;
4670         }
4671         if (type & BTRFS_BLOCK_GROUP_RAID6) {
4672                 raid_stripe_len = find_raid56_stripe_len(ndevs - 2,
4673                                  btrfs_super_stripesize(info->super_copy));
4674                 data_stripes = num_stripes - 2;
4675         }
4676
4677         /*
4678          * Use the number of data stripes to figure out how big this chunk
4679          * is really going to be in terms of logical address space,
4680          * and compare that answer with the max chunk size
4681          */
4682         if (stripe_size * data_stripes > max_chunk_size) {
4683                 u64 mask = (1ULL << 24) - 1;
4684
4685                 stripe_size = div_u64(max_chunk_size, data_stripes);
4686
4687                 /* bump the answer up to a 16MB boundary */
4688                 stripe_size = (stripe_size + mask) & ~mask;
4689
4690                 /* but don't go higher than the limits we found
4691                  * while searching for free extents
4692                  */
4693                 if (stripe_size > devices_info[ndevs-1].max_avail)
4694                         stripe_size = devices_info[ndevs-1].max_avail;
4695         }
4696
4697         /* align to BTRFS_STRIPE_LEN */
4698         stripe_size = div_u64(stripe_size, raid_stripe_len);
4699         stripe_size *= raid_stripe_len;
4700
4701         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
4702         if (!map) {
4703                 ret = -ENOMEM;
4704                 goto error;
4705         }
4706         map->num_stripes = num_stripes;
4707
4708         for (i = 0; i < ndevs; ++i) {
4709                 for (j = 0; j < dev_stripes; ++j) {
4710                         int s = i * dev_stripes + j;
4711                         map->stripes[s].dev = devices_info[i].dev;
4712                         map->stripes[s].physical = devices_info[i].dev_offset +
4713                                                    j * stripe_size;
4714                 }
4715         }
4716         map->sector_size = extent_root->sectorsize;
4717         map->stripe_len = raid_stripe_len;
4718         map->io_align = raid_stripe_len;
4719         map->io_width = raid_stripe_len;
4720         map->type = type;
4721         map->sub_stripes = sub_stripes;
4722
4723         num_bytes = stripe_size * data_stripes;
4724
4725         trace_btrfs_chunk_alloc(info->chunk_root, map, start, num_bytes);
4726
4727         em = alloc_extent_map();
4728         if (!em) {
4729                 kfree(map);
4730                 ret = -ENOMEM;
4731                 goto error;
4732         }
4733         set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
4734         em->map_lookup = map;
4735         em->start = start;
4736         em->len = num_bytes;
4737         em->block_start = 0;
4738         em->block_len = em->len;
4739         em->orig_block_len = stripe_size;
4740
4741         em_tree = &extent_root->fs_info->mapping_tree.map_tree;
4742         write_lock(&em_tree->lock);
4743         ret = add_extent_mapping(em_tree, em, 0);
4744         if (!ret) {
4745                 list_add_tail(&em->list, &trans->transaction->pending_chunks);
4746                 atomic_inc(&em->refs);
4747         }
4748         write_unlock(&em_tree->lock);
4749         if (ret) {
4750                 free_extent_map(em);
4751                 goto error;
4752         }
4753
4754         ret = btrfs_make_block_group(trans, extent_root, 0, type,
4755                                      BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4756                                      start, num_bytes);
4757         if (ret)
4758                 goto error_del_extent;
4759
4760         for (i = 0; i < map->num_stripes; i++) {
4761                 num_bytes = map->stripes[i].dev->bytes_used + stripe_size;
4762                 btrfs_device_set_bytes_used(map->stripes[i].dev, num_bytes);
4763                 map->stripes[i].dev->has_pending_chunks = true;
4764         }
4765
4766         spin_lock(&extent_root->fs_info->free_chunk_lock);
4767         extent_root->fs_info->free_chunk_space -= (stripe_size *
4768                                                    map->num_stripes);
4769         spin_unlock(&extent_root->fs_info->free_chunk_lock);
4770
4771         free_extent_map(em);
4772         check_raid56_incompat_flag(extent_root->fs_info, type);
4773
4774         kfree(devices_info);
4775         return 0;
4776
4777 error_del_extent:
4778         write_lock(&em_tree->lock);
4779         remove_extent_mapping(em_tree, em);
4780         write_unlock(&em_tree->lock);
4781
4782         /* One for our allocation */
4783         free_extent_map(em);
4784         /* One for the tree reference */
4785         free_extent_map(em);
4786         /* One for the pending_chunks list reference */
4787         free_extent_map(em);
4788 error:
4789         kfree(devices_info);
4790         return ret;
4791 }
4792
4793 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans,
4794                                 struct btrfs_root *extent_root,
4795                                 u64 chunk_offset, u64 chunk_size)
4796 {
4797         struct btrfs_key key;
4798         struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
4799         struct btrfs_device *device;
4800         struct btrfs_chunk *chunk;
4801         struct btrfs_stripe *stripe;
4802         struct extent_map_tree *em_tree;
4803         struct extent_map *em;
4804         struct map_lookup *map;
4805         size_t item_size;
4806         u64 dev_offset;
4807         u64 stripe_size;
4808         int i = 0;
4809         int ret;
4810
4811         em_tree = &extent_root->fs_info->mapping_tree.map_tree;
4812         read_lock(&em_tree->lock);
4813         em = lookup_extent_mapping(em_tree, chunk_offset, chunk_size);
4814         read_unlock(&em_tree->lock);
4815
4816         if (!em) {
4817                 btrfs_crit(extent_root->fs_info, "unable to find logical "
4818                            "%Lu len %Lu", chunk_offset, chunk_size);
4819                 return -EINVAL;
4820         }
4821
4822         if (em->start != chunk_offset || em->len != chunk_size) {
4823                 btrfs_crit(extent_root->fs_info, "found a bad mapping, wanted"
4824                           " %Lu-%Lu, found %Lu-%Lu", chunk_offset,
4825                           chunk_size, em->start, em->len);
4826                 free_extent_map(em);
4827                 return -EINVAL;
4828         }
4829
4830         map = em->map_lookup;
4831         item_size = btrfs_chunk_item_size(map->num_stripes);
4832         stripe_size = em->orig_block_len;
4833
4834         chunk = kzalloc(item_size, GFP_NOFS);
4835         if (!chunk) {
4836                 ret = -ENOMEM;
4837                 goto out;
4838         }
4839
4840         for (i = 0; i < map->num_stripes; i++) {
4841                 device = map->stripes[i].dev;
4842                 dev_offset = map->stripes[i].physical;
4843
4844                 ret = btrfs_update_device(trans, device);
4845                 if (ret)
4846                         goto out;
4847                 ret = btrfs_alloc_dev_extent(trans, device,
4848                                              chunk_root->root_key.objectid,
4849                                              BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4850                                              chunk_offset, dev_offset,
4851                                              stripe_size);
4852                 if (ret)
4853                         goto out;
4854         }
4855
4856         stripe = &chunk->stripe;
4857         for (i = 0; i < map->num_stripes; i++) {
4858                 device = map->stripes[i].dev;
4859                 dev_offset = map->stripes[i].physical;
4860
4861                 btrfs_set_stack_stripe_devid(stripe, device->devid);
4862                 btrfs_set_stack_stripe_offset(stripe, dev_offset);
4863                 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
4864                 stripe++;
4865         }
4866
4867         btrfs_set_stack_chunk_length(chunk, chunk_size);
4868         btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
4869         btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
4870         btrfs_set_stack_chunk_type(chunk, map->type);
4871         btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
4872         btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
4873         btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
4874         btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
4875         btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
4876
4877         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
4878         key.type = BTRFS_CHUNK_ITEM_KEY;
4879         key.offset = chunk_offset;
4880
4881         ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
4882         if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
4883                 /*
4884                  * TODO: Cleanup of inserted chunk root in case of
4885                  * failure.
4886                  */
4887                 ret = btrfs_add_system_chunk(chunk_root, &key, chunk,
4888                                              item_size);
4889         }
4890
4891 out:
4892         kfree(chunk);
4893         free_extent_map(em);
4894         return ret;
4895 }
4896
4897 /*
4898  * Chunk allocation falls into two parts. The first part does works
4899  * that make the new allocated chunk useable, but not do any operation
4900  * that modifies the chunk tree. The second part does the works that
4901  * require modifying the chunk tree. This division is important for the
4902  * bootstrap process of adding storage to a seed btrfs.
4903  */
4904 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4905                       struct btrfs_root *extent_root, u64 type)
4906 {
4907         u64 chunk_offset;
4908
4909         ASSERT(mutex_is_locked(&extent_root->fs_info->chunk_mutex));
4910         chunk_offset = find_next_chunk(extent_root->fs_info);
4911         return __btrfs_alloc_chunk(trans, extent_root, chunk_offset, type);
4912 }
4913
4914 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
4915                                          struct btrfs_root *root,
4916                                          struct btrfs_device *device)
4917 {
4918         u64 chunk_offset;
4919         u64 sys_chunk_offset;
4920         u64 alloc_profile;
4921         struct btrfs_fs_info *fs_info = root->fs_info;
4922         struct btrfs_root *extent_root = fs_info->extent_root;
4923         int ret;
4924
4925         chunk_offset = find_next_chunk(fs_info);
4926         alloc_profile = btrfs_get_alloc_profile(extent_root, 0);
4927         ret = __btrfs_alloc_chunk(trans, extent_root, chunk_offset,
4928                                   alloc_profile);
4929         if (ret)
4930                 return ret;
4931
4932         sys_chunk_offset = find_next_chunk(root->fs_info);
4933         alloc_profile = btrfs_get_alloc_profile(fs_info->chunk_root, 0);
4934         ret = __btrfs_alloc_chunk(trans, extent_root, sys_chunk_offset,
4935                                   alloc_profile);
4936         return ret;
4937 }
4938
4939 static inline int btrfs_chunk_max_errors(struct map_lookup *map)
4940 {
4941         int max_errors;
4942
4943         if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
4944                          BTRFS_BLOCK_GROUP_RAID10 |
4945                          BTRFS_BLOCK_GROUP_RAID5)) {
4946                 max_errors = 1;
4947         } else if (map->type & BTRFS_BLOCK_GROUP_RAID6) {
4948                 max_errors = 2;
4949         } else {
4950                 max_errors = 0;
4951         }
4952
4953         return max_errors;
4954 }
4955
4956 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
4957 {
4958         struct extent_map *em;
4959         struct map_lookup *map;
4960         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
4961         int readonly = 0;
4962         int miss_ndevs = 0;
4963         int i;
4964
4965         read_lock(&map_tree->map_tree.lock);
4966         em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
4967         read_unlock(&map_tree->map_tree.lock);
4968         if (!em)
4969                 return 1;
4970
4971         map = em->map_lookup;
4972         for (i = 0; i < map->num_stripes; i++) {
4973                 if (map->stripes[i].dev->missing) {
4974                         miss_ndevs++;
4975                         continue;
4976                 }
4977
4978                 if (!map->stripes[i].dev->writeable) {
4979                         readonly = 1;
4980                         goto end;
4981                 }
4982         }
4983
4984         /*
4985          * If the number of missing devices is larger than max errors,
4986          * we can not write the data into that chunk successfully, so
4987          * set it readonly.
