fs/btrfs/volumes.c

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