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 <asm/div64.h>
  30 #include "compat.h"
  31 #include "ctree.h"
  32 #include "extent_map.h"
  33 #include "disk-io.h"
  34 #include "transaction.h"
  35 #include "print-tree.h"
  36 #include "volumes.h"
  37 #include "raid56.h"
  38 #include "async-thread.h"
  39 #include "check-integrity.h"
  40 #include "rcu-string.h"
  41 #include "math.h"
  42 #include "dev-replace.h"
  43
  44 static int init_first_rw_device(struct btrfs_trans_handle *trans,
  45                                 struct btrfs_root *root,
  46                                 struct btrfs_device *device);
  47 static int btrfs_relocate_sys_chunks(struct btrfs_root *root);
  48 static void __btrfs_reset_dev_stats(struct btrfs_device *dev);
  49 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev);
  50 static void btrfs_dev_stat_print_on_load(struct btrfs_device *device);
  51
  52 static DEFINE_MUTEX(uuid_mutex);
  53 static LIST_HEAD(fs_uuids);
  54
  55 static void lock_chunks(struct btrfs_root *root)
  56 {
  57         mutex_lock(&root->fs_info->chunk_mutex);
  58 }
  59
  60 static void unlock_chunks(struct btrfs_root *root)
  61 {
  62         mutex_unlock(&root->fs_info->chunk_mutex);
  63 }
  64
  65 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
  66 {
  67         struct btrfs_device *device;
  68         WARN_ON(fs_devices->opened);
  69         while (!list_empty(&fs_devices->devices)) {
  70                 device = list_entry(fs_devices->devices.next,
  71                                     struct btrfs_device, dev_list);
  72                 list_del(&device->dev_list);
  73                 rcu_string_free(device->name);
  74                 kfree(device);
  75         }
  76         kfree(fs_devices);
  77 }
  78
  79 static void btrfs_kobject_uevent(struct block_device *bdev,
  80                                  enum kobject_action action)
  81 {
  82         int ret;
  83
  84         ret = kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, action);
  85         if (ret)
  86                 pr_warn("Sending event '%d' to kobject: '%s' (%p): failed\n",
  87                         action,
  88                         kobject_name(&disk_to_dev(bdev->bd_disk)->kobj),
  89                         &disk_to_dev(bdev->bd_disk)->kobj);
  90 }
  91
  92 void btrfs_cleanup_fs_uuids(void)
  93 {
  94         struct btrfs_fs_devices *fs_devices;
  95
  96         while (!list_empty(&fs_uuids)) {
  97                 fs_devices = list_entry(fs_uuids.next,
  98                                         struct btrfs_fs_devices, list);
  99                 list_del(&fs_devices->list);
 100                 free_fs_devices(fs_devices);
 101         }
 102 }
 103
 104 static noinline struct btrfs_device *__find_device(struct list_head *head,
 105                                                    u64 devid, u8 *uuid)
 106 {
 107         struct btrfs_device *dev;
 108
 109         list_for_each_entry(dev, head, dev_list) {
 110                 if (dev->devid == devid &&
 111                     (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
 112                         return dev;
 113                 }
 114         }
 115         return NULL;
 116 }
 117
 118 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
 119 {
 120         struct btrfs_fs_devices *fs_devices;
 121
 122         list_for_each_entry(fs_devices, &fs_uuids, list) {
 123                 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
 124                         return fs_devices;
 125         }
 126         return NULL;
 127 }
 128
 129 static int
 130 btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder,
 131                       int flush, struct block_device **bdev,
 132                       struct buffer_head **bh)
 133 {
 134         int ret;
 135
 136         *bdev = blkdev_get_by_path(device_path, flags, holder);
 137
 138         if (IS_ERR(*bdev)) {
 139                 ret = PTR_ERR(*bdev);
 140                 printk(KERN_INFO "btrfs: open %s failed\n", device_path);
 141                 goto error;
 142         }
 143
 144         if (flush)
 145                 filemap_write_and_wait((*bdev)->bd_inode->i_mapping);
 146         ret = set_blocksize(*bdev, 4096);
 147         if (ret) {
 148                 blkdev_put(*bdev, flags);
 149                 goto error;
 150         }
 151         invalidate_bdev(*bdev);
 152         *bh = btrfs_read_dev_super(*bdev);
 153         if (!*bh) {
 154                 ret = -EINVAL;
 155                 blkdev_put(*bdev, flags);
 156                 goto error;
 157         }
 158
 159         return 0;
 160
 161 error:
 162         *bdev = NULL;
 163         *bh = NULL;
 164         return ret;
 165 }
 166
 167 static void requeue_list(struct btrfs_pending_bios *pending_bios,
 168                         struct bio *head, struct bio *tail)
 169 {
 170
 171         struct bio *old_head;
 172
 173         old_head = pending_bios->head;
 174         pending_bios->head = head;
 175         if (pending_bios->tail)
 176                 tail->bi_next = old_head;
 177         else
 178                 pending_bios->tail = tail;
 179 }
 180
 181 /*
 182  * we try to collect pending bios for a device so we don't get a large
 183  * number of procs sending bios down to the same device.  This greatly
 184  * improves the schedulers ability to collect and merge the bios.
 185  *
 186  * But, it also turns into a long list of bios to process and that is sure
 187  * to eventually make the worker thread block.  The solution here is to
 188  * make some progress and then put this work struct back at the end of
 189  * the list if the block device is congested.  This way, multiple devices
 190  * can make progress from a single worker thread.
 191  */
 192 static noinline void run_scheduled_bios(struct btrfs_device *device)
 193 {
 194         struct bio *pending;
 195         struct backing_dev_info *bdi;
 196         struct btrfs_fs_info *fs_info;
 197         struct btrfs_pending_bios *pending_bios;
 198         struct bio *tail;
 199         struct bio *cur;
 200         int again = 0;
 201         unsigned long num_run;
 202         unsigned long batch_run = 0;
 203         unsigned long limit;
 204         unsigned long last_waited = 0;
 205         int force_reg = 0;
 206         int sync_pending = 0;
 207         struct blk_plug plug;
 208
 209         /*
 210          * this function runs all the bios we've collected for
 211          * a particular device.  We don't want to wander off to
 212          * another device without first sending all of these down.
 213          * So, setup a plug here and finish it off before we return
 214          */
 215         blk_start_plug(&plug);
 216
 217         bdi = blk_get_backing_dev_info(device->bdev);
 218         fs_info = device->dev_root->fs_info;
 219         limit = btrfs_async_submit_limit(fs_info);
 220         limit = limit * 2 / 3;
 221
 222 loop:
 223         spin_lock(&device->io_lock);
 224
 225 loop_lock:
 226         num_run = 0;
 227
 228         /* take all the bios off the list at once and process them
 229          * later on (without the lock held).  But, remember the
 230          * tail and other pointers so the bios can be properly reinserted
 231          * into the list if we hit congestion
 232          */
 233         if (!force_reg && device->pending_sync_bios.head) {
 234                 pending_bios = &device->pending_sync_bios;
 235                 force_reg = 1;
 236         } else {
 237                 pending_bios = &device->pending_bios;
 238                 force_reg = 0;
 239         }
 240
 241         pending = pending_bios->head;
 242         tail = pending_bios->tail;
 243         WARN_ON(pending && !tail);
 244
 245         /*
 246          * if pending was null this time around, no bios need processing
 247          * at all and we can stop.  Otherwise it'll loop back up again
 248          * and do an additional check so no bios are missed.
 249          *
 250          * device->running_pending is used to synchronize with the
 251          * schedule_bio code.
 252          */
 253         if (device->pending_sync_bios.head == NULL &&
 254             device->pending_bios.head == NULL) {
 255                 again = 0;
 256                 device->running_pending = 0;
 257         } else {
 258                 again = 1;
 259                 device->running_pending = 1;
 260         }
 261
 262         pending_bios->head = NULL;
 263         pending_bios->tail = NULL;
 264
 265         spin_unlock(&device->io_lock);
 266
 267         while (pending) {
 268
 269                 rmb();
 270                 /* we want to work on both lists, but do more bios on the
 271                  * sync list than the regular list
 272                  */
 273                 if ((num_run > 32 &&
 274                     pending_bios != &device->pending_sync_bios &&
 275                     device->pending_sync_bios.head) ||
 276                    (num_run > 64 && pending_bios == &device->pending_sync_bios &&
 277                     device->pending_bios.head)) {
 278                         spin_lock(&device->io_lock);
 279                         requeue_list(pending_bios, pending, tail);
 280                         goto loop_lock;
 281                 }
 282
 283                 cur = pending;
 284                 pending = pending->bi_next;
 285                 cur->bi_next = NULL;
 286
 287                 if (atomic_dec_return(&fs_info->nr_async_bios) < limit &&
 288                     waitqueue_active(&fs_info->async_submit_wait))
 289                         wake_up(&fs_info->async_submit_wait);
 290
 291                 BUG_ON(atomic_read(&cur->bi_cnt) == 0);
 292
 293                 /*
 294                  * if we're doing the sync list, record that our
 295                  * plug has some sync requests on it
 296                  *
 297                  * If we're doing the regular list and there are
 298                  * sync requests sitting around, unplug before
 299                  * we add more
 300                  */
 301                 if (pending_bios == &device->pending_sync_bios) {
 302                         sync_pending = 1;
 303                 } else if (sync_pending) {
 304                         blk_finish_plug(&plug);
 305                         blk_start_plug(&plug);
 306                         sync_pending = 0;
 307                 }
 308
 309                 btrfsic_submit_bio(cur->bi_rw, cur);
 310                 num_run++;
 311                 batch_run++;
 312                 if (need_resched())
 313                         cond_resched();
 314
 315                 /*
 316                  * we made progress, there is more work to do and the bdi
 317                  * is now congested.  Back off and let other work structs
 318                  * run instead
 319                  */
 320                 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
 321                     fs_info->fs_devices->open_devices > 1) {
 322                         struct io_context *ioc;
 323
 324                         ioc = current->io_context;
 325
 326                         /*
 327                          * the main goal here is that we don't want to
 328                          * block if we're going to be able to submit
 329                          * more requests without blocking.
 330                          *
 331                          * This code does two great things, it pokes into
 332                          * the elevator code from a filesystem _and_
 333                          * it makes assumptions about how batching works.
 334                          */
 335                         if (ioc && ioc->nr_batch_requests > 0 &&
 336                             time_before(jiffies, ioc->last_waited + HZ/50UL) &&
 337                             (last_waited == 0 ||
 338                              ioc->last_waited == last_waited)) {
 339                                 /*
 340                                  * we want to go through our batch of
 341                                  * requests and stop.  So, we copy out
 342                                  * the ioc->last_waited time and test
 343                                  * against it before looping
 344                                  */
 345                                 last_waited = ioc->last_waited;
 346                                 if (need_resched())
 347                                         cond_resched();
 348                                 continue;
 349                         }
 350                         spin_lock(&device->io_lock);
 351                         requeue_list(pending_bios, pending, tail);
 352                         device->running_pending = 1;
 353
 354                         spin_unlock(&device->io_lock);
 355                         btrfs_requeue_work(&device->work);
 356                         goto done;
 357                 }
 358                 /* unplug every 64 requests just for good measure */
 359                 if (batch_run % 64 == 0) {
 360                         blk_finish_plug(&plug);
 361                         blk_start_plug(&plug);
 362                         sync_pending = 0;
 363                 }
 364         }
 365
 366         cond_resched();
 367         if (again)
 368                 goto loop;
 369
 370         spin_lock(&device->io_lock);
 371         if (device->pending_bios.head || device->pending_sync_bios.head)
 372                 goto loop_lock;
 373         spin_unlock(&device->io_lock);
 374
 375 done:
 376         blk_finish_plug(&plug);
 377 }
 378
 379 static void pending_bios_fn(struct btrfs_work *work)
 380 {
 381         struct btrfs_device *device;
 382
 383         device = container_of(work, struct btrfs_device, work);
 384         run_scheduled_bios(device);
 385 }
 386
 387 static noinline int device_list_add(const char *path,
 388                            struct btrfs_super_block *disk_super,
 389                            u64 devid, struct btrfs_fs_devices **fs_devices_ret)
 390 {
 391         struct btrfs_device *device;
 392         struct btrfs_fs_devices *fs_devices;
 393         struct rcu_string *name;
 394         u64 found_transid = btrfs_super_generation(disk_super);
 395
 396         fs_devices = find_fsid(disk_super->fsid);
 397         if (!fs_devices) {
 398                 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
 399                 if (!fs_devices)
 400                         return -ENOMEM;
 401                 INIT_LIST_HEAD(&fs_devices->devices);
 402                 INIT_LIST_HEAD(&fs_devices->alloc_list);
 403                 list_add(&fs_devices->list, &fs_uuids);
 404                 memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
 405                 fs_devices->latest_devid = devid;
 406                 fs_devices->latest_trans = found_transid;
 407                 mutex_init(&fs_devices->device_list_mutex);
 408                 device = NULL;
 409         } else {
 410                 device = __find_device(&fs_devices->devices, devid,
 411                                        disk_super->dev_item.uuid);
 412         }
 413         if (!device) {
 414                 if (fs_devices->opened)
 415                         return -EBUSY;
 416
 417                 device = kzalloc(sizeof(*device), GFP_NOFS);
 418                 if (!device) {
 419                         /* we can safely leave the fs_devices entry around */
 420                         return -ENOMEM;
 421                 }
 422                 device->devid = devid;
 423                 device->dev_stats_valid = 0;
 424                 device->work.func = pending_bios_fn;
 425                 memcpy(device->uuid, disk_super->dev_item.uuid,
 426                        BTRFS_UUID_SIZE);
 427                 spin_lock_init(&device->io_lock);
 428
 429                 name = rcu_string_strdup(path, GFP_NOFS);
 430                 if (!name) {
 431                         kfree(device);
 432                         return -ENOMEM;
 433                 }
 434                 rcu_assign_pointer(device->name, name);
 435                 INIT_LIST_HEAD(&device->dev_alloc_list);
 436
 437                 /* init readahead state */
 438                 spin_lock_init(&device->reada_lock);
 439                 device->reada_curr_zone = NULL;
 440                 atomic_set(&device->reada_in_flight, 0);
 441                 device->reada_next = 0;
 442                 INIT_RADIX_TREE(&device->reada_zones, GFP_NOFS & ~__GFP_WAIT);
 443                 INIT_RADIX_TREE(&device->reada_extents, GFP_NOFS & ~__GFP_WAIT);
 444
 445                 mutex_lock(&fs_devices->device_list_mutex);
 446                 list_add_rcu(&device->dev_list, &fs_devices->devices);
 447                 mutex_unlock(&fs_devices->device_list_mutex);
 448
 449                 device->fs_devices = fs_devices;
 450                 fs_devices->num_devices++;
 451         } else if (!device->name || strcmp(device->name->str, path)) {
 452                 name = rcu_string_strdup(path, GFP_NOFS);
 453                 if (!name)
 454                         return -ENOMEM;
 455                 rcu_string_free(device->name);
 456                 rcu_assign_pointer(device->name, name);
 457                 if (device->missing) {
 458                         fs_devices->missing_devices--;
 459                         device->missing = 0;
 460                 }
 461         }
 462
 463         if (found_transid > fs_devices->latest_trans) {
 464                 fs_devices->latest_devid = devid;
 465                 fs_devices->latest_trans = found_transid;
 466         }
 467         *fs_devices_ret = fs_devices;
 468         return 0;
 469 }
 470
 471 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
 472 {
 473         struct btrfs_fs_devices *fs_devices;
 474         struct btrfs_device *device;
 475         struct btrfs_device *orig_dev;
 476
 477         fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
 478         if (!fs_devices)
 479                 return ERR_PTR(-ENOMEM);
 480
 481         INIT_LIST_HEAD(&fs_devices->devices);
 482         INIT_LIST_HEAD(&fs_devices->alloc_list);
 483         INIT_LIST_HEAD(&fs_devices->list);
 484         mutex_init(&fs_devices->device_list_mutex);
 485         fs_devices->latest_devid = orig->latest_devid;
 486         fs_devices->latest_trans = orig->latest_trans;
 487         fs_devices->total_devices = orig->total_devices;
 488         memcpy(fs_devices->fsid, orig->fsid, sizeof(fs_devices->fsid));
 489
 490         /* We have held the volume lock, it is safe to get the devices. */
 491         list_for_each_entry(orig_dev, &orig->devices, dev_list) {
 492                 struct rcu_string *name;
 493
 494                 device = kzalloc(sizeof(*device), GFP_NOFS);
 495                 if (!device)
 496                         goto error;
 497
 498                 /*
 499                  * This is ok to do without rcu read locked because we hold the
 500                  * uuid mutex so nothing we touch in here is going to disappear.
 501                  */
 502                 name = rcu_string_strdup(orig_dev->name->str, GFP_NOFS);
 503                 if (!name) {
 504                         kfree(device);
 505                         goto error;
 506                 }
 507                 rcu_assign_pointer(device->name, name);
 508
 509                 device->devid = orig_dev->devid;
 510                 device->work.func = pending_bios_fn;
 511                 memcpy(device->uuid, orig_dev->uuid, sizeof(device->uuid));
 512                 spin_lock_init(&device->io_lock);
 513                 INIT_LIST_HEAD(&device->dev_list);
 514                 INIT_LIST_HEAD(&device->dev_alloc_list);
 515
 516                 list_add(&device->dev_list, &fs_devices->devices);
 517                 device->fs_devices = fs_devices;
 518                 fs_devices->num_devices++;
 519         }
 520         return fs_devices;
 521 error:
 522         free_fs_devices(fs_devices);
 523         return ERR_PTR(-ENOMEM);
 524 }
 525
 526 void btrfs_close_extra_devices(struct btrfs_fs_info *fs_info,
 527                                struct btrfs_fs_devices *fs_devices, int step)
 528 {
 529         struct btrfs_device *device, *next;
 530
 531         struct block_device *latest_bdev = NULL;
 532         u64 latest_devid = 0;
 533         u64 latest_transid = 0;
 534
 535         mutex_lock(&uuid_mutex);
 536 again:
 537         /* This is the initialized path, it is safe to release the devices. */
 538         list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
 539                 if (device->in_fs_metadata) {
 540                         if (!device->is_tgtdev_for_dev_replace &&
 541                             (!latest_transid ||
 542                              device->generation > latest_transid)) {
 543                                 latest_devid = device->devid;
 544                                 latest_transid = device->generation;
 545                                 latest_bdev = device->bdev;
 546                         }
 547                         continue;
 548                 }
 549
 550                 if (device->devid == BTRFS_DEV_REPLACE_DEVID) {
 551                         /*
 552                          * In the first step, keep the device which has
 553                          * the correct fsid and the devid that is used
 554                          * for the dev_replace procedure.
 555                          * In the second step, the dev_replace state is
 556                          * read from the device tree and it is known
 557                          * whether the procedure is really active or
 558                          * not, which means whether this device is
 559                          * used or whether it should be removed.
 560                          */
 561                         if (step == 0 || device->is_tgtdev_for_dev_replace) {
 562                                 continue;
 563                         }
 564                 }
 565                 if (device->bdev) {
 566                         blkdev_put(device->bdev, device->mode);
 567                         device->bdev = NULL;
 568                         fs_devices->open_devices--;
 569                 }
 570                 if (device->writeable) {
 571                         list_del_init(&device->dev_alloc_list);
 572                         device->writeable = 0;
 573                         if (!device->is_tgtdev_for_dev_replace)
 574                                 fs_devices->rw_devices--;
 575                 }
 576                 list_del_init(&device->dev_list);
 577                 fs_devices->num_devices--;
 578                 rcu_string_free(device->name);
 579                 kfree(device);
 580         }
 581
 582         if (fs_devices->seed) {
 583                 fs_devices = fs_devices->seed;
 584                 goto again;
 585         }
 586
 587         fs_devices->latest_bdev = latest_bdev;
 588         fs_devices->latest_devid = latest_devid;
 589         fs_devices->latest_trans = latest_transid;
 590
 591         mutex_unlock(&uuid_mutex);
 592 }
 593
 594 static void __free_device(struct work_struct *work)
 595 {
 596         struct btrfs_device *device;
 597
 598         device = container_of(work, struct btrfs_device, rcu_work);
 599
 600         if (device->bdev)
 601                 blkdev_put(device->bdev, device->mode);
 602
 603         rcu_string_free(device->name);
 604         kfree(device);
 605 }
 606
 607 static void free_device(struct rcu_head *head)
 608 {
 609         struct btrfs_device *device;
 610
 611         device = container_of(head, struct btrfs_device, rcu);
 612
 613         INIT_WORK(&device->rcu_work, __free_device);
 614         schedule_work(&device->rcu_work);
 615 }
 616
 617 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
 618 {
 619         struct btrfs_device *device;
 620
 621         if (--fs_devices->opened > 0)
 622                 return 0;
 623
 624         mutex_lock(&fs_devices->device_list_mutex);
 625         list_for_each_entry(device, &fs_devices->devices, dev_list) {
 626                 struct btrfs_device *new_device;
 627                 struct rcu_string *name;
 628
 629                 if (device->bdev)
 630                         fs_devices->open_devices--;
 631
 632                 if (device->writeable && !device->is_tgtdev_for_dev_replace) {
 633                         list_del_init(&device->dev_alloc_list);
 634                         fs_devices->rw_devices--;
 635                 }
 636
 637                 if (device->can_discard)
 638                         fs_devices->num_can_discard--;
 639
 640                 new_device = kmalloc(sizeof(*new_device), GFP_NOFS);
 641                 BUG_ON(!new_device); /* -ENOMEM */
 642                 memcpy(new_device, device, sizeof(*new_device));
 643
 644                 /* Safe because we are under uuid_mutex */
 645                 if (device->name) {
 646                         name = rcu_string_strdup(device->name->str, GFP_NOFS);
 647                         BUG_ON(device->name && !name); /* -ENOMEM */
 648                         rcu_assign_pointer(new_device->name, name);
 649                 }
 650                 new_device->bdev = NULL;
 651                 new_device->writeable = 0;
 652                 new_device->in_fs_metadata = 0;
 653                 new_device->can_discard = 0;
 654                 spin_lock_init(&new_device->io_lock);
 655                 list_replace_rcu(&device->dev_list, &new_device->dev_list);
 656
 657                 call_rcu(&device->rcu, free_device);
 658         }
 659         mutex_unlock(&fs_devices->device_list_mutex);
 660
 661         WARN_ON(fs_devices->open_devices);
 662         WARN_ON(fs_devices->rw_devices);
 663         fs_devices->opened = 0;
 664         fs_devices->seeding = 0;
 665
 666         return 0;
 667 }
 668
 669 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
 670 {
 671         struct btrfs_fs_devices *seed_devices = NULL;
 672         int ret;
 673
 674         mutex_lock(&uuid_mutex);
 675         ret = __btrfs_close_devices(fs_devices);
 676         if (!fs_devices->opened) {
 677                 seed_devices = fs_devices->seed;
 678                 fs_devices->seed = NULL;
 679         }
 680         mutex_unlock(&uuid_mutex);
 681
 682         while (seed_devices) {
 683                 fs_devices = seed_devices;
 684                 seed_devices = fs_devices->seed;
 685                 __btrfs_close_devices(fs_devices);
 686                 free_fs_devices(fs_devices);
 687         }
 688         /*
 689          * Wait for rcu kworkers under __btrfs_close_devices
 690          * to finish all blkdev_puts so device is really
 691          * free when umount is done.
 692          */
 693         rcu_barrier();
 694         return ret;
 695 }
 696
 697 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
 698                                 fmode_t flags, void *holder)
 699 {
 700         struct request_queue *q;
 701         struct block_device *bdev;
 702         struct list_head *head = &fs_devices->devices;
 703         struct btrfs_device *device;
 704         struct block_device *latest_bdev = NULL;
 705         struct buffer_head *bh;
 706         struct btrfs_super_block *disk_super;
 707         u64 latest_devid = 0;
 708         u64 latest_transid = 0;
 709         u64 devid;
 710         int seeding = 1;
 711         int ret = 0;
 712
 713         flags |= FMODE_EXCL;
 714
 715         list_for_each_entry(device, head, dev_list) {
 716                 if (device->bdev)
 717                         continue;
 718                 if (!device->name)
 719                         continue;
 720
 721                 /* Just open everything we can; ignore failures here */
 722                 if (btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1,
 723                                             &bdev, &bh))
 724                         continue;
 725
 726                 disk_super = (struct btrfs_super_block *)bh->b_data;
 727                 devid = btrfs_stack_device_id(&disk_super->dev_item);
 728                 if (devid != device->devid)
 729                         goto error_brelse;
 730
 731                 if (memcmp(device->uuid, disk_super->dev_item.uuid,
 732                            BTRFS_UUID_SIZE))
 733                         goto error_brelse;
 734
 735                 device->generation = btrfs_super_generation(disk_super);
 736                 if (!latest_transid || device->generation > latest_transid) {
 737                         latest_devid = devid;
 738                         latest_transid = device->generation;
 739                         latest_bdev = bdev;
 740                 }
 741
 742                 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
 743                         device->writeable = 0;
 744                 } else {
 745                         device->writeable = !bdev_read_only(bdev);
 746                         seeding = 0;
 747                 }
 748
 749                 q = bdev_get_queue(bdev);
 750                 if (blk_queue_discard(q)) {
 751                         device->can_discard = 1;
 752                         fs_devices->num_can_discard++;
 753                 }
 754
 755                 device->bdev = bdev;
 756                 device->in_fs_metadata = 0;
 757                 device->mode = flags;
 758
 759                 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
 760                         fs_devices->rotating = 1;
 761
 762                 fs_devices->open_devices++;
 763                 if (device->writeable && !device->is_tgtdev_for_dev_replace) {
 764                         fs_devices->rw_devices++;
 765                         list_add(&device->dev_alloc_list,
 766                                  &fs_devices->alloc_list);
 767                 }
 768                 brelse(bh);
 769                 continue;
 770
 771 error_brelse:
 772                 brelse(bh);
 773                 blkdev_put(bdev, flags);
 774                 continue;
 775         }
 776         if (fs_devices->open_devices == 0) {
 777                 ret = -EINVAL;
 778                 goto out;
 779         }
 780         fs_devices->seeding = seeding;
 781         fs_devices->opened = 1;
 782         fs_devices->latest_bdev = latest_bdev;
 783         fs_devices->latest_devid = latest_devid;
 784         fs_devices->latest_trans = latest_transid;
 785         fs_devices->total_rw_bytes = 0;
 786 out:
 787         return ret;
 788 }
 789
 790 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
 791                        fmode_t flags, void *holder)
 792 {
 793         int ret;
 794
 795         mutex_lock(&uuid_mutex);
 796         if (fs_devices->opened) {
 797                 fs_devices->opened++;
 798                 ret = 0;
 799         } else {
 800                 ret = __btrfs_open_devices(fs_devices, flags, holder);
 801         }
 802         mutex_unlock(&uuid_mutex);
 803         return ret;
 804 }
 805
 806 /*
 807  * Look for a btrfs signature on a device. This may be called out of the mount path
 808  * and we are not allowed to call set_blocksize during the scan. The superblock
 809  * is read via pagecache
 810  */
 811 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
 812                           struct btrfs_fs_devices **fs_devices_ret)
 813 {
 814         struct btrfs_super_block *disk_super;
 815         struct block_device *bdev;
 816         struct page *page;
 817         void *p;
 818         int ret = -EINVAL;
 819         u64 devid;
 820         u64 transid;
 821         u64 total_devices;
 822         u64 bytenr;
 823         pgoff_t index;
 824
 825         /*
 826          * we would like to check all the supers, but that would make
 827          * a btrfs mount succeed after a mkfs from a different FS.
