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_load(struct btrfs_device *device);
  50
  51 static DEFINE_MUTEX(uuid_mutex);
  52 static LIST_HEAD(fs_uuids);
  53
  54 static void lock_chunks(struct btrfs_root *root)
  55 {
  56         mutex_lock(&root->fs_info->chunk_mutex);
  57 }
  58
  59 static void unlock_chunks(struct btrfs_root *root)
  60 {
  61         mutex_unlock(&root->fs_info->chunk_mutex);
  62 }
  63
  64 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
  65 {
  66         struct btrfs_device *device;
  67         WARN_ON(fs_devices->opened);
  68         while (!list_empty(&fs_devices->devices)) {
  69                 device = list_entry(fs_devices->devices.next,
  70                                     struct btrfs_device, dev_list);
  71                 list_del(&device->dev_list);
  72                 rcu_string_free(device->name);
  73                 kfree(device);
  74         }
  75         kfree(fs_devices);
  76 }
  77
  78 static void btrfs_kobject_uevent(struct block_device *bdev,
  79                                  enum kobject_action action)
  80 {
  81         int ret;
  82
  83         ret = kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, action);
  84         if (ret)
  85                 pr_warn("Sending event '%d' to kobject: '%s' (%p): failed\n",
  86                         action,
  87                         kobject_name(&disk_to_dev(bdev->bd_disk)->kobj),
  88                         &disk_to_dev(bdev->bd_disk)->kobj);
  89 }
  90
  91 void btrfs_cleanup_fs_uuids(void)
  92 {
  93         struct btrfs_fs_devices *fs_devices;
  94
  95         while (!list_empty(&fs_uuids)) {
  96                 fs_devices = list_entry(fs_uuids.next,
  97                                         struct btrfs_fs_devices, list);
  98                 list_del(&fs_devices->list);
  99                 free_fs_devices(fs_devices);
 100         }
 101 }
 102
 103 static noinline struct btrfs_device *__find_device(struct list_head *head,
 104                                                    u64 devid, u8 *uuid)
 105 {
 106         struct btrfs_device *dev;
 107
 108         list_for_each_entry(dev, head, dev_list) {
 109                 if (dev->devid == devid &&
 110                     (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
 111                         return dev;
 112                 }
 113         }
 114         return NULL;
 115 }
 116
 117 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
 118 {
 119         struct btrfs_fs_devices *fs_devices;
 120
 121         list_for_each_entry(fs_devices, &fs_uuids, list) {
 122                 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
 123                         return fs_devices;
 124         }
 125         return NULL;
 126 }
 127
 128 static int
 129 btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder,
 130                       int flush, struct block_device **bdev,
 131                       struct buffer_head **bh)
 132 {
 133         int ret;
 134
 135         *bdev = blkdev_get_by_path(device_path, flags, holder);
 136
 137         if (IS_ERR(*bdev)) {
 138                 ret = PTR_ERR(*bdev);
 139                 printk(KERN_INFO "btrfs: open %s failed\n", device_path);
 140                 goto error;
 141         }
 142
 143         if (flush)
 144                 filemap_write_and_wait((*bdev)->bd_inode->i_mapping);
 145         ret = set_blocksize(*bdev, 4096);
 146         if (ret) {
 147                 blkdev_put(*bdev, flags);
 148                 goto error;
 149         }
 150         invalidate_bdev(*bdev);
 151         *bh = btrfs_read_dev_super(*bdev);
 152         if (!*bh) {
 153                 ret = -EINVAL;
 154                 blkdev_put(*bdev, flags);
 155                 goto error;
 156         }
 157
 158         return 0;
 159
 160 error:
 161         *bdev = NULL;
 162         *bh = NULL;
 163         return ret;
 164 }
 165
 166 static void requeue_list(struct btrfs_pending_bios *pending_bios,
 167                         struct bio *head, struct bio *tail)
 168 {
 169
 170         struct bio *old_head;
 171
 172         old_head = pending_bios->head;
 173         pending_bios->head = head;
 174         if (pending_bios->tail)
 175                 tail->bi_next = old_head;
 176         else
 177                 pending_bios->tail = tail;
 178 }
 179
 180 /*
 181  * we try to collect pending bios for a device so we don't get a large
 182  * number of procs sending bios down to the same device.  This greatly
 183  * improves the schedulers ability to collect and merge the bios.
 184  *
 185  * But, it also turns into a long list of bios to process and that is sure
 186  * to eventually make the worker thread block.  The solution here is to
 187  * make some progress and then put this work struct back at the end of
 188  * the list if the block device is congested.  This way, multiple devices
 189  * can make progress from a single worker thread.
 190  */
 191 static noinline void run_scheduled_bios(struct btrfs_device *device)
 192 {
 193         struct bio *pending;
 194         struct backing_dev_info *bdi;
 195         struct btrfs_fs_info *fs_info;
 196         struct btrfs_pending_bios *pending_bios;
 197         struct bio *tail;
 198         struct bio *cur;
 199         int again = 0;
 200         unsigned long num_run;
 201         unsigned long batch_run = 0;
 202         unsigned long limit;
 203         unsigned long last_waited = 0;
 204         int force_reg = 0;
 205         int sync_pending = 0;
 206         struct blk_plug plug;
 207
 208         /*
 209          * this function runs all the bios we've collected for
 210          * a particular device.  We don't want to wander off to
 211          * another device without first sending all of these down.
 212          * So, setup a plug here and finish it off before we return
 213          */
 214         blk_start_plug(&plug);
 215
 216         bdi = blk_get_backing_dev_info(device->bdev);
 217         fs_info = device->dev_root->fs_info;
 218         limit = btrfs_async_submit_limit(fs_info);
 219         limit = limit * 2 / 3;
 220
 221 loop:
 222         spin_lock(&device->io_lock);
 223
 224 loop_lock:
 225         num_run = 0;
 226
 227         /* take all the bios off the list at once and process them
 228          * later on (without the lock held).  But, remember the
 229          * tail and other pointers so the bios can be properly reinserted
 230          * into the list if we hit congestion
 231          */
 232         if (!force_reg && device->pending_sync_bios.head) {
 233                 pending_bios = &device->pending_sync_bios;
 234                 force_reg = 1;
 235         } else {
 236                 pending_bios = &device->pending_bios;
 237                 force_reg = 0;
 238         }
 239
 240         pending = pending_bios->head;
 241         tail = pending_bios->tail;
 242         WARN_ON(pending && !tail);
 243
 244         /*
 245          * if pending was null this time around, no bios need processing
 246          * at all and we can stop.  Otherwise it'll loop back up again
 247          * and do an additional check so no bios are missed.
 248          *
 249          * device->running_pending is used to synchronize with the
 250          * schedule_bio code.
 251          */
 252         if (device->pending_sync_bios.head == NULL &&
 253             device->pending_bios.head == NULL) {
 254                 again = 0;
 255                 device->running_pending = 0;
 256         } else {
 257                 again = 1;
 258                 device->running_pending = 1;
 259         }
 260
 261         pending_bios->head = NULL;
 262         pending_bios->tail = NULL;
 263
 264         spin_unlock(&device->io_lock);
 265
 266         while (pending) {
 267
 268                 rmb();
 269                 /* we want to work on both lists, but do more bios on the
 270                  * sync list than the regular list
 271                  */
 272                 if ((num_run > 32 &&
 273                     pending_bios != &device->pending_sync_bios &&
 274                     device->pending_sync_bios.head) ||
 275                    (num_run > 64 && pending_bios == &device->pending_sync_bios &&
 276                     device->pending_bios.head)) {
 277                         spin_lock(&device->io_lock);
 278                         requeue_list(pending_bios, pending, tail);
 279                         goto loop_lock;
 280                 }
 281
 282                 cur = pending;
 283                 pending = pending->bi_next;
 284                 cur->bi_next = NULL;
 285
 286                 if (atomic_dec_return(&fs_info->nr_async_bios) < limit &&
 287                     waitqueue_active(&fs_info->async_submit_wait))
 288                         wake_up(&fs_info->async_submit_wait);
 289
 290                 BUG_ON(atomic_read(&cur->bi_cnt) == 0);
 291
 292                 /*
 293                  * if we're doing the sync list, record that our
 294                  * plug has some sync requests on it
 295                  *
 296                  * If we're doing the regular list and there are
 297                  * sync requests sitting around, unplug before
 298                  * we add more
 299                  */
 300                 if (pending_bios == &device->pending_sync_bios) {
 301                         sync_pending = 1;
 302                 } else if (sync_pending) {
 303                         blk_finish_plug(&plug);
 304                         blk_start_plug(&plug);
 305                         sync_pending = 0;
 306                 }
 307
 308                 btrfsic_submit_bio(cur->bi_rw, cur);
 309                 num_run++;
 310                 batch_run++;
 311                 if (need_resched())
 312                         cond_resched();
 313
 314                 /*
 315                  * we made progress, there is more work to do and the bdi
 316                  * is now congested.  Back off and let other work structs
 317                  * run instead
 318                  */
 319                 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
 320                     fs_info->fs_devices->open_devices > 1) {
 321                         struct io_context *ioc;
 322
 323                         ioc = current->io_context;
 324
 325                         /*
 326                          * the main goal here is that we don't want to
 327                          * block if we're going to be able to submit
 328                          * more requests without blocking.
 329                          *
 330                          * This code does two great things, it pokes into
 331                          * the elevator code from a filesystem _and_
 332                          * it makes assumptions about how batching works.
 333                          */
 334                         if (ioc && ioc->nr_batch_requests > 0 &&
 335                             time_before(jiffies, ioc->last_waited + HZ/50UL) &&
 336                             (last_waited == 0 ||
 337                              ioc->last_waited == last_waited)) {
 338                                 /*
 339                                  * we want to go through our batch of
 340                                  * requests and stop.  So, we copy out
 341                                  * the ioc->last_waited time and test
 342                                  * against it before looping
 343                                  */
 344                                 last_waited = ioc->last_waited;
 345                                 if (need_resched())
 346                                         cond_resched();
 347                                 continue;
 348                         }
 349                         spin_lock(&device->io_lock);
 350                         requeue_list(pending_bios, pending, tail);
 351                         device->running_pending = 1;
 352
 353                         spin_unlock(&device->io_lock);
 354                         btrfs_requeue_work(&device->work);
 355                         goto done;
 356                 }
 357                 /* unplug every 64 requests just for good measure */
 358                 if (batch_run % 64 == 0) {
 359                         blk_finish_plug(&plug);
 360                         blk_start_plug(&plug);
 361                         sync_pending = 0;
 362                 }
 363         }
 364
 365         cond_resched();
 366         if (again)
 367                 goto loop;
 368
 369         spin_lock(&device->io_lock);
 370         if (device->pending_bios.head || device->pending_sync_bios.head)
 371                 goto loop_lock;
 372         spin_unlock(&device->io_lock);
 373
 374 done:
 375         blk_finish_plug(&plug);
 376 }
 377
 378 static void pending_bios_fn(struct btrfs_work *work)
 379 {
 380         struct btrfs_device *device;
 381
 382         device = container_of(work, struct btrfs_device, work);
 383         run_scheduled_bios(device);
 384 }
 385
 386 static noinline int device_list_add(const char *path,
 387                            struct btrfs_super_block *disk_super,
 388                            u64 devid, struct btrfs_fs_devices **fs_devices_ret)
 389 {
 390         struct btrfs_device *device;
 391         struct btrfs_fs_devices *fs_devices;
 392         struct rcu_string *name;
 393         u64 found_transid = btrfs_super_generation(disk_super);
 394
 395         fs_devices = find_fsid(disk_super->fsid);
 396         if (!fs_devices) {
 397                 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
 398                 if (!fs_devices)
 399                         return -ENOMEM;
 400                 INIT_LIST_HEAD(&fs_devices->devices);
 401                 INIT_LIST_HEAD(&fs_devices->alloc_list);
 402                 list_add(&fs_devices->list, &fs_uuids);
 403                 memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
 404                 fs_devices->latest_devid = devid;
 405                 fs_devices->latest_trans = found_transid;
 406                 mutex_init(&fs_devices->device_list_mutex);
 407                 device = NULL;
 408         } else {
 409                 device = __find_device(&fs_devices->devices, devid,
 410                                        disk_super->dev_item.uuid);
 411         }
 412         if (!device) {
 413                 if (fs_devices->opened)
 414                         return -EBUSY;
 415
 416                 device = kzalloc(sizeof(*device), GFP_NOFS);
 417                 if (!device) {
 418                         /* we can safely leave the fs_devices entry around */
 419                         return -ENOMEM;
 420                 }
 421                 device->devid = devid;
 422                 device->dev_stats_valid = 0;
 423                 device->work.func = pending_bios_fn;
 424                 memcpy(device->uuid, disk_super->dev_item.uuid,
 425                        BTRFS_UUID_SIZE);
 426                 spin_lock_init(&device->io_lock);
 427
 428                 name = rcu_string_strdup(path, GFP_NOFS);
 429                 if (!name) {
 430                         kfree(device);
 431                         return -ENOMEM;
 432                 }
 433                 rcu_assign_pointer(device->name, name);
 434                 INIT_LIST_HEAD(&device->dev_alloc_list);
 435
 436                 /* init readahead state */
 437                 spin_lock_init(&device->reada_lock);
 438                 device->reada_curr_zone = NULL;
 439                 atomic_set(&device->reada_in_flight, 0);
 440                 device->reada_next = 0;
 441                 INIT_RADIX_TREE(&device->reada_zones, GFP_NOFS & ~__GFP_WAIT);
 442                 INIT_RADIX_TREE(&device->reada_extents, GFP_NOFS & ~__GFP_WAIT);
 443
 444                 mutex_lock(&fs_devices->device_list_mutex);
 445                 list_add_rcu(&device->dev_list, &fs_devices->devices);
 446                 mutex_unlock(&fs_devices->device_list_mutex);
 447
 448                 device->fs_devices = fs_devices;
 449                 fs_devices->num_devices++;
 450         } else if (!device->name || strcmp(device->name->str, path)) {
 451                 name = rcu_string_strdup(path, GFP_NOFS);
 452                 if (!name)
 453                         return -ENOMEM;
 454                 rcu_string_free(device->name);
 455                 rcu_assign_pointer(device->name, name);
 456                 if (device->missing) {
 457                         fs_devices->missing_devices--;
 458                         device->missing = 0;
 459                 }
 460         }
 461
 462         if (found_transid > fs_devices->latest_trans) {
 463                 fs_devices->latest_devid = devid;
 464                 fs_devices->latest_trans = found_transid;
 465         }
 466         *fs_devices_ret = fs_devices;
 467         return 0;
 468 }
 469
 470 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
 471 {
 472         struct btrfs_fs_devices *fs_devices;
 473         struct btrfs_device *device;
 474         struct btrfs_device *orig_dev;
 475
 476         fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
 477         if (!fs_devices)
 478                 return ERR_PTR(-ENOMEM);
 479
 480         INIT_LIST_HEAD(&fs_devices->devices);
 481         INIT_LIST_HEAD(&fs_devices->alloc_list);
 482         INIT_LIST_HEAD(&fs_devices->list);
 483         mutex_init(&fs_devices->device_list_mutex);
 484         fs_devices->latest_devid = orig->latest_devid;
 485         fs_devices->latest_trans = orig->latest_trans;
 486         fs_devices->total_devices = orig->total_devices;
 487         memcpy(fs_devices->fsid, orig->fsid, sizeof(fs_devices->fsid));
 488
 489         /* We have held the volume lock, it is safe to get the devices. */
 490         list_for_each_entry(orig_dev, &orig->devices, dev_list) {
 491                 struct rcu_string *name;
 492
 493                 device = kzalloc(sizeof(*device), GFP_NOFS);
 494                 if (!device)
 495                         goto error;
 496
 497                 /*
 498                  * This is ok to do without rcu read locked because we hold the
 499                  * uuid mutex so nothing we touch in here is going to disappear.
 500                  */
 501                 name = rcu_string_strdup(orig_dev->name->str, GFP_NOFS);
 502                 if (!name) {
 503                         kfree(device);
 504                         goto error;
 505                 }
 506                 rcu_assign_pointer(device->name, name);
 507
 508                 device->devid = orig_dev->devid;
 509                 device->work.func = pending_bios_fn;
 510                 memcpy(device->uuid, orig_dev->uuid, sizeof(device->uuid));
 511                 spin_lock_init(&device->io_lock);
 512                 INIT_LIST_HEAD(&device->dev_list);
 513                 INIT_LIST_HEAD(&device->dev_alloc_list);
 514
 515                 list_add(&device->dev_list, &fs_devices->devices);
 516                 device->fs_devices = fs_devices;
 517                 fs_devices->num_devices++;
 518         }
 519         return fs_devices;
 520 error:
 521         free_fs_devices(fs_devices);
 522         return ERR_PTR(-ENOMEM);
 523 }
 524
 525 void btrfs_close_extra_devices(struct btrfs_fs_info *fs_info,
 526                                struct btrfs_fs_devices *fs_devices, int step)
 527 {
 528         struct btrfs_device *device, *next;
 529
 530         struct block_device *latest_bdev = NULL;
 531         u64 latest_devid = 0;
 532         u64 latest_transid = 0;
 533
 534         mutex_lock(&uuid_mutex);
 535 again:
 536         /* This is the initialized path, it is safe to release the devices. */
 537         list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
 538                 if (device->in_fs_metadata) {
 539                         if (!device->is_tgtdev_for_dev_replace &&
 540                             (!latest_transid ||
 541                              device->generation > latest_transid)) {
 542                                 latest_devid = device->devid;
 543                                 latest_transid = device->generation;
 544                                 latest_bdev = device->bdev;
 545                         }
 546                         continue;
 547                 }
 548
 549                 if (device->devid == BTRFS_DEV_REPLACE_DEVID) {
 550                         /*
 551                          * In the first step, keep the device which has
 552                          * the correct fsid and the devid that is used
 553                          * for the dev_replace procedure.
 554                          * In the second step, the dev_replace state is
 555                          * read from the device tree and it is known
 556                          * whether the procedure is really active or
 557                          * not, which means whether this device is
 558                          * used or whether it should be removed.
 559                          */
 560                         if (step == 0 || device->is_tgtdev_for_dev_replace) {
 561                                 continue;
 562                         }
 563                 }
 564                 if (device->bdev) {
 565                         blkdev_put(device->bdev, device->mode);
 566                         device->bdev = NULL;
 567                         fs_devices->open_devices--;
 568                 }
 569                 if (device->writeable) {
 570                         list_del_init(&device->dev_alloc_list);
 571                         device->writeable = 0;
 572                         if (!device->is_tgtdev_for_dev_replace)
 573                                 fs_devices->rw_devices--;
 574                 }
 575                 list_del_init(&device->dev_list);
 576                 fs_devices->num_devices--;
 577                 rcu_string_free(device->name);
 578                 kfree(device);
 579         }
 580
 581         if (fs_devices->seed) {
 582                 fs_devices = fs_devices->seed;
 583                 goto again;
 584         }
 585
 586         fs_devices->latest_bdev = latest_bdev;
 587         fs_devices->latest_devid = latest_devid;
 588         fs_devices->latest_trans = latest_transid;
 589
 590         mutex_unlock(&uuid_mutex);
 591 }
 592
 593 static void __free_device(struct work_struct *work)
 594 {
 595         struct btrfs_device *device;
 596
 597         device = container_of(work, struct btrfs_device, rcu_work);
 598
 599         if (device->bdev)
 600                 blkdev_put(device->bdev, device->mode);
 601
 602         rcu_string_free(device->name);
 603         kfree(device);
 604 }
 605
 606 static void free_device(struct rcu_head *head)
 607 {
 608         struct btrfs_device *device;
 609
 610         device = container_of(head, struct btrfs_device, rcu);
 611
 612         INIT_WORK(&device->rcu_work, __free_device);
 613         schedule_work(&device->rcu_work);
 614 }
 615
 616 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
 617 {
 618         struct btrfs_device *device;
 619
 620         if (--fs_devices->opened > 0)
 621                 return 0;
 622
 623         mutex_lock(&fs_devices->device_list_mutex);
 624         list_for_each_entry(device, &fs_devices->devices, dev_list) {
 625                 struct btrfs_device *new_device;
 626                 struct rcu_string *name;
 627
 628                 if (device->bdev)
 629                         fs_devices->open_devices--;
 630
 631                 if (device->writeable && !device->is_tgtdev_for_dev_replace) {
 632                         list_del_init(&device->dev_alloc_list);
 633                         fs_devices->rw_devices--;
 634                 }
 635
 636                 if (device->can_discard)
 637                         fs_devices->num_can_discard--;
 638
 639                 new_device = kmalloc(sizeof(*new_device), GFP_NOFS);
 640                 BUG_ON(!new_device); /* -ENOMEM */
 641                 memcpy(new_device, device, sizeof(*new_device));
 642
 643                 /* Safe because we are under uuid_mutex */
 644                 if (device->name) {
 645                         name = rcu_string_strdup(device->name->str, GFP_NOFS);
 646                         BUG_ON(device->name && !name); /* -ENOMEM */
 647                         rcu_assign_pointer(new_device->name, name);
 648                 }
 649                 new_device->bdev = NULL;
 650                 new_device->writeable = 0;
 651                 new_device->in_fs_metadata = 0;
 652                 new_device->can_discard = 0;
 653                 spin_lock_init(&new_device->io_lock);
 654                 list_replace_rcu(&device->dev_list, &new_device->dev_list);
 655
 656                 call_rcu(&device->rcu, free_device);
 657         }
 658         mutex_unlock(&fs_devices->device_list_mutex);
 659
 660         WARN_ON(fs_devices->open_devices);
 661         WARN_ON(fs_devices->rw_devices);
 662         fs_devices->opened = 0;
 663         fs_devices->seeding = 0;
 664
 665         return 0;
 666 }
 667
 668 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
 669 {
 670         struct btrfs_fs_devices *seed_devices = NULL;
 671         int ret;
 672
 673         mutex_lock(&uuid_mutex);
 674         ret = __btrfs_close_devices(fs_devices);
 675         if (!fs_devices->opened) {
 676                 seed_devices = fs_devices->seed;
 677                 fs_devices->seed = NULL;
 678         }
 679         mutex_unlock(&uuid_mutex);
 680
 681         while (seed_devices) {
 682                 fs_devices = seed_devices;
 683                 seed_devices = fs_devices->seed;
 684                 __btrfs_close_devices(fs_devices);
 685                 free_fs_devices(fs_devices);
 686         }
 687         return ret;
 688 }
 689
 690 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
 691                                 fmode_t flags, void *holder)
 692 {
 693         struct request_queue *q;
 694         struct block_device *bdev;
 695         struct list_head *head = &fs_devices->devices;
 696         struct btrfs_device *device;
 697         struct block_device *latest_bdev = NULL;
 698         struct buffer_head *bh;
 699         struct btrfs_super_block *disk_super;
 700         u64 latest_devid = 0;
 701         u64 latest_transid = 0;
 702         u64 devid;
 703         int seeding = 1;
 704         int ret = 0;
 705
 706         flags |= FMODE_EXCL;
 707
 708         list_for_each_entry(device, head, dev_list) {
 709                 if (device->bdev)
 710                         continue;
 711                 if (!device->name)
 712                         continue;
 713
 714                 ret = btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1,
 715                                             &bdev, &bh);
 716                 if (ret)
 717                         continue;
 718
 719                 disk_super = (struct btrfs_super_block *)bh->b_data;
 720                 devid = btrfs_stack_device_id(&disk_super->dev_item);
 721                 if (devid != device->devid)
 722                         goto error_brelse;
 723
 724                 if (memcmp(device->uuid, disk_super->dev_item.uuid,
 725                            BTRFS_UUID_SIZE))
 726                         goto error_brelse;
 727
 728                 device->generation = btrfs_super_generation(disk_super);
 729                 if (!latest_transid || device->generation > latest_transid) {
 730                         latest_devid = devid;
 731                         latest_transid = device->generation;
 732                         latest_bdev = bdev;
 733                 }
 734
 735                 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
 736                         device->writeable = 0;
 737                 } else {
 738                         device->writeable = !bdev_read_only(bdev);
 739                         seeding = 0;
 740                 }
 741
 742                 q = bdev_get_queue(bdev);
 743                 if (blk_queue_discard(q)) {
 744                         device->can_discard = 1;
 745                         fs_devices->num_can_discard++;
 746                 }
 747
 748                 device->bdev = bdev;
 749                 device->in_fs_metadata = 0;
 750                 device->mode = flags;
 751
 752                 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
 753                         fs_devices->rotating = 1;
 754
 755                 fs_devices->open_devices++;
 756                 if (device->writeable && !device->is_tgtdev_for_dev_replace) {
 757                         fs_devices->rw_devices++;
 758                         list_add(&device->dev_alloc_list,
 759                                  &fs_devices->alloc_list);
 760                 }
 761                 brelse(bh);
 762                 continue;
 763
 764 error_brelse:
 765                 brelse(bh);
 766                 blkdev_put(bdev, flags);
 767                 continue;
 768         }
 769         if (fs_devices->open_devices == 0) {
 770                 ret = -EINVAL;
 771                 goto out;
 772         }
 773         fs_devices->seeding = seeding;
 774         fs_devices->opened = 1;
 775         fs_devices->latest_bdev = latest_bdev;
 776         fs_devices->latest_devid = latest_devid;
 777         fs_devices->latest_trans = latest_transid;
 778         fs_devices->total_rw_bytes = 0;
 779 out:
 780         return ret;
 781 }
 782
 783 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
 784                        fmode_t flags, void *holder)
 785 {
 786         int ret;
 787
 788         mutex_lock(&uuid_mutex);
 789         if (fs_devices->opened) {
 790                 fs_devices->opened++;
 791                 ret = 0;
 792         } else {
 793                 ret = __btrfs_open_devices(fs_devices, flags, holder);
 794         }
 795         mutex_unlock(&uuid_mutex);
 796         return ret;
 797 }
 798
 799 /*
 800  * Look for a btrfs signature on a device. This may be called out of the mount path
 801  * and we are not allowed to call set_blocksize during the scan. The superblock
 802  * is read via pagecache
 803  */
 804 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
 805                           struct btrfs_fs_devices **fs_devices_ret)
 806 {
 807         struct btrfs_super_block *disk_super;
 808         struct block_device *bdev;
 809         struct page *page;
 810         void *p;
 811         int ret = -EINVAL;
 812         u64 devid;
 813         u64 transid;
 814         u64 total_devices;
 815         u64 bytenr;
 816         pgoff_t index;
 817
 818         /*
 819          * we would like to check all the supers, but that would make
 820          * a btrfs mount succeed after a mkfs from a different FS.
