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 static int btrfs_rm_dev_item(struct btrfs_root *root,
1388                              struct btrfs_device *device)
1389 {
1390         int ret;
1391         struct btrfs_path *path;
1392         struct btrfs_key key;
1393         struct btrfs_trans_handle *trans;
1394
1395         root = root->fs_info->chunk_root;
1396
1397         path = btrfs_alloc_path();
1398         if (!path)
1399                 return -ENOMEM;
1400
1401         trans = btrfs_start_transaction(root, 0);
1402         if (IS_ERR(trans)) {
1403                 btrfs_free_path(path);
1404                 return PTR_ERR(trans);
1405         }
1406         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1407         key.type = BTRFS_DEV_ITEM_KEY;
1408         key.offset = device->devid;
1409         lock_chunks(root);
1410
1411         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1412         if (ret < 0)
1413                 goto out;
1414
1415         if (ret > 0) {
1416                 ret = -ENOENT;
1417                 goto out;
1418         }
1419
1420         ret = btrfs_del_item(trans, root, path);
1421         if (ret)
1422                 goto out;
1423 out:
1424         btrfs_free_path(path);
1425         unlock_chunks(root);
1426         btrfs_commit_transaction(trans, root);
1427         return ret;
1428 }
1429
1430 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
1431 {
1432         struct btrfs_device *device;
1433         struct btrfs_device *next_device;
1434         struct block_device *bdev;
1435         struct buffer_head *bh = NULL;
1436         struct btrfs_super_block *disk_super;
1437         struct btrfs_fs_devices *cur_devices;
1438         u64 all_avail;
1439         u64 devid;
1440         u64 num_devices;
1441         u8 *dev_uuid;
1442         unsigned seq;
1443         int ret = 0;
1444         bool clear_super = false;
1445
1446         mutex_lock(&uuid_mutex);
1447
1448         do {
1449                 seq = read_seqbegin(&root->fs_info->profiles_lock);
1450
1451                 all_avail = root->fs_info->avail_data_alloc_bits |
1452                             root->fs_info->avail_system_alloc_bits |
1453                             root->fs_info->avail_metadata_alloc_bits;
1454         } while (read_seqretry(&root->fs_info->profiles_lock, seq));
1455
1456         num_devices = root->fs_info->fs_devices->num_devices;
1457         btrfs_dev_replace_lock(&root->fs_info->dev_replace);
1458         if (btrfs_dev_replace_is_ongoing(&root->fs_info->dev_replace)) {
1459                 WARN_ON(num_devices < 1);
1460                 num_devices--;
1461         }
1462         btrfs_dev_replace_unlock(&root->fs_info->dev_replace);
1463
1464         if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) && num_devices <= 4) {
1465                 printk(KERN_ERR "btrfs: unable to go below four devices "
1466                        "on raid10\n");
1467                 ret = -EINVAL;
1468                 goto out;
1469         }
1470
1471         if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) && num_devices <= 2) {
1472                 printk(KERN_ERR "btrfs: unable to go below two "
1473                        "devices on raid1\n");
1474                 ret = -EINVAL;
1475                 goto out;
1476         }
1477
1478         if ((all_avail & BTRFS_BLOCK_GROUP_RAID5) &&
1479             root->fs_info->fs_devices->rw_devices <= 2) {
1480                 printk(KERN_ERR "btrfs: unable to go below two "
1481                        "devices on raid5\n");
1482                 ret = -EINVAL;
1483                 goto out;
1484         }
1485         if ((all_avail & BTRFS_BLOCK_GROUP_RAID6) &&
1486             root->fs_info->fs_devices->rw_devices <= 3) {
1487                 printk(KERN_ERR "btrfs: unable to go below three "
1488                        "devices on raid6\n");
1489                 ret = -EINVAL;
1490                 goto out;
1491         }
1492
1493         if (strcmp(device_path, "missing") == 0) {
1494                 struct list_head *devices;
1495                 struct btrfs_device *tmp;
1496
1497                 device = NULL;
1498                 devices = &root->fs_info->fs_devices->devices;
1499                 /*
1500                  * It is safe to read the devices since the volume_mutex
1501                  * is held.
1502                  */
1503                 list_for_each_entry(tmp, devices, dev_list) {
1504                         if (tmp->in_fs_metadata &&
1505                             !tmp->is_tgtdev_for_dev_replace &&
1506                             !tmp->bdev) {
1507                                 device = tmp;
1508                                 break;
1509                         }
1510                 }
1511                 bdev = NULL;
1512                 bh = NULL;
1513                 disk_super = NULL;
1514                 if (!device) {
1515                         printk(KERN_ERR "btrfs: no missing devices found to "
1516                                "remove\n");
1517                         goto out;
1518                 }
1519         } else {
1520                 ret = btrfs_get_bdev_and_sb(device_path,
1521                                             FMODE_WRITE | FMODE_EXCL,
1522                                             root->fs_info->bdev_holder, 0,
1523                                             &bdev, &bh);
1524                 if (ret)
1525                         goto out;
1526                 disk_super = (struct btrfs_super_block *)bh->b_data;
1527                 devid = btrfs_stack_device_id(&disk_super->dev_item);
1528                 dev_uuid = disk_super->dev_item.uuid;
1529                 device = btrfs_find_device(root->fs_info, devid, dev_uuid,
1530                                            disk_super->fsid);
1531                 if (!device) {
1532                         ret = -ENOENT;
1533                         goto error_brelse;
1534                 }
1535         }
1536
1537         if (device->is_tgtdev_for_dev_replace) {
1538                 pr_err("btrfs: unable to remove the dev_replace target dev\n");
1539                 ret = -EINVAL;
1540                 goto error_brelse;
1541         }
1542
1543         if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1544                 printk(KERN_ERR "btrfs: unable to remove the only writeable "
1545                        "device\n");
1546                 ret = -EINVAL;
1547                 goto error_brelse;
1548         }
1549
1550         if (device->writeable) {
1551                 lock_chunks(root);
1552                 list_del_init(&device->dev_alloc_list);
1553                 unlock_chunks(root);
1554                 root->fs_info->fs_devices->rw_devices--;
1555                 clear_super = true;
1556         }
1557
1558         ret = btrfs_shrink_device(device, 0);
1559         if (ret)
1560                 goto error_undo;
1561
1562         /*
1563          * TODO: the superblock still includes this device in its num_devices
1564          * counter although write_all_supers() is not locked out. This
1565          * could give a filesystem state which requires a degraded mount.
1566          */
1567         ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1568         if (ret)
1569                 goto error_undo;
1570
1571         spin_lock(&root->fs_info->free_chunk_lock);
1572         root->fs_info->free_chunk_space = device->total_bytes -
1573                 device->bytes_used;
1574         spin_unlock(&root->fs_info->free_chunk_lock);
1575
1576         device->in_fs_metadata = 0;
1577         btrfs_scrub_cancel_dev(root->fs_info, device);
1578
1579         /*
1580          * the device list mutex makes sure that we don't change
1581          * the device list while someone else is writing out all
1582          * the device supers.
1583          */
1584
1585         cur_devices = device->fs_devices;
1586         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1587         list_del_rcu(&device->dev_list);
1588
1589         device->fs_devices->num_devices--;
1590         device->fs_devices->total_devices--;
1591
1592         if (device->missing)
1593                 root->fs_info->fs_devices->missing_devices--;
1594
1595         next_device = list_entry(root->fs_info->fs_devices->devices.next,
1596                                  struct btrfs_device, dev_list);
1597         if (device->bdev == root->fs_info->sb->s_bdev)
1598                 root->fs_info->sb->s_bdev = next_device->bdev;
1599         if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1600                 root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1601
1602         if (device->bdev)
1603                 device->fs_devices->open_devices--;
1604
1605         call_rcu(&device->rcu, free_device);
1606         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1607
1608         num_devices = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
1609         btrfs_set_super_num_devices(root->fs_info->super_copy, num_devices);
1610
1611         if (cur_devices->open_devices == 0) {
1612                 struct btrfs_fs_devices *fs_devices;
1613                 fs_devices = root->fs_info->fs_devices;
1614                 while (fs_devices) {
1615                         if (fs_devices->seed == cur_devices)
1616                                 break;
1617                         fs_devices = fs_devices->seed;
1618                 }
1619                 fs_devices->seed = cur_devices->seed;
1620                 cur_devices->seed = NULL;
1621                 lock_chunks(root);
1622                 __btrfs_close_devices(cur_devices);
1623                 unlock_chunks(root);
1624                 free_fs_devices(cur_devices);
1625         }
1626
1627         root->fs_info->num_tolerated_disk_barrier_failures =
1628                 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
1629
1630         /*
1631          * at this point, the device is zero sized.  We want to
1632          * remove it from the devices list and zero out the old super
1633          */
1634         if (clear_super && disk_super) {
1635                 /* make sure this device isn't detected as part of
1636                  * the FS anymore
1637                  */
1638                 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1639                 set_buffer_dirty(bh);
1640                 sync_dirty_buffer(bh);
1641         }
1642
1643         ret = 0;
1644
1645         /* Notify udev that device has changed */
1646         if (bdev)
1647                 btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
1648
1649 error_brelse:
1650         brelse(bh);
1651         if (bdev)
1652                 blkdev_put(bdev, FMODE_READ | FMODE_EXCL);
1653 out:
1654         mutex_unlock(&uuid_mutex);
1655         return ret;
1656 error_undo:
1657         if (device->writeable) {
1658                 lock_chunks(root);
1659                 list_add(&device->dev_alloc_list,
1660                          &root->fs_info->fs_devices->alloc_list);
1661                 unlock_chunks(root);
1662                 root->fs_info->fs_devices->rw_devices++;
1663         }
1664         goto error_brelse;
1665 }
1666
1667 void btrfs_rm_dev_replace_srcdev(struct btrfs_fs_info *fs_info,
1668                                  struct btrfs_device *srcdev)
1669 {
1670         WARN_ON(!mutex_is_locked(&fs_info->fs_devices->device_list_mutex));
1671         list_del_rcu(&srcdev->dev_list);
1672         list_del_rcu(&srcdev->dev_alloc_list);
1673         fs_info->fs_devices->num_devices--;
1674         if (srcdev->missing) {
1675                 fs_info->fs_devices->missing_devices--;
1676                 fs_info->fs_devices->rw_devices++;
1677         }
1678         if (srcdev->can_discard)
1679                 fs_info->fs_devices->num_can_discard--;
1680         if (srcdev->bdev)
1681                 fs_info->fs_devices->open_devices--;
1682
1683         call_rcu(&srcdev->rcu, free_device);
1684 }
1685
1686 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
1687                                       struct btrfs_device *tgtdev)
1688 {
1689         struct btrfs_device *next_device;
1690
1691         WARN_ON(!tgtdev);
1692         mutex_lock(&fs_info->fs_devices->device_list_mutex);
1693         if (tgtdev->bdev) {
1694                 btrfs_scratch_superblock(tgtdev);
1695                 fs_info->fs_devices->open_devices--;
1696         }
1697         fs_info->fs_devices->num_devices--;
1698         if (tgtdev->can_discard)
1699                 fs_info->fs_devices->num_can_discard++;
1700
1701         next_device = list_entry(fs_info->fs_devices->devices.next,
1702                                  struct btrfs_device, dev_list);
1703         if (tgtdev->bdev == fs_info->sb->s_bdev)
1704                 fs_info->sb->s_bdev = next_device->bdev;
1705         if (tgtdev->bdev == fs_info->fs_devices->latest_bdev)
1706                 fs_info->fs_devices->latest_bdev = next_device->bdev;
1707         list_del_rcu(&tgtdev->dev_list);
1708
1709         call_rcu(&tgtdev->rcu, free_device);
1710
1711         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
1712 }
1713
1714 static int btrfs_find_device_by_path(struct btrfs_root *root, char *device_path,
1715                                      struct btrfs_device **device)
1716 {
1717         int ret = 0;
1718         struct btrfs_super_block *disk_super;
1719         u64 devid;
1720         u8 *dev_uuid;
1721         struct block_device *bdev;
1722         struct buffer_head *bh;
1723
1724         *device = NULL;
1725         ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
1726                                     root->fs_info->bdev_holder, 0, &bdev, &bh);
1727         if (ret)
1728                 return ret;
1729         disk_super = (struct btrfs_super_block *)bh->b_data;
1730         devid = btrfs_stack_device_id(&disk_super->dev_item);
1731         dev_uuid = disk_super->dev_item.uuid;
1732         *device = btrfs_find_device(root->fs_info, devid, dev_uuid,
1733                                     disk_super->fsid);
1734         brelse(bh);
1735         if (!*device)
1736                 ret = -ENOENT;
1737         blkdev_put(bdev, FMODE_READ);
1738         return ret;
1739 }
1740
1741 int btrfs_find_device_missing_or_by_path(struct btrfs_root *root,
1742                                          char *device_path,
1743                                          struct btrfs_device **device)
1744 {
1745         *device = NULL;
1746         if (strcmp(device_path, "missing") == 0) {
1747                 struct list_head *devices;
1748                 struct btrfs_device *tmp;
1749
1750                 devices = &root->fs_info->fs_devices->devices;
1751                 /*
1752                  * It is safe to read the devices since the volume_mutex
1753                  * is held by the caller.
1754                  */
1755                 list_for_each_entry(tmp, devices, dev_list) {
1756                         if (tmp->in_fs_metadata && !tmp->bdev) {
1757                                 *device = tmp;
1758                                 break;
1759                         }
1760                 }
1761
1762                 if (!*device) {
1763                         pr_err("btrfs: no missing device found\n");
1764                         return -ENOENT;
1765                 }
1766
1767                 return 0;
1768         } else {
1769                 return btrfs_find_device_by_path(root, device_path, device);
1770         }
1771 }
1772
1773 /*
1774  * does all the dirty work required for changing file system's UUID.
1775  */
1776 static int btrfs_prepare_sprout(struct btrfs_root *root)
1777 {
1778         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
1779         struct btrfs_fs_devices *old_devices;
1780         struct btrfs_fs_devices *seed_devices;
1781         struct btrfs_super_block *disk_super = root->fs_info->super_copy;
1782         struct btrfs_device *device;
1783         u64 super_flags;
1784
1785         BUG_ON(!mutex_is_locked(&uuid_mutex));
1786         if (!fs_devices->seeding)
1787                 return -EINVAL;
1788
1789         seed_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
1790         if (!seed_devices)
1791                 return -ENOMEM;
1792
1793         old_devices = clone_fs_devices(fs_devices);
1794         if (IS_ERR(old_devices)) {
1795                 kfree(seed_devices);
1796                 return PTR_ERR(old_devices);
1797         }
1798
1799         list_add(&old_devices->list, &fs_uuids);
1800
1801         memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
1802         seed_devices->opened = 1;
1803         INIT_LIST_HEAD(&seed_devices->devices);
1804         INIT_LIST_HEAD(&seed_devices->alloc_list);
1805         mutex_init(&seed_devices->device_list_mutex);
1806
1807         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1808         list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
1809                               synchronize_rcu);
1810         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1811
1812         list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
1813         list_for_each_entry(device, &seed_devices->devices, dev_list) {
1814                 device->fs_devices = seed_devices;
1815         }
1816
1817         fs_devices->seeding = 0;
1818         fs_devices->num_devices = 0;
1819         fs_devices->open_devices = 0;
1820         fs_devices->total_devices = 0;
1821         fs_devices->seed = seed_devices;
1822
1823         generate_random_uuid(fs_devices->fsid);
1824         memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1825         memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1826         super_flags = btrfs_super_flags(disk_super) &
1827                       ~BTRFS_SUPER_FLAG_SEEDING;
1828         btrfs_set_super_flags(disk_super, super_flags);
1829
1830         return 0;
1831 }
1832
1833 /*
1834  * strore the expected generation for seed devices in device items.
