fs/btrfs/volumes.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18 #include <linux/sched.h>
  19 #include <linux/bio.h>
  20 #include <linux/slab.h>
  21 #include <linux/buffer_head.h>
  22 #include <linux/blkdev.h>
  23 #include <linux/random.h>
  24 #include <linux/iocontext.h>
  25 #include <linux/capability.h>
  26 #include <linux/ratelimit.h>
  27 #include <linux/kthread.h>
  28 #include <linux/raid/pq.h>
  29 #include <linux/semaphore.h>
  30 #include <asm/div64.h>
  31 #include "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 #include "sysfs.h"
  44
  45 static int init_first_rw_device(struct btrfs_trans_handle *trans,
  46                                 struct btrfs_root *root,
  47                                 struct btrfs_device *device);
  48 static int btrfs_relocate_sys_chunks(struct btrfs_root *root);
  49 static void __btrfs_reset_dev_stats(struct btrfs_device *dev);
  50 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev);
  51 static void btrfs_dev_stat_print_on_load(struct btrfs_device *device);
  52
  53 static DEFINE_MUTEX(uuid_mutex);
  54 static LIST_HEAD(fs_uuids);
  55
  56 static void lock_chunks(struct btrfs_root *root)
  57 {
  58         mutex_lock(&root->fs_info->chunk_mutex);
  59 }
  60
  61 static void unlock_chunks(struct btrfs_root *root)
  62 {
  63         mutex_unlock(&root->fs_info->chunk_mutex);
  64 }
  65
  66 static struct btrfs_fs_devices *__alloc_fs_devices(void)
  67 {
  68         struct btrfs_fs_devices *fs_devs;
  69
  70         fs_devs = kzalloc(sizeof(*fs_devs), GFP_NOFS);
  71         if (!fs_devs)
  72                 return ERR_PTR(-ENOMEM);
  73
  74         mutex_init(&fs_devs->device_list_mutex);
  75
  76         INIT_LIST_HEAD(&fs_devs->devices);
  77         INIT_LIST_HEAD(&fs_devs->alloc_list);
  78         INIT_LIST_HEAD(&fs_devs->list);
  79
  80         return fs_devs;
  81 }
  82
  83 /**
  84  * alloc_fs_devices - allocate struct btrfs_fs_devices
  85  * @fsid:       a pointer to UUID for this FS.  If NULL a new UUID is
  86  *              generated.
  87  *
  88  * Return: a pointer to a new &struct btrfs_fs_devices on success;
  89  * ERR_PTR() on error.  Returned struct is not linked onto any lists and
  90  * can be destroyed with kfree() right away.
  91  */
  92 static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid)
  93 {
  94         struct btrfs_fs_devices *fs_devs;
  95
  96         fs_devs = __alloc_fs_devices();
  97         if (IS_ERR(fs_devs))
  98                 return fs_devs;
  99
 100         if (fsid)
 101                 memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE);
 102         else
 103                 generate_random_uuid(fs_devs->fsid);
 104
 105         return fs_devs;
 106 }
 107
 108 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
 109 {
 110         struct btrfs_device *device;
 111         WARN_ON(fs_devices->opened);
 112         while (!list_empty(&fs_devices->devices)) {
 113                 device = list_entry(fs_devices->devices.next,
 114                                     struct btrfs_device, dev_list);
 115                 list_del(&device->dev_list);
 116                 rcu_string_free(device->name);
 117                 kfree(device);
 118         }
 119         kfree(fs_devices);
 120 }
 121
 122 static void btrfs_kobject_uevent(struct block_device *bdev,
 123                                  enum kobject_action action)
 124 {
 125         int ret;
 126
 127         ret = kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, action);
 128         if (ret)
 129                 pr_warn("BTRFS: Sending event '%d' to kobject: '%s' (%p): failed\n",
 130                         action,
 131                         kobject_name(&disk_to_dev(bdev->bd_disk)->kobj),
 132                         &disk_to_dev(bdev->bd_disk)->kobj);
 133 }
 134
 135 void btrfs_cleanup_fs_uuids(void)
 136 {
 137         struct btrfs_fs_devices *fs_devices;
 138
 139         while (!list_empty(&fs_uuids)) {
 140                 fs_devices = list_entry(fs_uuids.next,
 141                                         struct btrfs_fs_devices, list);
 142                 list_del(&fs_devices->list);
 143                 free_fs_devices(fs_devices);
 144         }
 145 }
 146
 147 static struct btrfs_device *__alloc_device(void)
 148 {
 149         struct btrfs_device *dev;
 150
 151         dev = kzalloc(sizeof(*dev), GFP_NOFS);
 152         if (!dev)
 153                 return ERR_PTR(-ENOMEM);
 154
 155         INIT_LIST_HEAD(&dev->dev_list);
 156         INIT_LIST_HEAD(&dev->dev_alloc_list);
 157
 158         spin_lock_init(&dev->io_lock);
 159
 160         spin_lock_init(&dev->reada_lock);
 161         atomic_set(&dev->reada_in_flight, 0);
 162         INIT_RADIX_TREE(&dev->reada_zones, GFP_NOFS & ~__GFP_WAIT);
 163         INIT_RADIX_TREE(&dev->reada_extents, GFP_NOFS & ~__GFP_WAIT);
 164
 165         return dev;
 166 }
 167
 168 static noinline struct btrfs_device *__find_device(struct list_head *head,
 169                                                    u64 devid, u8 *uuid)
 170 {
 171         struct btrfs_device *dev;
 172
 173         list_for_each_entry(dev, head, dev_list) {
 174                 if (dev->devid == devid &&
 175                     (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
 176                         return dev;
 177                 }
 178         }
 179         return NULL;
 180 }
 181
 182 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
 183 {
 184         struct btrfs_fs_devices *fs_devices;
 185
 186         list_for_each_entry(fs_devices, &fs_uuids, list) {
 187                 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
 188                         return fs_devices;
 189         }
 190         return NULL;
 191 }
 192
 193 static int
 194 btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder,
 195                       int flush, struct block_device **bdev,
 196                       struct buffer_head **bh)
 197 {
 198         int ret;
 199
 200         *bdev = blkdev_get_by_path(device_path, flags, holder);
 201
 202         if (IS_ERR(*bdev)) {
 203                 ret = PTR_ERR(*bdev);
 204                 printk(KERN_INFO "BTRFS: open %s failed\n", device_path);
 205                 goto error;
 206         }
 207
 208         if (flush)
 209                 filemap_write_and_wait((*bdev)->bd_inode->i_mapping);
 210         ret = set_blocksize(*bdev, 4096);
 211         if (ret) {
 212                 blkdev_put(*bdev, flags);
 213                 goto error;
 214         }
 215         invalidate_bdev(*bdev);
 216         *bh = btrfs_read_dev_super(*bdev);
 217         if (!*bh) {
 218                 ret = -EINVAL;
 219                 blkdev_put(*bdev, flags);
 220                 goto error;
 221         }
 222
 223         return 0;
 224
 225 error:
 226         *bdev = NULL;
 227         *bh = NULL;
 228         return ret;
 229 }
 230
 231 static void requeue_list(struct btrfs_pending_bios *pending_bios,
 232                         struct bio *head, struct bio *tail)
 233 {
 234
 235         struct bio *old_head;
 236
 237         old_head = pending_bios->head;
 238         pending_bios->head = head;
 239         if (pending_bios->tail)
 240                 tail->bi_next = old_head;
 241         else
 242                 pending_bios->tail = tail;
 243 }
 244
 245 /*
 246  * we try to collect pending bios for a device so we don't get a large
 247  * number of procs sending bios down to the same device.  This greatly
 248  * improves the schedulers ability to collect and merge the bios.
 249  *
 250  * But, it also turns into a long list of bios to process and that is sure
 251  * to eventually make the worker thread block.  The solution here is to
 252  * make some progress and then put this work struct back at the end of
 253  * the list if the block device is congested.  This way, multiple devices
 254  * can make progress from a single worker thread.
 255  */
 256 static noinline void run_scheduled_bios(struct btrfs_device *device)
 257 {
 258         struct bio *pending;
 259         struct backing_dev_info *bdi;
 260         struct btrfs_fs_info *fs_info;
 261         struct btrfs_pending_bios *pending_bios;
 262         struct bio *tail;
 263         struct bio *cur;
 264         int again = 0;
 265         unsigned long num_run;
 266         unsigned long batch_run = 0;
 267         unsigned long limit;
 268         unsigned long last_waited = 0;
 269         int force_reg = 0;
 270         int sync_pending = 0;
 271         struct blk_plug plug;
 272
 273         /*
 274          * this function runs all the bios we've collected for
 275          * a particular device.  We don't want to wander off to
 276          * another device without first sending all of these down.
 277          * So, setup a plug here and finish it off before we return
 278          */
 279         blk_start_plug(&plug);
 280
 281         bdi = blk_get_backing_dev_info(device->bdev);
 282         fs_info = device->dev_root->fs_info;
 283         limit = btrfs_async_submit_limit(fs_info);
 284         limit = limit * 2 / 3;
 285
 286 loop:
 287         spin_lock(&device->io_lock);
 288
 289 loop_lock:
 290         num_run = 0;
 291
 292         /* take all the bios off the list at once and process them
 293          * later on (without the lock held).  But, remember the
 294          * tail and other pointers so the bios can be properly reinserted
 295          * into the list if we hit congestion
 296          */
 297         if (!force_reg && device->pending_sync_bios.head) {
 298                 pending_bios = &device->pending_sync_bios;
 299                 force_reg = 1;
 300         } else {
 301                 pending_bios = &device->pending_bios;
 302                 force_reg = 0;
 303         }
 304
 305         pending = pending_bios->head;
 306         tail = pending_bios->tail;
 307         WARN_ON(pending && !tail);
 308
 309         /*
 310          * if pending was null this time around, no bios need processing
 311          * at all and we can stop.  Otherwise it'll loop back up again
 312          * and do an additional check so no bios are missed.
 313          *
 314          * device->running_pending is used to synchronize with the
 315          * schedule_bio code.