4988          */
4989         if (miss_ndevs > btrfs_chunk_max_errors(map))
4990                 readonly = 1;
4991 end:
4992         free_extent_map(em);
4993         return readonly;
4994 }
4995
4996 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
4997 {
4998         extent_map_tree_init(&tree->map_tree);
4999 }
5000
5001 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
5002 {
5003         struct extent_map *em;
5004
5005         while (1) {
5006                 write_lock(&tree->map_tree.lock);
5007                 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
5008                 if (em)
5009                         remove_extent_mapping(&tree->map_tree, em);
5010                 write_unlock(&tree->map_tree.lock);
5011                 if (!em)
5012                         break;
5013                 /* once for us */
5014                 free_extent_map(em);
5015                 /* once for the tree */
5016                 free_extent_map(em);
5017         }
5018 }
5019
5020 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5021 {
5022         struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
5023         struct extent_map *em;
5024         struct map_lookup *map;
5025         struct extent_map_tree *em_tree = &map_tree->map_tree;
5026         int ret;
5027
5028         read_lock(&em_tree->lock);
5029         em = lookup_extent_mapping(em_tree, logical, len);
5030         read_unlock(&em_tree->lock);
5031
5032         /*
5033          * We could return errors for these cases, but that could get ugly and
5034          * we'd probably do the same thing which is just not do anything else
5035          * and exit, so return 1 so the callers don't try to use other copies.
5036          */
5037         if (!em) {
5038                 btrfs_crit(fs_info, "No mapping for %Lu-%Lu", logical,
5039                             logical+len);
5040                 return 1;
5041         }
5042
5043         if (em->start > logical || em->start + em->len < logical) {
5044                 btrfs_crit(fs_info, "Invalid mapping for %Lu-%Lu, got "
5045                             "%Lu-%Lu", logical, logical+len, em->start,
5046                             em->start + em->len);
5047                 free_extent_map(em);
5048                 return 1;
5049         }
5050
5051         map = em->map_lookup;
5052         if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
5053                 ret = map->num_stripes;
5054         else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5055                 ret = map->sub_stripes;
5056         else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
5057                 ret = 2;
5058         else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5059                 /*
5060                  * There could be two corrupted data stripes, we need
5061                  * to loop retry in order to rebuild the correct data.
5062                  *
5063                  * Fail a stripe at a time on every retry except the
5064                  * stripe under reconstruction.
5065                  */
5066                 ret = map->num_stripes;
5067         else
5068                 ret = 1;
5069         free_extent_map(em);
5070
5071         btrfs_dev_replace_lock(&fs_info->dev_replace);
5072         if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace))
5073                 ret++;
5074         btrfs_dev_replace_unlock(&fs_info->dev_replace);
5075
5076         return ret;
5077 }
5078
5079 unsigned long btrfs_full_stripe_len(struct btrfs_root *root,
5080                                     struct btrfs_mapping_tree *map_tree,
5081                                     u64 logical)
5082 {
5083         struct extent_map *em;
5084         struct map_lookup *map;
5085         struct extent_map_tree *em_tree = &map_tree->map_tree;
5086         unsigned long len = root->sectorsize;
5087
5088         read_lock(&em_tree->lock);
5089         em = lookup_extent_mapping(em_tree, logical, len);
5090         read_unlock(&em_tree->lock);
5091         BUG_ON(!em);
5092
5093         BUG_ON(em->start > logical || em->start + em->len < logical);
5094         map = em->map_lookup;
5095         if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5096                 len = map->stripe_len * nr_data_stripes(map);
5097         free_extent_map(em);
5098         return len;
5099 }
5100
5101 int btrfs_is_parity_mirror(struct btrfs_mapping_tree *map_tree,
5102                            u64 logical, u64 len, int mirror_num)
5103 {
5104         struct extent_map *em;
5105         struct map_lookup *map;
5106         struct extent_map_tree *em_tree = &map_tree->map_tree;
5107         int ret = 0;
5108
5109         read_lock(&em_tree->lock);
5110         em = lookup_extent_mapping(em_tree, logical, len);
5111         read_unlock(&em_tree->lock);
5112         BUG_ON(!em);
5113
5114         BUG_ON(em->start > logical || em->start + em->len < logical);
5115         map = em->map_lookup;
5116         if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5117                 ret = 1;
5118         free_extent_map(em);
5119         return ret;
5120 }
5121
5122 static int find_live_mirror(struct btrfs_fs_info *fs_info,
5123                             struct map_lookup *map, int first, int num,
5124                             int optimal, int dev_replace_is_ongoing)
5125 {
5126         int i;
5127         int tolerance;
5128         struct btrfs_device *srcdev;
5129
5130         if (dev_replace_is_ongoing &&
5131             fs_info->dev_replace.cont_reading_from_srcdev_mode ==
5132              BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
5133                 srcdev = fs_info->dev_replace.srcdev;
5134         else
5135                 srcdev = NULL;
5136
5137         /*
5138          * try to avoid the drive that is the source drive for a
5139          * dev-replace procedure, only choose it if no other non-missing
5140          * mirror is available
5141          */
5142         for (tolerance = 0; tolerance < 2; tolerance++) {
5143                 if (map->stripes[optimal].dev->bdev &&
5144                     (tolerance || map->stripes[optimal].dev != srcdev))
5145                         return optimal;
5146                 for (i = first; i < first + num; i++) {
5147                         if (map->stripes[i].dev->bdev &&
5148                             (tolerance || map->stripes[i].dev != srcdev))
5149                                 return i;
5150                 }
5151         }
5152
5153         /* we couldn't find one that doesn't fail.  Just return something
5154          * and the io error handling code will clean up eventually
5155          */
5156         return optimal;
5157 }
5158
5159 static inline int parity_smaller(u64 a, u64 b)
5160 {
5161         return a > b;
5162 }
5163
5164 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
5165 static void sort_parity_stripes(struct btrfs_bio *bbio, int num_stripes)
5166 {
5167         struct btrfs_bio_stripe s;
5168         int i;
5169         u64 l;
5170         int again = 1;
5171
5172         while (again) {
5173                 again = 0;
5174                 for (i = 0; i < num_stripes - 1; i++) {
5175                         if (parity_smaller(bbio->raid_map[i],
5176                                            bbio->raid_map[i+1])) {
5177                                 s = bbio->stripes[i];
5178                                 l = bbio->raid_map[i];
5179                                 bbio->stripes[i] = bbio->stripes[i+1];
5180                                 bbio->raid_map[i] = bbio->raid_map[i+1];
5181                                 bbio->stripes[i+1] = s;
5182                                 bbio->raid_map[i+1] = l;
5183
5184                                 again = 1;
5185                         }
5186                 }
5187         }
5188 }
5189
5190 static struct btrfs_bio *alloc_btrfs_bio(int total_stripes, int real_stripes)
5191 {
5192         struct btrfs_bio *bbio = kzalloc(
5193                  /* the size of the btrfs_bio */
5194                 sizeof(struct btrfs_bio) +
5195                 /* plus the variable array for the stripes */
5196                 sizeof(struct btrfs_bio_stripe) * (total_stripes) +
5197                 /* plus the variable array for the tgt dev */
5198                 sizeof(int) * (real_stripes) +
5199                 /*
5200                  * plus the raid_map, which includes both the tgt dev
5201                  * and the stripes
5202                  */
5203                 sizeof(u64) * (total_stripes),
5204                 GFP_NOFS|__GFP_NOFAIL);
5205
5206         atomic_set(&bbio->error, 0);
5207         atomic_set(&bbio->refs, 1);
5208
5209         return bbio;
5210 }
5211
5212 void btrfs_get_bbio(struct btrfs_bio *bbio)
5213 {
5214         WARN_ON(!atomic_read(&bbio->refs));
5215         atomic_inc(&bbio->refs);
5216 }
5217
5218 void btrfs_put_bbio(struct btrfs_bio *bbio)
5219 {
5220         if (!bbio)
5221                 return;
5222         if (atomic_dec_and_test(&bbio->refs))
5223                 kfree(bbio);
5224 }
5225
5226 static int __btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
5227                              u64 logical, u64 *length,
5228                              struct btrfs_bio **bbio_ret,
5229                              int mirror_num, int need_raid_map)
5230 {
5231         struct extent_map *em;
5232         struct map_lookup *map;
5233         struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
5234         struct extent_map_tree *em_tree = &map_tree->map_tree;
5235         u64 offset;
5236         u64 stripe_offset;
5237         u64 stripe_end_offset;
5238         u64 stripe_nr;
5239         u64 stripe_nr_orig;
5240         u64 stripe_nr_end;
5241         u64 stripe_len;
5242         u32 stripe_index;
5243         int i;
5244         int ret = 0;
5245         int num_stripes;
5246         int max_errors = 0;
5247         int tgtdev_indexes = 0;
5248         struct btrfs_bio *bbio = NULL;
5249         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
5250         int dev_replace_is_ongoing = 0;
5251         int num_alloc_stripes;
5252         int patch_the_first_stripe_for_dev_replace = 0;
5253         u64 physical_to_patch_in_first_stripe = 0;
5254         u64 raid56_full_stripe_start = (u64)-1;
5255
5256         read_lock(&em_tree->lock);
5257         em = lookup_extent_mapping(em_tree, logical, *length);
5258         read_unlock(&em_tree->lock);
5259
5260         if (!em) {
5261                 btrfs_crit(fs_info, "unable to find logical %llu len %llu",
5262                         logical, *length);
5263                 return -EINVAL;
5264         }
5265
5266         if (em->start > logical || em->start + em->len < logical) {
5267                 btrfs_crit(fs_info, "found a bad mapping, wanted %Lu, "
5268                            "found %Lu-%Lu", logical, em->start,
5269                            em->start + em->len);
5270                 free_extent_map(em);
5271                 return -EINVAL;
5272         }
5273
5274         map = em->map_lookup;
5275         offset = logical - em->start;
5276
5277         stripe_len = map->stripe_len;
5278         stripe_nr = offset;
5279         /*
5280          * stripe_nr counts the total number of stripes we have to stride
5281          * to get to this block
5282          */
5283         stripe_nr = div64_u64(stripe_nr, stripe_len);
5284
5285         stripe_offset = stripe_nr * stripe_len;
5286         BUG_ON(offset < stripe_offset);
5287
5288         /* stripe_offset is the offset of this block in its stripe*/
5289         stripe_offset = offset - stripe_offset;
5290
5291         /* if we're here for raid56, we need to know the stripe aligned start */
5292         if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5293                 unsigned long full_stripe_len = stripe_len * nr_data_stripes(map);
5294                 raid56_full_stripe_start = offset;
5295
5296                 /* allow a write of a full stripe, but make sure we don't
5297                  * allow straddling of stripes
5298                  */
5299                 raid56_full_stripe_start = div64_u64(raid56_full_stripe_start,
5300                                 full_stripe_len);
5301                 raid56_full_stripe_start *= full_stripe_len;
5302         }
5303
5304         if (rw & REQ_DISCARD) {
5305                 /* we don't discard raid56 yet */
5306                 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5307                         ret = -EOPNOTSUPP;
5308                         goto out;
5309                 }
5310                 *length = min_t(u64, em->len - offset, *length);
5311         } else if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
5312                 u64 max_len;
5313                 /* For writes to RAID[56], allow a full stripeset across all disks.