1835  */
1836 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
1837                                struct btrfs_root *root)
1838 {
1839         struct btrfs_path *path;
1840         struct extent_buffer *leaf;
1841         struct btrfs_dev_item *dev_item;
1842         struct btrfs_device *device;
1843         struct btrfs_key key;
1844         u8 fs_uuid[BTRFS_UUID_SIZE];
1845         u8 dev_uuid[BTRFS_UUID_SIZE];
1846         u64 devid;
1847         int ret;
1848
1849         path = btrfs_alloc_path();
1850         if (!path)
1851                 return -ENOMEM;
1852
1853         root = root->fs_info->chunk_root;
1854         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1855         key.offset = 0;
1856         key.type = BTRFS_DEV_ITEM_KEY;
1857
1858         while (1) {
1859                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1860                 if (ret < 0)
1861                         goto error;
1862
1863                 leaf = path->nodes[0];
1864 next_slot:
1865                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1866                         ret = btrfs_next_leaf(root, path);
1867                         if (ret > 0)
1868                                 break;
1869                         if (ret < 0)
1870                                 goto error;
1871                         leaf = path->nodes[0];
1872                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1873                         btrfs_release_path(path);
1874                         continue;
1875                 }
1876
1877                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1878                 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
1879                     key.type != BTRFS_DEV_ITEM_KEY)
1880                         break;
1881
1882                 dev_item = btrfs_item_ptr(leaf, path->slots[0],
1883                                           struct btrfs_dev_item);
1884                 devid = btrfs_device_id(leaf, dev_item);
1885                 read_extent_buffer(leaf, dev_uuid,
1886                                    (unsigned long)btrfs_device_uuid(dev_item),
1887                                    BTRFS_UUID_SIZE);
1888                 read_extent_buffer(leaf, fs_uuid,
1889                                    (unsigned long)btrfs_device_fsid(dev_item),
1890                                    BTRFS_UUID_SIZE);
1891                 device = btrfs_find_device(root->fs_info, devid, dev_uuid,
1892                                            fs_uuid);
1893                 BUG_ON(!device); /* Logic error */
1894
1895                 if (device->fs_devices->seeding) {
1896                         btrfs_set_device_generation(leaf, dev_item,
1897                                                     device->generation);
1898                         btrfs_mark_buffer_dirty(leaf);
1899                 }
1900
1901                 path->slots[0]++;
1902                 goto next_slot;
1903         }
1904         ret = 0;
1905 error:
1906         btrfs_free_path(path);
1907         return ret;
1908 }
1909
1910 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
1911 {
1912         struct request_queue *q;
1913         struct btrfs_trans_handle *trans;
1914         struct btrfs_device *device;
1915         struct block_device *bdev;
1916         struct list_head *devices;
1917         struct super_block *sb = root->fs_info->sb;
1918         struct rcu_string *name;
1919         u64 total_bytes;
1920         int seeding_dev = 0;
1921         int ret = 0;
1922
1923         if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
1924                 return -EROFS;
1925
1926         bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
1927                                   root->fs_info->bdev_holder);
1928         if (IS_ERR(bdev))
1929                 return PTR_ERR(bdev);
1930
1931         if (root->fs_info->fs_devices->seeding) {
1932                 seeding_dev = 1;
1933                 down_write(&sb->s_umount);
1934                 mutex_lock(&uuid_mutex);
1935         }
1936
1937         filemap_write_and_wait(bdev->bd_inode->i_mapping);
1938
1939         devices = &root->fs_info->fs_devices->devices;
1940
1941         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1942         list_for_each_entry(device, devices, dev_list) {
1943                 if (device->bdev == bdev) {
1944                         ret = -EEXIST;
1945                         mutex_unlock(
1946                                 &root->fs_info->fs_devices->device_list_mutex);
1947                         goto error;
1948                 }
1949         }
1950         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1951
1952         device = kzalloc(sizeof(*device), GFP_NOFS);
1953         if (!device) {
1954                 /* we can safely leave the fs_devices entry around */
1955                 ret = -ENOMEM;
1956                 goto error;
1957         }
1958
1959         name = rcu_string_strdup(device_path, GFP_NOFS);
1960         if (!name) {
1961                 kfree(device);
1962                 ret = -ENOMEM;
1963                 goto error;
1964         }
1965         rcu_assign_pointer(device->name, name);
1966
1967         ret = find_next_devid(root, &device->devid);
1968         if (ret) {
1969                 rcu_string_free(device->name);
1970                 kfree(device);
1971                 goto error;
1972         }
1973
1974         trans = btrfs_start_transaction(root, 0);
1975         if (IS_ERR(trans)) {
1976                 rcu_string_free(device->name);
1977                 kfree(device);
1978                 ret = PTR_ERR(trans);
1979                 goto error;
1980         }
1981
1982         lock_chunks(root);
1983
1984         q = bdev_get_queue(bdev);
1985         if (blk_queue_discard(q))
1986                 device->can_discard = 1;
1987         device->writeable = 1;
1988         device->work.func = pending_bios_fn;
1989         generate_random_uuid(device->uuid);
1990         spin_lock_init(&device->io_lock);
1991         device->generation = trans->transid;
1992         device->io_width = root->sectorsize;
1993         device->io_align = root->sectorsize;
1994         device->sector_size = root->sectorsize;
1995         device->total_bytes = i_size_read(bdev->bd_inode);
1996         device->disk_total_bytes = device->total_bytes;
1997         device->dev_root = root->fs_info->dev_root;
1998         device->bdev = bdev;
1999         device->in_fs_metadata = 1;
2000         device->is_tgtdev_for_dev_replace = 0;
2001         device->mode = FMODE_EXCL;
2002         set_blocksize(device->bdev, 4096);
2003
2004         if (seeding_dev) {
2005                 sb->s_flags &= ~MS_RDONLY;
2006                 ret = btrfs_prepare_sprout(root);
2007                 BUG_ON(ret); /* -ENOMEM */
2008         }
2009
2010         device->fs_devices = root->fs_info->fs_devices;
2011
2012         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2013         list_add_rcu(&device->dev_list, &root->fs_info->fs_devices->devices);
2014         list_add(&device->dev_alloc_list,
2015                  &root->fs_info->fs_devices->alloc_list);
2016         root->fs_info->fs_devices->num_devices++;
2017         root->fs_info->fs_devices->open_devices++;
2018         root->fs_info->fs_devices->rw_devices++;
2019         root->fs_info->fs_devices->total_devices++;
2020         if (device->can_discard)
2021                 root->fs_info->fs_devices->num_can_discard++;
2022         root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
2023
2024         spin_lock(&root->fs_info->free_chunk_lock);
2025         root->fs_info->free_chunk_space += device->total_bytes;
2026         spin_unlock(&root->fs_info->free_chunk_lock);
2027
2028         if (!blk_queue_nonrot(bdev_get_queue(bdev)))
2029                 root->fs_info->fs_devices->rotating = 1;
2030
2031         total_bytes = btrfs_super_total_bytes(root->fs_info->super_copy);
2032         btrfs_set_super_total_bytes(root->fs_info->super_copy,
2033                                     total_bytes + device->total_bytes);
2034
2035         total_bytes = btrfs_super_num_devices(root->fs_info->super_copy);
2036         btrfs_set_super_num_devices(root->fs_info->super_copy,
2037                                     total_bytes + 1);
2038         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2039
2040         if (seeding_dev) {
2041                 ret = init_first_rw_device(trans, root, device);
2042                 if (ret) {
2043                         btrfs_abort_transaction(trans, root, ret);
2044                         goto error_trans;
2045                 }
2046                 ret = btrfs_finish_sprout(trans, root);
2047                 if (ret) {
2048                         btrfs_abort_transaction(trans, root, ret);
2049                         goto error_trans;
2050                 }
2051         } else {
2052                 ret = btrfs_add_device(trans, root, device);
2053                 if (ret) {
2054                         btrfs_abort_transaction(trans, root, ret);
2055                         goto error_trans;
2056                 }
2057         }
2058
2059         /*
2060          * we've got more storage, clear any full flags on the space
2061          * infos
2062          */
2063         btrfs_clear_space_info_full(root->fs_info);
2064
2065         unlock_chunks(root);
2066         root->fs_info->num_tolerated_disk_barrier_failures =
2067                 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
2068         ret = btrfs_commit_transaction(trans, root);
2069
2070         if (seeding_dev) {
2071                 mutex_unlock(&uuid_mutex);
2072                 up_write(&sb->s_umount);
2073
2074                 if (ret) /* transaction commit */
2075                         return ret;
2076
2077                 ret = btrfs_relocate_sys_chunks(root);
2078                 if (ret < 0)
2079                         btrfs_error(root->fs_info, ret,
2080                                     "Failed to relocate sys chunks after "
2081                                     "device initialization. This can be fixed "
2082                                     "using the \"btrfs balance\" command.");
2083                 trans = btrfs_attach_transaction(root);
2084                 if (IS_ERR(trans)) {
2085                         if (PTR_ERR(trans) == -ENOENT)
2086                                 return 0;
2087                         return PTR_ERR(trans);
2088                 }
2089                 ret = btrfs_commit_transaction(trans, root);
2090         }
2091
2092         return ret;
2093
2094 error_trans:
2095         unlock_chunks(root);
2096         btrfs_end_transaction(trans, root);
2097         rcu_string_free(device->name);
2098         kfree(device);
2099 error:
2100         blkdev_put(bdev, FMODE_EXCL);
2101         if (seeding_dev) {
2102                 mutex_unlock(&uuid_mutex);
2103                 up_write(&sb->s_umount);
2104         }
2105         return ret;
2106 }
2107
2108 int btrfs_init_dev_replace_tgtdev(struct btrfs_root *root, char *device_path,
2109                                   struct btrfs_device **device_out)
2110 {
2111         struct request_queue *q;
2112         struct btrfs_device *device;
2113         struct block_device *bdev;
2114         struct btrfs_fs_info *fs_info = root->fs_info;
2115         struct list_head *devices;
2116         struct rcu_string *name;
2117         int ret = 0;
2118
2119         *device_out = NULL;
2120         if (fs_info->fs_devices->seeding)
2121                 return -EINVAL;
2122
2123         bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2124                                   fs_info->bdev_holder);
2125         if (IS_ERR(bdev))
2126                 return PTR_ERR(bdev);
2127
2128         filemap_write_and_wait(bdev->bd_inode->i_mapping);
2129
2130         devices = &fs_info->fs_devices->devices;
2131         list_for_each_entry(device, devices, dev_list) {
2132                 if (device->bdev == bdev) {
2133                         ret = -EEXIST;
2134                         goto error;
2135                 }
2136         }
2137
2138         device = kzalloc(sizeof(*device), GFP_NOFS);
2139         if (!device) {
2140                 ret = -ENOMEM;
2141                 goto error;
2142         }
2143
2144         name = rcu_string_strdup(device_path, GFP_NOFS);
2145         if (!name) {
2146                 kfree(device);
2147                 ret = -ENOMEM;
2148                 goto error;
2149         }
2150         rcu_assign_pointer(device->name, name);
2151
2152         q = bdev_get_queue(bdev);
2153         if (blk_queue_discard(q))
2154                 device->can_discard = 1;
2155         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2156         device->writeable = 1;
2157         device->work.func = pending_bios_fn;
2158         generate_random_uuid(device->uuid);
2159         device->devid = BTRFS_DEV_REPLACE_DEVID;
2160         spin_lock_init(&device->io_lock);
2161         device->generation = 0;
2162         device->io_width = root->sectorsize;
2163         device->io_align = root->sectorsize;
2164         device->sector_size = root->sectorsize;
2165         device->total_bytes = i_size_read(bdev->bd_inode);
2166         device->disk_total_bytes = device->total_bytes;
2167         device->dev_root = fs_info->dev_root;
2168         device->bdev = bdev;
2169         device->in_fs_metadata = 1;
2170         device->is_tgtdev_for_dev_replace = 1;
2171         device->mode = FMODE_EXCL;
2172         set_blocksize(device->bdev, 4096);
2173         device->fs_devices = fs_info->fs_devices;
2174         list_add(&device->dev_list, &fs_info->fs_devices->devices);
2175         fs_info->fs_devices->num_devices++;
2176         fs_info->fs_devices->open_devices++;
2177         if (device->can_discard)
2178                 fs_info->fs_devices->num_can_discard++;
2179         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2180
2181         *device_out = device;
2182         return ret;
2183
2184 error:
2185         blkdev_put(bdev, FMODE_EXCL);
2186         return ret;
2187 }
2188
2189 void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info *fs_info,
2190                                               struct btrfs_device *tgtdev)
2191 {
2192         WARN_ON(fs_info->fs_devices->rw_devices == 0);
2193         tgtdev->io_width = fs_info->dev_root->sectorsize;
2194         tgtdev->io_align = fs_info->dev_root->sectorsize;
2195         tgtdev->sector_size = fs_info->dev_root->sectorsize;
2196         tgtdev->dev_root = fs_info->dev_root;
2197         tgtdev->in_fs_metadata = 1;
2198 }
2199
2200 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2201                                         struct btrfs_device *device)
2202 {
2203         int ret;
2204         struct btrfs_path *path;
2205         struct btrfs_root *root;
2206         struct btrfs_dev_item *dev_item;
2207         struct extent_buffer *leaf;
2208         struct btrfs_key key;
2209
2210         root = device->dev_root->fs_info->chunk_root;
2211
2212         path = btrfs_alloc_path();
2213         if (!path)
2214                 return -ENOMEM;
2215
2216         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2217         key.type = BTRFS_DEV_ITEM_KEY;
2218         key.offset = device->devid;
2219
2220         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2221         if (ret < 0)
2222                 goto out;
2223
2224         if (ret > 0) {
2225                 ret = -ENOENT;
2226                 goto out;
2227         }
2228
2229         leaf = path->nodes[0];
2230         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2231
2232         btrfs_set_device_id(leaf, dev_item, device->devid);
2233         btrfs_set_device_type(leaf, dev_item, device->type);
2234         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2235         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2236         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2237         btrfs_set_device_total_bytes(leaf, dev_item, device->disk_total_bytes);
2238         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
2239         btrfs_mark_buffer_dirty(leaf);
2240
2241 out:
2242         btrfs_free_path(path);
2243         return ret;
2244 }
2245
2246 static int __btrfs_grow_device(struct btrfs_trans_handle *trans,
2247                       struct btrfs_device *device, u64 new_size)
2248 {
2249         struct btrfs_super_block *super_copy =
2250                 device->dev_root->fs_info->super_copy;
2251         u64 old_total = btrfs_super_total_bytes(super_copy);
2252         u64 diff = new_size - device->total_bytes;
2253
2254         if (!device->writeable)
2255                 return -EACCES;
2256         if (new_size <= device->total_bytes ||
2257             device->is_tgtdev_for_dev_replace)
2258                 return -EINVAL;
2259
2260         btrfs_set_super_total_bytes(super_copy, old_total + diff);
2261         device->fs_devices->total_rw_bytes += diff;
2262
2263         device->total_bytes = new_size;
2264         device->disk_total_bytes = new_size;
2265         btrfs_clear_space_info_full(device->dev_root->fs_info);
2266
2267         return btrfs_update_device(trans, device);
2268 }
2269
2270 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2271                       struct btrfs_device *device, u64 new_size)
2272 {
2273         int ret;
2274         lock_chunks(device->dev_root);
2275         ret = __btrfs_grow_device(trans, device, new_size);
2276         unlock_chunks(device->dev_root);
2277         return ret;
2278 }
2279
2280 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
2281                             struct btrfs_root *root,
2282                             u64 chunk_tree, u64 chunk_objectid,
2283                             u64 chunk_offset)
2284 {
2285         int ret;
2286         struct btrfs_path *path;
2287         struct btrfs_key key;
2288
2289         root = root->fs_info->chunk_root;
2290         path = btrfs_alloc_path();
2291         if (!path)
2292                 return -ENOMEM;
2293
2294         key.objectid = chunk_objectid;
2295         key.offset = chunk_offset;
2296         key.type = BTRFS_CHUNK_ITEM_KEY;
2297
2298         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2299         if (ret < 0)
2300                 goto out;
2301         else if (ret > 0) { /* Logic error or corruption */
2302                 btrfs_error(root->fs_info, -ENOENT,
2303                             "Failed lookup while freeing chunk.");
2304                 ret = -ENOENT;
2305                 goto out;
2306         }
2307
2308         ret = btrfs_del_item(trans, root, path);
2309         if (ret < 0)
2310                 btrfs_error(root->fs_info, ret,
2311                             "Failed to delete chunk item.");
2312 out:
2313         btrfs_free_path(path);
2314         return ret;
2315 }
2316
2317 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
2318                         chunk_offset)
2319 {
2320         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
2321         struct btrfs_disk_key *disk_key;
2322         struct btrfs_chunk *chunk;
2323         u8 *ptr;
2324         int ret = 0;
2325         u32 num_stripes;
2326         u32 array_size;
2327         u32 len = 0;
2328         u32 cur;
2329         struct btrfs_key key;
2330
2331         array_size = btrfs_super_sys_array_size(super_copy);
2332
2333         ptr = super_copy->sys_chunk_array;
2334         cur = 0;
2335
2336         while (cur < array_size) {
2337                 disk_key = (struct btrfs_disk_key *)ptr;
2338                 btrfs_disk_key_to_cpu(&key, disk_key);
2339
2340                 len = sizeof(*disk_key);
2341
2342                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2343                         chunk = (struct btrfs_chunk *)(ptr + len);
2344                         num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2345                         len += btrfs_chunk_item_size(num_stripes);
2346                 } else {
2347                         ret = -EIO;
2348                         break;
2349                 }
2350                 if (key.objectid == chunk_objectid &&
2351                     key.offset == chunk_offset) {
2352                         memmove(ptr, ptr + len, array_size - (cur + len));
2353                         array_size -= len;
2354                         btrfs_set_super_sys_array_size(super_copy, array_size);
2355                 } else {
2356                         ptr += len;
2357                         cur += len;
2358                 }
2359         }
2360         return ret;
2361 }
2362
2363 static int btrfs_relocate_chunk(struct btrfs_root *root,
2364                          u64 chunk_tree, u64 chunk_objectid,
2365                          u64 chunk_offset)
2366 {
2367         struct extent_map_tree *em_tree;
2368         struct btrfs_root *extent_root;
2369         struct btrfs_trans_handle *trans;
2370         struct extent_map *em;
2371         struct map_lookup *map;
2372         int ret;
2373         int i;
2374
2375         root = root->fs_info->chunk_root;
2376         extent_root = root->fs_info->extent_root;
2377         em_tree = &root->fs_info->mapping_tree.map_tree;
2378
2379         ret = btrfs_can_relocate(extent_root, chunk_offset);
2380         if (ret)
2381                 return -ENOSPC;
2382
2383         /* step one, relocate all the extents inside this chunk */
2384         ret = btrfs_relocate_block_group(extent_root, chunk_offset);
2385         if (ret)
2386                 return ret;
2387
2388         trans = btrfs_start_transaction(root, 0);
2389         if (IS_ERR(trans)) {
2390                 ret = PTR_ERR(trans);
2391                 btrfs_std_error(root->fs_info, ret);
2392                 return ret;
2393         }
2394
2395         lock_chunks(root);
2396
2397         /*
2398          * step two, delete the device extents and the
2399          * chunk tree entries
2400          */
2401         read_lock(&em_tree->lock);
2402         em = lookup_extent_mapping(em_tree, chunk_offset, 1);
2403         read_unlock(&em_tree->lock);
2404
2405         BUG_ON(!em || em->start > chunk_offset ||
2406                em->start + em->len < chunk_offset);
2407         map = (struct map_lookup *)em->bdev;
2408
2409         for (i = 0; i < map->num_stripes; i++) {
2410                 ret = btrfs_free_dev_extent(trans, map->stripes[i].dev,
2411                                             map->stripes[i].physical);
2412                 BUG_ON(ret);
2413
2414                 if (map->stripes[i].dev) {
2415                         ret = btrfs_update_device(trans, map->stripes[i].dev);
2416                         BUG_ON(ret);
2417                 }
2418         }
2419         ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
2420                                chunk_offset);
2421
2422         BUG_ON(ret);
2423
2424         trace_btrfs_chunk_free(root, map, chunk_offset, em->len);
2425
2426         if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2427                 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
2428                 BUG_ON(ret);
2429         }
2430
2431         ret = btrfs_remove_block_group(trans, extent_root, chunk_offset);
2432         BUG_ON(ret);
2433
2434         write_lock(&em_tree->lock);
2435         remove_extent_mapping(em_tree, em);
2436         write_unlock(&em_tree->lock);
2437
2438         kfree(map);
2439         em->bdev = NULL;
2440
2441         /* once for the tree */
2442         free_extent_map(em);
2443         /* once for us */
2444         free_extent_map(em);
2445
2446         unlock_chunks(root);
2447         btrfs_end_transaction(trans, root);
2448         return 0;
2449 }
2450
2451 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
2452 {
2453         struct btrfs_root *chunk_root = root->fs_info->chunk_root;
2454         struct btrfs_path *path;
2455         struct extent_buffer *leaf;
2456         struct btrfs_chunk *chunk;
2457         struct btrfs_key key;
2458         struct btrfs_key found_key;
2459         u64 chunk_tree = chunk_root->root_key.objectid;
2460         u64 chunk_type;
2461         bool retried = false;
2462         int failed = 0;
2463         int ret;
2464
2465         path = btrfs_alloc_path();
2466         if (!path)
2467                 return -ENOMEM;
2468
2469 again:
2470         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2471         key.offset = (u64)-1;
2472         key.type = BTRFS_CHUNK_ITEM_KEY;
2473
2474         while (1) {
2475                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2476                 if (ret < 0)
2477                         goto error;
2478                 BUG_ON(ret == 0); /* Corruption */
2479
2480                 ret = btrfs_previous_item(chunk_root, path, key.objectid,
2481                                           key.type);
2482                 if (ret < 0)
2483                         goto error;
2484                 if (ret > 0)
2485                         break;
2486
2487                 leaf = path->nodes[0];
2488                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2489
2490                 chunk = btrfs_item_ptr(leaf, path->slots[0],
2491                                        struct btrfs_chunk);
2492                 chunk_type = btrfs_chunk_type(leaf, chunk);
2493                 btrfs_release_path(path);
2494
2495                 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
2496                         ret = btrfs_relocate_chunk(chunk_root, chunk_tree,
2497                                                    found_key.objectid,
2498                                                    found_key.offset);
2499                         if (ret == -ENOSPC)
2500                                 failed++;
2501                         else if (ret)
2502                                 BUG();
2503                 }
2504
2505                 if (found_key.offset == 0)
2506                         break;
2507                 key.offset = found_key.offset - 1;
2508         }
2509         ret = 0;
2510         if (failed && !retried) {
2511                 failed = 0;
2512                 retried = true;
2513                 goto again;
2514         } else if (failed && retried) {
2515                 WARN_ON(1);
2516                 ret = -ENOSPC;
2517         }
2518 error:
2519         btrfs_free_path(path);
2520         return ret;
2521 }
2522
2523 static int insert_balance_item(struct btrfs_root *root,
2524                                struct btrfs_balance_control *bctl)
2525 {
2526         struct btrfs_trans_handle *trans;
2527         struct btrfs_balance_item *item;
2528         struct btrfs_disk_balance_args disk_bargs;
2529         struct btrfs_path *path;
2530         struct extent_buffer *leaf;
2531         struct btrfs_key key;
2532         int ret, err;
2533
2534         path = btrfs_alloc_path();
2535         if (!path)
2536                 return -ENOMEM;
2537
2538         trans = btrfs_start_transaction(root, 0);
2539         if (IS_ERR(trans)) {
2540                 btrfs_free_path(path);
2541                 return PTR_ERR(trans);
2542         }
2543
2544         key.objectid = BTRFS_BALANCE_OBJECTID;
2545         key.type = BTRFS_BALANCE_ITEM_KEY;
2546         key.offset = 0;
2547
2548         ret = btrfs_insert_empty_item(trans, root, path, &key,
2549                                       sizeof(*item));
2550         if (ret)
2551                 goto out;
2552
2553         leaf = path->nodes[0];
2554         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
2555
2556         memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
2557
2558         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
2559         btrfs_set_balance_data(leaf, item, &disk_bargs);
2560         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
2561         btrfs_set_balance_meta(leaf, item, &disk_bargs);
2562         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
2563         btrfs_set_balance_sys(leaf, item, &disk_bargs);
2564
2565         btrfs_set_balance_flags(leaf, item, bctl->flags);
2566
2567         btrfs_mark_buffer_dirty(leaf);
2568 out:
2569         btrfs_free_path(path);
2570         err = btrfs_commit_transaction(trans, root);
2571         if (err && !ret)
2572                 ret = err;
2573         return ret;
2574 }
2575
2576 static int del_balance_item(struct btrfs_root *root)
2577 {
2578         struct btrfs_trans_handle *trans;
2579         struct btrfs_path *path;
2580         struct btrfs_key key;
2581         int ret, err;
2582
2583         path = btrfs_alloc_path();
2584         if (!path)
2585                 return -ENOMEM;
2586
2587         trans = btrfs_start_transaction(root, 0);
2588         if (IS_ERR(trans)) {
2589                 btrfs_free_path(path);
2590                 return PTR_ERR(trans);
2591         }
2592
2593         key.objectid = BTRFS_BALANCE_OBJECTID;
2594         key.type = BTRFS_BALANCE_ITEM_KEY;
2595         key.offset = 0;
2596
2597         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2598         if (ret < 0)
2599                 goto out;
2600         if (ret > 0) {
2601                 ret = -ENOENT;
2602                 goto out;
2603         }
2604
2605         ret = btrfs_del_item(trans, root, path);
2606 out:
2607         btrfs_free_path(path);
2608         err = btrfs_commit_transaction(trans, root);
2609         if (err && !ret)
2610                 ret = err;
2611         return ret;
2612 }
2613
2614 /*
2615  * This is a heuristic used to reduce the number of chunks balanced on
2616  * resume after balance was interrupted.