 821          * So, we need to add a special mount option to scan for
 822          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
 823          */
 824         bytenr = btrfs_sb_offset(0);
 825         flags |= FMODE_EXCL;
 826         mutex_lock(&uuid_mutex);
 827
 828         bdev = blkdev_get_by_path(path, flags, holder);
 829
 830         if (IS_ERR(bdev)) {
 831                 ret = PTR_ERR(bdev);
 832                 goto error;
 833         }
 834
 835         /* make sure our super fits in the device */
 836         if (bytenr + PAGE_CACHE_SIZE >= i_size_read(bdev->bd_inode))
 837                 goto error_bdev_put;
 838
 839         /* make sure our super fits in the page */
 840         if (sizeof(*disk_super) > PAGE_CACHE_SIZE)
 841                 goto error_bdev_put;
 842
 843         /* make sure our super doesn't straddle pages on disk */
 844         index = bytenr >> PAGE_CACHE_SHIFT;
 845         if ((bytenr + sizeof(*disk_super) - 1) >> PAGE_CACHE_SHIFT != index)
 846                 goto error_bdev_put;
 847
 848         /* pull in the page with our super */
 849         page = read_cache_page_gfp(bdev->bd_inode->i_mapping,
 850                                    index, GFP_NOFS);
 851
 852         if (IS_ERR_OR_NULL(page))
 853                 goto error_bdev_put;
 854
 855         p = kmap(page);
 856
 857         /* align our pointer to the offset of the super block */
 858         disk_super = p + (bytenr & ~PAGE_CACHE_MASK);
 859
 860         if (btrfs_super_bytenr(disk_super) != bytenr ||
 861             disk_super->magic != cpu_to_le64(BTRFS_MAGIC))
 862                 goto error_unmap;
 863
 864         devid = btrfs_stack_device_id(&disk_super->dev_item);
 865         transid = btrfs_super_generation(disk_super);
 866         total_devices = btrfs_super_num_devices(disk_super);
 867
 868         if (disk_super->label[0]) {
 869                 if (disk_super->label[BTRFS_LABEL_SIZE - 1])
 870                         disk_super->label[BTRFS_LABEL_SIZE - 1] = '\0';
 871                 printk(KERN_INFO "device label %s ", disk_super->label);
 872         } else {
 873                 printk(KERN_INFO "device fsid %pU ", disk_super->fsid);
 874         }
 875
 876         printk(KERN_CONT "devid %llu transid %llu %s\n",
 877                (unsigned long long)devid, (unsigned long long)transid, path);
 878
 879         ret = device_list_add(path, disk_super, devid, fs_devices_ret);
 880         if (!ret && fs_devices_ret)
 881                 (*fs_devices_ret)->total_devices = total_devices;
 882
 883 error_unmap:
 884         kunmap(page);
 885         page_cache_release(page);
 886
 887 error_bdev_put:
 888         blkdev_put(bdev, flags);
 889 error:
 890         mutex_unlock(&uuid_mutex);
 891         return ret;
 892 }
 893
 894 /* helper to account the used device space in the range */
 895 int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
 896                                    u64 end, u64 *length)
 897 {
 898         struct btrfs_key key;
 899         struct btrfs_root *root = device->dev_root;
 900         struct btrfs_dev_extent *dev_extent;
 901         struct btrfs_path *path;
 902         u64 extent_end;
 903         int ret;
 904         int slot;
 905         struct extent_buffer *l;
 906
 907         *length = 0;
 908
 909         if (start >= device->total_bytes || device->is_tgtdev_for_dev_replace)
 910                 return 0;
 911
 912         path = btrfs_alloc_path();
 913         if (!path)
 914                 return -ENOMEM;
 915         path->reada = 2;
 916
 917         key.objectid = device->devid;
 918         key.offset = start;
 919         key.type = BTRFS_DEV_EXTENT_KEY;
 920
 921         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 922         if (ret < 0)
 923                 goto out;
 924         if (ret > 0) {
 925                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
 926                 if (ret < 0)
 927                         goto out;
 928         }
 929
 930         while (1) {
 931                 l = path->nodes[0];
 932                 slot = path->slots[0];
 933                 if (slot >= btrfs_header_nritems(l)) {
 934                         ret = btrfs_next_leaf(root, path);
 935                         if (ret == 0)
 936                                 continue;
 937                         if (ret < 0)
 938                                 goto out;
 939
 940                         break;
 941                 }
 942                 btrfs_item_key_to_cpu(l, &key, slot);
 943
 944                 if (key.objectid < device->devid)
 945                         goto next;
 946
 947                 if (key.objectid > device->devid)
 948                         break;
 949
 950                 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
 951                         goto next;
 952
 953                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
 954                 extent_end = key.offset + btrfs_dev_extent_length(l,
 955                                                                   dev_extent);
 956                 if (key.offset <= start && extent_end > end) {
 957                         *length = end - start + 1;
 958                         break;
 959                 } else if (key.offset <= start && extent_end > start)
 960                         *length += extent_end - start;
 961                 else if (key.offset > start && extent_end <= end)
 962                         *length += extent_end - key.offset;
 963                 else if (key.offset > start && key.offset <= end) {
 964                         *length += end - key.offset + 1;
 965                         break;
 966                 } else if (key.offset > end)
 967                         break;
 968
 969 next:
 970                 path->slots[0]++;
 971         }
 972         ret = 0;
 973 out:
 974         btrfs_free_path(path);
 975         return ret;
 976 }
 977
 978 /*
 979  * find_free_dev_extent - find free space in the specified device
 980  * @device:     the device which we search the free space in
 981  * @num_bytes:  the size of the free space that we need
 982  * @start:      store the start of the free space.
 983  * @len:        the size of the free space. that we find, or the size of the max
 984  *              free space if we don't find suitable free space
 985  *
 986  * this uses a pretty simple search, the expectation is that it is
 987  * called very infrequently and that a given device has a small number
 988  * of extents
 989  *
 990  * @start is used to store the start of the free space if we find. But if we
 991  * don't find suitable free space, it will be used to store the start position
 992  * of the max free space.
 993  *
 994  * @len is used to store the size of the free space that we find.
 995  * But if we don't find suitable free space, it is used to store the size of
 996  * the max free space.
 997  */
 998 int find_free_dev_extent(struct btrfs_device *device, u64 num_bytes,
 999                          u64 *start, u64 *len)
1000 {
1001         struct btrfs_key key;
1002         struct btrfs_root *root = device->dev_root;
1003         struct btrfs_dev_extent *dev_extent;
1004         struct btrfs_path *path;
1005         u64 hole_size;
1006         u64 max_hole_start;
1007         u64 max_hole_size;
1008         u64 extent_end;
1009         u64 search_start;
1010         u64 search_end = device->total_bytes;
1011         int ret;
1012         int slot;
1013         struct extent_buffer *l;
1014
1015         /* FIXME use last free of some kind */
1016
1017         /* we don't want to overwrite the superblock on the drive,
1018          * so we make sure to start at an offset of at least 1MB
1019          */
1020         search_start = max(root->fs_info->alloc_start, 1024ull * 1024);
1021
1022         max_hole_start = search_start;
1023         max_hole_size = 0;
1024         hole_size = 0;
1025
1026         if (search_start >= search_end || device->is_tgtdev_for_dev_replace) {
1027                 ret = -ENOSPC;
1028                 goto error;
1029         }
1030
1031         path = btrfs_alloc_path();
1032         if (!path) {
1033                 ret = -ENOMEM;
1034                 goto error;
1035         }
1036         path->reada = 2;
1037
1038         key.objectid = device->devid;
1039         key.offset = search_start;
1040         key.type = BTRFS_DEV_EXTENT_KEY;
1041
1042         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1043         if (ret < 0)
1044                 goto out;
1045         if (ret > 0) {
1046                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1047                 if (ret < 0)
1048                         goto out;
1049         }
1050
1051         while (1) {
1052                 l = path->nodes[0];
1053                 slot = path->slots[0];
1054                 if (slot >= btrfs_header_nritems(l)) {
1055                         ret = btrfs_next_leaf(root, path);
1056                         if (ret == 0)
1057                                 continue;
1058                         if (ret < 0)
1059                                 goto out;
1060
1061                         break;
1062                 }
1063                 btrfs_item_key_to_cpu(l, &key, slot);
1064
1065                 if (key.objectid < device->devid)
1066                         goto next;
1067
1068                 if (key.objectid > device->devid)
1069                         break;
1070
1071                 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
1072                         goto next;
1073
1074                 if (key.offset > search_start) {
1075                         hole_size = key.offset - search_start;
1076
1077                         if (hole_size > max_hole_size) {
1078                                 max_hole_start = search_start;
1079                                 max_hole_size = hole_size;
1080                         }
1081
1082                         /*
1083                          * If this free space is greater than which we need,
1084                          * it must be the max free space that we have found
1085                          * until now, so max_hole_start must point to the start
1086                          * of this free space and the length of this free space
1087                          * is stored in max_hole_size. Thus, we return
1088                          * max_hole_start and max_hole_size and go back to the
1089                          * caller.
1090                          */
1091                         if (hole_size >= num_bytes) {
1092                                 ret = 0;
1093                                 goto out;
1094                         }
1095                 }
1096
1097                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1098                 extent_end = key.offset + btrfs_dev_extent_length(l,
1099                                                                   dev_extent);
1100                 if (extent_end > search_start)
1101                         search_start = extent_end;
1102 next:
1103                 path->slots[0]++;
1104                 cond_resched();
1105         }
1106
1107         /*
1108          * At this point, search_start should be the end of
1109          * allocated dev extents, and when shrinking the device,
1110          * search_end may be smaller than search_start.
1111          */
1112         if (search_end > search_start)
1113                 hole_size = search_end - search_start;
1114
1115         if (hole_size > max_hole_size) {
1116                 max_hole_start = search_start;
1117                 max_hole_size = hole_size;
1118         }
1119
1120         /* See above. */
1121         if (hole_size < num_bytes)
1122                 ret = -ENOSPC;
1123         else
1124                 ret = 0;
1125
1126 out:
1127         btrfs_free_path(path);
1128 error:
1129         *start = max_hole_start;
1130         if (len)
1131                 *len = max_hole_size;
1132         return ret;
1133 }
1134
1135 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1136                           struct btrfs_device *device,
1137                           u64 start)
1138 {
1139         int ret;
1140         struct btrfs_path *path;
1141         struct btrfs_root *root = device->dev_root;
1142         struct btrfs_key key;
1143         struct btrfs_key found_key;
1144         struct extent_buffer *leaf = NULL;
1145         struct btrfs_dev_extent *extent = NULL;
1146
1147         path = btrfs_alloc_path();
1148         if (!path)
1149                 return -ENOMEM;
1150
1151         key.objectid = device->devid;
1152         key.offset = start;
1153         key.type = BTRFS_DEV_EXTENT_KEY;
1154 again:
1155         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1156         if (ret > 0) {
1157                 ret = btrfs_previous_item(root, path, key.objectid,
1158                                           BTRFS_DEV_EXTENT_KEY);
1159                 if (ret)
1160                         goto out;
1161                 leaf = path->nodes[0];
1162                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1163                 extent = btrfs_item_ptr(leaf, path->slots[0],
1164                                         struct btrfs_dev_extent);
1165                 BUG_ON(found_key.offset > start || found_key.offset +
1166                        btrfs_dev_extent_length(leaf, extent) < start);
1167                 key = found_key;
1168                 btrfs_release_path(path);
1169                 goto again;
1170         } else if (ret == 0) {
1171                 leaf = path->nodes[0];
1172                 extent = btrfs_item_ptr(leaf, path->slots[0],
1173                                         struct btrfs_dev_extent);
1174         } else {
1175                 btrfs_error(root->fs_info, ret, "Slot search failed");
1176                 goto out;
1177         }
1178
1179         if (device->bytes_used > 0) {
1180                 u64 len = btrfs_dev_extent_length(leaf, extent);
1181                 device->bytes_used -= len;
1182                 spin_lock(&root->fs_info->free_chunk_lock);
1183                 root->fs_info->free_chunk_space += len;
1184                 spin_unlock(&root->fs_info->free_chunk_lock);
1185         }
1186         ret = btrfs_del_item(trans, root, path);
1187         if (ret) {
1188                 btrfs_error(root->fs_info, ret,
1189                             "Failed to remove dev extent item");
1190         }
1191 out:
1192         btrfs_free_path(path);
1193         return ret;
1194 }
1195
1196 int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1197                            struct btrfs_device *device,
1198                            u64 chunk_tree, u64 chunk_objectid,
1199                            u64 chunk_offset, u64 start, u64 num_bytes)
1200 {
1201         int ret;
1202         struct btrfs_path *path;
1203         struct btrfs_root *root = device->dev_root;
1204         struct btrfs_dev_extent *extent;
1205         struct extent_buffer *leaf;
1206         struct btrfs_key key;
1207
1208         WARN_ON(!device->in_fs_metadata);
1209         WARN_ON(device->is_tgtdev_for_dev_replace);
1210         path = btrfs_alloc_path();
1211         if (!path)
1212                 return -ENOMEM;
1213
1214         key.objectid = device->devid;
1215         key.offset = start;
1216         key.type = BTRFS_DEV_EXTENT_KEY;
1217         ret = btrfs_insert_empty_item(trans, root, path, &key,
1218                                       sizeof(*extent));
1219         if (ret)
1220                 goto out;
1221
1222         leaf = path->nodes[0];
1223         extent = btrfs_item_ptr(leaf, path->slots[0],
1224                                 struct btrfs_dev_extent);
1225         btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
1226         btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
1227         btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1228
1229         write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
1230                     (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
1231                     BTRFS_UUID_SIZE);
1232
1233         btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1234         btrfs_mark_buffer_dirty(leaf);
1235 out:
1236         btrfs_free_path(path);
1237         return ret;
1238 }
1239
1240 static noinline int find_next_chunk(struct btrfs_root *root,
1241                                     u64 objectid, u64 *offset)
1242 {
1243         struct btrfs_path *path;
1244         int ret;
1245         struct btrfs_key key;
1246         struct btrfs_chunk *chunk;
1247         struct btrfs_key found_key;
1248
1249         path = btrfs_alloc_path();
1250         if (!path)
1251                 return -ENOMEM;
1252
1253         key.objectid = objectid;
1254         key.offset = (u64)-1;
1255         key.type = BTRFS_CHUNK_ITEM_KEY;
1256
1257         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1258         if (ret < 0)
1259                 goto error;
1260
1261         BUG_ON(ret == 0); /* Corruption */
1262
1263         ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
1264         if (ret) {
1265                 *offset = 0;
1266         } else {
1267                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1268                                       path->slots[0]);
1269                 if (found_key.objectid != objectid)
1270                         *offset = 0;
1271                 else {
1272                         chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
1273                                                struct btrfs_chunk);
1274                         *offset = found_key.offset +
1275                                 btrfs_chunk_length(path->nodes[0], chunk);
1276                 }
1277         }
1278         ret = 0;
1279 error:
1280         btrfs_free_path(path);
1281         return ret;
1282 }
1283
1284 static noinline int find_next_devid(struct btrfs_root *root, u64 *objectid)
1285 {
1286         int ret;
1287         struct btrfs_key key;
1288         struct btrfs_key found_key;
1289         struct btrfs_path *path;
1290
1291         root = root->fs_info->chunk_root;
1292
1293         path = btrfs_alloc_path();
1294         if (!path)
1295                 return -ENOMEM;
1296
1297         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1298         key.type = BTRFS_DEV_ITEM_KEY;
1299         key.offset = (u64)-1;
1300
1301         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1302         if (ret < 0)
1303                 goto error;
1304
1305         BUG_ON(ret == 0); /* Corruption */
1306
1307         ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
1308                                   BTRFS_DEV_ITEM_KEY);
1309         if (ret) {
1310                 *objectid = 1;
1311         } else {
1312                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1313                                       path->slots[0]);
1314                 *objectid = found_key.offset + 1;
1315         }
1316         ret = 0;
1317 error:
1318         btrfs_free_path(path);
1319         return ret;
1320 }
1321
1322 /*
1323  * the device information is stored in the chunk root
1324  * the btrfs_device struct should be fully filled in
1325  */
1326 int btrfs_add_device(struct btrfs_trans_handle *trans,
1327                      struct btrfs_root *root,
1328                      struct btrfs_device *device)
1329 {
1330         int ret;
1331         struct btrfs_path *path;
1332         struct btrfs_dev_item *dev_item;
1333         struct extent_buffer *leaf;
1334         struct btrfs_key key;
1335         unsigned long ptr;
1336
1337         root = root->fs_info->chunk_root;
1338
1339         path = btrfs_alloc_path();
1340         if (!path)
1341                 return -ENOMEM;
1342
1343         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1344         key.type = BTRFS_DEV_ITEM_KEY;
1345         key.offset = device->devid;
1346
1347         ret = btrfs_insert_empty_item(trans, root, path, &key,
1348                                       sizeof(*dev_item));
1349         if (ret)
1350                 goto out;
1351
1352         leaf = path->nodes[0];
1353         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1354
1355         btrfs_set_device_id(leaf, dev_item, device->devid);
1356         btrfs_set_device_generation(leaf, dev_item, 0);
1357         btrfs_set_device_type(leaf, dev_item, device->type);
1358         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1359         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1360         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1361         btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
1362         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1363         btrfs_set_device_group(leaf, dev_item, 0);
1364         btrfs_set_device_seek_speed(leaf, dev_item, 0);
1365         btrfs_set_device_bandwidth(leaf, dev_item, 0);
1366         btrfs_set_device_start_offset(leaf, dev_item, 0);
1367
1368         ptr = (unsigned long)btrfs_device_uuid(dev_item);
1369         write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1370         ptr = (unsigned long)btrfs_device_fsid(dev_item);
1371         write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1372         btrfs_mark_buffer_dirty(leaf);
1373
1374         ret = 0;
1375 out:
1376         btrfs_free_path(path);
1377         return ret;
1378 }
1379
1380 static int btrfs_rm_dev_item(struct btrfs_root *root,
1381                              struct btrfs_device *device)
1382 {
1383         int ret;
1384         struct btrfs_path *path;
1385         struct btrfs_key key;
1386         struct btrfs_trans_handle *trans;
1387
1388         root = root->fs_info->chunk_root;
1389
1390         path = btrfs_alloc_path();
1391         if (!path)
1392                 return -ENOMEM;
1393
1394         trans = btrfs_start_transaction(root, 0);
1395         if (IS_ERR(trans)) {
1396                 btrfs_free_path(path);
1397                 return PTR_ERR(trans);
1398         }
1399         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1400         key.type = BTRFS_DEV_ITEM_KEY;
1401         key.offset = device->devid;
1402         lock_chunks(root);
1403
1404         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1405         if (ret < 0)
1406                 goto out;
1407
1408         if (ret > 0) {
1409                 ret = -ENOENT;
1410                 goto out;
1411         }
1412
1413         ret = btrfs_del_item(trans, root, path);
1414         if (ret)
1415                 goto out;
1416 out:
1417         btrfs_free_path(path);
1418         unlock_chunks(root);
1419         btrfs_commit_transaction(trans, root);
1420         return ret;
1421 }
1422
1423 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
1424 {
1425         struct btrfs_device *device;
1426         struct btrfs_device *next_device;
1427         struct block_device *bdev;
1428         struct buffer_head *bh = NULL;
1429         struct btrfs_super_block *disk_super;
1430         struct btrfs_fs_devices *cur_devices;
1431         u64 all_avail;
1432         u64 devid;
1433         u64 num_devices;
1434         u8 *dev_uuid;
1435         unsigned seq;
1436         int ret = 0;
1437         bool clear_super = false;
1438
1439         mutex_lock(&uuid_mutex);
1440
1441         do {
1442                 seq = read_seqbegin(&root->fs_info->profiles_lock);
1443
1444                 all_avail = root->fs_info->avail_data_alloc_bits |
1445                             root->fs_info->avail_system_alloc_bits |
1446                             root->fs_info->avail_metadata_alloc_bits;
1447         } while (read_seqretry(&root->fs_info->profiles_lock, seq));
1448
1449         num_devices = root->fs_info->fs_devices->num_devices;
1450         btrfs_dev_replace_lock(&root->fs_info->dev_replace);
1451         if (btrfs_dev_replace_is_ongoing(&root->fs_info->dev_replace)) {
1452                 WARN_ON(num_devices < 1);
1453                 num_devices--;
1454         }
1455         btrfs_dev_replace_unlock(&root->fs_info->dev_replace);
1456
1457         if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) && num_devices <= 4) {
1458                 printk(KERN_ERR "btrfs: unable to go below four devices "
1459                        "on raid10\n");
1460                 ret = -EINVAL;
1461                 goto out;
1462         }
1463
1464         if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) && num_devices <= 2) {
1465                 printk(KERN_ERR "btrfs: unable to go below two "
1466                        "devices on raid1\n");
1467                 ret = -EINVAL;
1468                 goto out;
1469         }
1470
1471         if ((all_avail & BTRFS_BLOCK_GROUP_RAID5) &&
1472             root->fs_info->fs_devices->rw_devices <= 2) {
1473                 printk(KERN_ERR "btrfs: unable to go below two "
1474                        "devices on raid5\n");
1475                 ret = -EINVAL;
1476                 goto out;
1477         }
1478         if ((all_avail & BTRFS_BLOCK_GROUP_RAID6) &&
1479             root->fs_info->fs_devices->rw_devices <= 3) {
1480                 printk(KERN_ERR "btrfs: unable to go below three "
1481                        "devices on raid6\n");
1482                 ret = -EINVAL;
1483                 goto out;
1484         }
1485
1486         if (strcmp(device_path, "missing") == 0) {
1487                 struct list_head *devices;
1488                 struct btrfs_device *tmp;
1489
1490                 device = NULL;
1491                 devices = &root->fs_info->fs_devices->devices;
1492                 /*
1493                  * It is safe to read the devices since the volume_mutex
1494                  * is held.
1495                  */
1496                 list_for_each_entry(tmp, devices, dev_list) {
1497                         if (tmp->in_fs_metadata &&
1498                             !tmp->is_tgtdev_for_dev_replace &&
1499                             !tmp->bdev) {
1500                                 device = tmp;
1501                                 break;
1502                         }
1503                 }
1504                 bdev = NULL;
1505                 bh = NULL;
1506                 disk_super = NULL;
1507                 if (!device) {
1508                         printk(KERN_ERR "btrfs: no missing devices found to "
1509                                "remove\n");
1510                         goto out;
1511                 }
1512         } else {
1513                 ret = btrfs_get_bdev_and_sb(device_path,
1514                                             FMODE_WRITE | FMODE_EXCL,
1515                                             root->fs_info->bdev_holder, 0,
1516                                             &bdev, &bh);
1517                 if (ret)
1518                         goto out;
1519                 disk_super = (struct btrfs_super_block *)bh->b_data;
1520                 devid = btrfs_stack_device_id(&disk_super->dev_item);
1521                 dev_uuid = disk_super->dev_item.uuid;
1522                 device = btrfs_find_device(root->fs_info, devid, dev_uuid,
1523                                            disk_super->fsid);
1524                 if (!device) {
1525                         ret = -ENOENT;
1526                         goto error_brelse;
1527                 }
1528         }
1529
1530         if (device->is_tgtdev_for_dev_replace) {
1531                 pr_err("btrfs: unable to remove the dev_replace target dev\n");
1532                 ret = -EINVAL;
1533                 goto error_brelse;
1534         }
1535
1536         if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1537                 printk(KERN_ERR "btrfs: unable to remove the only writeable "
1538                        "device\n");
1539                 ret = -EINVAL;
1540                 goto error_brelse;
1541         }
1542
1543         if (device->writeable) {
1544                 lock_chunks(root);
1545                 list_del_init(&device->dev_alloc_list);
1546                 unlock_chunks(root);
1547                 root->fs_info->fs_devices->rw_devices--;
1548                 clear_super = true;
1549         }
1550
1551         ret = btrfs_shrink_device(device, 0);
1552         if (ret)
1553                 goto error_undo;
1554
1555         /*
1556          * TODO: the superblock still includes this device in its num_devices
1557          * counter although write_all_supers() is not locked out. This
1558          * could give a filesystem state which requires a degraded mount.