 828          * So, we need to add a special mount option to scan for
 829          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
 830          */
 831         bytenr = btrfs_sb_offset(0);
 832         flags |= FMODE_EXCL;
 833         mutex_lock(&uuid_mutex);
 834
 835         bdev = blkdev_get_by_path(path, flags, holder);
 836
 837         if (IS_ERR(bdev)) {
 838                 ret = PTR_ERR(bdev);
 839                 goto error;
 840         }
 841
 842         /* make sure our super fits in the device */
 843         if (bytenr + PAGE_CACHE_SIZE >= i_size_read(bdev->bd_inode))
 844                 goto error_bdev_put;
 845
 846         /* make sure our super fits in the page */
 847         if (sizeof(*disk_super) > PAGE_CACHE_SIZE)
 848                 goto error_bdev_put;
 849
 850         /* make sure our super doesn't straddle pages on disk */
 851         index = bytenr >> PAGE_CACHE_SHIFT;
 852         if ((bytenr + sizeof(*disk_super) - 1) >> PAGE_CACHE_SHIFT != index)
 853                 goto error_bdev_put;
 854
 855         /* pull in the page with our super */
 856         page = read_cache_page_gfp(bdev->bd_inode->i_mapping,
 857                                    index, GFP_NOFS);
 858
 859         if (IS_ERR_OR_NULL(page))
 860                 goto error_bdev_put;
 861
 862         p = kmap(page);
 863
 864         /* align our pointer to the offset of the super block */
 865         disk_super = p + (bytenr & ~PAGE_CACHE_MASK);
 866
 867         if (btrfs_super_bytenr(disk_super) != bytenr ||
 868             disk_super->magic != cpu_to_le64(BTRFS_MAGIC))
 869                 goto error_unmap;
 870
 871         devid = btrfs_stack_device_id(&disk_super->dev_item);
 872         transid = btrfs_super_generation(disk_super);
 873         total_devices = btrfs_super_num_devices(disk_super);
 874
 875         if (disk_super->label[0]) {
 876                 if (disk_super->label[BTRFS_LABEL_SIZE - 1])
 877                         disk_super->label[BTRFS_LABEL_SIZE - 1] = '\0';
 878                 printk(KERN_INFO "device label %s ", disk_super->label);
 879         } else {
 880                 printk(KERN_INFO "device fsid %pU ", disk_super->fsid);
 881         }
 882
 883         printk(KERN_CONT "devid %llu transid %llu %s\n",
 884                (unsigned long long)devid, (unsigned long long)transid, path);
 885
 886         ret = device_list_add(path, disk_super, devid, fs_devices_ret);
 887         if (!ret && fs_devices_ret)
 888                 (*fs_devices_ret)->total_devices = total_devices;
 889
 890 error_unmap:
 891         kunmap(page);
 892         page_cache_release(page);
 893
 894 error_bdev_put:
 895         blkdev_put(bdev, flags);
 896 error:
 897         mutex_unlock(&uuid_mutex);
 898         return ret;
 899 }
 900
 901 /* helper to account the used device space in the range */
 902 int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
 903                                    u64 end, u64 *length)
 904 {
 905         struct btrfs_key key;
 906         struct btrfs_root *root = device->dev_root;
 907         struct btrfs_dev_extent *dev_extent;
 908         struct btrfs_path *path;
 909         u64 extent_end;
 910         int ret;
 911         int slot;
 912         struct extent_buffer *l;
 913
 914         *length = 0;
 915
 916         if (start >= device->total_bytes || device->is_tgtdev_for_dev_replace)
 917                 return 0;
 918
 919         path = btrfs_alloc_path();
 920         if (!path)
 921                 return -ENOMEM;
 922         path->reada = 2;
 923
 924         key.objectid = device->devid;
 925         key.offset = start;
 926         key.type = BTRFS_DEV_EXTENT_KEY;
 927
 928         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 929         if (ret < 0)
 930                 goto out;
 931         if (ret > 0) {
 932                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
 933                 if (ret < 0)
 934                         goto out;
 935         }
 936
 937         while (1) {
 938                 l = path->nodes[0];
 939                 slot = path->slots[0];
 940                 if (slot >= btrfs_header_nritems(l)) {
 941                         ret = btrfs_next_leaf(root, path);
 942                         if (ret == 0)
 943                                 continue;
 944                         if (ret < 0)
 945                                 goto out;
 946
 947                         break;
 948                 }
 949                 btrfs_item_key_to_cpu(l, &key, slot);
 950
 951                 if (key.objectid < device->devid)
 952                         goto next;
 953
 954                 if (key.objectid > device->devid)
 955                         break;
 956
 957                 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
 958                         goto next;
 959
 960                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
 961                 extent_end = key.offset + btrfs_dev_extent_length(l,
 962                                                                   dev_extent);
 963                 if (key.offset <= start && extent_end > end) {
 964                         *length = end - start + 1;
 965                         break;
 966                 } else if (key.offset <= start && extent_end > start)
 967                         *length += extent_end - start;
 968                 else if (key.offset > start && extent_end <= end)
 969                         *length += extent_end - key.offset;
 970                 else if (key.offset > start && key.offset <= end) {
 971                         *length += end - key.offset + 1;
 972                         break;
 973                 } else if (key.offset > end)
 974                         break;
 975
 976 next:
 977                 path->slots[0]++;
 978         }
 979         ret = 0;
 980 out:
 981         btrfs_free_path(path);
 982         return ret;
 983 }
 984
 985 /*
 986  * find_free_dev_extent - find free space in the specified device
 987  * @device:     the device which we search the free space in
 988  * @num_bytes:  the size of the free space that we need
 989  * @start:      store the start of the free space.
 990  * @len:        the size of the free space. that we find, or the size of the max
 991  *              free space if we don't find suitable free space
 992  *
 993  * this uses a pretty simple search, the expectation is that it is
 994  * called very infrequently and that a given device has a small number
 995  * of extents
 996  *
 997  * @start is used to store the start of the free space if we find. But if we
 998  * don't find suitable free space, it will be used to store the start position
 999  * of the max free space.
1000  *
1001  * @len is used to store the size of the free space that we find.
1002  * But if we don't find suitable free space, it is used to store the size of
1003  * the max free space.
1004  */
1005 int find_free_dev_extent(struct btrfs_device *device, u64 num_bytes,
1006                          u64 *start, u64 *len)
1007 {
1008         struct btrfs_key key;
1009         struct btrfs_root *root = device->dev_root;
1010         struct btrfs_dev_extent *dev_extent;
1011         struct btrfs_path *path;
1012         u64 hole_size;
1013         u64 max_hole_start;
1014         u64 max_hole_size;
1015         u64 extent_end;
1016         u64 search_start;
1017         u64 search_end = device->total_bytes;
1018         int ret;
1019         int slot;
1020         struct extent_buffer *l;
1021
1022         /* FIXME use last free of some kind */
1023
1024         /* we don't want to overwrite the superblock on the drive,
1025          * so we make sure to start at an offset of at least 1MB
1026          */
1027         search_start = max(root->fs_info->alloc_start, 1024ull * 1024);
1028
1029         max_hole_start = search_start;
1030         max_hole_size = 0;
1031         hole_size = 0;
1032
1033         if (search_start >= search_end || device->is_tgtdev_for_dev_replace) {
1034                 ret = -ENOSPC;
1035                 goto error;
1036         }
1037
1038         path = btrfs_alloc_path();
1039         if (!path) {
1040                 ret = -ENOMEM;
1041                 goto error;
1042         }
1043         path->reada = 2;
1044
1045         key.objectid = device->devid;
1046         key.offset = search_start;
1047         key.type = BTRFS_DEV_EXTENT_KEY;
1048
1049         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1050         if (ret < 0)
1051                 goto out;
1052         if (ret > 0) {
1053                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1054                 if (ret < 0)
1055                         goto out;
1056         }
1057
1058         while (1) {
1059                 l = path->nodes[0];
1060                 slot = path->slots[0];
1061                 if (slot >= btrfs_header_nritems(l)) {
1062                         ret = btrfs_next_leaf(root, path);
1063                         if (ret == 0)
1064                                 continue;
1065                         if (ret < 0)
1066                                 goto out;
1067
1068                         break;
1069                 }
1070                 btrfs_item_key_to_cpu(l, &key, slot);
1071
1072                 if (key.objectid < device->devid)
1073                         goto next;
1074
1075                 if (key.objectid > device->devid)
1076                         break;
1077
1078                 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
1079                         goto next;
1080
1081                 if (key.offset > search_start) {
1082                         hole_size = key.offset - search_start;
1083
1084                         if (hole_size > max_hole_size) {
1085                                 max_hole_start = search_start;
1086                                 max_hole_size = hole_size;
1087                         }
1088
1089                         /*
1090                          * If this free space is greater than which we need,
1091                          * it must be the max free space that we have found
1092                          * until now, so max_hole_start must point to the start
1093                          * of this free space and the length of this free space
1094                          * is stored in max_hole_size. Thus, we return
1095                          * max_hole_start and max_hole_size and go back to the
1096                          * caller.
1097                          */
1098                         if (hole_size >= num_bytes) {
1099                                 ret = 0;
1100                                 goto out;
1101                         }
1102                 }
1103
1104                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1105                 extent_end = key.offset + btrfs_dev_extent_length(l,
1106                                                                   dev_extent);
1107                 if (extent_end > search_start)
1108                         search_start = extent_end;
1109 next:
1110                 path->slots[0]++;
1111                 cond_resched();
1112         }
1113
1114         /*
1115          * At this point, search_start should be the end of
1116          * allocated dev extents, and when shrinking the device,
1117          * search_end may be smaller than search_start.
1118          */
1119         if (search_end > search_start)
1120                 hole_size = search_end - search_start;
1121
1122         if (hole_size > max_hole_size) {
1123                 max_hole_start = search_start;
1124                 max_hole_size = hole_size;
1125         }
1126
1127         /* See above. */
1128         if (hole_size < num_bytes)
1129                 ret = -ENOSPC;
1130         else
1131                 ret = 0;
1132
1133 out:
1134         btrfs_free_path(path);
1135 error:
1136         *start = max_hole_start;
1137         if (len)
1138                 *len = max_hole_size;
1139         return ret;
1140 }
1141
1142 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1143                           struct btrfs_device *device,
1144                           u64 start)
1145 {
1146         int ret;
1147         struct btrfs_path *path;
1148         struct btrfs_root *root = device->dev_root;
1149         struct btrfs_key key;
1150         struct btrfs_key found_key;
1151         struct extent_buffer *leaf = NULL;
1152         struct btrfs_dev_extent *extent = NULL;
1153
1154         path = btrfs_alloc_path();
1155         if (!path)
1156                 return -ENOMEM;
1157
1158         key.objectid = device->devid;
1159         key.offset = start;
1160         key.type = BTRFS_DEV_EXTENT_KEY;
1161 again:
1162         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1163         if (ret > 0) {
1164                 ret = btrfs_previous_item(root, path, key.objectid,
1165                                           BTRFS_DEV_EXTENT_KEY);
1166                 if (ret)
1167                         goto out;
1168                 leaf = path->nodes[0];
1169                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1170                 extent = btrfs_item_ptr(leaf, path->slots[0],
1171                                         struct btrfs_dev_extent);
1172                 BUG_ON(found_key.offset > start || found_key.offset +
1173                        btrfs_dev_extent_length(leaf, extent) < start);
1174                 key = found_key;
1175                 btrfs_release_path(path);
1176                 goto again;
1177         } else if (ret == 0) {
1178                 leaf = path->nodes[0];
1179                 extent = btrfs_item_ptr(leaf, path->slots[0],
1180                                         struct btrfs_dev_extent);
1181         } else {
1182                 btrfs_error(root->fs_info, ret, "Slot search failed");
1183                 goto out;
1184         }
1185
1186         if (device->bytes_used > 0) {
1187                 u64 len = btrfs_dev_extent_length(leaf, extent);
1188                 device->bytes_used -= len;
1189                 spin_lock(&root->fs_info->free_chunk_lock);
1190                 root->fs_info->free_chunk_space += len;
1191                 spin_unlock(&root->fs_info->free_chunk_lock);
1192         }
1193         ret = btrfs_del_item(trans, root, path);
1194         if (ret) {
1195                 btrfs_error(root->fs_info, ret,
1196                             "Failed to remove dev extent item");
1197         }
1198 out:
1199         btrfs_free_path(path);
1200         return ret;
1201 }
1202
1203 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1204                                   struct btrfs_device *device,
1205                                   u64 chunk_tree, u64 chunk_objectid,
1206                                   u64 chunk_offset, u64 start, u64 num_bytes)
1207 {
1208         int ret;
1209         struct btrfs_path *path;
1210         struct btrfs_root *root = device->dev_root;
1211         struct btrfs_dev_extent *extent;
1212         struct extent_buffer *leaf;
1213         struct btrfs_key key;
1214
1215         WARN_ON(!device->in_fs_metadata);
1216         WARN_ON(device->is_tgtdev_for_dev_replace);
1217         path = btrfs_alloc_path();
1218         if (!path)
1219                 return -ENOMEM;
1220
1221         key.objectid = device->devid;
1222         key.offset = start;
1223         key.type = BTRFS_DEV_EXTENT_KEY;
1224         ret = btrfs_insert_empty_item(trans, root, path, &key,
1225                                       sizeof(*extent));
1226         if (ret)
1227                 goto out;
1228
1229         leaf = path->nodes[0];
1230         extent = btrfs_item_ptr(leaf, path->slots[0],
1231                                 struct btrfs_dev_extent);
1232         btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
1233         btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
1234         btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1235
1236         write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
1237                     (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
1238                     BTRFS_UUID_SIZE);
1239
1240         btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1241         btrfs_mark_buffer_dirty(leaf);
1242 out:
1243         btrfs_free_path(path);
1244         return ret;
1245 }
1246
1247 static noinline int find_next_chunk(struct btrfs_root *root,
1248                                     u64 objectid, u64 *offset)
1249 {
1250         struct btrfs_path *path;
1251         int ret;
1252         struct btrfs_key key;
1253         struct btrfs_chunk *chunk;
1254         struct btrfs_key found_key;
1255
1256         path = btrfs_alloc_path();
1257         if (!path)
1258                 return -ENOMEM;
1259
1260         key.objectid = objectid;
1261         key.offset = (u64)-1;
1262         key.type = BTRFS_CHUNK_ITEM_KEY;
1263
1264         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1265         if (ret < 0)
1266                 goto error;
1267
1268         BUG_ON(ret == 0); /* Corruption */
1269
1270         ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
1271         if (ret) {
1272                 *offset = 0;
1273         } else {
1274                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1275                                       path->slots[0]);
1276                 if (found_key.objectid != objectid)
1277                         *offset = 0;
1278                 else {
1279                         chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
1280                                                struct btrfs_chunk);
1281                         *offset = found_key.offset +
1282                                 btrfs_chunk_length(path->nodes[0], chunk);
1283                 }
1284         }
1285         ret = 0;
1286 error:
1287         btrfs_free_path(path);
1288         return ret;
1289 }
1290
1291 static noinline int find_next_devid(struct btrfs_root *root, u64 *objectid)
1292 {
1293         int ret;
1294         struct btrfs_key key;
1295         struct btrfs_key found_key;
1296         struct btrfs_path *path;
1297
1298         root = root->fs_info->chunk_root;
1299
1300         path = btrfs_alloc_path();
1301         if (!path)
1302                 return -ENOMEM;
1303
1304         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1305         key.type = BTRFS_DEV_ITEM_KEY;
1306         key.offset = (u64)-1;
1307
1308         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1309         if (ret < 0)
1310                 goto error;
1311
1312         BUG_ON(ret == 0); /* Corruption */
1313
1314         ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
1315                                   BTRFS_DEV_ITEM_KEY);
1316         if (ret) {
1317                 *objectid = 1;
1318         } else {
1319                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1320                                       path->slots[0]);
1321                 *objectid = found_key.offset + 1;
1322         }
1323         ret = 0;
1324 error:
1325         btrfs_free_path(path);
1326         return ret;
1327 }
1328
1329 /*
1330  * the device information is stored in the chunk root
1331  * the btrfs_device struct should be fully filled in
1332  */
1333 static int btrfs_add_device(struct btrfs_trans_handle *trans,
1334                             struct btrfs_root *root,
1335                             struct btrfs_device *device)
1336 {
1337         int ret;
1338         struct btrfs_path *path;
1339         struct btrfs_dev_item *dev_item;
1340         struct extent_buffer *leaf;
1341         struct btrfs_key key;
1342         unsigned long ptr;
1343
1344         root = root->fs_info->chunk_root;
1345
1346         path = btrfs_alloc_path();
1347         if (!path)
1348                 return -ENOMEM;
1349
1350         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1351         key.type = BTRFS_DEV_ITEM_KEY;
1352         key.offset = device->devid;
1353
1354         ret = btrfs_insert_empty_item(trans, root, path, &key,
1355                                       sizeof(*dev_item));
1356         if (ret)
1357                 goto out;
1358
1359         leaf = path->nodes[0];
1360         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1361
1362         btrfs_set_device_id(leaf, dev_item, device->devid);
1363         btrfs_set_device_generation(leaf, dev_item, 0);
1364         btrfs_set_device_type(leaf, dev_item, device->type);
1365         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1366         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1367         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1368         btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
1369         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1370         btrfs_set_device_group(leaf, dev_item, 0);
1371         btrfs_set_device_seek_speed(leaf, dev_item, 0);
1372         btrfs_set_device_bandwidth(leaf, dev_item, 0);
1373         btrfs_set_device_start_offset(leaf, dev_item, 0);
1374
1375         ptr = (unsigned long)btrfs_device_uuid(dev_item);
1376         write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1377         ptr = (unsigned long)btrfs_device_fsid(dev_item);
1378         write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1379         btrfs_mark_buffer_dirty(leaf);
1380
1381         ret = 0;
1382 out:
1383         btrfs_free_path(path);
1384         return ret;
1385 }
1386
1387 /*
1388  * Function to update ctime/mtime for a given device path.
1389  * Mainly used for ctime/mtime based probe like libblkid.
1390  */
1391 static void update_dev_time(char *path_name)
1392 {
1393         struct file *filp;
1394
1395         filp = filp_open(path_name, O_RDWR, 0);
1396         if (!filp)
1397                 return;
1398         file_update_time(filp);
1399         filp_close(filp, NULL);
1400         return;
1401 }
1402
1403 static int btrfs_rm_dev_item(struct btrfs_root *root,
1404                              struct btrfs_device *device)
1405 {
1406         int ret;
1407         struct btrfs_path *path;
1408         struct btrfs_key key;
1409         struct btrfs_trans_handle *trans;
1410
1411         root = root->fs_info->chunk_root;
1412
1413         path = btrfs_alloc_path();
1414         if (!path)
1415                 return -ENOMEM;
1416
1417         trans = btrfs_start_transaction(root, 0);
1418         if (IS_ERR(trans)) {
1419                 btrfs_free_path(path);
1420                 return PTR_ERR(trans);
1421         }
1422         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1423         key.type = BTRFS_DEV_ITEM_KEY;
1424         key.offset = device->devid;
1425         lock_chunks(root);
1426
1427         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1428         if (ret < 0)
1429                 goto out;
1430
1431         if (ret > 0) {
1432                 ret = -ENOENT;
1433                 goto out;
1434         }
1435
1436         ret = btrfs_del_item(trans, root, path);
1437         if (ret)
1438                 goto out;
1439 out:
1440         btrfs_free_path(path);
1441         unlock_chunks(root);
1442         btrfs_commit_transaction(trans, root);
1443         return ret;
1444 }
1445
1446 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
1447 {
1448         struct btrfs_device *device;
1449         struct btrfs_device *next_device;
1450         struct block_device *bdev;
1451         struct buffer_head *bh = NULL;
1452         struct btrfs_super_block *disk_super;
1453         struct btrfs_fs_devices *cur_devices;
1454         u64 all_avail;
1455         u64 devid;
1456         u64 num_devices;
1457         u8 *dev_uuid;
1458         unsigned seq;
1459         int ret = 0;
1460         bool clear_super = false;
1461
1462         mutex_lock(&uuid_mutex);
1463
1464         do {
1465                 seq = read_seqbegin(&root->fs_info->profiles_lock);
1466
1467                 all_avail = root->fs_info->avail_data_alloc_bits |
1468                             root->fs_info->avail_system_alloc_bits |
1469                             root->fs_info->avail_metadata_alloc_bits;
1470         } while (read_seqretry(&root->fs_info->profiles_lock, seq));
1471
1472         num_devices = root->fs_info->fs_devices->num_devices;
1473         btrfs_dev_replace_lock(&root->fs_info->dev_replace);
1474         if (btrfs_dev_replace_is_ongoing(&root->fs_info->dev_replace)) {
1475                 WARN_ON(num_devices < 1);
1476                 num_devices--;
1477         }
1478         btrfs_dev_replace_unlock(&root->fs_info->dev_replace);
1479
1480         if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) && num_devices <= 4) {
1481                 printk(KERN_ERR "btrfs: unable to go below four devices "
1482                        "on raid10\n");
1483                 ret = -EINVAL;
1484                 goto out;
1485         }
1486
1487         if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) && num_devices <= 2) {
1488                 printk(KERN_ERR "btrfs: unable to go below two "
1489                        "devices on raid1\n");
1490                 ret = -EINVAL;
1491                 goto out;
1492         }
1493
1494         if ((all_avail & BTRFS_BLOCK_GROUP_RAID5) &&
1495             root->fs_info->fs_devices->rw_devices <= 2) {
1496                 printk(KERN_ERR "btrfs: unable to go below two "
1497                        "devices on raid5\n");
1498                 ret = -EINVAL;
1499                 goto out;
1500         }
1501         if ((all_avail & BTRFS_BLOCK_GROUP_RAID6) &&
1502             root->fs_info->fs_devices->rw_devices <= 3) {
1503                 printk(KERN_ERR "btrfs: unable to go below three "
1504                        "devices on raid6\n");
1505                 ret = -EINVAL;
1506                 goto out;
1507         }
1508
1509         if (strcmp(device_path, "missing") == 0) {
1510                 struct list_head *devices;
1511                 struct btrfs_device *tmp;
1512
1513                 device = NULL;
1514                 devices = &root->fs_info->fs_devices->devices;
1515                 /*
1516                  * It is safe to read the devices since the volume_mutex
1517                  * is held.
1518                  */
1519                 list_for_each_entry(tmp, devices, dev_list) {
1520                         if (tmp->in_fs_metadata &&
1521                             !tmp->is_tgtdev_for_dev_replace &&
1522                             !tmp->bdev) {
1523                                 device = tmp;
1524                                 break;
1525                         }
1526                 }
1527                 bdev = NULL;
1528                 bh = NULL;
1529                 disk_super = NULL;
1530                 if (!device) {
1531                         printk(KERN_ERR "btrfs: no missing devices found to "
1532                                "remove\n");
1533                         goto out;
1534                 }
1535         } else {
1536                 ret = btrfs_get_bdev_and_sb(device_path,
1537                                             FMODE_WRITE | FMODE_EXCL,
1538                                             root->fs_info->bdev_holder, 0,
1539                                             &bdev, &bh);
1540                 if (ret)
1541                         goto out;
1542                 disk_super = (struct btrfs_super_block *)bh->b_data;
1543                 devid = btrfs_stack_device_id(&disk_super->dev_item);
1544                 dev_uuid = disk_super->dev_item.uuid;
1545                 device = btrfs_find_device(root->fs_info, devid, dev_uuid,
1546                                            disk_super->fsid);
1547                 if (!device) {
1548                         ret = -ENOENT;
1549                         goto error_brelse;
1550                 }
1551         }
1552
1553         if (device->is_tgtdev_for_dev_replace) {
1554                 pr_err("btrfs: unable to remove the dev_replace target dev\n");
1555                 ret = -EINVAL;
1556                 goto error_brelse;
1557         }
1558
1559         if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1560                 printk(KERN_ERR "btrfs: unable to remove the only writeable "
1561                        "device\n");
1562                 ret = -EINVAL;
1563                 goto error_brelse;
1564         }
1565
1566         if (device->writeable) {
1567                 lock_chunks(root);
1568                 list_del_init(&device->dev_alloc_list);
1569                 unlock_chunks(root);
1570                 root->fs_info->fs_devices->rw_devices--;
1571                 clear_super = true;
1572         }
1573
1574         ret = btrfs_shrink_device(device, 0);
1575         if (ret)
1576                 goto error_undo;
1577
1578         /*
1579          * TODO: the superblock still includes this device in its num_devices
1580          * counter although write_all_supers() is not locked out. This
1581          * could give a filesystem state which requires a degraded mount.