2617  */
2618 static void update_balance_args(struct btrfs_balance_control *bctl)
2619 {
2620         /*
2621          * Turn on soft mode for chunk types that were being converted.
2622          */
2623         if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
2624                 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
2625         if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
2626                 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
2627         if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
2628                 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
2629
2630         /*
2631          * Turn on usage filter if is not already used.  The idea is
2632          * that chunks that we have already balanced should be
2633          * reasonably full.  Don't do it for chunks that are being
2634          * converted - that will keep us from relocating unconverted
2635          * (albeit full) chunks.
2636          */
2637         if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2638             !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2639                 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
2640                 bctl->data.usage = 90;
2641         }
2642         if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2643             !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2644                 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
2645                 bctl->sys.usage = 90;
2646         }
2647         if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2648             !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2649                 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
2650                 bctl->meta.usage = 90;
2651         }
2652 }
2653
2654 /*
2655  * Should be called with both balance and volume mutexes held to
2656  * serialize other volume operations (add_dev/rm_dev/resize) with
2657  * restriper.  Same goes for unset_balance_control.
2658  */
2659 static void set_balance_control(struct btrfs_balance_control *bctl)
2660 {
2661         struct btrfs_fs_info *fs_info = bctl->fs_info;
2662
2663         BUG_ON(fs_info->balance_ctl);
2664
2665         spin_lock(&fs_info->balance_lock);
2666         fs_info->balance_ctl = bctl;
2667         spin_unlock(&fs_info->balance_lock);
2668 }
2669
2670 static void unset_balance_control(struct btrfs_fs_info *fs_info)
2671 {
2672         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
2673
2674         BUG_ON(!fs_info->balance_ctl);
2675
2676         spin_lock(&fs_info->balance_lock);
2677         fs_info->balance_ctl = NULL;
2678         spin_unlock(&fs_info->balance_lock);
2679
2680         kfree(bctl);
2681 }
2682
2683 /*
2684  * Balance filters.  Return 1 if chunk should be filtered out
2685  * (should not be balanced).
2686  */
2687 static int chunk_profiles_filter(u64 chunk_type,
2688                                  struct btrfs_balance_args *bargs)
2689 {
2690         chunk_type = chunk_to_extended(chunk_type) &
2691                                 BTRFS_EXTENDED_PROFILE_MASK;
2692
2693         if (bargs->profiles & chunk_type)
2694                 return 0;
2695
2696         return 1;
2697 }
2698
2699 static int chunk_usage_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
2700                               struct btrfs_balance_args *bargs)
2701 {
2702         struct btrfs_block_group_cache *cache;
2703         u64 chunk_used, user_thresh;
2704         int ret = 1;
2705
2706         cache = btrfs_lookup_block_group(fs_info, chunk_offset);
2707         chunk_used = btrfs_block_group_used(&cache->item);
2708
2709         if (bargs->usage == 0)
2710                 user_thresh = 1;
2711         else if (bargs->usage > 100)
2712                 user_thresh = cache->key.offset;
2713         else
2714                 user_thresh = div_factor_fine(cache->key.offset,
2715                                               bargs->usage);
2716
2717         if (chunk_used < user_thresh)
2718                 ret = 0;
2719
2720         btrfs_put_block_group(cache);
2721         return ret;
2722 }
2723
2724 static int chunk_devid_filter(struct extent_buffer *leaf,
2725                               struct btrfs_chunk *chunk,
2726                               struct btrfs_balance_args *bargs)
2727 {
2728         struct btrfs_stripe *stripe;
2729         int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2730         int i;
2731
2732         for (i = 0; i < num_stripes; i++) {
2733                 stripe = btrfs_stripe_nr(chunk, i);
2734                 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
2735                         return 0;
2736         }
2737
2738         return 1;
2739 }
2740
2741 /* [pstart, pend) */
2742 static int chunk_drange_filter(struct extent_buffer *leaf,
2743                                struct btrfs_chunk *chunk,
2744                                u64 chunk_offset,
2745                                struct btrfs_balance_args *bargs)
2746 {
2747         struct btrfs_stripe *stripe;
2748         int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2749         u64 stripe_offset;
2750         u64 stripe_length;
2751         int factor;
2752         int i;
2753
2754         if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
2755                 return 0;
2756
2757         if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP |
2758              BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) {
2759                 factor = num_stripes / 2;
2760         } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) {
2761                 factor = num_stripes - 1;
2762         } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) {
2763                 factor = num_stripes - 2;
2764         } else {
2765                 factor = num_stripes;
2766         }
2767
2768         for (i = 0; i < num_stripes; i++) {
2769                 stripe = btrfs_stripe_nr(chunk, i);
2770                 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
2771                         continue;
2772
2773                 stripe_offset = btrfs_stripe_offset(leaf, stripe);
2774                 stripe_length = btrfs_chunk_length(leaf, chunk);
2775                 do_div(stripe_length, factor);
2776
2777                 if (stripe_offset < bargs->pend &&
2778                     stripe_offset + stripe_length > bargs->pstart)
2779                         return 0;
2780         }
2781
2782         return 1;
2783 }
2784
2785 /* [vstart, vend) */
2786 static int chunk_vrange_filter(struct extent_buffer *leaf,
2787                                struct btrfs_chunk *chunk,
2788                                u64 chunk_offset,
2789                                struct btrfs_balance_args *bargs)
2790 {
2791         if (chunk_offset < bargs->vend &&
2792             chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
2793                 /* at least part of the chunk is inside this vrange */
2794                 return 0;
2795
2796         return 1;
2797 }
2798
2799 static int chunk_soft_convert_filter(u64 chunk_type,
2800                                      struct btrfs_balance_args *bargs)
2801 {
2802         if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
2803                 return 0;
2804
2805         chunk_type = chunk_to_extended(chunk_type) &
2806                                 BTRFS_EXTENDED_PROFILE_MASK;
2807
2808         if (bargs->target == chunk_type)
2809                 return 1;
2810
2811         return 0;
2812 }
2813
2814 static int should_balance_chunk(struct btrfs_root *root,
2815                                 struct extent_buffer *leaf,
2816                                 struct btrfs_chunk *chunk, u64 chunk_offset)
2817 {
2818         struct btrfs_balance_control *bctl = root->fs_info->balance_ctl;
2819         struct btrfs_balance_args *bargs = NULL;
2820         u64 chunk_type = btrfs_chunk_type(leaf, chunk);
2821
2822         /* type filter */
2823         if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
2824               (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
2825                 return 0;
2826         }
2827
2828         if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
2829                 bargs = &bctl->data;
2830         else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
2831                 bargs = &bctl->sys;
2832         else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
2833                 bargs = &bctl->meta;
2834
2835         /* profiles filter */
2836         if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
2837             chunk_profiles_filter(chunk_type, bargs)) {
2838                 return 0;
2839         }
2840
2841         /* usage filter */
2842         if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
2843             chunk_usage_filter(bctl->fs_info, chunk_offset, bargs)) {
2844                 return 0;
2845         }
2846
2847         /* devid filter */
2848         if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
2849             chunk_devid_filter(leaf, chunk, bargs)) {
2850                 return 0;
2851         }
2852
2853         /* drange filter, makes sense only with devid filter */
2854         if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
2855             chunk_drange_filter(leaf, chunk, chunk_offset, bargs)) {
2856                 return 0;
2857         }
2858
2859         /* vrange filter */
2860         if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
2861             chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
2862                 return 0;
2863         }
2864
2865         /* soft profile changing mode */
2866         if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
2867             chunk_soft_convert_filter(chunk_type, bargs)) {
2868                 return 0;
2869         }
2870
2871         return 1;
2872 }
2873
2874 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
2875 {
2876         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
2877         struct btrfs_root *chunk_root = fs_info->chunk_root;
2878         struct btrfs_root *dev_root = fs_info->dev_root;
2879         struct list_head *devices;
2880         struct btrfs_device *device;
2881         u64 old_size;
2882         u64 size_to_free;
2883         struct btrfs_chunk *chunk;
2884         struct btrfs_path *path;
2885         struct btrfs_key key;
2886         struct btrfs_key found_key;
2887         struct btrfs_trans_handle *trans;
2888         struct extent_buffer *leaf;
2889         int slot;
2890         int ret;
2891         int enospc_errors = 0;
2892         bool counting = true;
2893
2894         /* step one make some room on all the devices */
2895         devices = &fs_info->fs_devices->devices;
2896         list_for_each_entry(device, devices, dev_list) {
2897                 old_size = device->total_bytes;
2898                 size_to_free = div_factor(old_size, 1);
2899                 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
2900                 if (!device->writeable ||
2901                     device->total_bytes - device->bytes_used > size_to_free ||
2902                     device->is_tgtdev_for_dev_replace)
2903                         continue;
2904
2905                 ret = btrfs_shrink_device(device, old_size - size_to_free);
2906                 if (ret == -ENOSPC)
2907                         break;
2908                 BUG_ON(ret);
2909
2910                 trans = btrfs_start_transaction(dev_root, 0);
2911                 BUG_ON(IS_ERR(trans));
2912
2913                 ret = btrfs_grow_device(trans, device, old_size);
2914                 BUG_ON(ret);
2915
2916                 btrfs_end_transaction(trans, dev_root);
2917         }
2918
2919         /* step two, relocate all the chunks */
2920         path = btrfs_alloc_path();
2921         if (!path) {
2922                 ret = -ENOMEM;
2923                 goto error;
2924         }
2925
2926         /* zero out stat counters */
2927         spin_lock(&fs_info->balance_lock);
2928         memset(&bctl->stat, 0, sizeof(bctl->stat));
2929         spin_unlock(&fs_info->balance_lock);
2930 again:
2931         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2932         key.offset = (u64)-1;
2933         key.type = BTRFS_CHUNK_ITEM_KEY;
2934
2935         while (1) {
2936                 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
2937                     atomic_read(&fs_info->balance_cancel_req)) {
2938                         ret = -ECANCELED;
2939                         goto error;
2940                 }
2941
2942                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2943                 if (ret < 0)
2944                         goto error;
2945
2946                 /*
2947                  * this shouldn't happen, it means the last relocate
2948                  * failed
2949                  */
2950                 if (ret == 0)
2951                         BUG(); /* FIXME break ? */
2952
2953                 ret = btrfs_previous_item(chunk_root, path, 0,
2954                                           BTRFS_CHUNK_ITEM_KEY);
2955                 if (ret) {
2956                         ret = 0;
2957                         break;
2958                 }
2959
2960                 leaf = path->nodes[0];
2961                 slot = path->slots[0];
2962                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2963
2964                 if (found_key.objectid != key.objectid)
2965                         break;
2966
2967                 /* chunk zero is special */
2968                 if (found_key.offset == 0)
2969                         break;
2970
2971                 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
2972
2973                 if (!counting) {
2974                         spin_lock(&fs_info->balance_lock);
2975                         bctl->stat.considered++;
2976                         spin_unlock(&fs_info->balance_lock);
2977                 }
2978
2979                 ret = should_balance_chunk(chunk_root, leaf, chunk,
2980                                            found_key.offset);
2981                 btrfs_release_path(path);
2982                 if (!ret)
2983                         goto loop;
2984
2985                 if (counting) {
2986                         spin_lock(&fs_info->balance_lock);
2987                         bctl->stat.expected++;
2988                         spin_unlock(&fs_info->balance_lock);
2989                         goto loop;
2990                 }
2991
2992                 ret = btrfs_relocate_chunk(chunk_root,
2993                                            chunk_root->root_key.objectid,
2994                                            found_key.objectid,
2995                                            found_key.offset);
2996                 if (ret && ret != -ENOSPC)
2997                         goto error;
2998                 if (ret == -ENOSPC) {
2999                         enospc_errors++;
3000                 } else {
3001                         spin_lock(&fs_info->balance_lock);
3002                         bctl->stat.completed++;
3003                         spin_unlock(&fs_info->balance_lock);
3004                 }
3005 loop:
3006                 key.offset = found_key.offset - 1;
3007         }
3008
3009         if (counting) {
3010                 btrfs_release_path(path);
3011                 counting = false;
3012                 goto again;
3013         }
3014 error:
3015         btrfs_free_path(path);
3016         if (enospc_errors) {
3017                 printk(KERN_INFO "btrfs: %d enospc errors during balance\n",
3018                        enospc_errors);
3019                 if (!ret)
3020                         ret = -ENOSPC;
3021         }
3022
3023         return ret;
3024 }
3025
3026 /**
3027  * alloc_profile_is_valid - see if a given profile is valid and reduced
3028  * @flags: profile to validate
3029  * @extended: if true @flags is treated as an extended profile
3030  */
3031 static int alloc_profile_is_valid(u64 flags, int extended)
3032 {
3033         u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
3034                                BTRFS_BLOCK_GROUP_PROFILE_MASK);
3035
3036         flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
3037
3038         /* 1) check that all other bits are zeroed */
3039         if (flags & ~mask)
3040                 return 0;
3041
3042         /* 2) see if profile is reduced */
3043         if (flags == 0)
3044                 return !extended; /* "0" is valid for usual profiles */
3045
3046         /* true if exactly one bit set */
3047         return (flags & (flags - 1)) == 0;
3048 }
3049
3050 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
3051 {
3052         /* cancel requested || normal exit path */
3053         return atomic_read(&fs_info->balance_cancel_req) ||
3054                 (atomic_read(&fs_info->balance_pause_req) == 0 &&
3055                  atomic_read(&fs_info->balance_cancel_req) == 0);
3056 }
3057
3058 static void __cancel_balance(struct btrfs_fs_info *fs_info)
3059 {
3060         int ret;
3061
3062         unset_balance_control(fs_info);
3063         ret = del_balance_item(fs_info->tree_root);
3064         if (ret)
3065                 btrfs_std_error(fs_info, ret);
3066
3067         atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3068 }
3069
3070 void update_ioctl_balance_args(struct btrfs_fs_info *fs_info, int lock,
3071                                struct btrfs_ioctl_balance_args *bargs);
3072
3073 /*
3074  * Should be called with both balance and volume mutexes held
3075  */
3076 int btrfs_balance(struct btrfs_balance_control *bctl,
3077                   struct btrfs_ioctl_balance_args *bargs)
3078 {
3079         struct btrfs_fs_info *fs_info = bctl->fs_info;
3080         u64 allowed;
3081         int mixed = 0;
3082         int ret;
3083         u64 num_devices;
3084         unsigned seq;
3085
3086         if (btrfs_fs_closing(fs_info) ||
3087             atomic_read(&fs_info->balance_pause_req) ||
3088             atomic_read(&fs_info->balance_cancel_req)) {
3089                 ret = -EINVAL;
3090                 goto out;
3091         }
3092
3093         allowed = btrfs_super_incompat_flags(fs_info->super_copy);
3094         if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
3095                 mixed = 1;
3096
3097         /*
3098          * In case of mixed groups both data and meta should be picked,
3099          * and identical options should be given for both of them.