5314                    For other RAID types and for RAID[56] reads, just allow a single
5315                    stripe (on a single disk). */
5316                 if ((map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
5317                     (rw & REQ_WRITE)) {
5318                         max_len = stripe_len * nr_data_stripes(map) -
5319                                 (offset - raid56_full_stripe_start);
5320                 } else {
5321                         /* we limit the length of each bio to what fits in a stripe */
5322                         max_len = stripe_len - stripe_offset;
5323                 }
5324                 *length = min_t(u64, em->len - offset, max_len);
5325         } else {
5326                 *length = em->len - offset;
5327         }
5328
5329         /* This is for when we're called from btrfs_merge_bio_hook() and all
5330            it cares about is the length */
5331         if (!bbio_ret)
5332                 goto out;
5333
5334         btrfs_dev_replace_lock(dev_replace);
5335         dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
5336         if (!dev_replace_is_ongoing)
5337                 btrfs_dev_replace_unlock(dev_replace);
5338
5339         if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
5340             !(rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) &&
5341             dev_replace->tgtdev != NULL) {
5342                 /*
5343                  * in dev-replace case, for repair case (that's the only
5344                  * case where the mirror is selected explicitly when
5345                  * calling btrfs_map_block), blocks left of the left cursor
5346                  * can also be read from the target drive.
5347                  * For REQ_GET_READ_MIRRORS, the target drive is added as
5348                  * the last one to the array of stripes. For READ, it also
5349                  * needs to be supported using the same mirror number.
5350                  * If the requested block is not left of the left cursor,
5351                  * EIO is returned. This can happen because btrfs_num_copies()
5352                  * returns one more in the dev-replace case.
5353                  */
5354                 u64 tmp_length = *length;
5355                 struct btrfs_bio *tmp_bbio = NULL;
5356                 int tmp_num_stripes;
5357                 u64 srcdev_devid = dev_replace->srcdev->devid;
5358                 int index_srcdev = 0;
5359                 int found = 0;
5360                 u64 physical_of_found = 0;
5361
5362                 ret = __btrfs_map_block(fs_info, REQ_GET_READ_MIRRORS,
5363                              logical, &tmp_length, &tmp_bbio, 0, 0);
5364                 if (ret) {
5365                         WARN_ON(tmp_bbio != NULL);
5366                         goto out;
5367                 }
5368
5369                 tmp_num_stripes = tmp_bbio->num_stripes;
5370                 if (mirror_num > tmp_num_stripes) {
5371                         /*
5372                          * REQ_GET_READ_MIRRORS does not contain this
5373                          * mirror, that means that the requested area
5374                          * is not left of the left cursor
5375                          */
5376                         ret = -EIO;
5377                         btrfs_put_bbio(tmp_bbio);
5378                         goto out;
5379                 }
5380
5381                 /*
5382                  * process the rest of the function using the mirror_num
5383                  * of the source drive. Therefore look it up first.
5384                  * At the end, patch the device pointer to the one of the
5385                  * target drive.
5386                  */
5387                 for (i = 0; i < tmp_num_stripes; i++) {
5388                         if (tmp_bbio->stripes[i].dev->devid == srcdev_devid) {
5389                                 /*
5390                                  * In case of DUP, in order to keep it
5391                                  * simple, only add the mirror with the
5392                                  * lowest physical address
5393                                  */
5394                                 if (found &&
5395                                     physical_of_found <=
5396                                      tmp_bbio->stripes[i].physical)
5397                                         continue;
5398                                 index_srcdev = i;
5399                                 found = 1;
5400                                 physical_of_found =
5401                                         tmp_bbio->stripes[i].physical;
5402                         }
5403                 }
5404
5405                 if (found) {
5406                         mirror_num = index_srcdev + 1;
5407                         patch_the_first_stripe_for_dev_replace = 1;
5408                         physical_to_patch_in_first_stripe = physical_of_found;
5409                 } else {
5410                         WARN_ON(1);
5411                         ret = -EIO;
5412                         btrfs_put_bbio(tmp_bbio);
5413                         goto out;
5414                 }
5415
5416                 btrfs_put_bbio(tmp_bbio);
5417         } else if (mirror_num > map->num_stripes) {
5418                 mirror_num = 0;
5419         }
5420
5421         num_stripes = 1;
5422         stripe_index = 0;
5423         stripe_nr_orig = stripe_nr;
5424         stripe_nr_end = ALIGN(offset + *length, map->stripe_len);
5425         stripe_nr_end = div_u64(stripe_nr_end, map->stripe_len);
5426         stripe_end_offset = stripe_nr_end * map->stripe_len -
5427                             (offset + *length);
5428
5429         if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5430                 if (rw & REQ_DISCARD)
5431                         num_stripes = min_t(u64, map->num_stripes,
5432                                             stripe_nr_end - stripe_nr_orig);
5433                 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5434                                 &stripe_index);
5435                 if (!(rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)))
5436                         mirror_num = 1;
5437         } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
5438                 if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS))
5439                         num_stripes = map->num_stripes;
5440                 else if (mirror_num)
5441                         stripe_index = mirror_num - 1;
5442                 else {
5443                         stripe_index = find_live_mirror(fs_info, map, 0,
5444                                             map->num_stripes,
5445                                             current->pid % map->num_stripes,
5446                                             dev_replace_is_ongoing);
5447                         mirror_num = stripe_index + 1;
5448                 }
5449
5450         } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
5451                 if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) {
5452                         num_stripes = map->num_stripes;
5453                 } else if (mirror_num) {
5454                         stripe_index = mirror_num - 1;
5455                 } else {
5456                         mirror_num = 1;
5457                 }
5458
5459         } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5460                 u32 factor = map->num_stripes / map->sub_stripes;
5461
5462                 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
5463                 stripe_index *= map->sub_stripes;
5464
5465                 if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS))
5466                         num_stripes = map->sub_stripes;
5467                 else if (rw & REQ_DISCARD)
5468                         num_stripes = min_t(u64, map->sub_stripes *
5469                                             (stripe_nr_end - stripe_nr_orig),
5470                                             map->num_stripes);
5471                 else if (mirror_num)
5472                         stripe_index += mirror_num - 1;
5473                 else {
5474                         int old_stripe_index = stripe_index;
5475                         stripe_index = find_live_mirror(fs_info, map,
5476                                               stripe_index,
5477                                               map->sub_stripes, stripe_index +
5478                                               current->pid % map->sub_stripes,
5479                                               dev_replace_is_ongoing);
5480                         mirror_num = stripe_index - old_stripe_index + 1;
5481                 }
5482
5483         } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5484                 if (need_raid_map &&
5485                     ((rw & (REQ_WRITE | REQ_GET_READ_MIRRORS)) ||
5486                      mirror_num > 1)) {
5487                         /* push stripe_nr back to the start of the full stripe */
5488                         stripe_nr = div_u64(raid56_full_stripe_start,
5489                                         stripe_len * nr_data_stripes(map));
5490
5491                         /* RAID[56] write or recovery. Return all stripes */
5492                         num_stripes = map->num_stripes;
5493                         max_errors = nr_parity_stripes(map);
5494
5495                         *length = map->stripe_len;
5496                         stripe_index = 0;
5497                         stripe_offset = 0;
5498                 } else {
5499                         /*
5500                          * Mirror #0 or #1 means the original data block.
5501                          * Mirror #2 is RAID5 parity block.
5502                          * Mirror #3 is RAID6 Q block.
5503                          */
5504                         stripe_nr = div_u64_rem(stripe_nr,
5505                                         nr_data_stripes(map), &stripe_index);
5506                         if (mirror_num > 1)
5507                                 stripe_index = nr_data_stripes(map) +
5508                                                 mirror_num - 2;
5509
5510                         /* We distribute the parity blocks across stripes */
5511                         div_u64_rem(stripe_nr + stripe_index, map->num_stripes,
5512                                         &stripe_index);
5513                         if (!(rw & (REQ_WRITE | REQ_DISCARD |
5514                                     REQ_GET_READ_MIRRORS)) && mirror_num <= 1)
5515                                 mirror_num = 1;
5516                 }
5517         } else {
5518                 /*
5519                  * after this, stripe_nr is the number of stripes on this
5520                  * device we have to walk to find the data, and stripe_index is
5521                  * the number of our device in the stripe array
5522                  */
5523                 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5524                                 &stripe_index);
5525                 mirror_num = stripe_index + 1;
5526         }
5527         BUG_ON(stripe_index >= map->num_stripes);
5528
5529         num_alloc_stripes = num_stripes;
5530         if (dev_replace_is_ongoing) {
5531                 if (rw & (REQ_WRITE | REQ_DISCARD))
5532                         num_alloc_stripes <<= 1;
5533                 if (rw & REQ_GET_READ_MIRRORS)
5534                         num_alloc_stripes++;
5535                 tgtdev_indexes = num_stripes;
5536         }
5537
5538         bbio = alloc_btrfs_bio(num_alloc_stripes, tgtdev_indexes);
5539         if (!bbio) {
5540                 ret = -ENOMEM;
5541                 goto out;
5542         }
5543         if (dev_replace_is_ongoing)
5544                 bbio->tgtdev_map = (int *)(bbio->stripes + num_alloc_stripes);
5545
5546         /* build raid_map */
5547         if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK &&
5548             need_raid_map && ((rw & (REQ_WRITE | REQ_GET_READ_MIRRORS)) ||
5549             mirror_num > 1)) {
5550                 u64 tmp;
5551                 unsigned rot;
5552
5553                 bbio->raid_map = (u64 *)((void *)bbio->stripes +
5554                                  sizeof(struct btrfs_bio_stripe) *
5555                                  num_alloc_stripes +
5556                                  sizeof(int) * tgtdev_indexes);
5557
5558                 /* Work out the disk rotation on this stripe-set */
5559                 div_u64_rem(stripe_nr, num_stripes, &rot);
5560
5561                 /* Fill in the logical address of each stripe */
5562                 tmp = stripe_nr * nr_data_stripes(map);
5563                 for (i = 0; i < nr_data_stripes(map); i++)
5564                         bbio->raid_map[(i+rot) % num_stripes] =
5565                                 em->start + (tmp + i) * map->stripe_len;
5566
5567                 bbio->raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE;
5568                 if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5569                         bbio->raid_map[(i+rot+1) % num_stripes] =
5570                                 RAID6_Q_STRIPE;
5571         }
5572
5573         if (rw & REQ_DISCARD) {
5574                 u32 factor = 0;
5575                 u32 sub_stripes = 0;
5576                 u64 stripes_per_dev = 0;
5577                 u32 remaining_stripes = 0;
5578                 u32 last_stripe = 0;
5579
5580                 if (map->type &
5581                     (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID10)) {
5582                         if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5583                                 sub_stripes = 1;
5584                         else
5585                                 sub_stripes = map->sub_stripes;
5586
5587                         factor = map->num_stripes / sub_stripes;
5588                         stripes_per_dev = div_u64_rem(stripe_nr_end -
5589                                                       stripe_nr_orig,
5590                                                       factor,
5591                                                       &remaining_stripes);
5592                         div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
5593                         last_stripe *= sub_stripes;
5594                 }
5595
5596                 for (i = 0; i < num_stripes; i++) {
5597                         bbio->stripes[i].physical =
5598                                 map->stripes[stripe_index].physical +
5599                                 stripe_offset + stripe_nr * map->stripe_len;
5600                         bbio->stripes[i].dev = map->stripes[stripe_index].dev;
5601
5602                         if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5603                                          BTRFS_BLOCK_GROUP_RAID10)) {
5604                                 bbio->stripes[i].length = stripes_per_dev *
5605                                                           map->stripe_len;
5606
5607                                 if (i / sub_stripes < remaining_stripes)
5608                                         bbio->stripes[i].length +=
5609                                                 map->stripe_len;
5610
5611                                 /*
5612                                  * Special for the first stripe and
5613                                  * the last stripe:
5614                                  *
5615                                  * |-------|...|-------|
5616                                  *     |----------|
5617                                  *    off     end_off
5618                                  */
5619                                 if (i < sub_stripes)
5620                                         bbio->stripes[i].length -=
5621                                                 stripe_offset;
5622
5623                                 if (stripe_index >= last_stripe &&
5624                                     stripe_index <= (last_stripe +
5625                                                      sub_stripes - 1))
5626                                         bbio->stripes[i].length -=
5627                                                 stripe_end_offset;
5628
5629                                 if (i == sub_stripes - 1)
5630                                         stripe_offset = 0;
5631                         } else
5632                                 bbio->stripes[i].length = *length;
5633
5634                         stripe_index++;
5635                         if (stripe_index == map->num_stripes) {
5636                                 /* This could only happen for RAID0/10 */
5637                                 stripe_index = 0;
5638                                 stripe_nr++;
5639                         }
5640                 }
5641         } else {
5642                 for (i = 0; i < num_stripes; i++) {
5643                         bbio->stripes[i].physical =
5644                                 map->stripes[stripe_index].physical +
5645                                 stripe_offset +
5646                                 stripe_nr * map->stripe_len;
5647                         bbio->stripes[i].dev =
5648                                 map->stripes[stripe_index].dev;
5649                         stripe_index++;
5650                 }
5651         }
5652
5653         if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS))
5654                 max_errors = btrfs_chunk_max_errors(map);
5655
5656         if (bbio->raid_map)
5657                 sort_parity_stripes(bbio, num_stripes);
5658
5659         tgtdev_indexes = 0;
5660         if (dev_replace_is_ongoing && (rw & (REQ_WRITE | REQ_DISCARD)) &&
5661             dev_replace->tgtdev != NULL) {
5662                 int index_where_to_add;
5663                 u64 srcdev_devid = dev_replace->srcdev->devid;
5664
5665                 /*
5666                  * duplicate the write operations while the dev replace
5667                  * procedure is running. Since the copying of the old disk
5668                  * to the new disk takes place at run time while the
5669                  * filesystem is mounted writable, the regular write
5670                  * operations to the old disk have to be duplicated to go
5671                  * to the new disk as well.