1582          */
1583         ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1584         if (ret)
1585                 goto error_undo;
1586
1587         spin_lock(&root->fs_info->free_chunk_lock);
1588         root->fs_info->free_chunk_space = device->total_bytes -
1589                 device->bytes_used;
1590         spin_unlock(&root->fs_info->free_chunk_lock);
1591
1592         device->in_fs_metadata = 0;
1593         btrfs_scrub_cancel_dev(root->fs_info, device);
1594
1595         /*
1596          * the device list mutex makes sure that we don't change
1597          * the device list while someone else is writing out all
1598          * the device supers.
1599          */
1600
1601         cur_devices = device->fs_devices;
1602         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1603         list_del_rcu(&device->dev_list);
1604
1605         device->fs_devices->num_devices--;
1606         device->fs_devices->total_devices--;
1607
1608         if (device->missing)
1609                 root->fs_info->fs_devices->missing_devices--;
1610
1611         next_device = list_entry(root->fs_info->fs_devices->devices.next,
1612                                  struct btrfs_device, dev_list);
1613         if (device->bdev == root->fs_info->sb->s_bdev)
1614                 root->fs_info->sb->s_bdev = next_device->bdev;
1615         if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1616                 root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1617
1618         if (device->bdev)
1619                 device->fs_devices->open_devices--;
1620
1621         call_rcu(&device->rcu, free_device);
1622         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1623
1624         num_devices = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
1625         btrfs_set_super_num_devices(root->fs_info->super_copy, num_devices);
1626
1627         if (cur_devices->open_devices == 0) {
1628                 struct btrfs_fs_devices *fs_devices;
1629                 fs_devices = root->fs_info->fs_devices;
1630                 while (fs_devices) {
1631                         if (fs_devices->seed == cur_devices) {
1632                                 fs_devices->seed = cur_devices->seed;
1633                                 break;
1634                         }
1635                         fs_devices = fs_devices->seed;
1636                 }
1637                 cur_devices->seed = NULL;
1638                 lock_chunks(root);
1639                 __btrfs_close_devices(cur_devices);
1640                 unlock_chunks(root);
1641                 free_fs_devices(cur_devices);
1642         }
1643
1644         root->fs_info->num_tolerated_disk_barrier_failures =
1645                 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
1646
1647         /*
1648          * at this point, the device is zero sized.  We want to
1649          * remove it from the devices list and zero out the old super
1650          */
1651         if (clear_super && disk_super) {
1652                 /* make sure this device isn't detected as part of
1653                  * the FS anymore
1654                  */
1655                 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1656                 set_buffer_dirty(bh);
1657                 sync_dirty_buffer(bh);
1658         }
1659
1660         ret = 0;
1661
1662         if (bdev) {
1663                 /* Notify udev that device has changed */
1664                 btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
1665
1666                 /* Update ctime/mtime for device path for libblkid */
1667                 update_dev_time(device_path);
1668         }
1669
1670 error_brelse:
1671         brelse(bh);
1672         if (bdev)
1673                 blkdev_put(bdev, FMODE_READ | FMODE_EXCL);
1674 out:
1675         mutex_unlock(&uuid_mutex);
1676         return ret;
1677 error_undo:
1678         if (device->writeable) {
1679                 lock_chunks(root);
1680                 list_add(&device->dev_alloc_list,
1681                          &root->fs_info->fs_devices->alloc_list);
1682                 unlock_chunks(root);
1683                 root->fs_info->fs_devices->rw_devices++;
1684         }
1685         goto error_brelse;
1686 }
1687
1688 void btrfs_rm_dev_replace_srcdev(struct btrfs_fs_info *fs_info,
1689                                  struct btrfs_device *srcdev)
1690 {
1691         WARN_ON(!mutex_is_locked(&fs_info->fs_devices->device_list_mutex));
1692         list_del_rcu(&srcdev->dev_list);
1693         list_del_rcu(&srcdev->dev_alloc_list);
1694         fs_info->fs_devices->num_devices--;
1695         if (srcdev->missing) {
1696                 fs_info->fs_devices->missing_devices--;
1697                 fs_info->fs_devices->rw_devices++;
1698         }
1699         if (srcdev->can_discard)
1700                 fs_info->fs_devices->num_can_discard--;
1701         if (srcdev->bdev)
1702                 fs_info->fs_devices->open_devices--;
1703
1704         call_rcu(&srcdev->rcu, free_device);
1705 }
1706
1707 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
1708                                       struct btrfs_device *tgtdev)
1709 {
1710         struct btrfs_device *next_device;
1711
1712         WARN_ON(!tgtdev);
1713         mutex_lock(&fs_info->fs_devices->device_list_mutex);
1714         if (tgtdev->bdev) {
1715                 btrfs_scratch_superblock(tgtdev);
1716                 fs_info->fs_devices->open_devices--;
1717         }
1718         fs_info->fs_devices->num_devices--;
1719         if (tgtdev->can_discard)
1720                 fs_info->fs_devices->num_can_discard++;
1721
1722         next_device = list_entry(fs_info->fs_devices->devices.next,
1723                                  struct btrfs_device, dev_list);
1724         if (tgtdev->bdev == fs_info->sb->s_bdev)
1725                 fs_info->sb->s_bdev = next_device->bdev;
1726         if (tgtdev->bdev == fs_info->fs_devices->latest_bdev)
1727                 fs_info->fs_devices->latest_bdev = next_device->bdev;
1728         list_del_rcu(&tgtdev->dev_list);
1729
1730         call_rcu(&tgtdev->rcu, free_device);
1731
1732         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
1733 }
1734
1735 static int btrfs_find_device_by_path(struct btrfs_root *root, char *device_path,
1736                                      struct btrfs_device **device)
1737 {
1738         int ret = 0;
1739         struct btrfs_super_block *disk_super;
1740         u64 devid;
1741         u8 *dev_uuid;
1742         struct block_device *bdev;
1743         struct buffer_head *bh;
1744
1745         *device = NULL;
1746         ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
1747                                     root->fs_info->bdev_holder, 0, &bdev, &bh);
1748         if (ret)
1749                 return ret;
1750         disk_super = (struct btrfs_super_block *)bh->b_data;
1751         devid = btrfs_stack_device_id(&disk_super->dev_item);
1752         dev_uuid = disk_super->dev_item.uuid;
1753         *device = btrfs_find_device(root->fs_info, devid, dev_uuid,
1754                                     disk_super->fsid);
1755         brelse(bh);
1756         if (!*device)
1757                 ret = -ENOENT;
1758         blkdev_put(bdev, FMODE_READ);
1759         return ret;
1760 }
1761
1762 int btrfs_find_device_missing_or_by_path(struct btrfs_root *root,
1763                                          char *device_path,
1764                                          struct btrfs_device **device)
1765 {
1766         *device = NULL;
1767         if (strcmp(device_path, "missing") == 0) {
1768                 struct list_head *devices;
1769                 struct btrfs_device *tmp;
1770
1771                 devices = &root->fs_info->fs_devices->devices;
1772                 /*
1773                  * It is safe to read the devices since the volume_mutex
1774                  * is held by the caller.
1775                  */
1776                 list_for_each_entry(tmp, devices, dev_list) {
1777                         if (tmp->in_fs_metadata && !tmp->bdev) {
1778                                 *device = tmp;
1779                                 break;
1780                         }
1781                 }
1782
1783                 if (!*device) {
1784                         pr_err("btrfs: no missing device found\n");
1785                         return -ENOENT;
1786                 }
1787
1788                 return 0;
1789         } else {
1790                 return btrfs_find_device_by_path(root, device_path, device);
1791         }
1792 }
1793
1794 /*
1795  * does all the dirty work required for changing file system's UUID.
1796  */
1797 static int btrfs_prepare_sprout(struct btrfs_root *root)
1798 {
1799         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
1800         struct btrfs_fs_devices *old_devices;
1801         struct btrfs_fs_devices *seed_devices;
1802         struct btrfs_super_block *disk_super = root->fs_info->super_copy;
1803         struct btrfs_device *device;
1804         u64 super_flags;
1805
1806         BUG_ON(!mutex_is_locked(&uuid_mutex));
1807         if (!fs_devices->seeding)
1808                 return -EINVAL;
1809
1810         seed_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
1811         if (!seed_devices)
1812                 return -ENOMEM;
1813
1814         old_devices = clone_fs_devices(fs_devices);
1815         if (IS_ERR(old_devices)) {
1816                 kfree(seed_devices);
1817                 return PTR_ERR(old_devices);
1818         }
1819
1820         list_add(&old_devices->list, &fs_uuids);
1821
1822         memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
1823         seed_devices->opened = 1;
1824         INIT_LIST_HEAD(&seed_devices->devices);
1825         INIT_LIST_HEAD(&seed_devices->alloc_list);
1826         mutex_init(&seed_devices->device_list_mutex);
1827
1828         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1829         list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
1830                               synchronize_rcu);
1831         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1832
1833         list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
1834         list_for_each_entry(device, &seed_devices->devices, dev_list) {
1835                 device->fs_devices = seed_devices;
1836         }
1837
1838         fs_devices->seeding = 0;
1839         fs_devices->num_devices = 0;
1840         fs_devices->open_devices = 0;
1841         fs_devices->seed = seed_devices;
1842
1843         generate_random_uuid(fs_devices->fsid);
1844         memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1845         memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1846         super_flags = btrfs_super_flags(disk_super) &
1847                       ~BTRFS_SUPER_FLAG_SEEDING;
1848         btrfs_set_super_flags(disk_super, super_flags);
1849
1850         return 0;
1851 }
1852
1853 /*
1854  * strore the expected generation for seed devices in device items.
1855  */
1856 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
1857                                struct btrfs_root *root)
1858 {
1859         struct btrfs_path *path;
1860         struct extent_buffer *leaf;
1861         struct btrfs_dev_item *dev_item;
1862         struct btrfs_device *device;
1863         struct btrfs_key key;
1864         u8 fs_uuid[BTRFS_UUID_SIZE];
1865         u8 dev_uuid[BTRFS_UUID_SIZE];
1866         u64 devid;
1867         int ret;
1868
1869         path = btrfs_alloc_path();
1870         if (!path)
1871                 return -ENOMEM;
1872
1873         root = root->fs_info->chunk_root;
1874         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1875         key.offset = 0;
1876         key.type = BTRFS_DEV_ITEM_KEY;
1877
1878         while (1) {
1879                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1880                 if (ret < 0)
1881                         goto error;
1882
1883                 leaf = path->nodes[0];
1884 next_slot:
1885                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1886                         ret = btrfs_next_leaf(root, path);
1887                         if (ret > 0)
1888                                 break;
1889                         if (ret < 0)
1890                                 goto error;
1891                         leaf = path->nodes[0];
1892                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1893                         btrfs_release_path(path);
1894                         continue;
1895                 }
1896
1897                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1898                 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
1899                     key.type != BTRFS_DEV_ITEM_KEY)
1900                         break;
1901
1902                 dev_item = btrfs_item_ptr(leaf, path->slots[0],
1903                                           struct btrfs_dev_item);
1904                 devid = btrfs_device_id(leaf, dev_item);
1905                 read_extent_buffer(leaf, dev_uuid,
1906                                    (unsigned long)btrfs_device_uuid(dev_item),
1907                                    BTRFS_UUID_SIZE);
1908                 read_extent_buffer(leaf, fs_uuid,
1909                                    (unsigned long)btrfs_device_fsid(dev_item),
1910                                    BTRFS_UUID_SIZE);
1911                 device = btrfs_find_device(root->fs_info, devid, dev_uuid,
1912                                            fs_uuid);
1913                 BUG_ON(!device); /* Logic error */
1914
1915                 if (device->fs_devices->seeding) {
1916                         btrfs_set_device_generation(leaf, dev_item,
1917                                                     device->generation);
1918                         btrfs_mark_buffer_dirty(leaf);
1919                 }
1920
1921                 path->slots[0]++;
1922                 goto next_slot;
1923         }
1924         ret = 0;
1925 error:
1926         btrfs_free_path(path);
1927         return ret;
1928 }
1929
1930 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
1931 {
1932         struct request_queue *q;
1933         struct btrfs_trans_handle *trans;
1934         struct btrfs_device *device;
1935         struct block_device *bdev;
1936         struct list_head *devices;
1937         struct super_block *sb = root->fs_info->sb;
1938         struct rcu_string *name;
1939         u64 total_bytes;
1940         int seeding_dev = 0;
1941         int ret = 0;
1942
1943         if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
1944                 return -EROFS;
1945
1946         bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
1947                                   root->fs_info->bdev_holder);
1948         if (IS_ERR(bdev))
1949                 return PTR_ERR(bdev);
1950
1951         if (root->fs_info->fs_devices->seeding) {
1952                 seeding_dev = 1;
1953                 down_write(&sb->s_umount);
1954                 mutex_lock(&uuid_mutex);
1955         }
1956
1957         filemap_write_and_wait(bdev->bd_inode->i_mapping);
1958
1959         devices = &root->fs_info->fs_devices->devices;
1960
1961         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1962         list_for_each_entry(device, devices, dev_list) {
1963                 if (device->bdev == bdev) {
1964                         ret = -EEXIST;
1965                         mutex_unlock(
1966                                 &root->fs_info->fs_devices->device_list_mutex);
1967                         goto error;
1968                 }
1969         }
1970         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1971
1972         device = kzalloc(sizeof(*device), GFP_NOFS);
1973         if (!device) {
1974                 /* we can safely leave the fs_devices entry around */
1975                 ret = -ENOMEM;
1976                 goto error;
1977         }
1978
1979         name = rcu_string_strdup(device_path, GFP_NOFS);
1980         if (!name) {
1981                 kfree(device);
1982                 ret = -ENOMEM;
1983                 goto error;
1984         }
1985         rcu_assign_pointer(device->name, name);
1986
1987         ret = find_next_devid(root, &device->devid);
1988         if (ret) {
1989                 rcu_string_free(device->name);
1990                 kfree(device);
1991                 goto error;
1992         }
1993
1994         trans = btrfs_start_transaction(root, 0);
1995         if (IS_ERR(trans)) {
1996                 rcu_string_free(device->name);
1997                 kfree(device);
1998                 ret = PTR_ERR(trans);
1999                 goto error;
2000         }
2001
2002         lock_chunks(root);
2003
2004         q = bdev_get_queue(bdev);
2005         if (blk_queue_discard(q))
2006                 device->can_discard = 1;
2007         device->writeable = 1;
2008         device->work.func = pending_bios_fn;
2009         generate_random_uuid(device->uuid);
2010         spin_lock_init(&device->io_lock);
2011         device->generation = trans->transid;
2012         device->io_width = root->sectorsize;
2013         device->io_align = root->sectorsize;
2014         device->sector_size = root->sectorsize;
2015         device->total_bytes = i_size_read(bdev->bd_inode);
2016         device->disk_total_bytes = device->total_bytes;
2017         device->dev_root = root->fs_info->dev_root;
2018         device->bdev = bdev;
2019         device->in_fs_metadata = 1;
2020         device->is_tgtdev_for_dev_replace = 0;
2021         device->mode = FMODE_EXCL;
2022         set_blocksize(device->bdev, 4096);
2023
2024         if (seeding_dev) {
2025                 sb->s_flags &= ~MS_RDONLY;
2026                 ret = btrfs_prepare_sprout(root);
2027                 BUG_ON(ret); /* -ENOMEM */
2028         }
2029
2030         device->fs_devices = root->fs_info->fs_devices;
2031
2032         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2033         list_add_rcu(&device->dev_list, &root->fs_info->fs_devices->devices);
2034         list_add(&device->dev_alloc_list,
2035                  &root->fs_info->fs_devices->alloc_list);
2036         root->fs_info->fs_devices->num_devices++;
2037         root->fs_info->fs_devices->open_devices++;
2038         root->fs_info->fs_devices->rw_devices++;
2039         root->fs_info->fs_devices->total_devices++;
2040         if (device->can_discard)
2041                 root->fs_info->fs_devices->num_can_discard++;
2042         root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
2043
2044         spin_lock(&root->fs_info->free_chunk_lock);
2045         root->fs_info->free_chunk_space += device->total_bytes;
2046         spin_unlock(&root->fs_info->free_chunk_lock);
2047
2048         if (!blk_queue_nonrot(bdev_get_queue(bdev)))
2049                 root->fs_info->fs_devices->rotating = 1;
2050
2051         total_bytes = btrfs_super_total_bytes(root->fs_info->super_copy);
2052         btrfs_set_super_total_bytes(root->fs_info->super_copy,
2053                                     total_bytes + device->total_bytes);
2054
2055         total_bytes = btrfs_super_num_devices(root->fs_info->super_copy);
2056         btrfs_set_super_num_devices(root->fs_info->super_copy,
2057                                     total_bytes + 1);
2058         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2059
2060         if (seeding_dev) {
2061                 ret = init_first_rw_device(trans, root, device);
2062                 if (ret) {
2063                         btrfs_abort_transaction(trans, root, ret);
2064                         goto error_trans;
2065                 }
2066                 ret = btrfs_finish_sprout(trans, root);
2067                 if (ret) {
2068                         btrfs_abort_transaction(trans, root, ret);
2069                         goto error_trans;
2070                 }
2071         } else {
2072                 ret = btrfs_add_device(trans, root, device);
2073                 if (ret) {
2074                         btrfs_abort_transaction(trans, root, ret);
2075                         goto error_trans;
2076                 }
2077         }
2078
2079         /*
2080          * we've got more storage, clear any full flags on the space
2081          * infos
2082          */
2083         btrfs_clear_space_info_full(root->fs_info);
2084
2085         unlock_chunks(root);
2086         root->fs_info->num_tolerated_disk_barrier_failures =
2087                 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
2088         ret = btrfs_commit_transaction(trans, root);
2089
2090         if (seeding_dev) {
2091                 mutex_unlock(&uuid_mutex);
2092                 up_write(&sb->s_umount);
2093
2094                 if (ret) /* transaction commit */
2095                         return ret;
2096
2097                 ret = btrfs_relocate_sys_chunks(root);
2098                 if (ret < 0)
2099                         btrfs_error(root->fs_info, ret,
2100                                     "Failed to relocate sys chunks after "
2101                                     "device initialization. This can be fixed "
2102                                     "using the \"btrfs balance\" command.");
2103                 trans = btrfs_attach_transaction(root);
2104                 if (IS_ERR(trans)) {
2105                         if (PTR_ERR(trans) == -ENOENT)
2106                                 return 0;
2107                         return PTR_ERR(trans);
2108                 }
2109                 ret = btrfs_commit_transaction(trans, root);
2110         }
2111
2112         /* Update ctime/mtime for libblkid */
2113         update_dev_time(device_path);
2114         return ret;
2115
2116 error_trans:
2117         unlock_chunks(root);
2118         btrfs_end_transaction(trans, root);
2119         rcu_string_free(device->name);
2120         kfree(device);
2121 error:
2122         blkdev_put(bdev, FMODE_EXCL);
2123         if (seeding_dev) {
2124                 mutex_unlock(&uuid_mutex);
2125                 up_write(&sb->s_umount);
2126         }
2127         return ret;
2128 }
2129
2130 int btrfs_init_dev_replace_tgtdev(struct btrfs_root *root, char *device_path,
2131                                   struct btrfs_device **device_out)
2132 {
2133         struct request_queue *q;
2134         struct btrfs_device *device;
2135         struct block_device *bdev;
2136         struct btrfs_fs_info *fs_info = root->fs_info;
2137         struct list_head *devices;
2138         struct rcu_string *name;
2139         int ret = 0;
2140
2141         *device_out = NULL;
2142         if (fs_info->fs_devices->seeding)
2143                 return -EINVAL;
2144
2145         bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2146                                   fs_info->bdev_holder);
2147         if (IS_ERR(bdev))
2148                 return PTR_ERR(bdev);
2149
2150         filemap_write_and_wait(bdev->bd_inode->i_mapping);
2151
2152         devices = &fs_info->fs_devices->devices;
2153         list_for_each_entry(device, devices, dev_list) {
2154                 if (device->bdev == bdev) {
2155                         ret = -EEXIST;
2156                         goto error;
2157                 }
2158         }
2159
2160         device = kzalloc(sizeof(*device), GFP_NOFS);
2161         if (!device) {
2162                 ret = -ENOMEM;
2163                 goto error;
2164         }
2165
2166         name = rcu_string_strdup(device_path, GFP_NOFS);
2167         if (!name) {
2168                 kfree(device);
2169                 ret = -ENOMEM;
2170                 goto error;
2171         }
2172         rcu_assign_pointer(device->name, name);
2173
2174         q = bdev_get_queue(bdev);
2175         if (blk_queue_discard(q))
2176                 device->can_discard = 1;
2177         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2178         device->writeable = 1;
2179         device->work.func = pending_bios_fn;
2180         generate_random_uuid(device->uuid);
2181         device->devid = BTRFS_DEV_REPLACE_DEVID;
2182         spin_lock_init(&device->io_lock);
2183         device->generation = 0;
2184         device->io_width = root->sectorsize;
2185         device->io_align = root->sectorsize;
2186         device->sector_size = root->sectorsize;
2187         device->total_bytes = i_size_read(bdev->bd_inode);
2188         device->disk_total_bytes = device->total_bytes;
2189         device->dev_root = fs_info->dev_root;
2190         device->bdev = bdev;
2191         device->in_fs_metadata = 1;
2192         device->is_tgtdev_for_dev_replace = 1;
2193         device->mode = FMODE_EXCL;
2194         set_blocksize(device->bdev, 4096);
2195         device->fs_devices = fs_info->fs_devices;
2196         list_add(&device->dev_list, &fs_info->fs_devices->devices);
2197         fs_info->fs_devices->num_devices++;
2198         fs_info->fs_devices->open_devices++;
2199         if (device->can_discard)
2200                 fs_info->fs_devices->num_can_discard++;
2201         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2202
2203         *device_out = device;
2204         return ret;
2205
2206 error:
2207         blkdev_put(bdev, FMODE_EXCL);
2208         return ret;
2209 }
2210
2211 void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info *fs_info,
2212                                               struct btrfs_device *tgtdev)
2213 {
2214         WARN_ON(fs_info->fs_devices->rw_devices == 0);
2215         tgtdev->io_width = fs_info->dev_root->sectorsize;
2216         tgtdev->io_align = fs_info->dev_root->sectorsize;
2217         tgtdev->sector_size = fs_info->dev_root->sectorsize;
2218         tgtdev->dev_root = fs_info->dev_root;
2219         tgtdev->in_fs_metadata = 1;
2220 }
2221
2222 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2223                                         struct btrfs_device *device)
2224 {
2225         int ret;
2226         struct btrfs_path *path;
2227         struct btrfs_root *root;
2228         struct btrfs_dev_item *dev_item;
2229         struct extent_buffer *leaf;
2230         struct btrfs_key key;
2231
2232         root = device->dev_root->fs_info->chunk_root;
2233
2234         path = btrfs_alloc_path();
2235         if (!path)
2236                 return -ENOMEM;
2237
2238         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2239         key.type = BTRFS_DEV_ITEM_KEY;
2240         key.offset = device->devid;
2241
2242         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2243         if (ret < 0)
2244                 goto out;
2245
2246         if (ret > 0) {
2247                 ret = -ENOENT;
2248                 goto out;
2249         }
2250
2251         leaf = path->nodes[0];
2252         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2253
2254         btrfs_set_device_id(leaf, dev_item, device->devid);
2255         btrfs_set_device_type(leaf, dev_item, device->type);
2256         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2257         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2258         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2259         btrfs_set_device_total_bytes(leaf, dev_item, device->disk_total_bytes);
2260         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
2261         btrfs_mark_buffer_dirty(leaf);
2262
2263 out:
2264         btrfs_free_path(path);
2265         return ret;
2266 }
2267
2268 static int __btrfs_grow_device(struct btrfs_trans_handle *trans,
2269                       struct btrfs_device *device, u64 new_size)
2270 {
2271         struct btrfs_super_block *super_copy =
2272                 device->dev_root->fs_info->super_copy;
2273         u64 old_total = btrfs_super_total_bytes(super_copy);
2274         u64 diff = new_size - device->total_bytes;
2275
2276         if (!device->writeable)
2277                 return -EACCES;
2278         if (new_size <= device->total_bytes ||
2279             device->is_tgtdev_for_dev_replace)
2280                 return -EINVAL;
2281
2282         btrfs_set_super_total_bytes(super_copy, old_total + diff);
2283         device->fs_devices->total_rw_bytes += diff;
2284
2285         device->total_bytes = new_size;
2286         device->disk_total_bytes = new_size;
2287         btrfs_clear_space_info_full(device->dev_root->fs_info);
2288
2289         return btrfs_update_device(trans, device);
2290 }
2291
2292 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2293                       struct btrfs_device *device, u64 new_size)
2294 {
2295         int ret;
2296         lock_chunks(device->dev_root);
2297         ret = __btrfs_grow_device(trans, device, new_size);
2298         unlock_chunks(device->dev_root);
2299         return ret;
2300 }
2301
2302 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
2303                             struct btrfs_root *root,
2304                             u64 chunk_tree, u64 chunk_objectid,
2305                             u64 chunk_offset)
2306 {
2307         int ret;
2308         struct btrfs_path *path;
2309         struct btrfs_key key;
2310
2311         root = root->fs_info->chunk_root;
2312         path = btrfs_alloc_path();
2313         if (!path)
2314                 return -ENOMEM;
2315
2316         key.objectid = chunk_objectid;
2317         key.offset = chunk_offset;
2318         key.type = BTRFS_CHUNK_ITEM_KEY;
2319
2320         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2321         if (ret < 0)
2322                 goto out;
2323         else if (ret > 0) { /* Logic error or corruption */
2324                 btrfs_error(root->fs_info, -ENOENT,
2325                             "Failed lookup while freeing chunk.");