3100          */
3101         allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
3102         if (mixed && (bctl->flags & allowed)) {
3103                 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
3104                     !(bctl->flags & BTRFS_BALANCE_METADATA) ||
3105                     memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
3106                         printk(KERN_ERR "btrfs: with mixed groups data and "
3107                                "metadata balance options must be the same\n");
3108                         ret = -EINVAL;
3109                         goto out;
3110                 }
3111         }
3112
3113         num_devices = fs_info->fs_devices->num_devices;
3114         btrfs_dev_replace_lock(&fs_info->dev_replace);
3115         if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
3116                 BUG_ON(num_devices < 1);
3117                 num_devices--;
3118         }
3119         btrfs_dev_replace_unlock(&fs_info->dev_replace);
3120         allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE;
3121         if (num_devices == 1)
3122                 allowed |= BTRFS_BLOCK_GROUP_DUP;
3123         else if (num_devices < 4)
3124                 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1);
3125         else
3126                 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
3127                                 BTRFS_BLOCK_GROUP_RAID10 |
3128                                 BTRFS_BLOCK_GROUP_RAID5 |
3129                                 BTRFS_BLOCK_GROUP_RAID6);
3130
3131         if ((bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3132             (!alloc_profile_is_valid(bctl->data.target, 1) ||
3133              (bctl->data.target & ~allowed))) {
3134                 printk(KERN_ERR "btrfs: unable to start balance with target "
3135                        "data profile %llu\n",
3136                        (unsigned long long)bctl->data.target);
3137                 ret = -EINVAL;
3138                 goto out;
3139         }
3140         if ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3141             (!alloc_profile_is_valid(bctl->meta.target, 1) ||
3142              (bctl->meta.target & ~allowed))) {
3143                 printk(KERN_ERR "btrfs: unable to start balance with target "
3144                        "metadata profile %llu\n",
3145                        (unsigned long long)bctl->meta.target);
3146                 ret = -EINVAL;
3147                 goto out;
3148         }
3149         if ((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3150             (!alloc_profile_is_valid(bctl->sys.target, 1) ||
3151              (bctl->sys.target & ~allowed))) {
3152                 printk(KERN_ERR "btrfs: unable to start balance with target "
3153                        "system profile %llu\n",
3154                        (unsigned long long)bctl->sys.target);
3155                 ret = -EINVAL;
3156                 goto out;
3157         }
3158
3159         /* allow dup'ed data chunks only in mixed mode */
3160         if (!mixed && (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3161             (bctl->data.target & BTRFS_BLOCK_GROUP_DUP)) {
3162                 printk(KERN_ERR "btrfs: dup for data is not allowed\n");
3163                 ret = -EINVAL;
3164                 goto out;
3165         }
3166
3167         /* allow to reduce meta or sys integrity only if force set */
3168         allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3169                         BTRFS_BLOCK_GROUP_RAID10 |
3170                         BTRFS_BLOCK_GROUP_RAID5 |
3171                         BTRFS_BLOCK_GROUP_RAID6;
3172         do {
3173                 seq = read_seqbegin(&fs_info->profiles_lock);
3174
3175                 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3176                      (fs_info->avail_system_alloc_bits & allowed) &&
3177                      !(bctl->sys.target & allowed)) ||
3178                     ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3179                      (fs_info->avail_metadata_alloc_bits & allowed) &&
3180                      !(bctl->meta.target & allowed))) {
3181                         if (bctl->flags & BTRFS_BALANCE_FORCE) {
3182                                 printk(KERN_INFO "btrfs: force reducing metadata "
3183                                        "integrity\n");
3184                         } else {
3185                                 printk(KERN_ERR "btrfs: balance will reduce metadata "
3186                                        "integrity, use force if you want this\n");
3187                                 ret = -EINVAL;
3188                                 goto out;
3189                         }
3190                 }
3191         } while (read_seqretry(&fs_info->profiles_lock, seq));
3192
3193         if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3194                 int num_tolerated_disk_barrier_failures;
3195                 u64 target = bctl->sys.target;
3196
3197                 num_tolerated_disk_barrier_failures =
3198                         btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3199                 if (num_tolerated_disk_barrier_failures > 0 &&
3200                     (target &
3201                      (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID0 |
3202                       BTRFS_AVAIL_ALLOC_BIT_SINGLE)))
3203                         num_tolerated_disk_barrier_failures = 0;
3204                 else if (num_tolerated_disk_barrier_failures > 1 &&
3205                          (target &
3206                           (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)))
3207                         num_tolerated_disk_barrier_failures = 1;
3208
3209                 fs_info->num_tolerated_disk_barrier_failures =
3210                         num_tolerated_disk_barrier_failures;
3211         }
3212
3213         ret = insert_balance_item(fs_info->tree_root, bctl);
3214         if (ret && ret != -EEXIST)
3215                 goto out;
3216
3217         if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
3218                 BUG_ON(ret == -EEXIST);
3219                 set_balance_control(bctl);
3220         } else {
3221                 BUG_ON(ret != -EEXIST);
3222                 spin_lock(&fs_info->balance_lock);
3223                 update_balance_args(bctl);
3224                 spin_unlock(&fs_info->balance_lock);
3225         }
3226
3227         atomic_inc(&fs_info->balance_running);
3228         mutex_unlock(&fs_info->balance_mutex);
3229
3230         ret = __btrfs_balance(fs_info);
3231
3232         mutex_lock(&fs_info->balance_mutex);
3233         atomic_dec(&fs_info->balance_running);
3234
3235         if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3236                 fs_info->num_tolerated_disk_barrier_failures =
3237                         btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3238         }
3239
3240         if (bargs) {
3241                 memset(bargs, 0, sizeof(*bargs));
3242                 update_ioctl_balance_args(fs_info, 0, bargs);
3243         }
3244
3245         if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
3246             balance_need_close(fs_info)) {
3247                 __cancel_balance(fs_info);
3248         }
3249
3250         wake_up(&fs_info->balance_wait_q);
3251
3252         return ret;
3253 out:
3254         if (bctl->flags & BTRFS_BALANCE_RESUME)
3255                 __cancel_balance(fs_info);
3256         else {
3257                 kfree(bctl);
3258                 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3259         }
3260         return ret;
3261 }
3262
3263 static int balance_kthread(void *data)
3264 {
3265         struct btrfs_fs_info *fs_info = data;
3266         int ret = 0;
3267
3268         mutex_lock(&fs_info->volume_mutex);
3269         mutex_lock(&fs_info->balance_mutex);
3270
3271         if (fs_info->balance_ctl) {
3272                 printk(KERN_INFO "btrfs: continuing balance\n");
3273                 ret = btrfs_balance(fs_info->balance_ctl, NULL);
3274         }
3275
3276         mutex_unlock(&fs_info->balance_mutex);
3277         mutex_unlock(&fs_info->volume_mutex);
3278
3279         return ret;
3280 }
3281
3282 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
3283 {
3284         struct task_struct *tsk;
3285
3286         spin_lock(&fs_info->balance_lock);
3287         if (!fs_info->balance_ctl) {
3288                 spin_unlock(&fs_info->balance_lock);
3289                 return 0;
3290         }
3291         spin_unlock(&fs_info->balance_lock);
3292
3293         if (btrfs_test_opt(fs_info->tree_root, SKIP_BALANCE)) {
3294                 printk(KERN_INFO "btrfs: force skipping balance\n");
3295                 return 0;
3296         }
3297
3298         tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
3299         if (IS_ERR(tsk))
3300                 return PTR_ERR(tsk);
3301
3302         return 0;
3303 }
3304
3305 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
3306 {
3307         struct btrfs_balance_control *bctl;
3308         struct btrfs_balance_item *item;
3309         struct btrfs_disk_balance_args disk_bargs;
3310         struct btrfs_path *path;
3311         struct extent_buffer *leaf;
3312         struct btrfs_key key;
3313         int ret;
3314
3315         path = btrfs_alloc_path();
3316         if (!path)
3317                 return -ENOMEM;
3318
3319         key.objectid = BTRFS_BALANCE_OBJECTID;
3320         key.type = BTRFS_BALANCE_ITEM_KEY;
3321         key.offset = 0;
3322
3323         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
3324         if (ret < 0)
3325                 goto out;
3326         if (ret > 0) { /* ret = -ENOENT; */
3327                 ret = 0;
3328                 goto out;
3329         }
3330
3331         bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
3332         if (!bctl) {
3333                 ret = -ENOMEM;
3334                 goto out;
3335         }
3336
3337         leaf = path->nodes[0];
3338         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3339
3340         bctl->fs_info = fs_info;
3341         bctl->flags = btrfs_balance_flags(leaf, item);
3342         bctl->flags |= BTRFS_BALANCE_RESUME;
3343
3344         btrfs_balance_data(leaf, item, &disk_bargs);
3345         btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
3346         btrfs_balance_meta(leaf, item, &disk_bargs);
3347         btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
3348         btrfs_balance_sys(leaf, item, &disk_bargs);
3349         btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
3350
3351         WARN_ON(atomic_xchg(&fs_info->mutually_exclusive_operation_running, 1));
3352
3353         mutex_lock(&fs_info->volume_mutex);
3354         mutex_lock(&fs_info->balance_mutex);
3355
3356         set_balance_control(bctl);
3357
3358         mutex_unlock(&fs_info->balance_mutex);
3359         mutex_unlock(&fs_info->volume_mutex);
3360 out:
3361         btrfs_free_path(path);
3362         return ret;
3363 }
3364
3365 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
3366 {
3367         int ret = 0;
3368
3369         mutex_lock(&fs_info->balance_mutex);
3370         if (!fs_info->balance_ctl) {
3371                 mutex_unlock(&fs_info->balance_mutex);
3372                 return -ENOTCONN;
3373         }
3374
3375         if (atomic_read(&fs_info->balance_running)) {
3376                 atomic_inc(&fs_info->balance_pause_req);
3377                 mutex_unlock(&fs_info->balance_mutex);
3378
3379                 wait_event(fs_info->balance_wait_q,
3380                            atomic_read(&fs_info->balance_running) == 0);
3381
3382                 mutex_lock(&fs_info->balance_mutex);
3383                 /* we are good with balance_ctl ripped off from under us */
3384                 BUG_ON(atomic_read(&fs_info->balance_running));
3385                 atomic_dec(&fs_info->balance_pause_req);
3386         } else {
3387                 ret = -ENOTCONN;
3388         }
3389
3390         mutex_unlock(&fs_info->balance_mutex);
3391         return ret;
3392 }
3393
3394 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
3395 {
3396         mutex_lock(&fs_info->balance_mutex);
3397         if (!fs_info->balance_ctl) {
3398                 mutex_unlock(&fs_info->balance_mutex);
3399                 return -ENOTCONN;
3400         }
3401
3402         atomic_inc(&fs_info->balance_cancel_req);
3403         /*
3404          * if we are running just wait and return, balance item is
3405          * deleted in btrfs_balance in this case
3406          */
3407         if (atomic_read(&fs_info->balance_running)) {
3408                 mutex_unlock(&fs_info->balance_mutex);
3409                 wait_event(fs_info->balance_wait_q,
3410                            atomic_read(&fs_info->balance_running) == 0);
3411                 mutex_lock(&fs_info->balance_mutex);
3412         } else {
3413                 /* __cancel_balance needs volume_mutex */
3414                 mutex_unlock(&fs_info->balance_mutex);
3415                 mutex_lock(&fs_info->volume_mutex);
3416                 mutex_lock(&fs_info->balance_mutex);
3417
3418                 if (fs_info->balance_ctl)
3419                         __cancel_balance(fs_info);
3420
3421                 mutex_unlock(&fs_info->volume_mutex);
3422         }
3423
3424         BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running));
3425         atomic_dec(&fs_info->balance_cancel_req);
3426         mutex_unlock(&fs_info->balance_mutex);
3427         return 0;
3428 }
3429
3430 /*
3431  * shrinking a device means finding all of the device extents past
3432  * the new size, and then following the back refs to the chunks.
3433  * The chunk relocation code actually frees the device extent
3434  */
3435 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
3436 {
3437         struct btrfs_trans_handle *trans;
3438         struct btrfs_root *root = device->dev_root;
3439         struct btrfs_dev_extent *dev_extent = NULL;
3440         struct btrfs_path *path;
3441         u64 length;
3442         u64 chunk_tree;
3443         u64 chunk_objectid;
3444         u64 chunk_offset;
3445         int ret;
3446         int slot;
3447         int failed = 0;
3448         bool retried = false;
3449         struct extent_buffer *l;
3450         struct btrfs_key key;
3451         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
3452         u64 old_total = btrfs_super_total_bytes(super_copy);
3453         u64 old_size = device->total_bytes;
3454         u64 diff = device->total_bytes - new_size;
3455
3456         if (device->is_tgtdev_for_dev_replace)
3457                 return -EINVAL;
3458
3459         path = btrfs_alloc_path();
3460         if (!path)
3461                 return -ENOMEM;
3462
3463         path->reada = 2;
3464
3465         lock_chunks(root);
3466
3467         device->total_bytes = new_size;
3468         if (device->writeable) {
3469                 device->fs_devices->total_rw_bytes -= diff;
3470                 spin_lock(&root->fs_info->free_chunk_lock);
3471                 root->fs_info->free_chunk_space -= diff;
3472                 spin_unlock(&root->fs_info->free_chunk_lock);
3473         }
3474         unlock_chunks(root);
3475
3476 again:
3477         key.objectid = device->devid;
3478         key.offset = (u64)-1;
3479         key.type = BTRFS_DEV_EXTENT_KEY;
3480
3481         do {
3482                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3483                 if (ret < 0)
3484                         goto done;
3485
3486                 ret = btrfs_previous_item(root, path, 0, key.type);
3487                 if (ret < 0)
3488                         goto done;
3489                 if (ret) {
3490                         ret = 0;
3491                         btrfs_release_path(path);
3492                         break;
3493                 }
3494
3495                 l = path->nodes[0];
3496                 slot = path->slots[0];
3497                 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
3498
3499                 if (key.objectid != device->devid) {
3500                         btrfs_release_path(path);
3501                         break;
3502                 }
3503
3504                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
3505                 length = btrfs_dev_extent_length(l, dev_extent);
3506
3507                 if (key.offset + length <= new_size) {
3508                         btrfs_release_path(path);
3509                         break;
3510                 }
3511
3512                 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
3513                 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
3514                 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
3515                 btrfs_release_path(path);
3516
3517                 ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
3518                                            chunk_offset);
3519                 if (ret && ret != -ENOSPC)
3520                         goto done;
3521                 if (ret == -ENOSPC)
3522                         failed++;
3523         } while (key.offset-- > 0);
3524
3525         if (failed && !retried) {
3526                 failed = 0;
3527                 retried = true;
3528                 goto again;
3529         } else if (failed && retried) {
3530                 ret = -ENOSPC;
3531                 lock_chunks(root);
3532
3533                 device->total_bytes = old_size;
3534                 if (device->writeable)
3535                         device->fs_devices->total_rw_bytes += diff;
3536                 spin_lock(&root->fs_info->free_chunk_lock);
3537                 root->fs_info->free_chunk_space += diff;
3538                 spin_unlock(&root->fs_info->free_chunk_lock);
3539                 unlock_chunks(root);
3540                 goto done;
3541         }
3542
3543         /* Shrinking succeeded, else we would be at "done". */
3544         trans = btrfs_start_transaction(root, 0);
3545         if (IS_ERR(trans)) {
3546                 ret = PTR_ERR(trans);
3547                 goto done;
3548         }
3549
3550         lock_chunks(root);
3551
3552         device->disk_total_bytes = new_size;
3553         /* Now btrfs_update_device() will change the on-disk size. */
3554         ret = btrfs_update_device(trans, device);
3555         if (ret) {
3556                 unlock_chunks(root);
3557                 btrfs_end_transaction(trans, root);
3558                 goto done;
3559         }
3560         WARN_ON(diff > old_total);
3561         btrfs_set_super_total_bytes(super_copy, old_total - diff);
3562         unlock_chunks(root);
3563         btrfs_end_transaction(trans, root);
3564 done:
3565         btrfs_free_path(path);
3566         return ret;
3567 }
3568
3569 static int btrfs_add_system_chunk(struct btrfs_root *root,
3570                            struct btrfs_key *key,
3571                            struct btrfs_chunk *chunk, int item_size)
3572 {
3573         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
3574         struct btrfs_disk_key disk_key;
3575         u32 array_size;
3576         u8 *ptr;
3577
3578         array_size = btrfs_super_sys_array_size(super_copy);
3579         if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
3580                 return -EFBIG;
3581
3582         ptr = super_copy->sys_chunk_array + array_size;
3583         btrfs_cpu_key_to_disk(&disk_key, key);
3584         memcpy(ptr, &disk_key, sizeof(disk_key));
3585         ptr += sizeof(disk_key);
3586         memcpy(ptr, chunk, item_size);
3587         item_size += sizeof(disk_key);
3588         btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
3589         return 0;
3590 }
3591
3592 /*
3593  * sort the devices in descending order by max_avail, total_avail
3594  */
3595 static int btrfs_cmp_device_info(const void *a, const void *b)
3596 {
3597         const struct btrfs_device_info *di_a = a;
3598         const struct btrfs_device_info *di_b = b;
3599
3600         if (di_a->max_avail > di_b->max_avail)
3601                 return -1;
3602         if (di_a->max_avail < di_b->max_avail)
3603                 return 1;
3604         if (di_a->total_avail > di_b->total_avail)
3605                 return -1;
3606         if (di_a->total_avail < di_b->total_avail)
3607                 return 1;
3608         return 0;
3609 }
3610
3611 static struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = {
3612         [BTRFS_RAID_RAID10] = {
3613                 .sub_stripes    = 2,
3614                 .dev_stripes    = 1,
3615                 .devs_max       = 0,    /* 0 == as many as possible */
3616                 .devs_min       = 4,
3617                 .devs_increment = 2,
3618                 .ncopies        = 2,
3619         },
3620         [BTRFS_RAID_RAID1] = {
3621                 .sub_stripes    = 1,
3622                 .dev_stripes    = 1,
3623                 .devs_max       = 2,
3624                 .devs_min       = 2,
3625                 .devs_increment = 2,
3626                 .ncopies        = 2,
3627         },
3628         [BTRFS_RAID_DUP] = {
3629                 .sub_stripes    = 1,
3630                 .dev_stripes    = 2,
3631                 .devs_max       = 1,
3632                 .devs_min       = 1,
3633                 .devs_increment = 1,
3634                 .ncopies        = 2,
3635         },
3636         [BTRFS_RAID_RAID0] = {
3637                 .sub_stripes    = 1,
3638                 .dev_stripes    = 1,
3639                 .devs_max       = 0,
3640                 .devs_min       = 2,
3641                 .devs_increment = 1,
3642                 .ncopies        = 1,
3643         },
3644         [BTRFS_RAID_SINGLE] = {
3645                 .sub_stripes    = 1,
3646                 .dev_stripes    = 1,
3647                 .devs_max       = 1,
3648                 .devs_min       = 1,
3649                 .devs_increment = 1,
3650                 .ncopies        = 1,
3651         },
3652         [BTRFS_RAID_RAID5] = {
3653                 .sub_stripes    = 1,
3654                 .dev_stripes    = 1,
3655                 .devs_max       = 0,
3656                 .devs_min       = 2,
3657                 .devs_increment = 1,
3658                 .ncopies        = 2,
3659         },
3660         [BTRFS_RAID_RAID6] = {
3661                 .sub_stripes    = 1,
3662                 .dev_stripes    = 1,
3663                 .devs_max       = 0,
3664                 .devs_min       = 3,
3665                 .devs_increment = 1,
3666                 .ncopies        = 3,
3667         },
3668 };
3669
3670 static u32 find_raid56_stripe_len(u32 data_devices, u32 dev_stripe_target)
3671 {
3672         /* TODO allow them to set a preferred stripe size */
3673         return 64 * 1024;
3674 }
3675
3676 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
3677 {
3678         if (!(type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)))
3679                 return;
3680
3681         btrfs_set_fs_incompat(info, RAID56);
3682 }
3683
3684 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
3685                                struct btrfs_root *extent_root,
3686                                struct map_lookup **map_ret,
3687                                u64 *num_bytes_out, u64 *stripe_size_out,
3688                                u64 start, u64 type)
3689 {
3690         struct btrfs_fs_info *info = extent_root->fs_info;
3691         struct btrfs_fs_devices *fs_devices = info->fs_devices;
3692         struct list_head *cur;
3693         struct map_lookup *map = NULL;
3694         struct extent_map_tree *em_tree;
3695         struct extent_map *em;
3696         struct btrfs_device_info *devices_info = NULL;
3697         u64 total_avail;
3698         int num_stripes;        /* total number of stripes to allocate */
3699         int data_stripes;       /* number of stripes that count for
3700                                    block group size */
3701         int sub_stripes;        /* sub_stripes info for map */
3702         int dev_stripes;        /* stripes per dev */
3703         int devs_max;           /* max devs to use */
3704         int devs_min;           /* min devs needed */
3705         int devs_increment;     /* ndevs has to be a multiple of this */
3706         int ncopies;            /* how many copies to data has */
3707         int ret;
3708         u64 max_stripe_size;
3709         u64 max_chunk_size;
3710         u64 stripe_size;
3711         u64 num_bytes;
3712         u64 raid_stripe_len = BTRFS_STRIPE_LEN;
3713         int ndevs;
3714         int i;
3715         int j;
3716         int index;
3717
3718         BUG_ON(!alloc_profile_is_valid(type, 0));
3719
3720         if (list_empty(&fs_devices->alloc_list))
3721                 return -ENOSPC;
3722
3723         index = __get_raid_index(type);
3724
3725         sub_stripes = btrfs_raid_array[index].sub_stripes;
3726         dev_stripes = btrfs_raid_array[index].dev_stripes;
3727         devs_max = btrfs_raid_array[index].devs_max;
3728         devs_min = btrfs_raid_array[index].devs_min;
3729         devs_increment = btrfs_raid_array[index].devs_increment;
3730         ncopies = btrfs_raid_array[index].ncopies;
3731
3732         if (type & BTRFS_BLOCK_GROUP_DATA) {
3733                 max_stripe_size = 1024 * 1024 * 1024;
3734                 max_chunk_size = 10 * max_stripe_size;
3735         } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
3736                 /* for larger filesystems, use larger metadata chunks */
3737                 if (fs_devices->total_rw_bytes > 50ULL * 1024 * 1024 * 1024)
3738                         max_stripe_size = 1024 * 1024 * 1024;
3739                 else
3740                         max_stripe_size = 256 * 1024 * 1024;
3741                 max_chunk_size = max_stripe_size;
3742         } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
3743                 max_stripe_size = 32 * 1024 * 1024;
3744                 max_chunk_size = 2 * max_stripe_size;
3745         } else {
3746                 printk(KERN_ERR "btrfs: invalid chunk type 0x%llx requested\n",
3747                        type);
3748                 BUG_ON(1);
3749         }
3750
3751         /* we don't want a chunk larger than 10% of writeable space */
3752         max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
3753                              max_chunk_size);
3754
3755         devices_info = kzalloc(sizeof(*devices_info) * fs_devices->rw_devices,
3756                                GFP_NOFS);
3757         if (!devices_info)
3758                 return -ENOMEM;
3759
3760         cur = fs_devices->alloc_list.next;
3761
3762         /*
3763          * in the first pass through the devices list, we gather information
3764          * about the available holes on each device.