5672                  * Note that device->missing is handled by the caller, and
5673                  * that the write to the old disk is already set up in the
5674                  * stripes array.
5675                  */
5676                 index_where_to_add = num_stripes;
5677                 for (i = 0; i < num_stripes; i++) {
5678                         if (bbio->stripes[i].dev->devid == srcdev_devid) {
5679                                 /* write to new disk, too */
5680                                 struct btrfs_bio_stripe *new =
5681                                         bbio->stripes + index_where_to_add;
5682                                 struct btrfs_bio_stripe *old =
5683                                         bbio->stripes + i;
5684
5685                                 new->physical = old->physical;
5686                                 new->length = old->length;
5687                                 new->dev = dev_replace->tgtdev;
5688                                 bbio->tgtdev_map[i] = index_where_to_add;
5689                                 index_where_to_add++;
5690                                 max_errors++;
5691                                 tgtdev_indexes++;
5692                         }
5693                 }
5694                 num_stripes = index_where_to_add;
5695         } else if (dev_replace_is_ongoing && (rw & REQ_GET_READ_MIRRORS) &&
5696                    dev_replace->tgtdev != NULL) {
5697                 u64 srcdev_devid = dev_replace->srcdev->devid;
5698                 int index_srcdev = 0;
5699                 int found = 0;
5700                 u64 physical_of_found = 0;
5701
5702                 /*
5703                  * During the dev-replace procedure, the target drive can
5704                  * also be used to read data in case it is needed to repair
5705                  * a corrupt block elsewhere. This is possible if the
5706                  * requested area is left of the left cursor. In this area,
5707                  * the target drive is a full copy of the source drive.
5708                  */
5709                 for (i = 0; i < num_stripes; i++) {
5710                         if (bbio->stripes[i].dev->devid == srcdev_devid) {
5711                                 /*
5712                                  * In case of DUP, in order to keep it
5713                                  * simple, only add the mirror with the
5714                                  * lowest physical address
5715                                  */
5716                                 if (found &&
5717                                     physical_of_found <=
5718                                      bbio->stripes[i].physical)
5719                                         continue;
5720                                 index_srcdev = i;
5721                                 found = 1;
5722                                 physical_of_found = bbio->stripes[i].physical;
5723                         }
5724                 }
5725                 if (found) {
5726                         if (physical_of_found + map->stripe_len <=
5727                             dev_replace->cursor_left) {
5728                                 struct btrfs_bio_stripe *tgtdev_stripe =
5729                                         bbio->stripes + num_stripes;
5730
5731                                 tgtdev_stripe->physical = physical_of_found;
5732                                 tgtdev_stripe->length =
5733                                         bbio->stripes[index_srcdev].length;
5734                                 tgtdev_stripe->dev = dev_replace->tgtdev;
5735                                 bbio->tgtdev_map[index_srcdev] = num_stripes;
5736
5737                                 tgtdev_indexes++;
5738                                 num_stripes++;
5739                         }
5740                 }
5741         }
5742
5743         *bbio_ret = bbio;
5744         bbio->map_type = map->type;
5745         bbio->num_stripes = num_stripes;
5746         bbio->max_errors = max_errors;
5747         bbio->mirror_num = mirror_num;
5748         bbio->num_tgtdevs = tgtdev_indexes;
5749
5750         /*
5751          * this is the case that REQ_READ && dev_replace_is_ongoing &&
5752          * mirror_num == num_stripes + 1 && dev_replace target drive is
5753          * available as a mirror
5754          */
5755         if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) {
5756                 WARN_ON(num_stripes > 1);
5757                 bbio->stripes[0].dev = dev_replace->tgtdev;
5758                 bbio->stripes[0].physical = physical_to_patch_in_first_stripe;
5759                 bbio->mirror_num = map->num_stripes + 1;
5760         }
5761 out:
5762         if (dev_replace_is_ongoing)
5763                 btrfs_dev_replace_unlock(dev_replace);
5764         free_extent_map(em);
5765         return ret;
5766 }
5767
5768 int btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
5769                       u64 logical, u64 *length,
5770                       struct btrfs_bio **bbio_ret, int mirror_num)
5771 {
5772         return __btrfs_map_block(fs_info, rw, logical, length, bbio_ret,
5773                                  mirror_num, 0);
5774 }
5775
5776 /* For Scrub/replace */
5777 int btrfs_map_sblock(struct btrfs_fs_info *fs_info, int rw,
5778                      u64 logical, u64 *length,
5779                      struct btrfs_bio **bbio_ret, int mirror_num,
5780                      int need_raid_map)
5781 {
5782         return __btrfs_map_block(fs_info, rw, logical, length, bbio_ret,
5783                                  mirror_num, need_raid_map);
5784 }
5785
5786 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
5787                      u64 chunk_start, u64 physical, u64 devid,
5788                      u64 **logical, int *naddrs, int *stripe_len)
5789 {
5790         struct extent_map_tree *em_tree = &map_tree->map_tree;
5791         struct extent_map *em;
5792         struct map_lookup *map;
5793         u64 *buf;
5794         u64 bytenr;
5795         u64 length;
5796         u64 stripe_nr;
5797         u64 rmap_len;
5798         int i, j, nr = 0;
5799
5800         read_lock(&em_tree->lock);
5801         em = lookup_extent_mapping(em_tree, chunk_start, 1);
5802         read_unlock(&em_tree->lock);
5803
5804         if (!em) {
5805                 printk(KERN_ERR "BTRFS: couldn't find em for chunk %Lu\n",
5806                        chunk_start);
5807                 return -EIO;
5808         }
5809
5810         if (em->start != chunk_start) {
5811                 printk(KERN_ERR "BTRFS: bad chunk start, em=%Lu, wanted=%Lu\n",
5812                        em->start, chunk_start);
5813                 free_extent_map(em);
5814                 return -EIO;
5815         }
5816         map = em->map_lookup;
5817
5818         length = em->len;
5819         rmap_len = map->stripe_len;
5820
5821         if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5822                 length = div_u64(length, map->num_stripes / map->sub_stripes);
5823         else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5824                 length = div_u64(length, map->num_stripes);
5825         else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5826                 length = div_u64(length, nr_data_stripes(map));
5827                 rmap_len = map->stripe_len * nr_data_stripes(map);
5828         }
5829
5830         buf = kcalloc(map->num_stripes, sizeof(u64), GFP_NOFS);
5831         BUG_ON(!buf); /* -ENOMEM */
5832
5833         for (i = 0; i < map->num_stripes; i++) {
5834                 if (devid && map->stripes[i].dev->devid != devid)
5835                         continue;
5836                 if (map->stripes[i].physical > physical ||
5837                     map->stripes[i].physical + length <= physical)
5838                         continue;
5839
5840                 stripe_nr = physical - map->stripes[i].physical;
5841                 stripe_nr = div_u64(stripe_nr, map->stripe_len);
5842
5843                 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5844                         stripe_nr = stripe_nr * map->num_stripes + i;
5845                         stripe_nr = div_u64(stripe_nr, map->sub_stripes);
5846                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5847                         stripe_nr = stripe_nr * map->num_stripes + i;
5848                 } /* else if RAID[56], multiply by nr_data_stripes().