1559          */
1560         ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1561         if (ret)
1562                 goto error_undo;
1563
1564         spin_lock(&root->fs_info->free_chunk_lock);
1565         root->fs_info->free_chunk_space = device->total_bytes -
1566                 device->bytes_used;
1567         spin_unlock(&root->fs_info->free_chunk_lock);
1568
1569         device->in_fs_metadata = 0;
1570         btrfs_scrub_cancel_dev(root->fs_info, device);
1571
1572         /*
1573          * the device list mutex makes sure that we don't change
1574          * the device list while someone else is writing out all
1575          * the device supers.
1576          */
1577
1578         cur_devices = device->fs_devices;
1579         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1580         list_del_rcu(&device->dev_list);
1581
1582         device->fs_devices->num_devices--;
1583         device->fs_devices->total_devices--;
1584
1585         if (device->missing)
1586                 root->fs_info->fs_devices->missing_devices--;
1587
1588         next_device = list_entry(root->fs_info->fs_devices->devices.next,
1589                                  struct btrfs_device, dev_list);
1590         if (device->bdev == root->fs_info->sb->s_bdev)
1591                 root->fs_info->sb->s_bdev = next_device->bdev;
1592         if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1593                 root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1594
1595         if (device->bdev)
1596                 device->fs_devices->open_devices--;
1597
1598         call_rcu(&device->rcu, free_device);
1599         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1600
1601         num_devices = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
1602         btrfs_set_super_num_devices(root->fs_info->super_copy, num_devices);
1603
1604         if (cur_devices->open_devices == 0) {
1605                 struct btrfs_fs_devices *fs_devices;
1606                 fs_devices = root->fs_info->fs_devices;
1607                 while (fs_devices) {
1608                         if (fs_devices->seed == cur_devices)
1609                                 break;
1610                         fs_devices = fs_devices->seed;
1611                 }
1612                 fs_devices->seed = cur_devices->seed;
1613                 cur_devices->seed = NULL;
1614                 lock_chunks(root);
1615                 __btrfs_close_devices(cur_devices);
1616                 unlock_chunks(root);
1617                 free_fs_devices(cur_devices);
1618         }
1619
1620         root->fs_info->num_tolerated_disk_barrier_failures =
1621                 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
1622
1623         /*
1624          * at this point, the device is zero sized.  We want to
1625          * remove it from the devices list and zero out the old super
1626          */
1627         if (clear_super && disk_super) {
1628                 /* make sure this device isn't detected as part of
1629                  * the FS anymore
1630                  */
1631                 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1632                 set_buffer_dirty(bh);
1633                 sync_dirty_buffer(bh);
1634         }
1635
1636         ret = 0;
1637
1638         /* Notify udev that device has changed */
1639         if (bdev)
1640                 btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
1641
1642 error_brelse:
1643         brelse(bh);
1644         if (bdev)
1645                 blkdev_put(bdev, FMODE_READ | FMODE_EXCL);
1646 out:
1647         mutex_unlock(&uuid_mutex);
1648         return ret;
1649 error_undo:
1650         if (device->writeable) {
1651                 lock_chunks(root);
1652                 list_add(&device->dev_alloc_list,
1653                          &root->fs_info->fs_devices->alloc_list);
1654                 unlock_chunks(root);
1655                 root->fs_info->fs_devices->rw_devices++;
1656         }
1657         goto error_brelse;
1658 }
1659
1660 void btrfs_rm_dev_replace_srcdev(struct btrfs_fs_info *fs_info,
1661                                  struct btrfs_device *srcdev)
1662 {
1663         WARN_ON(!mutex_is_locked(&fs_info->fs_devices->device_list_mutex));
1664         list_del_rcu(&srcdev->dev_list);
1665         list_del_rcu(&srcdev->dev_alloc_list);
1666         fs_info->fs_devices->num_devices--;
1667         if (srcdev->missing) {
1668                 fs_info->fs_devices->missing_devices--;
1669                 fs_info->fs_devices->rw_devices++;
1670         }
1671         if (srcdev->can_discard)
1672                 fs_info->fs_devices->num_can_discard--;
1673         if (srcdev->bdev)
1674                 fs_info->fs_devices->open_devices--;
1675
1676         call_rcu(&srcdev->rcu, free_device);
1677 }
1678
1679 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
1680                                       struct btrfs_device *tgtdev)
1681 {
1682         struct btrfs_device *next_device;
1683
1684         WARN_ON(!tgtdev);
1685         mutex_lock(&fs_info->fs_devices->device_list_mutex);
1686         if (tgtdev->bdev) {
1687                 btrfs_scratch_superblock(tgtdev);
1688                 fs_info->fs_devices->open_devices--;
1689         }
1690         fs_info->fs_devices->num_devices--;
1691         if (tgtdev->can_discard)
1692                 fs_info->fs_devices->num_can_discard++;
1693
1694         next_device = list_entry(fs_info->fs_devices->devices.next,
1695                                  struct btrfs_device, dev_list);
1696         if (tgtdev->bdev == fs_info->sb->s_bdev)
1697                 fs_info->sb->s_bdev = next_device->bdev;
1698         if (tgtdev->bdev == fs_info->fs_devices->latest_bdev)
1699                 fs_info->fs_devices->latest_bdev = next_device->bdev;
1700         list_del_rcu(&tgtdev->dev_list);
1701
1702         call_rcu(&tgtdev->rcu, free_device);
1703
1704         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
1705 }
1706
1707 int btrfs_find_device_by_path(struct btrfs_root *root, char *device_path,
1708                               struct btrfs_device **device)
1709 {
1710         int ret = 0;
1711         struct btrfs_super_block *disk_super;
1712         u64 devid;
1713         u8 *dev_uuid;
1714         struct block_device *bdev;
1715         struct buffer_head *bh;
1716
1717         *device = NULL;
1718         ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
1719                                     root->fs_info->bdev_holder, 0, &bdev, &bh);
1720         if (ret)
1721                 return ret;
1722         disk_super = (struct btrfs_super_block *)bh->b_data;
1723         devid = btrfs_stack_device_id(&disk_super->dev_item);
1724         dev_uuid = disk_super->dev_item.uuid;
1725         *device = btrfs_find_device(root->fs_info, devid, dev_uuid,
1726                                     disk_super->fsid);
1727         brelse(bh);
1728         if (!*device)
1729                 ret = -ENOENT;
1730         blkdev_put(bdev, FMODE_READ);
1731         return ret;
1732 }
1733
1734 int btrfs_find_device_missing_or_by_path(struct btrfs_root *root,
1735                                          char *device_path,
1736                                          struct btrfs_device **device)
1737 {
1738         *device = NULL;
1739         if (strcmp(device_path, "missing") == 0) {
1740                 struct list_head *devices;
1741                 struct btrfs_device *tmp;
1742
1743                 devices = &root->fs_info->fs_devices->devices;
1744                 /*
1745                  * It is safe to read the devices since the volume_mutex
1746                  * is held by the caller.
1747                  */
1748                 list_for_each_entry(tmp, devices, dev_list) {
1749                         if (tmp->in_fs_metadata && !tmp->bdev) {
1750                                 *device = tmp;
1751                                 break;
1752                         }
1753                 }
1754
1755                 if (!*device) {
1756                         pr_err("btrfs: no missing device found\n");
1757                         return -ENOENT;
1758                 }
1759
1760                 return 0;
1761         } else {
1762                 return btrfs_find_device_by_path(root, device_path, device);
1763         }
1764 }
1765
1766 /*
1767  * does all the dirty work required for changing file system's UUID.
1768  */
1769 static int btrfs_prepare_sprout(struct btrfs_root *root)
1770 {
1771         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
1772         struct btrfs_fs_devices *old_devices;
1773         struct btrfs_fs_devices *seed_devices;
1774         struct btrfs_super_block *disk_super = root->fs_info->super_copy;
1775         struct btrfs_device *device;
1776         u64 super_flags;
1777
1778         BUG_ON(!mutex_is_locked(&uuid_mutex));
1779         if (!fs_devices->seeding)
1780                 return -EINVAL;
1781
1782         seed_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
1783         if (!seed_devices)
1784                 return -ENOMEM;
1785
1786         old_devices = clone_fs_devices(fs_devices);
1787         if (IS_ERR(old_devices)) {
1788                 kfree(seed_devices);
1789                 return PTR_ERR(old_devices);
1790         }
1791
1792         list_add(&old_devices->list, &fs_uuids);
1793
1794         memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
1795         seed_devices->opened = 1;
1796         INIT_LIST_HEAD(&seed_devices->devices);
1797         INIT_LIST_HEAD(&seed_devices->alloc_list);
1798         mutex_init(&seed_devices->device_list_mutex);
1799
1800         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1801         list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
1802                               synchronize_rcu);
1803         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1804
1805         list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
1806         list_for_each_entry(device, &seed_devices->devices, dev_list) {
1807                 device->fs_devices = seed_devices;
1808         }
1809
1810         fs_devices->seeding = 0;
1811         fs_devices->num_devices = 0;
1812         fs_devices->open_devices = 0;
1813         fs_devices->total_devices = 0;
1814         fs_devices->seed = seed_devices;
1815
1816         generate_random_uuid(fs_devices->fsid);
1817         memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1818         memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1819         super_flags = btrfs_super_flags(disk_super) &
1820                       ~BTRFS_SUPER_FLAG_SEEDING;
1821         btrfs_set_super_flags(disk_super, super_flags);
1822
1823         return 0;
1824 }
1825
1826 /*
1827  * strore the expected generation for seed devices in device items.
1828  */
1829 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
1830                                struct btrfs_root *root)
1831 {
1832         struct btrfs_path *path;
1833         struct extent_buffer *leaf;
1834         struct btrfs_dev_item *dev_item;
1835         struct btrfs_device *device;
1836         struct btrfs_key key;
1837         u8 fs_uuid[BTRFS_UUID_SIZE];
1838         u8 dev_uuid[BTRFS_UUID_SIZE];
1839         u64 devid;
1840         int ret;
1841
1842         path = btrfs_alloc_path();
1843         if (!path)
1844                 return -ENOMEM;
1845
1846         root = root->fs_info->chunk_root;
1847         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1848         key.offset = 0;
1849         key.type = BTRFS_DEV_ITEM_KEY;
1850
1851         while (1) {
1852                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1853                 if (ret < 0)
1854                         goto error;
1855
1856                 leaf = path->nodes[0];
1857 next_slot:
1858                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1859                         ret = btrfs_next_leaf(root, path);
1860                         if (ret > 0)
1861                                 break;
1862                         if (ret < 0)
1863                                 goto error;
1864                         leaf = path->nodes[0];
1865                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1866                         btrfs_release_path(path);
1867                         continue;
1868                 }
1869
1870                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1871                 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
1872                     key.type != BTRFS_DEV_ITEM_KEY)
1873                         break;
1874
1875                 dev_item = btrfs_item_ptr(leaf, path->slots[0],
1876                                           struct btrfs_dev_item);
1877                 devid = btrfs_device_id(leaf, dev_item);
1878                 read_extent_buffer(leaf, dev_uuid,
1879                                    (unsigned long)btrfs_device_uuid(dev_item),
1880                                    BTRFS_UUID_SIZE);
1881                 read_extent_buffer(leaf, fs_uuid,
1882                                    (unsigned long)btrfs_device_fsid(dev_item),
1883                                    BTRFS_UUID_SIZE);
1884                 device = btrfs_find_device(root->fs_info, devid, dev_uuid,
1885                                            fs_uuid);
1886                 BUG_ON(!device); /* Logic error */
1887
1888                 if (device->fs_devices->seeding) {
1889                         btrfs_set_device_generation(leaf, dev_item,
1890                                                     device->generation);
1891                         btrfs_mark_buffer_dirty(leaf);
1892                 }
1893
1894                 path->slots[0]++;
1895                 goto next_slot;
1896         }
1897         ret = 0;
1898 error:
1899         btrfs_free_path(path);
1900         return ret;
1901 }
1902
1903 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
1904 {
1905         struct request_queue *q;
1906         struct btrfs_trans_handle *trans;
1907         struct btrfs_device *device;
1908         struct block_device *bdev;
1909         struct list_head *devices;
1910         struct super_block *sb = root->fs_info->sb;
1911         struct rcu_string *name;
1912         u64 total_bytes;
1913         int seeding_dev = 0;
1914         int ret = 0;
1915
1916         if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
1917                 return -EROFS;
1918
1919         bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
1920                                   root->fs_info->bdev_holder);
1921         if (IS_ERR(bdev))
1922                 return PTR_ERR(bdev);
1923
1924         if (root->fs_info->fs_devices->seeding) {
1925                 seeding_dev = 1;
1926                 down_write(&sb->s_umount);
1927                 mutex_lock(&uuid_mutex);
1928         }
1929
1930         filemap_write_and_wait(bdev->bd_inode->i_mapping);
1931
1932         devices = &root->fs_info->fs_devices->devices;
1933
1934         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1935         list_for_each_entry(device, devices, dev_list) {
1936                 if (device->bdev == bdev) {
1937                         ret = -EEXIST;
1938                         mutex_unlock(
1939                                 &root->fs_info->fs_devices->device_list_mutex);
1940                         goto error;
1941                 }
1942         }
1943         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1944
1945         device = kzalloc(sizeof(*device), GFP_NOFS);
1946         if (!device) {
1947                 /* we can safely leave the fs_devices entry around */
1948                 ret = -ENOMEM;
1949                 goto error;
1950         }
1951
1952         name = rcu_string_strdup(device_path, GFP_NOFS);
1953         if (!name) {
1954                 kfree(device);
1955                 ret = -ENOMEM;
1956                 goto error;
1957         }
1958         rcu_assign_pointer(device->name, name);
1959
1960         ret = find_next_devid(root, &device->devid);
1961         if (ret) {
1962                 rcu_string_free(device->name);
1963                 kfree(device);
1964                 goto error;
1965         }
1966
1967         trans = btrfs_start_transaction(root, 0);
1968         if (IS_ERR(trans)) {
1969                 rcu_string_free(device->name);
1970                 kfree(device);
1971                 ret = PTR_ERR(trans);
1972                 goto error;
1973         }
1974
1975         lock_chunks(root);
1976
1977         q = bdev_get_queue(bdev);
1978         if (blk_queue_discard(q))
1979                 device->can_discard = 1;
1980         device->writeable = 1;
1981         device->work.func = pending_bios_fn;
1982         generate_random_uuid(device->uuid);
1983         spin_lock_init(&device->io_lock);
1984         device->generation = trans->transid;
1985         device->io_width = root->sectorsize;
1986         device->io_align = root->sectorsize;
1987         device->sector_size = root->sectorsize;
1988         device->total_bytes = i_size_read(bdev->bd_inode);
1989         device->disk_total_bytes = device->total_bytes;
1990         device->dev_root = root->fs_info->dev_root;
1991         device->bdev = bdev;
1992         device->in_fs_metadata = 1;
1993         device->is_tgtdev_for_dev_replace = 0;
1994         device->mode = FMODE_EXCL;
1995         set_blocksize(device->bdev, 4096);
1996
1997         if (seeding_dev) {
1998                 sb->s_flags &= ~MS_RDONLY;
1999                 ret = btrfs_prepare_sprout(root);
2000                 BUG_ON(ret); /* -ENOMEM */
2001         }
2002
2003         device->fs_devices = root->fs_info->fs_devices;
2004
2005         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2006         list_add_rcu(&device->dev_list, &root->fs_info->fs_devices->devices);
2007         list_add(&device->dev_alloc_list,
2008                  &root->fs_info->fs_devices->alloc_list);
2009         root->fs_info->fs_devices->num_devices++;
2010         root->fs_info->fs_devices->open_devices++;
2011         root->fs_info->fs_devices->rw_devices++;
2012         root->fs_info->fs_devices->total_devices++;
2013         if (device->can_discard)
2014                 root->fs_info->fs_devices->num_can_discard++;
2015         root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
2016
2017         spin_lock(&root->fs_info->free_chunk_lock);
2018         root->fs_info->free_chunk_space += device->total_bytes;
2019         spin_unlock(&root->fs_info->free_chunk_lock);
2020
2021         if (!blk_queue_nonrot(bdev_get_queue(bdev)))
2022                 root->fs_info->fs_devices->rotating = 1;
2023
2024         total_bytes = btrfs_super_total_bytes(root->fs_info->super_copy);
2025         btrfs_set_super_total_bytes(root->fs_info->super_copy,
2026                                     total_bytes + device->total_bytes);
2027
2028         total_bytes = btrfs_super_num_devices(root->fs_info->super_copy);
2029         btrfs_set_super_num_devices(root->fs_info->super_copy,
2030                                     total_bytes + 1);
2031         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2032
2033         if (seeding_dev) {
2034                 ret = init_first_rw_device(trans, root, device);
2035                 if (ret) {
2036                         btrfs_abort_transaction(trans, root, ret);
2037                         goto error_trans;
2038                 }
2039                 ret = btrfs_finish_sprout(trans, root);
2040                 if (ret) {
2041                         btrfs_abort_transaction(trans, root, ret);
2042                         goto error_trans;
2043                 }
2044         } else {
2045                 ret = btrfs_add_device(trans, root, device);
2046                 if (ret) {
2047                         btrfs_abort_transaction(trans, root, ret);
2048                         goto error_trans;
2049                 }
2050         }
2051
2052         /*
2053          * we've got more storage, clear any full flags on the space
2054          * infos
2055          */
2056         btrfs_clear_space_info_full(root->fs_info);
2057
2058         unlock_chunks(root);
2059         root->fs_info->num_tolerated_disk_barrier_failures =
2060                 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
2061         ret = btrfs_commit_transaction(trans, root);
2062
2063         if (seeding_dev) {
2064                 mutex_unlock(&uuid_mutex);
2065                 up_write(&sb->s_umount);
2066
2067                 if (ret) /* transaction commit */
2068                         return ret;
2069
2070                 ret = btrfs_relocate_sys_chunks(root);
2071                 if (ret < 0)
2072                         btrfs_error(root->fs_info, ret,
2073                                     "Failed to relocate sys chunks after "
2074                                     "device initialization. This can be fixed "
2075                                     "using the \"btrfs balance\" command.");
2076                 trans = btrfs_attach_transaction(root);
2077                 if (IS_ERR(trans)) {
2078                         if (PTR_ERR(trans) == -ENOENT)
2079                                 return 0;
2080                         return PTR_ERR(trans);
2081                 }
2082                 ret = btrfs_commit_transaction(trans, root);
2083         }
2084
2085         return ret;
2086
2087 error_trans:
2088         unlock_chunks(root);
2089         btrfs_end_transaction(trans, root);
2090         rcu_string_free(device->name);
2091         kfree(device);
2092 error:
2093         blkdev_put(bdev, FMODE_EXCL);
2094         if (seeding_dev) {
2095                 mutex_unlock(&uuid_mutex);
2096                 up_write(&sb->s_umount);
2097         }
2098         return ret;
2099 }
2100
2101 int btrfs_init_dev_replace_tgtdev(struct btrfs_root *root, char *device_path,
2102                                   struct btrfs_device **device_out)
2103 {
2104         struct request_queue *q;
2105         struct btrfs_device *device;
2106         struct block_device *bdev;
2107         struct btrfs_fs_info *fs_info = root->fs_info;
2108         struct list_head *devices;
2109         struct rcu_string *name;
2110         int ret = 0;
2111
2112         *device_out = NULL;
2113         if (fs_info->fs_devices->seeding)
2114                 return -EINVAL;
2115
2116         bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2117                                   fs_info->bdev_holder);
2118         if (IS_ERR(bdev))
2119                 return PTR_ERR(bdev);
2120
2121         filemap_write_and_wait(bdev->bd_inode->i_mapping);
2122
2123         devices = &fs_info->fs_devices->devices;
2124         list_for_each_entry(device, devices, dev_list) {
2125                 if (device->bdev == bdev) {
2126                         ret = -EEXIST;
2127                         goto error;
2128                 }
2129         }
2130
2131         device = kzalloc(sizeof(*device), GFP_NOFS);
2132         if (!device) {
2133                 ret = -ENOMEM;
2134                 goto error;
2135         }
2136
2137         name = rcu_string_strdup(device_path, GFP_NOFS);
2138         if (!name) {
2139                 kfree(device);
2140                 ret = -ENOMEM;
2141                 goto error;
2142         }
2143         rcu_assign_pointer(device->name, name);
2144
2145         q = bdev_get_queue(bdev);
2146         if (blk_queue_discard(q))
2147                 device->can_discard = 1;
2148         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2149         device->writeable = 1;
2150         device->work.func = pending_bios_fn;
2151         generate_random_uuid(device->uuid);
2152         device->devid = BTRFS_DEV_REPLACE_DEVID;
2153         spin_lock_init(&device->io_lock);
2154         device->generation = 0;
2155         device->io_width = root->sectorsize;
2156         device->io_align = root->sectorsize;
2157         device->sector_size = root->sectorsize;
2158         device->total_bytes = i_size_read(bdev->bd_inode);
2159         device->disk_total_bytes = device->total_bytes;
2160         device->dev_root = fs_info->dev_root;
2161         device->bdev = bdev;
2162         device->in_fs_metadata = 1;
2163         device->is_tgtdev_for_dev_replace = 1;
2164         device->mode = FMODE_EXCL;
2165         set_blocksize(device->bdev, 4096);
2166         device->fs_devices = fs_info->fs_devices;
2167         list_add(&device->dev_list, &fs_info->fs_devices->devices);
2168         fs_info->fs_devices->num_devices++;
2169         fs_info->fs_devices->open_devices++;
2170         if (device->can_discard)
2171                 fs_info->fs_devices->num_can_discard++;
2172         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2173
2174         *device_out = device;
2175         return ret;
2176
2177 error:
2178         blkdev_put(bdev, FMODE_EXCL);
2179         return ret;
2180 }
2181
2182 void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info *fs_info,
2183                                               struct btrfs_device *tgtdev)
2184 {
2185         WARN_ON(fs_info->fs_devices->rw_devices == 0);
2186         tgtdev->io_width = fs_info->dev_root->sectorsize;
2187         tgtdev->io_align = fs_info->dev_root->sectorsize;
2188         tgtdev->sector_size = fs_info->dev_root->sectorsize;
2189         tgtdev->dev_root = fs_info->dev_root;
2190         tgtdev->in_fs_metadata = 1;
2191 }
2192
2193 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2194                                         struct btrfs_device *device)
2195 {
2196         int ret;
2197         struct btrfs_path *path;
2198         struct btrfs_root *root;
2199         struct btrfs_dev_item *dev_item;
2200         struct extent_buffer *leaf;
2201         struct btrfs_key key;
2202
2203         root = device->dev_root->fs_info->chunk_root;
2204
2205         path = btrfs_alloc_path();
2206         if (!path)
2207                 return -ENOMEM;
2208
2209         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2210         key.type = BTRFS_DEV_ITEM_KEY;
2211         key.offset = device->devid;
2212
2213         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2214         if (ret < 0)
2215                 goto out;
2216
2217         if (ret > 0) {
2218                 ret = -ENOENT;
2219                 goto out;
2220         }
2221
2222         leaf = path->nodes[0];
2223         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2224
2225         btrfs_set_device_id(leaf, dev_item, device->devid);
2226         btrfs_set_device_type(leaf, dev_item, device->type);
2227         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2228         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2229         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2230         btrfs_set_device_total_bytes(leaf, dev_item, device->disk_total_bytes);
2231         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
2232         btrfs_mark_buffer_dirty(leaf);
2233
2234 out:
2235         btrfs_free_path(path);
2236         return ret;
2237 }
2238
2239 static int __btrfs_grow_device(struct btrfs_trans_handle *trans,
2240                       struct btrfs_device *device, u64 new_size)
2241 {
2242         struct btrfs_super_block *super_copy =
2243                 device->dev_root->fs_info->super_copy;
2244         u64 old_total = btrfs_super_total_bytes(super_copy);
2245         u64 diff = new_size - device->total_bytes;
2246
2247         if (!device->writeable)
2248                 return -EACCES;
2249         if (new_size <= device->total_bytes ||
2250             device->is_tgtdev_for_dev_replace)
2251                 return -EINVAL;
2252
2253         btrfs_set_super_total_bytes(super_copy, old_total + diff);
2254         device->fs_devices->total_rw_bytes += diff;
2255
2256         device->total_bytes = new_size;
2257         device->disk_total_bytes = new_size;
2258         btrfs_clear_space_info_full(device->dev_root->fs_info);
2259
2260         return btrfs_update_device(trans, device);
2261 }
2262
2263 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2264                       struct btrfs_device *device, u64 new_size)
2265 {
2266         int ret;
2267         lock_chunks(device->dev_root);
2268         ret = __btrfs_grow_device(trans, device, new_size);
2269         unlock_chunks(device->dev_root);
2270         return ret;
2271 }
2272
2273 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
2274                             struct btrfs_root *root,
2275                             u64 chunk_tree, u64 chunk_objectid,
2276                             u64 chunk_offset)
2277 {
2278         int ret;
2279         struct btrfs_path *path;
2280         struct btrfs_key key;
2281
2282         root = root->fs_info->chunk_root;
2283         path = btrfs_alloc_path();
2284         if (!path)
2285                 return -ENOMEM;
2286
2287         key.objectid = chunk_objectid;
2288         key.offset = chunk_offset;
2289         key.type = BTRFS_CHUNK_ITEM_KEY;
2290
2291         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2292         if (ret < 0)
2293                 goto out;
2294         else if (ret > 0) { /* Logic error or corruption */
2295                 btrfs_error(root->fs_info, -ENOENT,
2296                             "Failed lookup while freeing chunk.");
2297                 ret = -ENOENT;
2298                 goto out;