 316          */
 317         if (device->pending_sync_bios.head == NULL &&
 318             device->pending_bios.head == NULL) {
 319                 again = 0;
 320                 device->running_pending = 0;
 321         } else {
 322                 again = 1;
 323                 device->running_pending = 1;
 324         }
 325
 326         pending_bios->head = NULL;
 327         pending_bios->tail = NULL;
 328
 329         spin_unlock(&device->io_lock);
 330
 331         while (pending) {
 332
 333                 rmb();
 334                 /* we want to work on both lists, but do more bios on the
 335                  * sync list than the regular list
 336                  */
 337                 if ((num_run > 32 &&
 338                     pending_bios != &device->pending_sync_bios &&
 339                     device->pending_sync_bios.head) ||
 340                    (num_run > 64 && pending_bios == &device->pending_sync_bios &&
 341                     device->pending_bios.head)) {
 342                         spin_lock(&device->io_lock);
 343                         requeue_list(pending_bios, pending, tail);
 344                         goto loop_lock;
 345                 }
 346
 347                 cur = pending;
 348                 pending = pending->bi_next;
 349                 cur->bi_next = NULL;
 350
 351                 if (atomic_dec_return(&fs_info->nr_async_bios) < limit &&
 352                     waitqueue_active(&fs_info->async_submit_wait))
 353                         wake_up(&fs_info->async_submit_wait);
 354
 355                 BUG_ON(atomic_read(&cur->bi_cnt) == 0);
 356
 357                 /*
 358                  * if we're doing the sync list, record that our
 359                  * plug has some sync requests on it
 360                  *
 361                  * If we're doing the regular list and there are
 362                  * sync requests sitting around, unplug before
 363                  * we add more
 364                  */
 365                 if (pending_bios == &device->pending_sync_bios) {
 366                         sync_pending = 1;
 367                 } else if (sync_pending) {
 368                         blk_finish_plug(&plug);
 369                         blk_start_plug(&plug);
 370                         sync_pending = 0;
 371                 }
 372
 373                 btrfsic_submit_bio(cur->bi_rw, cur);
 374                 num_run++;
 375                 batch_run++;
 376                 if (need_resched())
 377                         cond_resched();
 378
 379                 /*
 380                  * we made progress, there is more work to do and the bdi
 381                  * is now congested.  Back off and let other work structs
 382                  * run instead
 383                  */
 384                 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
 385                     fs_info->fs_devices->open_devices > 1) {
 386                         struct io_context *ioc;
 387
 388                         ioc = current->io_context;
 389
 390                         /*
 391                          * the main goal here is that we don't want to
 392                          * block if we're going to be able to submit
 393                          * more requests without blocking.
 394                          *
 395                          * This code does two great things, it pokes into
 396                          * the elevator code from a filesystem _and_
 397                          * it makes assumptions about how batching works.
 398                          */
 399                         if (ioc && ioc->nr_batch_requests > 0 &&
 400                             time_before(jiffies, ioc->last_waited + HZ/50UL) &&
 401                             (last_waited == 0 ||
 402                              ioc->last_waited == last_waited)) {
 403                                 /*
 404                                  * we want to go through our batch of
 405                                  * requests and stop.  So, we copy out
 406                                  * the ioc->last_waited time and test
 407                                  * against it before looping
 408                                  */
 409                                 last_waited = ioc->last_waited;
 410                                 if (need_resched())
 411                                         cond_resched();
 412                                 continue;
 413                         }
 414                         spin_lock(&device->io_lock);
 415                         requeue_list(pending_bios, pending, tail);
 416                         device->running_pending = 1;
 417
 418                         spin_unlock(&device->io_lock);
 419                         btrfs_queue_work(fs_info->submit_workers,
 420                                          &device->work);
 421                         goto done;
 422                 }
 423                 /* unplug every 64 requests just for good measure */
 424                 if (batch_run % 64 == 0) {
 425                         blk_finish_plug(&plug);
 426                         blk_start_plug(&plug);
 427                         sync_pending = 0;
 428                 }
 429         }
 430
 431         cond_resched();
 432         if (again)
 433                 goto loop;
 434
 435         spin_lock(&device->io_lock);
 436         if (device->pending_bios.head || device->pending_sync_bios.head)
 437                 goto loop_lock;
 438         spin_unlock(&device->io_lock);
 439
 440 done:
 441         blk_finish_plug(&plug);
 442 }
 443
 444 static void pending_bios_fn(struct btrfs_work *work)
 445 {
 446         struct btrfs_device *device;
 447
 448         device = container_of(work, struct btrfs_device, work);
 449         run_scheduled_bios(device);
 450 }
 451
 452 /*
 453  * Add new device to list of registered devices
 454  *
 455  * Returns:
 456  * 1   - first time device is seen
 457  * 0   - device already known
 458  * < 0 - error
 459  */
 460 static noinline int device_list_add(const char *path,
 461                            struct btrfs_super_block *disk_super,
 462                            u64 devid, struct btrfs_fs_devices **fs_devices_ret)
 463 {
 464         struct btrfs_device *device;
 465         struct btrfs_fs_devices *fs_devices;
 466         struct rcu_string *name;
 467         int ret = 0;
 468         u64 found_transid = btrfs_super_generation(disk_super);
 469
 470         fs_devices = find_fsid(disk_super->fsid);
 471         if (!fs_devices) {
 472                 fs_devices = alloc_fs_devices(disk_super->fsid);
 473                 if (IS_ERR(fs_devices))
 474                         return PTR_ERR(fs_devices);
 475
 476                 list_add(&fs_devices->list, &fs_uuids);
 477                 fs_devices->latest_devid = devid;
 478                 fs_devices->latest_trans = found_transid;
 479
 480                 device = NULL;
 481         } else {
 482                 device = __find_device(&fs_devices->devices, devid,
 483                                        disk_super->dev_item.uuid);
 484         }
 485         if (!device) {
 486                 if (fs_devices->opened)
 487                         return -EBUSY;
 488
 489                 device = btrfs_alloc_device(NULL, &devid,
 490                                             disk_super->dev_item.uuid);
 491                 if (IS_ERR(device)) {
 492                         /* we can safely leave the fs_devices entry around */
 493                         return PTR_ERR(device);
 494                 }
 495
 496                 name = rcu_string_strdup(path, GFP_NOFS);
 497                 if (!name) {
 498                         kfree(device);
 499                         return -ENOMEM;
 500                 }
 501                 rcu_assign_pointer(device->name, name);
 502
 503                 mutex_lock(&fs_devices->device_list_mutex);
 504                 list_add_rcu(&device->dev_list, &fs_devices->devices);
 505                 fs_devices->num_devices++;
 506                 mutex_unlock(&fs_devices->device_list_mutex);
 507
 508                 ret = 1;
 509                 device->fs_devices = fs_devices;
 510         } else if (!device->name || strcmp(device->name->str, path)) {
 511                 /*
 512                  * When FS is already mounted.
 513                  * 1. If you are here and if the device->name is NULL that
 514                  *    means this device was missing at time of FS mount.
 515                  * 2. If you are here and if the device->name is different
 516                  *    from 'path' that means either
 517                  *      a. The same device disappeared and reappeared with
 518                  *         different name. or
 519                  *      b. The missing-disk-which-was-replaced, has
 520                  *         reappeared now.
 521                  *
 522                  * We must allow 1 and 2a above. But 2b would be a spurious
 523                  * and unintentional.
 524                  *
 525                  * Further in case of 1 and 2a above, the disk at 'path'
 526                  * would have missed some transaction when it was away and
 527                  * in case of 2a the stale bdev has to be updated as well.
 528                  * 2b must not be allowed at all time.
 529                  */
 530
 531                 /*
 532                  * As of now don't allow update to btrfs_fs_device through
 533                  * the btrfs dev scan cli, after FS has been mounted.
 534                  */
 535                 if (fs_devices->opened) {
 536                         return -EBUSY;
 537                 } else {
 538                         /*
 539                          * That is if the FS is _not_ mounted and if you
 540                          * are here, that means there is more than one
 541                          * disk with same uuid and devid.We keep the one
 542                          * with larger generation number or the last-in if
 543                          * generation are equal.
 544                          */
 545                         if (found_transid < device->generation)
 546                                 return -EEXIST;
 547                 }
 548
 549                 name = rcu_string_strdup(path, GFP_NOFS);
 550                 if (!name)
 551                         return -ENOMEM;
 552                 rcu_string_free(device->name);
 553                 rcu_assign_pointer(device->name, name);
 554                 if (device->missing) {
 555                         fs_devices->missing_devices--;
 556                         device->missing = 0;
 557                 }
 558         }
 559
 560         /*
 561          * Unmount does not free the btrfs_device struct but would zero
 562          * generation along with most of the other members. So just update
 563          * it back. We need it to pick the disk with largest generation
 564          * (as above).
 565          */
 566         if (!fs_devices->opened)
 567                 device->generation = found_transid;
 568
 569         if (found_transid > fs_devices->latest_trans) {
 570                 fs_devices->latest_devid = devid;
 571                 fs_devices->latest_trans = found_transid;
 572         }
 573         *fs_devices_ret = fs_devices;
 574
 575         return ret;
 576 }
 577
 578 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
 579 {
 580         struct btrfs_fs_devices *fs_devices;
 581         struct btrfs_device *device;
 582         struct btrfs_device *orig_dev;
 583
 584         fs_devices = alloc_fs_devices(orig->fsid);
 585         if (IS_ERR(fs_devices))
 586                 return fs_devices;
 587
 588         fs_devices->latest_devid = orig->latest_devid;
 589         fs_devices->latest_trans = orig->latest_trans;
 590         fs_devices->total_devices = orig->total_devices;
 591
 592         /* We have held the volume lock, it is safe to get the devices. */
 593         list_for_each_entry(orig_dev, &orig->devices, dev_list) {
 594                 struct rcu_string *name;
 595
 596                 device = btrfs_alloc_device(NULL, &orig_dev->devid,
 597                                             orig_dev->uuid);
 598                 if (IS_ERR(device))
 599                         goto error;
 600
 601                 /*
 602                  * This is ok to do without rcu read locked because we hold the
 603                  * uuid mutex so nothing we touch in here is going to disappear.
 604                  */
 605                 if (orig_dev->name) {
 606                         name = rcu_string_strdup(orig_dev->name->str, GFP_NOFS);
 607                         if (!name) {
 608                                 kfree(device);
 609                                 goto error;
 610                         }
 611                         rcu_assign_pointer(device->name, name);
 612                 }
 613
 614                 list_add(&device->dev_list, &fs_devices->devices);
 615                 device->fs_devices = fs_devices;
 616                 fs_devices->num_devices++;
 617         }
 618         return fs_devices;
 619 error:
 620         free_fs_devices(fs_devices);
 621         return ERR_PTR(-ENOMEM);
 622 }
 623
 624 void btrfs_close_extra_devices(struct btrfs_fs_info *fs_info,
 625                                struct btrfs_fs_devices *fs_devices, int step)
 626 {
 627         struct btrfs_device *device, *next;
 628
 629         struct block_device *latest_bdev = NULL;
 630         u64 latest_devid = 0;
 631         u64 latest_transid = 0;
 632
 633         mutex_lock(&uuid_mutex);
 634 again:
 635         /* This is the initialized path, it is safe to release the devices. */
 636         list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
 637                 if (device->in_fs_metadata) {
 638                         if (!device->is_tgtdev_for_dev_replace &&
 639                             (!latest_transid ||
 640                              device->generation > latest_transid)) {
 641                                 latest_devid = device->devid;
 642                                 latest_transid = device->generation;
 643                                 latest_bdev = device->bdev;
 644                         }
 645                         continue;
 646                 }
 647
 648                 if (device->devid == BTRFS_DEV_REPLACE_DEVID) {
 649                         /*
 650                          * In the first step, keep the device which has
 651                          * the correct fsid and the devid that is used
 652                          * for the dev_replace procedure.