5849                    * Alternatively, just use rmap_len below instead of
5850                    * map->stripe_len */
5851
5852                 bytenr = chunk_start + stripe_nr * rmap_len;
5853                 WARN_ON(nr >= map->num_stripes);
5854                 for (j = 0; j < nr; j++) {
5855                         if (buf[j] == bytenr)
5856                                 break;
5857                 }
5858                 if (j == nr) {
5859                         WARN_ON(nr >= map->num_stripes);
5860                         buf[nr++] = bytenr;
5861                 }
5862         }
5863
5864         *logical = buf;
5865         *naddrs = nr;
5866         *stripe_len = rmap_len;
5867
5868         free_extent_map(em);
5869         return 0;
5870 }
5871
5872 static inline void btrfs_end_bbio(struct btrfs_bio *bbio, struct bio *bio)
5873 {
5874         bio->bi_private = bbio->private;
5875         bio->bi_end_io = bbio->end_io;
5876         bio_endio(bio);
5877
5878         btrfs_put_bbio(bbio);
5879 }
5880
5881 static void btrfs_end_bio(struct bio *bio)
5882 {
5883         struct btrfs_bio *bbio = bio->bi_private;
5884         int is_orig_bio = 0;
5885
5886         if (bio->bi_error) {
5887                 atomic_inc(&bbio->error);
5888                 if (bio->bi_error == -EIO || bio->bi_error == -EREMOTEIO) {
5889                         unsigned int stripe_index =
5890                                 btrfs_io_bio(bio)->stripe_index;
5891                         struct btrfs_device *dev;
5892
5893                         BUG_ON(stripe_index >= bbio->num_stripes);
5894                         dev = bbio->stripes[stripe_index].dev;
5895                         if (dev->bdev) {
5896                                 if (bio->bi_rw & WRITE)
5897                                         btrfs_dev_stat_inc(dev,
5898                                                 BTRFS_DEV_STAT_WRITE_ERRS);
5899                                 else
5900                                         btrfs_dev_stat_inc(dev,
5901                                                 BTRFS_DEV_STAT_READ_ERRS);
5902                                 if ((bio->bi_rw & WRITE_FLUSH) == WRITE_FLUSH)
5903                                         btrfs_dev_stat_inc(dev,
5904                                                 BTRFS_DEV_STAT_FLUSH_ERRS);
5905                                 btrfs_dev_stat_print_on_error(dev);
5906                         }
5907                 }
5908         }
5909
5910         if (bio == bbio->orig_bio)
5911                 is_orig_bio = 1;
5912
5913         btrfs_bio_counter_dec(bbio->fs_info);
5914
5915         if (atomic_dec_and_test(&bbio->stripes_pending)) {
5916                 if (!is_orig_bio) {
5917                         bio_put(bio);
5918                         bio = bbio->orig_bio;
5919                 }
5920
5921                 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
5922                 /* only send an error to the higher layers if it is
5923                  * beyond the tolerance of the btrfs bio
5924                  */
5925                 if (atomic_read(&bbio->error) > bbio->max_errors) {
5926                         bio->bi_error = -EIO;
5927                 } else {
5928                         /*
5929                          * this bio is actually up to date, we didn't
5930                          * go over the max number of errors
5931                          */
5932                         bio->bi_error = 0;
5933                 }
5934
5935                 btrfs_end_bbio(bbio, bio);
5936         } else if (!is_orig_bio) {
5937                 bio_put(bio);
5938         }
5939 }
5940
5941 /*
5942  * see run_scheduled_bios for a description of why bios are collected for
5943  * async submit.
5944  *
5945  * This will add one bio to the pending list for a device and make sure
5946  * the work struct is scheduled.
5947  */
5948 static noinline void btrfs_schedule_bio(struct btrfs_root *root,
5949                                         struct btrfs_device *device,
5950                                         int rw, struct bio *bio)
5951 {
5952         int should_queue = 1;
5953         struct btrfs_pending_bios *pending_bios;
5954
5955         if (device->missing || !device->bdev) {
5956                 bio_io_error(bio);
5957                 return;
5958         }
5959
5960         /* don't bother with additional async steps for reads, right now */
5961         if (!(rw & REQ_WRITE)) {
5962                 bio_get(bio);
5963                 btrfsic_submit_bio(rw, bio);
5964                 bio_put(bio);
5965                 return;
5966         }
5967
5968         /*
5969          * nr_async_bios allows us to reliably return congestion to the
5970          * higher layers.  Otherwise, the async bio makes it appear we have
5971          * made progress against dirty pages when we've really just put it
5972          * on a queue for later
5973          */
5974         atomic_inc(&root->fs_info->nr_async_bios);
5975         WARN_ON(bio->bi_next);
5976         bio->bi_next = NULL;
5977         bio->bi_rw |= rw;
5978
5979         spin_lock(&device->io_lock);
5980         if (bio->bi_rw & REQ_SYNC)
5981                 pending_bios = &device->pending_sync_bios;
5982         else
5983                 pending_bios = &device->pending_bios;
5984
5985         if (pending_bios->tail)
5986                 pending_bios->tail->bi_next = bio;
5987
5988         pending_bios->tail = bio;
5989         if (!pending_bios->head)
5990                 pending_bios->head = bio;
5991         if (device->running_pending)
5992                 should_queue = 0;
5993
5994         spin_unlock(&device->io_lock);
5995
5996         if (should_queue)
5997                 btrfs_queue_work(root->fs_info->submit_workers,
5998                                  &device->work);
5999 }
6000
6001 static void submit_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio,
6002                               struct bio *bio, u64 physical, int dev_nr,
6003                               int rw, int async)
6004 {
6005         struct btrfs_device *dev = bbio->stripes[dev_nr].dev;
6006
6007         bio->bi_private = bbio;
6008         btrfs_io_bio(bio)->stripe_index = dev_nr;
6009         bio->bi_end_io = btrfs_end_bio;
6010         bio->bi_iter.bi_sector = physical >> 9;
6011 #ifdef DEBUG
6012         {
6013                 struct rcu_string *name;
6014
6015                 rcu_read_lock();
6016                 name = rcu_dereference(dev->name);
6017                 pr_debug("btrfs_map_bio: rw %d, sector=%llu, dev=%lu "
6018                          "(%s id %llu), size=%u\n", rw,
6019                          (u64)bio->bi_iter.bi_sector, (u_long)dev->bdev->bd_dev,
6020                          name->str, dev->devid, bio->bi_iter.bi_size);
6021                 rcu_read_unlock();
6022         }
6023 #endif
6024         bio->bi_bdev = dev->bdev;
6025
6026         btrfs_bio_counter_inc_noblocked(root->fs_info);
6027
6028         if (async)
6029                 btrfs_schedule_bio(root, dev, rw, bio);
6030         else
6031                 btrfsic_submit_bio(rw, bio);
6032 }
6033
6034 static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical)
6035 {
6036         atomic_inc(&bbio->error);
6037         if (atomic_dec_and_test(&bbio->stripes_pending)) {
6038                 /* Shoud be the original bio. */
6039                 WARN_ON(bio != bbio->orig_bio);
6040
6041                 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
6042                 bio->bi_iter.bi_sector = logical >> 9;
6043                 bio->bi_error = -EIO;
6044                 btrfs_end_bbio(bbio, bio);
6045         }
6046 }
6047
6048 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
6049                   int mirror_num, int async_submit)
6050 {
6051         struct btrfs_device *dev;
6052         struct bio *first_bio = bio;
6053         u64 logical = (u64)bio->bi_iter.bi_sector << 9;
6054         u64 length = 0;
6055         u64 map_length;
6056         int ret;
6057         int dev_nr;
6058         int total_devs;
6059         struct btrfs_bio *bbio = NULL;
6060
6061         length = bio->bi_iter.bi_size;
6062         map_length = length;
6063
6064         btrfs_bio_counter_inc_blocked(root->fs_info);
6065         ret = __btrfs_map_block(root->fs_info, rw, logical, &map_length, &bbio,
6066                               mirror_num, 1);
6067         if (ret) {
6068                 btrfs_bio_counter_dec(root->fs_info);
6069                 return ret;
6070         }
6071
6072         total_devs = bbio->num_stripes;
6073         bbio->orig_bio = first_bio;
6074         bbio->private = first_bio->bi_private;
6075         bbio->end_io = first_bio->bi_end_io;
6076         bbio->fs_info = root->fs_info;
6077         atomic_set(&bbio->stripes_pending, bbio->num_stripes);
6078
6079         if (bbio->raid_map) {
6080                 /* In this case, map_length has been set to the length of
6081                    a single stripe; not the whole write */
6082                 if (rw & WRITE) {
6083                         ret = raid56_parity_write(root, bio, bbio, map_length);
6084                 } else {
6085                         ret = raid56_parity_recover(root, bio, bbio, map_length,
6086                                                     mirror_num, 1);
6087                 }
6088
6089                 btrfs_bio_counter_dec(root->fs_info);
6090                 return ret;
6091         }
6092
6093         if (map_length < length) {
6094                 btrfs_crit(root->fs_info, "mapping failed logical %llu bio len %llu len %llu",
6095                         logical, length, map_length);
6096                 BUG();
6097         }
6098
6099         for (dev_nr = 0; dev_nr < total_devs; dev_nr++) {
6100                 dev = bbio->stripes[dev_nr].dev;
6101                 if (!dev || !dev->bdev || (rw & WRITE && !dev->writeable)) {
6102                         bbio_error(bbio, first_bio, logical);
6103                         continue;
6104                 }
6105
6106                 if (dev_nr < total_devs - 1) {
6107                         bio = btrfs_bio_clone(first_bio, GFP_NOFS);
6108                         BUG_ON(!bio); /* -ENOMEM */
6109                 } else
6110                         bio = first_bio;
6111
6112                 submit_stripe_bio(root, bbio, bio,
6113                                   bbio->stripes[dev_nr].physical, dev_nr, rw,
6114                                   async_submit);
6115         }
6116         btrfs_bio_counter_dec(root->fs_info);
6117         return 0;
6118 }
6119
6120 struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid,
6121                                        u8 *uuid, u8 *fsid)
6122 {
6123         struct btrfs_device *device;
6124         struct btrfs_fs_devices *cur_devices;
6125
6126         cur_devices = fs_info->fs_devices;
6127         while (cur_devices) {
6128                 if (!fsid ||
6129                     !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
6130                         device = __find_device(&cur_devices->devices,
6131                                                devid, uuid);
6132                         if (device)
6133                                 return device;
6134                 }
6135                 cur_devices = cur_devices->seed;
6136         }
6137         return NULL;
6138 }
6139
6140 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
6141                                             struct btrfs_fs_devices *fs_devices,
6142                                             u64 devid, u8 *dev_uuid)
6143 {
6144         struct btrfs_device *device;
6145
6146         device = btrfs_alloc_device(NULL, &devid, dev_uuid);
6147         if (IS_ERR(device))
6148                 return NULL;
6149
6150         list_add(&device->dev_list, &fs_devices->devices);
6151         device->fs_devices = fs_devices;
6152         fs_devices->num_devices++;
6153
6154         device->missing = 1;
6155         fs_devices->missing_devices++;
6156
6157         return device;
6158 }
6159
6160 /**
6161  * btrfs_alloc_device - allocate struct btrfs_device
6162  * @fs_info:    used only for generating a new devid, can be NULL if
6163  *              devid is provided (i.e. @devid != NULL).
6164  * @devid:      a pointer to devid for this device.  If NULL a new devid
6165  *              is generated.
6166  * @uuid:       a pointer to UUID for this device.  If NULL a new UUID
6167  *              is generated.
6168  *
6169  * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
6170  * on error.  Returned struct is not linked onto any lists and can be
6171  * destroyed with kfree() right away.