2299         }
2300
2301         ret = btrfs_del_item(trans, root, path);
2302         if (ret < 0)
2303                 btrfs_error(root->fs_info, ret,
2304                             "Failed to delete chunk item.");
2305 out:
2306         btrfs_free_path(path);
2307         return ret;
2308 }
2309
2310 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
2311                         chunk_offset)
2312 {
2313         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
2314         struct btrfs_disk_key *disk_key;
2315         struct btrfs_chunk *chunk;
2316         u8 *ptr;
2317         int ret = 0;
2318         u32 num_stripes;
2319         u32 array_size;
2320         u32 len = 0;
2321         u32 cur;
2322         struct btrfs_key key;
2323
2324         array_size = btrfs_super_sys_array_size(super_copy);
2325
2326         ptr = super_copy->sys_chunk_array;
2327         cur = 0;
2328
2329         while (cur < array_size) {
2330                 disk_key = (struct btrfs_disk_key *)ptr;
2331                 btrfs_disk_key_to_cpu(&key, disk_key);
2332
2333                 len = sizeof(*disk_key);
2334
2335                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2336                         chunk = (struct btrfs_chunk *)(ptr + len);
2337                         num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2338                         len += btrfs_chunk_item_size(num_stripes);
2339                 } else {
2340                         ret = -EIO;
2341                         break;
2342                 }
2343                 if (key.objectid == chunk_objectid &&
2344                     key.offset == chunk_offset) {
2345                         memmove(ptr, ptr + len, array_size - (cur + len));
2346                         array_size -= len;
2347                         btrfs_set_super_sys_array_size(super_copy, array_size);
2348                 } else {
2349                         ptr += len;
2350                         cur += len;
2351                 }
2352         }
2353         return ret;
2354 }
2355
2356 static int btrfs_relocate_chunk(struct btrfs_root *root,
2357                          u64 chunk_tree, u64 chunk_objectid,
2358                          u64 chunk_offset)
2359 {
2360         struct extent_map_tree *em_tree;
2361         struct btrfs_root *extent_root;
2362         struct btrfs_trans_handle *trans;
2363         struct extent_map *em;
2364         struct map_lookup *map;
2365         int ret;
2366         int i;
2367
2368         root = root->fs_info->chunk_root;
2369         extent_root = root->fs_info->extent_root;
2370         em_tree = &root->fs_info->mapping_tree.map_tree;
2371
2372         ret = btrfs_can_relocate(extent_root, chunk_offset);
2373         if (ret)
2374                 return -ENOSPC;
2375
2376         /* step one, relocate all the extents inside this chunk */
2377         ret = btrfs_relocate_block_group(extent_root, chunk_offset);
2378         if (ret)
2379                 return ret;
2380
2381         trans = btrfs_start_transaction(root, 0);
2382         BUG_ON(IS_ERR(trans));
2383
2384         lock_chunks(root);
2385
2386         /*
2387          * step two, delete the device extents and the
2388          * chunk tree entries
2389          */
2390         read_lock(&em_tree->lock);
2391         em = lookup_extent_mapping(em_tree, chunk_offset, 1);
2392         read_unlock(&em_tree->lock);
2393
2394         BUG_ON(!em || em->start > chunk_offset ||
2395                em->start + em->len < chunk_offset);
2396         map = (struct map_lookup *)em->bdev;
2397
2398         for (i = 0; i < map->num_stripes; i++) {
2399                 ret = btrfs_free_dev_extent(trans, map->stripes[i].dev,
2400                                             map->stripes[i].physical);
2401                 BUG_ON(ret);
2402
2403                 if (map->stripes[i].dev) {
2404                         ret = btrfs_update_device(trans, map->stripes[i].dev);
2405                         BUG_ON(ret);
2406                 }
2407         }
2408         ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
2409                                chunk_offset);
2410
2411         BUG_ON(ret);
2412
2413         trace_btrfs_chunk_free(root, map, chunk_offset, em->len);
2414
2415         if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2416                 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
2417                 BUG_ON(ret);
2418         }
2419
2420         ret = btrfs_remove_block_group(trans, extent_root, chunk_offset);
2421         BUG_ON(ret);
2422
2423         write_lock(&em_tree->lock);
2424         remove_extent_mapping(em_tree, em);
2425         write_unlock(&em_tree->lock);
2426
2427         kfree(map);
2428         em->bdev = NULL;
2429
2430         /* once for the tree */
2431         free_extent_map(em);
2432         /* once for us */
2433         free_extent_map(em);
2434
2435         unlock_chunks(root);
2436         btrfs_end_transaction(trans, root);
2437         return 0;
2438 }
2439
2440 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
2441 {
2442         struct btrfs_root *chunk_root = root->fs_info->chunk_root;
2443         struct btrfs_path *path;
2444         struct extent_buffer *leaf;
2445         struct btrfs_chunk *chunk;
2446         struct btrfs_key key;
2447         struct btrfs_key found_key;
2448         u64 chunk_tree = chunk_root->root_key.objectid;
2449         u64 chunk_type;
2450         bool retried = false;
2451         int failed = 0;
2452         int ret;
2453
2454         path = btrfs_alloc_path();
2455         if (!path)
2456                 return -ENOMEM;
2457
2458 again:
2459         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2460         key.offset = (u64)-1;
2461         key.type = BTRFS_CHUNK_ITEM_KEY;
2462
2463         while (1) {
2464                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2465                 if (ret < 0)
2466                         goto error;
2467                 BUG_ON(ret == 0); /* Corruption */
2468
2469                 ret = btrfs_previous_item(chunk_root, path, key.objectid,
2470                                           key.type);
2471                 if (ret < 0)
2472                         goto error;
2473                 if (ret > 0)
2474                         break;
2475
2476                 leaf = path->nodes[0];
2477                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2478
2479                 chunk = btrfs_item_ptr(leaf, path->slots[0],
2480                                        struct btrfs_chunk);
2481                 chunk_type = btrfs_chunk_type(leaf, chunk);
2482                 btrfs_release_path(path);
2483
2484                 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
2485                         ret = btrfs_relocate_chunk(chunk_root, chunk_tree,
2486                                                    found_key.objectid,
2487                                                    found_key.offset);
2488                         if (ret == -ENOSPC)
2489                                 failed++;
2490                         else if (ret)
2491                                 BUG();
2492                 }
2493
2494                 if (found_key.offset == 0)
2495                         break;
2496                 key.offset = found_key.offset - 1;
2497         }
2498         ret = 0;
2499         if (failed && !retried) {
2500                 failed = 0;
2501                 retried = true;
2502                 goto again;
2503         } else if (failed && retried) {
2504                 WARN_ON(1);
2505                 ret = -ENOSPC;
2506         }
2507 error:
2508         btrfs_free_path(path);
2509         return ret;
2510 }
2511
2512 static int insert_balance_item(struct btrfs_root *root,
2513                                struct btrfs_balance_control *bctl)
2514 {
2515         struct btrfs_trans_handle *trans;
2516         struct btrfs_balance_item *item;
2517         struct btrfs_disk_balance_args disk_bargs;
2518         struct btrfs_path *path;
2519         struct extent_buffer *leaf;
2520         struct btrfs_key key;
2521         int ret, err;
2522
2523         path = btrfs_alloc_path();
2524         if (!path)
2525                 return -ENOMEM;
2526
2527         trans = btrfs_start_transaction(root, 0);
2528         if (IS_ERR(trans)) {
2529                 btrfs_free_path(path);
2530                 return PTR_ERR(trans);
2531         }
2532
2533         key.objectid = BTRFS_BALANCE_OBJECTID;
2534         key.type = BTRFS_BALANCE_ITEM_KEY;
2535         key.offset = 0;
2536
2537         ret = btrfs_insert_empty_item(trans, root, path, &key,
2538                                       sizeof(*item));
2539         if (ret)
2540                 goto out;
2541
2542         leaf = path->nodes[0];
2543         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
2544
2545         memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
2546
2547         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
2548         btrfs_set_balance_data(leaf, item, &disk_bargs);
2549         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
2550         btrfs_set_balance_meta(leaf, item, &disk_bargs);
2551         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
2552         btrfs_set_balance_sys(leaf, item, &disk_bargs);
2553
2554         btrfs_set_balance_flags(leaf, item, bctl->flags);
2555
2556         btrfs_mark_buffer_dirty(leaf);
2557 out:
2558         btrfs_free_path(path);
2559         err = btrfs_commit_transaction(trans, root);
2560         if (err && !ret)
2561                 ret = err;
2562         return ret;
2563 }
2564
2565 static int del_balance_item(struct btrfs_root *root)
2566 {
2567         struct btrfs_trans_handle *trans;
2568         struct btrfs_path *path;
2569         struct btrfs_key key;
2570         int ret, err;
2571
2572         path = btrfs_alloc_path();
2573         if (!path)
2574                 return -ENOMEM;
2575
2576         trans = btrfs_start_transaction(root, 0);
2577         if (IS_ERR(trans)) {
2578                 btrfs_free_path(path);
2579                 return PTR_ERR(trans);
2580         }
2581
2582         key.objectid = BTRFS_BALANCE_OBJECTID;
2583         key.type = BTRFS_BALANCE_ITEM_KEY;
2584         key.offset = 0;
2585
2586         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2587         if (ret < 0)
2588                 goto out;
2589         if (ret > 0) {
2590                 ret = -ENOENT;
2591                 goto out;
2592         }
2593
2594         ret = btrfs_del_item(trans, root, path);
2595 out:
2596         btrfs_free_path(path);
2597         err = btrfs_commit_transaction(trans, root);
2598         if (err && !ret)
2599                 ret = err;
2600         return ret;
2601 }
2602
2603 /*
2604  * This is a heuristic used to reduce the number of chunks balanced on
2605  * resume after balance was interrupted.
2606  */
2607 static void update_balance_args(struct btrfs_balance_control *bctl)
2608 {
2609         /*
2610          * Turn on soft mode for chunk types that were being converted.
2611          */
2612         if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
2613                 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
2614         if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
2615                 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
2616         if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
2617                 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
2618
2619         /*
2620          * Turn on usage filter if is not already used.  The idea is
2621          * that chunks that we have already balanced should be
2622          * reasonably full.  Don't do it for chunks that are being
2623          * converted - that will keep us from relocating unconverted
2624          * (albeit full) chunks.
2625          */
2626         if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2627             !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2628                 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
2629                 bctl->data.usage = 90;
2630         }
2631         if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2632             !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2633                 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
2634                 bctl->sys.usage = 90;
2635         }
2636         if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2637             !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2638                 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
2639                 bctl->meta.usage = 90;
2640         }
2641 }
2642
2643 /*
2644  * Should be called with both balance and volume mutexes held to
2645  * serialize other volume operations (add_dev/rm_dev/resize) with
2646  * restriper.  Same goes for unset_balance_control.
2647  */
2648 static void set_balance_control(struct btrfs_balance_control *bctl)
2649 {
2650         struct btrfs_fs_info *fs_info = bctl->fs_info;
2651
2652         BUG_ON(fs_info->balance_ctl);
2653
2654         spin_lock(&fs_info->balance_lock);
2655         fs_info->balance_ctl = bctl;
2656         spin_unlock(&fs_info->balance_lock);
2657 }
2658
2659 static void unset_balance_control(struct btrfs_fs_info *fs_info)
2660 {
2661         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
2662
2663         BUG_ON(!fs_info->balance_ctl);
2664
2665         spin_lock(&fs_info->balance_lock);
2666         fs_info->balance_ctl = NULL;
2667         spin_unlock(&fs_info->balance_lock);
2668
2669         kfree(bctl);
2670 }
2671
2672 /*
2673  * Balance filters.  Return 1 if chunk should be filtered out
2674  * (should not be balanced).
2675  */
2676 static int chunk_profiles_filter(u64 chunk_type,
2677                                  struct btrfs_balance_args *bargs)
2678 {
2679         chunk_type = chunk_to_extended(chunk_type) &
2680                                 BTRFS_EXTENDED_PROFILE_MASK;
2681
2682         if (bargs->profiles & chunk_type)
2683                 return 0;
2684
2685         return 1;
2686 }
2687
2688 static int chunk_usage_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
2689                               struct btrfs_balance_args *bargs)
2690 {
2691         struct btrfs_block_group_cache *cache;
2692         u64 chunk_used, user_thresh;
2693         int ret = 1;
2694
2695         cache = btrfs_lookup_block_group(fs_info, chunk_offset);
2696         chunk_used = btrfs_block_group_used(&cache->item);
2697
2698         if (bargs->usage == 0)
2699                 user_thresh = 1;
2700         else if (bargs->usage > 100)
2701                 user_thresh = cache->key.offset;
2702         else
2703                 user_thresh = div_factor_fine(cache->key.offset,
2704                                               bargs->usage);
2705
2706         if (chunk_used < user_thresh)
2707                 ret = 0;
2708
2709         btrfs_put_block_group(cache);
2710         return ret;
2711 }
2712
2713 static int chunk_devid_filter(struct extent_buffer *leaf,
2714                               struct btrfs_chunk *chunk,
2715                               struct btrfs_balance_args *bargs)
2716 {
2717         struct btrfs_stripe *stripe;
2718         int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2719         int i;
2720
2721         for (i = 0; i < num_stripes; i++) {
2722                 stripe = btrfs_stripe_nr(chunk, i);
2723                 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
2724                         return 0;
2725         }
2726
2727         return 1;
2728 }
2729
2730 /* [pstart, pend) */
2731 static int chunk_drange_filter(struct extent_buffer *leaf,
2732                                struct btrfs_chunk *chunk,
2733                                u64 chunk_offset,
2734                                struct btrfs_balance_args *bargs)
2735 {
2736         struct btrfs_stripe *stripe;
2737         int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2738         u64 stripe_offset;
2739         u64 stripe_length;
2740         int factor;
2741         int i;
2742
2743         if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
2744                 return 0;
2745
2746         if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP |
2747              BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) {
2748                 factor = num_stripes / 2;
2749         } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) {
2750                 factor = num_stripes - 1;
2751         } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) {
2752                 factor = num_stripes - 2;
2753         } else {
2754                 factor = num_stripes;
2755         }
2756
2757         for (i = 0; i < num_stripes; i++) {
2758                 stripe = btrfs_stripe_nr(chunk, i);
2759                 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
2760                         continue;
2761
2762                 stripe_offset = btrfs_stripe_offset(leaf, stripe);
2763                 stripe_length = btrfs_chunk_length(leaf, chunk);
2764                 do_div(stripe_length, factor);
2765
2766                 if (stripe_offset < bargs->pend &&
2767                     stripe_offset + stripe_length > bargs->pstart)
2768                         return 0;
2769         }
2770
2771         return 1;
2772 }
2773
2774 /* [vstart, vend) */
2775 static int chunk_vrange_filter(struct extent_buffer *leaf,
2776                                struct btrfs_chunk *chunk,
2777                                u64 chunk_offset,
2778                                struct btrfs_balance_args *bargs)
2779 {
2780         if (chunk_offset < bargs->vend &&
2781             chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
2782                 /* at least part of the chunk is inside this vrange */
2783                 return 0;
2784
2785         return 1;
2786 }
2787
2788 static int chunk_soft_convert_filter(u64 chunk_type,
2789                                      struct btrfs_balance_args *bargs)
2790 {
2791         if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
2792                 return 0;
2793
2794         chunk_type = chunk_to_extended(chunk_type) &
2795                                 BTRFS_EXTENDED_PROFILE_MASK;
2796
2797         if (bargs->target == chunk_type)
2798                 return 1;
2799
2800         return 0;
2801 }
2802
2803 static int should_balance_chunk(struct btrfs_root *root,
2804                                 struct extent_buffer *leaf,
2805                                 struct btrfs_chunk *chunk, u64 chunk_offset)
2806 {
2807         struct btrfs_balance_control *bctl = root->fs_info->balance_ctl;
2808         struct btrfs_balance_args *bargs = NULL;
2809         u64 chunk_type = btrfs_chunk_type(leaf, chunk);
2810
2811         /* type filter */
2812         if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
2813               (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
2814                 return 0;
2815         }
2816
2817         if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
2818                 bargs = &bctl->data;
2819         else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
2820                 bargs = &bctl->sys;
2821         else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
2822                 bargs = &bctl->meta;
2823
2824         /* profiles filter */
2825         if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
2826             chunk_profiles_filter(chunk_type, bargs)) {
2827                 return 0;
2828         }
2829
2830         /* usage filter */
2831         if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
2832             chunk_usage_filter(bctl->fs_info, chunk_offset, bargs)) {
2833                 return 0;
2834         }
2835
2836         /* devid filter */
2837         if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
2838             chunk_devid_filter(leaf, chunk, bargs)) {
2839                 return 0;
2840         }
2841
2842         /* drange filter, makes sense only with devid filter */
2843         if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
2844             chunk_drange_filter(leaf, chunk, chunk_offset, bargs)) {
2845                 return 0;
2846         }
2847
2848         /* vrange filter */
2849         if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
2850             chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
2851                 return 0;
2852         }
2853
2854         /* soft profile changing mode */
2855         if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
2856             chunk_soft_convert_filter(chunk_type, bargs)) {
2857                 return 0;
2858         }
2859
2860         return 1;
2861 }
2862
2863 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
2864 {
2865         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
2866         struct btrfs_root *chunk_root = fs_info->chunk_root;
2867         struct btrfs_root *dev_root = fs_info->dev_root;
2868         struct list_head *devices;
2869         struct btrfs_device *device;
2870         u64 old_size;
2871         u64 size_to_free;
2872         struct btrfs_chunk *chunk;
2873         struct btrfs_path *path;
2874         struct btrfs_key key;
2875         struct btrfs_key found_key;
2876         struct btrfs_trans_handle *trans;
2877         struct extent_buffer *leaf;
2878         int slot;
2879         int ret;
2880         int enospc_errors = 0;
2881         bool counting = true;
2882
2883         /* step one make some room on all the devices */
2884         devices = &fs_info->fs_devices->devices;
2885         list_for_each_entry(device, devices, dev_list) {
2886                 old_size = device->total_bytes;
2887                 size_to_free = div_factor(old_size, 1);
2888                 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
2889                 if (!device->writeable ||
2890                     device->total_bytes - device->bytes_used > size_to_free ||
2891                     device->is_tgtdev_for_dev_replace)
2892                         continue;
2893
2894                 ret = btrfs_shrink_device(device, old_size - size_to_free);
2895                 if (ret == -ENOSPC)
2896                         break;
2897                 BUG_ON(ret);
2898
2899                 trans = btrfs_start_transaction(dev_root, 0);
2900                 BUG_ON(IS_ERR(trans));
2901
2902                 ret = btrfs_grow_device(trans, device, old_size);
2903                 BUG_ON(ret);
2904
2905                 btrfs_end_transaction(trans, dev_root);
2906         }
2907
2908         /* step two, relocate all the chunks */
2909         path = btrfs_alloc_path();
2910         if (!path) {
2911                 ret = -ENOMEM;
2912                 goto error;
2913         }
2914
2915         /* zero out stat counters */
2916         spin_lock(&fs_info->balance_lock);
2917         memset(&bctl->stat, 0, sizeof(bctl->stat));
2918         spin_unlock(&fs_info->balance_lock);
2919 again:
2920         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2921         key.offset = (u64)-1;
2922         key.type = BTRFS_CHUNK_ITEM_KEY;
2923
2924         while (1) {
2925                 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
2926                     atomic_read(&fs_info->balance_cancel_req)) {
2927                         ret = -ECANCELED;
2928                         goto error;
2929                 }
2930
2931                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2932                 if (ret < 0)
2933                         goto error;
2934
2935                 /*
2936                  * this shouldn't happen, it means the last relocate
2937                  * failed
2938                  */
2939                 if (ret == 0)
2940                         BUG(); /* FIXME break ? */
2941
2942                 ret = btrfs_previous_item(chunk_root, path, 0,
2943                                           BTRFS_CHUNK_ITEM_KEY);
2944                 if (ret) {
2945                         ret = 0;
2946                         break;
2947                 }
2948
2949                 leaf = path->nodes[0];
2950                 slot = path->slots[0];
2951                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2952
2953                 if (found_key.objectid != key.objectid)
2954                         break;
2955
2956                 /* chunk zero is special */
2957                 if (found_key.offset == 0)
2958                         break;
2959
2960                 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
2961
2962                 if (!counting) {
2963                         spin_lock(&fs_info->balance_lock);
2964                         bctl->stat.considered++;
2965                         spin_unlock(&fs_info->balance_lock);
2966                 }
2967
2968                 ret = should_balance_chunk(chunk_root, leaf, chunk,
2969                                            found_key.offset);
2970                 btrfs_release_path(path);
2971                 if (!ret)
2972                         goto loop;
2973
2974                 if (counting) {
2975                         spin_lock(&fs_info->balance_lock);
2976                         bctl->stat.expected++;
2977                         spin_unlock(&fs_info->balance_lock);
2978                         goto loop;
2979                 }
2980
2981                 ret = btrfs_relocate_chunk(chunk_root,
2982                                            chunk_root->root_key.objectid,
2983                                            found_key.objectid,
2984                                            found_key.offset);
2985                 if (ret && ret != -ENOSPC)
2986                         goto error;
2987                 if (ret == -ENOSPC) {
2988                         enospc_errors++;
2989                 } else {
2990                         spin_lock(&fs_info->balance_lock);
2991                         bctl->stat.completed++;
2992                         spin_unlock(&fs_info->balance_lock);
2993                 }
2994 loop:
2995                 key.offset = found_key.offset - 1;
2996         }
2997
2998         if (counting) {
2999                 btrfs_release_path(path);
3000                 counting = false;
3001                 goto again;
3002         }
3003 error:
3004         btrfs_free_path(path);
3005         if (enospc_errors) {
3006                 printk(KERN_INFO "btrfs: %d enospc errors during balance\n",
3007                        enospc_errors);
3008                 if (!ret)
3009                         ret = -ENOSPC;
3010         }
3011
3012         return ret;
3013 }
3014
3015 /**
3016  * alloc_profile_is_valid - see if a given profile is valid and reduced
3017  * @flags: profile to validate
3018  * @extended: if true @flags is treated as an extended profile
3019  */
3020 static int alloc_profile_is_valid(u64 flags, int extended)
3021 {
3022         u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
3023                                BTRFS_BLOCK_GROUP_PROFILE_MASK);
3024
3025         flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
3026
3027         /* 1) check that all other bits are zeroed */
3028         if (flags & ~mask)
3029                 return 0;
3030
3031         /* 2) see if profile is reduced */
3032         if (flags == 0)
3033                 return !extended; /* "0" is valid for usual profiles */
3034
3035         /* true if exactly one bit set */
3036         return (flags & (flags - 1)) == 0;
3037 }
3038
3039 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
3040 {
3041         /* cancel requested || normal exit path */
3042         return atomic_read(&fs_info->balance_cancel_req) ||
3043                 (atomic_read(&fs_info->balance_pause_req) == 0 &&
3044                  atomic_read(&fs_info->balance_cancel_req) == 0);
3045 }
3046
3047 static void __cancel_balance(struct btrfs_fs_info *fs_info)
3048 {
3049         int ret;
3050
3051         unset_balance_control(fs_info);
3052         ret = del_balance_item(fs_info->tree_root);
3053         BUG_ON(ret);
3054
3055         atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3056 }
3057
3058 void update_ioctl_balance_args(struct btrfs_fs_info *fs_info, int lock,
3059                                struct btrfs_ioctl_balance_args *bargs);
3060
3061 /*
3062  * Should be called with both balance and volume mutexes held
3063  */
3064 int btrfs_balance(struct btrfs_balance_control *bctl,
3065                   struct btrfs_ioctl_balance_args *bargs)
3066 {
3067         struct btrfs_fs_info *fs_info = bctl->fs_info;
3068         u64 allowed;
3069         int mixed = 0;
3070         int ret;
3071         u64 num_devices;
3072         unsigned seq;
3073
3074         if (btrfs_fs_closing(fs_info) ||
3075             atomic_read(&fs_info->balance_pause_req) ||
3076             atomic_read(&fs_info->balance_cancel_req)) {
3077                 ret = -EINVAL;
3078                 goto out;
3079         }
3080
3081         allowed = btrfs_super_incompat_flags(fs_info->super_copy);
3082         if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
3083                 mixed = 1;
3084
3085         /*
3086          * In case of mixed groups both data and meta should be picked,
3087          * and identical options should be given for both of them.