 653                          * In the second step, the dev_replace state is
 654                          * read from the device tree and it is known
 655                          * whether the procedure is really active or
 656                          * not, which means whether this device is
 657                          * used or whether it should be removed.
 658                          */
 659                         if (step == 0 || device->is_tgtdev_for_dev_replace) {
 660                                 continue;
 661                         }
 662                 }
 663                 if (device->bdev) {
 664                         blkdev_put(device->bdev, device->mode);
 665                         device->bdev = NULL;
 666                         fs_devices->open_devices--;
 667                 }
 668                 if (device->writeable) {
 669                         list_del_init(&device->dev_alloc_list);
 670                         device->writeable = 0;
 671                         if (!device->is_tgtdev_for_dev_replace)
 672                                 fs_devices->rw_devices--;
 673                 }
 674                 list_del_init(&device->dev_list);
 675                 fs_devices->num_devices--;
 676                 rcu_string_free(device->name);
 677                 kfree(device);
 678         }
 679
 680         if (fs_devices->seed) {
 681                 fs_devices = fs_devices->seed;
 682                 goto again;
 683         }
 684
 685         fs_devices->latest_bdev = latest_bdev;
 686         fs_devices->latest_devid = latest_devid;
 687         fs_devices->latest_trans = latest_transid;
 688
 689         mutex_unlock(&uuid_mutex);
 690 }
 691
 692 static void __free_device(struct work_struct *work)
 693 {
 694         struct btrfs_device *device;
 695
 696         device = container_of(work, struct btrfs_device, rcu_work);
 697
 698         if (device->bdev)
 699                 blkdev_put(device->bdev, device->mode);
 700
 701         rcu_string_free(device->name);
 702         kfree(device);
 703 }
 704
 705 static void free_device(struct rcu_head *head)
 706 {
 707         struct btrfs_device *device;
 708
 709         device = container_of(head, struct btrfs_device, rcu);
 710
 711         INIT_WORK(&device->rcu_work, __free_device);
 712         schedule_work(&device->rcu_work);
 713 }
 714
 715 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
 716 {
 717         struct btrfs_device *device;
 718
 719         if (--fs_devices->opened > 0)
 720                 return 0;
 721
 722         mutex_lock(&fs_devices->device_list_mutex);
 723         list_for_each_entry(device, &fs_devices->devices, dev_list) {
 724                 struct btrfs_device *new_device;
 725                 struct rcu_string *name;
 726
 727                 if (device->bdev)
 728                         fs_devices->open_devices--;
 729
 730                 if (device->writeable &&
 731                     device->devid != BTRFS_DEV_REPLACE_DEVID) {
 732                         list_del_init(&device->dev_alloc_list);
 733                         fs_devices->rw_devices--;
 734                 }
 735
 736                 if (device->can_discard)
 737                         fs_devices->num_can_discard--;
 738                 if (device->missing)
 739                         fs_devices->missing_devices--;
 740
 741                 new_device = btrfs_alloc_device(NULL, &device->devid,
 742                                                 device->uuid);
 743                 BUG_ON(IS_ERR(new_device)); /* -ENOMEM */
 744
 745                 /* Safe because we are under uuid_mutex */
 746                 if (device->name) {
 747                         name = rcu_string_strdup(device->name->str, GFP_NOFS);
 748                         BUG_ON(!name); /* -ENOMEM */
 749                         rcu_assign_pointer(new_device->name, name);
 750                 }
 751
 752                 list_replace_rcu(&device->dev_list, &new_device->dev_list);
 753                 new_device->fs_devices = device->fs_devices;
 754
 755                 call_rcu(&device->rcu, free_device);
 756         }
 757         mutex_unlock(&fs_devices->device_list_mutex);
 758
 759         WARN_ON(fs_devices->open_devices);
 760         WARN_ON(fs_devices->rw_devices);
 761         fs_devices->opened = 0;
 762         fs_devices->seeding = 0;
 763
 764         return 0;
 765 }
 766
 767 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
 768 {
 769         struct btrfs_fs_devices *seed_devices = NULL;
 770         int ret;
 771
 772         mutex_lock(&uuid_mutex);
 773         ret = __btrfs_close_devices(fs_devices);
 774         if (!fs_devices->opened) {
 775                 seed_devices = fs_devices->seed;
 776                 fs_devices->seed = NULL;
 777         }
 778         mutex_unlock(&uuid_mutex);
 779
 780         while (seed_devices) {
 781                 fs_devices = seed_devices;
 782                 seed_devices = fs_devices->seed;
 783                 __btrfs_close_devices(fs_devices);
 784                 free_fs_devices(fs_devices);
 785         }
 786         /*
 787          * Wait for rcu kworkers under __btrfs_close_devices
 788          * to finish all blkdev_puts so device is really
 789          * free when umount is done.
 790          */
 791         rcu_barrier();
 792         return ret;
 793 }
 794
 795 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
 796                                 fmode_t flags, void *holder)
 797 {
 798         struct request_queue *q;
 799         struct block_device *bdev;
 800         struct list_head *head = &fs_devices->devices;
 801         struct btrfs_device *device;
 802         struct block_device *latest_bdev = NULL;
 803         struct buffer_head *bh;
 804         struct btrfs_super_block *disk_super;
 805         u64 latest_devid = 0;
 806         u64 latest_transid = 0;
 807         u64 devid;
 808         int seeding = 1;
 809         int ret = 0;
 810
 811         flags |= FMODE_EXCL;
 812
 813         list_for_each_entry(device, head, dev_list) {
 814                 if (device->bdev)
 815                         continue;
 816                 if (!device->name)
 817                         continue;
 818
 819                 /* Just open everything we can; ignore failures here */
 820                 if (btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1,
 821                                             &bdev, &bh))
 822                         continue;
 823
 824                 disk_super = (struct btrfs_super_block *)bh->b_data;
 825                 devid = btrfs_stack_device_id(&disk_super->dev_item);
 826                 if (devid != device->devid)
 827                         goto error_brelse;
 828
 829                 if (memcmp(device->uuid, disk_super->dev_item.uuid,
 830                            BTRFS_UUID_SIZE))
 831                         goto error_brelse;
 832
 833                 device->generation = btrfs_super_generation(disk_super);
 834                 if (!latest_transid || device->generation > latest_transid) {
 835                         latest_devid = devid;
 836                         latest_transid = device->generation;
 837                         latest_bdev = bdev;
 838                 }
 839
 840                 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
 841                         device->writeable = 0;
 842                 } else {
 843                         device->writeable = !bdev_read_only(bdev);
 844                         seeding = 0;
 845                 }
 846
 847                 q = bdev_get_queue(bdev);
 848                 if (blk_queue_discard(q)) {
 849                         device->can_discard = 1;
 850                         fs_devices->num_can_discard++;
 851                 }
 852
 853                 device->bdev = bdev;
 854                 device->in_fs_metadata = 0;
 855                 device->mode = flags;
 856
 857                 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
 858                         fs_devices->rotating = 1;
 859
 860                 fs_devices->open_devices++;
 861                 if (device->writeable &&
 862                     device->devid != BTRFS_DEV_REPLACE_DEVID) {
 863                         fs_devices->rw_devices++;
 864                         list_add(&device->dev_alloc_list,
 865                                  &fs_devices->alloc_list);
 866                 }
 867                 brelse(bh);
 868                 continue;
 869
 870 error_brelse:
 871                 brelse(bh);
 872                 blkdev_put(bdev, flags);
 873                 continue;
 874         }
 875         if (fs_devices->open_devices == 0) {
 876                 ret = -EINVAL;
 877                 goto out;
 878         }
 879         fs_devices->seeding = seeding;
 880         fs_devices->opened = 1;
 881         fs_devices->latest_bdev = latest_bdev;
 882         fs_devices->latest_devid = latest_devid;
 883         fs_devices->latest_trans = latest_transid;
 884         fs_devices->total_rw_bytes = 0;
 885 out:
 886         return ret;
 887 }
 888
 889 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
 890                        fmode_t flags, void *holder)
 891 {
 892         int ret;
 893
 894         mutex_lock(&uuid_mutex);
 895         if (fs_devices->opened) {
 896                 fs_devices->opened++;
 897                 ret = 0;
 898         } else {
 899                 ret = __btrfs_open_devices(fs_devices, flags, holder);
 900         }
 901         mutex_unlock(&uuid_mutex);
 902         return ret;
 903 }
 904
 905 /*
 906  * Look for a btrfs signature on a device. This may be called out of the mount path
 907  * and we are not allowed to call set_blocksize during the scan. The superblock
 908  * is read via pagecache
 909  */
 910 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
 911                           struct btrfs_fs_devices **fs_devices_ret)
 912 {
 913         struct btrfs_super_block *disk_super;
 914         struct block_device *bdev;
 915         struct page *page;
 916         void *p;
 917         int ret = -EINVAL;
 918         u64 devid;
 919         u64 transid;
 920         u64 total_devices;
 921         u64 bytenr;
 922         pgoff_t index;
 923
 924         /*
 925          * we would like to check all the supers, but that would make
 926          * a btrfs mount succeed after a mkfs from a different FS.