2326                 ret = -ENOENT;
2327                 goto out;
2328         }
2329
2330         ret = btrfs_del_item(trans, root, path);
2331         if (ret < 0)
2332                 btrfs_error(root->fs_info, ret,
2333                             "Failed to delete chunk item.");
2334 out:
2335         btrfs_free_path(path);
2336         return ret;
2337 }
2338
2339 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
2340                         chunk_offset)
2341 {
2342         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
2343         struct btrfs_disk_key *disk_key;
2344         struct btrfs_chunk *chunk;
2345         u8 *ptr;
2346         int ret = 0;
2347         u32 num_stripes;
2348         u32 array_size;
2349         u32 len = 0;
2350         u32 cur;
2351         struct btrfs_key key;
2352
2353         array_size = btrfs_super_sys_array_size(super_copy);
2354
2355         ptr = super_copy->sys_chunk_array;
2356         cur = 0;
2357
2358         while (cur < array_size) {
2359                 disk_key = (struct btrfs_disk_key *)ptr;
2360                 btrfs_disk_key_to_cpu(&key, disk_key);
2361
2362                 len = sizeof(*disk_key);
2363
2364                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2365                         chunk = (struct btrfs_chunk *)(ptr + len);
2366                         num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2367                         len += btrfs_chunk_item_size(num_stripes);
2368                 } else {
2369                         ret = -EIO;
2370                         break;
2371                 }
2372                 if (key.objectid == chunk_objectid &&
2373                     key.offset == chunk_offset) {
2374                         memmove(ptr, ptr + len, array_size - (cur + len));
2375                         array_size -= len;
2376                         btrfs_set_super_sys_array_size(super_copy, array_size);
2377                 } else {
2378                         ptr += len;
2379                         cur += len;
2380                 }
2381         }
2382         return ret;
2383 }
2384
2385 static int btrfs_relocate_chunk(struct btrfs_root *root,
2386                          u64 chunk_tree, u64 chunk_objectid,
2387                          u64 chunk_offset)
2388 {
2389         struct extent_map_tree *em_tree;
2390         struct btrfs_root *extent_root;
2391         struct btrfs_trans_handle *trans;
2392         struct extent_map *em;
2393         struct map_lookup *map;
2394         int ret;
2395         int i;
2396
2397         root = root->fs_info->chunk_root;
2398         extent_root = root->fs_info->extent_root;
2399         em_tree = &root->fs_info->mapping_tree.map_tree;
2400
2401         ret = btrfs_can_relocate(extent_root, chunk_offset);
2402         if (ret)
2403                 return -ENOSPC;
2404
2405         /* step one, relocate all the extents inside this chunk */
2406         ret = btrfs_relocate_block_group(extent_root, chunk_offset);
2407         if (ret)
2408                 return ret;
2409
2410         trans = btrfs_start_transaction(root, 0);
2411         if (IS_ERR(trans)) {
2412                 ret = PTR_ERR(trans);
2413                 btrfs_std_error(root->fs_info, ret);
2414                 return ret;
2415         }
2416
2417         lock_chunks(root);
2418
2419         /*
2420          * step two, delete the device extents and the
2421          * chunk tree entries
2422          */
2423         read_lock(&em_tree->lock);
2424         em = lookup_extent_mapping(em_tree, chunk_offset, 1);
2425         read_unlock(&em_tree->lock);
2426
2427         BUG_ON(!em || em->start > chunk_offset ||
2428                em->start + em->len < chunk_offset);
2429         map = (struct map_lookup *)em->bdev;
2430
2431         for (i = 0; i < map->num_stripes; i++) {
2432                 ret = btrfs_free_dev_extent(trans, map->stripes[i].dev,
2433                                             map->stripes[i].physical);
2434                 BUG_ON(ret);
2435
2436                 if (map->stripes[i].dev) {
2437                         ret = btrfs_update_device(trans, map->stripes[i].dev);
2438                         BUG_ON(ret);
2439                 }
2440         }
2441         ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
2442                                chunk_offset);
2443
2444         BUG_ON(ret);
2445
2446         trace_btrfs_chunk_free(root, map, chunk_offset, em->len);
2447
2448         if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2449                 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
2450                 BUG_ON(ret);
2451         }
2452
2453         ret = btrfs_remove_block_group(trans, extent_root, chunk_offset);
2454         BUG_ON(ret);
2455
2456         write_lock(&em_tree->lock);
2457         remove_extent_mapping(em_tree, em);
2458         write_unlock(&em_tree->lock);
2459
2460         kfree(map);
2461         em->bdev = NULL;
2462
2463         /* once for the tree */
2464         free_extent_map(em);
2465         /* once for us */
2466         free_extent_map(em);
2467
2468         unlock_chunks(root);
2469         btrfs_end_transaction(trans, root);
2470         return 0;
2471 }
2472
2473 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
2474 {
2475         struct btrfs_root *chunk_root = root->fs_info->chunk_root;
2476         struct btrfs_path *path;
2477         struct extent_buffer *leaf;
2478         struct btrfs_chunk *chunk;
2479         struct btrfs_key key;
2480         struct btrfs_key found_key;
2481         u64 chunk_tree = chunk_root->root_key.objectid;
2482         u64 chunk_type;
2483         bool retried = false;
2484         int failed = 0;
2485         int ret;
2486
2487         path = btrfs_alloc_path();
2488         if (!path)
2489                 return -ENOMEM;
2490
2491 again:
2492         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2493         key.offset = (u64)-1;
2494         key.type = BTRFS_CHUNK_ITEM_KEY;
2495
2496         while (1) {
2497                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2498                 if (ret < 0)
2499                         goto error;
2500                 BUG_ON(ret == 0); /* Corruption */
2501
2502                 ret = btrfs_previous_item(chunk_root, path, key.objectid,
2503                                           key.type);
2504                 if (ret < 0)
2505                         goto error;
2506                 if (ret > 0)
2507                         break;
2508
2509                 leaf = path->nodes[0];
2510                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2511
2512                 chunk = btrfs_item_ptr(leaf, path->slots[0],
2513                                        struct btrfs_chunk);
2514                 chunk_type = btrfs_chunk_type(leaf, chunk);
2515                 btrfs_release_path(path);
2516
2517                 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
2518                         ret = btrfs_relocate_chunk(chunk_root, chunk_tree,
2519                                                    found_key.objectid,
2520                                                    found_key.offset);
2521                         if (ret == -ENOSPC)
2522                                 failed++;
2523                         else if (ret)
2524                                 BUG();
2525                 }
2526
2527                 if (found_key.offset == 0)
2528                         break;
2529                 key.offset = found_key.offset - 1;
2530         }
2531         ret = 0;
2532         if (failed && !retried) {
2533                 failed = 0;
2534                 retried = true;
2535                 goto again;
2536         } else if (failed && retried) {
2537                 WARN_ON(1);
2538                 ret = -ENOSPC;
2539         }
2540 error:
2541         btrfs_free_path(path);
2542         return ret;
2543 }
2544
2545 static int insert_balance_item(struct btrfs_root *root,
2546                                struct btrfs_balance_control *bctl)
2547 {
2548         struct btrfs_trans_handle *trans;
2549         struct btrfs_balance_item *item;
2550         struct btrfs_disk_balance_args disk_bargs;
2551         struct btrfs_path *path;
2552         struct extent_buffer *leaf;
2553         struct btrfs_key key;
2554         int ret, err;
2555
2556         path = btrfs_alloc_path();
2557         if (!path)
2558                 return -ENOMEM;
2559
2560         trans = btrfs_start_transaction(root, 0);
2561         if (IS_ERR(trans)) {
2562                 btrfs_free_path(path);
2563                 return PTR_ERR(trans);
2564         }
2565
2566         key.objectid = BTRFS_BALANCE_OBJECTID;
2567         key.type = BTRFS_BALANCE_ITEM_KEY;
2568         key.offset = 0;
2569
2570         ret = btrfs_insert_empty_item(trans, root, path, &key,
2571                                       sizeof(*item));
2572         if (ret)
2573                 goto out;
2574
2575         leaf = path->nodes[0];
2576         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
2577
2578         memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
2579
2580         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
2581         btrfs_set_balance_data(leaf, item, &disk_bargs);
2582         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
2583         btrfs_set_balance_meta(leaf, item, &disk_bargs);
2584         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
2585         btrfs_set_balance_sys(leaf, item, &disk_bargs);
2586
2587         btrfs_set_balance_flags(leaf, item, bctl->flags);
2588
2589         btrfs_mark_buffer_dirty(leaf);
2590 out:
2591         btrfs_free_path(path);
2592         err = btrfs_commit_transaction(trans, root);
2593         if (err && !ret)
2594                 ret = err;
2595         return ret;
2596 }
2597
2598 static int del_balance_item(struct btrfs_root *root)
2599 {
2600         struct btrfs_trans_handle *trans;
2601         struct btrfs_path *path;
2602         struct btrfs_key key;
2603         int ret, err;
2604
2605         path = btrfs_alloc_path();
2606         if (!path)
2607                 return -ENOMEM;
2608
2609         trans = btrfs_start_transaction(root, 0);
2610         if (IS_ERR(trans)) {
2611                 btrfs_free_path(path);
2612                 return PTR_ERR(trans);
2613         }
2614
2615         key.objectid = BTRFS_BALANCE_OBJECTID;
2616         key.type = BTRFS_BALANCE_ITEM_KEY;
2617         key.offset = 0;
2618
2619         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2620         if (ret < 0)
2621                 goto out;
2622         if (ret > 0) {
2623                 ret = -ENOENT;
2624                 goto out;
2625         }
2626
2627         ret = btrfs_del_item(trans, root, path);
2628 out:
2629         btrfs_free_path(path);
2630         err = btrfs_commit_transaction(trans, root);
2631         if (err && !ret)
2632                 ret = err;
2633         return ret;
2634 }
2635
2636 /*
2637  * This is a heuristic used to reduce the number of chunks balanced on
2638  * resume after balance was interrupted.
2639  */
2640 static void update_balance_args(struct btrfs_balance_control *bctl)
2641 {
2642         /*
2643          * Turn on soft mode for chunk types that were being converted.
2644          */
2645         if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
2646                 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
2647         if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
2648                 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
2649         if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
2650                 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
2651
2652         /*
2653          * Turn on usage filter if is not already used.  The idea is
2654          * that chunks that we have already balanced should be
2655          * reasonably full.  Don't do it for chunks that are being
2656          * converted - that will keep us from relocating unconverted
2657          * (albeit full) chunks.
2658          */
2659         if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2660             !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2661                 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
2662                 bctl->data.usage = 90;
2663         }
2664         if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2665             !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2666                 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
2667                 bctl->sys.usage = 90;
2668         }
2669         if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2670             !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2671                 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
2672                 bctl->meta.usage = 90;
2673         }
2674 }
2675
2676 /*
2677  * Should be called with both balance and volume mutexes held to
2678  * serialize other volume operations (add_dev/rm_dev/resize) with
2679  * restriper.  Same goes for unset_balance_control.
2680  */
2681 static void set_balance_control(struct btrfs_balance_control *bctl)
2682 {
2683         struct btrfs_fs_info *fs_info = bctl->fs_info;
2684
2685         BUG_ON(fs_info->balance_ctl);
2686
2687         spin_lock(&fs_info->balance_lock);
2688         fs_info->balance_ctl = bctl;
2689         spin_unlock(&fs_info->balance_lock);
2690 }
2691
2692 static void unset_balance_control(struct btrfs_fs_info *fs_info)
2693 {
2694         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
2695
2696         BUG_ON(!fs_info->balance_ctl);
2697
2698         spin_lock(&fs_info->balance_lock);
2699         fs_info->balance_ctl = NULL;
2700         spin_unlock(&fs_info->balance_lock);
2701
2702         kfree(bctl);
2703 }
2704
2705 /*
2706  * Balance filters.  Return 1 if chunk should be filtered out
2707  * (should not be balanced).
2708  */
2709 static int chunk_profiles_filter(u64 chunk_type,
2710                                  struct btrfs_balance_args *bargs)
2711 {
2712         chunk_type = chunk_to_extended(chunk_type) &
2713                                 BTRFS_EXTENDED_PROFILE_MASK;
2714
2715         if (bargs->profiles & chunk_type)
2716                 return 0;
2717
2718         return 1;
2719 }
2720
2721 static int chunk_usage_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
2722                               struct btrfs_balance_args *bargs)
2723 {
2724         struct btrfs_block_group_cache *cache;
2725         u64 chunk_used, user_thresh;
2726         int ret = 1;
2727
2728         cache = btrfs_lookup_block_group(fs_info, chunk_offset);
2729         chunk_used = btrfs_block_group_used(&cache->item);
2730
2731         if (bargs->usage == 0)
2732                 user_thresh = 1;
2733         else if (bargs->usage > 100)
2734                 user_thresh = cache->key.offset;
2735         else
2736                 user_thresh = div_factor_fine(cache->key.offset,
2737                                               bargs->usage);
2738
2739         if (chunk_used < user_thresh)
2740                 ret = 0;
2741
2742         btrfs_put_block_group(cache);
2743         return ret;
2744 }
2745
2746 static int chunk_devid_filter(struct extent_buffer *leaf,
2747                               struct btrfs_chunk *chunk,
2748                               struct btrfs_balance_args *bargs)
2749 {
2750         struct btrfs_stripe *stripe;
2751         int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2752         int i;
2753
2754         for (i = 0; i < num_stripes; i++) {
2755                 stripe = btrfs_stripe_nr(chunk, i);
2756                 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
2757                         return 0;
2758         }
2759
2760         return 1;
2761 }
2762
2763 /* [pstart, pend) */
2764 static int chunk_drange_filter(struct extent_buffer *leaf,
2765                                struct btrfs_chunk *chunk,
2766                                u64 chunk_offset,
2767                                struct btrfs_balance_args *bargs)
2768 {
2769         struct btrfs_stripe *stripe;
2770         int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2771         u64 stripe_offset;
2772         u64 stripe_length;
2773         int factor;
2774         int i;
2775
2776         if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
2777                 return 0;
2778
2779         if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP |
2780              BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) {
2781                 factor = num_stripes / 2;
2782         } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) {
2783                 factor = num_stripes - 1;
2784         } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) {
2785                 factor = num_stripes - 2;
2786         } else {
2787                 factor = num_stripes;
2788         }
2789
2790         for (i = 0; i < num_stripes; i++) {
2791                 stripe = btrfs_stripe_nr(chunk, i);
2792                 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
2793                         continue;
2794
2795                 stripe_offset = btrfs_stripe_offset(leaf, stripe);
2796                 stripe_length = btrfs_chunk_length(leaf, chunk);
2797                 do_div(stripe_length, factor);
2798
2799                 if (stripe_offset < bargs->pend &&
2800                     stripe_offset + stripe_length > bargs->pstart)
2801                         return 0;
2802         }
2803
2804         return 1;
2805 }
2806
2807 /* [vstart, vend) */
2808 static int chunk_vrange_filter(struct extent_buffer *leaf,
2809                                struct btrfs_chunk *chunk,
2810                                u64 chunk_offset,
2811                                struct btrfs_balance_args *bargs)
2812 {
2813         if (chunk_offset < bargs->vend &&
2814             chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
2815                 /* at least part of the chunk is inside this vrange */
2816                 return 0;
2817
2818         return 1;
2819 }
2820
2821 static int chunk_soft_convert_filter(u64 chunk_type,
2822                                      struct btrfs_balance_args *bargs)
2823 {
2824         if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
2825                 return 0;
2826
2827         chunk_type = chunk_to_extended(chunk_type) &
2828                                 BTRFS_EXTENDED_PROFILE_MASK;
2829
2830         if (bargs->target == chunk_type)
2831                 return 1;
2832
2833         return 0;
2834 }
2835
2836 static int should_balance_chunk(struct btrfs_root *root,
2837                                 struct extent_buffer *leaf,
2838                                 struct btrfs_chunk *chunk, u64 chunk_offset)
2839 {
2840         struct btrfs_balance_control *bctl = root->fs_info->balance_ctl;
2841         struct btrfs_balance_args *bargs = NULL;
2842         u64 chunk_type = btrfs_chunk_type(leaf, chunk);
2843
2844         /* type filter */
2845         if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
2846               (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
2847                 return 0;
2848         }
2849
2850         if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
2851                 bargs = &bctl->data;
2852         else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
2853                 bargs = &bctl->sys;
2854         else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
2855                 bargs = &bctl->meta;
2856
2857         /* profiles filter */
2858         if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
2859             chunk_profiles_filter(chunk_type, bargs)) {
2860                 return 0;
2861         }
2862
2863         /* usage filter */
2864         if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
2865             chunk_usage_filter(bctl->fs_info, chunk_offset, bargs)) {
2866                 return 0;
2867         }
2868
2869         /* devid filter */
2870         if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
2871             chunk_devid_filter(leaf, chunk, bargs)) {
2872                 return 0;
2873         }
2874
2875         /* drange filter, makes sense only with devid filter */
2876         if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
2877             chunk_drange_filter(leaf, chunk, chunk_offset, bargs)) {
2878                 return 0;
2879         }
2880
2881         /* vrange filter */
2882         if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
2883             chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
2884                 return 0;
2885         }
2886
2887         /* soft profile changing mode */
2888         if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
2889             chunk_soft_convert_filter(chunk_type, bargs)) {
2890                 return 0;
2891         }
2892
2893         return 1;
2894 }
2895
2896 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
2897 {
2898         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
2899         struct btrfs_root *chunk_root = fs_info->chunk_root;
2900         struct btrfs_root *dev_root = fs_info->dev_root;
2901         struct list_head *devices;
2902         struct btrfs_device *device;
2903         u64 old_size;
2904         u64 size_to_free;
2905         struct btrfs_chunk *chunk;
2906         struct btrfs_path *path;
2907         struct btrfs_key key;
2908         struct btrfs_key found_key;
2909         struct btrfs_trans_handle *trans;
2910         struct extent_buffer *leaf;
2911         int slot;
2912         int ret;
2913         int enospc_errors = 0;
2914         bool counting = true;
2915
2916         /* step one make some room on all the devices */
2917         devices = &fs_info->fs_devices->devices;
2918         list_for_each_entry(device, devices, dev_list) {
2919                 old_size = device->total_bytes;
2920                 size_to_free = div_factor(old_size, 1);
2921                 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
2922                 if (!device->writeable ||
2923                     device->total_bytes - device->bytes_used > size_to_free ||
2924                     device->is_tgtdev_for_dev_replace)
2925                         continue;
2926
2927                 ret = btrfs_shrink_device(device, old_size - size_to_free);
2928                 if (ret == -ENOSPC)
2929                         break;
2930                 BUG_ON(ret);
2931
2932                 trans = btrfs_start_transaction(dev_root, 0);
2933                 BUG_ON(IS_ERR(trans));
2934
2935                 ret = btrfs_grow_device(trans, device, old_size);
2936                 BUG_ON(ret);
2937
2938                 btrfs_end_transaction(trans, dev_root);
2939         }
2940
2941         /* step two, relocate all the chunks */
2942         path = btrfs_alloc_path();
2943         if (!path) {
2944                 ret = -ENOMEM;
2945                 goto error;
2946         }
2947
2948         /* zero out stat counters */
2949         spin_lock(&fs_info->balance_lock);
2950         memset(&bctl->stat, 0, sizeof(bctl->stat));
2951         spin_unlock(&fs_info->balance_lock);
2952 again:
2953         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2954         key.offset = (u64)-1;
2955         key.type = BTRFS_CHUNK_ITEM_KEY;
2956
2957         while (1) {
2958                 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
2959                     atomic_read(&fs_info->balance_cancel_req)) {
2960                         ret = -ECANCELED;
2961                         goto error;
2962                 }
2963
2964                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2965                 if (ret < 0)
2966                         goto error;
2967
2968                 /*
2969                  * this shouldn't happen, it means the last relocate
2970                  * failed
2971                  */
2972                 if (ret == 0)
2973                         BUG(); /* FIXME break ? */
2974
2975                 ret = btrfs_previous_item(chunk_root, path, 0,
2976                                           BTRFS_CHUNK_ITEM_KEY);
2977                 if (ret) {
2978                         ret = 0;
2979                         break;
2980                 }
2981
2982                 leaf = path->nodes[0];
2983                 slot = path->slots[0];
2984                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2985
2986                 if (found_key.objectid != key.objectid)
2987                         break;
2988
2989                 /* chunk zero is special */
2990                 if (found_key.offset == 0)
2991                         break;
2992
2993                 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
2994
2995                 if (!counting) {
2996                         spin_lock(&fs_info->balance_lock);
2997                         bctl->stat.considered++;
2998                         spin_unlock(&fs_info->balance_lock);
2999                 }
3000
3001                 ret = should_balance_chunk(chunk_root, leaf, chunk,
3002                                            found_key.offset);
3003                 btrfs_release_path(path);
3004                 if (!ret)
3005                         goto loop;
3006
3007                 if (counting) {
3008                         spin_lock(&fs_info->balance_lock);
3009                         bctl->stat.expected++;
3010                         spin_unlock(&fs_info->balance_lock);
3011                         goto loop;
3012                 }
3013
3014                 ret = btrfs_relocate_chunk(chunk_root,
3015                                            chunk_root->root_key.objectid,
3016                                            found_key.objectid,
3017                                            found_key.offset);
3018                 if (ret && ret != -ENOSPC)
3019                         goto error;
3020                 if (ret == -ENOSPC) {
3021                         enospc_errors++;
3022                 } else {
3023                         spin_lock(&fs_info->balance_lock);
3024                         bctl->stat.completed++;
3025                         spin_unlock(&fs_info->balance_lock);
3026                 }
3027 loop:
3028                 key.offset = found_key.offset - 1;
3029         }
3030
3031         if (counting) {
3032                 btrfs_release_path(path);
3033                 counting = false;
3034                 goto again;
3035         }
3036 error:
3037         btrfs_free_path(path);
3038         if (enospc_errors) {
3039                 printk(KERN_INFO "btrfs: %d enospc errors during balance\n",
3040                        enospc_errors);
3041                 if (!ret)
3042                         ret = -ENOSPC;
3043         }
3044
3045         return ret;
3046 }
3047
3048 /**
3049  * alloc_profile_is_valid - see if a given profile is valid and reduced
3050  * @flags: profile to validate
3051  * @extended: if true @flags is treated as an extended profile
3052  */
3053 static int alloc_profile_is_valid(u64 flags, int extended)
3054 {
3055         u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
3056                                BTRFS_BLOCK_GROUP_PROFILE_MASK);
3057
3058         flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
3059
3060         /* 1) check that all other bits are zeroed */
3061         if (flags & ~mask)
3062                 return 0;
3063
3064         /* 2) see if profile is reduced */
3065         if (flags == 0)
3066                 return !extended; /* "0" is valid for usual profiles */
3067
3068         /* true if exactly one bit set */
3069         return (flags & (flags - 1)) == 0;
3070 }
3071
3072 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
3073 {
3074         /* cancel requested || normal exit path */
3075         return atomic_read(&fs_info->balance_cancel_req) ||
3076                 (atomic_read(&fs_info->balance_pause_req) == 0 &&
3077                  atomic_read(&fs_info->balance_cancel_req) == 0);
3078 }
3079
3080 static void __cancel_balance(struct btrfs_fs_info *fs_info)
3081 {
3082         int ret;
3083
3084         unset_balance_control(fs_info);
3085         ret = del_balance_item(fs_info->tree_root);
3086         if (ret)
3087                 btrfs_std_error(fs_info, ret);
3088
3089         atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3090 }
3091
3092 void update_ioctl_balance_args(struct btrfs_fs_info *fs_info, int lock,
3093                                struct btrfs_ioctl_balance_args *bargs);
3094
3095 /*
3096  * Should be called with both balance and volume mutexes held
3097  */
3098 int btrfs_balance(struct btrfs_balance_control *bctl,
3099                   struct btrfs_ioctl_balance_args *bargs)
3100 {
3101         struct btrfs_fs_info *fs_info = bctl->fs_info;
3102         u64 allowed;
3103         int mixed = 0;
3104         int ret;
3105         u64 num_devices;
3106         unsigned seq;
3107
3108         if (btrfs_fs_closing(fs_info) ||
3109             atomic_read(&fs_info->balance_pause_req) ||
3110             atomic_read(&fs_info->balance_cancel_req)) {
3111                 ret = -EINVAL;
3112                 goto out;
3113         }
3114
3115         allowed = btrfs_super_incompat_flags(fs_info->super_copy);
3116         if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
3117                 mixed = 1;
3118
3119         /*
3120          * In case of mixed groups both data and meta should be picked,
3121          * and identical options should be given for both of them.