3088          */
3089         allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
3090         if (mixed && (bctl->flags & allowed)) {
3091                 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
3092                     !(bctl->flags & BTRFS_BALANCE_METADATA) ||
3093                     memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
3094                         printk(KERN_ERR "btrfs: with mixed groups data and "
3095                                "metadata balance options must be the same\n");
3096                         ret = -EINVAL;
3097                         goto out;
3098                 }
3099         }
3100
3101         num_devices = fs_info->fs_devices->num_devices;
3102         btrfs_dev_replace_lock(&fs_info->dev_replace);
3103         if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
3104                 BUG_ON(num_devices < 1);
3105                 num_devices--;
3106         }
3107         btrfs_dev_replace_unlock(&fs_info->dev_replace);
3108         allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE;
3109         if (num_devices == 1)
3110                 allowed |= BTRFS_BLOCK_GROUP_DUP;
3111         else if (num_devices < 4)
3112                 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1);
3113         else
3114                 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
3115                                 BTRFS_BLOCK_GROUP_RAID10 |
3116                                 BTRFS_BLOCK_GROUP_RAID5 |
3117                                 BTRFS_BLOCK_GROUP_RAID6);
3118
3119         if ((bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3120             (!alloc_profile_is_valid(bctl->data.target, 1) ||
3121              (bctl->data.target & ~allowed))) {
3122                 printk(KERN_ERR "btrfs: unable to start balance with target "
3123                        "data profile %llu\n",
3124                        (unsigned long long)bctl->data.target);
3125                 ret = -EINVAL;
3126                 goto out;
3127         }
3128         if ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3129             (!alloc_profile_is_valid(bctl->meta.target, 1) ||
3130              (bctl->meta.target & ~allowed))) {
3131                 printk(KERN_ERR "btrfs: unable to start balance with target "
3132                        "metadata profile %llu\n",
3133                        (unsigned long long)bctl->meta.target);
3134                 ret = -EINVAL;
3135                 goto out;
3136         }
3137         if ((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3138             (!alloc_profile_is_valid(bctl->sys.target, 1) ||
3139              (bctl->sys.target & ~allowed))) {
3140                 printk(KERN_ERR "btrfs: unable to start balance with target "
3141                        "system profile %llu\n",
3142                        (unsigned long long)bctl->sys.target);
3143                 ret = -EINVAL;
3144                 goto out;
3145         }
3146
3147         /* allow dup'ed data chunks only in mixed mode */
3148         if (!mixed && (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3149             (bctl->data.target & BTRFS_BLOCK_GROUP_DUP)) {
3150                 printk(KERN_ERR "btrfs: dup for data is not allowed\n");
3151                 ret = -EINVAL;
3152                 goto out;
3153         }
3154
3155         /* allow to reduce meta or sys integrity only if force set */
3156         allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3157                         BTRFS_BLOCK_GROUP_RAID10 |
3158                         BTRFS_BLOCK_GROUP_RAID5 |
3159                         BTRFS_BLOCK_GROUP_RAID6;
3160         do {
3161                 seq = read_seqbegin(&fs_info->profiles_lock);
3162
3163                 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3164                      (fs_info->avail_system_alloc_bits & allowed) &&
3165                      !(bctl->sys.target & allowed)) ||
3166                     ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3167                      (fs_info->avail_metadata_alloc_bits & allowed) &&
3168                      !(bctl->meta.target & allowed))) {
3169                         if (bctl->flags & BTRFS_BALANCE_FORCE) {
3170                                 printk(KERN_INFO "btrfs: force reducing metadata "
3171                                        "integrity\n");
3172                         } else {
3173                                 printk(KERN_ERR "btrfs: balance will reduce metadata "
3174                                        "integrity, use force if you want this\n");
3175                                 ret = -EINVAL;
3176                                 goto out;
3177                         }
3178                 }
3179         } while (read_seqretry(&fs_info->profiles_lock, seq));
3180
3181         if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3182                 int num_tolerated_disk_barrier_failures;
3183                 u64 target = bctl->sys.target;
3184
3185                 num_tolerated_disk_barrier_failures =
3186                         btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3187                 if (num_tolerated_disk_barrier_failures > 0 &&
3188                     (target &
3189                      (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID0 |
3190                       BTRFS_AVAIL_ALLOC_BIT_SINGLE)))
3191                         num_tolerated_disk_barrier_failures = 0;
3192                 else if (num_tolerated_disk_barrier_failures > 1 &&
3193                          (target &
3194                           (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)))
3195                         num_tolerated_disk_barrier_failures = 1;
3196
3197                 fs_info->num_tolerated_disk_barrier_failures =
3198                         num_tolerated_disk_barrier_failures;
3199         }
3200
3201         ret = insert_balance_item(fs_info->tree_root, bctl);
3202         if (ret && ret != -EEXIST)
3203                 goto out;
3204
3205         if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
3206                 BUG_ON(ret == -EEXIST);
3207                 set_balance_control(bctl);
3208         } else {
3209                 BUG_ON(ret != -EEXIST);
3210                 spin_lock(&fs_info->balance_lock);
3211                 update_balance_args(bctl);
3212                 spin_unlock(&fs_info->balance_lock);
3213         }
3214
3215         atomic_inc(&fs_info->balance_running);
3216         mutex_unlock(&fs_info->balance_mutex);
3217
3218         ret = __btrfs_balance(fs_info);
3219
3220         mutex_lock(&fs_info->balance_mutex);
3221         atomic_dec(&fs_info->balance_running);
3222
3223         if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3224                 fs_info->num_tolerated_disk_barrier_failures =
3225                         btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3226         }
3227
3228         if (bargs) {
3229                 memset(bargs, 0, sizeof(*bargs));
3230                 update_ioctl_balance_args(fs_info, 0, bargs);
3231         }
3232
3233         wake_up(&fs_info->balance_wait_q);
3234
3235         return ret;
3236 out:
3237         if (bctl->flags & BTRFS_BALANCE_RESUME)
3238                 __cancel_balance(fs_info);
3239         else {
3240                 kfree(bctl);
3241                 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3242         }
3243         return ret;
3244 }
3245
3246 static int balance_kthread(void *data)
3247 {
3248         struct btrfs_fs_info *fs_info = data;
3249         int ret = 0;
3250
3251         mutex_lock(&fs_info->volume_mutex);
3252         mutex_lock(&fs_info->balance_mutex);
3253
3254         if (fs_info->balance_ctl) {
3255                 printk(KERN_INFO "btrfs: continuing balance\n");
3256                 ret = btrfs_balance(fs_info->balance_ctl, NULL);
3257         }
3258
3259         mutex_unlock(&fs_info->balance_mutex);
3260         mutex_unlock(&fs_info->volume_mutex);
3261
3262         return ret;
3263 }
3264
3265 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
3266 {
3267         struct task_struct *tsk;
3268
3269         spin_lock(&fs_info->balance_lock);
3270         if (!fs_info->balance_ctl) {
3271                 spin_unlock(&fs_info->balance_lock);
3272                 return 0;
3273         }
3274         spin_unlock(&fs_info->balance_lock);
3275
3276         if (btrfs_test_opt(fs_info->tree_root, SKIP_BALANCE)) {
3277                 printk(KERN_INFO "btrfs: force skipping balance\n");
3278                 return 0;
3279         }
3280
3281         tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
3282         if (IS_ERR(tsk))
3283                 return PTR_ERR(tsk);
3284
3285         return 0;
3286 }
3287
3288 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
3289 {
3290         struct btrfs_balance_control *bctl;
3291         struct btrfs_balance_item *item;
3292         struct btrfs_disk_balance_args disk_bargs;
3293         struct btrfs_path *path;
3294         struct extent_buffer *leaf;
3295         struct btrfs_key key;
3296         int ret;
3297
3298         path = btrfs_alloc_path();
3299         if (!path)
3300                 return -ENOMEM;
3301
3302         key.objectid = BTRFS_BALANCE_OBJECTID;
3303         key.type = BTRFS_BALANCE_ITEM_KEY;
3304         key.offset = 0;
3305
3306         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
3307         if (ret < 0)
3308                 goto out;
3309         if (ret > 0) { /* ret = -ENOENT; */
3310                 ret = 0;
3311                 goto out;
3312         }
3313
3314         bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
3315         if (!bctl) {
3316                 ret = -ENOMEM;
3317                 goto out;
3318         }
3319
3320         leaf = path->nodes[0];
3321         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3322
3323         bctl->fs_info = fs_info;
3324         bctl->flags = btrfs_balance_flags(leaf, item);
3325         bctl->flags |= BTRFS_BALANCE_RESUME;
3326
3327         btrfs_balance_data(leaf, item, &disk_bargs);
3328         btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
3329         btrfs_balance_meta(leaf, item, &disk_bargs);
3330         btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
3331         btrfs_balance_sys(leaf, item, &disk_bargs);
3332         btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
3333
3334         WARN_ON(atomic_xchg(&fs_info->mutually_exclusive_operation_running, 1));
3335
3336         mutex_lock(&fs_info->volume_mutex);
3337         mutex_lock(&fs_info->balance_mutex);
3338
3339         set_balance_control(bctl);
3340
3341         mutex_unlock(&fs_info->balance_mutex);
3342         mutex_unlock(&fs_info->volume_mutex);
3343 out:
3344         btrfs_free_path(path);
3345         return ret;
3346 }
3347
3348 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
3349 {
3350         int ret = 0;
3351
3352         mutex_lock(&fs_info->balance_mutex);
3353         if (!fs_info->balance_ctl) {
3354                 mutex_unlock(&fs_info->balance_mutex);
3355                 return -ENOTCONN;
3356         }
3357
3358         if (atomic_read(&fs_info->balance_running)) {
3359                 atomic_inc(&fs_info->balance_pause_req);
3360                 mutex_unlock(&fs_info->balance_mutex);
3361
3362                 wait_event(fs_info->balance_wait_q,
3363                            atomic_read(&fs_info->balance_running) == 0);
3364
3365                 mutex_lock(&fs_info->balance_mutex);
3366                 /* we are good with balance_ctl ripped off from under us */
3367                 BUG_ON(atomic_read(&fs_info->balance_running));
3368                 atomic_dec(&fs_info->balance_pause_req);
3369         } else {
3370                 ret = -ENOTCONN;
3371         }
3372
3373         mutex_unlock(&fs_info->balance_mutex);
3374         return ret;
3375 }
3376
3377 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
3378 {
3379         mutex_lock(&fs_info->balance_mutex);
3380         if (!fs_info->balance_ctl) {
3381                 mutex_unlock(&fs_info->balance_mutex);
3382                 return -ENOTCONN;
3383         }
3384
3385         atomic_inc(&fs_info->balance_cancel_req);
3386         /*
3387          * if we are running just wait and return, balance item is
3388          * deleted in btrfs_balance in this case
3389          */
3390         if (atomic_read(&fs_info->balance_running)) {
3391                 mutex_unlock(&fs_info->balance_mutex);
3392                 wait_event(fs_info->balance_wait_q,
3393                            atomic_read(&fs_info->balance_running) == 0);
3394                 mutex_lock(&fs_info->balance_mutex);
3395         } else {
3396                 /* __cancel_balance needs volume_mutex */
3397                 mutex_unlock(&fs_info->balance_mutex);
3398                 mutex_lock(&fs_info->volume_mutex);
3399                 mutex_lock(&fs_info->balance_mutex);
3400
3401                 if (fs_info->balance_ctl)
3402                         __cancel_balance(fs_info);
3403
3404                 mutex_unlock(&fs_info->volume_mutex);
3405         }
3406
3407         BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running));
3408         atomic_dec(&fs_info->balance_cancel_req);
3409         mutex_unlock(&fs_info->balance_mutex);
3410         return 0;
3411 }
3412
3413 /*
3414  * shrinking a device means finding all of the device extents past
3415  * the new size, and then following the back refs to the chunks.
3416  * The chunk relocation code actually frees the device extent
3417  */
3418 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
3419 {
3420         struct btrfs_trans_handle *trans;
3421         struct btrfs_root *root = device->dev_root;
3422         struct btrfs_dev_extent *dev_extent = NULL;
3423         struct btrfs_path *path;
3424         u64 length;
3425         u64 chunk_tree;
3426         u64 chunk_objectid;
3427         u64 chunk_offset;
3428         int ret;
3429         int slot;
3430         int failed = 0;
3431         bool retried = false;
3432         struct extent_buffer *l;
3433         struct btrfs_key key;
3434         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
3435         u64 old_total = btrfs_super_total_bytes(super_copy);
3436         u64 old_size = device->total_bytes;
3437         u64 diff = device->total_bytes - new_size;
3438
3439         if (device->is_tgtdev_for_dev_replace)
3440                 return -EINVAL;
3441
3442         path = btrfs_alloc_path();
3443         if (!path)
3444                 return -ENOMEM;
3445
3446         path->reada = 2;
3447
3448         lock_chunks(root);
3449
3450         device->total_bytes = new_size;
3451         if (device->writeable) {
3452                 device->fs_devices->total_rw_bytes -= diff;
3453                 spin_lock(&root->fs_info->free_chunk_lock);
3454                 root->fs_info->free_chunk_space -= diff;
3455                 spin_unlock(&root->fs_info->free_chunk_lock);
3456         }
3457         unlock_chunks(root);
3458
3459 again:
3460         key.objectid = device->devid;
3461         key.offset = (u64)-1;
3462         key.type = BTRFS_DEV_EXTENT_KEY;
3463
3464         do {
3465                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3466                 if (ret < 0)
3467                         goto done;
3468
3469                 ret = btrfs_previous_item(root, path, 0, key.type);
3470                 if (ret < 0)
3471                         goto done;
3472                 if (ret) {
3473                         ret = 0;
3474                         btrfs_release_path(path);
3475                         break;
3476                 }
3477
3478                 l = path->nodes[0];
3479                 slot = path->slots[0];
3480                 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
3481
3482                 if (key.objectid != device->devid) {
3483                         btrfs_release_path(path);
3484                         break;
3485                 }
3486
3487                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
3488                 length = btrfs_dev_extent_length(l, dev_extent);
3489
3490                 if (key.offset + length <= new_size) {
3491                         btrfs_release_path(path);
3492                         break;
3493                 }
3494
3495                 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
3496                 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
3497                 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
3498                 btrfs_release_path(path);
3499
3500                 ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
3501                                            chunk_offset);
3502                 if (ret && ret != -ENOSPC)
3503                         goto done;
3504                 if (ret == -ENOSPC)
3505                         failed++;
3506         } while (key.offset-- > 0);
3507
3508         if (failed && !retried) {
3509                 failed = 0;
3510                 retried = true;
3511                 goto again;
3512         } else if (failed && retried) {
3513                 ret = -ENOSPC;
3514                 lock_chunks(root);
3515
3516                 device->total_bytes = old_size;
3517                 if (device->writeable)
3518                         device->fs_devices->total_rw_bytes += diff;
3519                 spin_lock(&root->fs_info->free_chunk_lock);
3520                 root->fs_info->free_chunk_space += diff;
3521                 spin_unlock(&root->fs_info->free_chunk_lock);
3522                 unlock_chunks(root);
3523                 goto done;
3524         }
3525
3526         /* Shrinking succeeded, else we would be at "done". */
3527         trans = btrfs_start_transaction(root, 0);
3528         if (IS_ERR(trans)) {
3529                 ret = PTR_ERR(trans);
3530                 goto done;
3531         }
3532
3533         lock_chunks(root);
3534
3535         device->disk_total_bytes = new_size;
3536         /* Now btrfs_update_device() will change the on-disk size. */
3537         ret = btrfs_update_device(trans, device);
3538         if (ret) {
3539                 unlock_chunks(root);
3540                 btrfs_end_transaction(trans, root);
3541                 goto done;
3542         }
3543         WARN_ON(diff > old_total);
3544         btrfs_set_super_total_bytes(super_copy, old_total - diff);
3545         unlock_chunks(root);
3546         btrfs_end_transaction(trans, root);
3547 done:
3548         btrfs_free_path(path);
3549         return ret;
3550 }
3551
3552 static int btrfs_add_system_chunk(struct btrfs_root *root,
3553                            struct btrfs_key *key,
3554                            struct btrfs_chunk *chunk, int item_size)
3555 {
3556         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
3557         struct btrfs_disk_key disk_key;
3558         u32 array_size;
3559         u8 *ptr;
3560
3561         array_size = btrfs_super_sys_array_size(super_copy);
3562         if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
3563                 return -EFBIG;
3564
3565         ptr = super_copy->sys_chunk_array + array_size;
3566         btrfs_cpu_key_to_disk(&disk_key, key);
3567         memcpy(ptr, &disk_key, sizeof(disk_key));
3568         ptr += sizeof(disk_key);
3569         memcpy(ptr, chunk, item_size);
3570         item_size += sizeof(disk_key);
3571         btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
3572         return 0;
3573 }
3574
3575 /*
3576  * sort the devices in descending order by max_avail, total_avail
3577  */
3578 static int btrfs_cmp_device_info(const void *a, const void *b)
3579 {
3580         const struct btrfs_device_info *di_a = a;
3581         const struct btrfs_device_info *di_b = b;
3582
3583         if (di_a->max_avail > di_b->max_avail)
3584                 return -1;
3585         if (di_a->max_avail < di_b->max_avail)
3586                 return 1;
3587         if (di_a->total_avail > di_b->total_avail)
3588                 return -1;
3589         if (di_a->total_avail < di_b->total_avail)
3590                 return 1;
3591         return 0;
3592 }
3593
3594 struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = {
3595         [BTRFS_RAID_RAID10] = {
3596                 .sub_stripes    = 2,
3597                 .dev_stripes    = 1,
3598                 .devs_max       = 0,    /* 0 == as many as possible */
3599                 .devs_min       = 4,
3600                 .devs_increment = 2,
3601                 .ncopies        = 2,
3602         },
3603         [BTRFS_RAID_RAID1] = {
3604                 .sub_stripes    = 1,
3605                 .dev_stripes    = 1,
3606                 .devs_max       = 2,
3607                 .devs_min       = 2,
3608                 .devs_increment = 2,
3609                 .ncopies        = 2,
3610         },
3611         [BTRFS_RAID_DUP] = {
3612                 .sub_stripes    = 1,
3613                 .dev_stripes    = 2,
3614                 .devs_max       = 1,
3615                 .devs_min       = 1,
3616                 .devs_increment = 1,
3617                 .ncopies        = 2,
3618         },
3619         [BTRFS_RAID_RAID0] = {
3620                 .sub_stripes    = 1,
3621                 .dev_stripes    = 1,
3622                 .devs_max       = 0,
3623                 .devs_min       = 2,
3624                 .devs_increment = 1,
3625                 .ncopies        = 1,
3626         },
3627         [BTRFS_RAID_SINGLE] = {
3628                 .sub_stripes    = 1,
3629                 .dev_stripes    = 1,
3630                 .devs_max       = 1,
3631                 .devs_min       = 1,
3632                 .devs_increment = 1,
3633                 .ncopies        = 1,
3634         },
3635         [BTRFS_RAID_RAID5] = {
3636                 .sub_stripes    = 1,
3637                 .dev_stripes    = 1,
3638                 .devs_max       = 0,
3639                 .devs_min       = 2,
3640                 .devs_increment = 1,
3641                 .ncopies        = 2,
3642         },
3643         [BTRFS_RAID_RAID6] = {
3644                 .sub_stripes    = 1,
3645                 .dev_stripes    = 1,
3646                 .devs_max       = 0,
3647                 .devs_min       = 3,
3648                 .devs_increment = 1,
3649                 .ncopies        = 3,
3650         },
3651 };
3652
3653 static u32 find_raid56_stripe_len(u32 data_devices, u32 dev_stripe_target)
3654 {
3655         /* TODO allow them to set a preferred stripe size */
3656         return 64 * 1024;
3657 }
3658
3659 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
3660 {
3661         u64 features;
3662
3663         if (!(type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)))
3664                 return;
3665
3666         features = btrfs_super_incompat_flags(info->super_copy);
3667         if (features & BTRFS_FEATURE_INCOMPAT_RAID56)
3668                 return;
3669
3670         features |= BTRFS_FEATURE_INCOMPAT_RAID56;
3671         btrfs_set_super_incompat_flags(info->super_copy, features);
3672         printk(KERN_INFO "btrfs: setting RAID5/6 feature flag\n");
3673 }
3674
3675 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
3676                                struct btrfs_root *extent_root,
3677                                struct map_lookup **map_ret,
3678                                u64 *num_bytes_out, u64 *stripe_size_out,
3679                                u64 start, u64 type)
3680 {
3681         struct btrfs_fs_info *info = extent_root->fs_info;
3682         struct btrfs_fs_devices *fs_devices = info->fs_devices;
3683         struct list_head *cur;
3684         struct map_lookup *map = NULL;
3685         struct extent_map_tree *em_tree;
3686         struct extent_map *em;
3687         struct btrfs_device_info *devices_info = NULL;
3688         u64 total_avail;
3689         int num_stripes;        /* total number of stripes to allocate */
3690         int data_stripes;       /* number of stripes that count for
3691                                    block group size */
3692         int sub_stripes;        /* sub_stripes info for map */
3693         int dev_stripes;        /* stripes per dev */
3694         int devs_max;           /* max devs to use */
3695         int devs_min;           /* min devs needed */
3696         int devs_increment;     /* ndevs has to be a multiple of this */
3697         int ncopies;            /* how many copies to data has */
3698         int ret;
3699         u64 max_stripe_size;
3700         u64 max_chunk_size;
3701         u64 stripe_size;
3702         u64 num_bytes;
3703         u64 raid_stripe_len = BTRFS_STRIPE_LEN;
3704         int ndevs;
3705         int i;
3706         int j;
3707         int index;
3708
3709         BUG_ON(!alloc_profile_is_valid(type, 0));
3710
3711         if (list_empty(&fs_devices->alloc_list))
3712                 return -ENOSPC;
3713
3714         index = __get_raid_index(type);
3715
3716         sub_stripes = btrfs_raid_array[index].sub_stripes;
3717         dev_stripes = btrfs_raid_array[index].dev_stripes;
3718         devs_max = btrfs_raid_array[index].devs_max;
3719         devs_min = btrfs_raid_array[index].devs_min;
3720         devs_increment = btrfs_raid_array[index].devs_increment;
3721         ncopies = btrfs_raid_array[index].ncopies;
3722
3723         if (type & BTRFS_BLOCK_GROUP_DATA) {
3724                 max_stripe_size = 1024 * 1024 * 1024;
3725                 max_chunk_size = 10 * max_stripe_size;
3726         } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
3727                 /* for larger filesystems, use larger metadata chunks */
3728                 if (fs_devices->total_rw_bytes > 50ULL * 1024 * 1024 * 1024)
3729                         max_stripe_size = 1024 * 1024 * 1024;
3730                 else
3731                         max_stripe_size = 256 * 1024 * 1024;
3732                 max_chunk_size = max_stripe_size;
3733         } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
3734                 max_stripe_size = 32 * 1024 * 1024;
3735                 max_chunk_size = 2 * max_stripe_size;
3736         } else {
3737                 printk(KERN_ERR "btrfs: invalid chunk type 0x%llx requested\n",
3738                        type);
3739                 BUG_ON(1);
3740         }
3741
3742         /* we don't want a chunk larger than 10% of writeable space */
3743         max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
3744                              max_chunk_size);
3745
3746         devices_info = kzalloc(sizeof(*devices_info) * fs_devices->rw_devices,
3747                                GFP_NOFS);
3748         if (!devices_info)
3749                 return -ENOMEM;
3750
3751         cur = fs_devices->alloc_list.next;
3752
3753         /*
3754          * in the first pass through the devices list, we gather information
3755          * about the available holes on each device.
3756          */
3757         ndevs = 0;
3758         while (cur != &fs_devices->alloc_list) {
3759                 struct btrfs_device *device;
3760                 u64 max_avail;
3761                 u64 dev_offset;
3762
3763                 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
3764
3765                 cur = cur->next;
3766
3767                 if (!device->writeable) {
3768                         WARN(1, KERN_ERR
3769                                "btrfs: read-only device in alloc_list\n");
3770                         continue;
3771                 }
3772
3773                 if (!device->in_fs_metadata ||
3774                     device->is_tgtdev_for_dev_replace)
3775                         continue;
3776
3777                 if (device->total_bytes > device->bytes_used)
3778                         total_avail = device->total_bytes - device->bytes_used;
3779                 else
3780                         total_avail = 0;
3781
3782                 /* If there is no space on this device, skip it. */
3783                 if (total_avail == 0)
3784                         continue;
3785
3786                 ret = find_free_dev_extent(device,
3787                                            max_stripe_size * dev_stripes,
3788                                            &dev_offset, &max_avail);
3789                 if (ret && ret != -ENOSPC)
3790                         goto error;
3791
3792                 if (ret == 0)
3793                         max_avail = max_stripe_size * dev_stripes;
3794
3795                 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes)
3796                         continue;
3797
3798                 if (ndevs == fs_devices->rw_devices) {
3799                         WARN(1, "%s: found more than %llu devices\n",
3800                              __func__, fs_devices->rw_devices);
3801                         break;
3802                 }
3803                 devices_info[ndevs].dev_offset = dev_offset;
3804                 devices_info[ndevs].max_avail = max_avail;
3805                 devices_info[ndevs].total_avail = total_avail;
3806                 devices_info[ndevs].dev = device;
3807                 ++ndevs;
3808         }
3809
3810         /*
3811          * now sort the devices by hole size / available space
3812          */
3813         sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
3814              btrfs_cmp_device_info, NULL);
3815
3816         /* round down to number of usable stripes */
3817         ndevs -= ndevs % devs_increment;
3818
3819         if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) {
3820                 ret = -ENOSPC;
3821                 goto error;
3822         }
3823
3824         if (devs_max && ndevs > devs_max)
3825                 ndevs = devs_max;
3826         /*
3827          * the primary goal is to maximize the number of stripes, so use as many
3828          * devices as possible, even if the stripes are not maximum sized.