 927          * So, we need to add a special mount option to scan for
 928          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
 929          */
 930         bytenr = btrfs_sb_offset(0);
 931         flags |= FMODE_EXCL;
 932         mutex_lock(&uuid_mutex);
 933
 934         bdev = blkdev_get_by_path(path, flags, holder);
 935
 936         if (IS_ERR(bdev)) {
 937                 ret = PTR_ERR(bdev);
 938                 goto error;
 939         }
 940
 941         /* make sure our super fits in the device */
 942         if (bytenr + PAGE_CACHE_SIZE >= i_size_read(bdev->bd_inode))
 943                 goto error_bdev_put;
 944
 945         /* make sure our super fits in the page */
 946         if (sizeof(*disk_super) > PAGE_CACHE_SIZE)
 947                 goto error_bdev_put;
 948
 949         /* make sure our super doesn't straddle pages on disk */
 950         index = bytenr >> PAGE_CACHE_SHIFT;
 951         if ((bytenr + sizeof(*disk_super) - 1) >> PAGE_CACHE_SHIFT != index)
 952                 goto error_bdev_put;
 953
 954         /* pull in the page with our super */
 955         page = read_cache_page_gfp(bdev->bd_inode->i_mapping,
 956                                    index, GFP_NOFS);
 957
 958         if (IS_ERR_OR_NULL(page))
 959                 goto error_bdev_put;
 960
 961         p = kmap(page);
 962
 963         /* align our pointer to the offset of the super block */
 964         disk_super = p + (bytenr & ~PAGE_CACHE_MASK);
 965
 966         if (btrfs_super_bytenr(disk_super) != bytenr ||
 967             btrfs_super_magic(disk_super) != BTRFS_MAGIC)
 968                 goto error_unmap;
 969
 970         devid = btrfs_stack_device_id(&disk_super->dev_item);
 971         transid = btrfs_super_generation(disk_super);
 972         total_devices = btrfs_super_num_devices(disk_super);
 973
 974         ret = device_list_add(path, disk_super, devid, fs_devices_ret);
 975         if (ret > 0) {
 976                 if (disk_super->label[0]) {
 977                         if (disk_super->label[BTRFS_LABEL_SIZE - 1])
 978                                 disk_super->label[BTRFS_LABEL_SIZE - 1] = '\0';
 979                         printk(KERN_INFO "BTRFS: device label %s ", disk_super->label);
 980                 } else {
 981                         printk(KERN_INFO "BTRFS: device fsid %pU ", disk_super->fsid);
 982                 }
 983
 984                 printk(KERN_CONT "devid %llu transid %llu %s\n", devid, transid, path);
 985                 ret = 0;
 986         }
 987         if (!ret && fs_devices_ret)
 988                 (*fs_devices_ret)->total_devices = total_devices;
 989
 990 error_unmap:
 991         kunmap(page);
 992         page_cache_release(page);
 993
 994 error_bdev_put:
 995         blkdev_put(bdev, flags);
 996 error:
 997         mutex_unlock(&uuid_mutex);
 998         return ret;
 999 }
1000
1001 /* helper to account the used device space in the range */
1002 int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
1003                                    u64 end, u64 *length)
1004 {
1005         struct btrfs_key key;
1006         struct btrfs_root *root = device->dev_root;
1007         struct btrfs_dev_extent *dev_extent;
1008         struct btrfs_path *path;
1009         u64 extent_end;
1010         int ret;
1011         int slot;
1012         struct extent_buffer *l;
1013
1014         *length = 0;
1015
1016         if (start >= device->total_bytes || device->is_tgtdev_for_dev_replace)
1017                 return 0;
1018
1019         path = btrfs_alloc_path();
1020         if (!path)
1021                 return -ENOMEM;
1022         path->reada = 2;
1023
1024         key.objectid = device->devid;
1025         key.offset = start;
1026         key.type = BTRFS_DEV_EXTENT_KEY;
1027
1028         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1029         if (ret < 0)
1030                 goto out;
1031         if (ret > 0) {
1032                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1033                 if (ret < 0)
1034                         goto out;
1035         }
1036
1037         while (1) {
1038                 l = path->nodes[0];
1039                 slot = path->slots[0];
1040                 if (slot >= btrfs_header_nritems(l)) {
1041                         ret = btrfs_next_leaf(root, path);
1042                         if (ret == 0)
1043                                 continue;
1044                         if (ret < 0)
1045                                 goto out;
1046
1047                         break;
1048                 }
1049                 btrfs_item_key_to_cpu(l, &key, slot);
1050
1051                 if (key.objectid < device->devid)
1052                         goto next;
1053
1054                 if (key.objectid > device->devid)
1055                         break;
1056
1057                 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
1058                         goto next;
1059
1060                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1061                 extent_end = key.offset + btrfs_dev_extent_length(l,
1062                                                                   dev_extent);
1063                 if (key.offset <= start && extent_end > end) {
1064                         *length = end - start + 1;
1065                         break;
1066                 } else if (key.offset <= start && extent_end > start)
1067                         *length += extent_end - start;
1068                 else if (key.offset > start && extent_end <= end)
1069                         *length += extent_end - key.offset;
1070                 else if (key.offset > start && key.offset <= end) {
1071                         *length += end - key.offset + 1;
1072                         break;
1073                 } else if (key.offset > end)
1074                         break;
1075
1076 next:
1077                 path->slots[0]++;
1078         }
1079         ret = 0;
1080 out:
1081         btrfs_free_path(path);
1082         return ret;
1083 }
1084
1085 static int contains_pending_extent(struct btrfs_trans_handle *trans,
1086                                    struct btrfs_device *device,
1087                                    u64 *start, u64 len)
1088 {
1089         struct extent_map *em;
1090         int ret = 0;
1091
1092         list_for_each_entry(em, &trans->transaction->pending_chunks, list) {
1093                 struct map_lookup *map;
1094                 int i;
1095
1096                 map = (struct map_lookup *)em->bdev;
1097                 for (i = 0; i < map->num_stripes; i++) {
1098                         if (map->stripes[i].dev != device)
1099                                 continue;
1100                         if (map->stripes[i].physical >= *start + len ||
1101                             map->stripes[i].physical + em->orig_block_len <=
1102                             *start)
1103                                 continue;
1104                         *start = map->stripes[i].physical +
1105                                 em->orig_block_len;
1106                         ret = 1;
1107                 }
1108         }
1109
1110         return ret;
1111 }
1112
1113
1114 /*
1115  * find_free_dev_extent - find free space in the specified device
1116  * @device:     the device which we search the free space in
1117  * @num_bytes:  the size of the free space that we need
1118  * @start:      store the start of the free space.
1119  * @len:        the size of the free space. that we find, or the size of the max
1120  *              free space if we don't find suitable free space
1121  *
1122  * this uses a pretty simple search, the expectation is that it is
1123  * called very infrequently and that a given device has a small number
1124  * of extents
1125  *
1126  * @start is used to store the start of the free space if we find. But if we
1127  * don't find suitable free space, it will be used to store the start position
1128  * of the max free space.
1129  *
1130  * @len is used to store the size of the free space that we find.
1131  * But if we don't find suitable free space, it is used to store the size of
1132  * the max free space.
1133  */
1134 int find_free_dev_extent(struct btrfs_trans_handle *trans,
1135                          struct btrfs_device *device, u64 num_bytes,
1136                          u64 *start, u64 *len)
1137 {
1138         struct btrfs_key key;
1139         struct btrfs_root *root = device->dev_root;
1140         struct btrfs_dev_extent *dev_extent;
1141         struct btrfs_path *path;
1142         u64 hole_size;
1143         u64 max_hole_start;
1144         u64 max_hole_size;
1145         u64 extent_end;
1146         u64 search_start;
1147         u64 search_end = device->total_bytes;
1148         int ret;
1149         int slot;
1150         struct extent_buffer *l;
1151
1152         /* FIXME use last free of some kind */
1153
1154         /* we don't want to overwrite the superblock on the drive,
1155          * so we make sure to start at an offset of at least 1MB
1156          */
1157         search_start = max(root->fs_info->alloc_start, 1024ull * 1024);
1158
1159         path = btrfs_alloc_path();
1160         if (!path)
1161                 return -ENOMEM;
1162 again:
1163         max_hole_start = search_start;
1164         max_hole_size = 0;
1165         hole_size = 0;
1166
1167         if (search_start >= search_end || device->is_tgtdev_for_dev_replace) {
1168                 ret = -ENOSPC;
1169                 goto out;
1170         }
1171
1172         path->reada = 2;
1173         path->search_commit_root = 1;
1174         path->skip_locking = 1;
1175
1176         key.objectid = device->devid;
1177         key.offset = search_start;
1178         key.type = BTRFS_DEV_EXTENT_KEY;
1179
1180         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1181         if (ret < 0)
1182                 goto out;
1183         if (ret > 0) {
1184                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1185                 if (ret < 0)
1186                         goto out;
1187         }
1188
1189         while (1) {
1190                 l = path->nodes[0];
1191                 slot = path->slots[0];
1192                 if (slot >= btrfs_header_nritems(l)) {
1193                         ret = btrfs_next_leaf(root, path);
1194                         if (ret == 0)
1195                                 continue;
1196                         if (ret < 0)
1197                                 goto out;
1198
1199                         break;
1200                 }
1201                 btrfs_item_key_to_cpu(l, &key, slot);
1202
1203                 if (key.objectid < device->devid)
1204                         goto next;
1205
1206                 if (key.objectid > device->devid)
1207                         break;
1208
1209                 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
1210                         goto next;
1211
1212                 if (key.offset > search_start) {
1213                         hole_size = key.offset - search_start;
1214
1215                         /*
1216                          * Have to check before we set max_hole_start, otherwise
1217                          * we could end up sending back this offset anyway.
1218                          */
1219                         if (contains_pending_extent(trans, device,
1220                                                     &search_start,
1221                                                     hole_size))
1222                                 hole_size = 0;
1223
1224                         if (hole_size > max_hole_size) {
1225                                 max_hole_start = search_start;
1226                                 max_hole_size = hole_size;
1227                         }
1228
1229                         /*
1230                          * If this free space is greater than which we need,
1231                          * it must be the max free space that we have found
1232                          * until now, so max_hole_start must point to the start
1233                          * of this free space and the length of this free space
1234                          * is stored in max_hole_size. Thus, we return
1235                          * max_hole_start and max_hole_size and go back to the
1236                          * caller.
1237                          */
1238                         if (hole_size >= num_bytes) {
1239                                 ret = 0;
1240                                 goto out;
1241                         }
1242                 }
1243
1244                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1245                 extent_end = key.offset + btrfs_dev_extent_length(l,
1246                                                                   dev_extent);
1247                 if (extent_end > search_start)
1248                         search_start = extent_end;
1249 next:
1250                 path->slots[0]++;
1251                 cond_resched();
1252         }
1253
1254         /*
1255          * At this point, search_start should be the end of
1256          * allocated dev extents, and when shrinking the device,
1257          * search_end may be smaller than search_start.