3122          */
3123         allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
3124         if (mixed && (bctl->flags & allowed)) {
3125                 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
3126                     !(bctl->flags & BTRFS_BALANCE_METADATA) ||
3127                     memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
3128                         printk(KERN_ERR "btrfs: with mixed groups data and "
3129                                "metadata balance options must be the same\n");
3130                         ret = -EINVAL;
3131                         goto out;
3132                 }
3133         }
3134
3135         num_devices = fs_info->fs_devices->num_devices;
3136         btrfs_dev_replace_lock(&fs_info->dev_replace);
3137         if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
3138                 BUG_ON(num_devices < 1);
3139                 num_devices--;
3140         }
3141         btrfs_dev_replace_unlock(&fs_info->dev_replace);
3142         allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE;
3143         if (num_devices == 1)
3144                 allowed |= BTRFS_BLOCK_GROUP_DUP;
3145         else if (num_devices > 1)
3146                 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1);
3147         if (num_devices > 2)
3148                 allowed |= BTRFS_BLOCK_GROUP_RAID5;
3149         if (num_devices > 3)
3150                 allowed |= (BTRFS_BLOCK_GROUP_RAID10 |
3151                             BTRFS_BLOCK_GROUP_RAID6);
3152         if ((bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3153             (!alloc_profile_is_valid(bctl->data.target, 1) ||
3154              (bctl->data.target & ~allowed))) {
3155                 printk(KERN_ERR "btrfs: unable to start balance with target "
3156                        "data profile %llu\n",
3157                        (unsigned long long)bctl->data.target);
3158                 ret = -EINVAL;
3159                 goto out;
3160         }
3161         if ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3162             (!alloc_profile_is_valid(bctl->meta.target, 1) ||
3163              (bctl->meta.target & ~allowed))) {
3164                 printk(KERN_ERR "btrfs: unable to start balance with target "
3165                        "metadata profile %llu\n",
3166                        (unsigned long long)bctl->meta.target);
3167                 ret = -EINVAL;
3168                 goto out;
3169         }
3170         if ((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3171             (!alloc_profile_is_valid(bctl->sys.target, 1) ||
3172              (bctl->sys.target & ~allowed))) {
3173                 printk(KERN_ERR "btrfs: unable to start balance with target "
3174                        "system profile %llu\n",
3175                        (unsigned long long)bctl->sys.target);
3176                 ret = -EINVAL;
3177                 goto out;
3178         }
3179
3180         /* allow dup'ed data chunks only in mixed mode */
3181         if (!mixed && (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3182             (bctl->data.target & BTRFS_BLOCK_GROUP_DUP)) {
3183                 printk(KERN_ERR "btrfs: dup for data is not allowed\n");
3184                 ret = -EINVAL;
3185                 goto out;
3186         }
3187
3188         /* allow to reduce meta or sys integrity only if force set */
3189         allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3190                         BTRFS_BLOCK_GROUP_RAID10 |
3191                         BTRFS_BLOCK_GROUP_RAID5 |
3192                         BTRFS_BLOCK_GROUP_RAID6;
3193         do {
3194                 seq = read_seqbegin(&fs_info->profiles_lock);
3195
3196                 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3197                      (fs_info->avail_system_alloc_bits & allowed) &&
3198                      !(bctl->sys.target & allowed)) ||
3199                     ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3200                      (fs_info->avail_metadata_alloc_bits & allowed) &&
3201                      !(bctl->meta.target & allowed))) {
3202                         if (bctl->flags & BTRFS_BALANCE_FORCE) {
3203                                 printk(KERN_INFO "btrfs: force reducing metadata "
3204                                        "integrity\n");
3205                         } else {
3206                                 printk(KERN_ERR "btrfs: balance will reduce metadata "
3207                                        "integrity, use force if you want this\n");
3208                                 ret = -EINVAL;
3209                                 goto out;
3210                         }
3211                 }
3212         } while (read_seqretry(&fs_info->profiles_lock, seq));
3213
3214         if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3215                 int num_tolerated_disk_barrier_failures;
3216                 u64 target = bctl->sys.target;
3217
3218                 num_tolerated_disk_barrier_failures =
3219                         btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3220                 if (num_tolerated_disk_barrier_failures > 0 &&
3221                     (target &
3222                      (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID0 |
3223                       BTRFS_AVAIL_ALLOC_BIT_SINGLE)))
3224                         num_tolerated_disk_barrier_failures = 0;
3225                 else if (num_tolerated_disk_barrier_failures > 1 &&
3226                          (target &
3227                           (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)))
3228                         num_tolerated_disk_barrier_failures = 1;
3229
3230                 fs_info->num_tolerated_disk_barrier_failures =
3231                         num_tolerated_disk_barrier_failures;
3232         }
3233
3234         ret = insert_balance_item(fs_info->tree_root, bctl);
3235         if (ret && ret != -EEXIST)
3236                 goto out;
3237
3238         if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
3239                 BUG_ON(ret == -EEXIST);
3240                 set_balance_control(bctl);
3241         } else {
3242                 BUG_ON(ret != -EEXIST);
3243                 spin_lock(&fs_info->balance_lock);
3244                 update_balance_args(bctl);
3245                 spin_unlock(&fs_info->balance_lock);
3246         }
3247
3248         atomic_inc(&fs_info->balance_running);
3249         mutex_unlock(&fs_info->balance_mutex);
3250
3251         ret = __btrfs_balance(fs_info);
3252
3253         mutex_lock(&fs_info->balance_mutex);
3254         atomic_dec(&fs_info->balance_running);
3255
3256         if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3257                 fs_info->num_tolerated_disk_barrier_failures =
3258                         btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3259         }
3260
3261         if (bargs) {
3262                 memset(bargs, 0, sizeof(*bargs));
3263                 update_ioctl_balance_args(fs_info, 0, bargs);
3264         }
3265
3266         if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
3267             balance_need_close(fs_info)) {
3268                 __cancel_balance(fs_info);
3269         }
3270
3271         wake_up(&fs_info->balance_wait_q);
3272
3273         return ret;
3274 out:
3275         if (bctl->flags & BTRFS_BALANCE_RESUME)
3276                 __cancel_balance(fs_info);
3277         else {
3278                 kfree(bctl);
3279                 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3280         }
3281         return ret;
3282 }
3283
3284 static int balance_kthread(void *data)
3285 {
3286         struct btrfs_fs_info *fs_info = data;
3287         int ret = 0;
3288
3289         mutex_lock(&fs_info->volume_mutex);
3290         mutex_lock(&fs_info->balance_mutex);
3291
3292         if (fs_info->balance_ctl) {
3293                 printk(KERN_INFO "btrfs: continuing balance\n");
3294                 ret = btrfs_balance(fs_info->balance_ctl, NULL);
3295         }
3296
3297         mutex_unlock(&fs_info->balance_mutex);
3298         mutex_unlock(&fs_info->volume_mutex);
3299
3300         return ret;
3301 }
3302
3303 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
3304 {
3305         struct task_struct *tsk;
3306
3307         spin_lock(&fs_info->balance_lock);
3308         if (!fs_info->balance_ctl) {
3309                 spin_unlock(&fs_info->balance_lock);
3310                 return 0;
3311         }
3312         spin_unlock(&fs_info->balance_lock);
3313
3314         if (btrfs_test_opt(fs_info->tree_root, SKIP_BALANCE)) {
3315                 printk(KERN_INFO "btrfs: force skipping balance\n");
3316                 return 0;
3317         }
3318
3319         tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
3320         if (IS_ERR(tsk))
3321                 return PTR_ERR(tsk);
3322
3323         return 0;
3324 }
3325
3326 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
3327 {
3328         struct btrfs_balance_control *bctl;
3329         struct btrfs_balance_item *item;
3330         struct btrfs_disk_balance_args disk_bargs;
3331         struct btrfs_path *path;
3332         struct extent_buffer *leaf;
3333         struct btrfs_key key;
3334         int ret;
3335
3336         path = btrfs_alloc_path();
3337         if (!path)
3338                 return -ENOMEM;
3339
3340         key.objectid = BTRFS_BALANCE_OBJECTID;
3341         key.type = BTRFS_BALANCE_ITEM_KEY;
3342         key.offset = 0;
3343
3344         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
3345         if (ret < 0)
3346                 goto out;
3347         if (ret > 0) { /* ret = -ENOENT; */
3348                 ret = 0;
3349                 goto out;
3350         }
3351
3352         bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
3353         if (!bctl) {
3354                 ret = -ENOMEM;
3355                 goto out;
3356         }
3357
3358         leaf = path->nodes[0];
3359         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3360
3361         bctl->fs_info = fs_info;
3362         bctl->flags = btrfs_balance_flags(leaf, item);
3363         bctl->flags |= BTRFS_BALANCE_RESUME;
3364
3365         btrfs_balance_data(leaf, item, &disk_bargs);
3366         btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
3367         btrfs_balance_meta(leaf, item, &disk_bargs);
3368         btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
3369         btrfs_balance_sys(leaf, item, &disk_bargs);
3370         btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
3371
3372         WARN_ON(atomic_xchg(&fs_info->mutually_exclusive_operation_running, 1));
3373
3374         mutex_lock(&fs_info->volume_mutex);
3375         mutex_lock(&fs_info->balance_mutex);
3376
3377         set_balance_control(bctl);
3378
3379         mutex_unlock(&fs_info->balance_mutex);
3380         mutex_unlock(&fs_info->volume_mutex);
3381 out:
3382         btrfs_free_path(path);
3383         return ret;
3384 }
3385
3386 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
3387 {
3388         int ret = 0;
3389
3390         mutex_lock(&fs_info->balance_mutex);
3391         if (!fs_info->balance_ctl) {
3392                 mutex_unlock(&fs_info->balance_mutex);
3393                 return -ENOTCONN;
3394         }
3395
3396         if (atomic_read(&fs_info->balance_running)) {
3397                 atomic_inc(&fs_info->balance_pause_req);
3398                 mutex_unlock(&fs_info->balance_mutex);
3399
3400                 wait_event(fs_info->balance_wait_q,
3401                            atomic_read(&fs_info->balance_running) == 0);
3402
3403                 mutex_lock(&fs_info->balance_mutex);
3404                 /* we are good with balance_ctl ripped off from under us */
3405                 BUG_ON(atomic_read(&fs_info->balance_running));
3406                 atomic_dec(&fs_info->balance_pause_req);
3407         } else {
3408                 ret = -ENOTCONN;
3409         }
3410
3411         mutex_unlock(&fs_info->balance_mutex);
3412         return ret;
3413 }
3414
3415 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
3416 {
3417         mutex_lock(&fs_info->balance_mutex);
3418         if (!fs_info->balance_ctl) {
3419                 mutex_unlock(&fs_info->balance_mutex);
3420                 return -ENOTCONN;
3421         }
3422
3423         atomic_inc(&fs_info->balance_cancel_req);
3424         /*
3425          * if we are running just wait and return, balance item is
3426          * deleted in btrfs_balance in this case
3427          */
3428         if (atomic_read(&fs_info->balance_running)) {
3429                 mutex_unlock(&fs_info->balance_mutex);
3430                 wait_event(fs_info->balance_wait_q,
3431                            atomic_read(&fs_info->balance_running) == 0);
3432                 mutex_lock(&fs_info->balance_mutex);
3433         } else {
3434                 /* __cancel_balance needs volume_mutex */
3435                 mutex_unlock(&fs_info->balance_mutex);
3436                 mutex_lock(&fs_info->volume_mutex);
3437                 mutex_lock(&fs_info->balance_mutex);
3438
3439                 if (fs_info->balance_ctl)
3440                         __cancel_balance(fs_info);
3441
3442                 mutex_unlock(&fs_info->volume_mutex);
3443         }
3444
3445         BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running));
3446         atomic_dec(&fs_info->balance_cancel_req);
3447         mutex_unlock(&fs_info->balance_mutex);
3448         return 0;
3449 }
3450
3451 /*
3452  * shrinking a device means finding all of the device extents past
3453  * the new size, and then following the back refs to the chunks.
3454  * The chunk relocation code actually frees the device extent
3455  */
3456 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
3457 {
3458         struct btrfs_trans_handle *trans;
3459         struct btrfs_root *root = device->dev_root;
3460         struct btrfs_dev_extent *dev_extent = NULL;
3461         struct btrfs_path *path;
3462         u64 length;
3463         u64 chunk_tree;
3464         u64 chunk_objectid;
3465         u64 chunk_offset;
3466         int ret;
3467         int slot;
3468         int failed = 0;
3469         bool retried = false;
3470         struct extent_buffer *l;
3471         struct btrfs_key key;
3472         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
3473         u64 old_total = btrfs_super_total_bytes(super_copy);
3474         u64 old_size = device->total_bytes;
3475         u64 diff = device->total_bytes - new_size;
3476
3477         if (device->is_tgtdev_for_dev_replace)
3478                 return -EINVAL;
3479
3480         path = btrfs_alloc_path();
3481         if (!path)
3482                 return -ENOMEM;
3483
3484         path->reada = 2;
3485
3486         lock_chunks(root);
3487
3488         device->total_bytes = new_size;
3489         if (device->writeable) {
3490                 device->fs_devices->total_rw_bytes -= diff;
3491                 spin_lock(&root->fs_info->free_chunk_lock);
3492                 root->fs_info->free_chunk_space -= diff;
3493                 spin_unlock(&root->fs_info->free_chunk_lock);
3494         }
3495         unlock_chunks(root);
3496
3497 again:
3498         key.objectid = device->devid;
3499         key.offset = (u64)-1;
3500         key.type = BTRFS_DEV_EXTENT_KEY;
3501
3502         do {
3503                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3504                 if (ret < 0)
3505                         goto done;
3506
3507                 ret = btrfs_previous_item(root, path, 0, key.type);
3508                 if (ret < 0)
3509                         goto done;
3510                 if (ret) {
3511                         ret = 0;
3512                         btrfs_release_path(path);
3513                         break;
3514                 }
3515
3516                 l = path->nodes[0];
3517                 slot = path->slots[0];
3518                 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
3519
3520                 if (key.objectid != device->devid) {
3521                         btrfs_release_path(path);
3522                         break;
3523                 }
3524
3525                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
3526                 length = btrfs_dev_extent_length(l, dev_extent);
3527
3528                 if (key.offset + length <= new_size) {
3529                         btrfs_release_path(path);
3530                         break;
3531                 }
3532
3533                 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
3534                 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
3535                 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
3536                 btrfs_release_path(path);
3537
3538                 ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
3539                                            chunk_offset);
3540                 if (ret && ret != -ENOSPC)
3541                         goto done;
3542                 if (ret == -ENOSPC)
3543                         failed++;
3544         } while (key.offset-- > 0);
3545
3546         if (failed && !retried) {
3547                 failed = 0;
3548                 retried = true;
3549                 goto again;
3550         } else if (failed && retried) {
3551                 ret = -ENOSPC;
3552                 lock_chunks(root);
3553
3554                 device->total_bytes = old_size;
3555                 if (device->writeable)
3556                         device->fs_devices->total_rw_bytes += diff;
3557                 spin_lock(&root->fs_info->free_chunk_lock);
3558                 root->fs_info->free_chunk_space += diff;
3559                 spin_unlock(&root->fs_info->free_chunk_lock);
3560                 unlock_chunks(root);
3561                 goto done;
3562         }
3563
3564         /* Shrinking succeeded, else we would be at "done". */
3565         trans = btrfs_start_transaction(root, 0);
3566         if (IS_ERR(trans)) {
3567                 ret = PTR_ERR(trans);
3568                 goto done;
3569         }
3570
3571         lock_chunks(root);
3572
3573         device->disk_total_bytes = new_size;
3574         /* Now btrfs_update_device() will change the on-disk size. */
3575         ret = btrfs_update_device(trans, device);
3576         if (ret) {
3577                 unlock_chunks(root);
3578                 btrfs_end_transaction(trans, root);
3579                 goto done;
3580         }
3581         WARN_ON(diff > old_total);
3582         btrfs_set_super_total_bytes(super_copy, old_total - diff);
3583         unlock_chunks(root);
3584         btrfs_end_transaction(trans, root);
3585 done:
3586         btrfs_free_path(path);
3587         return ret;
3588 }
3589
3590 static int btrfs_add_system_chunk(struct btrfs_root *root,
3591                            struct btrfs_key *key,
3592                            struct btrfs_chunk *chunk, int item_size)
3593 {
3594         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
3595         struct btrfs_disk_key disk_key;
3596         u32 array_size;
3597         u8 *ptr;
3598
3599         array_size = btrfs_super_sys_array_size(super_copy);
3600         if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
3601                 return -EFBIG;
3602
3603         ptr = super_copy->sys_chunk_array + array_size;
3604         btrfs_cpu_key_to_disk(&disk_key, key);
3605         memcpy(ptr, &disk_key, sizeof(disk_key));
3606         ptr += sizeof(disk_key);
3607         memcpy(ptr, chunk, item_size);
3608         item_size += sizeof(disk_key);
3609         btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
3610         return 0;
3611 }
3612
3613 /*
3614  * sort the devices in descending order by max_avail, total_avail
3615  */
3616 static int btrfs_cmp_device_info(const void *a, const void *b)
3617 {
3618         const struct btrfs_device_info *di_a = a;
3619         const struct btrfs_device_info *di_b = b;
3620
3621         if (di_a->max_avail > di_b->max_avail)
3622                 return -1;
3623         if (di_a->max_avail < di_b->max_avail)
3624                 return 1;
3625         if (di_a->total_avail > di_b->total_avail)
3626                 return -1;
3627         if (di_a->total_avail < di_b->total_avail)
3628                 return 1;
3629         return 0;
3630 }
3631
3632 static struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = {
3633         [BTRFS_RAID_RAID10] = {
3634                 .sub_stripes    = 2,
3635                 .dev_stripes    = 1,
3636                 .devs_max       = 0,    /* 0 == as many as possible */
3637                 .devs_min       = 4,
3638                 .devs_increment = 2,
3639                 .ncopies        = 2,
3640         },
3641         [BTRFS_RAID_RAID1] = {
3642                 .sub_stripes    = 1,
3643                 .dev_stripes    = 1,
3644                 .devs_max       = 2,
3645                 .devs_min       = 2,
3646                 .devs_increment = 2,
3647                 .ncopies        = 2,
3648         },
3649         [BTRFS_RAID_DUP] = {
3650                 .sub_stripes    = 1,
3651                 .dev_stripes    = 2,
3652                 .devs_max       = 1,
3653                 .devs_min       = 1,
3654                 .devs_increment = 1,
3655                 .ncopies        = 2,
3656         },
3657         [BTRFS_RAID_RAID0] = {
3658                 .sub_stripes    = 1,
3659                 .dev_stripes    = 1,
3660                 .devs_max       = 0,
3661                 .devs_min       = 2,
3662                 .devs_increment = 1,
3663                 .ncopies        = 1,
3664         },
3665         [BTRFS_RAID_SINGLE] = {
3666                 .sub_stripes    = 1,
3667                 .dev_stripes    = 1,
3668                 .devs_max       = 1,
3669                 .devs_min       = 1,
3670                 .devs_increment = 1,
3671                 .ncopies        = 1,
3672         },
3673         [BTRFS_RAID_RAID5] = {
3674                 .sub_stripes    = 1,
3675                 .dev_stripes    = 1,
3676                 .devs_max       = 0,
3677                 .devs_min       = 2,
3678                 .devs_increment = 1,
3679                 .ncopies        = 2,
3680         },
3681         [BTRFS_RAID_RAID6] = {
3682                 .sub_stripes    = 1,
3683                 .dev_stripes    = 1,
3684                 .devs_max       = 0,
3685                 .devs_min       = 3,
3686                 .devs_increment = 1,
3687                 .ncopies        = 3,
3688         },
3689 };
3690
3691 static u32 find_raid56_stripe_len(u32 data_devices, u32 dev_stripe_target)
3692 {
3693         /* TODO allow them to set a preferred stripe size */
3694         return 64 * 1024;
3695 }
3696
3697 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
3698 {
3699         if (!(type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)))
3700                 return;
3701
3702         btrfs_set_fs_incompat(info, RAID56);
3703 }
3704
3705 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
3706                                struct btrfs_root *extent_root,
3707                                struct map_lookup **map_ret,
3708                                u64 *num_bytes_out, u64 *stripe_size_out,
3709                                u64 start, u64 type)
3710 {
3711         struct btrfs_fs_info *info = extent_root->fs_info;
3712         struct btrfs_fs_devices *fs_devices = info->fs_devices;
3713         struct list_head *cur;
3714         struct map_lookup *map = NULL;
3715         struct extent_map_tree *em_tree;
3716         struct extent_map *em;
3717         struct btrfs_device_info *devices_info = NULL;
3718         u64 total_avail;
3719         int num_stripes;        /* total number of stripes to allocate */
3720         int data_stripes;       /* number of stripes that count for
3721                                    block group size */
3722         int sub_stripes;        /* sub_stripes info for map */
3723         int dev_stripes;        /* stripes per dev */
3724         int devs_max;           /* max devs to use */
3725         int devs_min;           /* min devs needed */
3726         int devs_increment;     /* ndevs has to be a multiple of this */
3727         int ncopies;            /* how many copies to data has */
3728         int ret;
3729         u64 max_stripe_size;
3730         u64 max_chunk_size;
3731         u64 stripe_size;
3732         u64 num_bytes;
3733         u64 raid_stripe_len = BTRFS_STRIPE_LEN;
3734         int ndevs;
3735         int i;
3736         int j;
3737         int index;
3738
3739         BUG_ON(!alloc_profile_is_valid(type, 0));
3740
3741         if (list_empty(&fs_devices->alloc_list))
3742                 return -ENOSPC;
3743
3744         index = __get_raid_index(type);
3745
3746         sub_stripes = btrfs_raid_array[index].sub_stripes;
3747         dev_stripes = btrfs_raid_array[index].dev_stripes;
3748         devs_max = btrfs_raid_array[index].devs_max;
3749         devs_min = btrfs_raid_array[index].devs_min;
3750         devs_increment = btrfs_raid_array[index].devs_increment;
3751         ncopies = btrfs_raid_array[index].ncopies;
3752
3753         if (type & BTRFS_BLOCK_GROUP_DATA) {
3754                 max_stripe_size = 1024 * 1024 * 1024;
3755                 max_chunk_size = 10 * max_stripe_size;
3756         } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
3757                 /* for larger filesystems, use larger metadata chunks */
3758                 if (fs_devices->total_rw_bytes > 50ULL * 1024 * 1024 * 1024)
3759                         max_stripe_size = 1024 * 1024 * 1024;
3760                 else
3761                         max_stripe_size = 256 * 1024 * 1024;
3762                 max_chunk_size = max_stripe_size;
3763         } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
3764                 max_stripe_size = 32 * 1024 * 1024;
3765                 max_chunk_size = 2 * max_stripe_size;
3766         } else {
3767                 printk(KERN_ERR "btrfs: invalid chunk type 0x%llx requested\n",
3768                        type);
3769                 BUG_ON(1);
3770         }
3771
3772         /* we don't want a chunk larger than 10% of writeable space */
3773         max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
3774                              max_chunk_size);
3775
3776         devices_info = kzalloc(sizeof(*devices_info) * fs_devices->rw_devices,
3777                                GFP_NOFS);
3778         if (!devices_info)
3779                 return -ENOMEM;
3780
3781         cur = fs_devices->alloc_list.next;
3782
3783         /*
3784          * in the first pass through the devices list, we gather information
3785          * about the available holes on each device.
3786          */
3787         ndevs = 0;
3788         while (cur != &fs_devices->alloc_list) {
3789                 struct btrfs_device *device;
3790                 u64 max_avail;
3791                 u64 dev_offset;
3792
3793                 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
3794
3795                 cur = cur->next;
3796
3797                 if (!device->writeable) {
3798                         WARN(1, KERN_ERR
3799                                "btrfs: read-only device in alloc_list\n");
3800                         continue;
3801                 }
3802
3803                 if (!device->in_fs_metadata ||
3804                     device->is_tgtdev_for_dev_replace)
3805                         continue;
3806
3807                 if (device->total_bytes > device->bytes_used)
3808                         total_avail = device->total_bytes - device->bytes_used;
3809                 else
3810                         total_avail = 0;
3811
3812                 /* If there is no space on this device, skip it. */
3813                 if (total_avail == 0)
3814                         continue;
3815
3816                 ret = find_free_dev_extent(device,
3817                                            max_stripe_size * dev_stripes,
3818                                            &dev_offset, &max_avail);
3819                 if (ret && ret != -ENOSPC)
3820                         goto error;
3821
3822                 if (ret == 0)
3823                         max_avail = max_stripe_size * dev_stripes;
3824
3825                 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes)
3826                         continue;
3827
3828                 if (ndevs == fs_devices->rw_devices) {
3829                         WARN(1, "%s: found more than %llu devices\n",
3830                              __func__, fs_devices->rw_devices);
3831                         break;
3832                 }
3833                 devices_info[ndevs].dev_offset = dev_offset;
3834                 devices_info[ndevs].max_avail = max_avail;
3835                 devices_info[ndevs].total_avail = total_avail;
3836                 devices_info[ndevs].dev = device;
3837                 ++ndevs;
3838         }
3839
3840         /*
3841          * now sort the devices by hole size / available space
3842          */
3843         sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
3844              btrfs_cmp_device_info, NULL);
3845
3846         /* round down to number of usable stripes */
3847         ndevs -= ndevs % devs_increment;
3848
3849         if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) {
3850                 ret = -ENOSPC;
3851                 goto error;
3852         }
3853
3854         if (devs_max && ndevs > devs_max)
3855                 ndevs = devs_max;
3856         /*
3857          * the primary goal is to maximize the number of stripes, so use as many
3858          * devices as possible, even if the stripes are not maximum sized.
