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