1258          */
1259         if (search_end > search_start)
1260                 hole_size = search_end - search_start;
1261
1262         if (hole_size > max_hole_size) {
1263                 max_hole_start = search_start;
1264                 max_hole_size = hole_size;
1265         }
1266
1267         if (contains_pending_extent(trans, device, &search_start, hole_size)) {
1268                 btrfs_release_path(path);
1269                 goto again;
1270         }
1271
1272         /* See above. */
1273         if (hole_size < num_bytes)
1274                 ret = -ENOSPC;
1275         else
1276                 ret = 0;
1277
1278 out:
1279         btrfs_free_path(path);
1280         *start = max_hole_start;
1281         if (len)
1282                 *len = max_hole_size;
1283         return ret;
1284 }
1285
1286 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1287                           struct btrfs_device *device,
1288                           u64 start)
1289 {
1290         int ret;
1291         struct btrfs_path *path;
1292         struct btrfs_root *root = device->dev_root;
1293         struct btrfs_key key;
1294         struct btrfs_key found_key;
1295         struct extent_buffer *leaf = NULL;
1296         struct btrfs_dev_extent *extent = NULL;
1297
1298         path = btrfs_alloc_path();
1299         if (!path)
1300                 return -ENOMEM;
1301
1302         key.objectid = device->devid;
1303         key.offset = start;
1304         key.type = BTRFS_DEV_EXTENT_KEY;
1305 again:
1306         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1307         if (ret > 0) {
1308                 ret = btrfs_previous_item(root, path, key.objectid,
1309                                           BTRFS_DEV_EXTENT_KEY);
1310                 if (ret)
1311                         goto out;
1312                 leaf = path->nodes[0];
1313                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1314                 extent = btrfs_item_ptr(leaf, path->slots[0],
1315                                         struct btrfs_dev_extent);
1316                 BUG_ON(found_key.offset > start || found_key.offset +
1317                        btrfs_dev_extent_length(leaf, extent) < start);
1318                 key = found_key;
1319                 btrfs_release_path(path);
1320                 goto again;
1321         } else if (ret == 0) {
1322                 leaf = path->nodes[0];
1323                 extent = btrfs_item_ptr(leaf, path->slots[0],
1324                                         struct btrfs_dev_extent);
1325         } else {
1326                 btrfs_error(root->fs_info, ret, "Slot search failed");
1327                 goto out;
1328         }
1329
1330         if (device->bytes_used > 0) {
1331                 u64 len = btrfs_dev_extent_length(leaf, extent);
1332                 device->bytes_used -= len;
1333                 spin_lock(&root->fs_info->free_chunk_lock);
1334                 root->fs_info->free_chunk_space += len;
1335                 spin_unlock(&root->fs_info->free_chunk_lock);
1336         }
1337         ret = btrfs_del_item(trans, root, path);
1338         if (ret) {
1339                 btrfs_error(root->fs_info, ret,
1340                             "Failed to remove dev extent item");
1341         }
1342 out:
1343         btrfs_free_path(path);
1344         return ret;
1345 }
1346
1347 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1348                                   struct btrfs_device *device,
1349                                   u64 chunk_tree, u64 chunk_objectid,
1350                                   u64 chunk_offset, u64 start, u64 num_bytes)
1351 {
1352         int ret;
1353         struct btrfs_path *path;
1354         struct btrfs_root *root = device->dev_root;
1355         struct btrfs_dev_extent *extent;
1356         struct extent_buffer *leaf;
1357         struct btrfs_key key;
1358
1359         WARN_ON(!device->in_fs_metadata);
1360         WARN_ON(device->is_tgtdev_for_dev_replace);
1361         path = btrfs_alloc_path();
1362         if (!path)
1363                 return -ENOMEM;
1364
1365         key.objectid = device->devid;
1366         key.offset = start;
1367         key.type = BTRFS_DEV_EXTENT_KEY;
1368         ret = btrfs_insert_empty_item(trans, root, path, &key,
1369                                       sizeof(*extent));
1370         if (ret)
1371                 goto out;
1372
1373         leaf = path->nodes[0];
1374         extent = btrfs_item_ptr(leaf, path->slots[0],
1375                                 struct btrfs_dev_extent);
1376         btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
1377         btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
1378         btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1379
1380         write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
1381                     btrfs_dev_extent_chunk_tree_uuid(extent), BTRFS_UUID_SIZE);
1382
1383         btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1384         btrfs_mark_buffer_dirty(leaf);
1385 out:
1386         btrfs_free_path(path);
1387         return ret;
1388 }
1389
1390 static u64 find_next_chunk(struct btrfs_fs_info *fs_info)
1391 {
1392         struct extent_map_tree *em_tree;
1393         struct extent_map *em;
1394         struct rb_node *n;
1395         u64 ret = 0;
1396
1397         em_tree = &fs_info->mapping_tree.map_tree;
1398         read_lock(&em_tree->lock);
1399         n = rb_last(&em_tree->map);
1400         if (n) {
1401                 em = rb_entry(n, struct extent_map, rb_node);
1402                 ret = em->start + em->len;
1403         }
1404         read_unlock(&em_tree->lock);
1405
1406         return ret;
1407 }
1408
1409 static noinline int find_next_devid(struct btrfs_fs_info *fs_info,
1410                                     u64 *devid_ret)
1411 {
1412         int ret;
1413         struct btrfs_key key;
1414         struct btrfs_key found_key;
1415         struct btrfs_path *path;
1416
1417         path = btrfs_alloc_path();
1418         if (!path)
1419                 return -ENOMEM;
1420
1421         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1422         key.type = BTRFS_DEV_ITEM_KEY;
1423         key.offset = (u64)-1;
1424
1425         ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
1426         if (ret < 0)
1427                 goto error;
1428
1429         BUG_ON(ret == 0); /* Corruption */
1430
1431         ret = btrfs_previous_item(fs_info->chunk_root, path,
1432                                   BTRFS_DEV_ITEMS_OBJECTID,
1433                                   BTRFS_DEV_ITEM_KEY);
1434         if (ret) {
1435                 *devid_ret = 1;
1436         } else {
1437                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1438                                       path->slots[0]);
1439                 *devid_ret = found_key.offset + 1;
1440         }
1441         ret = 0;
1442 error:
1443         btrfs_free_path(path);
1444         return ret;
1445 }
1446
1447 /*
1448  * the device information is stored in the chunk root
1449  * the btrfs_device struct should be fully filled in
1450  */
1451 static int btrfs_add_device(struct btrfs_trans_handle *trans,
1452                             struct btrfs_root *root,
1453                             struct btrfs_device *device)
1454 {
1455         int ret;
1456         struct btrfs_path *path;
1457         struct btrfs_dev_item *dev_item;
1458         struct extent_buffer *leaf;
1459         struct btrfs_key key;
1460         unsigned long ptr;
1461
1462         root = root->fs_info->chunk_root;
1463
1464         path = btrfs_alloc_path();
1465         if (!path)
1466                 return -ENOMEM;
1467
1468         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1469         key.type = BTRFS_DEV_ITEM_KEY;
1470         key.offset = device->devid;
1471
1472         ret = btrfs_insert_empty_item(trans, root, path, &key,
1473                                       sizeof(*dev_item));
1474         if (ret)
1475                 goto out;
1476
1477         leaf = path->nodes[0];
1478         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1479
1480         btrfs_set_device_id(leaf, dev_item, device->devid);
1481         btrfs_set_device_generation(leaf, dev_item, 0);
1482         btrfs_set_device_type(leaf, dev_item, device->type);
1483         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1484         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1485         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1486         btrfs_set_device_total_bytes(leaf, dev_item, device->disk_total_bytes);
1487         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1488         btrfs_set_device_group(leaf, dev_item, 0);
1489         btrfs_set_device_seek_speed(leaf, dev_item, 0);
1490         btrfs_set_device_bandwidth(leaf, dev_item, 0);
1491         btrfs_set_device_start_offset(leaf, dev_item, 0);
1492
1493         ptr = btrfs_device_uuid(dev_item);
1494         write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1495         ptr = btrfs_device_fsid(dev_item);
1496         write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1497         btrfs_mark_buffer_dirty(leaf);
1498
1499         ret = 0;
1500 out:
1501         btrfs_free_path(path);
1502         return ret;
1503 }
1504
1505 /*
1506  * Function to update ctime/mtime for a given device path.
1507  * Mainly used for ctime/mtime based probe like libblkid.
1508  */
1509 static void update_dev_time(char *path_name)
1510 {
1511         struct file *filp;
1512
1513         filp = filp_open(path_name, O_RDWR, 0);
1514         if (!filp)
1515                 return;
1516         file_update_time(filp);
1517         filp_close(filp, NULL);
1518         return;
1519 }
1520
1521 static int btrfs_rm_dev_item(struct btrfs_root *root,
1522                              struct btrfs_device *device)
1523 {
1524         int ret;
1525         struct btrfs_path *path;
1526         struct btrfs_key key;
1527         struct btrfs_trans_handle *trans;
1528
1529         root = root->fs_info->chunk_root;
1530
1531         path = btrfs_alloc_path();
1532         if (!path)
1533                 return -ENOMEM;
1534
1535         trans = btrfs_start_transaction(root, 0);
1536         if (IS_ERR(trans)) {
1537                 btrfs_free_path(path);
1538                 return PTR_ERR(trans);
1539         }
1540         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1541         key.type = BTRFS_DEV_ITEM_KEY;
1542         key.offset = device->devid;
1543         lock_chunks(root);
1544
1545         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1546         if (ret < 0)
1547                 goto out;
1548
1549         if (ret > 0) {
1550                 ret = -ENOENT;
1551                 goto out;
1552         }
1553
1554         ret = btrfs_del_item(trans, root, path);
1555         if (ret)
1556                 goto out;
1557 out:
1558         btrfs_free_path(path);
1559         unlock_chunks(root);
1560         btrfs_commit_transaction(trans, root);
1561         return ret;
1562 }
1563
1564 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
1565 {
1566         struct btrfs_device *device;
1567         struct btrfs_device *next_device;
1568         struct block_device *bdev;
1569         struct buffer_head *bh = NULL;
1570         struct btrfs_super_block *disk_super;
1571         struct btrfs_fs_devices *cur_devices;
1572         u64 all_avail;
1573         u64 devid;
1574         u64 num_devices;
1575         u8 *dev_uuid;
1576         unsigned seq;
1577         int ret = 0;
1578         bool clear_super = false;
1579
1580         mutex_lock(&uuid_mutex);
1581
1582         do {
1583                 seq = read_seqbegin(&root->fs_info->profiles_lock);
1584
1585                 all_avail = root->fs_info->avail_data_alloc_bits |
1586                             root->fs_info->avail_system_alloc_bits |
1587                             root->fs_info->avail_metadata_alloc_bits;
1588         } while (read_seqretry(&root->fs_info->profiles_lock, seq));
1589
1590         num_devices = root->fs_info->fs_devices->num_devices;
1591         btrfs_dev_replace_lock(&root->fs_info->dev_replace);
1592         if (btrfs_dev_replace_is_ongoing(&root->fs_info->dev_replace)) {
1593                 WARN_ON(num_devices < 1);
1594                 num_devices--;
1595         }
1596         btrfs_dev_replace_unlock(&root->fs_info->dev_replace);
1597
1598         if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) && num_devices <= 4) {
1599                 ret = BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET;
1600                 goto out;
1601         }
1602
1603         if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) && num_devices <= 2) {
1604                 ret = BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET;
1605                 goto out;
1606         }
1607
1608         if ((all_avail & BTRFS_BLOCK_GROUP_RAID5) &&
1609             root->fs_info->fs_devices->rw_devices <= 2) {
1610                 ret = BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET;
1611                 goto out;
1612         }
1613         if ((all_avail & BTRFS_BLOCK_GROUP_RAID6) &&
1614             root->fs_info->fs_devices->rw_devices <= 3) {
1615                 ret = BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET;
1616                 goto out;
1617         }
1618
1619         if (strcmp(device_path, "missing") == 0) {
1620                 struct list_head *devices;
1621                 struct btrfs_device *tmp;
1622
1623                 device = NULL;
1624                 devices = &root->fs_info->fs_devices->devices;
1625                 /*
1626                  * It is safe to read the devices since the volume_mutex
1627                  * is held.