3765          */
3766         ndevs = 0;
3767         while (cur != &fs_devices->alloc_list) {
3768                 struct btrfs_device *device;
3769                 u64 max_avail;
3770                 u64 dev_offset;
3771
3772                 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
3773
3774                 cur = cur->next;
3775
3776                 if (!device->writeable) {
3777                         WARN(1, KERN_ERR
3778                                "btrfs: read-only device in alloc_list\n");
3779                         continue;
3780                 }
3781
3782                 if (!device->in_fs_metadata ||
3783                     device->is_tgtdev_for_dev_replace)
3784                         continue;
3785
3786                 if (device->total_bytes > device->bytes_used)
3787                         total_avail = device->total_bytes - device->bytes_used;
3788                 else
3789                         total_avail = 0;
3790
3791                 /* If there is no space on this device, skip it. */
3792                 if (total_avail == 0)
3793                         continue;
3794
3795                 ret = find_free_dev_extent(device,
3796                                            max_stripe_size * dev_stripes,
3797                                            &dev_offset, &max_avail);
3798                 if (ret && ret != -ENOSPC)
3799                         goto error;
3800
3801                 if (ret == 0)
3802                         max_avail = max_stripe_size * dev_stripes;
3803
3804                 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes)
3805                         continue;
3806
3807                 if (ndevs == fs_devices->rw_devices) {
3808                         WARN(1, "%s: found more than %llu devices\n",
3809                              __func__, fs_devices->rw_devices);
3810                         break;
3811                 }
3812                 devices_info[ndevs].dev_offset = dev_offset;
3813                 devices_info[ndevs].max_avail = max_avail;
3814                 devices_info[ndevs].total_avail = total_avail;
3815                 devices_info[ndevs].dev = device;
3816                 ++ndevs;
3817         }
3818
3819         /*
3820          * now sort the devices by hole size / available space
3821          */
3822         sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
3823              btrfs_cmp_device_info, NULL);
3824
3825         /* round down to number of usable stripes */
3826         ndevs -= ndevs % devs_increment;
3827
3828         if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) {
3829                 ret = -ENOSPC;
3830                 goto error;
3831         }
3832
3833         if (devs_max && ndevs > devs_max)
3834                 ndevs = devs_max;
3835         /*
3836          * the primary goal is to maximize the number of stripes, so use as many
3837          * devices as possible, even if the stripes are not maximum sized.
3838          */
3839         stripe_size = devices_info[ndevs-1].max_avail;
3840         num_stripes = ndevs * dev_stripes;
3841
3842         /*
3843          * this will have to be fixed for RAID1 and RAID10 over
3844          * more drives
3845          */
3846         data_stripes = num_stripes / ncopies;
3847
3848         if (type & BTRFS_BLOCK_GROUP_RAID5) {
3849                 raid_stripe_len = find_raid56_stripe_len(ndevs - 1,
3850                                  btrfs_super_stripesize(info->super_copy));
3851                 data_stripes = num_stripes - 1;
3852         }
3853         if (type & BTRFS_BLOCK_GROUP_RAID6) {
3854                 raid_stripe_len = find_raid56_stripe_len(ndevs - 2,
3855                                  btrfs_super_stripesize(info->super_copy));
3856                 data_stripes = num_stripes - 2;
3857         }
3858
3859         /*
3860          * Use the number of data stripes to figure out how big this chunk
3861          * is really going to be in terms of logical address space,
3862          * and compare that answer with the max chunk size
3863          */
3864         if (stripe_size * data_stripes > max_chunk_size) {
3865                 u64 mask = (1ULL << 24) - 1;
3866                 stripe_size = max_chunk_size;
3867                 do_div(stripe_size, data_stripes);
3868
3869                 /* bump the answer up to a 16MB boundary */
3870                 stripe_size = (stripe_size + mask) & ~mask;
3871
3872                 /* but don't go higher than the limits we found
3873                  * while searching for free extents
3874                  */
3875                 if (stripe_size > devices_info[ndevs-1].max_avail)
3876                         stripe_size = devices_info[ndevs-1].max_avail;
3877         }
3878
3879         do_div(stripe_size, dev_stripes);
3880
3881         /* align to BTRFS_STRIPE_LEN */
3882         do_div(stripe_size, raid_stripe_len);
3883         stripe_size *= raid_stripe_len;
3884
3885         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
3886         if (!map) {
3887                 ret = -ENOMEM;
3888                 goto error;
3889         }
3890         map->num_stripes = num_stripes;
3891
3892         for (i = 0; i < ndevs; ++i) {
3893                 for (j = 0; j < dev_stripes; ++j) {
3894                         int s = i * dev_stripes + j;
3895                         map->stripes[s].dev = devices_info[i].dev;
3896                         map->stripes[s].physical = devices_info[i].dev_offset +
3897                                                    j * stripe_size;
3898                 }
3899         }
3900         map->sector_size = extent_root->sectorsize;
3901         map->stripe_len = raid_stripe_len;
3902         map->io_align = raid_stripe_len;
3903         map->io_width = raid_stripe_len;
3904         map->type = type;
3905         map->sub_stripes = sub_stripes;
3906
3907         *map_ret = map;
3908         num_bytes = stripe_size * data_stripes;
3909
3910         *stripe_size_out = stripe_size;
3911         *num_bytes_out = num_bytes;
3912
3913         trace_btrfs_chunk_alloc(info->chunk_root, map, start, num_bytes);
3914
3915         em = alloc_extent_map();
3916         if (!em) {
3917                 ret = -ENOMEM;
3918                 goto error;
3919         }
3920         em->bdev = (struct block_device *)map;
3921         em->start = start;
3922         em->len = num_bytes;
3923         em->block_start = 0;
3924         em->block_len = em->len;
3925
3926         em_tree = &extent_root->fs_info->mapping_tree.map_tree;
3927         write_lock(&em_tree->lock);
3928         ret = add_extent_mapping(em_tree, em, 0);
3929         write_unlock(&em_tree->lock);
3930         if (ret) {
3931                 free_extent_map(em);
3932                 goto error;
3933         }
3934
3935         for (i = 0; i < map->num_stripes; ++i) {
3936                 struct btrfs_device *device;
3937                 u64 dev_offset;
3938
3939                 device = map->stripes[i].dev;
3940                 dev_offset = map->stripes[i].physical;
3941
3942                 ret = btrfs_alloc_dev_extent(trans, device,
3943                                 info->chunk_root->root_key.objectid,
3944                                 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
3945                                 start, dev_offset, stripe_size);
3946                 if (ret)
3947                         goto error_dev_extent;
3948         }
3949
3950         ret = btrfs_make_block_group(trans, extent_root, 0, type,
3951                                      BTRFS_FIRST_CHUNK_TREE_OBJECTID,
3952                                      start, num_bytes);
3953         if (ret) {
3954                 i = map->num_stripes - 1;
3955                 goto error_dev_extent;
3956         }
3957
3958         free_extent_map(em);
3959         check_raid56_incompat_flag(extent_root->fs_info, type);
3960
3961         kfree(devices_info);
3962         return 0;
3963
3964 error_dev_extent:
3965         for (; i >= 0; i--) {
3966                 struct btrfs_device *device;
3967                 int err;
3968
3969                 device = map->stripes[i].dev;
3970                 err = btrfs_free_dev_extent(trans, device, start);
3971                 if (err) {
3972                         btrfs_abort_transaction(trans, extent_root, err);
3973                         break;
3974                 }
3975         }
3976         write_lock(&em_tree->lock);
3977         remove_extent_mapping(em_tree, em);
3978         write_unlock(&em_tree->lock);
3979
3980         /* One for our allocation */
3981         free_extent_map(em);
3982         /* One for the tree reference */
3983         free_extent_map(em);
3984 error:
3985         kfree(map);
3986         kfree(devices_info);
3987         return ret;
3988 }
3989
3990 static int __finish_chunk_alloc(struct btrfs_trans_handle *trans,
3991                                 struct btrfs_root *extent_root,
3992                                 struct map_lookup *map, u64 chunk_offset,
3993                                 u64 chunk_size, u64 stripe_size)
3994 {
3995         u64 dev_offset;
3996         struct btrfs_key key;
3997         struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
3998         struct btrfs_device *device;
3999         struct btrfs_chunk *chunk;
4000         struct btrfs_stripe *stripe;
4001         size_t item_size = btrfs_chunk_item_size(map->num_stripes);
4002         int index = 0;
4003         int ret;
4004
4005         chunk = kzalloc(item_size, GFP_NOFS);
4006         if (!chunk)
4007                 return -ENOMEM;
4008
4009         index = 0;
4010         while (index < map->num_stripes) {
4011                 device = map->stripes[index].dev;
4012                 device->bytes_used += stripe_size;
4013                 ret = btrfs_update_device(trans, device);
4014                 if (ret)
4015                         goto out_free;
4016                 index++;
4017         }
4018
4019         spin_lock(&extent_root->fs_info->free_chunk_lock);
4020         extent_root->fs_info->free_chunk_space -= (stripe_size *
4021                                                    map->num_stripes);
4022         spin_unlock(&extent_root->fs_info->free_chunk_lock);
4023
4024         index = 0;
4025         stripe = &chunk->stripe;
4026         while (index < map->num_stripes) {
4027                 device = map->stripes[index].dev;
4028                 dev_offset = map->stripes[index].physical;
4029
4030                 btrfs_set_stack_stripe_devid(stripe, device->devid);
4031                 btrfs_set_stack_stripe_offset(stripe, dev_offset);
4032                 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
4033                 stripe++;
4034                 index++;
4035         }
4036
4037         btrfs_set_stack_chunk_length(chunk, chunk_size);
4038         btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
4039         btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
4040         btrfs_set_stack_chunk_type(chunk, map->type);
4041         btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
4042         btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
4043         btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
4044         btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
4045         btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
4046
4047         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
4048         key.type = BTRFS_CHUNK_ITEM_KEY;
4049         key.offset = chunk_offset;
4050
4051         ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
4052
4053         if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
4054                 /*
4055                  * TODO: Cleanup of inserted chunk root in case of
4056                  * failure.
4057                  */
4058                 ret = btrfs_add_system_chunk(chunk_root, &key, chunk,
4059                                              item_size);
4060         }
4061
4062 out_free:
4063         kfree(chunk);
4064         return ret;
4065 }
4066
4067 /*
4068  * Chunk allocation falls into two parts. The first part does works
4069  * that make the new allocated chunk useable, but not do any operation
4070  * that modifies the chunk tree. The second part does the works that
4071  * require modifying the chunk tree. This division is important for the
4072  * bootstrap process of adding storage to a seed btrfs.
4073  */
4074 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4075                       struct btrfs_root *extent_root, u64 type)
4076 {
4077         u64 chunk_offset;
4078         u64 chunk_size;
4079         u64 stripe_size;
4080         struct map_lookup *map;
4081         struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
4082         int ret;
4083
4084         ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4085                               &chunk_offset);
4086         if (ret)
4087                 return ret;
4088
4089         ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
4090                                   &stripe_size, chunk_offset, type);
4091         if (ret)
4092                 return ret;
4093
4094         ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
4095                                    chunk_size, stripe_size);
4096         if (ret)
4097                 return ret;
4098         return 0;
4099 }
4100
4101 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
4102                                          struct btrfs_root *root,
4103                                          struct btrfs_device *device)
4104 {
4105         u64 chunk_offset;
4106         u64 sys_chunk_offset;
4107         u64 chunk_size;
4108         u64 sys_chunk_size;
4109         u64 stripe_size;
4110         u64 sys_stripe_size;
4111         u64 alloc_profile;
4112         struct map_lookup *map;
4113         struct map_lookup *sys_map;
4114         struct btrfs_fs_info *fs_info = root->fs_info;
4115         struct btrfs_root *extent_root = fs_info->extent_root;
4116         int ret;
4117
4118         ret = find_next_chunk(fs_info->chunk_root,
4119                               BTRFS_FIRST_CHUNK_TREE_OBJECTID, &chunk_offset);
4120         if (ret)
4121                 return ret;
4122
4123         alloc_profile = btrfs_get_alloc_profile(extent_root, 0);
4124         ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
4125                                   &stripe_size, chunk_offset, alloc_profile);
4126         if (ret)
4127                 return ret;
4128
4129         sys_chunk_offset = chunk_offset + chunk_size;
4130
4131         alloc_profile = btrfs_get_alloc_profile(fs_info->chunk_root, 0);
4132         ret = __btrfs_alloc_chunk(trans, extent_root, &sys_map,
4133                                   &sys_chunk_size, &sys_stripe_size,
4134                                   sys_chunk_offset, alloc_profile);
4135         if (ret) {
4136                 btrfs_abort_transaction(trans, root, ret);
4137                 goto out;
4138         }
4139
4140         ret = btrfs_add_device(trans, fs_info->chunk_root, device);
4141         if (ret) {
4142                 btrfs_abort_transaction(trans, root, ret);
4143                 goto out;
4144         }
4145
4146         /*
4147          * Modifying chunk tree needs allocating new blocks from both
4148          * system block group and metadata block group. So we only can
4149          * do operations require modifying the chunk tree after both
4150          * block groups were created.
4151          */
4152         ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
4153                                    chunk_size, stripe_size);
4154         if (ret) {
4155                 btrfs_abort_transaction(trans, root, ret);
4156                 goto out;
4157         }
4158
4159         ret = __finish_chunk_alloc(trans, extent_root, sys_map,
4160                                    sys_chunk_offset, sys_chunk_size,
4161                                    sys_stripe_size);
4162         if (ret)
4163                 btrfs_abort_transaction(trans, root, ret);
4164
4165 out:
4166
4167         return ret;
4168 }
4169
4170 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
4171 {
4172         struct extent_map *em;
4173         struct map_lookup *map;
4174         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
4175         int readonly = 0;
4176         int i;
4177
4178         read_lock(&map_tree->map_tree.lock);
4179         em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
4180         read_unlock(&map_tree->map_tree.lock);
4181         if (!em)
4182                 return 1;
4183
4184         if (btrfs_test_opt(root, DEGRADED)) {
4185                 free_extent_map(em);
4186                 return 0;
4187         }
4188
4189         map = (struct map_lookup *)em->bdev;
4190         for (i = 0; i < map->num_stripes; i++) {
4191                 if (!map->stripes[i].dev->writeable) {
4192                         readonly = 1;
4193                         break;
4194                 }
4195         }
4196         free_extent_map(em);
4197         return readonly;
4198 }
4199
4200 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
4201 {
4202         extent_map_tree_init(&tree->map_tree);
4203 }
4204
4205 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
4206 {
4207         struct extent_map *em;
4208
4209         while (1) {
4210                 write_lock(&tree->map_tree.lock);
4211                 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
4212                 if (em)
4213                         remove_extent_mapping(&tree->map_tree, em);
4214                 write_unlock(&tree->map_tree.lock);
4215                 if (!em)
4216                         break;
4217                 kfree(em->bdev);
4218                 /* once for us */
4219                 free_extent_map(em);
4220                 /* once for the tree */
4221                 free_extent_map(em);
4222         }
4223 }
4224
4225 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
4226 {
4227         struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
4228         struct extent_map *em;
4229         struct map_lookup *map;
4230         struct extent_map_tree *em_tree = &map_tree->map_tree;
4231         int ret;
4232
4233         read_lock(&em_tree->lock);
4234         em = lookup_extent_mapping(em_tree, logical, len);
4235         read_unlock(&em_tree->lock);
4236
4237         /*
4238          * We could return errors for these cases, but that could get ugly and
4239          * we'd probably do the same thing which is just not do anything else
4240          * and exit, so return 1 so the callers don't try to use other copies.
4241          */
4242         if (!em) {
4243                 btrfs_emerg(fs_info, "No mapping for %Lu-%Lu\n", logical,
4244                             logical+len);
4245                 return 1;
4246         }
4247
4248         if (em->start > logical || em->start + em->len < logical) {
4249                 btrfs_emerg(fs_info, "Invalid mapping for %Lu-%Lu, got "
4250                             "%Lu-%Lu\n", logical, logical+len, em->start,
4251                             em->start + em->len);
4252                 return 1;
4253         }
4254
4255         map = (struct map_lookup *)em->bdev;
4256         if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
4257                 ret = map->num_stripes;
4258         else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
4259                 ret = map->sub_stripes;
4260         else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
4261                 ret = 2;
4262         else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
4263                 ret = 3;
4264         else
4265                 ret = 1;
4266         free_extent_map(em);
4267
4268         btrfs_dev_replace_lock(&fs_info->dev_replace);
4269         if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace))
4270                 ret++;
4271         btrfs_dev_replace_unlock(&fs_info->dev_replace);
4272
4273         return ret;
4274 }
4275
4276 unsigned long btrfs_full_stripe_len(struct btrfs_root *root,
4277                                     struct btrfs_mapping_tree *map_tree,
4278                                     u64 logical)
4279 {
4280         struct extent_map *em;
4281         struct map_lookup *map;
4282         struct extent_map_tree *em_tree = &map_tree->map_tree;
4283         unsigned long len = root->sectorsize;
4284
4285         read_lock(&em_tree->lock);
4286         em = lookup_extent_mapping(em_tree, logical, len);
4287         read_unlock(&em_tree->lock);
4288         BUG_ON(!em);
4289
4290         BUG_ON(em->start > logical || em->start + em->len < logical);
4291         map = (struct map_lookup *)em->bdev;
4292         if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
4293                          BTRFS_BLOCK_GROUP_RAID6)) {
4294                 len = map->stripe_len * nr_data_stripes(map);
4295         }
4296         free_extent_map(em);
4297         return len;
4298 }
4299
4300 int btrfs_is_parity_mirror(struct btrfs_mapping_tree *map_tree,
4301                            u64 logical, u64 len, int mirror_num)
4302 {
4303         struct extent_map *em;
4304         struct map_lookup *map;
4305         struct extent_map_tree *em_tree = &map_tree->map_tree;
4306         int ret = 0;
4307
4308         read_lock(&em_tree->lock);
4309         em = lookup_extent_mapping(em_tree, logical, len);
4310         read_unlock(&em_tree->lock);
4311         BUG_ON(!em);
4312
4313         BUG_ON(em->start > logical || em->start + em->len < logical);
4314         map = (struct map_lookup *)em->bdev;
4315         if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
4316                          BTRFS_BLOCK_GROUP_RAID6))
4317                 ret = 1;
4318         free_extent_map(em);
4319         return ret;
4320 }
4321
4322 static int find_live_mirror(struct btrfs_fs_info *fs_info,
4323                             struct map_lookup *map, int first, int num,
4324                             int optimal, int dev_replace_is_ongoing)
4325 {
4326         int i;
4327         int tolerance;
4328         struct btrfs_device *srcdev;
4329
4330         if (dev_replace_is_ongoing &&
4331             fs_info->dev_replace.cont_reading_from_srcdev_mode ==
4332              BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
4333                 srcdev = fs_info->dev_replace.srcdev;
4334         else
4335                 srcdev = NULL;
4336
4337         /*
4338          * try to avoid the drive that is the source drive for a
4339          * dev-replace procedure, only choose it if no other non-missing
4340          * mirror is available
4341          */
4342         for (tolerance = 0; tolerance < 2; tolerance++) {
4343                 if (map->stripes[optimal].dev->bdev &&
4344                     (tolerance || map->stripes[optimal].dev != srcdev))
4345                         return optimal;
4346                 for (i = first; i < first + num; i++) {
4347                         if (map->stripes[i].dev->bdev &&
4348                             (tolerance || map->stripes[i].dev != srcdev))
4349                                 return i;
4350                 }
4351         }
4352
4353         /* we couldn't find one that doesn't fail.  Just return something
4354          * and the io error handling code will clean up eventually
4355          */
4356         return optimal;
4357 }
4358
4359 static inline int parity_smaller(u64 a, u64 b)
4360 {
4361         return a > b;
4362 }
4363
4364 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
4365 static void sort_parity_stripes(struct btrfs_bio *bbio, u64 *raid_map)
4366 {
4367         struct btrfs_bio_stripe s;
4368         int i;
4369         u64 l;
4370         int again = 1;
4371
4372         while (again) {
4373                 again = 0;
4374                 for (i = 0; i < bbio->num_stripes - 1; i++) {
4375                         if (parity_smaller(raid_map[i], raid_map[i+1])) {
4376                                 s = bbio->stripes[i];
4377                                 l = raid_map[i];
4378                                 bbio->stripes[i] = bbio->stripes[i+1];
4379                                 raid_map[i] = raid_map[i+1];
4380                                 bbio->stripes[i+1] = s;
4381                                 raid_map[i+1] = l;
4382                                 again = 1;
4383                         }
4384                 }
4385         }
4386 }
4387
4388 static int __btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
4389                              u64 logical, u64 *length,
4390                              struct btrfs_bio **bbio_ret,
4391                              int mirror_num, u64 **raid_map_ret)
4392 {
4393         struct extent_map *em;
4394         struct map_lookup *map;
4395         struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
4396         struct extent_map_tree *em_tree = &map_tree->map_tree;
4397         u64 offset;
4398         u64 stripe_offset;
4399         u64 stripe_end_offset;
4400         u64 stripe_nr;
4401         u64 stripe_nr_orig;
4402         u64 stripe_nr_end;
4403         u64 stripe_len;
4404         u64 *raid_map = NULL;
4405         int stripe_index;
4406         int i;
4407         int ret = 0;
4408         int num_stripes;
4409         int max_errors = 0;
4410         struct btrfs_bio *bbio = NULL;
4411         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
4412         int dev_replace_is_ongoing = 0;
4413         int num_alloc_stripes;
4414         int patch_the_first_stripe_for_dev_replace = 0;
4415         u64 physical_to_patch_in_first_stripe = 0;
4416         u64 raid56_full_stripe_start = (u64)-1;
4417
4418         read_lock(&em_tree->lock);
4419         em = lookup_extent_mapping(em_tree, logical, *length);
4420         read_unlock(&em_tree->lock);
4421
4422         if (!em) {
4423                 btrfs_crit(fs_info, "unable to find logical %llu len %llu",
4424                         (unsigned long long)logical,
4425                         (unsigned long long)*length);
4426                 return -EINVAL;
4427         }
4428
4429         if (em->start > logical || em->start + em->len < logical) {
4430                 btrfs_crit(fs_info, "found a bad mapping, wanted %Lu, "
4431                            "found %Lu-%Lu\n", logical, em->start,
4432                            em->start + em->len);
4433                 return -EINVAL;
4434         }
4435
4436         map = (struct map_lookup *)em->bdev;
4437         offset = logical - em->start;
4438
4439         if (mirror_num > map->num_stripes)
4440                 mirror_num = 0;
4441
4442         stripe_len = map->stripe_len;
4443         stripe_nr = offset;
4444         /*
4445          * stripe_nr counts the total number of stripes we have to stride
4446          * to get to this block
4447          */
4448         do_div(stripe_nr, stripe_len);
4449
4450         stripe_offset = stripe_nr * stripe_len;
4451         BUG_ON(offset < stripe_offset);
4452
4453         /* stripe_offset is the offset of this block in its stripe*/
4454         stripe_offset = offset - stripe_offset;
4455
4456         /* if we're here for raid56, we need to know the stripe aligned start */
4457         if (map->type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)) {
4458                 unsigned long full_stripe_len = stripe_len * nr_data_stripes(map);
4459                 raid56_full_stripe_start = offset;
4460
4461                 /* allow a write of a full stripe, but make sure we don't
4462                  * allow straddling of stripes
4463                  */
4464                 do_div(raid56_full_stripe_start, full_stripe_len);
4465                 raid56_full_stripe_start *= full_stripe_len;
4466         }
4467
4468         if (rw & REQ_DISCARD) {
4469                 /* we don't discard raid56 yet */
4470                 if (map->type &
4471                     (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)) {
4472                         ret = -EOPNOTSUPP;
4473                         goto out;
4474                 }
4475                 *length = min_t(u64, em->len - offset, *length);
4476         } else if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
4477                 u64 max_len;
4478                 /* For writes to RAID[56], allow a full stripeset across all disks.