3829          */
3830         stripe_size = devices_info[ndevs-1].max_avail;
3831         num_stripes = ndevs * dev_stripes;
3832
3833         /*
3834          * this will have to be fixed for RAID1 and RAID10 over
3835          * more drives
3836          */
3837         data_stripes = num_stripes / ncopies;
3838
3839         if (type & BTRFS_BLOCK_GROUP_RAID5) {
3840                 raid_stripe_len = find_raid56_stripe_len(ndevs - 1,
3841                                  btrfs_super_stripesize(info->super_copy));
3842                 data_stripes = num_stripes - 1;
3843         }
3844         if (type & BTRFS_BLOCK_GROUP_RAID6) {
3845                 raid_stripe_len = find_raid56_stripe_len(ndevs - 2,
3846                                  btrfs_super_stripesize(info->super_copy));
3847                 data_stripes = num_stripes - 2;
3848         }
3849
3850         /*
3851          * Use the number of data stripes to figure out how big this chunk
3852          * is really going to be in terms of logical address space,
3853          * and compare that answer with the max chunk size
3854          */
3855         if (stripe_size * data_stripes > max_chunk_size) {
3856                 u64 mask = (1ULL << 24) - 1;
3857                 stripe_size = max_chunk_size;
3858                 do_div(stripe_size, data_stripes);
3859
3860                 /* bump the answer up to a 16MB boundary */
3861                 stripe_size = (stripe_size + mask) & ~mask;
3862
3863                 /* but don't go higher than the limits we found
3864                  * while searching for free extents
3865                  */
3866                 if (stripe_size > devices_info[ndevs-1].max_avail)
3867                         stripe_size = devices_info[ndevs-1].max_avail;
3868         }
3869
3870         do_div(stripe_size, dev_stripes);
3871
3872         /* align to BTRFS_STRIPE_LEN */
3873         do_div(stripe_size, raid_stripe_len);
3874         stripe_size *= raid_stripe_len;
3875
3876         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
3877         if (!map) {
3878                 ret = -ENOMEM;
3879                 goto error;
3880         }
3881         map->num_stripes = num_stripes;
3882
3883         for (i = 0; i < ndevs; ++i) {
3884                 for (j = 0; j < dev_stripes; ++j) {
3885                         int s = i * dev_stripes + j;
3886                         map->stripes[s].dev = devices_info[i].dev;
3887                         map->stripes[s].physical = devices_info[i].dev_offset +
3888                                                    j * stripe_size;
3889                 }
3890         }
3891         map->sector_size = extent_root->sectorsize;
3892         map->stripe_len = raid_stripe_len;
3893         map->io_align = raid_stripe_len;
3894         map->io_width = raid_stripe_len;
3895         map->type = type;
3896         map->sub_stripes = sub_stripes;
3897
3898         *map_ret = map;
3899         num_bytes = stripe_size * data_stripes;
3900
3901         *stripe_size_out = stripe_size;
3902         *num_bytes_out = num_bytes;
3903
3904         trace_btrfs_chunk_alloc(info->chunk_root, map, start, num_bytes);
3905
3906         em = alloc_extent_map();
3907         if (!em) {
3908                 ret = -ENOMEM;
3909                 goto error;
3910         }
3911         em->bdev = (struct block_device *)map;
3912         em->start = start;
3913         em->len = num_bytes;
3914         em->block_start = 0;
3915         em->block_len = em->len;
3916
3917         em_tree = &extent_root->fs_info->mapping_tree.map_tree;
3918         write_lock(&em_tree->lock);
3919         ret = add_extent_mapping(em_tree, em);
3920         write_unlock(&em_tree->lock);
3921         if (ret) {
3922                 free_extent_map(em);
3923                 goto error;
3924         }
3925
3926         for (i = 0; i < map->num_stripes; ++i) {
3927                 struct btrfs_device *device;
3928                 u64 dev_offset;
3929
3930                 device = map->stripes[i].dev;
3931                 dev_offset = map->stripes[i].physical;
3932
3933                 ret = btrfs_alloc_dev_extent(trans, device,
3934                                 info->chunk_root->root_key.objectid,
3935                                 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
3936                                 start, dev_offset, stripe_size);
3937                 if (ret)
3938                         goto error_dev_extent;
3939         }
3940
3941         ret = btrfs_make_block_group(trans, extent_root, 0, type,
3942                                      BTRFS_FIRST_CHUNK_TREE_OBJECTID,
3943                                      start, num_bytes);
3944         if (ret) {
3945                 i = map->num_stripes - 1;
3946                 goto error_dev_extent;
3947         }
3948
3949         free_extent_map(em);
3950         check_raid56_incompat_flag(extent_root->fs_info, type);
3951
3952         kfree(devices_info);
3953         return 0;
3954
3955 error_dev_extent:
3956         for (; i >= 0; i--) {
3957                 struct btrfs_device *device;
3958                 int err;
3959
3960                 device = map->stripes[i].dev;
3961                 err = btrfs_free_dev_extent(trans, device, start);
3962                 if (err) {
3963                         btrfs_abort_transaction(trans, extent_root, err);
3964                         break;
3965                 }
3966         }
3967         write_lock(&em_tree->lock);
3968         remove_extent_mapping(em_tree, em);
3969         write_unlock(&em_tree->lock);
3970
3971         /* One for our allocation */
3972         free_extent_map(em);
3973         /* One for the tree reference */
3974         free_extent_map(em);
3975 error:
3976         kfree(map);
3977         kfree(devices_info);
3978         return ret;
3979 }
3980
3981 static int __finish_chunk_alloc(struct btrfs_trans_handle *trans,
3982                                 struct btrfs_root *extent_root,
3983                                 struct map_lookup *map, u64 chunk_offset,
3984                                 u64 chunk_size, u64 stripe_size)
3985 {
3986         u64 dev_offset;
3987         struct btrfs_key key;
3988         struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
3989         struct btrfs_device *device;
3990         struct btrfs_chunk *chunk;
3991         struct btrfs_stripe *stripe;
3992         size_t item_size = btrfs_chunk_item_size(map->num_stripes);
3993         int index = 0;
3994         int ret;
3995
3996         chunk = kzalloc(item_size, GFP_NOFS);
3997         if (!chunk)
3998                 return -ENOMEM;
3999
4000         index = 0;
4001         while (index < map->num_stripes) {
4002                 device = map->stripes[index].dev;
4003                 device->bytes_used += stripe_size;
4004                 ret = btrfs_update_device(trans, device);
4005                 if (ret)
4006                         goto out_free;
4007                 index++;
4008         }
4009
4010         spin_lock(&extent_root->fs_info->free_chunk_lock);
4011         extent_root->fs_info->free_chunk_space -= (stripe_size *
4012                                                    map->num_stripes);
4013         spin_unlock(&extent_root->fs_info->free_chunk_lock);
4014
4015         index = 0;
4016         stripe = &chunk->stripe;
4017         while (index < map->num_stripes) {
4018                 device = map->stripes[index].dev;
4019                 dev_offset = map->stripes[index].physical;
4020
4021                 btrfs_set_stack_stripe_devid(stripe, device->devid);
4022                 btrfs_set_stack_stripe_offset(stripe, dev_offset);
4023                 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
4024                 stripe++;
4025                 index++;
4026         }
4027
4028         btrfs_set_stack_chunk_length(chunk, chunk_size);
4029         btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
4030         btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
4031         btrfs_set_stack_chunk_type(chunk, map->type);
4032         btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
4033         btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
4034         btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
4035         btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
4036         btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
4037
4038         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
4039         key.type = BTRFS_CHUNK_ITEM_KEY;
4040         key.offset = chunk_offset;
4041
4042         ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
4043
4044         if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
4045                 /*
4046                  * TODO: Cleanup of inserted chunk root in case of
4047                  * failure.
4048                  */
4049                 ret = btrfs_add_system_chunk(chunk_root, &key, chunk,
4050                                              item_size);
4051         }
4052
4053 out_free:
4054         kfree(chunk);
4055         return ret;
4056 }
4057
4058 /*
4059  * Chunk allocation falls into two parts. The first part does works
4060  * that make the new allocated chunk useable, but not do any operation
4061  * that modifies the chunk tree. The second part does the works that
4062  * require modifying the chunk tree. This division is important for the
4063  * bootstrap process of adding storage to a seed btrfs.
4064  */
4065 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4066                       struct btrfs_root *extent_root, u64 type)
4067 {
4068         u64 chunk_offset;
4069         u64 chunk_size;
4070         u64 stripe_size;
4071         struct map_lookup *map;
4072         struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
4073         int ret;
4074
4075         ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4076                               &chunk_offset);
4077         if (ret)
4078                 return ret;
4079
4080         ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
4081                                   &stripe_size, chunk_offset, type);
4082         if (ret)
4083                 return ret;
4084
4085         ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
4086                                    chunk_size, stripe_size);
4087         if (ret)
4088                 return ret;
4089         return 0;
4090 }
4091
4092 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
4093                                          struct btrfs_root *root,
4094                                          struct btrfs_device *device)
4095 {
4096         u64 chunk_offset;
4097         u64 sys_chunk_offset;
4098         u64 chunk_size;
4099         u64 sys_chunk_size;
4100         u64 stripe_size;
4101         u64 sys_stripe_size;
4102         u64 alloc_profile;
4103         struct map_lookup *map;
4104         struct map_lookup *sys_map;
4105         struct btrfs_fs_info *fs_info = root->fs_info;
4106         struct btrfs_root *extent_root = fs_info->extent_root;
4107         int ret;
4108
4109         ret = find_next_chunk(fs_info->chunk_root,
4110                               BTRFS_FIRST_CHUNK_TREE_OBJECTID, &chunk_offset);
4111         if (ret)
4112                 return ret;
4113
4114         alloc_profile = btrfs_get_alloc_profile(extent_root, 0);
4115         ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
4116                                   &stripe_size, chunk_offset, alloc_profile);
4117         if (ret)
4118                 return ret;
4119
4120         sys_chunk_offset = chunk_offset + chunk_size;
4121
4122         alloc_profile = btrfs_get_alloc_profile(fs_info->chunk_root, 0);
4123         ret = __btrfs_alloc_chunk(trans, extent_root, &sys_map,
4124                                   &sys_chunk_size, &sys_stripe_size,
4125                                   sys_chunk_offset, alloc_profile);
4126         if (ret) {
4127                 btrfs_abort_transaction(trans, root, ret);
4128                 goto out;
4129         }
4130
4131         ret = btrfs_add_device(trans, fs_info->chunk_root, device);
4132         if (ret) {
4133                 btrfs_abort_transaction(trans, root, ret);
4134                 goto out;
4135         }
4136
4137         /*
4138          * Modifying chunk tree needs allocating new blocks from both
4139          * system block group and metadata block group. So we only can
4140          * do operations require modifying the chunk tree after both
4141          * block groups were created.
4142          */
4143         ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
4144                                    chunk_size, stripe_size);
4145         if (ret) {
4146                 btrfs_abort_transaction(trans, root, ret);
4147                 goto out;
4148         }
4149
4150         ret = __finish_chunk_alloc(trans, extent_root, sys_map,
4151                                    sys_chunk_offset, sys_chunk_size,
4152                                    sys_stripe_size);
4153         if (ret)
4154                 btrfs_abort_transaction(trans, root, ret);
4155
4156 out:
4157
4158         return ret;
4159 }
4160
4161 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
4162 {
4163         struct extent_map *em;
4164         struct map_lookup *map;
4165         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
4166         int readonly = 0;
4167         int i;
4168
4169         read_lock(&map_tree->map_tree.lock);
4170         em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
4171         read_unlock(&map_tree->map_tree.lock);
4172         if (!em)
4173                 return 1;
4174
4175         if (btrfs_test_opt(root, DEGRADED)) {
4176                 free_extent_map(em);
4177                 return 0;
4178         }
4179
4180         map = (struct map_lookup *)em->bdev;
4181         for (i = 0; i < map->num_stripes; i++) {
4182                 if (!map->stripes[i].dev->writeable) {
4183                         readonly = 1;
4184                         break;
4185                 }
4186         }
4187         free_extent_map(em);
4188         return readonly;
4189 }
4190
4191 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
4192 {
4193         extent_map_tree_init(&tree->map_tree);
4194 }
4195
4196 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
4197 {
4198         struct extent_map *em;
4199
4200         while (1) {
4201                 write_lock(&tree->map_tree.lock);
4202                 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
4203                 if (em)
4204                         remove_extent_mapping(&tree->map_tree, em);
4205                 write_unlock(&tree->map_tree.lock);
4206                 if (!em)
4207                         break;
4208                 kfree(em->bdev);
4209                 /* once for us */
4210                 free_extent_map(em);
4211                 /* once for the tree */
4212                 free_extent_map(em);
4213         }
4214 }
4215
4216 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
4217 {
4218         struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
4219         struct extent_map *em;
4220         struct map_lookup *map;
4221         struct extent_map_tree *em_tree = &map_tree->map_tree;
4222         int ret;
4223
4224         read_lock(&em_tree->lock);
4225         em = lookup_extent_mapping(em_tree, logical, len);
4226         read_unlock(&em_tree->lock);
4227         BUG_ON(!em);
4228
4229         BUG_ON(em->start > logical || em->start + em->len < logical);
4230         map = (struct map_lookup *)em->bdev;
4231         if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
4232                 ret = map->num_stripes;
4233         else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
4234                 ret = map->sub_stripes;
4235         else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
4236                 ret = 2;
4237         else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
4238                 ret = 3;
4239         else
4240                 ret = 1;
4241         free_extent_map(em);
4242
4243         btrfs_dev_replace_lock(&fs_info->dev_replace);
4244         if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace))
4245                 ret++;
4246         btrfs_dev_replace_unlock(&fs_info->dev_replace);
4247
4248         return ret;
4249 }
4250
4251 unsigned long btrfs_full_stripe_len(struct btrfs_root *root,
4252                                     struct btrfs_mapping_tree *map_tree,
4253                                     u64 logical)
4254 {
4255         struct extent_map *em;
4256         struct map_lookup *map;
4257         struct extent_map_tree *em_tree = &map_tree->map_tree;
4258         unsigned long len = root->sectorsize;
4259
4260         read_lock(&em_tree->lock);
4261         em = lookup_extent_mapping(em_tree, logical, len);
4262         read_unlock(&em_tree->lock);
4263         BUG_ON(!em);
4264
4265         BUG_ON(em->start > logical || em->start + em->len < logical);
4266         map = (struct map_lookup *)em->bdev;
4267         if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
4268                          BTRFS_BLOCK_GROUP_RAID6)) {
4269                 len = map->stripe_len * nr_data_stripes(map);
4270         }
4271         free_extent_map(em);
4272         return len;
4273 }
4274
4275 int btrfs_is_parity_mirror(struct btrfs_mapping_tree *map_tree,
4276                            u64 logical, u64 len, int mirror_num)
4277 {
4278         struct extent_map *em;
4279         struct map_lookup *map;
4280         struct extent_map_tree *em_tree = &map_tree->map_tree;
4281         int ret = 0;
4282
4283         read_lock(&em_tree->lock);
4284         em = lookup_extent_mapping(em_tree, logical, len);
4285         read_unlock(&em_tree->lock);
4286         BUG_ON(!em);
4287
4288         BUG_ON(em->start > logical || em->start + em->len < logical);
4289         map = (struct map_lookup *)em->bdev;
4290         if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
4291                          BTRFS_BLOCK_GROUP_RAID6))
4292                 ret = 1;
4293         free_extent_map(em);
4294         return ret;
4295 }
4296
4297 static int find_live_mirror(struct btrfs_fs_info *fs_info,
4298                             struct map_lookup *map, int first, int num,
4299                             int optimal, int dev_replace_is_ongoing)
4300 {
4301         int i;
4302         int tolerance;
4303         struct btrfs_device *srcdev;
4304
4305         if (dev_replace_is_ongoing &&
4306             fs_info->dev_replace.cont_reading_from_srcdev_mode ==
4307              BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
4308                 srcdev = fs_info->dev_replace.srcdev;
4309         else
4310                 srcdev = NULL;
4311
4312         /*
4313          * try to avoid the drive that is the source drive for a
4314          * dev-replace procedure, only choose it if no other non-missing
4315          * mirror is available
4316          */
4317         for (tolerance = 0; tolerance < 2; tolerance++) {
4318                 if (map->stripes[optimal].dev->bdev &&
4319                     (tolerance || map->stripes[optimal].dev != srcdev))
4320                         return optimal;
4321                 for (i = first; i < first + num; i++) {
4322                         if (map->stripes[i].dev->bdev &&
4323                             (tolerance || map->stripes[i].dev != srcdev))
4324                                 return i;
4325                 }
4326         }
4327
4328         /* we couldn't find one that doesn't fail.  Just return something
4329          * and the io error handling code will clean up eventually
4330          */
4331         return optimal;
4332 }
4333
4334 static inline int parity_smaller(u64 a, u64 b)
4335 {
4336         return a > b;
4337 }
4338
4339 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
4340 static void sort_parity_stripes(struct btrfs_bio *bbio, u64 *raid_map)
4341 {
4342         struct btrfs_bio_stripe s;
4343         int i;
4344         u64 l;
4345         int again = 1;
4346
4347         while (again) {
4348                 again = 0;
4349                 for (i = 0; i < bbio->num_stripes - 1; i++) {
4350                         if (parity_smaller(raid_map[i], raid_map[i+1])) {
4351                                 s = bbio->stripes[i];
4352                                 l = raid_map[i];
4353                                 bbio->stripes[i] = bbio->stripes[i+1];
4354                                 raid_map[i] = raid_map[i+1];
4355                                 bbio->stripes[i+1] = s;
4356                                 raid_map[i+1] = l;
4357                                 again = 1;
4358                         }
4359                 }
4360         }
4361 }
4362
4363 static int __btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
4364                              u64 logical, u64 *length,
4365                              struct btrfs_bio **bbio_ret,
4366                              int mirror_num, u64 **raid_map_ret)
4367 {
4368         struct extent_map *em;
4369         struct map_lookup *map;
4370         struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
4371         struct extent_map_tree *em_tree = &map_tree->map_tree;
4372         u64 offset;
4373         u64 stripe_offset;
4374         u64 stripe_end_offset;
4375         u64 stripe_nr;
4376         u64 stripe_nr_orig;
4377         u64 stripe_nr_end;
4378         u64 stripe_len;
4379         u64 *raid_map = NULL;
4380         int stripe_index;
4381         int i;
4382         int ret = 0;
4383         int num_stripes;
4384         int max_errors = 0;
4385         struct btrfs_bio *bbio = NULL;
4386         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
4387         int dev_replace_is_ongoing = 0;
4388         int num_alloc_stripes;
4389         int patch_the_first_stripe_for_dev_replace = 0;
4390         u64 physical_to_patch_in_first_stripe = 0;
4391         u64 raid56_full_stripe_start = (u64)-1;
4392
4393         read_lock(&em_tree->lock);
4394         em = lookup_extent_mapping(em_tree, logical, *length);
4395         read_unlock(&em_tree->lock);
4396
4397         if (!em) {
4398                 printk(KERN_CRIT "btrfs: unable to find logical %llu len %llu\n",
4399                        (unsigned long long)logical,
4400                        (unsigned long long)*length);
4401                 BUG();
4402         }
4403
4404         BUG_ON(em->start > logical || em->start + em->len < logical);
4405         map = (struct map_lookup *)em->bdev;
4406         offset = logical - em->start;
4407
4408         if (mirror_num > map->num_stripes)
4409                 mirror_num = 0;
4410
4411         stripe_len = map->stripe_len;
4412         stripe_nr = offset;
4413         /*
4414          * stripe_nr counts the total number of stripes we have to stride
4415          * to get to this block
4416          */
4417         do_div(stripe_nr, stripe_len);
4418
4419         stripe_offset = stripe_nr * stripe_len;
4420         BUG_ON(offset < stripe_offset);
4421
4422         /* stripe_offset is the offset of this block in its stripe*/
4423         stripe_offset = offset - stripe_offset;
4424
4425         /* if we're here for raid56, we need to know the stripe aligned start */
4426         if (map->type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)) {
4427                 unsigned long full_stripe_len = stripe_len * nr_data_stripes(map);
4428                 raid56_full_stripe_start = offset;
4429
4430                 /* allow a write of a full stripe, but make sure we don't
4431                  * allow straddling of stripes
4432                  */
4433                 do_div(raid56_full_stripe_start, full_stripe_len);
4434                 raid56_full_stripe_start *= full_stripe_len;
4435         }
4436
4437         if (rw & REQ_DISCARD) {
4438                 /* we don't discard raid56 yet */
4439                 if (map->type &
4440                     (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)) {
4441                         ret = -EOPNOTSUPP;
4442                         goto out;
4443                 }
4444                 *length = min_t(u64, em->len - offset, *length);
4445         } else if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
4446                 u64 max_len;
4447                 /* For writes to RAID[56], allow a full stripeset across all disks.
4448                    For other RAID types and for RAID[56] reads, just allow a single
4449                    stripe (on a single disk). */
4450                 if (map->type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6) &&
4451                     (rw & REQ_WRITE)) {
4452                         max_len = stripe_len * nr_data_stripes(map) -
4453                                 (offset - raid56_full_stripe_start);
4454                 } else {
4455                         /* we limit the length of each bio to what fits in a stripe */
4456                         max_len = stripe_len - stripe_offset;
4457                 }
4458                 *length = min_t(u64, em->len - offset, max_len);
4459         } else {
4460                 *length = em->len - offset;
4461         }
4462
4463         /* This is for when we're called from btrfs_merge_bio_hook() and all
4464            it cares about is the length */
4465         if (!bbio_ret)
4466                 goto out;
4467
4468         btrfs_dev_replace_lock(dev_replace);
4469         dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
4470         if (!dev_replace_is_ongoing)
4471                 btrfs_dev_replace_unlock(dev_replace);
4472
4473         if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
4474             !(rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) &&
4475             dev_replace->tgtdev != NULL) {
4476                 /*
4477                  * in dev-replace case, for repair case (that's the only
4478                  * case where the mirror is selected explicitly when
4479                  * calling btrfs_map_block), blocks left of the left cursor
4480                  * can also be read from the target drive.
4481                  * For REQ_GET_READ_MIRRORS, the target drive is added as
4482                  * the last one to the array of stripes. For READ, it also
4483                  * needs to be supported using the same mirror number.
4484                  * If the requested block is not left of the left cursor,
4485                  * EIO is returned. This can happen because btrfs_num_copies()
4486                  * returns one more in the dev-replace case.
4487                  */
4488                 u64 tmp_length = *length;
4489                 struct btrfs_bio *tmp_bbio = NULL;
4490                 int tmp_num_stripes;
4491                 u64 srcdev_devid = dev_replace->srcdev->devid;
4492                 int index_srcdev = 0;
4493                 int found = 0;
4494                 u64 physical_of_found = 0;
4495
4496                 ret = __btrfs_map_block(fs_info, REQ_GET_READ_MIRRORS,
4497                              logical, &tmp_length, &tmp_bbio, 0, NULL);
4498                 if (ret) {
4499                         WARN_ON(tmp_bbio != NULL);
4500                         goto out;
4501                 }
4502
4503                 tmp_num_stripes = tmp_bbio->num_stripes;
4504                 if (mirror_num > tmp_num_stripes) {
4505                         /*
4506                          * REQ_GET_READ_MIRRORS does not contain this
4507                          * mirror, that means that the requested area
4508                          * is not left of the left cursor
4509                          */
4510                         ret = -EIO;
4511                         kfree(tmp_bbio);
4512                         goto out;
4513                 }
4514
4515                 /*
4516                  * process the rest of the function using the mirror_num
4517                  * of the source drive. Therefore look it up first.
4518                  * At the end, patch the device pointer to the one of the
4519                  * target drive.
