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