4479                    For other RAID types and for RAID[56] reads, just allow a single
4480                    stripe (on a single disk). */
4481                 if (map->type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6) &&
4482                     (rw & REQ_WRITE)) {
4483                         max_len = stripe_len * nr_data_stripes(map) -
4484                                 (offset - raid56_full_stripe_start);
4485                 } else {
4486                         /* we limit the length of each bio to what fits in a stripe */
4487                         max_len = stripe_len - stripe_offset;
4488                 }
4489                 *length = min_t(u64, em->len - offset, max_len);
4490         } else {
4491                 *length = em->len - offset;
4492         }
4493
4494         /* This is for when we're called from btrfs_merge_bio_hook() and all
4495            it cares about is the length */
4496         if (!bbio_ret)
4497                 goto out;
4498
4499         btrfs_dev_replace_lock(dev_replace);
4500         dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
4501         if (!dev_replace_is_ongoing)
4502                 btrfs_dev_replace_unlock(dev_replace);
4503
4504         if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
4505             !(rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) &&
4506             dev_replace->tgtdev != NULL) {
4507                 /*
4508                  * in dev-replace case, for repair case (that's the only
4509                  * case where the mirror is selected explicitly when
4510                  * calling btrfs_map_block), blocks left of the left cursor
4511                  * can also be read from the target drive.
4512                  * For REQ_GET_READ_MIRRORS, the target drive is added as
4513                  * the last one to the array of stripes. For READ, it also
4514                  * needs to be supported using the same mirror number.
4515                  * If the requested block is not left of the left cursor,
4516                  * EIO is returned. This can happen because btrfs_num_copies()
4517                  * returns one more in the dev-replace case.
4518                  */
4519                 u64 tmp_length = *length;
4520                 struct btrfs_bio *tmp_bbio = NULL;
4521                 int tmp_num_stripes;
4522                 u64 srcdev_devid = dev_replace->srcdev->devid;
4523                 int index_srcdev = 0;
4524                 int found = 0;
4525                 u64 physical_of_found = 0;
4526
4527                 ret = __btrfs_map_block(fs_info, REQ_GET_READ_MIRRORS,
4528                              logical, &tmp_length, &tmp_bbio, 0, NULL);
4529                 if (ret) {
4530                         WARN_ON(tmp_bbio != NULL);
4531                         goto out;
4532                 }
4533
4534                 tmp_num_stripes = tmp_bbio->num_stripes;
4535                 if (mirror_num > tmp_num_stripes) {
4536                         /*
4537                          * REQ_GET_READ_MIRRORS does not contain this
4538                          * mirror, that means that the requested area
4539                          * is not left of the left cursor
4540                          */
4541                         ret = -EIO;
4542                         kfree(tmp_bbio);
4543                         goto out;
4544                 }
4545
4546                 /*
4547                  * process the rest of the function using the mirror_num
4548                  * of the source drive. Therefore look it up first.
4549                  * At the end, patch the device pointer to the one of the
4550                  * target drive.
4551                  */
4552                 for (i = 0; i < tmp_num_stripes; i++) {
4553                         if (tmp_bbio->stripes[i].dev->devid == srcdev_devid) {
4554                                 /*
4555                                  * In case of DUP, in order to keep it
4556                                  * simple, only add the mirror with the
4557                                  * lowest physical address
4558                                  */
4559                                 if (found &&
4560                                     physical_of_found <=
4561                                      tmp_bbio->stripes[i].physical)
4562                                         continue;
4563                                 index_srcdev = i;
4564                                 found = 1;
4565                                 physical_of_found =
4566                                         tmp_bbio->stripes[i].physical;
4567                         }
4568                 }
4569
4570                 if (found) {
4571                         mirror_num = index_srcdev + 1;
4572                         patch_the_first_stripe_for_dev_replace = 1;
4573                         physical_to_patch_in_first_stripe = physical_of_found;
4574                 } else {
4575                         WARN_ON(1);
4576                         ret = -EIO;
4577                         kfree(tmp_bbio);
4578                         goto out;
4579                 }
4580
4581                 kfree(tmp_bbio);
4582         } else if (mirror_num > map->num_stripes) {
4583                 mirror_num = 0;
4584         }
4585
4586         num_stripes = 1;
4587         stripe_index = 0;
4588         stripe_nr_orig = stripe_nr;
4589         stripe_nr_end = ALIGN(offset + *length, map->stripe_len);
4590         do_div(stripe_nr_end, map->stripe_len);
4591         stripe_end_offset = stripe_nr_end * map->stripe_len -
4592                             (offset + *length);
4593
4594         if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
4595                 if (rw & REQ_DISCARD)
4596                         num_stripes = min_t(u64, map->num_stripes,
4597                                             stripe_nr_end - stripe_nr_orig);
4598                 stripe_index = do_div(stripe_nr, map->num_stripes);
4599         } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
4600                 if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS))
4601                         num_stripes = map->num_stripes;
4602                 else if (mirror_num)
4603                         stripe_index = mirror_num - 1;
4604                 else {
4605                         stripe_index = find_live_mirror(fs_info, map, 0,
4606                                             map->num_stripes,
4607                                             current->pid % map->num_stripes,
4608                                             dev_replace_is_ongoing);
4609                         mirror_num = stripe_index + 1;
4610                 }
4611
4612         } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
4613                 if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) {
4614                         num_stripes = map->num_stripes;
4615                 } else if (mirror_num) {
4616                         stripe_index = mirror_num - 1;
4617                 } else {
4618                         mirror_num = 1;
4619                 }
4620
4621         } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
4622                 int factor = map->num_stripes / map->sub_stripes;
4623
4624                 stripe_index = do_div(stripe_nr, factor);
4625                 stripe_index *= map->sub_stripes;
4626
4627                 if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS))
4628                         num_stripes = map->sub_stripes;
4629                 else if (rw & REQ_DISCARD)
4630                         num_stripes = min_t(u64, map->sub_stripes *
4631                                             (stripe_nr_end - stripe_nr_orig),
4632                                             map->num_stripes);
4633                 else if (mirror_num)
4634                         stripe_index += mirror_num - 1;
4635                 else {
4636                         int old_stripe_index = stripe_index;
4637                         stripe_index = find_live_mirror(fs_info, map,
4638                                               stripe_index,
4639                                               map->sub_stripes, stripe_index +
4640                                               current->pid % map->sub_stripes,
4641                                               dev_replace_is_ongoing);
4642                         mirror_num = stripe_index - old_stripe_index + 1;
4643                 }
4644
4645         } else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
4646                                 BTRFS_BLOCK_GROUP_RAID6)) {
4647                 u64 tmp;
4648
4649                 if (bbio_ret && ((rw & REQ_WRITE) || mirror_num > 1)
4650                     && raid_map_ret) {
4651                         int i, rot;
4652
4653                         /* push stripe_nr back to the start of the full stripe */
4654                         stripe_nr = raid56_full_stripe_start;
4655                         do_div(stripe_nr, stripe_len);
4656
4657                         stripe_index = do_div(stripe_nr, nr_data_stripes(map));
4658
4659                         /* RAID[56] write or recovery. Return all stripes */
4660                         num_stripes = map->num_stripes;
4661                         max_errors = nr_parity_stripes(map);
4662
4663                         raid_map = kmalloc(sizeof(u64) * num_stripes,
4664                                            GFP_NOFS);
4665                         if (!raid_map) {
4666                                 ret = -ENOMEM;
4667                                 goto out;
4668                         }
4669
4670                         /* Work out the disk rotation on this stripe-set */
4671                         tmp = stripe_nr;
4672                         rot = do_div(tmp, num_stripes);
4673
4674                         /* Fill in the logical address of each stripe */
4675                         tmp = stripe_nr * nr_data_stripes(map);
4676                         for (i = 0; i < nr_data_stripes(map); i++)
4677                                 raid_map[(i+rot) % num_stripes] =
4678                                         em->start + (tmp + i) * map->stripe_len;
4679
4680                         raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE;
4681                         if (map->type & BTRFS_BLOCK_GROUP_RAID6)
4682                                 raid_map[(i+rot+1) % num_stripes] =
4683                                         RAID6_Q_STRIPE;
4684
4685                         *length = map->stripe_len;
4686                         stripe_index = 0;
4687                         stripe_offset = 0;
4688                 } else {
4689                         /*
4690                          * Mirror #0 or #1 means the original data block.
4691                          * Mirror #2 is RAID5 parity block.
4692                          * Mirror #3 is RAID6 Q block.
4693                          */
4694                         stripe_index = do_div(stripe_nr, nr_data_stripes(map));
4695                         if (mirror_num > 1)
4696                                 stripe_index = nr_data_stripes(map) +
4697                                                 mirror_num - 2;
4698
4699                         /* We distribute the parity blocks across stripes */
4700                         tmp = stripe_nr + stripe_index;
4701                         stripe_index = do_div(tmp, map->num_stripes);
4702                 }
4703         } else {
4704                 /*
4705                  * after this do_div call, stripe_nr is the number of stripes
4706                  * on this device we have to walk to find the data, and
4707                  * stripe_index is the number of our device in the stripe array
4708                  */
4709                 stripe_index = do_div(stripe_nr, map->num_stripes);
4710                 mirror_num = stripe_index + 1;
4711         }
4712         BUG_ON(stripe_index >= map->num_stripes);
4713
4714         num_alloc_stripes = num_stripes;
4715         if (dev_replace_is_ongoing) {
4716                 if (rw & (REQ_WRITE | REQ_DISCARD))
4717                         num_alloc_stripes <<= 1;
4718                 if (rw & REQ_GET_READ_MIRRORS)
4719                         num_alloc_stripes++;
4720         }
4721         bbio = kzalloc(btrfs_bio_size(num_alloc_stripes), GFP_NOFS);
4722         if (!bbio) {
4723                 ret = -ENOMEM;
4724                 goto out;
4725         }
4726         atomic_set(&bbio->error, 0);
4727
4728         if (rw & REQ_DISCARD) {
4729                 int factor = 0;
4730                 int sub_stripes = 0;
4731                 u64 stripes_per_dev = 0;
4732                 u32 remaining_stripes = 0;
4733                 u32 last_stripe = 0;
4734
4735                 if (map->type &
4736                     (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID10)) {
4737                         if (map->type & BTRFS_BLOCK_GROUP_RAID0)
4738                                 sub_stripes = 1;
4739                         else
4740                                 sub_stripes = map->sub_stripes;
4741
4742                         factor = map->num_stripes / sub_stripes;
4743                         stripes_per_dev = div_u64_rem(stripe_nr_end -
4744                                                       stripe_nr_orig,
4745                                                       factor,
4746                                                       &remaining_stripes);
4747                         div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
4748                         last_stripe *= sub_stripes;
4749                 }
4750
4751                 for (i = 0; i < num_stripes; i++) {
4752                         bbio->stripes[i].physical =
4753                                 map->stripes[stripe_index].physical +
4754                                 stripe_offset + stripe_nr * map->stripe_len;
4755                         bbio->stripes[i].dev = map->stripes[stripe_index].dev;
4756
4757                         if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
4758                                          BTRFS_BLOCK_GROUP_RAID10)) {
4759                                 bbio->stripes[i].length = stripes_per_dev *
4760                                                           map->stripe_len;
4761
4762                                 if (i / sub_stripes < remaining_stripes)
4763                                         bbio->stripes[i].length +=
4764                                                 map->stripe_len;
4765
4766                                 /*
4767                                  * Special for the first stripe and
4768                                  * the last stripe:
4769                                  *
4770                                  * |-------|...|-------|
4771                                  *     |----------|
4772                                  *    off     end_off
4773                                  */
4774                                 if (i < sub_stripes)
4775                                         bbio->stripes[i].length -=
4776                                                 stripe_offset;
4777
4778                                 if (stripe_index >= last_stripe &&
4779                                     stripe_index <= (last_stripe +
4780                                                      sub_stripes - 1))
4781                                         bbio->stripes[i].length -=
4782                                                 stripe_end_offset;
4783
4784                                 if (i == sub_stripes - 1)
4785                                         stripe_offset = 0;
4786                         } else
4787                                 bbio->stripes[i].length = *length;
4788
4789                         stripe_index++;
4790                         if (stripe_index == map->num_stripes) {
4791                                 /* This could only happen for RAID0/10 */
4792                                 stripe_index = 0;
4793                                 stripe_nr++;
4794                         }
4795                 }
4796         } else {
4797                 for (i = 0; i < num_stripes; i++) {
4798                         bbio->stripes[i].physical =
4799                                 map->stripes[stripe_index].physical +
4800                                 stripe_offset +
4801                                 stripe_nr * map->stripe_len;
4802                         bbio->stripes[i].dev =
4803                                 map->stripes[stripe_index].dev;
4804                         stripe_index++;
4805                 }
4806         }
4807
4808         if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS)) {
4809                 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
4810                                  BTRFS_BLOCK_GROUP_RAID10 |
4811                                  BTRFS_BLOCK_GROUP_RAID5 |
4812                                  BTRFS_BLOCK_GROUP_DUP)) {
4813                         max_errors = 1;
4814                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID6) {
4815                         max_errors = 2;
4816                 }
4817         }
4818
4819         if (dev_replace_is_ongoing && (rw & (REQ_WRITE | REQ_DISCARD)) &&
4820             dev_replace->tgtdev != NULL) {
4821                 int index_where_to_add;
4822                 u64 srcdev_devid = dev_replace->srcdev->devid;
4823
4824                 /*
4825                  * duplicate the write operations while the dev replace
4826                  * procedure is running. Since the copying of the old disk
4827                  * to the new disk takes place at run time while the
4828                  * filesystem is mounted writable, the regular write
4829                  * operations to the old disk have to be duplicated to go
4830                  * to the new disk as well.
4831                  * Note that device->missing is handled by the caller, and
4832                  * that the write to the old disk is already set up in the
4833                  * stripes array.
4834                  */
4835                 index_where_to_add = num_stripes;
4836                 for (i = 0; i < num_stripes; i++) {
4837                         if (bbio->stripes[i].dev->devid == srcdev_devid) {
4838                                 /* write to new disk, too */
4839                                 struct btrfs_bio_stripe *new =
4840                                         bbio->stripes + index_where_to_add;
4841                                 struct btrfs_bio_stripe *old =
4842                                         bbio->stripes + i;
4843
4844                                 new->physical = old->physical;
4845                                 new->length = old->length;
4846                                 new->dev = dev_replace->tgtdev;
4847                                 index_where_to_add++;
4848                                 max_errors++;
4849                         }
4850                 }
4851                 num_stripes = index_where_to_add;
4852         } else if (dev_replace_is_ongoing && (rw & REQ_GET_READ_MIRRORS) &&
4853                    dev_replace->tgtdev != NULL) {
4854                 u64 srcdev_devid = dev_replace->srcdev->devid;
4855                 int index_srcdev = 0;
4856                 int found = 0;
4857                 u64 physical_of_found = 0;
4858
4859                 /*
4860                  * During the dev-replace procedure, the target drive can
4861                  * also be used to read data in case it is needed to repair
4862                  * a corrupt block elsewhere. This is possible if the
4863                  * requested area is left of the left cursor. In this area,
4864                  * the target drive is a full copy of the source drive.