4520                  */
4521                 for (i = 0; i < tmp_num_stripes; i++) {
4522                         if (tmp_bbio->stripes[i].dev->devid == srcdev_devid) {
4523                                 /*
4524                                  * In case of DUP, in order to keep it
4525                                  * simple, only add the mirror with the
4526                                  * lowest physical address
4527                                  */
4528                                 if (found &&
4529                                     physical_of_found <=
4530                                      tmp_bbio->stripes[i].physical)
4531                                         continue;
4532                                 index_srcdev = i;
4533                                 found = 1;
4534                                 physical_of_found =
4535                                         tmp_bbio->stripes[i].physical;
4536                         }
4537                 }
4538
4539                 if (found) {
4540                         mirror_num = index_srcdev + 1;
4541                         patch_the_first_stripe_for_dev_replace = 1;
4542                         physical_to_patch_in_first_stripe = physical_of_found;
4543                 } else {
4544                         WARN_ON(1);
4545                         ret = -EIO;
4546                         kfree(tmp_bbio);
4547                         goto out;
4548                 }
4549
4550                 kfree(tmp_bbio);
4551         } else if (mirror_num > map->num_stripes) {
4552                 mirror_num = 0;
4553         }
4554
4555         num_stripes = 1;
4556         stripe_index = 0;
4557         stripe_nr_orig = stripe_nr;
4558         stripe_nr_end = ALIGN(offset + *length, map->stripe_len);
4559         do_div(stripe_nr_end, map->stripe_len);
4560         stripe_end_offset = stripe_nr_end * map->stripe_len -
4561                             (offset + *length);
4562
4563         if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
4564                 if (rw & REQ_DISCARD)
4565                         num_stripes = min_t(u64, map->num_stripes,
4566                                             stripe_nr_end - stripe_nr_orig);
4567                 stripe_index = do_div(stripe_nr, map->num_stripes);
4568         } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
4569                 if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS))
4570                         num_stripes = map->num_stripes;
4571                 else if (mirror_num)
4572                         stripe_index = mirror_num - 1;
4573                 else {
4574                         stripe_index = find_live_mirror(fs_info, map, 0,
4575                                             map->num_stripes,
4576                                             current->pid % map->num_stripes,
4577                                             dev_replace_is_ongoing);
4578                         mirror_num = stripe_index + 1;
4579                 }
4580
4581         } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
4582                 if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) {
4583                         num_stripes = map->num_stripes;
4584                 } else if (mirror_num) {
4585                         stripe_index = mirror_num - 1;
4586                 } else {
4587                         mirror_num = 1;
4588                 }
4589
4590         } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
4591                 int factor = map->num_stripes / map->sub_stripes;
4592
4593                 stripe_index = do_div(stripe_nr, factor);
4594                 stripe_index *= map->sub_stripes;
4595
4596                 if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS))
4597                         num_stripes = map->sub_stripes;
4598                 else if (rw & REQ_DISCARD)
4599                         num_stripes = min_t(u64, map->sub_stripes *
4600                                             (stripe_nr_end - stripe_nr_orig),
4601                                             map->num_stripes);
4602                 else if (mirror_num)
4603                         stripe_index += mirror_num - 1;
4604                 else {
4605                         int old_stripe_index = stripe_index;
4606                         stripe_index = find_live_mirror(fs_info, map,
4607                                               stripe_index,
4608                                               map->sub_stripes, stripe_index +
4609                                               current->pid % map->sub_stripes,
4610                                               dev_replace_is_ongoing);
4611                         mirror_num = stripe_index - old_stripe_index + 1;
4612                 }
4613
4614         } else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
4615                                 BTRFS_BLOCK_GROUP_RAID6)) {
4616                 u64 tmp;
4617
4618                 if (bbio_ret && ((rw & REQ_WRITE) || mirror_num > 1)
4619                     && raid_map_ret) {
4620                         int i, rot;
4621
4622                         /* push stripe_nr back to the start of the full stripe */
4623                         stripe_nr = raid56_full_stripe_start;
4624                         do_div(stripe_nr, stripe_len);
4625
4626                         stripe_index = do_div(stripe_nr, nr_data_stripes(map));
4627
4628                         /* RAID[56] write or recovery. Return all stripes */
4629                         num_stripes = map->num_stripes;
4630                         max_errors = nr_parity_stripes(map);
4631
4632                         raid_map = kmalloc(sizeof(u64) * num_stripes,
4633                                            GFP_NOFS);
4634                         if (!raid_map) {
4635                                 ret = -ENOMEM;
4636                                 goto out;
4637                         }
4638
4639                         /* Work out the disk rotation on this stripe-set */
4640                         tmp = stripe_nr;
4641                         rot = do_div(tmp, num_stripes);
4642
4643                         /* Fill in the logical address of each stripe */
4644                         tmp = stripe_nr * nr_data_stripes(map);
4645                         for (i = 0; i < nr_data_stripes(map); i++)
4646                                 raid_map[(i+rot) % num_stripes] =
4647                                         em->start + (tmp + i) * map->stripe_len;
4648
4649                         raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE;
4650                         if (map->type & BTRFS_BLOCK_GROUP_RAID6)
4651                                 raid_map[(i+rot+1) % num_stripes] =
4652                                         RAID6_Q_STRIPE;
4653
4654                         *length = map->stripe_len;
4655                         stripe_index = 0;
4656                         stripe_offset = 0;
4657                 } else {
4658                         /*
4659                          * Mirror #0 or #1 means the original data block.
4660                          * Mirror #2 is RAID5 parity block.
4661                          * Mirror #3 is RAID6 Q block.
4662                          */
4663                         stripe_index = do_div(stripe_nr, nr_data_stripes(map));
4664                         if (mirror_num > 1)
4665                                 stripe_index = nr_data_stripes(map) +
4666                                                 mirror_num - 2;
4667
4668                         /* We distribute the parity blocks across stripes */
4669                         tmp = stripe_nr + stripe_index;
4670                         stripe_index = do_div(tmp, map->num_stripes);
4671                 }
4672         } else {
4673                 /*
4674                  * after this do_div call, stripe_nr is the number of stripes
4675                  * on this device we have to walk to find the data, and
4676                  * stripe_index is the number of our device in the stripe array
4677                  */
4678                 stripe_index = do_div(stripe_nr, map->num_stripes);
4679                 mirror_num = stripe_index + 1;
4680         }
4681         BUG_ON(stripe_index >= map->num_stripes);
4682
4683         num_alloc_stripes = num_stripes;
4684         if (dev_replace_is_ongoing) {
4685                 if (rw & (REQ_WRITE | REQ_DISCARD))
4686                         num_alloc_stripes <<= 1;
4687                 if (rw & REQ_GET_READ_MIRRORS)
4688                         num_alloc_stripes++;
4689         }
4690         bbio = kzalloc(btrfs_bio_size(num_alloc_stripes), GFP_NOFS);
4691         if (!bbio) {
4692                 ret = -ENOMEM;
4693                 goto out;
4694         }
4695         atomic_set(&bbio->error, 0);
4696
4697         if (rw & REQ_DISCARD) {
4698                 int factor = 0;
4699                 int sub_stripes = 0;
4700                 u64 stripes_per_dev = 0;
4701                 u32 remaining_stripes = 0;
4702                 u32 last_stripe = 0;
4703
4704                 if (map->type &
4705                     (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID10)) {
4706                         if (map->type & BTRFS_BLOCK_GROUP_RAID0)
4707                                 sub_stripes = 1;
4708                         else
4709                                 sub_stripes = map->sub_stripes;
4710
4711                         factor = map->num_stripes / sub_stripes;
4712                         stripes_per_dev = div_u64_rem(stripe_nr_end -
4713                                                       stripe_nr_orig,
4714                                                       factor,
4715                                                       &remaining_stripes);
4716                         div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
4717                         last_stripe *= sub_stripes;
4718                 }
4719
4720                 for (i = 0; i < num_stripes; i++) {
4721                         bbio->stripes[i].physical =
4722                                 map->stripes[stripe_index].physical +
4723                                 stripe_offset + stripe_nr * map->stripe_len;
4724                         bbio->stripes[i].dev = map->stripes[stripe_index].dev;
4725
4726                         if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
4727                                          BTRFS_BLOCK_GROUP_RAID10)) {
4728                                 bbio->stripes[i].length = stripes_per_dev *
4729                                                           map->stripe_len;
4730
4731                                 if (i / sub_stripes < remaining_stripes)
4732                                         bbio->stripes[i].length +=
4733                                                 map->stripe_len;
4734
4735                                 /*
4736                                  * Special for the first stripe and
4737                                  * the last stripe:
4738                                  *
4739                                  * |-------|...|-------|
4740                                  *     |----------|
4741                                  *    off     end_off
4742                                  */
4743                                 if (i < sub_stripes)
4744                                         bbio->stripes[i].length -=
4745                                                 stripe_offset;
4746
4747                                 if (stripe_index >= last_stripe &&
4748                                     stripe_index <= (last_stripe +
4749                                                      sub_stripes - 1))
4750                                         bbio->stripes[i].length -=
4751                                                 stripe_end_offset;
4752
4753                                 if (i == sub_stripes - 1)
4754                                         stripe_offset = 0;
4755                         } else
4756                                 bbio->stripes[i].length = *length;
4757
4758                         stripe_index++;
4759                         if (stripe_index == map->num_stripes) {
4760                                 /* This could only happen for RAID0/10 */
4761                                 stripe_index = 0;
4762                                 stripe_nr++;
4763                         }
4764                 }
4765         } else {
4766                 for (i = 0; i < num_stripes; i++) {
4767                         bbio->stripes[i].physical =
4768                                 map->stripes[stripe_index].physical +
4769                                 stripe_offset +
4770                                 stripe_nr * map->stripe_len;
4771                         bbio->stripes[i].dev =
4772                                 map->stripes[stripe_index].dev;
4773                         stripe_index++;
4774                 }
4775         }
4776
4777         if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS)) {
4778                 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
4779                                  BTRFS_BLOCK_GROUP_RAID10 |
4780                                  BTRFS_BLOCK_GROUP_RAID5 |
4781                                  BTRFS_BLOCK_GROUP_DUP)) {
4782                         max_errors = 1;
4783                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID6) {
4784                         max_errors = 2;
4785                 }
4786         }
4787
4788         if (dev_replace_is_ongoing && (rw & (REQ_WRITE | REQ_DISCARD)) &&
4789             dev_replace->tgtdev != NULL) {
4790                 int index_where_to_add;
4791                 u64 srcdev_devid = dev_replace->srcdev->devid;
4792
4793                 /*
4794                  * duplicate the write operations while the dev replace
4795                  * procedure is running. Since the copying of the old disk
4796                  * to the new disk takes place at run time while the
4797                  * filesystem is mounted writable, the regular write
4798                  * operations to the old disk have to be duplicated to go
4799                  * to the new disk as well.
4800                  * Note that device->missing is handled by the caller, and
4801                  * that the write to the old disk is already set up in the
4802                  * stripes array.
4803                  */
4804                 index_where_to_add = num_stripes;
4805                 for (i = 0; i < num_stripes; i++) {
4806                         if (bbio->stripes[i].dev->devid == srcdev_devid) {
4807                                 /* write to new disk, too */
4808                                 struct btrfs_bio_stripe *new =
4809                                         bbio->stripes + index_where_to_add;
4810                                 struct btrfs_bio_stripe *old =
4811                                         bbio->stripes + i;
4812
4813                                 new->physical = old->physical;
4814                                 new->length = old->length;
4815                                 new->dev = dev_replace->tgtdev;
4816                                 index_where_to_add++;
4817                                 max_errors++;
4818                         }
4819                 }
4820                 num_stripes = index_where_to_add;
4821         } else if (dev_replace_is_ongoing && (rw & REQ_GET_READ_MIRRORS) &&
4822                    dev_replace->tgtdev != NULL) {
4823                 u64 srcdev_devid = dev_replace->srcdev->devid;
4824                 int index_srcdev = 0;
4825                 int found = 0;
4826                 u64 physical_of_found = 0;
4827
4828                 /*
4829                  * During the dev-replace procedure, the target drive can
4830                  * also be used to read data in case it is needed to repair
4831                  * a corrupt block elsewhere. This is possible if the
4832                  * requested area is left of the left cursor. In this area,
4833                  * the target drive is a full copy of the source drive.
4834                  */
4835                 for (i = 0; i < num_stripes; i++) {
4836                         if (bbio->stripes[i].dev->devid == srcdev_devid) {
4837                                 /*
4838                                  * In case of DUP, in order to keep it
4839                                  * simple, only add the mirror with the
4840                                  * lowest physical address
4841                                  */
4842                                 if (found &&
4843                                     physical_of_found <=
4844                                      bbio->stripes[i].physical)
4845                                         continue;
4846                                 index_srcdev = i;
4847                                 found = 1;
4848                                 physical_of_found = bbio->stripes[i].physical;
4849                         }
4850                 }
4851                 if (found) {
4852                         u64 length = map->stripe_len;
4853
4854                         if (physical_of_found + length <=
4855                             dev_replace->cursor_left) {
4856                                 struct btrfs_bio_stripe *tgtdev_stripe =
4857                                         bbio->stripes + num_stripes;
4858
4859                                 tgtdev_stripe->physical = physical_of_found;
4860                                 tgtdev_stripe->length =
4861                                         bbio->stripes[index_srcdev].length;
4862                                 tgtdev_stripe->dev = dev_replace->tgtdev;
4863
4864                                 num_stripes++;
4865                         }
4866                 }
4867         }
4868
4869         *bbio_ret = bbio;
4870         bbio->num_stripes = num_stripes;
4871         bbio->max_errors = max_errors;
4872         bbio->mirror_num = mirror_num;
4873
4874         /*
4875          * this is the case that REQ_READ && dev_replace_is_ongoing &&
4876          * mirror_num == num_stripes + 1 && dev_replace target drive is
4877          * available as a mirror
4878          */
4879         if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) {
4880                 WARN_ON(num_stripes > 1);
4881                 bbio->stripes[0].dev = dev_replace->tgtdev;
4882                 bbio->stripes[0].physical = physical_to_patch_in_first_stripe;
4883                 bbio->mirror_num = map->num_stripes + 1;
4884         }
4885         if (raid_map) {
4886                 sort_parity_stripes(bbio, raid_map);
4887                 *raid_map_ret = raid_map;
4888         }
4889 out:
4890         if (dev_replace_is_ongoing)
4891                 btrfs_dev_replace_unlock(dev_replace);
4892         free_extent_map(em);
4893         return ret;
4894 }
4895
4896 int btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
4897                       u64 logical, u64 *length,
4898                       struct btrfs_bio **bbio_ret, int mirror_num)
4899 {
4900         return __btrfs_map_block(fs_info, rw, logical, length, bbio_ret,
4901                                  mirror_num, NULL);
4902 }
4903
4904 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
4905                      u64 chunk_start, u64 physical, u64 devid,
4906                      u64 **logical, int *naddrs, int *stripe_len)
4907 {
4908         struct extent_map_tree *em_tree = &map_tree->map_tree;
4909         struct extent_map *em;
4910         struct map_lookup *map;
4911         u64 *buf;
4912         u64 bytenr;
4913         u64 length;
4914         u64 stripe_nr;
4915         u64 rmap_len;
4916         int i, j, nr = 0;
4917
4918         read_lock(&em_tree->lock);
4919         em = lookup_extent_mapping(em_tree, chunk_start, 1);
4920         read_unlock(&em_tree->lock);
4921
4922         BUG_ON(!em || em->start != chunk_start);
4923         map = (struct map_lookup *)em->bdev;
4924
4925         length = em->len;
4926         rmap_len = map->stripe_len;
4927
4928         if (map->type & BTRFS_BLOCK_GROUP_RAID10)
4929                 do_div(length, map->num_stripes / map->sub_stripes);
4930         else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
4931                 do_div(length, map->num_stripes);
4932         else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
4933                               BTRFS_BLOCK_GROUP_RAID6)) {
4934                 do_div(length, nr_data_stripes(map));
4935                 rmap_len = map->stripe_len * nr_data_stripes(map);
4936         }
4937
4938         buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
4939         BUG_ON(!buf); /* -ENOMEM */
4940
4941         for (i = 0; i < map->num_stripes; i++) {
4942                 if (devid && map->stripes[i].dev->devid != devid)
4943                         continue;
4944                 if (map->stripes[i].physical > physical ||
4945                     map->stripes[i].physical + length <= physical)
4946                         continue;
4947
4948                 stripe_nr = physical - map->stripes[i].physical;
4949                 do_div(stripe_nr, map->stripe_len);
4950
4951                 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
4952                         stripe_nr = stripe_nr * map->num_stripes + i;
4953                         do_div(stripe_nr, map->sub_stripes);
4954                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
4955                         stripe_nr = stripe_nr * map->num_stripes + i;
4956                 } /* else if RAID[56], multiply by nr_data_stripes().
4957                    * Alternatively, just use rmap_len below instead of
4958                    * map->stripe_len */
4959
4960                 bytenr = chunk_start + stripe_nr * rmap_len;
4961                 WARN_ON(nr >= map->num_stripes);
4962                 for (j = 0; j < nr; j++) {
4963                         if (buf[j] == bytenr)
4964                                 break;
4965                 }
4966                 if (j == nr) {
4967                         WARN_ON(nr >= map->num_stripes);
4968                         buf[nr++] = bytenr;
4969                 }
4970         }
4971
4972         *logical = buf;
4973         *naddrs = nr;
4974         *stripe_len = rmap_len;
4975
4976         free_extent_map(em);
4977         return 0;
4978 }
4979
4980 static void *merge_stripe_index_into_bio_private(void *bi_private,
4981                                                  unsigned int stripe_index)
4982 {
4983         /*
4984          * with single, dup, RAID0, RAID1 and RAID10, stripe_index is
4985          * at most 1.
4986          * The alternative solution (instead of stealing bits from the
4987          * pointer) would be to allocate an intermediate structure
4988          * that contains the old private pointer plus the stripe_index.
4989          */
4990         BUG_ON((((uintptr_t)bi_private) & 3) != 0);
4991         BUG_ON(stripe_index > 3);
4992         return (void *)(((uintptr_t)bi_private) | stripe_index);
4993 }
4994
4995 static struct btrfs_bio *extract_bbio_from_bio_private(void *bi_private)
4996 {
4997         return (struct btrfs_bio *)(((uintptr_t)bi_private) & ~((uintptr_t)3));
4998 }
4999
5000 static unsigned int extract_stripe_index_from_bio_private(void *bi_private)
5001 {
5002         return (unsigned int)((uintptr_t)bi_private) & 3;
5003 }
5004
5005 static void btrfs_end_bio(struct bio *bio, int err)
5006 {
5007         struct btrfs_bio *bbio = extract_bbio_from_bio_private(bio->bi_private);
5008         int is_orig_bio = 0;
5009
5010         if (err) {
5011                 atomic_inc(&bbio->error);
5012                 if (err == -EIO || err == -EREMOTEIO) {
5013                         unsigned int stripe_index =
5014                                 extract_stripe_index_from_bio_private(
5015                                         bio->bi_private);
5016                         struct btrfs_device *dev;
5017
5018                         BUG_ON(stripe_index >= bbio->num_stripes);
5019                         dev = bbio->stripes[stripe_index].dev;
5020                         if (dev->bdev) {
5021                                 if (bio->bi_rw & WRITE)
5022                                         btrfs_dev_stat_inc(dev,
5023                                                 BTRFS_DEV_STAT_WRITE_ERRS);
5024                                 else
5025                                         btrfs_dev_stat_inc(dev,
5026                                                 BTRFS_DEV_STAT_READ_ERRS);
5027                                 if ((bio->bi_rw & WRITE_FLUSH) == WRITE_FLUSH)
5028                                         btrfs_dev_stat_inc(dev,
5029                                                 BTRFS_DEV_STAT_FLUSH_ERRS);
5030                                 btrfs_dev_stat_print_on_error(dev);
5031                         }
5032                 }
5033         }
5034
5035         if (bio == bbio->orig_bio)
5036                 is_orig_bio = 1;
5037
5038         if (atomic_dec_and_test(&bbio->stripes_pending)) {
5039                 if (!is_orig_bio) {
5040                         bio_put(bio);
5041                         bio = bbio->orig_bio;
5042                 }
5043                 bio->bi_private = bbio->private;
5044                 bio->bi_end_io = bbio->end_io;
5045                 bio->bi_bdev = (struct block_device *)
5046                                         (unsigned long)bbio->mirror_num;
5047                 /* only send an error to the higher layers if it is
5048                  * beyond the tolerance of the btrfs bio
5049                  */
5050                 if (atomic_read(&bbio->error) > bbio->max_errors) {
5051                         err = -EIO;
5052                 } else {
5053                         /*
5054                          * this bio is actually up to date, we didn't
5055                          * go over the max number of errors
5056                          */
5057                         set_bit(BIO_UPTODATE, &bio->bi_flags);
5058                         err = 0;
5059                 }
5060                 kfree(bbio);
5061
5062                 bio_endio(bio, err);
5063         } else if (!is_orig_bio) {
5064                 bio_put(bio);
5065         }
5066 }
5067
5068 struct async_sched {
5069         struct bio *bio;
5070         int rw;
5071         struct btrfs_fs_info *info;
5072         struct btrfs_work work;
5073 };
5074
5075 /*
5076  * see run_scheduled_bios for a description of why bios are collected for
5077  * async submit.
5078  *
5079  * This will add one bio to the pending list for a device and make sure
5080  * the work struct is scheduled.
5081  */
5082 noinline void btrfs_schedule_bio(struct btrfs_root *root,
5083                                  struct btrfs_device *device,
5084                                  int rw, struct bio *bio)
5085 {
5086         int should_queue = 1;
5087         struct btrfs_pending_bios *pending_bios;
5088
5089         if (device->missing || !device->bdev) {
5090                 bio_endio(bio, -EIO);
5091                 return;
5092         }
5093
5094         /* don't bother with additional async steps for reads, right now */
5095         if (!(rw & REQ_WRITE)) {
5096                 bio_get(bio);
5097                 btrfsic_submit_bio(rw, bio);
5098                 bio_put(bio);
5099                 return;
5100         }
5101
5102         /*
5103          * nr_async_bios allows us to reliably return congestion to the
5104          * higher layers.  Otherwise, the async bio makes it appear we have
5105          * made progress against dirty pages when we've really just put it
5106          * on a queue for later
5107          */
5108         atomic_inc(&root->fs_info->nr_async_bios);
5109         WARN_ON(bio->bi_next);
5110         bio->bi_next = NULL;
5111         bio->bi_rw |= rw;
5112
5113         spin_lock(&device->io_lock);
5114         if (bio->bi_rw & REQ_SYNC)
5115                 pending_bios = &device->pending_sync_bios;
5116         else
5117                 pending_bios = &device->pending_bios;
5118
5119         if (pending_bios->tail)
5120                 pending_bios->tail->bi_next = bio;
5121
5122         pending_bios->tail = bio;
5123         if (!pending_bios->head)
5124                 pending_bios->head = bio;
5125         if (device->running_pending)
5126                 should_queue = 0;
5127
5128         spin_unlock(&device->io_lock);
5129
5130         if (should_queue)
5131                 btrfs_queue_worker(&root->fs_info->submit_workers,
5132                                    &device->work);
5133 }
5134
5135 static int bio_size_ok(struct block_device *bdev, struct bio *bio,
5136                        sector_t sector)
5137 {
5138         struct bio_vec *prev;
5139         struct request_queue *q = bdev_get_queue(bdev);
5140         unsigned short max_sectors = queue_max_sectors(q);
5141         struct bvec_merge_data bvm = {
5142                 .bi_bdev = bdev,
5143                 .bi_sector = sector,
5144                 .bi_rw = bio->bi_rw,
5145         };
5146
5147         if (bio->bi_vcnt == 0) {
5148                 WARN_ON(1);
5149                 return 1;
5150         }
5151
5152         prev = &bio->bi_io_vec[bio->bi_vcnt - 1];
5153         if ((bio->bi_size >> 9) > max_sectors)
5154                 return 0;
5155
5156         if (!q->merge_bvec_fn)
5157                 return 1;
5158
5159         bvm.bi_size = bio->bi_size - prev->bv_len;
5160         if (q->merge_bvec_fn(q, &bvm, prev) < prev->bv_len)
5161                 return 0;
5162         return 1;
5163 }
5164
5165 static void submit_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio,
5166                               struct bio *bio, u64 physical, int dev_nr,
5167                               int rw, int async)
5168 {
5169         struct btrfs_device *dev = bbio->stripes[dev_nr].dev;
5170
5171         bio->bi_private = bbio;
5172         bio->bi_private = merge_stripe_index_into_bio_private(
5173                         bio->bi_private, (unsigned int)dev_nr);
5174         bio->bi_end_io = btrfs_end_bio;
5175         bio->bi_sector = physical >> 9;
5176 #ifdef DEBUG
5177         {
5178                 struct rcu_string *name;
5179
5180                 rcu_read_lock();
5181                 name = rcu_dereference(dev->name);
5182                 pr_debug("btrfs_map_bio: rw %d, sector=%llu, dev=%lu "
5183                          "(%s id %llu), size=%u\n", rw,
5184                          (u64)bio->bi_sector, (u_long)dev->bdev->bd_dev,
5185                          name->str, dev->devid, bio->bi_size);
5186                 rcu_read_unlock();
5187         }
5188 #endif
5189         bio->bi_bdev = dev->bdev;
5190         if (async)
5191                 btrfs_schedule_bio(root, dev, rw, bio);
5192         else
5193                 btrfsic_submit_bio(rw, bio);
5194 }
5195
5196 static int breakup_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio,
5197                               struct bio *first_bio, struct btrfs_device *dev,
5198                               int dev_nr, int rw, int async)
5199 {
5200         struct bio_vec *bvec = first_bio->bi_io_vec;
5201         struct bio *bio;
5202         int nr_vecs = bio_get_nr_vecs(dev->bdev);
5203         u64 physical = bbio->stripes[dev_nr].physical;
5204
5205 again:
5206         bio = btrfs_bio_alloc(dev->bdev, physical >> 9, nr_vecs, GFP_NOFS);
5207         if (!bio)
5208                 return -ENOMEM;
5209
5210         while (bvec <= (first_bio->bi_io_vec + first_bio->bi_vcnt - 1)) {
5211                 if (bio_add_page(bio, bvec->bv_page, bvec->bv_len,
5212                                  bvec->bv_offset) < bvec->bv_len) {
5213                         u64 len = bio->bi_size;
5214
5215                         atomic_inc(&bbio->stripes_pending);
5216                         submit_stripe_bio(root, bbio, bio, physical, dev_nr,
5217                                           rw, async);
5218                         physical += len;
5219                         goto again;
5220                 }
5221                 bvec++;
5222         }
5223
5224         submit_stripe_bio(root, bbio, bio, physical, dev_nr, rw, async);
5225         return 0;
5226 }
5227
5228 static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical)
5229 {
5230         atomic_inc(&bbio->error);
5231         if (atomic_dec_and_test(&bbio->stripes_pending)) {
5232                 bio->bi_private = bbio->private;
5233                 bio->bi_end_io = bbio->end_io;
5234                 bio->bi_bdev = (struct block_device *)
5235                         (unsigned long)bbio->mirror_num;
5236                 bio->bi_sector = logical >> 9;
5237                 kfree(bbio);
5238                 bio_endio(bio, -EIO);
5239         }
5240 }
5241
5242 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
5243                   int mirror_num, int async_submit)
5244 {
5245         struct btrfs_device *dev;
5246         struct bio *first_bio = bio;
5247         u64 logical = (u64)bio->bi_sector << 9;
5248         u64 length = 0;
5249         u64 map_length;
5250         u64 *raid_map = NULL;
5251         int ret;
5252         int dev_nr = 0;
5253         int total_devs = 1;
5254         struct btrfs_bio *bbio = NULL;
5255
5256         length = bio->bi_size;
5257         map_length = length;
5258
5259         ret = __btrfs_map_block(root->fs_info, rw, logical, &map_length, &bbio,
5260                               mirror_num, &raid_map);
5261         if (ret) /* -ENOMEM */
5262                 return ret;
5263
5264         total_devs = bbio->num_stripes;
5265         bbio->orig_bio = first_bio;
5266         bbio->private = first_bio->bi_private;
5267         bbio->end_io = first_bio->bi_end_io;
5268         atomic_set(&bbio->stripes_pending, bbio->num_stripes);
5269
5270         if (raid_map) {
5271                 /* In this case, map_length has been set to the length of
5272                    a single stripe; not the whole write */
5273                 if (rw & WRITE) {
5274                         return raid56_parity_write(root, bio, bbio,
5275                                                    raid_map, map_length);
5276                 } else {
5277                         return raid56_parity_recover(root, bio, bbio,
5278                                                      raid_map, map_length,
5279                                                      mirror_num);
5280                 }
5281         }
5282
5283         if (map_length < length) {
5284                 printk(KERN_CRIT "btrfs: mapping failed logical %llu bio len %llu "
5285                        "len %llu\n", (unsigned long long)logical,
5286                        (unsigned long long)length,
5287                        (unsigned long long)map_length);
5288                 BUG();
5289         }
5290
5291         while (dev_nr < total_devs) {
5292                 dev = bbio->stripes[dev_nr].dev;
5293                 if (!dev || !dev->bdev || (rw & WRITE && !dev->writeable)) {
5294                         bbio_error(bbio, first_bio, logical);
5295                         dev_nr++;
5296                         continue;
5297                 }
5298
5299                 /*
5300                  * Check and see if we're ok with this bio based on it's size
5301                  * and offset with the given device.