1628                  */
1629                 list_for_each_entry(tmp, devices, dev_list) {
1630                         if (tmp->in_fs_metadata &&
1631                             !tmp->is_tgtdev_for_dev_replace &&
1632                             !tmp->bdev) {
1633                                 device = tmp;
1634                                 break;
1635                         }
1636                 }
1637                 bdev = NULL;
1638                 bh = NULL;
1639                 disk_super = NULL;
1640                 if (!device) {
1641                         ret = BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
1642                         goto out;
1643                 }
1644         } else {
1645                 ret = btrfs_get_bdev_and_sb(device_path,
1646                                             FMODE_WRITE | FMODE_EXCL,
1647                                             root->fs_info->bdev_holder, 0,
1648                                             &bdev, &bh);
1649                 if (ret)
1650                         goto out;
1651                 disk_super = (struct btrfs_super_block *)bh->b_data;
1652                 devid = btrfs_stack_device_id(&disk_super->dev_item);
1653                 dev_uuid = disk_super->dev_item.uuid;
1654                 device = btrfs_find_device(root->fs_info, devid, dev_uuid,
1655                                            disk_super->fsid);
1656                 if (!device) {
1657                         ret = -ENOENT;
1658                         goto error_brelse;
1659                 }
1660         }
1661
1662         if (device->is_tgtdev_for_dev_replace) {
1663                 ret = BTRFS_ERROR_DEV_TGT_REPLACE;
1664                 goto error_brelse;
1665         }
1666
1667         if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1668                 ret = BTRFS_ERROR_DEV_ONLY_WRITABLE;
1669                 goto error_brelse;
1670         }
1671
1672         if (device->writeable) {
1673                 lock_chunks(root);
1674                 list_del_init(&device->dev_alloc_list);
1675                 unlock_chunks(root);
1676                 root->fs_info->fs_devices->rw_devices--;
1677                 clear_super = true;
1678         }
1679
1680         mutex_unlock(&uuid_mutex);
1681         ret = btrfs_shrink_device(device, 0);
1682         mutex_lock(&uuid_mutex);
1683         if (ret)
1684                 goto error_undo;
1685
1686         /*
1687          * TODO: the superblock still includes this device in its num_devices
1688          * counter although write_all_supers() is not locked out. This
1689          * could give a filesystem state which requires a degraded mount.
1690          */
1691         ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1692         if (ret)
1693                 goto error_undo;
1694
1695         spin_lock(&root->fs_info->free_chunk_lock);
1696         root->fs_info->free_chunk_space = device->total_bytes -
1697                 device->bytes_used;
1698         spin_unlock(&root->fs_info->free_chunk_lock);
1699
1700         device->in_fs_metadata = 0;
1701         btrfs_scrub_cancel_dev(root->fs_info, device);
1702
1703         /*
1704          * the device list mutex makes sure that we don't change
1705          * the device list while someone else is writing out all
1706          * the device supers. Whoever is writing all supers, should
1707          * lock the device list mutex before getting the number of
1708          * devices in the super block (super_copy). Conversely,
1709          * whoever updates the number of devices in the super block
1710          * (super_copy) should hold the device list mutex.
1711          */
1712
1713         cur_devices = device->fs_devices;
1714         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1715         list_del_rcu(&device->dev_list);
1716
1717         device->fs_devices->num_devices--;
1718         device->fs_devices->total_devices--;
1719
1720         if (device->missing)
1721                 device->fs_devices->missing_devices--;
1722
1723         next_device = list_entry(root->fs_info->fs_devices->devices.next,
1724                                  struct btrfs_device, dev_list);
1725         if (device->bdev == root->fs_info->sb->s_bdev)
1726                 root->fs_info->sb->s_bdev = next_device->bdev;
1727         if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1728                 root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1729
1730         if (device->bdev) {
1731                 device->fs_devices->open_devices--;
1732                 /* remove sysfs entry */
1733                 btrfs_kobj_rm_device(root->fs_info, device);
1734         }
1735
1736         call_rcu(&device->rcu, free_device);
1737
1738         num_devices = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
1739         btrfs_set_super_num_devices(root->fs_info->super_copy, num_devices);
1740         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1741
1742         if (cur_devices->open_devices == 0) {
1743                 struct btrfs_fs_devices *fs_devices;
1744                 fs_devices = root->fs_info->fs_devices;
1745                 while (fs_devices) {
1746                         if (fs_devices->seed == cur_devices) {
1747                                 fs_devices->seed = cur_devices->seed;
1748                                 break;
1749                         }
1750                         fs_devices = fs_devices->seed;
1751                 }
1752                 cur_devices->seed = NULL;
1753                 lock_chunks(root);
1754                 __btrfs_close_devices(cur_devices);
1755                 unlock_chunks(root);
1756                 free_fs_devices(cur_devices);
1757         }
1758
1759         root->fs_info->num_tolerated_disk_barrier_failures =
1760                 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
1761
1762         /*
1763          * at this point, the device is zero sized.  We want to
1764          * remove it from the devices list and zero out the old super
1765          */
1766         if (clear_super && disk_super) {
1767                 u64 bytenr;
1768                 int i;
1769
1770                 /* make sure this device isn't detected as part of
1771                  * the FS anymore
1772                  */
1773                 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1774                 set_buffer_dirty(bh);
1775                 sync_dirty_buffer(bh);
1776
1777                 /* clear the mirror copies of super block on the disk
1778                  * being removed, 0th copy is been taken care above and
1779                  * the below would take of the rest
1780                  */
1781                 for (i = 1; i < BTRFS_SUPER_MIRROR_MAX; i++) {
1782                         bytenr = btrfs_sb_offset(i);
1783                         if (bytenr + BTRFS_SUPER_INFO_SIZE >=
1784                                         i_size_read(bdev->bd_inode))
1785                                 break;
1786
1787                         brelse(bh);
1788                         bh = __bread(bdev, bytenr / 4096,
1789                                         BTRFS_SUPER_INFO_SIZE);
1790                         if (!bh)
1791                                 continue;
1792
1793                         disk_super = (struct btrfs_super_block *)bh->b_data;
1794
1795                         if (btrfs_super_bytenr(disk_super) != bytenr ||
1796                                 btrfs_super_magic(disk_super) != BTRFS_MAGIC) {
1797                                 continue;
1798                         }
1799                         memset(&disk_super->magic, 0,
1800                                                 sizeof(disk_super->magic));
1801                         set_buffer_dirty(bh);
1802                         sync_dirty_buffer(bh);
1803                 }
1804         }
1805
1806         ret = 0;
1807
1808         if (bdev) {
1809                 /* Notify udev that device has changed */
1810                 btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
1811
1812                 /* Update ctime/mtime for device path for libblkid */
1813                 update_dev_time(device_path);
1814         }
1815
1816 error_brelse:
1817         brelse(bh);
1818         if (bdev)
1819                 blkdev_put(bdev, FMODE_READ | FMODE_EXCL);
1820 out:
1821         mutex_unlock(&uuid_mutex);
1822         return ret;
1823 error_undo:
1824         if (device->writeable) {
1825                 lock_chunks(root);
1826                 list_add(&device->dev_alloc_list,
1827                          &root->fs_info->fs_devices->alloc_list);
1828                 unlock_chunks(root);
1829                 root->fs_info->fs_devices->rw_devices++;
1830         }
1831         goto error_brelse;
1832 }
1833
1834 void btrfs_rm_dev_replace_srcdev(struct btrfs_fs_info *fs_info,
1835                                  struct btrfs_device *srcdev)
1836 {
1837         WARN_ON(!mutex_is_locked(&fs_info->fs_devices->device_list_mutex));
1838
1839         list_del_rcu(&srcdev->dev_list);
1840         list_del_rcu(&srcdev->dev_alloc_list);
1841         fs_info->fs_devices->num_devices--;
1842         if (srcdev->missing) {
1843                 fs_info->fs_devices->missing_devices--;
1844                 fs_info->fs_devices->rw_devices++;
1845         }
1846         if (srcdev->can_discard)
1847                 fs_info->fs_devices->num_can_discard--;
1848         if (srcdev->bdev) {
1849                 fs_info->fs_devices->open_devices--;
1850
1851                 /*
1852                  * zero out the old super if it is not writable
1853                  * (e.g. seed device)
1854                  */
1855                 if (srcdev->writeable)
1856                         btrfs_scratch_superblock(srcdev);
1857         }
1858
1859         call_rcu(&srcdev->rcu, free_device);
1860 }
1861
1862 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
1863                                       struct btrfs_device *tgtdev)
1864 {
1865         struct btrfs_device *next_device;
1866
1867         WARN_ON(!tgtdev);
1868         mutex_lock(&fs_info->fs_devices->device_list_mutex);
1869         if (tgtdev->bdev) {
1870                 btrfs_scratch_superblock(tgtdev);
1871                 fs_info->fs_devices->open_devices--;
1872         }
1873         fs_info->fs_devices->num_devices--;
1874         if (tgtdev->can_discard)
1875                 fs_info->fs_devices->num_can_discard++;
1876
1877         next_device = list_entry(fs_info->fs_devices->devices.next,
1878                                  struct btrfs_device, dev_list);
1879         if (tgtdev->bdev == fs_info->sb->s_bdev)
1880                 fs_info->sb->s_bdev = next_device->bdev;
1881         if (tgtdev->bdev == fs_info->fs_devices->latest_bdev)
1882                 fs_info->fs_devices->latest_bdev = next_device->bdev;
1883         list_del_rcu(&tgtdev->dev_list);
1884
1885         call_rcu(&tgtdev->rcu, free_device);
1886
1887         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
1888 }
1889
1890 static int btrfs_find_device_by_path(struct btrfs_root *root, char *device_path,
1891                                      struct btrfs_device **device)
1892 {
1893         int ret = 0;
1894         struct btrfs_super_block *disk_super;
1895         u64 devid;
1896         u8 *dev_uuid;
1897         struct block_device *bdev;
1898         struct buffer_head *bh;
1899
1900         *device = NULL;
1901         ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
1902                                     root->fs_info->bdev_holder, 0, &bdev, &bh);
1903         if (ret)
1904                 return ret;
1905         disk_super = (struct btrfs_super_block *)bh->b_data;
1906         devid = btrfs_stack_device_id(&disk_super->dev_item);
1907         dev_uuid = disk_super->dev_item.uuid;
1908         *device = btrfs_find_device(root->fs_info, devid, dev_uuid,
1909                                     disk_super->fsid);
1910         brelse(bh);
1911         if (!*device)
1912                 ret = -ENOENT;
1913         blkdev_put(bdev, FMODE_READ);
1914         return ret;
1915 }
1916
1917 int btrfs_find_device_missing_or_by_path(struct btrfs_root *root,
1918                                          char *device_path,
1919                                          struct btrfs_device **device)
1920 {
1921         *device = NULL;
1922         if (strcmp(device_path, "missing") == 0) {
1923                 struct list_head *devices;
1924                 struct btrfs_device *tmp;
1925
1926                 devices = &root->fs_info->fs_devices->devices;
1927                 /*
1928                  * It is safe to read the devices since the volume_mutex
1929                  * is held by the caller.