3859          */
3860         stripe_size = devices_info[ndevs-1].max_avail;
3861         num_stripes = ndevs * dev_stripes;
3862
3863         /*
3864          * this will have to be fixed for RAID1 and RAID10 over
3865          * more drives
3866          */
3867         data_stripes = num_stripes / ncopies;
3868
3869         if (type & BTRFS_BLOCK_GROUP_RAID5) {
3870                 raid_stripe_len = find_raid56_stripe_len(ndevs - 1,
3871                                  btrfs_super_stripesize(info->super_copy));
3872                 data_stripes = num_stripes - 1;
3873         }
3874         if (type & BTRFS_BLOCK_GROUP_RAID6) {
3875                 raid_stripe_len = find_raid56_stripe_len(ndevs - 2,
3876                                  btrfs_super_stripesize(info->super_copy));
3877                 data_stripes = num_stripes - 2;
3878         }
3879
3880         /*
3881          * Use the number of data stripes to figure out how big this chunk
3882          * is really going to be in terms of logical address space,
3883          * and compare that answer with the max chunk size
3884          */
3885         if (stripe_size * data_stripes > max_chunk_size) {
3886                 u64 mask = (1ULL << 24) - 1;
3887                 stripe_size = max_chunk_size;
3888                 do_div(stripe_size, data_stripes);
3889
3890                 /* bump the answer up to a 16MB boundary */
3891                 stripe_size = (stripe_size + mask) & ~mask;
3892
3893                 /* but don't go higher than the limits we found
3894                  * while searching for free extents
3895                  */
3896                 if (stripe_size > devices_info[ndevs-1].max_avail)
3897                         stripe_size = devices_info[ndevs-1].max_avail;
3898         }
3899
3900         do_div(stripe_size, dev_stripes);
3901
3902         /* align to BTRFS_STRIPE_LEN */
3903         do_div(stripe_size, raid_stripe_len);
3904         stripe_size *= raid_stripe_len;
3905
3906         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
3907         if (!map) {
3908                 ret = -ENOMEM;
3909                 goto error;
3910         }
3911         map->num_stripes = num_stripes;
3912
3913         for (i = 0; i < ndevs; ++i) {
3914                 for (j = 0; j < dev_stripes; ++j) {
3915                         int s = i * dev_stripes + j;
3916                         map->stripes[s].dev = devices_info[i].dev;
3917                         map->stripes[s].physical = devices_info[i].dev_offset +
3918                                                    j * stripe_size;
3919                 }
3920         }
3921         map->sector_size = extent_root->sectorsize;
3922         map->stripe_len = raid_stripe_len;
3923         map->io_align = raid_stripe_len;
3924         map->io_width = raid_stripe_len;
3925         map->type = type;
3926         map->sub_stripes = sub_stripes;
3927
3928         *map_ret = map;
3929         num_bytes = stripe_size * data_stripes;
3930
3931         *stripe_size_out = stripe_size;
3932         *num_bytes_out = num_bytes;
3933
3934         trace_btrfs_chunk_alloc(info->chunk_root, map, start, num_bytes);
3935
3936         em = alloc_extent_map();
3937         if (!em) {
3938                 ret = -ENOMEM;
3939                 goto error;
3940         }
3941         em->bdev = (struct block_device *)map;
3942         em->start = start;
3943         em->len = num_bytes;
3944         em->block_start = 0;
3945         em->block_len = em->len;
3946
3947         em_tree = &extent_root->fs_info->mapping_tree.map_tree;
3948         write_lock(&em_tree->lock);
3949         ret = add_extent_mapping(em_tree, em, 0);
3950         write_unlock(&em_tree->lock);
3951         if (ret) {
3952                 free_extent_map(em);
3953                 goto error;
3954         }
3955
3956         for (i = 0; i < map->num_stripes; ++i) {
3957                 struct btrfs_device *device;
3958                 u64 dev_offset;
3959
3960                 device = map->stripes[i].dev;
3961                 dev_offset = map->stripes[i].physical;
3962
3963                 ret = btrfs_alloc_dev_extent(trans, device,
3964                                 info->chunk_root->root_key.objectid,
3965                                 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
3966                                 start, dev_offset, stripe_size);
3967                 if (ret)
3968                         goto error_dev_extent;
3969         }
3970
3971         ret = btrfs_make_block_group(trans, extent_root, 0, type,
3972                                      BTRFS_FIRST_CHUNK_TREE_OBJECTID,
3973                                      start, num_bytes);
3974         if (ret) {
3975                 i = map->num_stripes - 1;
3976                 goto error_dev_extent;
3977         }
3978
3979         free_extent_map(em);
3980         check_raid56_incompat_flag(extent_root->fs_info, type);
3981
3982         kfree(devices_info);
3983         return 0;
3984
3985 error_dev_extent:
3986         for (; i >= 0; i--) {
3987                 struct btrfs_device *device;
3988                 int err;
3989
3990                 device = map->stripes[i].dev;
3991                 err = btrfs_free_dev_extent(trans, device, start);
3992                 if (err) {
3993                         btrfs_abort_transaction(trans, extent_root, err);
3994                         break;
3995                 }
3996         }
3997         write_lock(&em_tree->lock);
3998         remove_extent_mapping(em_tree, em);
3999         write_unlock(&em_tree->lock);
4000
4001         /* One for our allocation */
4002         free_extent_map(em);
4003         /* One for the tree reference */
4004         free_extent_map(em);
4005 error:
4006         kfree(map);
4007         kfree(devices_info);
4008         return ret;
4009 }
4010
4011 static int __finish_chunk_alloc(struct btrfs_trans_handle *trans,
4012                                 struct btrfs_root *extent_root,
4013                                 struct map_lookup *map, u64 chunk_offset,
4014                                 u64 chunk_size, u64 stripe_size)
4015 {
4016         u64 dev_offset;
4017         struct btrfs_key key;
4018         struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
4019         struct btrfs_device *device;
4020         struct btrfs_chunk *chunk;
4021         struct btrfs_stripe *stripe;
4022         size_t item_size = btrfs_chunk_item_size(map->num_stripes);
4023         int index = 0;
4024         int ret;
4025
4026         chunk = kzalloc(item_size, GFP_NOFS);
4027         if (!chunk)
4028                 return -ENOMEM;
4029
4030         index = 0;
4031         while (index < map->num_stripes) {
4032                 device = map->stripes[index].dev;
4033                 device->bytes_used += stripe_size;
4034                 ret = btrfs_update_device(trans, device);
4035                 if (ret)
4036                         goto out_free;
4037                 index++;
4038         }
4039
4040         spin_lock(&extent_root->fs_info->free_chunk_lock);
4041         extent_root->fs_info->free_chunk_space -= (stripe_size *
4042                                                    map->num_stripes);
4043         spin_unlock(&extent_root->fs_info->free_chunk_lock);
4044
4045         index = 0;
4046         stripe = &chunk->stripe;
4047         while (index < map->num_stripes) {
4048                 device = map->stripes[index].dev;
4049                 dev_offset = map->stripes[index].physical;
4050
4051                 btrfs_set_stack_stripe_devid(stripe, device->devid);
4052                 btrfs_set_stack_stripe_offset(stripe, dev_offset);
4053                 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
4054                 stripe++;
4055                 index++;
4056         }
4057
4058         btrfs_set_stack_chunk_length(chunk, chunk_size);
4059         btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
4060         btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
4061         btrfs_set_stack_chunk_type(chunk, map->type);
4062         btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
4063         btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
4064         btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
4065         btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
4066         btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
4067
4068         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
4069         key.type = BTRFS_CHUNK_ITEM_KEY;
4070         key.offset = chunk_offset;
4071
4072         ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
4073
4074         if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
4075                 /*
4076                  * TODO: Cleanup of inserted chunk root in case of
4077                  * failure.
4078                  */
4079                 ret = btrfs_add_system_chunk(chunk_root, &key, chunk,
4080                                              item_size);
4081         }
4082
4083 out_free:
4084         kfree(chunk);
4085         return ret;
4086 }
4087
4088 /*
4089  * Chunk allocation falls into two parts. The first part does works
4090  * that make the new allocated chunk useable, but not do any operation
4091  * that modifies the chunk tree. The second part does the works that
4092  * require modifying the chunk tree. This division is important for the
4093  * bootstrap process of adding storage to a seed btrfs.
4094  */
4095 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4096                       struct btrfs_root *extent_root, u64 type)
4097 {
4098         u64 chunk_offset;
4099         u64 chunk_size;
4100         u64 stripe_size;
4101         struct map_lookup *map;
4102         struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
4103         int ret;
4104
4105         ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4106                               &chunk_offset);
4107         if (ret)
4108                 return ret;
4109
4110         ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
4111                                   &stripe_size, chunk_offset, type);
4112         if (ret)
4113                 return ret;
4114
4115         ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
4116                                    chunk_size, stripe_size);
4117         if (ret)
4118                 return ret;
4119         return 0;
4120 }
4121
4122 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
4123                                          struct btrfs_root *root,
4124                                          struct btrfs_device *device)
4125 {
4126         u64 chunk_offset;
4127         u64 sys_chunk_offset;
4128         u64 chunk_size;
4129         u64 sys_chunk_size;
4130         u64 stripe_size;
4131         u64 sys_stripe_size;
4132         u64 alloc_profile;
4133         struct map_lookup *map;
4134         struct map_lookup *sys_map;
4135         struct btrfs_fs_info *fs_info = root->fs_info;
4136         struct btrfs_root *extent_root = fs_info->extent_root;
4137         int ret;
4138
4139         ret = find_next_chunk(fs_info->chunk_root,
4140                               BTRFS_FIRST_CHUNK_TREE_OBJECTID, &chunk_offset);
4141         if (ret)
4142                 return ret;
4143
4144         alloc_profile = btrfs_get_alloc_profile(extent_root, 0);
4145         ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
4146                                   &stripe_size, chunk_offset, alloc_profile);
4147         if (ret)
4148                 return ret;
4149
4150         sys_chunk_offset = chunk_offset + chunk_size;
4151
4152         alloc_profile = btrfs_get_alloc_profile(fs_info->chunk_root, 0);
4153         ret = __btrfs_alloc_chunk(trans, extent_root, &sys_map,
4154                                   &sys_chunk_size, &sys_stripe_size,
4155                                   sys_chunk_offset, alloc_profile);
4156         if (ret) {
4157                 btrfs_abort_transaction(trans, root, ret);
4158                 goto out;
4159         }
4160
4161         ret = btrfs_add_device(trans, fs_info->chunk_root, device);
4162         if (ret) {
4163                 btrfs_abort_transaction(trans, root, ret);
4164                 goto out;
4165         }
4166
4167         /*
4168          * Modifying chunk tree needs allocating new blocks from both
4169          * system block group and metadata block group. So we only can
4170          * do operations require modifying the chunk tree after both
4171          * block groups were created.
4172          */
4173         ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
4174                                    chunk_size, stripe_size);
4175         if (ret) {
4176                 btrfs_abort_transaction(trans, root, ret);
4177                 goto out;
4178         }
4179
4180         ret = __finish_chunk_alloc(trans, extent_root, sys_map,
4181                                    sys_chunk_offset, sys_chunk_size,
4182                                    sys_stripe_size);
4183         if (ret)
4184                 btrfs_abort_transaction(trans, root, ret);
4185
4186 out:
4187
4188         return ret;
4189 }
4190
4191 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
4192 {
4193         struct extent_map *em;
4194         struct map_lookup *map;
4195         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
4196         int readonly = 0;
4197         int i;
4198
4199         read_lock(&map_tree->map_tree.lock);
4200         em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
4201         read_unlock(&map_tree->map_tree.lock);
4202         if (!em)
4203                 return 1;
4204
4205         if (btrfs_test_opt(root, DEGRADED)) {
4206                 free_extent_map(em);
4207                 return 0;
4208         }
4209
4210         map = (struct map_lookup *)em->bdev;
4211         for (i = 0; i < map->num_stripes; i++) {
4212                 if (!map->stripes[i].dev->writeable) {
4213                         readonly = 1;
4214                         break;
4215                 }
4216         }
4217         free_extent_map(em);
4218         return readonly;
4219 }
4220
4221 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
4222 {
4223         extent_map_tree_init(&tree->map_tree);
4224 }
4225
4226 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
4227 {
4228         struct extent_map *em;
4229
4230         while (1) {
4231                 write_lock(&tree->map_tree.lock);
4232                 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
4233                 if (em)
4234                         remove_extent_mapping(&tree->map_tree, em);
4235                 write_unlock(&tree->map_tree.lock);
4236                 if (!em)
4237                         break;
4238                 kfree(em->bdev);
4239                 /* once for us */
4240                 free_extent_map(em);
4241                 /* once for the tree */
4242                 free_extent_map(em);
4243         }
4244 }
4245
4246 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
4247 {
4248         struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
4249         struct extent_map *em;
4250         struct map_lookup *map;
4251         struct extent_map_tree *em_tree = &map_tree->map_tree;
4252         int ret;
4253
4254         read_lock(&em_tree->lock);
4255         em = lookup_extent_mapping(em_tree, logical, len);
4256         read_unlock(&em_tree->lock);
4257
4258         /*
4259          * We could return errors for these cases, but that could get ugly and
4260          * we'd probably do the same thing which is just not do anything else
4261          * and exit, so return 1 so the callers don't try to use other copies.
4262          */
4263         if (!em) {
4264                 btrfs_emerg(fs_info, "No mapping for %Lu-%Lu\n", logical,
4265                             logical+len);
4266                 return 1;
4267         }
4268
4269         if (em->start > logical || em->start + em->len < logical) {
4270                 btrfs_emerg(fs_info, "Invalid mapping for %Lu-%Lu, got "
4271                             "%Lu-%Lu\n", logical, logical+len, em->start,
4272                             em->start + em->len);
4273                 free_extent_map(em);
4274                 return 1;
4275         }
4276
4277         map = (struct map_lookup *)em->bdev;
4278         if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
4279                 ret = map->num_stripes;
4280         else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
4281                 ret = map->sub_stripes;
4282         else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
4283                 ret = 2;
4284         else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
4285                 ret = 3;
4286         else
4287                 ret = 1;
4288         free_extent_map(em);
4289
4290         btrfs_dev_replace_lock(&fs_info->dev_replace);
4291         if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace))
4292                 ret++;
4293         btrfs_dev_replace_unlock(&fs_info->dev_replace);
4294
4295         return ret;
4296 }
4297
4298 unsigned long btrfs_full_stripe_len(struct btrfs_root *root,
4299                                     struct btrfs_mapping_tree *map_tree,
4300                                     u64 logical)
4301 {
4302         struct extent_map *em;
4303         struct map_lookup *map;
4304         struct extent_map_tree *em_tree = &map_tree->map_tree;
4305         unsigned long len = root->sectorsize;
4306
4307         read_lock(&em_tree->lock);
4308         em = lookup_extent_mapping(em_tree, logical, len);
4309         read_unlock(&em_tree->lock);
4310         BUG_ON(!em);
4311
4312         BUG_ON(em->start > logical || em->start + em->len < logical);
4313         map = (struct map_lookup *)em->bdev;
4314         if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
4315                          BTRFS_BLOCK_GROUP_RAID6)) {
4316                 len = map->stripe_len * nr_data_stripes(map);
4317         }
4318         free_extent_map(em);
4319         return len;
4320 }
4321
4322 int btrfs_is_parity_mirror(struct btrfs_mapping_tree *map_tree,
4323                            u64 logical, u64 len, int mirror_num)
4324 {
4325         struct extent_map *em;
4326         struct map_lookup *map;
4327         struct extent_map_tree *em_tree = &map_tree->map_tree;
4328         int ret = 0;
4329
4330         read_lock(&em_tree->lock);
4331         em = lookup_extent_mapping(em_tree, logical, len);
4332         read_unlock(&em_tree->lock);
4333         BUG_ON(!em);
4334
4335         BUG_ON(em->start > logical || em->start + em->len < logical);
4336         map = (struct map_lookup *)em->bdev;
4337         if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
4338                          BTRFS_BLOCK_GROUP_RAID6))
4339                 ret = 1;
4340         free_extent_map(em);
4341         return ret;
4342 }
4343
4344 static int find_live_mirror(struct btrfs_fs_info *fs_info,
4345                             struct map_lookup *map, int first, int num,
4346                             int optimal, int dev_replace_is_ongoing)
4347 {
4348         int i;
4349         int tolerance;
4350         struct btrfs_device *srcdev;
4351
4352         if (dev_replace_is_ongoing &&
4353             fs_info->dev_replace.cont_reading_from_srcdev_mode ==
4354              BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
4355                 srcdev = fs_info->dev_replace.srcdev;
4356         else
4357                 srcdev = NULL;
4358
4359         /*
4360          * try to avoid the drive that is the source drive for a
4361          * dev-replace procedure, only choose it if no other non-missing
4362          * mirror is available
4363          */
4364         for (tolerance = 0; tolerance < 2; tolerance++) {
4365                 if (map->stripes[optimal].dev->bdev &&
4366                     (tolerance || map->stripes[optimal].dev != srcdev))
4367                         return optimal;
4368                 for (i = first; i < first + num; i++) {
4369                         if (map->stripes[i].dev->bdev &&
4370                             (tolerance || map->stripes[i].dev != srcdev))
4371                                 return i;
4372                 }
4373         }
4374
4375         /* we couldn't find one that doesn't fail.  Just return something
4376          * and the io error handling code will clean up eventually
4377          */
4378         return optimal;
4379 }
4380
4381 static inline int parity_smaller(u64 a, u64 b)
4382 {
4383         return a > b;
4384 }
4385
4386 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
4387 static void sort_parity_stripes(struct btrfs_bio *bbio, u64 *raid_map)
4388 {
4389         struct btrfs_bio_stripe s;
4390         int i;
4391         u64 l;
4392         int again = 1;
4393
4394         while (again) {
4395                 again = 0;
4396                 for (i = 0; i < bbio->num_stripes - 1; i++) {
4397                         if (parity_smaller(raid_map[i], raid_map[i+1])) {
4398                                 s = bbio->stripes[i];
4399                                 l = raid_map[i];
4400                                 bbio->stripes[i] = bbio->stripes[i+1];
4401                                 raid_map[i] = raid_map[i+1];
4402                                 bbio->stripes[i+1] = s;
4403                                 raid_map[i+1] = l;
4404                                 again = 1;
4405                         }
4406                 }
4407         }
4408 }
4409
4410 static int __btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
4411                              u64 logical, u64 *length,
4412                              struct btrfs_bio **bbio_ret,
4413                              int mirror_num, u64 **raid_map_ret)
4414 {
4415         struct extent_map *em;
4416         struct map_lookup *map;
4417         struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
4418         struct extent_map_tree *em_tree = &map_tree->map_tree;
4419         u64 offset;
4420         u64 stripe_offset;
4421         u64 stripe_end_offset;
4422         u64 stripe_nr;
4423         u64 stripe_nr_orig;
4424         u64 stripe_nr_end;
4425         u64 stripe_len;
4426         u64 *raid_map = NULL;
4427         int stripe_index;
4428         int i;
4429         int ret = 0;
4430         int num_stripes;
4431         int max_errors = 0;
4432         struct btrfs_bio *bbio = NULL;
4433         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
4434         int dev_replace_is_ongoing = 0;
4435         int num_alloc_stripes;
4436         int patch_the_first_stripe_for_dev_replace = 0;
4437         u64 physical_to_patch_in_first_stripe = 0;
4438         u64 raid56_full_stripe_start = (u64)-1;
4439
4440         read_lock(&em_tree->lock);
4441         em = lookup_extent_mapping(em_tree, logical, *length);
4442         read_unlock(&em_tree->lock);
4443
4444         if (!em) {
4445                 btrfs_crit(fs_info, "unable to find logical %llu len %llu",
4446                         (unsigned long long)logical,
4447                         (unsigned long long)*length);
4448                 return -EINVAL;
4449         }
4450
4451         if (em->start > logical || em->start + em->len < logical) {
4452                 btrfs_crit(fs_info, "found a bad mapping, wanted %Lu, "
4453                            "found %Lu-%Lu\n", logical, em->start,
4454                            em->start + em->len);
4455                 free_extent_map(em);
4456                 return -EINVAL;
4457         }
4458
4459         map = (struct map_lookup *)em->bdev;
4460         offset = logical - em->start;
4461
4462         if (mirror_num > map->num_stripes)
4463                 mirror_num = 0;
4464
4465         stripe_len = map->stripe_len;
4466         stripe_nr = offset;
4467         /*
4468          * stripe_nr counts the total number of stripes we have to stride
4469          * to get to this block
4470          */
4471         do_div(stripe_nr, stripe_len);
4472
4473         stripe_offset = stripe_nr * stripe_len;
4474         BUG_ON(offset < stripe_offset);
4475
4476         /* stripe_offset is the offset of this block in its stripe*/
4477         stripe_offset = offset - stripe_offset;
4478
4479         /* if we're here for raid56, we need to know the stripe aligned start */
4480         if (map->type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)) {
4481                 unsigned long full_stripe_len = stripe_len * nr_data_stripes(map);
4482                 raid56_full_stripe_start = offset;
4483
4484                 /* allow a write of a full stripe, but make sure we don't
4485                  * allow straddling of stripes
4486                  */
4487                 do_div(raid56_full_stripe_start, full_stripe_len);
4488                 raid56_full_stripe_start *= full_stripe_len;
4489         }
4490
4491         if (rw & REQ_DISCARD) {
4492                 /* we don't discard raid56 yet */
4493                 if (map->type &
4494                     (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)) {
4495                         ret = -EOPNOTSUPP;
4496                         goto out;
4497                 }
4498                 *length = min_t(u64, em->len - offset, *length);
4499         } else if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
4500                 u64 max_len;
4501                 /* For writes to RAID[56], allow a full stripeset across all disks.
4502                    For other RAID types and for RAID[56] reads, just allow a single
4503                    stripe (on a single disk). */
4504                 if (map->type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6) &&
4505                     (rw & REQ_WRITE)) {
4506                         max_len = stripe_len * nr_data_stripes(map) -
4507                                 (offset - raid56_full_stripe_start);
4508                 } else {
4509                         /* we limit the length of each bio to what fits in a stripe */
4510                         max_len = stripe_len - stripe_offset;
4511                 }
4512                 *length = min_t(u64, em->len - offset, max_len);
4513         } else {
4514                 *length = em->len - offset;
4515         }
4516
4517         /* This is for when we're called from btrfs_merge_bio_hook() and all
4518            it cares about is the length */
4519         if (!bbio_ret)
4520                 goto out;
4521
4522         btrfs_dev_replace_lock(dev_replace);
4523         dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
4524         if (!dev_replace_is_ongoing)
4525                 btrfs_dev_replace_unlock(dev_replace);
4526
4527         if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
4528             !(rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) &&
4529             dev_replace->tgtdev != NULL) {
4530                 /*
4531                  * in dev-replace case, for repair case (that's the only
4532                  * case where the mirror is selected explicitly when
4533                  * calling btrfs_map_block), blocks left of the left cursor
4534                  * can also be read from the target drive.
4535                  * For REQ_GET_READ_MIRRORS, the target drive is added as
4536                  * the last one to the array of stripes. For READ, it also
4537                  * needs to be supported using the same mirror number.
4538                  * If the requested block is not left of the left cursor,
4539                  * EIO is returned. This can happen because btrfs_num_copies()
4540                  * returns one more in the dev-replace case.
4541                  */
4542                 u64 tmp_length = *length;
4543                 struct btrfs_bio *tmp_bbio = NULL;
4544                 int tmp_num_stripes;
4545                 u64 srcdev_devid = dev_replace->srcdev->devid;
4546                 int index_srcdev = 0;
4547                 int found = 0;
4548                 u64 physical_of_found = 0;
4549
4550                 ret = __btrfs_map_block(fs_info, REQ_GET_READ_MIRRORS,
4551                              logical, &tmp_length, &tmp_bbio, 0, NULL);
4552                 if (ret) {
4553                         WARN_ON(tmp_bbio != NULL);
4554                         goto out;
4555                 }
4556
4557                 tmp_num_stripes = tmp_bbio->num_stripes;
4558                 if (mirror_num > tmp_num_stripes) {
4559                         /*
4560                          * REQ_GET_READ_MIRRORS does not contain this
4561                          * mirror, that means that the requested area
4562                          * is not left of the left cursor
4563                          */
4564                         ret = -EIO;
4565                         kfree(tmp_bbio);
4566                         goto out;
4567                 }
4568
4569                 /*
4570                  * process the rest of the function using the mirror_num
4571                  * of the source drive. Therefore look it up first.
4572                  * At the end, patch the device pointer to the one of the
4573                  * target drive.