5302                  */
5303                 if (!bio_size_ok(dev->bdev, first_bio,
5304                                  bbio->stripes[dev_nr].physical >> 9)) {
5305                         ret = breakup_stripe_bio(root, bbio, first_bio, dev,
5306                                                  dev_nr, rw, async_submit);
5307                         BUG_ON(ret);
5308                         dev_nr++;
5309                         continue;
5310                 }
5311
5312                 if (dev_nr < total_devs - 1) {
5313                         bio = bio_clone(first_bio, GFP_NOFS);
5314                         BUG_ON(!bio); /* -ENOMEM */
5315                 } else {
5316                         bio = first_bio;
5317                 }
5318
5319                 submit_stripe_bio(root, bbio, bio,
5320                                   bbio->stripes[dev_nr].physical, dev_nr, rw,
5321                                   async_submit);
5322                 dev_nr++;
5323         }
5324         return 0;
5325 }
5326
5327 struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid,
5328                                        u8 *uuid, u8 *fsid)
5329 {
5330         struct btrfs_device *device;
5331         struct btrfs_fs_devices *cur_devices;
5332
5333         cur_devices = fs_info->fs_devices;
5334         while (cur_devices) {
5335                 if (!fsid ||
5336                     !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
5337                         device = __find_device(&cur_devices->devices,
5338                                                devid, uuid);
5339                         if (device)
5340                                 return device;
5341                 }
5342                 cur_devices = cur_devices->seed;
5343         }
5344         return NULL;
5345 }
5346
5347 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
5348                                             u64 devid, u8 *dev_uuid)
5349 {
5350         struct btrfs_device *device;
5351         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
5352
5353         device = kzalloc(sizeof(*device), GFP_NOFS);
5354         if (!device)
5355                 return NULL;
5356         list_add(&device->dev_list,
5357                  &fs_devices->devices);
5358         device->dev_root = root->fs_info->dev_root;
5359         device->devid = devid;
5360         device->work.func = pending_bios_fn;
5361         device->fs_devices = fs_devices;
5362         device->missing = 1;
5363         fs_devices->num_devices++;
5364         fs_devices->missing_devices++;
5365         spin_lock_init(&device->io_lock);
5366         INIT_LIST_HEAD(&device->dev_alloc_list);
5367         memcpy(device->uuid, dev_uuid, BTRFS_UUID_SIZE);
5368         return device;
5369 }
5370
5371 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
5372                           struct extent_buffer *leaf,
5373                           struct btrfs_chunk *chunk)
5374 {
5375         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
5376         struct map_lookup *map;
5377         struct extent_map *em;
5378         u64 logical;
5379         u64 length;
5380         u64 devid;
5381         u8 uuid[BTRFS_UUID_SIZE];
5382         int num_stripes;
5383         int ret;
5384         int i;
5385
5386         logical = key->offset;
5387         length = btrfs_chunk_length(leaf, chunk);
5388
5389         read_lock(&map_tree->map_tree.lock);
5390         em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
5391         read_unlock(&map_tree->map_tree.lock);
5392
5393         /* already mapped? */
5394         if (em && em->start <= logical && em->start + em->len > logical) {
5395                 free_extent_map(em);
5396                 return 0;
5397         } else if (em) {
5398                 free_extent_map(em);
5399         }
5400
5401         em = alloc_extent_map();
5402         if (!em)
5403                 return -ENOMEM;
5404         num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
5405         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
5406         if (!map) {
5407                 free_extent_map(em);
5408                 return -ENOMEM;
5409         }
5410
5411         em->bdev = (struct block_device *)map;
5412         em->start = logical;
5413         em->len = length;
5414         em->orig_start = 0;
5415         em->block_start = 0;
5416         em->block_len = em->len;
5417
5418         map->num_stripes = num_stripes;
5419         map->io_width = btrfs_chunk_io_width(leaf, chunk);
5420         map->io_align = btrfs_chunk_io_align(leaf, chunk);
5421         map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
5422         map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
5423         map->type = btrfs_chunk_type(leaf, chunk);
5424         map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
5425         for (i = 0; i < num_stripes; i++) {
5426                 map->stripes[i].physical =
5427                         btrfs_stripe_offset_nr(leaf, chunk, i);
5428                 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
5429                 read_extent_buffer(leaf, uuid, (unsigned long)
5430                                    btrfs_stripe_dev_uuid_nr(chunk, i),
5431                                    BTRFS_UUID_SIZE);
5432                 map->stripes[i].dev = btrfs_find_device(root->fs_info, devid,
5433                                                         uuid, NULL);
5434                 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
5435                         kfree(map);
5436                         free_extent_map(em);
5437                         return -EIO;
5438                 }
5439                 if (!map->stripes[i].dev) {
5440                         map->stripes[i].dev =
5441                                 add_missing_dev(root, devid, uuid);
5442                         if (!map->stripes[i].dev) {
5443                                 kfree(map);
5444                                 free_extent_map(em);
5445                                 return -EIO;
5446                         }
5447                 }
5448                 map->stripes[i].dev->in_fs_metadata = 1;
5449         }
5450
5451         write_lock(&map_tree->map_tree.lock);
5452         ret = add_extent_mapping(&map_tree->map_tree, em);
5453         write_unlock(&map_tree->map_tree.lock);
5454         BUG_ON(ret); /* Tree corruption */
5455         free_extent_map(em);
5456
5457         return 0;
5458 }
5459
5460 static void fill_device_from_item(struct extent_buffer *leaf,
5461                                  struct btrfs_dev_item *dev_item,
5462                                  struct btrfs_device *device)
5463 {
5464         unsigned long ptr;
5465
5466         device->devid = btrfs_device_id(leaf, dev_item);
5467         device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
5468         device->total_bytes = device->disk_total_bytes;
5469         device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
5470         device->type = btrfs_device_type(leaf, dev_item);
5471         device->io_align = btrfs_device_io_align(leaf, dev_item);
5472         device->io_width = btrfs_device_io_width(leaf, dev_item);
5473         device->sector_size = btrfs_device_sector_size(leaf, dev_item);
5474         WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
5475         device->is_tgtdev_for_dev_replace = 0;
5476
5477         ptr = (unsigned long)btrfs_device_uuid(dev_item);
5478         read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
5479 }
5480
5481 static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
5482 {
5483         struct btrfs_fs_devices *fs_devices;
5484         int ret;
5485
5486         BUG_ON(!mutex_is_locked(&uuid_mutex));
5487
5488         fs_devices = root->fs_info->fs_devices->seed;
5489         while (fs_devices) {
5490                 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
5491                         ret = 0;
5492                         goto out;
5493                 }
5494                 fs_devices = fs_devices->seed;
5495         }
5496
5497         fs_devices = find_fsid(fsid);
5498         if (!fs_devices) {
5499                 ret = -ENOENT;
5500                 goto out;
5501         }
5502
5503         fs_devices = clone_fs_devices(fs_devices);
5504         if (IS_ERR(fs_devices)) {
5505                 ret = PTR_ERR(fs_devices);
5506                 goto out;
5507         }
5508
5509         ret = __btrfs_open_devices(fs_devices, FMODE_READ,
5510                                    root->fs_info->bdev_holder);
5511         if (ret) {
5512                 free_fs_devices(fs_devices);
5513                 goto out;
5514         }
5515
5516         if (!fs_devices->seeding) {
5517                 __btrfs_close_devices(fs_devices);
5518                 free_fs_devices(fs_devices);
5519                 ret = -EINVAL;
5520                 goto out;
5521         }
5522
5523         fs_devices->seed = root->fs_info->fs_devices->seed;
5524         root->fs_info->fs_devices->seed = fs_devices;
5525 out:
5526         return ret;
5527 }
5528
5529 static int read_one_dev(struct btrfs_root *root,
5530                         struct extent_buffer *leaf,
5531                         struct btrfs_dev_item *dev_item)
5532 {
5533         struct btrfs_device *device;
5534         u64 devid;
5535         int ret;
5536         u8 fs_uuid[BTRFS_UUID_SIZE];
5537         u8 dev_uuid[BTRFS_UUID_SIZE];
5538
5539         devid = btrfs_device_id(leaf, dev_item);
5540         read_extent_buffer(leaf, dev_uuid,
5541                            (unsigned long)btrfs_device_uuid(dev_item),
5542                            BTRFS_UUID_SIZE);
5543         read_extent_buffer(leaf, fs_uuid,
5544                            (unsigned long)btrfs_device_fsid(dev_item),
5545                            BTRFS_UUID_SIZE);
5546
5547         if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
5548                 ret = open_seed_devices(root, fs_uuid);
5549                 if (ret && !btrfs_test_opt(root, DEGRADED))
5550                         return ret;
5551         }
5552
5553         device = btrfs_find_device(root->fs_info, devid, dev_uuid, fs_uuid);
5554         if (!device || !device->bdev) {
5555                 if (!btrfs_test_opt(root, DEGRADED))
5556                         return -EIO;
5557
5558                 if (!device) {
5559                         printk(KERN_WARNING "warning devid %llu missing\n",
5560                                (unsigned long long)devid);
5561                         device = add_missing_dev(root, devid, dev_uuid);
5562                         if (!device)
5563                                 return -ENOMEM;
5564                 } else if (!device->missing) {
5565                         /*
5566                          * this happens when a device that was properly setup
5567                          * in the device info lists suddenly goes bad.
5568                          * device->bdev is NULL, and so we have to set
5569                          * device->missing to one here
5570                          */
5571                         root->fs_info->fs_devices->missing_devices++;
5572                         device->missing = 1;
5573                 }
5574         }
5575
5576         if (device->fs_devices != root->fs_info->fs_devices) {
5577                 BUG_ON(device->writeable);
5578                 if (device->generation !=
5579                     btrfs_device_generation(leaf, dev_item))
5580                         return -EINVAL;
5581         }
5582
5583         fill_device_from_item(leaf, dev_item, device);
5584         device->dev_root = root->fs_info->dev_root;
5585         device->in_fs_metadata = 1;
5586         if (device->writeable && !device->is_tgtdev_for_dev_replace) {
5587                 device->fs_devices->total_rw_bytes += device->total_bytes;
5588                 spin_lock(&root->fs_info->free_chunk_lock);
5589                 root->fs_info->free_chunk_space += device->total_bytes -
5590                         device->bytes_used;
5591                 spin_unlock(&root->fs_info->free_chunk_lock);
5592         }
5593         ret = 0;
5594         return ret;
5595 }
5596
5597 int btrfs_read_sys_array(struct btrfs_root *root)
5598 {
5599         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
5600         struct extent_buffer *sb;
5601         struct btrfs_disk_key *disk_key;
5602         struct btrfs_chunk *chunk;
5603         u8 *ptr;
5604         unsigned long sb_ptr;
5605         int ret = 0;
5606         u32 num_stripes;
5607         u32 array_size;
5608         u32 len = 0;
5609         u32 cur;
5610         struct btrfs_key key;
5611
5612         sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
5613                                           BTRFS_SUPER_INFO_SIZE);
5614         if (!sb)
5615                 return -ENOMEM;
5616         btrfs_set_buffer_uptodate(sb);
5617         btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
5618         /*
5619          * The sb extent buffer is artifical and just used to read the system array.
5620          * btrfs_set_buffer_uptodate() call does not properly mark all it's
5621          * pages up-to-date when the page is larger: extent does not cover the
5622          * whole page and consequently check_page_uptodate does not find all
5623          * the page's extents up-to-date (the hole beyond sb),
5624          * write_extent_buffer then triggers a WARN_ON.
5625          *
5626          * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
5627          * but sb spans only this function. Add an explicit SetPageUptodate call
5628          * to silence the warning eg. on PowerPC 64.
5629          */
5630         if (PAGE_CACHE_SIZE > BTRFS_SUPER_INFO_SIZE)
5631                 SetPageUptodate(sb->pages[0]);
5632
5633         write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
5634         array_size = btrfs_super_sys_array_size(super_copy);
5635
5636         ptr = super_copy->sys_chunk_array;
5637         sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
5638         cur = 0;
5639
5640         while (cur < array_size) {
5641                 disk_key = (struct btrfs_disk_key *)ptr;
5642                 btrfs_disk_key_to_cpu(&key, disk_key);
5643
5644                 len = sizeof(*disk_key); ptr += len;
5645                 sb_ptr += len;
5646                 cur += len;
5647
5648                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
5649                         chunk = (struct btrfs_chunk *)sb_ptr;
5650                         ret = read_one_chunk(root, &key, sb, chunk);
5651                         if (ret)
5652                                 break;
5653                         num_stripes = btrfs_chunk_num_stripes(sb, chunk);
5654                         len = btrfs_chunk_item_size(num_stripes);
5655                 } else {
5656                         ret = -EIO;
5657                         break;
5658                 }
5659                 ptr += len;
5660                 sb_ptr += len;
5661                 cur += len;
5662         }
5663         free_extent_buffer(sb);
5664         return ret;
5665 }
5666
5667 int btrfs_read_chunk_tree(struct btrfs_root *root)
5668 {
5669         struct btrfs_path *path;
5670         struct extent_buffer *leaf;
5671         struct btrfs_key key;
5672         struct btrfs_key found_key;
5673         int ret;
5674         int slot;
5675
5676         root = root->fs_info->chunk_root;
5677
5678         path = btrfs_alloc_path();
5679         if (!path)
5680                 return -ENOMEM;
5681
5682         mutex_lock(&uuid_mutex);
5683         lock_chunks(root);
5684
5685         /* first we search for all of the device items, and then we
5686          * read in all of the chunk items.  This way we can create chunk
5687          * mappings that reference all of the devices that are afound
5688          */
5689         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
5690         key.offset = 0;
5691         key.type = 0;
5692 again:
5693         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5694         if (ret < 0)
5695                 goto error;
5696         while (1) {
5697                 leaf = path->nodes[0];
5698                 slot = path->slots[0];
5699                 if (slot >= btrfs_header_nritems(leaf)) {
5700                         ret = btrfs_next_leaf(root, path);
5701                         if (ret == 0)
5702                                 continue;
5703                         if (ret < 0)
5704                                 goto error;
5705                         break;
5706                 }
5707                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5708                 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
5709                         if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID)
5710                                 break;
5711                         if (found_key.type == BTRFS_DEV_ITEM_KEY) {
5712                                 struct btrfs_dev_item *dev_item;
5713                                 dev_item = btrfs_item_ptr(leaf, slot,
5714                                                   struct btrfs_dev_item);
5715                                 ret = read_one_dev(root, leaf, dev_item);
5716                                 if (ret)
5717                                         goto error;
5718                         }
5719                 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
5720                         struct btrfs_chunk *chunk;
5721                         chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
5722                         ret = read_one_chunk(root, &found_key, leaf, chunk);
5723                         if (ret)
5724                                 goto error;
5725                 }
5726                 path->slots[0]++;
5727         }
5728         if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
5729                 key.objectid = 0;
5730                 btrfs_release_path(path);
5731                 goto again;
5732         }
5733         ret = 0;
5734 error:
5735         unlock_chunks(root);
5736         mutex_unlock(&uuid_mutex);
5737
5738         btrfs_free_path(path);
5739         return ret;
5740 }
5741
5742 static void __btrfs_reset_dev_stats(struct btrfs_device *dev)
5743 {
5744         int i;
5745
5746         for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
5747                 btrfs_dev_stat_reset(dev, i);
5748 }
5749
5750 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
5751 {
5752         struct btrfs_key key;
5753         struct btrfs_key found_key;
5754         struct btrfs_root *dev_root = fs_info->dev_root;
5755         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
5756         struct extent_buffer *eb;
5757         int slot;
5758         int ret = 0;
5759         struct btrfs_device *device;
5760         struct btrfs_path *path = NULL;
5761         int i;
5762
5763         path = btrfs_alloc_path();
5764         if (!path) {
5765                 ret = -ENOMEM;
5766                 goto out;
5767         }
5768
5769         mutex_lock(&fs_devices->device_list_mutex);
5770         list_for_each_entry(device, &fs_devices->devices, dev_list) {
5771                 int item_size;
5772                 struct btrfs_dev_stats_item *ptr;
5773
5774                 key.objectid = 0;
5775                 key.type = BTRFS_DEV_STATS_KEY;
5776                 key.offset = device->devid;
5777                 ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
5778                 if (ret) {
5779                         __btrfs_reset_dev_stats(device);
5780                         device->dev_stats_valid = 1;
5781                         btrfs_release_path(path);
5782                         continue;
5783                 }
5784                 slot = path->slots[0];
5785                 eb = path->nodes[0];
5786                 btrfs_item_key_to_cpu(eb, &found_key, slot);
5787                 item_size = btrfs_item_size_nr(eb, slot);
5788
5789                 ptr = btrfs_item_ptr(eb, slot,
5790                                      struct btrfs_dev_stats_item);
5791
5792                 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
5793                         if (item_size >= (1 + i) * sizeof(__le64))
5794                                 btrfs_dev_stat_set(device, i,
5795                                         btrfs_dev_stats_value(eb, ptr, i));
5796                         else
5797                                 btrfs_dev_stat_reset(device, i);
5798                 }
5799
5800                 device->dev_stats_valid = 1;
5801                 btrfs_dev_stat_print_on_load(device);
5802                 btrfs_release_path(path);
5803         }
5804         mutex_unlock(&fs_devices->device_list_mutex);
5805
5806 out:
5807         btrfs_free_path(path);
5808         return ret < 0 ? ret : 0;
5809 }
5810
5811 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
5812                                 struct btrfs_root *dev_root,
5813                                 struct btrfs_device *device)
5814 {
5815         struct btrfs_path *path;
5816         struct btrfs_key key;
5817         struct extent_buffer *eb;
5818         struct btrfs_dev_stats_item *ptr;
5819         int ret;
5820         int i;
5821
5822         key.objectid = 0;
5823         key.type = BTRFS_DEV_STATS_KEY;
5824         key.offset = device->devid;
5825
5826         path = btrfs_alloc_path();
5827         BUG_ON(!path);
5828         ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
5829         if (ret < 0) {
5830                 printk_in_rcu(KERN_WARNING "btrfs: error %d while searching for dev_stats item for device %s!\n",
5831                               ret, rcu_str_deref(device->name));
5832                 goto out;
5833         }
5834
5835         if (ret == 0 &&
5836             btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
5837                 /* need to delete old one and insert a new one */
5838                 ret = btrfs_del_item(trans, dev_root, path);
5839                 if (ret != 0) {
5840                         printk_in_rcu(KERN_WARNING "btrfs: delete too small dev_stats item for device %s failed %d!\n",
5841                                       rcu_str_deref(device->name), ret);
5842                         goto out;
5843                 }
5844                 ret = 1;
5845         }
5846
5847         if (ret == 1) {
5848                 /* need to insert a new item */
5849                 btrfs_release_path(path);
5850                 ret = btrfs_insert_empty_item(trans, dev_root, path,
5851                                               &key, sizeof(*ptr));
5852                 if (ret < 0) {
5853                         printk_in_rcu(KERN_WARNING "btrfs: insert dev_stats item for device %s failed %d!\n",
5854                                       rcu_str_deref(device->name), ret);
5855                         goto out;
5856                 }
5857         }
5858
5859         eb = path->nodes[0];
5860         ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
5861         for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
5862                 btrfs_set_dev_stats_value(eb, ptr, i,
5863                                           btrfs_dev_stat_read(device, i));
5864         btrfs_mark_buffer_dirty(eb);
5865
5866 out:
5867         btrfs_free_path(path);
5868         return ret;
5869 }
5870
5871 /*
5872  * called from commit_transaction. Writes all changed device stats to disk.
5873  */
5874 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans,
5875                         struct btrfs_fs_info *fs_info)
5876 {
5877         struct btrfs_root *dev_root = fs_info->dev_root;
5878         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
5879         struct btrfs_device *device;
5880         int ret = 0;
5881
5882         mutex_lock(&fs_devices->device_list_mutex);
5883         list_for_each_entry(device, &fs_devices->devices, dev_list) {
5884                 if (!device->dev_stats_valid || !device->dev_stats_dirty)
5885                         continue;
5886
5887                 ret = update_dev_stat_item(trans, dev_root, device);
5888                 if (!ret)
5889                         device->dev_stats_dirty = 0;
5890         }
5891         mutex_unlock(&fs_devices->device_list_mutex);
5892
5893         return ret;
5894 }
5895
5896 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
5897 {
5898         btrfs_dev_stat_inc(dev, index);
5899         btrfs_dev_stat_print_on_error(dev);
5900 }
5901
5902 void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
5903 {
5904         if (!dev->dev_stats_valid)
5905                 return;
5906         printk_ratelimited_in_rcu(KERN_ERR
5907                            "btrfs: bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
5908                            rcu_str_deref(dev->name),
5909                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
5910                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
5911                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
5912                            btrfs_dev_stat_read(dev,
5913                                                BTRFS_DEV_STAT_CORRUPTION_ERRS),
5914                            btrfs_dev_stat_read(dev,
5915                                                BTRFS_DEV_STAT_GENERATION_ERRS));
5916 }
5917
5918 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
5919 {
5920         int i;
5921
5922         for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
5923                 if (btrfs_dev_stat_read(dev, i) != 0)
5924                         break;
5925         if (i == BTRFS_DEV_STAT_VALUES_MAX)
5926                 return; /* all values == 0, suppress message */
5927
5928         printk_in_rcu(KERN_INFO "btrfs: bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
5929                rcu_str_deref(dev->name),
5930                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
5931                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
5932                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
5933                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
5934                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
5935 }
5936
5937 int btrfs_get_dev_stats(struct btrfs_root *root,
5938                         struct btrfs_ioctl_get_dev_stats *stats)
5939 {
5940         struct btrfs_device *dev;
5941         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
5942         int i;
5943
5944         mutex_lock(&fs_devices->device_list_mutex);
5945         dev = btrfs_find_device(root->fs_info, stats->devid, NULL, NULL);
5946         mutex_unlock(&fs_devices->device_list_mutex);
5947
5948         if (!dev) {
5949                 printk(KERN_WARNING
5950                        "btrfs: get dev_stats failed, device not found\n");
5951                 return -ENODEV;
5952         } else if (!dev->dev_stats_valid) {
5953                 printk(KERN_WARNING
5954                        "btrfs: get dev_stats failed, not yet valid\n");
5955                 return -ENODEV;
5956         } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
5957                 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
5958                         if (stats->nr_items > i)
5959                                 stats->values[i] =
5960                                         btrfs_dev_stat_read_and_reset(dev, i);
5961                         else
5962                                 btrfs_dev_stat_reset(dev, i);
5963                 }
5964         } else {
5965                 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
5966                         if (stats->nr_items > i)
5967                                 stats->values[i] = btrfs_dev_stat_read(dev, i);
5968         }
5969         if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
5970                 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
5971         return 0;
5972 }
5973
5974 int btrfs_scratch_superblock(struct btrfs_device *device)
5975 {
5976         struct buffer_head *bh;
5977         struct btrfs_super_block *disk_super;
5978
5979         bh = btrfs_read_dev_super(device->bdev);
5980         if (!bh)
5981                 return -EINVAL;
5982         disk_super = (struct btrfs_super_block *)bh->b_data;
5983
5984         memset(&disk_super->magic, 0, sizeof(disk_super->magic));
5985         set_buffer_dirty(bh);
5986         sync_dirty_buffer(bh);
5987         brelse(bh);
5988
5989         return 0;
5990 }