1930                  */
1931                 list_for_each_entry(tmp, devices, dev_list) {
1932                         if (tmp->in_fs_metadata && !tmp->bdev) {
1933                                 *device = tmp;
1934                                 break;
1935                         }
1936                 }
1937
1938                 if (!*device) {
1939                         btrfs_err(root->fs_info, "no missing device found");
1940                         return -ENOENT;
1941                 }
1942
1943                 return 0;
1944         } else {
1945                 return btrfs_find_device_by_path(root, device_path, device);
1946         }
1947 }
1948
1949 /*
1950  * does all the dirty work required for changing file system's UUID.
1951  */
1952 static int btrfs_prepare_sprout(struct btrfs_root *root)
1953 {
1954         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
1955         struct btrfs_fs_devices *old_devices;
1956         struct btrfs_fs_devices *seed_devices;
1957         struct btrfs_super_block *disk_super = root->fs_info->super_copy;
1958         struct btrfs_device *device;
1959         u64 super_flags;
1960
1961         BUG_ON(!mutex_is_locked(&uuid_mutex));
1962         if (!fs_devices->seeding)
1963                 return -EINVAL;
1964
1965         seed_devices = __alloc_fs_devices();
1966         if (IS_ERR(seed_devices))
1967                 return PTR_ERR(seed_devices);
1968
1969         old_devices = clone_fs_devices(fs_devices);
1970         if (IS_ERR(old_devices)) {
1971                 kfree(seed_devices);
1972                 return PTR_ERR(old_devices);
1973         }
1974
1975         list_add(&old_devices->list, &fs_uuids);
1976
1977         memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
1978         seed_devices->opened = 1;
1979         INIT_LIST_HEAD(&seed_devices->devices);
1980         INIT_LIST_HEAD(&seed_devices->alloc_list);
1981         mutex_init(&seed_devices->device_list_mutex);
1982
1983         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1984         list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
1985                               synchronize_rcu);
1986
1987         list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
1988         list_for_each_entry(device, &seed_devices->devices, dev_list) {
1989                 device->fs_devices = seed_devices;
1990         }
1991
1992         fs_devices->seeding = 0;
1993         fs_devices->num_devices = 0;
1994         fs_devices->open_devices = 0;
1995         fs_devices->missing_devices = 0;
1996         fs_devices->num_can_discard = 0;
1997         fs_devices->rotating = 0;
1998         fs_devices->seed = seed_devices;
1999
2000         generate_random_uuid(fs_devices->fsid);
2001         memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2002         memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2003         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2004
2005         super_flags = btrfs_super_flags(disk_super) &
2006                       ~BTRFS_SUPER_FLAG_SEEDING;
2007         btrfs_set_super_flags(disk_super, super_flags);
2008
2009         return 0;
2010 }
2011
2012 /*
2013  * strore the expected generation for seed devices in device items.
2014  */
2015 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
2016                                struct btrfs_root *root)
2017 {
2018         struct btrfs_path *path;
2019         struct extent_buffer *leaf;
2020         struct btrfs_dev_item *dev_item;
2021         struct btrfs_device *device;
2022         struct btrfs_key key;
2023         u8 fs_uuid[BTRFS_UUID_SIZE];
2024         u8 dev_uuid[BTRFS_UUID_SIZE];
2025         u64 devid;
2026         int ret;
2027
2028         path = btrfs_alloc_path();
2029         if (!path)
2030                 return -ENOMEM;
2031
2032         root = root->fs_info->chunk_root;
2033         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2034         key.offset = 0;
2035         key.type = BTRFS_DEV_ITEM_KEY;
2036
2037         while (1) {
2038                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2039                 if (ret < 0)
2040                         goto error;
2041
2042                 leaf = path->nodes[0];
2043 next_slot:
2044                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2045                         ret = btrfs_next_leaf(root, path);
2046                         if (ret > 0)
2047                                 break;
2048                         if (ret < 0)
2049                                 goto error;
2050                         leaf = path->nodes[0];
2051                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2052                         btrfs_release_path(path);
2053                         continue;
2054                 }
2055
2056                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2057                 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
2058                     key.type != BTRFS_DEV_ITEM_KEY)
2059                         break;
2060
2061                 dev_item = btrfs_item_ptr(leaf, path->slots[0],
2062                                           struct btrfs_dev_item);
2063                 devid = btrfs_device_id(leaf, dev_item);
2064                 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
2065                                    BTRFS_UUID_SIZE);
2066                 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
2067                                    BTRFS_UUID_SIZE);
2068                 device = btrfs_find_device(root->fs_info, devid, dev_uuid,
2069                                            fs_uuid);
2070                 BUG_ON(!device); /* Logic error */
2071
2072                 if (device->fs_devices->seeding) {
2073                         btrfs_set_device_generation(leaf, dev_item,
2074                                                     device->generation);
2075                         btrfs_mark_buffer_dirty(leaf);
2076                 }
2077
2078                 path->slots[0]++;
2079                 goto next_slot;
2080         }
2081         ret = 0;
2082 error:
2083         btrfs_free_path(path);
2084         return ret;
2085 }
2086
2087 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
2088 {
2089         struct request_queue *q;
2090         struct btrfs_trans_handle *trans;
2091         struct btrfs_device *device;
2092         struct block_device *bdev;
2093         struct list_head *devices;
2094         struct super_block *sb = root->fs_info->sb;
2095         struct rcu_string *name;
2096         u64 total_bytes;
2097         int seeding_dev = 0;
2098         int ret = 0;
2099
2100         if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
2101                 return -EROFS;
2102
2103         bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2104                                   root->fs_info->bdev_holder);
2105         if (IS_ERR(bdev))
2106                 return PTR_ERR(bdev);
2107
2108         if (root->fs_info->fs_devices->seeding) {
2109                 seeding_dev = 1;
2110                 down_write(&sb->s_umount);
2111                 mutex_lock(&uuid_mutex);
2112         }
2113
2114         filemap_write_and_wait(bdev->bd_inode->i_mapping);
2115
2116         devices = &root->fs_info->fs_devices->devices;
2117
2118         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2119         list_for_each_entry(device, devices, dev_list) {
2120                 if (device->bdev == bdev) {
2121                         ret = -EEXIST;
2122                         mutex_unlock(
2123                                 &root->fs_info->fs_devices->device_list_mutex);
2124                         goto error;
2125                 }
2126         }
2127         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2128
2129         device = btrfs_alloc_device(root->fs_info, NULL, NULL);
2130         if (IS_ERR(device)) {
2131                 /* we can safely leave the fs_devices entry around */
2132                 ret = PTR_ERR(device);
2133                 goto error;
2134         }
2135
2136         name = rcu_string_strdup(device_path, GFP_NOFS);
2137         if (!name) {
2138                 kfree(device);
2139                 ret = -ENOMEM;
2140                 goto error;
2141         }
2142         rcu_assign_pointer(device->name, name);
2143
2144         trans = btrfs_start_transaction(root, 0);
2145         if (IS_ERR(trans)) {
2146                 rcu_string_free(device->name);
2147                 kfree(device);
2148                 ret = PTR_ERR(trans);
2149                 goto error;
2150         }
2151
2152         lock_chunks(root);
2153
2154         q = bdev_get_queue(bdev);
2155         if (blk_queue_discard(q))
2156                 device->can_discard = 1;
2157         device->writeable = 1;
2158         device->generation = trans->transid;
2159         device->io_width = root->sectorsize;
2160         device->io_align = root->sectorsize;
2161         device->sector_size = root->sectorsize;
2162         device->total_bytes = i_size_read(bdev->bd_inode);
2163         device->disk_total_bytes = device->total_bytes;
2164         device->dev_root = root->fs_info->dev_root;
2165         device->bdev = bdev;
2166         device->in_fs_metadata = 1;
2167         device->is_tgtdev_for_dev_replace = 0;
2168         device->mode = FMODE_EXCL;
2169         device->dev_stats_valid = 1;
2170         set_blocksize(device->bdev, 4096);
2171
2172         if (seeding_dev) {
2173                 sb->s_flags &= ~MS_RDONLY;
2174                 ret = btrfs_prepare_sprout(root);
2175                 BUG_ON(ret); /* -ENOMEM */
2176         }
2177
2178         device->fs_devices = root->fs_info->fs_devices;
2179
2180         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2181         list_add_rcu(&device->dev_list, &root->fs_info->fs_devices->devices);
2182         list_add(&device->dev_alloc_list,
2183                  &root->fs_info->fs_devices->alloc_list);
2184         root->fs_info->fs_devices->num_devices++;
2185         root->fs_info->fs_devices->open_devices++;
2186         root->fs_info->fs_devices->rw_devices++;
2187         root->fs_info->fs_devices->total_devices++;
2188         if (device->can_discard)
2189                 root->fs_info->fs_devices->num_can_discard++;
2190         root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
2191
2192         spin_lock(&root->fs_info->free_chunk_lock);
2193         root->fs_info->free_chunk_space += device->total_bytes;
2194         spin_unlock(&root->fs_info->free_chunk_lock);
2195
2196         if (!blk_queue_nonrot(bdev_get_queue(bdev)))
2197                 root->fs_info->fs_devices->rotating = 1;
2198
2199         total_bytes = btrfs_super_total_bytes(root->fs_info->super_copy);
2200         btrfs_set_super_total_bytes(root->fs_info->super_copy,
2201                                     total_bytes + device->total_bytes);
2202
2203         total_bytes = btrfs_super_num_devices(root->fs_info->super_copy);
2204         btrfs_set_super_num_devices(root->fs_info->super_copy,
2205                                     total_bytes + 1);