4574                  */
4575                 for (i = 0; i < tmp_num_stripes; i++) {
4576                         if (tmp_bbio->stripes[i].dev->devid == srcdev_devid) {
4577                                 /*
4578                                  * In case of DUP, in order to keep it
4579                                  * simple, only add the mirror with the
4580                                  * lowest physical address
4581                                  */
4582                                 if (found &&
4583                                     physical_of_found <=
4584                                      tmp_bbio->stripes[i].physical)
4585                                         continue;
4586                                 index_srcdev = i;
4587                                 found = 1;
4588                                 physical_of_found =
4589                                         tmp_bbio->stripes[i].physical;
4590                         }
4591                 }
4592
4593                 if (found) {
4594                         mirror_num = index_srcdev + 1;
4595                         patch_the_first_stripe_for_dev_replace = 1;
4596                         physical_to_patch_in_first_stripe = physical_of_found;
4597                 } else {
4598                         WARN_ON(1);
4599                         ret = -EIO;
4600                         kfree(tmp_bbio);
4601                         goto out;
4602                 }
4603
4604                 kfree(tmp_bbio);
4605         } else if (mirror_num > map->num_stripes) {
4606                 mirror_num = 0;
4607         }
4608
4609         num_stripes = 1;
4610         stripe_index = 0;
4611         stripe_nr_orig = stripe_nr;
4612         stripe_nr_end = ALIGN(offset + *length, map->stripe_len);
4613         do_div(stripe_nr_end, map->stripe_len);
4614         stripe_end_offset = stripe_nr_end * map->stripe_len -
4615                             (offset + *length);
4616
4617         if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
4618                 if (rw & REQ_DISCARD)
4619                         num_stripes = min_t(u64, map->num_stripes,
4620                                             stripe_nr_end - stripe_nr_orig);
4621                 stripe_index = do_div(stripe_nr, map->num_stripes);
4622         } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
4623                 if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS))
4624                         num_stripes = map->num_stripes;
4625                 else if (mirror_num)
4626                         stripe_index = mirror_num - 1;
4627                 else {
4628                         stripe_index = find_live_mirror(fs_info, map, 0,
4629                                             map->num_stripes,
4630                                             current->pid % map->num_stripes,
4631                                             dev_replace_is_ongoing);
4632                         mirror_num = stripe_index + 1;
4633                 }
4634
4635         } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
4636                 if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) {
4637                         num_stripes = map->num_stripes;
4638                 } else if (mirror_num) {
4639                         stripe_index = mirror_num - 1;
4640                 } else {
4641                         mirror_num = 1;
4642                 }
4643
4644         } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
4645                 int factor = map->num_stripes / map->sub_stripes;
4646
4647                 stripe_index = do_div(stripe_nr, factor);
4648                 stripe_index *= map->sub_stripes;
4649
4650                 if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS))
4651                         num_stripes = map->sub_stripes;
4652                 else if (rw & REQ_DISCARD)
4653                         num_stripes = min_t(u64, map->sub_stripes *
4654                                             (stripe_nr_end - stripe_nr_orig),
4655                                             map->num_stripes);
4656                 else if (mirror_num)
4657                         stripe_index += mirror_num - 1;
4658                 else {
4659                         int old_stripe_index = stripe_index;
4660                         stripe_index = find_live_mirror(fs_info, map,
4661                                               stripe_index,
4662                                               map->sub_stripes, stripe_index +
4663                                               current->pid % map->sub_stripes,
4664                                               dev_replace_is_ongoing);
4665                         mirror_num = stripe_index - old_stripe_index + 1;
4666                 }
4667
4668         } else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
4669                                 BTRFS_BLOCK_GROUP_RAID6)) {
4670                 u64 tmp;
4671
4672                 if (bbio_ret && ((rw & REQ_WRITE) || mirror_num > 1)
4673                     && raid_map_ret) {
4674                         int i, rot;
4675
4676                         /* push stripe_nr back to the start of the full stripe */
4677                         stripe_nr = raid56_full_stripe_start;
4678                         do_div(stripe_nr, stripe_len);
4679
4680                         stripe_index = do_div(stripe_nr, nr_data_stripes(map));
4681
4682                         /* RAID[56] write or recovery. Return all stripes */
4683                         num_stripes = map->num_stripes;
4684                         max_errors = nr_parity_stripes(map);
4685
4686                         raid_map = kmalloc(sizeof(u64) * num_stripes,
4687                                            GFP_NOFS);
4688                         if (!raid_map) {
4689                                 ret = -ENOMEM;
4690                                 goto out;
4691                         }
4692
4693                         /* Work out the disk rotation on this stripe-set */
4694                         tmp = stripe_nr;
4695                         rot = do_div(tmp, num_stripes);
4696
4697                         /* Fill in the logical address of each stripe */
4698                         tmp = stripe_nr * nr_data_stripes(map);
4699                         for (i = 0; i < nr_data_stripes(map); i++)
4700                                 raid_map[(i+rot) % num_stripes] =
4701                                         em->start + (tmp + i) * map->stripe_len;
4702
4703                         raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE;
4704                         if (map->type & BTRFS_BLOCK_GROUP_RAID6)
4705                                 raid_map[(i+rot+1) % num_stripes] =
4706                                         RAID6_Q_STRIPE;
4707
4708                         *length = map->stripe_len;
4709                         stripe_index = 0;
4710                         stripe_offset = 0;
4711                 } else {
4712                         /*
4713                          * Mirror #0 or #1 means the original data block.
4714                          * Mirror #2 is RAID5 parity block.
4715                          * Mirror #3 is RAID6 Q block.
4716                          */
4717                         stripe_index = do_div(stripe_nr, nr_data_stripes(map));
4718                         if (mirror_num > 1)
4719                                 stripe_index = nr_data_stripes(map) +
4720                                                 mirror_num - 2;
4721
4722                         /* We distribute the parity blocks across stripes */
4723                         tmp = stripe_nr + stripe_index;
4724                         stripe_index = do_div(tmp, map->num_stripes);
4725                 }
4726         } else {
4727                 /*
4728                  * after this do_div call, stripe_nr is the number of stripes
4729                  * on this device we have to walk to find the data, and
4730                  * stripe_index is the number of our device in the stripe array
4731                  */
4732                 stripe_index = do_div(stripe_nr, map->num_stripes);
4733                 mirror_num = stripe_index + 1;
4734         }
4735         BUG_ON(stripe_index >= map->num_stripes);
4736
4737         num_alloc_stripes = num_stripes;
4738         if (dev_replace_is_ongoing) {
4739                 if (rw & (REQ_WRITE | REQ_DISCARD))
4740                         num_alloc_stripes <<= 1;
4741                 if (rw & REQ_GET_READ_MIRRORS)
4742                         num_alloc_stripes++;
4743         }
4744         bbio = kzalloc(btrfs_bio_size(num_alloc_stripes), GFP_NOFS);
4745         if (!bbio) {
4746                 ret = -ENOMEM;
4747                 goto out;
4748         }
4749         atomic_set(&bbio->error, 0);
4750
4751         if (rw & REQ_DISCARD) {
4752                 int factor = 0;
4753                 int sub_stripes = 0;
4754                 u64 stripes_per_dev = 0;
4755                 u32 remaining_stripes = 0;
4756                 u32 last_stripe = 0;
4757
4758                 if (map->type &
4759                     (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID10)) {
4760                         if (map->type & BTRFS_BLOCK_GROUP_RAID0)
4761                                 sub_stripes = 1;
4762                         else
4763                                 sub_stripes = map->sub_stripes;
4764
4765                         factor = map->num_stripes / sub_stripes;
4766                         stripes_per_dev = div_u64_rem(stripe_nr_end -
4767                                                       stripe_nr_orig,
4768                                                       factor,
4769                                                       &remaining_stripes);
4770                         div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
4771                         last_stripe *= sub_stripes;
4772                 }
4773
4774                 for (i = 0; i < num_stripes; i++) {
4775                         bbio->stripes[i].physical =
4776                                 map->stripes[stripe_index].physical +
4777                                 stripe_offset + stripe_nr * map->stripe_len;
4778                         bbio->stripes[i].dev = map->stripes[stripe_index].dev;
4779
4780                         if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
4781                                          BTRFS_BLOCK_GROUP_RAID10)) {
4782                                 bbio->stripes[i].length = stripes_per_dev *
4783                                                           map->stripe_len;
4784
4785                                 if (i / sub_stripes < remaining_stripes)
4786                                         bbio->stripes[i].length +=
4787                                                 map->stripe_len;
4788
4789                                 /*
4790                                  * Special for the first stripe and
4791                                  * the last stripe:
4792                                  *
4793                                  * |-------|...|-------|
4794                                  *     |----------|
4795                                  *    off     end_off
4796                                  */
4797                                 if (i < sub_stripes)
4798                                         bbio->stripes[i].length -=
4799                                                 stripe_offset;
4800
4801                                 if (stripe_index >= last_stripe &&
4802                                     stripe_index <= (last_stripe +
4803                                                      sub_stripes - 1))
4804                                         bbio->stripes[i].length -=
4805                                                 stripe_end_offset;
4806
4807                                 if (i == sub_stripes - 1)
4808                                         stripe_offset = 0;
4809                         } else
4810                                 bbio->stripes[i].length = *length;
4811
4812                         stripe_index++;
4813                         if (stripe_index == map->num_stripes) {
4814                                 /* This could only happen for RAID0/10 */
4815                                 stripe_index = 0;
4816                                 stripe_nr++;
4817                         }
4818                 }
4819         } else {
4820                 for (i = 0; i < num_stripes; i++) {
4821                         bbio->stripes[i].physical =
4822                                 map->stripes[stripe_index].physical +
4823                                 stripe_offset +
4824                                 stripe_nr * map->stripe_len;
4825                         bbio->stripes[i].dev =
4826                                 map->stripes[stripe_index].dev;
4827                         stripe_index++;
4828                 }
4829         }
4830
4831         if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS)) {
4832                 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
4833                                  BTRFS_BLOCK_GROUP_RAID10 |
4834                                  BTRFS_BLOCK_GROUP_RAID5 |
4835                                  BTRFS_BLOCK_GROUP_DUP)) {
4836                         max_errors = 1;
4837                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID6) {
4838                         max_errors = 2;
4839                 }
4840         }
4841
4842         if (dev_replace_is_ongoing && (rw & (REQ_WRITE | REQ_DISCARD)) &&
4843             dev_replace->tgtdev != NULL) {
4844                 int index_where_to_add;
4845                 u64 srcdev_devid = dev_replace->srcdev->devid;
4846
4847                 /*
4848                  * duplicate the write operations while the dev replace
4849                  * procedure is running. Since the copying of the old disk
4850                  * to the new disk takes place at run time while the
4851                  * filesystem is mounted writable, the regular write
4852                  * operations to the old disk have to be duplicated to go
4853                  * to the new disk as well.
4854                  * Note that device->missing is handled by the caller, and
4855                  * that the write to the old disk is already set up in the
4856                  * stripes array.
4857                  */
4858                 index_where_to_add = num_stripes;
4859                 for (i = 0; i < num_stripes; i++) {
4860                         if (bbio->stripes[i].dev->devid == srcdev_devid) {
4861                                 /* write to new disk, too */
4862                                 struct btrfs_bio_stripe *new =
4863                                         bbio->stripes + index_where_to_add;
4864                                 struct btrfs_bio_stripe *old =
4865                                         bbio->stripes + i;
4866
4867                                 new->physical = old->physical;
4868                                 new->length = old->length;
4869                                 new->dev = dev_replace->tgtdev;
4870                                 index_where_to_add++;
4871                                 max_errors++;
4872                         }
4873                 }
4874                 num_stripes = index_where_to_add;
4875         } else if (dev_replace_is_ongoing && (rw & REQ_GET_READ_MIRRORS) &&
4876                    dev_replace->tgtdev != NULL) {
4877                 u64 srcdev_devid = dev_replace->srcdev->devid;
4878                 int index_srcdev = 0;
4879                 int found = 0;
4880                 u64 physical_of_found = 0;
4881
4882                 /*
4883                  * During the dev-replace procedure, the target drive can
4884                  * also be used to read data in case it is needed to repair
4885                  * a corrupt block elsewhere. This is possible if the
4886                  * requested area is left of the left cursor. In this area,
4887                  * the target drive is a full copy of the source drive.
4888                  */
4889                 for (i = 0; i < num_stripes; i++) {
4890                         if (bbio->stripes[i].dev->devid == srcdev_devid) {
4891                                 /*
4892                                  * In case of DUP, in order to keep it
4893                                  * simple, only add the mirror with the
4894                                  * lowest physical address
4895                                  */
4896                                 if (found &&
4897                                     physical_of_found <=
4898                                      bbio->stripes[i].physical)
4899                                         continue;
4900                                 index_srcdev = i;
4901                                 found = 1;
4902                                 physical_of_found = bbio->stripes[i].physical;
4903                         }
4904                 }
4905                 if (found) {
4906                         u64 length = map->stripe_len;
4907
4908                         if (physical_of_found + length <=
4909                             dev_replace->cursor_left) {
4910                                 struct btrfs_bio_stripe *tgtdev_stripe =
4911                                         bbio->stripes + num_stripes;
4912
4913                                 tgtdev_stripe->physical = physical_of_found;
4914                                 tgtdev_stripe->length =
4915                                         bbio->stripes[index_srcdev].length;
4916                                 tgtdev_stripe->dev = dev_replace->tgtdev;
4917
4918                                 num_stripes++;
4919                         }
4920                 }
4921         }
4922
4923         *bbio_ret = bbio;
4924         bbio->num_stripes = num_stripes;
4925         bbio->max_errors = max_errors;
4926         bbio->mirror_num = mirror_num;
4927
4928         /*
4929          * this is the case that REQ_READ && dev_replace_is_ongoing &&
4930          * mirror_num == num_stripes + 1 && dev_replace target drive is
4931          * available as a mirror
4932          */
4933         if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) {
4934                 WARN_ON(num_stripes > 1);
4935                 bbio->stripes[0].dev = dev_replace->tgtdev;
4936                 bbio->stripes[0].physical = physical_to_patch_in_first_stripe;
4937                 bbio->mirror_num = map->num_stripes + 1;
4938         }
4939         if (raid_map) {
4940                 sort_parity_stripes(bbio, raid_map);
4941                 *raid_map_ret = raid_map;
4942         }
4943 out:
4944         if (dev_replace_is_ongoing)
4945                 btrfs_dev_replace_unlock(dev_replace);
4946         free_extent_map(em);
4947         return ret;
4948 }
4949
4950 int btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
4951                       u64 logical, u64 *length,
4952                       struct btrfs_bio **bbio_ret, int mirror_num)
4953 {
4954         return __btrfs_map_block(fs_info, rw, logical, length, bbio_ret,
4955                                  mirror_num, NULL);
4956 }
4957
4958 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
4959                      u64 chunk_start, u64 physical, u64 devid,
4960                      u64 **logical, int *naddrs, int *stripe_len)
4961 {
4962         struct extent_map_tree *em_tree = &map_tree->map_tree;
4963         struct extent_map *em;
4964         struct map_lookup *map;
4965         u64 *buf;
4966         u64 bytenr;
4967         u64 length;
4968         u64 stripe_nr;
4969         u64 rmap_len;
4970         int i, j, nr = 0;
4971
4972         read_lock(&em_tree->lock);
4973         em = lookup_extent_mapping(em_tree, chunk_start, 1);
4974         read_unlock(&em_tree->lock);
4975
4976         if (!em) {
4977                 printk(KERN_ERR "btrfs: couldn't find em for chunk %Lu\n",
4978                        chunk_start);
4979                 return -EIO;
4980         }
4981
4982         if (em->start != chunk_start) {
4983                 printk(KERN_ERR "btrfs: bad chunk start, em=%Lu, wanted=%Lu\n",
4984                        em->start, chunk_start);
4985                 free_extent_map(em);
4986                 return -EIO;
4987         }
4988         map = (struct map_lookup *)em->bdev;
4989
4990         length = em->len;
4991         rmap_len = map->stripe_len;
4992
4993         if (map->type & BTRFS_BLOCK_GROUP_RAID10)
4994                 do_div(length, map->num_stripes / map->sub_stripes);
4995         else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
4996                 do_div(length, map->num_stripes);
4997         else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
4998                               BTRFS_BLOCK_GROUP_RAID6)) {
4999                 do_div(length, nr_data_stripes(map));
5000                 rmap_len = map->stripe_len * nr_data_stripes(map);
5001         }
5002
5003         buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
5004         BUG_ON(!buf); /* -ENOMEM */
5005
5006         for (i = 0; i < map->num_stripes; i++) {
5007                 if (devid && map->stripes[i].dev->devid != devid)
5008                         continue;
5009                 if (map->stripes[i].physical > physical ||
5010                     map->stripes[i].physical + length <= physical)
5011                         continue;
5012
5013                 stripe_nr = physical - map->stripes[i].physical;
5014                 do_div(stripe_nr, map->stripe_len);
5015
5016                 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5017                         stripe_nr = stripe_nr * map->num_stripes + i;
5018                         do_div(stripe_nr, map->sub_stripes);
5019                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5020                         stripe_nr = stripe_nr * map->num_stripes + i;
5021                 } /* else if RAID[56], multiply by nr_data_stripes().
5022                    * Alternatively, just use rmap_len below instead of
5023                    * map->stripe_len */
5024
5025                 bytenr = chunk_start + stripe_nr * rmap_len;
5026                 WARN_ON(nr >= map->num_stripes);
5027                 for (j = 0; j < nr; j++) {
5028                         if (buf[j] == bytenr)
5029                                 break;
5030                 }
5031                 if (j == nr) {
5032                         WARN_ON(nr >= map->num_stripes);
5033                         buf[nr++] = bytenr;
5034                 }
5035         }
5036
5037         *logical = buf;
5038         *naddrs = nr;
5039         *stripe_len = rmap_len;
5040
5041         free_extent_map(em);
5042         return 0;
5043 }
5044
5045 static void btrfs_end_bio(struct bio *bio, int err)
5046 {
5047         struct btrfs_bio *bbio = bio->bi_private;
5048         int is_orig_bio = 0;
5049
5050         if (err) {
5051                 atomic_inc(&bbio->error);
5052                 if (err == -EIO || err == -EREMOTEIO) {
5053                         unsigned int stripe_index =
5054                                 btrfs_io_bio(bio)->stripe_index;
5055                         struct btrfs_device *dev;
5056
5057                         BUG_ON(stripe_index >= bbio->num_stripes);
5058                         dev = bbio->stripes[stripe_index].dev;
5059                         if (dev->bdev) {
5060                                 if (bio->bi_rw & WRITE)
5061                                         btrfs_dev_stat_inc(dev,
5062                                                 BTRFS_DEV_STAT_WRITE_ERRS);
5063                                 else
5064                                         btrfs_dev_stat_inc(dev,
5065                                                 BTRFS_DEV_STAT_READ_ERRS);
5066                                 if ((bio->bi_rw & WRITE_FLUSH) == WRITE_FLUSH)
5067                                         btrfs_dev_stat_inc(dev,
5068                                                 BTRFS_DEV_STAT_FLUSH_ERRS);
5069                                 btrfs_dev_stat_print_on_error(dev);
5070                         }
5071                 }
5072         }
5073
5074         if (bio == bbio->orig_bio)
5075                 is_orig_bio = 1;
5076
5077         if (atomic_dec_and_test(&bbio->stripes_pending)) {
5078                 if (!is_orig_bio) {
5079                         bio_put(bio);
5080                         bio = bbio->orig_bio;
5081                 }
5082                 bio->bi_private = bbio->private;
5083                 bio->bi_end_io = bbio->end_io;
5084                 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
5085                 /* only send an error to the higher layers if it is
5086                  * beyond the tolerance of the btrfs bio
5087                  */
5088                 if (atomic_read(&bbio->error) > bbio->max_errors) {
5089                         err = -EIO;
5090                 } else {
5091                         /*
5092                          * this bio is actually up to date, we didn't
5093                          * go over the max number of errors
5094                          */
5095                         set_bit(BIO_UPTODATE, &bio->bi_flags);
5096                         err = 0;
5097                 }
5098                 kfree(bbio);
5099
5100                 bio_endio(bio, err);
5101         } else if (!is_orig_bio) {
5102                 bio_put(bio);
5103         }
5104 }
5105
5106 struct async_sched {
5107         struct bio *bio;
5108         int rw;
5109         struct btrfs_fs_info *info;
5110         struct btrfs_work work;
5111 };
5112
5113 /*
5114  * see run_scheduled_bios for a description of why bios are collected for
5115  * async submit.
5116  *
5117  * This will add one bio to the pending list for a device and make sure
5118  * the work struct is scheduled.
5119  */
5120 static noinline void btrfs_schedule_bio(struct btrfs_root *root,
5121                                         struct btrfs_device *device,
5122                                         int rw, struct bio *bio)
5123 {
5124         int should_queue = 1;
5125         struct btrfs_pending_bios *pending_bios;
5126
5127         if (device->missing || !device->bdev) {
5128                 bio_endio(bio, -EIO);
5129                 return;
5130         }
5131
5132         /* don't bother with additional async steps for reads, right now */
5133         if (!(rw & REQ_WRITE)) {
5134                 bio_get(bio);
5135                 btrfsic_submit_bio(rw, bio);
5136                 bio_put(bio);
5137                 return;
5138         }
5139
5140         /*
5141          * nr_async_bios allows us to reliably return congestion to the
5142          * higher layers.  Otherwise, the async bio makes it appear we have
5143          * made progress against dirty pages when we've really just put it
5144          * on a queue for later
5145          */
5146         atomic_inc(&root->fs_info->nr_async_bios);
5147         WARN_ON(bio->bi_next);
5148         bio->bi_next = NULL;
5149         bio->bi_rw |= rw;
5150
5151         spin_lock(&device->io_lock);
5152         if (bio->bi_rw & REQ_SYNC)
5153                 pending_bios = &device->pending_sync_bios;
5154         else
5155                 pending_bios = &device->pending_bios;
5156
5157         if (pending_bios->tail)
5158                 pending_bios->tail->bi_next = bio;
5159
5160         pending_bios->tail = bio;
5161         if (!pending_bios->head)
5162                 pending_bios->head = bio;
5163         if (device->running_pending)
5164                 should_queue = 0;
5165
5166         spin_unlock(&device->io_lock);
5167
5168         if (should_queue)
5169                 btrfs_queue_worker(&root->fs_info->submit_workers,
5170                                    &device->work);
5171 }
5172
5173 static int bio_size_ok(struct block_device *bdev, struct bio *bio,
5174                        sector_t sector)
5175 {
5176         struct bio_vec *prev;
5177         struct request_queue *q = bdev_get_queue(bdev);
5178         unsigned short max_sectors = queue_max_sectors(q);
5179         struct bvec_merge_data bvm = {
5180                 .bi_bdev = bdev,
5181                 .bi_sector = sector,
5182                 .bi_rw = bio->bi_rw,
5183         };
5184
5185         if (bio->bi_vcnt == 0) {
5186                 WARN_ON(1);
5187                 return 1;
5188         }
5189
5190         prev = &bio->bi_io_vec[bio->bi_vcnt - 1];
5191         if (bio_sectors(bio) > max_sectors)
5192                 return 0;
5193
5194         if (!q->merge_bvec_fn)
5195                 return 1;
5196
5197         bvm.bi_size = bio->bi_size - prev->bv_len;
5198         if (q->merge_bvec_fn(q, &bvm, prev) < prev->bv_len)
5199                 return 0;
5200         return 1;
5201 }
5202
5203 static void submit_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio,
5204                               struct bio *bio, u64 physical, int dev_nr,
5205                               int rw, int async)
5206 {
5207         struct btrfs_device *dev = bbio->stripes[dev_nr].dev;
5208
5209         bio->bi_private = bbio;
5210         btrfs_io_bio(bio)->stripe_index = dev_nr;
5211         bio->bi_end_io = btrfs_end_bio;
5212         bio->bi_sector = physical >> 9;
5213 #ifdef DEBUG
5214         {
5215                 struct rcu_string *name;
5216
5217                 rcu_read_lock();
5218                 name = rcu_dereference(dev->name);
5219                 pr_debug("btrfs_map_bio: rw %d, sector=%llu, dev=%lu "
5220                          "(%s id %llu), size=%u\n", rw,
5221                          (u64)bio->bi_sector, (u_long)dev->bdev->bd_dev,
5222                          name->str, dev->devid, bio->bi_size);
5223                 rcu_read_unlock();
5224         }
5225 #endif
5226         bio->bi_bdev = dev->bdev;
5227         if (async)
5228                 btrfs_schedule_bio(root, dev, rw, bio);
5229         else
5230                 btrfsic_submit_bio(rw, bio);
5231 }
5232
5233 static int breakup_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio,
5234                               struct bio *first_bio, struct btrfs_device *dev,
5235                               int dev_nr, int rw, int async)
5236 {
5237         struct bio_vec *bvec = first_bio->bi_io_vec;
5238         struct bio *bio;
5239         int nr_vecs = bio_get_nr_vecs(dev->bdev);
5240         u64 physical = bbio->stripes[dev_nr].physical;
5241
5242 again:
5243         bio = btrfs_bio_alloc(dev->bdev, physical >> 9, nr_vecs, GFP_NOFS);
5244         if (!bio)
5245                 return -ENOMEM;
5246
5247         while (bvec <= (first_bio->bi_io_vec + first_bio->bi_vcnt - 1)) {
5248                 if (bio_add_page(bio, bvec->bv_page, bvec->bv_len,
5249                                  bvec->bv_offset) < bvec->bv_len) {
5250                         u64 len = bio->bi_size;
5251
5252                         atomic_inc(&bbio->stripes_pending);
5253                         submit_stripe_bio(root, bbio, bio, physical, dev_nr,
5254                                           rw, async);
5255                         physical += len;
5256                         goto again;
5257                 }
5258                 bvec++;
5259         }
5260
5261         submit_stripe_bio(root, bbio, bio, physical, dev_nr, rw, async);
5262         return 0;
5263 }
5264
5265 static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical)
5266 {
5267         atomic_inc(&bbio->error);
5268         if (atomic_dec_and_test(&bbio->stripes_pending)) {
5269                 bio->bi_private = bbio->private;
5270                 bio->bi_end_io = bbio->end_io;
5271                 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
5272                 bio->bi_sector = logical >> 9;
5273                 kfree(bbio);
5274                 bio_endio(bio, -EIO);
5275         }
5276 }
5277
5278 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
5279                   int mirror_num, int async_submit)
5280 {
5281         struct btrfs_device *dev;
5282         struct bio *first_bio = bio;
5283         u64 logical = (u64)bio->bi_sector << 9;
5284         u64 length = 0;
5285         u64 map_length;
5286         u64 *raid_map = NULL;
5287         int ret;
5288         int dev_nr = 0;
5289         int total_devs = 1;
5290         struct btrfs_bio *bbio = NULL;
5291
5292         length = bio->bi_size;
5293         map_length = length;
5294
5295         ret = __btrfs_map_block(root->fs_info, rw, logical, &map_length, &bbio,
5296                               mirror_num, &raid_map);
5297         if (ret) /* -ENOMEM */
5298                 return ret;
5299
5300         total_devs = bbio->num_stripes;
5301         bbio->orig_bio = first_bio;
5302         bbio->private = first_bio->bi_private;
5303         bbio->end_io = first_bio->bi_end_io;
5304         atomic_set(&bbio->stripes_pending, bbio->num_stripes);
5305
5306         if (raid_map) {
5307                 /* In this case, map_length has been set to the length of
5308                    a single stripe; not the whole write */
5309                 if (rw & WRITE) {
5310                         return raid56_parity_write(root, bio, bbio,
5311                                                    raid_map, map_length);
5312                 } else {
5313                         return raid56_parity_recover(root, bio, bbio,
5314                                                      raid_map, map_length,
5315                                                      mirror_num);
5316                 }
5317         }
5318
5319         if (map_length < length) {
5320                 btrfs_crit(root->fs_info, "mapping failed logical %llu bio len %llu len %llu",
5321                         (unsigned long long)logical,
5322                         (unsigned long long)length,
5323                         (unsigned long long)map_length);
5324                 BUG();
5325         }
5326
5327         while (dev_nr < total_devs) {
5328                 dev = bbio->stripes[dev_nr].dev;
5329                 if (!dev || !dev->bdev || (rw & WRITE && !dev->writeable)) {
5330                         bbio_error(bbio, first_bio, logical);
5331                         dev_nr++;
5332                         continue;
5333                 }
5334
5335                 /*
5336                  * Check and see if we're ok with this bio based on it's size
5337                  * and offset with the given device.