4865                  */
4866                 for (i = 0; i < num_stripes; i++) {
4867                         if (bbio->stripes[i].dev->devid == srcdev_devid) {
4868                                 /*
4869                                  * In case of DUP, in order to keep it
4870                                  * simple, only add the mirror with the
4871                                  * lowest physical address
4872                                  */
4873                                 if (found &&
4874                                     physical_of_found <=
4875                                      bbio->stripes[i].physical)
4876                                         continue;
4877                                 index_srcdev = i;
4878                                 found = 1;
4879                                 physical_of_found = bbio->stripes[i].physical;
4880                         }
4881                 }
4882                 if (found) {
4883                         u64 length = map->stripe_len;
4884
4885                         if (physical_of_found + length <=
4886                             dev_replace->cursor_left) {
4887                                 struct btrfs_bio_stripe *tgtdev_stripe =
4888                                         bbio->stripes + num_stripes;
4889
4890                                 tgtdev_stripe->physical = physical_of_found;
4891                                 tgtdev_stripe->length =
4892                                         bbio->stripes[index_srcdev].length;
4893                                 tgtdev_stripe->dev = dev_replace->tgtdev;
4894
4895                                 num_stripes++;
4896                         }
4897                 }
4898         }
4899
4900         *bbio_ret = bbio;
4901         bbio->num_stripes = num_stripes;
4902         bbio->max_errors = max_errors;
4903         bbio->mirror_num = mirror_num;
4904
4905         /*
4906          * this is the case that REQ_READ && dev_replace_is_ongoing &&
4907          * mirror_num == num_stripes + 1 && dev_replace target drive is
4908          * available as a mirror
4909          */
4910         if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) {
4911                 WARN_ON(num_stripes > 1);
4912                 bbio->stripes[0].dev = dev_replace->tgtdev;
4913                 bbio->stripes[0].physical = physical_to_patch_in_first_stripe;
4914                 bbio->mirror_num = map->num_stripes + 1;
4915         }
4916         if (raid_map) {
4917                 sort_parity_stripes(bbio, raid_map);
4918                 *raid_map_ret = raid_map;
4919         }
4920 out:
4921         if (dev_replace_is_ongoing)
4922                 btrfs_dev_replace_unlock(dev_replace);
4923         free_extent_map(em);
4924         return ret;
4925 }
4926
4927 int btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
4928                       u64 logical, u64 *length,
4929                       struct btrfs_bio **bbio_ret, int mirror_num)
4930 {
4931         return __btrfs_map_block(fs_info, rw, logical, length, bbio_ret,
4932                                  mirror_num, NULL);
4933 }
4934
4935 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
4936                      u64 chunk_start, u64 physical, u64 devid,
4937                      u64 **logical, int *naddrs, int *stripe_len)
4938 {
4939         struct extent_map_tree *em_tree = &map_tree->map_tree;
4940         struct extent_map *em;
4941         struct map_lookup *map;
4942         u64 *buf;
4943         u64 bytenr;
4944         u64 length;
4945         u64 stripe_nr;
4946         u64 rmap_len;
4947         int i, j, nr = 0;
4948
4949         read_lock(&em_tree->lock);
4950         em = lookup_extent_mapping(em_tree, chunk_start, 1);
4951         read_unlock(&em_tree->lock);
4952
4953         if (!em) {
4954                 printk(KERN_ERR "btrfs: couldn't find em for chunk %Lu\n",
4955                        chunk_start);
4956                 return -EIO;
4957         }
4958
4959         if (em->start != chunk_start) {
4960                 printk(KERN_ERR "btrfs: bad chunk start, em=%Lu, wanted=%Lu\n",
4961                        em->start, chunk_start);
4962                 free_extent_map(em);
4963                 return -EIO;
4964         }
4965         map = (struct map_lookup *)em->bdev;
4966
4967         length = em->len;
4968         rmap_len = map->stripe_len;
4969
4970         if (map->type & BTRFS_BLOCK_GROUP_RAID10)
4971                 do_div(length, map->num_stripes / map->sub_stripes);
4972         else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
4973                 do_div(length, map->num_stripes);
4974         else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
4975                               BTRFS_BLOCK_GROUP_RAID6)) {
4976                 do_div(length, nr_data_stripes(map));
4977                 rmap_len = map->stripe_len * nr_data_stripes(map);
4978         }
4979
4980         buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
4981         BUG_ON(!buf); /* -ENOMEM */
4982
4983         for (i = 0; i < map->num_stripes; i++) {
4984                 if (devid && map->stripes[i].dev->devid != devid)
4985                         continue;
4986                 if (map->stripes[i].physical > physical ||
4987                     map->stripes[i].physical + length <= physical)
4988                         continue;
4989
4990                 stripe_nr = physical - map->stripes[i].physical;
4991                 do_div(stripe_nr, map->stripe_len);
4992
4993                 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
4994                         stripe_nr = stripe_nr * map->num_stripes + i;
4995                         do_div(stripe_nr, map->sub_stripes);
4996                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
4997                         stripe_nr = stripe_nr * map->num_stripes + i;
4998                 } /* else if RAID[56], multiply by nr_data_stripes().
4999                    * Alternatively, just use rmap_len below instead of
5000                    * map->stripe_len */
5001
5002                 bytenr = chunk_start + stripe_nr * rmap_len;
5003                 WARN_ON(nr >= map->num_stripes);
5004                 for (j = 0; j < nr; j++) {
5005                         if (buf[j] == bytenr)
5006                                 break;
5007                 }
5008                 if (j == nr) {
5009                         WARN_ON(nr >= map->num_stripes);
5010                         buf[nr++] = bytenr;
5011                 }
5012         }
5013
5014         *logical = buf;
5015         *naddrs = nr;
5016         *stripe_len = rmap_len;
5017
5018         free_extent_map(em);
5019         return 0;
5020 }
5021
5022 static void *merge_stripe_index_into_bio_private(void *bi_private,
5023                                                  unsigned int stripe_index)
5024 {
5025         /*
5026          * with single, dup, RAID0, RAID1 and RAID10, stripe_index is
5027          * at most 1.
5028          * The alternative solution (instead of stealing bits from the
5029          * pointer) would be to allocate an intermediate structure
5030          * that contains the old private pointer plus the stripe_index.
5031          */
5032         BUG_ON((((uintptr_t)bi_private) & 3) != 0);
5033         BUG_ON(stripe_index > 3);
5034         return (void *)(((uintptr_t)bi_private) | stripe_index);
5035 }
5036
5037 static struct btrfs_bio *extract_bbio_from_bio_private(void *bi_private)
5038 {
5039         return (struct btrfs_bio *)(((uintptr_t)bi_private) & ~((uintptr_t)3));
5040 }
5041
5042 static unsigned int extract_stripe_index_from_bio_private(void *bi_private)
5043 {
5044         return (unsigned int)((uintptr_t)bi_private) & 3;
5045 }
5046
5047 static void btrfs_end_bio(struct bio *bio, int err)
5048 {
5049         struct btrfs_bio *bbio = extract_bbio_from_bio_private(bio->bi_private);
5050         int is_orig_bio = 0;
5051
5052         if (err) {
5053                 atomic_inc(&bbio->error);
5054                 if (err == -EIO || err == -EREMOTEIO) {
5055                         unsigned int stripe_index =
5056                                 extract_stripe_index_from_bio_private(
5057                                         bio->bi_private);
5058                         struct btrfs_device *dev;
5059
5060                         BUG_ON(stripe_index >= bbio->num_stripes);
5061                         dev = bbio->stripes[stripe_index].dev;
5062                         if (dev->bdev) {
5063                                 if (bio->bi_rw & WRITE)
5064                                         btrfs_dev_stat_inc(dev,
5065                                                 BTRFS_DEV_STAT_WRITE_ERRS);
5066                                 else
5067                                         btrfs_dev_stat_inc(dev,
5068                                                 BTRFS_DEV_STAT_READ_ERRS);
5069                                 if ((bio->bi_rw & WRITE_FLUSH) == WRITE_FLUSH)
5070                                         btrfs_dev_stat_inc(dev,
5071                                                 BTRFS_DEV_STAT_FLUSH_ERRS);
5072                                 btrfs_dev_stat_print_on_error(dev);
5073                         }
5074                 }
5075         }
5076
5077         if (bio == bbio->orig_bio)
5078                 is_orig_bio = 1;
5079
5080         if (atomic_dec_and_test(&bbio->stripes_pending)) {
5081                 if (!is_orig_bio) {
5082                         bio_put(bio);
5083                         bio = bbio->orig_bio;
5084                 }
5085                 bio->bi_private = bbio->private;
5086                 bio->bi_end_io = bbio->end_io;
5087                 bio->bi_bdev = (struct block_device *)
5088                                         (unsigned long)bbio->mirror_num;
5089                 /* only send an error to the higher layers if it is
5090                  * beyond the tolerance of the btrfs bio
5091                  */
5092                 if (atomic_read(&bbio->error) > bbio->max_errors) {
5093                         err = -EIO;
5094                 } else {
5095                         /*
5096                          * this bio is actually up to date, we didn't
5097                          * go over the max number of errors
5098                          */
5099                         set_bit(BIO_UPTODATE, &bio->bi_flags);
5100                         err = 0;
5101                 }
5102                 kfree(bbio);
5103
5104                 bio_endio(bio, err);
5105         } else if (!is_orig_bio) {
5106                 bio_put(bio);
5107         }
5108 }
5109
5110 struct async_sched {
5111         struct bio *bio;
5112         int rw;
5113         struct btrfs_fs_info *info;
5114         struct btrfs_work work;
5115 };
5116
5117 /*
5118  * see run_scheduled_bios for a description of why bios are collected for
5119  * async submit.
5120  *
5121  * This will add one bio to the pending list for a device and make sure
5122  * the work struct is scheduled.
5123  */
5124 static noinline void btrfs_schedule_bio(struct btrfs_root *root,
5125                                         struct btrfs_device *device,
5126                                         int rw, struct bio *bio)
5127 {
5128         int should_queue = 1;
5129         struct btrfs_pending_bios *pending_bios;
5130
5131         if (device->missing || !device->bdev) {
5132                 bio_endio(bio, -EIO);
5133                 return;
5134         }
5135
5136         /* don't bother with additional async steps for reads, right now */
5137         if (!(rw & REQ_WRITE)) {
5138                 bio_get(bio);
5139                 btrfsic_submit_bio(rw, bio);
5140                 bio_put(bio);
5141                 return;
5142         }
5143
5144         /*
5145          * nr_async_bios allows us to reliably return congestion to the
5146          * higher layers.  Otherwise, the async bio makes it appear we have
5147          * made progress against dirty pages when we've really just put it
5148          * on a queue for later
5149          */
5150         atomic_inc(&root->fs_info->nr_async_bios);
5151         WARN_ON(bio->bi_next);
5152         bio->bi_next = NULL;
5153         bio->bi_rw |= rw;
5154
5155         spin_lock(&device->io_lock);
5156         if (bio->bi_rw & REQ_SYNC)
5157                 pending_bios = &device->pending_sync_bios;
5158         else
5159                 pending_bios = &device->pending_bios;
5160
5161         if (pending_bios->tail)
5162                 pending_bios->tail->bi_next = bio;
5163
5164         pending_bios->tail = bio;
5165         if (!pending_bios->head)
5166                 pending_bios->head = bio;
5167         if (device->running_pending)
5168                 should_queue = 0;
5169
5170         spin_unlock(&device->io_lock);
5171
5172         if (should_queue)
5173                 btrfs_queue_worker(&root->fs_info->submit_workers,
5174                                    &device->work);
5175 }
5176
5177 static int bio_size_ok(struct block_device *bdev, struct bio *bio,
5178                        sector_t sector)
5179 {
5180         struct bio_vec *prev;
5181         struct request_queue *q = bdev_get_queue(bdev);
5182         unsigned short max_sectors = queue_max_sectors(q);
5183         struct bvec_merge_data bvm = {
5184                 .bi_bdev = bdev,
5185                 .bi_sector = sector,
5186                 .bi_rw = bio->bi_rw,
5187         };
5188
5189         if (bio->bi_vcnt == 0) {
5190                 WARN_ON(1);
5191                 return 1;
5192         }
5193
5194         prev = &bio->bi_io_vec[bio->bi_vcnt - 1];
5195         if (bio_sectors(bio) > max_sectors)
5196                 return 0;
5197
5198         if (!q->merge_bvec_fn)
5199                 return 1;
5200
5201         bvm.bi_size = bio->bi_size - prev->bv_len;
5202         if (q->merge_bvec_fn(q, &bvm, prev) < prev->bv_len)
5203                 return 0;
5204         return 1;
5205 }
5206
5207 static void submit_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio,
5208                               struct bio *bio, u64 physical, int dev_nr,
5209                               int rw, int async)
5210 {
5211         struct btrfs_device *dev = bbio->stripes[dev_nr].dev;
5212
5213         bio->bi_private = bbio;
5214         bio->bi_private = merge_stripe_index_into_bio_private(
5215                         bio->bi_private, (unsigned int)dev_nr);
5216         bio->bi_end_io = btrfs_end_bio;
5217         bio->bi_sector = physical >> 9;
5218 #ifdef DEBUG
5219         {
5220                 struct rcu_string *name;
5221
5222                 rcu_read_lock();
5223                 name = rcu_dereference(dev->name);
5224                 pr_debug("btrfs_map_bio: rw %d, sector=%llu, dev=%lu "
5225                          "(%s id %llu), size=%u\n", rw,
5226                          (u64)bio->bi_sector, (u_long)dev->bdev->bd_dev,
5227                          name->str, dev->devid, bio->bi_size);
5228                 rcu_read_unlock();
5229         }
5230 #endif
5231         bio->bi_bdev = dev->bdev;
5232         if (async)
5233                 btrfs_schedule_bio(root, dev, rw, bio);
5234         else
5235                 btrfsic_submit_bio(rw, bio);
5236 }
5237
5238 static int breakup_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio,
5239                               struct bio *first_bio, struct btrfs_device *dev,
5240                               int dev_nr, int rw, int async)
5241 {
5242         struct bio_vec *bvec = first_bio->bi_io_vec;
5243         struct bio *bio;
5244         int nr_vecs = bio_get_nr_vecs(dev->bdev);
5245         u64 physical = bbio->stripes[dev_nr].physical;
5246
5247 again:
5248         bio = btrfs_bio_alloc(dev->bdev, physical >> 9, nr_vecs, GFP_NOFS);
5249         if (!bio)
5250                 return -ENOMEM;
5251
5252         while (bvec <= (first_bio->bi_io_vec + first_bio->bi_vcnt - 1)) {
5253                 if (bio_add_page(bio, bvec->bv_page, bvec->bv_len,
5254                                  bvec->bv_offset) < bvec->bv_len) {
5255                         u64 len = bio->bi_size;
5256
5257                         atomic_inc(&bbio->stripes_pending);
5258                         submit_stripe_bio(root, bbio, bio, physical, dev_nr,
5259                                           rw, async);
5260                         physical += len;
5261                         goto again;
5262                 }
5263                 bvec++;
5264         }
5265
5266         submit_stripe_bio(root, bbio, bio, physical, dev_nr, rw, async);
5267         return 0;
5268 }
5269
5270 static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical)
5271 {
5272         atomic_inc(&bbio->error);
5273         if (atomic_dec_and_test(&bbio->stripes_pending)) {
5274                 bio->bi_private = bbio->private;
5275                 bio->bi_end_io = bbio->end_io;
5276                 bio->bi_bdev = (struct block_device *)
5277                         (unsigned long)bbio->mirror_num;
5278                 bio->bi_sector = logical >> 9;
5279                 kfree(bbio);
5280                 bio_endio(bio, -EIO);
5281         }
5282 }
5283
5284 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
5285                   int mirror_num, int async_submit)
5286 {
5287         struct btrfs_device *dev;
5288         struct bio *first_bio = bio;
5289         u64 logical = (u64)bio->bi_sector << 9;
5290         u64 length = 0;
5291         u64 map_length;
5292         u64 *raid_map = NULL;
5293         int ret;
5294         int dev_nr = 0;
5295         int total_devs = 1;
5296         struct btrfs_bio *bbio = NULL;
5297
5298         length = bio->bi_size;
5299         map_length = length;
5300
5301         ret = __btrfs_map_block(root->fs_info, rw, logical, &map_length, &bbio,
5302                               mirror_num, &raid_map);
5303         if (ret) /* -ENOMEM */
5304                 return ret;
5305
5306         total_devs = bbio->num_stripes;
5307         bbio->orig_bio = first_bio;
5308         bbio->private = first_bio->bi_private;
5309         bbio->end_io = first_bio->bi_end_io;
5310         atomic_set(&bbio->stripes_pending, bbio->num_stripes);
5311
5312         if (raid_map) {
5313                 /* In this case, map_length has been set to the length of
5314                    a single stripe; not the whole write */
5315                 if (rw & WRITE) {
5316                         return raid56_parity_write(root, bio, bbio,
5317                                                    raid_map, map_length);
5318                 } else {
5319                         return raid56_parity_recover(root, bio, bbio,
5320                                                      raid_map, map_length,
5321                                                      mirror_num);
5322                 }
5323         }
5324
5325         if (map_length < length) {
5326                 btrfs_crit(root->fs_info, "mapping failed logical %llu bio len %llu len %llu",
5327                         (unsigned long long)logical,
5328                         (unsigned long long)length,
5329                         (unsigned long long)map_length);
5330                 BUG();
5331         }
5332
5333         while (dev_nr < total_devs) {
5334                 dev = bbio->stripes[dev_nr].dev;
5335                 if (!dev || !dev->bdev || (rw & WRITE && !dev->writeable)) {
5336                         bbio_error(bbio, first_bio, logical);
5337                         dev_nr++;
5338                         continue;
5339                 }
5340
5341                 /*
5342                  * Check and see if we're ok with this bio based on it's size
5343                  * and offset with the given device.