6172  */
6173 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
6174                                         const u64 *devid,
6175                                         const u8 *uuid)
6176 {
6177         struct btrfs_device *dev;
6178         u64 tmp;
6179
6180         if (WARN_ON(!devid && !fs_info))
6181                 return ERR_PTR(-EINVAL);
6182
6183         dev = __alloc_device();
6184         if (IS_ERR(dev))
6185                 return dev;
6186
6187         if (devid)
6188                 tmp = *devid;
6189         else {
6190                 int ret;
6191
6192                 ret = find_next_devid(fs_info, &tmp);
6193                 if (ret) {
6194                         kfree(dev);
6195                         return ERR_PTR(ret);
6196                 }
6197         }
6198         dev->devid = tmp;
6199
6200         if (uuid)
6201                 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE);
6202         else
6203                 generate_random_uuid(dev->uuid);
6204
6205         btrfs_init_work(&dev->work, btrfs_submit_helper,
6206                         pending_bios_fn, NULL, NULL);
6207
6208         return dev;
6209 }
6210
6211 /* Return -EIO if any error, otherwise return 0. */
6212 static int btrfs_check_chunk_valid(struct btrfs_root *root,
6213                                    struct extent_buffer *leaf,
6214                                    struct btrfs_chunk *chunk, u64 logical)
6215 {
6216         u64 length;
6217         u64 stripe_len;
6218         u16 num_stripes;
6219         u16 sub_stripes;
6220         u64 type;
6221         u64 features;
6222         bool mixed = false;
6223
6224         length = btrfs_chunk_length(leaf, chunk);
6225         stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6226         num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6227         sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
6228         type = btrfs_chunk_type(leaf, chunk);
6229
6230         if (!num_stripes) {
6231                 btrfs_err(root->fs_info, "invalid chunk num_stripes: %u",
6232                           num_stripes);
6233                 return -EIO;
6234         }
6235         if (!IS_ALIGNED(logical, root->sectorsize)) {
6236                 btrfs_err(root->fs_info,
6237                           "invalid chunk logical %llu", logical);
6238                 return -EIO;
6239         }
6240         if (btrfs_chunk_sector_size(leaf, chunk) != root->sectorsize) {
6241                 btrfs_err(root->fs_info, "invalid chunk sectorsize %u",
6242                           btrfs_chunk_sector_size(leaf, chunk));
6243                 return -EIO;
6244         }
6245         if (!length || !IS_ALIGNED(length, root->sectorsize)) {
6246                 btrfs_err(root->fs_info,
6247                         "invalid chunk length %llu", length);
6248                 return -EIO;
6249         }
6250         if (!is_power_of_2(stripe_len)) {
6251                 btrfs_err(root->fs_info, "invalid chunk stripe length: %llu",
6252                           stripe_len);
6253                 return -EIO;
6254         }
6255         if (~(BTRFS_BLOCK_GROUP_TYPE_MASK | BTRFS_BLOCK_GROUP_PROFILE_MASK) &
6256             type) {
6257                 btrfs_err(root->fs_info, "unrecognized chunk type: %llu",
6258                           ~(BTRFS_BLOCK_GROUP_TYPE_MASK |
6259                             BTRFS_BLOCK_GROUP_PROFILE_MASK) &
6260                           btrfs_chunk_type(leaf, chunk));
6261                 return -EIO;
6262         }
6263
6264         if ((type & BTRFS_BLOCK_GROUP_TYPE_MASK) == 0) {
6265                 btrfs_err(root->fs_info, "missing chunk type flag: 0x%llx", type);
6266                 return -EIO;
6267         }
6268
6269         if ((type & BTRFS_BLOCK_GROUP_SYSTEM) &&
6270             (type & (BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA))) {
6271                 btrfs_err(root->fs_info,
6272                         "system chunk with data or metadata type: 0x%llx", type);
6273                 return -EIO;
6274         }
6275
6276         features = btrfs_super_incompat_flags(root->fs_info->super_copy);
6277         if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
6278                 mixed = true;
6279
6280         if (!mixed) {
6281                 if ((type & BTRFS_BLOCK_GROUP_METADATA) &&
6282                     (type & BTRFS_BLOCK_GROUP_DATA)) {
6283                         btrfs_err(root->fs_info,
6284                         "mixed chunk type in non-mixed mode: 0x%llx", type);
6285                         return -EIO;
6286                 }
6287         }
6288
6289         if ((type & BTRFS_BLOCK_GROUP_RAID10 && sub_stripes != 2) ||
6290             (type & BTRFS_BLOCK_GROUP_RAID1 && num_stripes != 2) ||
6291             (type & BTRFS_BLOCK_GROUP_RAID5 && num_stripes < 2) ||
6292             (type & BTRFS_BLOCK_GROUP_RAID6 && num_stripes < 3) ||
6293             (type & BTRFS_BLOCK_GROUP_DUP && num_stripes != 2) ||
6294             ((type & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 &&
6295              num_stripes != 1)) {
6296                 btrfs_err(root->fs_info,
6297                         "invalid num_stripes:sub_stripes %u:%u for profile %llu",
6298                         num_stripes, sub_stripes,
6299                         type & BTRFS_BLOCK_GROUP_PROFILE_MASK);
6300                 return -EIO;
6301         }
6302
6303         return 0;
6304 }
6305
6306 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
6307                           struct extent_buffer *leaf,
6308                           struct btrfs_chunk *chunk)
6309 {
6310         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
6311         struct map_lookup *map;
6312         struct extent_map *em;
6313         u64 logical;
6314         u64 length;
6315         u64 stripe_len;
6316         u64 devid;
6317         u8 uuid[BTRFS_UUID_SIZE];
6318         int num_stripes;
6319         int ret;
6320         int i;
6321
6322         logical = key->offset;
6323         length = btrfs_chunk_length(leaf, chunk);
6324         stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6325         num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6326
6327         ret = btrfs_check_chunk_valid(root, leaf, chunk, logical);
6328         if (ret)
6329                 return ret;
6330
6331         read_lock(&map_tree->map_tree.lock);
6332         em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
6333         read_unlock(&map_tree->map_tree.lock);
6334
6335         /* already mapped? */
6336         if (em && em->start <= logical && em->start + em->len > logical) {
6337                 free_extent_map(em);
6338                 return 0;
6339         } else if (em) {
6340                 free_extent_map(em);
6341         }
6342
6343         em = alloc_extent_map();
6344         if (!em)
6345                 return -ENOMEM;
6346         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
6347         if (!map) {
6348                 free_extent_map(em);
6349                 return -ENOMEM;
6350         }
6351
6352         set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
6353         em->map_lookup = map;
6354         em->start = logical;
6355         em->len = length;
6356         em->orig_start = 0;
6357         em->block_start = 0;
6358         em->block_len = em->len;
6359
6360         map->num_stripes = num_stripes;
6361         map->io_width = btrfs_chunk_io_width(leaf, chunk);
6362         map->io_align = btrfs_chunk_io_align(leaf, chunk);
6363         map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
6364         map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6365         map->type = btrfs_chunk_type(leaf, chunk);
6366         map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
6367         for (i = 0; i < num_stripes; i++) {
6368                 map->stripes[i].physical =
6369                         btrfs_stripe_offset_nr(leaf, chunk, i);
6370                 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
6371                 read_extent_buffer(leaf, uuid, (unsigned long)
6372                                    btrfs_stripe_dev_uuid_nr(chunk, i),
6373                                    BTRFS_UUID_SIZE);
6374                 map->stripes[i].dev = btrfs_find_device(root->fs_info, devid,
6375                                                         uuid, NULL);
6376                 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
6377                         free_extent_map(em);
6378                         return -EIO;
6379                 }
6380                 if (!map->stripes[i].dev) {
6381                         map->stripes[i].dev =
6382                                 add_missing_dev(root, root->fs_info->fs_devices,
6383                                                 devid, uuid);
6384                         if (!map->stripes[i].dev) {
6385                                 free_extent_map(em);
6386                                 return -EIO;
6387                         }
6388                         btrfs_warn(root->fs_info, "devid %llu uuid %pU is missing",
6389                                                 devid, uuid);
6390                 }
6391                 map->stripes[i].dev->in_fs_metadata = 1;
6392         }
6393
6394         write_lock(&map_tree->map_tree.lock);
6395         ret = add_extent_mapping(&map_tree->map_tree, em, 0);
6396         write_unlock(&map_tree->map_tree.lock);
6397         BUG_ON(ret); /* Tree corruption */
6398         free_extent_map(em);
6399
6400         return 0;
6401 }
6402
6403 static void fill_device_from_item(struct extent_buffer *leaf,
6404                                  struct btrfs_dev_item *dev_item,
6405                                  struct btrfs_device *device)
6406 {
6407         unsigned long ptr;
6408
6409         device->devid = btrfs_device_id(leaf, dev_item);
6410         device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
6411         device->total_bytes = device->disk_total_bytes;
6412         device->commit_total_bytes = device->disk_total_bytes;
6413         device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
6414         device->commit_bytes_used = device->bytes_used;
6415         device->type = btrfs_device_type(leaf, dev_item);
6416         device->io_align = btrfs_device_io_align(leaf, dev_item);
6417         device->io_width = btrfs_device_io_width(leaf, dev_item);
6418         device->sector_size = btrfs_device_sector_size(leaf, dev_item);
6419         WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
6420         device->is_tgtdev_for_dev_replace = 0;
6421
6422         ptr = btrfs_device_uuid(dev_item);
6423         read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
6424 }
6425
6426 static struct btrfs_fs_devices *open_seed_devices(struct btrfs_root *root,
6427                                                   u8 *fsid)
6428 {
6429         struct btrfs_fs_devices *fs_devices;
6430         int ret;
6431
6432         BUG_ON(!mutex_is_locked(&uuid_mutex));
6433
6434         fs_devices = root->fs_info->fs_devices->seed;
6435         while (fs_devices) {
6436                 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE))
6437                         return fs_devices;
6438
6439                 fs_devices = fs_devices->seed;
6440         }
6441
6442         fs_devices = find_fsid(fsid);
6443         if (!fs_devices) {
6444                 if (!btrfs_test_opt(root, DEGRADED))
6445                         return ERR_PTR(-ENOENT);
6446
6447                 fs_devices = alloc_fs_devices(fsid);
6448                 if (IS_ERR(fs_devices))
6449                         return fs_devices;
6450
6451                 fs_devices->seeding = 1;
6452                 fs_devices->opened = 1;
6453                 return fs_devices;
6454         }
6455
6456         fs_devices = clone_fs_devices(fs_devices);
6457         if (IS_ERR(fs_devices))
6458                 return fs_devices;
6459
6460         ret = __btrfs_open_devices(fs_devices, FMODE_READ,
6461                                    root->fs_info->bdev_holder);
6462         if (ret) {
6463                 free_fs_devices(fs_devices);
6464                 fs_devices = ERR_PTR(ret);
6465                 goto out;
6466         }
6467
6468         if (!fs_devices->seeding) {
6469                 __btrfs_close_devices(fs_devices);
6470                 free_fs_devices(fs_devices);
6471                 fs_devices = ERR_PTR(-EINVAL);
6472                 goto out;
6473         }
6474
6475         fs_devices->seed = root->fs_info->fs_devices->seed;
6476         root->fs_info->fs_devices->seed = fs_devices;
6477 out:
6478         return fs_devices;
6479 }
6480
6481 static int read_one_dev(struct btrfs_root *root,
6482                         struct extent_buffer *leaf,
6483                         struct btrfs_dev_item *dev_item)
6484 {
6485         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
6486         struct btrfs_device *device;
6487         u64 devid;
6488         int ret;
6489         u8 fs_uuid[BTRFS_UUID_SIZE];
6490         u8 dev_uuid[BTRFS_UUID_SIZE];
6491
6492         devid = btrfs_device_id(leaf, dev_item);
6493         read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
6494                            BTRFS_UUID_SIZE);
6495         read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
6496                            BTRFS_UUID_SIZE);
6497
6498         if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
6499                 fs_devices = open_seed_devices(root, fs_uuid);
6500                 if (IS_ERR(fs_devices))
6501                         return PTR_ERR(fs_devices);
6502         }
6503
6504         device = btrfs_find_device(root->fs_info, devid, dev_uuid, fs_uuid);
6505         if (!device) {
6506                 if (!btrfs_test_opt(root, DEGRADED))
6507                         return -EIO;
6508
6509                 device = add_missing_dev(root, fs_devices, devid, dev_uuid);
6510                 if (!device)
6511                         return -ENOMEM;
6512                 btrfs_warn(root->fs_info, "devid %llu uuid %pU missing",
6513                                 devid, dev_uuid);
6514         } else {
6515                 if (!device->bdev && !btrfs_test_opt(root, DEGRADED))
6516                         return -EIO;
6517
6518                 if(!device->bdev && !device->missing) {
6519                         /*
6520                          * this happens when a device that was properly setup
6521                          * in the device info lists suddenly goes bad.