5338                  */
5339                 if (!bio_size_ok(dev->bdev, first_bio,
5340                                  bbio->stripes[dev_nr].physical >> 9)) {
5341                         ret = breakup_stripe_bio(root, bbio, first_bio, dev,
5342                                                  dev_nr, rw, async_submit);
5343                         BUG_ON(ret);
5344                         dev_nr++;
5345                         continue;
5346                 }
5347
5348                 if (dev_nr < total_devs - 1) {
5349                         bio = btrfs_bio_clone(first_bio, GFP_NOFS);
5350                         BUG_ON(!bio); /* -ENOMEM */
5351                 } else {
5352                         bio = first_bio;
5353                 }
5354
5355                 submit_stripe_bio(root, bbio, bio,
5356                                   bbio->stripes[dev_nr].physical, dev_nr, rw,
5357                                   async_submit);
5358                 dev_nr++;
5359         }
5360         return 0;
5361 }
5362
5363 struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid,
5364                                        u8 *uuid, u8 *fsid)
5365 {
5366         struct btrfs_device *device;
5367         struct btrfs_fs_devices *cur_devices;
5368
5369         cur_devices = fs_info->fs_devices;
5370         while (cur_devices) {
5371                 if (!fsid ||
5372                     !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
5373                         device = __find_device(&cur_devices->devices,
5374                                                devid, uuid);
5375                         if (device)
5376                                 return device;
5377                 }
5378                 cur_devices = cur_devices->seed;
5379         }
5380         return NULL;
5381 }
5382
5383 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
5384                                             u64 devid, u8 *dev_uuid)
5385 {
5386         struct btrfs_device *device;
5387         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
5388
5389         device = kzalloc(sizeof(*device), GFP_NOFS);
5390         if (!device)
5391                 return NULL;
5392         list_add(&device->dev_list,
5393                  &fs_devices->devices);
5394         device->dev_root = root->fs_info->dev_root;
5395         device->devid = devid;
5396         device->work.func = pending_bios_fn;
5397         device->fs_devices = fs_devices;
5398         device->missing = 1;
5399         fs_devices->num_devices++;
5400         fs_devices->missing_devices++;
5401         spin_lock_init(&device->io_lock);
5402         INIT_LIST_HEAD(&device->dev_alloc_list);
5403         memcpy(device->uuid, dev_uuid, BTRFS_UUID_SIZE);
5404         return device;
5405 }
5406
5407 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
5408                           struct extent_buffer *leaf,
5409                           struct btrfs_chunk *chunk)
5410 {
5411         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
5412         struct map_lookup *map;
5413         struct extent_map *em;
5414         u64 logical;
5415         u64 length;
5416         u64 devid;
5417         u8 uuid[BTRFS_UUID_SIZE];
5418         int num_stripes;
5419         int ret;
5420         int i;
5421
5422         logical = key->offset;
5423         length = btrfs_chunk_length(leaf, chunk);
5424
5425         read_lock(&map_tree->map_tree.lock);
5426         em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
5427         read_unlock(&map_tree->map_tree.lock);
5428
5429         /* already mapped? */
5430         if (em && em->start <= logical && em->start + em->len > logical) {
5431                 free_extent_map(em);
5432                 return 0;
5433         } else if (em) {
5434                 free_extent_map(em);
5435         }
5436
5437         em = alloc_extent_map();
5438         if (!em)
5439                 return -ENOMEM;
5440         num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
5441         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
5442         if (!map) {
5443                 free_extent_map(em);
5444                 return -ENOMEM;
5445         }
5446
5447         em->bdev = (struct block_device *)map;
5448         em->start = logical;
5449         em->len = length;
5450         em->orig_start = 0;
5451         em->block_start = 0;
5452         em->block_len = em->len;
5453
5454         map->num_stripes = num_stripes;
5455         map->io_width = btrfs_chunk_io_width(leaf, chunk);
5456         map->io_align = btrfs_chunk_io_align(leaf, chunk);
5457         map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
5458         map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
5459         map->type = btrfs_chunk_type(leaf, chunk);
5460         map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
5461         for (i = 0; i < num_stripes; i++) {
5462                 map->stripes[i].physical =
5463                         btrfs_stripe_offset_nr(leaf, chunk, i);
5464                 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
5465                 read_extent_buffer(leaf, uuid, (unsigned long)
5466                                    btrfs_stripe_dev_uuid_nr(chunk, i),
5467                                    BTRFS_UUID_SIZE);
5468                 map->stripes[i].dev = btrfs_find_device(root->fs_info, devid,
5469                                                         uuid, NULL);
5470                 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
5471                         kfree(map);
5472                         free_extent_map(em);
5473                         return -EIO;
5474                 }
5475                 if (!map->stripes[i].dev) {
5476                         map->stripes[i].dev =
5477                                 add_missing_dev(root, devid, uuid);
5478                         if (!map->stripes[i].dev) {
5479                                 kfree(map);
5480                                 free_extent_map(em);
5481                                 return -EIO;
5482                         }
5483                 }
5484                 map->stripes[i].dev->in_fs_metadata = 1;
5485         }
5486
5487         write_lock(&map_tree->map_tree.lock);
5488         ret = add_extent_mapping(&map_tree->map_tree, em, 0);
5489         write_unlock(&map_tree->map_tree.lock);
5490         BUG_ON(ret); /* Tree corruption */
5491         free_extent_map(em);
5492
5493         return 0;
5494 }
5495
5496 static void fill_device_from_item(struct extent_buffer *leaf,
5497                                  struct btrfs_dev_item *dev_item,
5498                                  struct btrfs_device *device)
5499 {
5500         unsigned long ptr;
5501
5502         device->devid = btrfs_device_id(leaf, dev_item);
5503         device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
5504         device->total_bytes = device->disk_total_bytes;
5505         device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
5506         device->type = btrfs_device_type(leaf, dev_item);
5507         device->io_align = btrfs_device_io_align(leaf, dev_item);
5508         device->io_width = btrfs_device_io_width(leaf, dev_item);
5509         device->sector_size = btrfs_device_sector_size(leaf, dev_item);
5510         WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
5511         device->is_tgtdev_for_dev_replace = 0;
5512
5513         ptr = (unsigned long)btrfs_device_uuid(dev_item);
5514         read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
5515 }
5516
5517 static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
5518 {
5519         struct btrfs_fs_devices *fs_devices;
5520         int ret;
5521
5522         BUG_ON(!mutex_is_locked(&uuid_mutex));
5523
5524         fs_devices = root->fs_info->fs_devices->seed;
5525         while (fs_devices) {
5526                 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
5527                         ret = 0;
5528                         goto out;
5529                 }
5530                 fs_devices = fs_devices->seed;
5531         }
5532
5533         fs_devices = find_fsid(fsid);
5534         if (!fs_devices) {
5535                 ret = -ENOENT;
5536                 goto out;
5537         }
5538
5539         fs_devices = clone_fs_devices(fs_devices);
5540         if (IS_ERR(fs_devices)) {
5541                 ret = PTR_ERR(fs_devices);
5542                 goto out;
5543         }
5544
5545         ret = __btrfs_open_devices(fs_devices, FMODE_READ,
5546                                    root->fs_info->bdev_holder);
5547         if (ret) {
5548                 free_fs_devices(fs_devices);
5549                 goto out;
5550         }
5551
5552         if (!fs_devices->seeding) {
5553                 __btrfs_close_devices(fs_devices);
5554                 free_fs_devices(fs_devices);
5555                 ret = -EINVAL;
5556                 goto out;
5557         }
5558
5559         fs_devices->seed = root->fs_info->fs_devices->seed;
5560         root->fs_info->fs_devices->seed = fs_devices;
5561 out:
5562         return ret;
5563 }
5564
5565 static int read_one_dev(struct btrfs_root *root,
5566                         struct extent_buffer *leaf,
5567                         struct btrfs_dev_item *dev_item)
5568 {
5569         struct btrfs_device *device;
5570         u64 devid;
5571         int ret;
5572         u8 fs_uuid[BTRFS_UUID_SIZE];
5573         u8 dev_uuid[BTRFS_UUID_SIZE];
5574
5575         devid = btrfs_device_id(leaf, dev_item);
5576         read_extent_buffer(leaf, dev_uuid,
5577                            (unsigned long)btrfs_device_uuid(dev_item),
5578                            BTRFS_UUID_SIZE);
5579         read_extent_buffer(leaf, fs_uuid,
5580                            (unsigned long)btrfs_device_fsid(dev_item),
5581                            BTRFS_UUID_SIZE);
5582
5583         if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
5584                 ret = open_seed_devices(root, fs_uuid);
5585                 if (ret && !btrfs_test_opt(root, DEGRADED))
5586                         return ret;
5587         }
5588
5589         device = btrfs_find_device(root->fs_info, devid, dev_uuid, fs_uuid);
5590         if (!device || !device->bdev) {
5591                 if (!btrfs_test_opt(root, DEGRADED))
5592                         return -EIO;
5593
5594                 if (!device) {
5595                         btrfs_warn(root->fs_info, "devid %llu missing",
5596                                 (unsigned long long)devid);
5597                         device = add_missing_dev(root, devid, dev_uuid);
5598                         if (!device)
5599                                 return -ENOMEM;
5600                 } else if (!device->missing) {
5601                         /*
5602                          * this happens when a device that was properly setup
5603                          * in the device info lists suddenly goes bad.
5604                          * device->bdev is NULL, and so we have to set
5605                          * device->missing to one here
5606                          */
5607                         root->fs_info->fs_devices->missing_devices++;
5608                         device->missing = 1;
5609                 }
5610         }
5611
5612         if (device->fs_devices != root->fs_info->fs_devices) {
5613                 BUG_ON(device->writeable);
5614                 if (device->generation !=
5615                     btrfs_device_generation(leaf, dev_item))
5616                         return -EINVAL;
5617         }
5618
5619         fill_device_from_item(leaf, dev_item, device);
5620         device->dev_root = root->fs_info->dev_root;
5621         device->in_fs_metadata = 1;
5622         if (device->writeable && !device->is_tgtdev_for_dev_replace) {
5623                 device->fs_devices->total_rw_bytes += device->total_bytes;
5624                 spin_lock(&root->fs_info->free_chunk_lock);
5625                 root->fs_info->free_chunk_space += device->total_bytes -
5626                         device->bytes_used;
5627                 spin_unlock(&root->fs_info->free_chunk_lock);
5628         }
5629         ret = 0;
5630         return ret;
5631 }
5632
5633 int btrfs_read_sys_array(struct btrfs_root *root)
5634 {
5635         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
5636         struct extent_buffer *sb;
5637         struct btrfs_disk_key *disk_key;
5638         struct btrfs_chunk *chunk;
5639         u8 *ptr;
5640         unsigned long sb_ptr;
5641         int ret = 0;
5642         u32 num_stripes;
5643         u32 array_size;
5644         u32 len = 0;
5645         u32 cur;
5646         struct btrfs_key key;
5647
5648         sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
5649                                           BTRFS_SUPER_INFO_SIZE);
5650         if (!sb)
5651                 return -ENOMEM;
5652         btrfs_set_buffer_uptodate(sb);
5653         btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
5654         /*
5655          * The sb extent buffer is artifical and just used to read the system array.
5656          * btrfs_set_buffer_uptodate() call does not properly mark all it's
5657          * pages up-to-date when the page is larger: extent does not cover the
5658          * whole page and consequently check_page_uptodate does not find all
5659          * the page's extents up-to-date (the hole beyond sb),
5660          * write_extent_buffer then triggers a WARN_ON.
5661          *
5662          * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
5663          * but sb spans only this function. Add an explicit SetPageUptodate call
5664          * to silence the warning eg. on PowerPC 64.
5665          */
5666         if (PAGE_CACHE_SIZE > BTRFS_SUPER_INFO_SIZE)
5667                 SetPageUptodate(sb->pages[0]);
5668
5669         write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
5670         array_size = btrfs_super_sys_array_size(super_copy);
5671
5672         ptr = super_copy->sys_chunk_array;
5673         sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
5674         cur = 0;
5675
5676         while (cur < array_size) {
5677                 disk_key = (struct btrfs_disk_key *)ptr;
5678                 btrfs_disk_key_to_cpu(&key, disk_key);
5679
5680                 len = sizeof(*disk_key); ptr += len;
5681                 sb_ptr += len;
5682                 cur += len;
5683
5684                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
5685                         chunk = (struct btrfs_chunk *)sb_ptr;
5686                         ret = read_one_chunk(root, &key, sb, chunk);
5687                         if (ret)
5688                                 break;
5689                         num_stripes = btrfs_chunk_num_stripes(sb, chunk);
5690                         len = btrfs_chunk_item_size(num_stripes);
5691                 } else {
5692                         ret = -EIO;
5693                         break;
5694                 }
5695                 ptr += len;
5696                 sb_ptr += len;
5697                 cur += len;
5698         }
5699         free_extent_buffer(sb);
5700         return ret;
5701 }
5702
5703 int btrfs_read_chunk_tree(struct btrfs_root *root)
5704 {
5705         struct btrfs_path *path;
5706         struct extent_buffer *leaf;
5707         struct btrfs_key key;
5708         struct btrfs_key found_key;
5709         int ret;
5710         int slot;
5711
5712         root = root->fs_info->chunk_root;
5713
5714         path = btrfs_alloc_path();
5715         if (!path)
5716                 return -ENOMEM;
5717
5718         mutex_lock(&uuid_mutex);
5719         lock_chunks(root);
5720
5721         /* first we search for all of the device items, and then we
5722          * read in all of the chunk items.  This way we can create chunk
5723          * mappings that reference all of the devices that are afound
5724          */
5725         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
5726         key.offset = 0;
5727         key.type = 0;
5728 again:
5729         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5730         if (ret < 0)
5731                 goto error;
5732         while (1) {
5733                 leaf = path->nodes[0];
5734                 slot = path->slots[0];
5735                 if (slot >= btrfs_header_nritems(leaf)) {
5736                         ret = btrfs_next_leaf(root, path);
5737                         if (ret == 0)
5738                                 continue;
5739                         if (ret < 0)
5740                                 goto error;
5741                         break;
5742                 }
5743                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5744                 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
5745                         if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID)
5746                                 break;
5747                         if (found_key.type == BTRFS_DEV_ITEM_KEY) {
5748                                 struct btrfs_dev_item *dev_item;
5749                                 dev_item = btrfs_item_ptr(leaf, slot,
5750                                                   struct btrfs_dev_item);
5751                                 ret = read_one_dev(root, leaf, dev_item);
5752                                 if (ret)
5753                                         goto error;
5754                         }
5755                 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
5756                         struct btrfs_chunk *chunk;
5757                         chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
5758                         ret = read_one_chunk(root, &found_key, leaf, chunk);
5759                         if (ret)
5760                                 goto error;
5761                 }
5762                 path->slots[0]++;
5763         }
5764         if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
5765                 key.objectid = 0;
5766                 btrfs_release_path(path);
5767                 goto again;
5768         }
5769         ret = 0;
5770 error:
5771         unlock_chunks(root);
5772         mutex_unlock(&uuid_mutex);
5773
5774         btrfs_free_path(path);
5775         return ret;
5776 }
5777
5778 static void __btrfs_reset_dev_stats(struct btrfs_device *dev)
5779 {
5780         int i;
5781
5782         for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
5783                 btrfs_dev_stat_reset(dev, i);
5784 }
5785
5786 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
5787 {
5788         struct btrfs_key key;
5789         struct btrfs_key found_key;
5790         struct btrfs_root *dev_root = fs_info->dev_root;
5791         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
5792         struct extent_buffer *eb;
5793         int slot;
5794         int ret = 0;
5795         struct btrfs_device *device;
5796         struct btrfs_path *path = NULL;
5797         int i;
5798
5799         path = btrfs_alloc_path();
5800         if (!path) {
5801                 ret = -ENOMEM;
5802                 goto out;
5803         }
5804
5805         mutex_lock(&fs_devices->device_list_mutex);
5806         list_for_each_entry(device, &fs_devices->devices, dev_list) {
5807                 int item_size;
5808                 struct btrfs_dev_stats_item *ptr;
5809
5810                 key.objectid = 0;
5811                 key.type = BTRFS_DEV_STATS_KEY;
5812                 key.offset = device->devid;
5813                 ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
5814                 if (ret) {
5815                         __btrfs_reset_dev_stats(device);
5816                         device->dev_stats_valid = 1;
5817                         btrfs_release_path(path);
5818                         continue;
5819                 }
5820                 slot = path->slots[0];
5821                 eb = path->nodes[0];
5822                 btrfs_item_key_to_cpu(eb, &found_key, slot);
5823                 item_size = btrfs_item_size_nr(eb, slot);
5824
5825                 ptr = btrfs_item_ptr(eb, slot,
5826                                      struct btrfs_dev_stats_item);
5827
5828                 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
5829                         if (item_size >= (1 + i) * sizeof(__le64))
5830                                 btrfs_dev_stat_set(device, i,
5831                                         btrfs_dev_stats_value(eb, ptr, i));
5832                         else
5833                                 btrfs_dev_stat_reset(device, i);
5834                 }
5835
5836                 device->dev_stats_valid = 1;
5837                 btrfs_dev_stat_print_on_load(device);
5838                 btrfs_release_path(path);
5839         }
5840         mutex_unlock(&fs_devices->device_list_mutex);
5841
5842 out:
5843         btrfs_free_path(path);
5844         return ret < 0 ? ret : 0;
5845 }
5846
5847 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
5848                                 struct btrfs_root *dev_root,
5849                                 struct btrfs_device *device)
5850 {
5851         struct btrfs_path *path;
5852         struct btrfs_key key;
5853         struct extent_buffer *eb;
5854         struct btrfs_dev_stats_item *ptr;
5855         int ret;
5856         int i;
5857
5858         key.objectid = 0;
5859         key.type = BTRFS_DEV_STATS_KEY;
5860         key.offset = device->devid;
5861
5862         path = btrfs_alloc_path();
5863         BUG_ON(!path);
5864         ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
5865         if (ret < 0) {
5866                 printk_in_rcu(KERN_WARNING "btrfs: error %d while searching for dev_stats item for device %s!\n",
5867                               ret, rcu_str_deref(device->name));
5868                 goto out;
5869         }
5870
5871         if (ret == 0 &&
5872             btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
5873                 /* need to delete old one and insert a new one */
5874                 ret = btrfs_del_item(trans, dev_root, path);
5875                 if (ret != 0) {
5876                         printk_in_rcu(KERN_WARNING "btrfs: delete too small dev_stats item for device %s failed %d!\n",
5877                                       rcu_str_deref(device->name), ret);
5878                         goto out;
5879                 }
5880                 ret = 1;
5881         }
5882
5883         if (ret == 1) {
5884                 /* need to insert a new item */
5885                 btrfs_release_path(path);
5886                 ret = btrfs_insert_empty_item(trans, dev_root, path,
5887                                               &key, sizeof(*ptr));
5888                 if (ret < 0) {
5889                         printk_in_rcu(KERN_WARNING "btrfs: insert dev_stats item for device %s failed %d!\n",
5890                                       rcu_str_deref(device->name), ret);
5891                         goto out;
5892                 }
5893         }
5894
5895         eb = path->nodes[0];
5896         ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
5897         for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
5898                 btrfs_set_dev_stats_value(eb, ptr, i,
5899                                           btrfs_dev_stat_read(device, i));
5900         btrfs_mark_buffer_dirty(eb);
5901
5902 out:
5903         btrfs_free_path(path);
5904         return ret;
5905 }
5906
5907 /*
5908  * called from commit_transaction. Writes all changed device stats to disk.
5909  */
5910 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans,
5911                         struct btrfs_fs_info *fs_info)
5912 {
5913         struct btrfs_root *dev_root = fs_info->dev_root;
5914         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
5915         struct btrfs_device *device;
5916         int ret = 0;
5917
5918         mutex_lock(&fs_devices->device_list_mutex);
5919         list_for_each_entry(device, &fs_devices->devices, dev_list) {
5920                 if (!device->dev_stats_valid || !device->dev_stats_dirty)
5921                         continue;
5922
5923                 ret = update_dev_stat_item(trans, dev_root, device);
5924                 if (!ret)
5925                         device->dev_stats_dirty = 0;
5926         }
5927         mutex_unlock(&fs_devices->device_list_mutex);
5928
5929         return ret;
5930 }
5931
5932 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
5933 {
5934         btrfs_dev_stat_inc(dev, index);
5935         btrfs_dev_stat_print_on_error(dev);
5936 }
5937
5938 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
5939 {
5940         if (!dev->dev_stats_valid)
5941                 return;
5942         printk_ratelimited_in_rcu(KERN_ERR
5943                            "btrfs: bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
5944                            rcu_str_deref(dev->name),
5945                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
5946                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
5947                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
5948                            btrfs_dev_stat_read(dev,
5949                                                BTRFS_DEV_STAT_CORRUPTION_ERRS),
5950                            btrfs_dev_stat_read(dev,
5951                                                BTRFS_DEV_STAT_GENERATION_ERRS));
5952 }
5953
5954 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
5955 {
5956         int i;
5957
5958         for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
5959                 if (btrfs_dev_stat_read(dev, i) != 0)
5960                         break;
5961         if (i == BTRFS_DEV_STAT_VALUES_MAX)
5962                 return; /* all values == 0, suppress message */
5963
5964         printk_in_rcu(KERN_INFO "btrfs: bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
5965                rcu_str_deref(dev->name),
5966                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
5967                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
5968                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
5969                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
5970                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
5971 }
5972
5973 int btrfs_get_dev_stats(struct btrfs_root *root,
5974                         struct btrfs_ioctl_get_dev_stats *stats)
5975 {
5976         struct btrfs_device *dev;
5977         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
5978         int i;
5979
5980         mutex_lock(&fs_devices->device_list_mutex);
5981         dev = btrfs_find_device(root->fs_info, stats->devid, NULL, NULL);
5982         mutex_unlock(&fs_devices->device_list_mutex);
5983
5984         if (!dev) {
5985                 printk(KERN_WARNING
5986                        "btrfs: get dev_stats failed, device not found\n");
5987                 return -ENODEV;
5988         } else if (!dev->dev_stats_valid) {
5989                 printk(KERN_WARNING
5990                        "btrfs: get dev_stats failed, not yet valid\n");
5991                 return -ENODEV;
5992         } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
5993                 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
5994                         if (stats->nr_items > i)
5995                                 stats->values[i] =
5996                                         btrfs_dev_stat_read_and_reset(dev, i);
5997                         else
5998                                 btrfs_dev_stat_reset(dev, i);
5999                 }
6000         } else {
6001                 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6002                         if (stats->nr_items > i)
6003                                 stats->values[i] = btrfs_dev_stat_read(dev, i);
6004         }
6005         if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
6006                 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
6007         return 0;
6008 }
6009
6010 int btrfs_scratch_superblock(struct btrfs_device *device)
6011 {
6012         struct buffer_head *bh;
6013         struct btrfs_super_block *disk_super;
6014
6015         bh = btrfs_read_dev_super(device->bdev);
6016         if (!bh)
6017                 return -EINVAL;
6018         disk_super = (struct btrfs_super_block *)bh->b_data;
6019
6020         memset(&disk_super->magic, 0, sizeof(disk_super->magic));
6021         set_buffer_dirty(bh);
6022         sync_dirty_buffer(bh);
6023         brelse(bh);
6024
6025         return 0;
6026 }