5344                  */
5345                 if (!bio_size_ok(dev->bdev, first_bio,
5346                                  bbio->stripes[dev_nr].physical >> 9)) {
5347                         ret = breakup_stripe_bio(root, bbio, first_bio, dev,
5348                                                  dev_nr, rw, async_submit);
5349                         BUG_ON(ret);
5350                         dev_nr++;
5351                         continue;
5352                 }
5353
5354                 if (dev_nr < total_devs - 1) {
5355                         bio = bio_clone(first_bio, GFP_NOFS);
5356                         BUG_ON(!bio); /* -ENOMEM */
5357                 } else {
5358                         bio = first_bio;
5359                 }
5360
5361                 submit_stripe_bio(root, bbio, bio,
5362                                   bbio->stripes[dev_nr].physical, dev_nr, rw,
5363                                   async_submit);
5364                 dev_nr++;
5365         }
5366         return 0;
5367 }
5368
5369 struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid,
5370                                        u8 *uuid, u8 *fsid)
5371 {
5372         struct btrfs_device *device;
5373         struct btrfs_fs_devices *cur_devices;
5374
5375         cur_devices = fs_info->fs_devices;
5376         while (cur_devices) {
5377                 if (!fsid ||
5378                     !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
5379                         device = __find_device(&cur_devices->devices,
5380                                                devid, uuid);
5381                         if (device)
5382                                 return device;
5383                 }
5384                 cur_devices = cur_devices->seed;
5385         }
5386         return NULL;
5387 }
5388
5389 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
5390                                             u64 devid, u8 *dev_uuid)
5391 {
5392         struct btrfs_device *device;
5393         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
5394
5395         device = kzalloc(sizeof(*device), GFP_NOFS);
5396         if (!device)
5397                 return NULL;
5398         list_add(&device->dev_list,
5399                  &fs_devices->devices);
5400         device->dev_root = root->fs_info->dev_root;
5401         device->devid = devid;
5402         device->work.func = pending_bios_fn;
5403         device->fs_devices = fs_devices;
5404         device->missing = 1;
5405         fs_devices->num_devices++;
5406         fs_devices->missing_devices++;
5407         spin_lock_init(&device->io_lock);
5408         INIT_LIST_HEAD(&device->dev_alloc_list);
5409         memcpy(device->uuid, dev_uuid, BTRFS_UUID_SIZE);
5410         return device;
5411 }
5412
5413 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
5414                           struct extent_buffer *leaf,
5415                           struct btrfs_chunk *chunk)
5416 {
5417         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
5418         struct map_lookup *map;
5419         struct extent_map *em;
5420         u64 logical;
5421         u64 length;
5422         u64 devid;
5423         u8 uuid[BTRFS_UUID_SIZE];
5424         int num_stripes;
5425         int ret;
5426         int i;
5427
5428         logical = key->offset;
5429         length = btrfs_chunk_length(leaf, chunk);
5430
5431         read_lock(&map_tree->map_tree.lock);
5432         em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
5433         read_unlock(&map_tree->map_tree.lock);
5434
5435         /* already mapped? */
5436         if (em && em->start <= logical && em->start + em->len > logical) {
5437                 free_extent_map(em);
5438                 return 0;
5439         } else if (em) {
5440                 free_extent_map(em);
5441         }
5442
5443         em = alloc_extent_map();
5444         if (!em)
5445                 return -ENOMEM;
5446         num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
5447         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
5448         if (!map) {
5449                 free_extent_map(em);
5450                 return -ENOMEM;
5451         }
5452
5453         em->bdev = (struct block_device *)map;
5454         em->start = logical;
5455         em->len = length;
5456         em->orig_start = 0;
5457         em->block_start = 0;
5458         em->block_len = em->len;
5459
5460         map->num_stripes = num_stripes;
5461         map->io_width = btrfs_chunk_io_width(leaf, chunk);
5462         map->io_align = btrfs_chunk_io_align(leaf, chunk);
5463         map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
5464         map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
5465         map->type = btrfs_chunk_type(leaf, chunk);
5466         map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
5467         for (i = 0; i < num_stripes; i++) {
5468                 map->stripes[i].physical =
5469                         btrfs_stripe_offset_nr(leaf, chunk, i);
5470                 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
5471                 read_extent_buffer(leaf, uuid, (unsigned long)
5472                                    btrfs_stripe_dev_uuid_nr(chunk, i),
5473                                    BTRFS_UUID_SIZE);
5474                 map->stripes[i].dev = btrfs_find_device(root->fs_info, devid,
5475                                                         uuid, NULL);
5476                 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
5477                         kfree(map);
5478                         free_extent_map(em);
5479                         return -EIO;
5480                 }
5481                 if (!map->stripes[i].dev) {
5482                         map->stripes[i].dev =
5483                                 add_missing_dev(root, devid, uuid);
5484                         if (!map->stripes[i].dev) {
5485                                 kfree(map);
5486                                 free_extent_map(em);
5487                                 return -EIO;
5488                         }
5489                 }
5490                 map->stripes[i].dev->in_fs_metadata = 1;
5491         }
5492
5493         write_lock(&map_tree->map_tree.lock);
5494         ret = add_extent_mapping(&map_tree->map_tree, em, 0);
5495         write_unlock(&map_tree->map_tree.lock);
5496         BUG_ON(ret); /* Tree corruption */
5497         free_extent_map(em);
5498
5499         return 0;
5500 }
5501
5502 static void fill_device_from_item(struct extent_buffer *leaf,
5503                                  struct btrfs_dev_item *dev_item,
5504                                  struct btrfs_device *device)
5505 {
5506         unsigned long ptr;
5507
5508         device->devid = btrfs_device_id(leaf, dev_item);
5509         device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
5510         device->total_bytes = device->disk_total_bytes;
5511         device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
5512         device->type = btrfs_device_type(leaf, dev_item);
5513         device->io_align = btrfs_device_io_align(leaf, dev_item);
5514         device->io_width = btrfs_device_io_width(leaf, dev_item);
5515         device->sector_size = btrfs_device_sector_size(leaf, dev_item);
5516         WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
5517         device->is_tgtdev_for_dev_replace = 0;
5518
5519         ptr = (unsigned long)btrfs_device_uuid(dev_item);
5520         read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
5521 }
5522
5523 static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
5524 {
5525         struct btrfs_fs_devices *fs_devices;
5526         int ret;
5527
5528         BUG_ON(!mutex_is_locked(&uuid_mutex));
5529
5530         fs_devices = root->fs_info->fs_devices->seed;
5531         while (fs_devices) {
5532                 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
5533                         ret = 0;
5534                         goto out;
5535                 }
5536                 fs_devices = fs_devices->seed;
5537         }
5538
5539         fs_devices = find_fsid(fsid);
5540         if (!fs_devices) {
5541                 ret = -ENOENT;
5542                 goto out;
5543         }
5544
5545         fs_devices = clone_fs_devices(fs_devices);
5546         if (IS_ERR(fs_devices)) {
5547                 ret = PTR_ERR(fs_devices);
5548                 goto out;
5549         }
5550
5551         ret = __btrfs_open_devices(fs_devices, FMODE_READ,
5552                                    root->fs_info->bdev_holder);
5553         if (ret) {
5554                 free_fs_devices(fs_devices);
5555                 goto out;
5556         }
5557
5558         if (!fs_devices->seeding) {
5559                 __btrfs_close_devices(fs_devices);
5560                 free_fs_devices(fs_devices);
5561                 ret = -EINVAL;
5562                 goto out;
5563         }
5564
5565         fs_devices->seed = root->fs_info->fs_devices->seed;
5566         root->fs_info->fs_devices->seed = fs_devices;
5567 out:
5568         return ret;
5569 }
5570
5571 static int read_one_dev(struct btrfs_root *root,
5572                         struct extent_buffer *leaf,
5573                         struct btrfs_dev_item *dev_item)
5574 {
5575         struct btrfs_device *device;
5576         u64 devid;
5577         int ret;
5578         u8 fs_uuid[BTRFS_UUID_SIZE];
5579         u8 dev_uuid[BTRFS_UUID_SIZE];
5580
5581         devid = btrfs_device_id(leaf, dev_item);
5582         read_extent_buffer(leaf, dev_uuid,
5583                            (unsigned long)btrfs_device_uuid(dev_item),
5584                            BTRFS_UUID_SIZE);
5585         read_extent_buffer(leaf, fs_uuid,
5586                            (unsigned long)btrfs_device_fsid(dev_item),
5587                            BTRFS_UUID_SIZE);
5588
5589         if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
5590                 ret = open_seed_devices(root, fs_uuid);
5591                 if (ret && !btrfs_test_opt(root, DEGRADED))
5592                         return ret;
5593         }
5594
5595         device = btrfs_find_device(root->fs_info, devid, dev_uuid, fs_uuid);
5596         if (!device || !device->bdev) {
5597                 if (!btrfs_test_opt(root, DEGRADED))
5598                         return -EIO;
5599
5600                 if (!device) {
5601                         btrfs_warn(root->fs_info, "devid %llu missing",
5602                                 (unsigned long long)devid);
5603                         device = add_missing_dev(root, devid, dev_uuid);
5604                         if (!device)
5605                                 return -ENOMEM;
5606                 } else if (!device->missing) {
5607                         /*
5608                          * this happens when a device that was properly setup
5609                          * in the device info lists suddenly goes bad.
5610                          * device->bdev is NULL, and so we have to set
5611                          * device->missing to one here
5612                          */
5613                         root->fs_info->fs_devices->missing_devices++;
5614                         device->missing = 1;
5615                 }
5616         }
5617
5618         if (device->fs_devices != root->fs_info->fs_devices) {
5619                 BUG_ON(device->writeable);
5620                 if (device->generation !=
5621                     btrfs_device_generation(leaf, dev_item))
5622                         return -EINVAL;
5623         }
5624
5625         fill_device_from_item(leaf, dev_item, device);
5626         device->dev_root = root->fs_info->dev_root;
5627         device->in_fs_metadata = 1;
5628         if (device->writeable && !device->is_tgtdev_for_dev_replace) {
5629                 device->fs_devices->total_rw_bytes += device->total_bytes;
5630                 spin_lock(&root->fs_info->free_chunk_lock);
5631                 root->fs_info->free_chunk_space += device->total_bytes -
5632                         device->bytes_used;
5633                 spin_unlock(&root->fs_info->free_chunk_lock);
5634         }
5635         ret = 0;
5636         return ret;
5637 }
5638
5639 int btrfs_read_sys_array(struct btrfs_root *root)
5640 {
5641         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
5642         struct extent_buffer *sb;
5643         struct btrfs_disk_key *disk_key;
5644         struct btrfs_chunk *chunk;
5645         u8 *ptr;
5646         unsigned long sb_ptr;
5647         int ret = 0;
5648         u32 num_stripes;
5649         u32 array_size;
5650         u32 len = 0;
5651         u32 cur;
5652         struct btrfs_key key;
5653
5654         sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
5655                                           BTRFS_SUPER_INFO_SIZE);
5656         if (!sb)
5657                 return -ENOMEM;
5658         btrfs_set_buffer_uptodate(sb);
5659         btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
5660         /*
5661          * The sb extent buffer is artifical and just used to read the system array.
5662          * btrfs_set_buffer_uptodate() call does not properly mark all it's
5663          * pages up-to-date when the page is larger: extent does not cover the
5664          * whole page and consequently check_page_uptodate does not find all
5665          * the page's extents up-to-date (the hole beyond sb),
5666          * write_extent_buffer then triggers a WARN_ON.
5667          *
5668          * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
5669          * but sb spans only this function. Add an explicit SetPageUptodate call
5670          * to silence the warning eg. on PowerPC 64.
5671          */
5672         if (PAGE_CACHE_SIZE > BTRFS_SUPER_INFO_SIZE)
5673                 SetPageUptodate(sb->pages[0]);
5674
5675         write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
5676         array_size = btrfs_super_sys_array_size(super_copy);
5677
5678         ptr = super_copy->sys_chunk_array;
5679         sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
5680         cur = 0;
5681
5682         while (cur < array_size) {
5683                 disk_key = (struct btrfs_disk_key *)ptr;
5684                 btrfs_disk_key_to_cpu(&key, disk_key);
5685
5686                 len = sizeof(*disk_key); ptr += len;
5687                 sb_ptr += len;
5688                 cur += len;
5689
5690                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
5691                         chunk = (struct btrfs_chunk *)sb_ptr;
5692                         ret = read_one_chunk(root, &key, sb, chunk);
5693                         if (ret)
5694                                 break;
5695                         num_stripes = btrfs_chunk_num_stripes(sb, chunk);
5696                         len = btrfs_chunk_item_size(num_stripes);
5697                 } else {
5698                         ret = -EIO;
5699                         break;
5700                 }
5701                 ptr += len;
5702                 sb_ptr += len;
5703                 cur += len;
5704         }
5705         free_extent_buffer(sb);
5706         return ret;
5707 }
5708
5709 int btrfs_read_chunk_tree(struct btrfs_root *root)
5710 {
5711         struct btrfs_path *path;
5712         struct extent_buffer *leaf;
5713         struct btrfs_key key;
5714         struct btrfs_key found_key;
5715         int ret;
5716         int slot;
5717
5718         root = root->fs_info->chunk_root;
5719
5720         path = btrfs_alloc_path();
5721         if (!path)
5722                 return -ENOMEM;
5723
5724         mutex_lock(&uuid_mutex);
5725         lock_chunks(root);
5726
5727         /* first we search for all of the device items, and then we
5728          * read in all of the chunk items.  This way we can create chunk
5729          * mappings that reference all of the devices that are afound
5730          */
5731         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
5732         key.offset = 0;
5733         key.type = 0;
5734 again:
5735         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5736         if (ret < 0)
5737                 goto error;
5738         while (1) {
5739                 leaf = path->nodes[0];
5740                 slot = path->slots[0];
5741                 if (slot >= btrfs_header_nritems(leaf)) {
5742                         ret = btrfs_next_leaf(root, path);
5743                         if (ret == 0)
5744                                 continue;
5745                         if (ret < 0)
5746                                 goto error;
5747                         break;
5748                 }
5749                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5750                 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
5751                         if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID)
5752                                 break;
5753                         if (found_key.type == BTRFS_DEV_ITEM_KEY) {
5754                                 struct btrfs_dev_item *dev_item;
5755                                 dev_item = btrfs_item_ptr(leaf, slot,
5756                                                   struct btrfs_dev_item);
5757                                 ret = read_one_dev(root, leaf, dev_item);
5758                                 if (ret)
5759                                         goto error;
5760                         }
5761                 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
5762                         struct btrfs_chunk *chunk;
5763                         chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
5764                         ret = read_one_chunk(root, &found_key, leaf, chunk);
5765                         if (ret)
5766                                 goto error;
5767                 }
5768                 path->slots[0]++;
5769         }
5770         if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
5771                 key.objectid = 0;
5772                 btrfs_release_path(path);
5773                 goto again;
5774         }
5775         ret = 0;
5776 error:
5777         unlock_chunks(root);
5778         mutex_unlock(&uuid_mutex);
5779
5780         btrfs_free_path(path);
5781         return ret;
5782 }
5783
5784 static void __btrfs_reset_dev_stats(struct btrfs_device *dev)
5785 {
5786         int i;
5787
5788         for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
5789                 btrfs_dev_stat_reset(dev, i);
5790 }
5791
5792 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
5793 {
5794         struct btrfs_key key;
5795         struct btrfs_key found_key;
5796         struct btrfs_root *dev_root = fs_info->dev_root;
5797         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
5798         struct extent_buffer *eb;
5799         int slot;
5800         int ret = 0;
5801         struct btrfs_device *device;
5802         struct btrfs_path *path = NULL;
5803         int i;
5804
5805         path = btrfs_alloc_path();
5806         if (!path) {
5807                 ret = -ENOMEM;
5808                 goto out;
5809         }
5810
5811         mutex_lock(&fs_devices->device_list_mutex);
5812         list_for_each_entry(device, &fs_devices->devices, dev_list) {
5813                 int item_size;
5814                 struct btrfs_dev_stats_item *ptr;
5815
5816                 key.objectid = 0;
5817                 key.type = BTRFS_DEV_STATS_KEY;
5818                 key.offset = device->devid;
5819                 ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
5820                 if (ret) {
5821                         __btrfs_reset_dev_stats(device);
5822                         device->dev_stats_valid = 1;
5823                         btrfs_release_path(path);
5824                         continue;
5825                 }
5826                 slot = path->slots[0];
5827                 eb = path->nodes[0];
5828                 btrfs_item_key_to_cpu(eb, &found_key, slot);
5829                 item_size = btrfs_item_size_nr(eb, slot);
5830
5831                 ptr = btrfs_item_ptr(eb, slot,
5832                                      struct btrfs_dev_stats_item);
5833
5834                 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
5835                         if (item_size >= (1 + i) * sizeof(__le64))
5836                                 btrfs_dev_stat_set(device, i,
5837                                         btrfs_dev_stats_value(eb, ptr, i));
5838                         else
5839                                 btrfs_dev_stat_reset(device, i);
5840                 }
5841
5842                 device->dev_stats_valid = 1;
5843                 btrfs_dev_stat_print_on_load(device);
5844                 btrfs_release_path(path);
5845         }
5846         mutex_unlock(&fs_devices->device_list_mutex);
5847
5848 out:
5849         btrfs_free_path(path);
5850         return ret < 0 ? ret : 0;
5851 }
5852
5853 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
5854                                 struct btrfs_root *dev_root,
5855                                 struct btrfs_device *device)
5856 {
5857         struct btrfs_path *path;
5858         struct btrfs_key key;
5859         struct extent_buffer *eb;
5860         struct btrfs_dev_stats_item *ptr;
5861         int ret;
5862         int i;
5863
5864         key.objectid = 0;
5865         key.type = BTRFS_DEV_STATS_KEY;
5866         key.offset = device->devid;
5867
5868         path = btrfs_alloc_path();
5869         BUG_ON(!path);
5870         ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
5871         if (ret < 0) {
5872                 printk_in_rcu(KERN_WARNING "btrfs: error %d while searching for dev_stats item for device %s!\n",
5873                               ret, rcu_str_deref(device->name));
5874                 goto out;
5875         }
5876
5877         if (ret == 0 &&
5878             btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
5879                 /* need to delete old one and insert a new one */
5880                 ret = btrfs_del_item(trans, dev_root, path);
5881                 if (ret != 0) {
5882                         printk_in_rcu(KERN_WARNING "btrfs: delete too small dev_stats item for device %s failed %d!\n",
5883                                       rcu_str_deref(device->name), ret);
5884                         goto out;
5885                 }
5886                 ret = 1;
5887         }
5888
5889         if (ret == 1) {
5890                 /* need to insert a new item */
5891                 btrfs_release_path(path);
5892                 ret = btrfs_insert_empty_item(trans, dev_root, path,
5893                                               &key, sizeof(*ptr));
5894                 if (ret < 0) {
5895                         printk_in_rcu(KERN_WARNING "btrfs: insert dev_stats item for device %s failed %d!\n",
5896                                       rcu_str_deref(device->name), ret);
5897                         goto out;
5898                 }
5899         }
5900
5901         eb = path->nodes[0];
5902         ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
5903         for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
5904                 btrfs_set_dev_stats_value(eb, ptr, i,
5905                                           btrfs_dev_stat_read(device, i));
5906         btrfs_mark_buffer_dirty(eb);
5907
5908 out:
5909         btrfs_free_path(path);
5910         return ret;
5911 }
5912
5913 /*
5914  * called from commit_transaction. Writes all changed device stats to disk.
5915  */
5916 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans,
5917                         struct btrfs_fs_info *fs_info)
5918 {
5919         struct btrfs_root *dev_root = fs_info->dev_root;
5920         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
5921         struct btrfs_device *device;
5922         int ret = 0;
5923
5924         mutex_lock(&fs_devices->device_list_mutex);
5925         list_for_each_entry(device, &fs_devices->devices, dev_list) {
5926                 if (!device->dev_stats_valid || !device->dev_stats_dirty)
5927                         continue;
5928
5929                 ret = update_dev_stat_item(trans, dev_root, device);
5930                 if (!ret)
5931                         device->dev_stats_dirty = 0;
5932         }
5933         mutex_unlock(&fs_devices->device_list_mutex);
5934
5935         return ret;
5936 }
5937
5938 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
5939 {
5940         btrfs_dev_stat_inc(dev, index);
5941         btrfs_dev_stat_print_on_error(dev);
5942 }
5943
5944 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
5945 {
5946         if (!dev->dev_stats_valid)
5947                 return;
5948         printk_ratelimited_in_rcu(KERN_ERR
5949                            "btrfs: bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
5950                            rcu_str_deref(dev->name),
5951                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
5952                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
5953                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
5954                            btrfs_dev_stat_read(dev,
5955                                                BTRFS_DEV_STAT_CORRUPTION_ERRS),
5956                            btrfs_dev_stat_read(dev,
5957                                                BTRFS_DEV_STAT_GENERATION_ERRS));
5958 }
5959
5960 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
5961 {
5962         int i;
5963
5964         for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
5965                 if (btrfs_dev_stat_read(dev, i) != 0)
5966                         break;
5967         if (i == BTRFS_DEV_STAT_VALUES_MAX)
5968                 return; /* all values == 0, suppress message */
5969
5970         printk_in_rcu(KERN_INFO "btrfs: bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
5971                rcu_str_deref(dev->name),
5972                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
5973                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
5974                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
5975                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
5976                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
5977 }
5978
5979 int btrfs_get_dev_stats(struct btrfs_root *root,
5980                         struct btrfs_ioctl_get_dev_stats *stats)
5981 {
5982         struct btrfs_device *dev;
5983         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
5984         int i;
5985
5986         mutex_lock(&fs_devices->device_list_mutex);
5987         dev = btrfs_find_device(root->fs_info, stats->devid, NULL, NULL);
5988         mutex_unlock(&fs_devices->device_list_mutex);
5989
5990         if (!dev) {
5991                 printk(KERN_WARNING
5992                        "btrfs: get dev_stats failed, device not found\n");
5993                 return -ENODEV;
5994         } else if (!dev->dev_stats_valid) {
5995                 printk(KERN_WARNING
5996                        "btrfs: get dev_stats failed, not yet valid\n");
5997                 return -ENODEV;
5998         } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
5999                 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
6000                         if (stats->nr_items > i)
6001                                 stats->values[i] =
6002                                         btrfs_dev_stat_read_and_reset(dev, i);
6003                         else
6004                                 btrfs_dev_stat_reset(dev, i);
6005                 }
6006         } else {
6007                 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6008                         if (stats->nr_items > i)
6009                                 stats->values[i] = btrfs_dev_stat_read(dev, i);
6010         }
6011         if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
6012                 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
6013         return 0;
6014 }
6015
6016 int btrfs_scratch_superblock(struct btrfs_device *device)
6017 {
6018         struct buffer_head *bh;
6019         struct btrfs_super_block *disk_super;
6020
6021         bh = btrfs_read_dev_super(device->bdev);
6022         if (!bh)
6023                 return -EINVAL;
6024         disk_super = (struct btrfs_super_block *)bh->b_data;
6025
6026         memset(&disk_super->magic, 0, sizeof(disk_super->magic));
6027         set_buffer_dirty(bh);
6028         sync_dirty_buffer(bh);
6029         brelse(bh);
6030
6031         return 0;
6032 }