6522                          * device->bdev is NULL, and so we have to set
6523                          * device->missing to one here
6524                          */
6525                         device->fs_devices->missing_devices++;
6526                         device->missing = 1;
6527                 }
6528
6529                 /* Move the device to its own fs_devices */
6530                 if (device->fs_devices != fs_devices) {
6531                         ASSERT(device->missing);
6532
6533                         list_move(&device->dev_list, &fs_devices->devices);
6534                         device->fs_devices->num_devices--;
6535                         fs_devices->num_devices++;
6536
6537                         device->fs_devices->missing_devices--;
6538                         fs_devices->missing_devices++;
6539
6540                         device->fs_devices = fs_devices;
6541                 }
6542         }
6543
6544         if (device->fs_devices != root->fs_info->fs_devices) {
6545                 BUG_ON(device->writeable);
6546                 if (device->generation !=
6547                     btrfs_device_generation(leaf, dev_item))
6548                         return -EINVAL;
6549         }
6550
6551         fill_device_from_item(leaf, dev_item, device);
6552         device->in_fs_metadata = 1;
6553         if (device->writeable && !device->is_tgtdev_for_dev_replace) {
6554                 device->fs_devices->total_rw_bytes += device->total_bytes;
6555                 spin_lock(&root->fs_info->free_chunk_lock);
6556                 root->fs_info->free_chunk_space += device->total_bytes -
6557                         device->bytes_used;
6558                 spin_unlock(&root->fs_info->free_chunk_lock);
6559         }
6560         ret = 0;
6561         return ret;
6562 }
6563
6564 int btrfs_read_sys_array(struct btrfs_root *root)
6565 {
6566         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
6567         struct extent_buffer *sb;
6568         struct btrfs_disk_key *disk_key;
6569         struct btrfs_chunk *chunk;
6570         u8 *array_ptr;
6571         unsigned long sb_array_offset;
6572         int ret = 0;
6573         u32 num_stripes;
6574         u32 array_size;
6575         u32 len = 0;
6576         u32 cur_offset;
6577         u64 type;
6578         struct btrfs_key key;
6579
6580         ASSERT(BTRFS_SUPER_INFO_SIZE <= root->nodesize);
6581         /*
6582          * This will create extent buffer of nodesize, superblock size is
6583          * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will
6584          * overallocate but we can keep it as-is, only the first page is used.
6585          */
6586         sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET);
6587         if (!sb)
6588                 return -ENOMEM;
6589         btrfs_set_buffer_uptodate(sb);
6590         btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
6591         /*
6592          * The sb extent buffer is artifical and just used to read the system array.
6593          * btrfs_set_buffer_uptodate() call does not properly mark all it's
6594          * pages up-to-date when the page is larger: extent does not cover the
6595          * whole page and consequently check_page_uptodate does not find all
6596          * the page's extents up-to-date (the hole beyond sb),
6597          * write_extent_buffer then triggers a WARN_ON.
6598          *
6599          * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
6600          * but sb spans only this function. Add an explicit SetPageUptodate call
6601          * to silence the warning eg. on PowerPC 64.
6602          */
6603         if (PAGE_CACHE_SIZE > BTRFS_SUPER_INFO_SIZE)
6604                 SetPageUptodate(sb->pages[0]);
6605
6606         write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
6607         array_size = btrfs_super_sys_array_size(super_copy);
6608
6609         array_ptr = super_copy->sys_chunk_array;
6610         sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array);
6611         cur_offset = 0;
6612
6613         while (cur_offset < array_size) {
6614                 disk_key = (struct btrfs_disk_key *)array_ptr;
6615                 len = sizeof(*disk_key);
6616                 if (cur_offset + len > array_size)
6617                         goto out_short_read;
6618
6619                 btrfs_disk_key_to_cpu(&key, disk_key);
6620
6621                 array_ptr += len;
6622                 sb_array_offset += len;
6623                 cur_offset += len;
6624
6625                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
6626                         chunk = (struct btrfs_chunk *)sb_array_offset;
6627                         /*
6628                          * At least one btrfs_chunk with one stripe must be
6629                          * present, exact stripe count check comes afterwards
6630                          */
6631                         len = btrfs_chunk_item_size(1);
6632                         if (cur_offset + len > array_size)
6633                                 goto out_short_read;
6634
6635                         num_stripes = btrfs_chunk_num_stripes(sb, chunk);
6636                         if (!num_stripes) {
6637                                 printk(KERN_ERR
6638             "BTRFS: invalid number of stripes %u in sys_array at offset %u\n",
6639                                         num_stripes, cur_offset);
6640                                 ret = -EIO;
6641                                 break;
6642                         }
6643
6644                         type = btrfs_chunk_type(sb, chunk);
6645                         if ((type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) {
6646                                 btrfs_err(root->fs_info,
6647                             "invalid chunk type %llu in sys_array at offset %u",
6648                                         type, cur_offset);
6649                                 ret = -EIO;
6650                                 break;
6651                         }
6652
6653                         len = btrfs_chunk_item_size(num_stripes);
6654                         if (cur_offset + len > array_size)
6655                                 goto out_short_read;
6656
6657                         ret = read_one_chunk(root, &key, sb, chunk);
6658                         if (ret)
6659                                 break;
6660                 } else {
6661                         ret = -EIO;
6662                         break;
6663                 }
6664                 array_ptr += len;
6665                 sb_array_offset += len;
6666                 cur_offset += len;
6667         }
6668         free_extent_buffer(sb);
6669         return ret;
6670
6671 out_short_read:
6672         printk(KERN_ERR "BTRFS: sys_array too short to read %u bytes at offset %u\n",
6673                         len, cur_offset);
6674         free_extent_buffer(sb);
6675         return -EIO;
6676 }
6677
6678 int btrfs_read_chunk_tree(struct btrfs_root *root)
6679 {
6680         struct btrfs_path *path;
6681         struct extent_buffer *leaf;
6682         struct btrfs_key key;
6683         struct btrfs_key found_key;
6684         int ret;
6685         int slot;
6686
6687         root = root->fs_info->chunk_root;
6688
6689         path = btrfs_alloc_path();
6690         if (!path)
6691                 return -ENOMEM;
6692
6693         mutex_lock(&uuid_mutex);
6694         lock_chunks(root);
6695
6696         /*
6697          * Read all device items, and then all the chunk items. All
6698          * device items are found before any chunk item (their object id
6699          * is smaller than the lowest possible object id for a chunk
6700          * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
6701          */
6702         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
6703         key.offset = 0;
6704         key.type = 0;
6705         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6706         if (ret < 0)
6707                 goto error;
6708         while (1) {
6709                 leaf = path->nodes[0];
6710                 slot = path->slots[0];
6711                 if (slot >= btrfs_header_nritems(leaf)) {
6712                         ret = btrfs_next_leaf(root, path);
6713                         if (ret == 0)
6714                                 continue;
6715                         if (ret < 0)
6716                                 goto error;
6717                         break;
6718                 }
6719                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6720                 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
6721                         struct btrfs_dev_item *dev_item;
6722                         dev_item = btrfs_item_ptr(leaf, slot,
6723                                                   struct btrfs_dev_item);
6724                         ret = read_one_dev(root, leaf, dev_item);
6725                         if (ret)
6726                                 goto error;
6727                 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
6728                         struct btrfs_chunk *chunk;
6729                         chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
6730                         ret = read_one_chunk(root, &found_key, leaf, chunk);
6731                         if (ret)
6732                                 goto error;
6733                 }
6734                 path->slots[0]++;
6735         }
6736         ret = 0;
6737 error:
6738         unlock_chunks(root);
6739         mutex_unlock(&uuid_mutex);
6740
6741         btrfs_free_path(path);
6742         return ret;
6743 }
6744
6745 void btrfs_init_devices_late(struct btrfs_fs_info *fs_info)
6746 {
6747         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6748         struct btrfs_device *device;
6749
6750         while (fs_devices) {
6751                 mutex_lock(&fs_devices->device_list_mutex);
6752                 list_for_each_entry(device, &fs_devices->devices, dev_list)
6753                         device->dev_root = fs_info->dev_root;
6754                 mutex_unlock(&fs_devices->device_list_mutex);
6755
6756                 fs_devices = fs_devices->seed;
6757         }
6758 }
6759
6760 static void __btrfs_reset_dev_stats(struct btrfs_device *dev)
6761 {
6762         int i;
6763
6764         for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6765                 btrfs_dev_stat_reset(dev, i);
6766 }
6767
6768 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
6769 {
6770         struct btrfs_key key;
6771         struct btrfs_key found_key;
6772         struct btrfs_root *dev_root = fs_info->dev_root;
6773         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6774         struct extent_buffer *eb;
6775         int slot;
6776         int ret = 0;
6777         struct btrfs_device *device;
6778         struct btrfs_path *path = NULL;
6779         int i;
6780
6781         path = btrfs_alloc_path();
6782         if (!path) {
6783                 ret = -ENOMEM;
6784                 goto out;
6785         }
6786
6787         mutex_lock(&fs_devices->device_list_mutex);
6788         list_for_each_entry(device, &fs_devices->devices, dev_list) {
6789                 int item_size;
6790                 struct btrfs_dev_stats_item *ptr;
6791
6792                 key.objectid = 0;
6793                 key.type = BTRFS_DEV_STATS_KEY;
6794                 key.offset = device->devid;
6795                 ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
6796                 if (ret) {
6797                         __btrfs_reset_dev_stats(device);
6798                         device->dev_stats_valid = 1;
6799                         btrfs_release_path(path);
6800                         continue;
6801                 }
6802                 slot = path->slots[0];
6803                 eb = path->nodes[0];
6804                 btrfs_item_key_to_cpu(eb, &found_key, slot);
6805                 item_size = btrfs_item_size_nr(eb, slot);
6806
6807                 ptr = btrfs_item_ptr(eb, slot,
6808                                      struct btrfs_dev_stats_item);
6809
6810                 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
6811                         if (item_size >= (1 + i) * sizeof(__le64))
6812                                 btrfs_dev_stat_set(device, i,
6813                                         btrfs_dev_stats_value(eb, ptr, i));
6814                         else
6815                                 btrfs_dev_stat_reset(device, i);
6816                 }
6817
6818                 device->dev_stats_valid = 1;
6819                 btrfs_dev_stat_print_on_load(device);
6820                 btrfs_release_path(path);
6821         }
6822         mutex_unlock(&fs_devices->device_list_mutex);
6823
6824 out:
6825         btrfs_free_path(path);
6826         return ret < 0 ? ret : 0;
6827 }
6828
6829 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
6830                                 struct btrfs_root *dev_root,
6831                                 struct btrfs_device *device)
6832 {
6833         struct btrfs_path *path;
6834         struct btrfs_key key;
6835         struct extent_buffer *eb;
6836         struct btrfs_dev_stats_item *ptr;
6837         int ret;
6838         int i;
6839
6840         key.objectid = 0;
6841         key.type = BTRFS_DEV_STATS_KEY;
6842         key.offset = device->devid;
6843
6844         path = btrfs_alloc_path();
6845         BUG_ON(!path);
6846         ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
6847         if (ret < 0) {
6848                 btrfs_warn_in_rcu(dev_root->fs_info,
6849                         "error %d while searching for dev_stats item for device %s",
6850                               ret, rcu_str_deref(device->name));
6851                 goto out;
6852         }
6853
6854         if (ret == 0 &&
6855             btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
6856                 /* need to delete old one and insert a new one */
6857                 ret = btrfs_del_item(trans, dev_root, path);
6858                 if (ret != 0) {
6859                         btrfs_warn_in_rcu(dev_root->fs_info,
6860                                 "delete too small dev_stats item for device %s failed %d",
6861                                       rcu_str_deref(device->name), ret);
6862                         goto out;
6863                 }
6864                 ret = 1;
6865         }
6866
6867         if (ret == 1) {
6868                 /* need to insert a new item */
6869                 btrfs_release_path(path);
6870                 ret = btrfs_insert_empty_item(trans, dev_root, path,
6871                                               &key, sizeof(*ptr));
6872                 if (ret < 0) {
6873                         btrfs_warn_in_rcu(dev_root->fs_info,
6874                                 "insert dev_stats item for device %s failed %d",
6875                                 rcu_str_deref(device->name), ret);
6876                         goto out;
6877                 }
6878         }
6879
6880         eb = path->nodes[0];
6881         ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
6882         for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6883                 btrfs_set_dev_stats_value(eb, ptr, i,
6884                                           btrfs_dev_stat_read(device, i));
6885         btrfs_mark_buffer_dirty(eb);
6886
6887 out:
6888         btrfs_free_path(path);
6889         return ret;
6890 }
6891
6892 /*
6893  * called from commit_transaction. Writes all changed device stats to disk.
6894  */
6895 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans,
6896                         struct btrfs_fs_info *fs_info)
6897 {
6898         struct btrfs_root *dev_root = fs_info->dev_root;
6899         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6900         struct btrfs_device *device;
6901         int stats_cnt;
6902         int ret = 0;
6903
6904         mutex_lock(&fs_devices->device_list_mutex);
6905         list_for_each_entry(device, &fs_devices->devices, dev_list) {
6906                 if (!device->dev_stats_valid || !btrfs_dev_stats_dirty(device))
6907                         continue;
6908
6909                 stats_cnt = atomic_read(&device->dev_stats_ccnt);
6910                 ret = update_dev_stat_item(trans, dev_root, device);
6911                 if (!ret)
6912                         atomic_sub(stats_cnt, &device->dev_stats_ccnt);
6913         }
6914         mutex_unlock(&fs_devices->device_list_mutex);
6915
6916         return ret;
6917 }
6918
6919 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
6920 {
6921         btrfs_dev_stat_inc(dev, index);
6922         btrfs_dev_stat_print_on_error(dev);
6923 }
6924
6925 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
6926 {
6927         if (!dev->dev_stats_valid)
6928                 return;
6929         btrfs_err_rl_in_rcu(dev->dev_root->fs_info,
6930                 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
6931                            rcu_str_deref(dev->name),
6932                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
6933                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
6934                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
6935                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
6936                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
6937 }
6938
6939 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
6940 {
6941         int i;
6942
6943         for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6944                 if (btrfs_dev_stat_read(dev, i) != 0)
6945                         break;
6946         if (i == BTRFS_DEV_STAT_VALUES_MAX)
6947                 return; /* all values == 0, suppress message */
6948
6949         btrfs_info_in_rcu(dev->dev_root->fs_info,
6950                 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
6951                rcu_str_deref(dev->name),
6952                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
6953                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
6954                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
6955                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
6956                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
6957 }
6958
6959 int btrfs_get_dev_stats(struct btrfs_root *root,
6960                         struct btrfs_ioctl_get_dev_stats *stats)
6961 {
6962         struct btrfs_device *dev;
6963         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
6964         int i;
6965
6966         mutex_lock(&fs_devices->device_list_mutex);
6967         dev = btrfs_find_device(root->fs_info, stats->devid, NULL, NULL);
6968         mutex_unlock(&fs_devices->device_list_mutex);
6969
6970         if (!dev) {
6971                 btrfs_warn(root->fs_info, "get dev_stats failed, device not found");
6972                 return -ENODEV;
6973         } else if (!dev->dev_stats_valid) {
6974                 btrfs_warn(root->fs_info, "get dev_stats failed, not yet valid");
6975                 return -ENODEV;
6976         } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
6977                 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
6978                         if (stats->nr_items > i)
6979                                 stats->values[i] =
6980                                         btrfs_dev_stat_read_and_reset(dev, i);
6981                         else
6982                                 btrfs_dev_stat_reset(dev, i);
6983                 }
6984         } else {
6985                 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6986                         if (stats->nr_items > i)
6987                                 stats->values[i] = btrfs_dev_stat_read(dev, i);
6988         }
6989         if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
6990                 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
6991         return 0;
6992 }
6993
6994 void btrfs_scratch_superblocks(struct block_device *bdev, char *device_path)
6995 {
6996         struct buffer_head *bh;
6997         struct btrfs_super_block *disk_super;
6998         int copy_num;
6999
7000         if (!bdev)
7001                 return;
7002
7003         for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX;
7004                 copy_num++) {
7005
7006                 if (btrfs_read_dev_one_super(bdev, copy_num, &bh))
7007                         continue;
7008
7009                 disk_super = (struct btrfs_super_block *)bh->b_data;
7010
7011                 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
7012                 set_buffer_dirty(bh);
7013                 sync_dirty_buffer(bh);
7014                 brelse(bh);
7015         }
7016
7017         /* Notify udev that device has changed */
7018         btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
7019
7020         /* Update ctime/mtime for device path for libblkid */
7021         update_dev_time(device_path);
7022 }
7023
7024 /*
7025  * Update the size of all devices, which is used for writing out the
7026  * super blocks.
7027  */
7028 void btrfs_update_commit_device_size(struct btrfs_fs_info *fs_info)
7029 {
7030         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7031         struct btrfs_device *curr, *next;
7032
7033         if (list_empty(&fs_devices->resized_devices))
7034                 return;
7035
7036         mutex_lock(&fs_devices->device_list_mutex);
7037         lock_chunks(fs_info->dev_root);
7038         list_for_each_entry_safe(curr, next, &fs_devices->resized_devices,
7039                                  resized_list) {
7040                 list_del_init(&curr->resized_list);
7041                 curr->commit_total_bytes = curr->disk_total_bytes;
7042         }
7043         unlock_chunks(fs_info->dev_root);
7044         mutex_unlock(&fs_devices->device_list_mutex);
7045 }
7046
7047 /* Must be invoked during the transaction commit */
7048 void btrfs_update_commit_device_bytes_used(struct btrfs_root *root,
7049                                         struct btrfs_transaction *transaction)
7050 {
7051         struct extent_map *em;
7052         struct map_lookup *map;
7053         struct btrfs_device *dev;
7054         int i;
7055
7056         if (list_empty(&transaction->pending_chunks))
7057                 return;
7058
7059         /* In order to kick the device replace finish process */
7060         lock_chunks(root);
7061         list_for_each_entry(em, &transaction->pending_chunks, list) {
7062                 map = em->map_lookup;
7063
7064                 for (i = 0; i < map->num_stripes; i++) {
7065                         dev = map->stripes[i].dev;
7066                         dev->commit_bytes_used = dev->bytes_used;
7067                         dev->has_pending_chunks = false;
7068                 }
7069         }
7070         unlock_chunks(root);
7071 }
7072
7073 void btrfs_set_fs_info_ptr(struct btrfs_fs_info *fs_info)
7074 {
7075         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7076         while (fs_devices) {
7077                 fs_devices->fs_info = fs_info;
7078                 fs_devices = fs_devices->seed;
7079         }
7080 }
7081
7082 void btrfs_reset_fs_info_ptr(struct btrfs_fs_info *fs_info)
7083 {
7084         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7085         while (fs_devices) {
7086                 fs_devices->fs_info = NULL;
7087                 fs_devices = fs_devices->seed;
7088         }
7089 }
7090
7091 void btrfs_close_one_device(struct btrfs_device *device)
7092 {
7093         struct btrfs_fs_devices *fs_devices = device->fs_devices;
7094         struct btrfs_device *new_device;
7095         struct rcu_string *name;
7096
7097         if (device->bdev)
7098                 fs_devices->open_devices--;
7099
7100         if (device->writeable &&
7101             device->devid != BTRFS_DEV_REPLACE_DEVID) {
7102                 list_del_init(&device->dev_alloc_list);
7103                 fs_devices->rw_devices--;
7104         }
7105
7106         if (device->missing)
7107                 fs_devices->missing_devices--;
7108
7109         new_device = btrfs_alloc_device(NULL, &device->devid,
7110                                         device->uuid);
7111         BUG_ON(IS_ERR(new_device)); /* -ENOMEM */
7112
7113         /* Safe because we are under uuid_mutex */
7114         if (device->name) {
7115                 name = rcu_string_strdup(device->name->str, GFP_NOFS);
7116                 BUG_ON(!name); /* -ENOMEM */
7117                 rcu_assign_pointer(new_device->name, name);
7118         }
7119
7120         list_replace_rcu(&device->dev_list, &new_device->dev_list);
7121         new_device->fs_devices = device->fs_devices;
7122
7123         call_rcu(&device->rcu, free_device);
7124 }