2206
2207         /* add sysfs device entry */
2208         btrfs_kobj_add_device(root->fs_info, device);
2209
2210         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2211
2212         if (seeding_dev) {
2213                 char fsid_buf[BTRFS_UUID_UNPARSED_SIZE];
2214                 ret = init_first_rw_device(trans, root, device);
2215                 if (ret) {
2216                         btrfs_abort_transaction(trans, root, ret);
2217                         goto error_trans;
2218                 }
2219                 ret = btrfs_finish_sprout(trans, root);
2220                 if (ret) {
2221                         btrfs_abort_transaction(trans, root, ret);
2222                         goto error_trans;
2223                 }
2224
2225                 /* Sprouting would change fsid of the mounted root,
2226                  * so rename the fsid on the sysfs
2227                  */
2228                 snprintf(fsid_buf, BTRFS_UUID_UNPARSED_SIZE, "%pU",
2229                                                 root->fs_info->fsid);
2230                 if (kobject_rename(&root->fs_info->super_kobj, fsid_buf))
2231                         goto error_trans;
2232         } else {
2233                 ret = btrfs_add_device(trans, root, device);
2234                 if (ret) {
2235                         btrfs_abort_transaction(trans, root, ret);
2236                         goto error_trans;
2237                 }
2238         }
2239
2240         /*
2241          * we've got more storage, clear any full flags on the space
2242          * infos
2243          */
2244         btrfs_clear_space_info_full(root->fs_info);
2245
2246         unlock_chunks(root);
2247         root->fs_info->num_tolerated_disk_barrier_failures =
2248                 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
2249         ret = btrfs_commit_transaction(trans, root);
2250
2251         if (seeding_dev) {
2252                 mutex_unlock(&uuid_mutex);
2253                 up_write(&sb->s_umount);
2254
2255                 if (ret) /* transaction commit */
2256                         return ret;
2257
2258                 ret = btrfs_relocate_sys_chunks(root);
2259                 if (ret < 0)
2260                         btrfs_error(root->fs_info, ret,
2261                                     "Failed to relocate sys chunks after "
2262                                     "device initialization. This can be fixed "
2263                                     "using the \"btrfs balance\" command.");
2264                 trans = btrfs_attach_transaction(root);
2265                 if (IS_ERR(trans)) {
2266                         if (PTR_ERR(trans) == -ENOENT)
2267                                 return 0;
2268                         return PTR_ERR(trans);
2269                 }
2270                 ret = btrfs_commit_transaction(trans, root);
2271         }
2272
2273         /* Update ctime/mtime for libblkid */
2274         update_dev_time(device_path);
2275         return ret;
2276
2277 error_trans:
2278         unlock_chunks(root);
2279         btrfs_end_transaction(trans, root);
2280         rcu_string_free(device->name);
2281         btrfs_kobj_rm_device(root->fs_info, device);
2282         kfree(device);
2283 error:
2284         blkdev_put(bdev, FMODE_EXCL);
2285         if (seeding_dev) {
2286                 mutex_unlock(&uuid_mutex);
2287                 up_write(&sb->s_umount);
2288         }
2289         return ret;
2290 }
2291
2292 int btrfs_init_dev_replace_tgtdev(struct btrfs_root *root, char *device_path,
2293                                   struct btrfs_device **device_out)
2294 {
2295         struct request_queue *q;
2296         struct btrfs_device *device;
2297         struct block_device *bdev;
2298         struct btrfs_fs_info *fs_info = root->fs_info;
2299         struct list_head *devices;
2300         struct rcu_string *name;
2301         u64 devid = BTRFS_DEV_REPLACE_DEVID;
2302         int ret = 0;
2303
2304         *device_out = NULL;
2305         if (fs_info->fs_devices->seeding)
2306                 return -EINVAL;
2307
2308         bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2309                                   fs_info->bdev_holder);
2310         if (IS_ERR(bdev))
2311                 return PTR_ERR(bdev);
2312
2313         filemap_write_and_wait(bdev->bd_inode->i_mapping);
2314
2315         devices = &fs_info->fs_devices->devices;
2316         list_for_each_entry(device, devices, dev_list) {
2317                 if (device->bdev == bdev) {
2318                         ret = -EEXIST;
2319                         goto error;
2320                 }
2321         }
2322
2323         device = btrfs_alloc_device(NULL, &devid, NULL);
2324         if (IS_ERR(device)) {
2325                 ret = PTR_ERR(device);
2326                 goto error;
2327         }
2328
2329         name = rcu_string_strdup(device_path, GFP_NOFS);
2330         if (!name) {
2331                 kfree(device);
2332                 ret = -ENOMEM;
2333                 goto error;
2334         }
2335         rcu_assign_pointer(device->name, name);
2336
2337         q = bdev_get_queue(bdev);
2338         if (blk_queue_discard(q))
2339                 device->can_discard = 1;
2340         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2341         device->writeable = 1;
2342         device->generation = 0;
2343         device->io_width = root->sectorsize;
2344         device->io_align = root->sectorsize;
2345         device->sector_size = root->sectorsize;
2346         device->total_bytes = i_size_read(bdev->bd_inode);
2347         device->disk_total_bytes = device->total_bytes;
2348         device->dev_root = fs_info->dev_root;
2349         device->bdev = bdev;
2350         device->in_fs_metadata = 1;
2351         device->is_tgtdev_for_dev_replace = 1;
2352         device->mode = FMODE_EXCL;
2353         device->dev_stats_valid = 1;
2354         set_blocksize(device->bdev, 4096);
2355         device->fs_devices = fs_info->fs_devices;
2356         list_add(&device->dev_list, &fs_info->fs_devices->devices);
2357         fs_info->fs_devices->num_devices++;
2358         fs_info->fs_devices->open_devices++;
2359         if (device->can_discard)
2360                 fs_info->fs_devices->num_can_discard++;
2361         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2362
2363         *device_out = device;
2364         return ret;
2365
2366 error:
2367         blkdev_put(bdev, FMODE_EXCL);
2368         return ret;
2369 }
2370
2371 void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info *fs_info,
2372                                               struct btrfs_device *tgtdev)
2373 {
2374         WARN_ON(fs_info->fs_devices->rw_devices == 0);
2375         tgtdev->io_width = fs_info->dev_root->sectorsize;
2376         tgtdev->io_align = fs_info->dev_root->sectorsize;
2377         tgtdev->sector_size = fs_info->dev_root->sectorsize;
2378         tgtdev->dev_root = fs_info->dev_root;
2379         tgtdev->in_fs_metadata = 1;
2380 }
2381
2382 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2383                                         struct btrfs_device *device)
2384 {
2385         int ret;
2386         struct btrfs_path *path;
2387         struct btrfs_root *root;
2388         struct btrfs_dev_item *dev_item;
2389         struct extent_buffer *leaf;
2390         struct btrfs_key key;
2391
2392         root = device->dev_root->fs_info->chunk_root;
2393
2394         path = btrfs_alloc_path();
2395         if (!path)
2396                 return -ENOMEM;
2397
2398         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2399         key.type = BTRFS_DEV_ITEM_KEY;
2400         key.offset = device->devid;
2401
2402         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2403         if (ret < 0)
2404                 goto out;
2405
2406         if (ret > 0) {
2407                 ret = -ENOENT;
2408                 goto out;
2409         }
2410
2411         leaf = path->nodes[0];
2412         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2413
2414         btrfs_set_device_id(leaf, dev_item, device->devid);
2415         btrfs_set_device_type(leaf, dev_item, device->type);
2416         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2417         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2418         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2419         btrfs_set_device_total_bytes(leaf, dev_item, device->disk_total_bytes);
2420         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
2421         btrfs_mark_buffer_dirty(leaf);
2422
2423 out:
2424         btrfs_free_path(path);
2425         return ret;
2426 }
2427
2428 static int __btrfs_grow_device(struct btrfs_trans_handle *trans,
2429                       struct btrfs_device *device, u64 new_size)
2430 {
2431         struct btrfs_super_block *super_copy =
2432                 device->dev_root->fs_info->super_copy;
2433         u64 old_total = btrfs_super_total_bytes(super_copy);
2434         u64 diff = new_size - device->total_bytes;
2435
2436         if (!device->writeable)
2437                 return -EACCES;
2438         if (new_size <= device->total_bytes ||
2439             device->is_tgtdev_for_dev_replace)
2440                 return -EINVAL;
2441
2442         btrfs_set_super_total_bytes(super_copy, old_total + diff);
2443         device->fs_devices->total_rw_bytes += diff;
2444
2445         device->total_bytes = new_size;
2446         device->disk_total_bytes = new_size;
2447         btrfs_clear_space_info_full(device->dev_root->fs_info);
2448
2449         return btrfs_update_device(trans, device);
2450 }
2451
2452 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2453                       struct btrfs_device *device, u64 new_size)
2454 {
2455         int ret;
2456         lock_chunks(device->dev_root);
2457         ret = __btrfs_grow_device(trans, device, new_size);
2458         unlock_chunks(device->dev_root);
2459         return ret;
2460 }
2461
2462 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
2463                             struct btrfs_root *root,
2464                             u64 chunk_tree, u64 chunk_objectid,
2465                             u64 chunk_offset)
2466 {
2467         int ret;
2468         struct btrfs_path *path;
2469         struct btrfs_key key;
2470
2471         root = root->fs_info->chunk_root;
2472         path = btrfs_alloc_path();
2473         if (!path)
2474                 return -ENOMEM;
2475
2476         key.objectid = chunk_objectid;
2477         key.offset = chunk_offset;
2478         key.type = BTRFS_CHUNK_ITEM_KEY;
2479
2480         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2481         if (ret < 0)
2482                 goto out;
2483         else if (ret > 0) { /* Logic error or corruption */
2484                 btrfs_error(root->fs_info, -ENOENT,
2485                             "Failed lookup while freeing chunk.");
2486                 ret = -ENOENT;
2487                 goto out;
2488         }
2489
2490         ret = btrfs_del_item(trans, root, path);
2491         if (ret < 0)
2492                 btrfs_error(root->fs_info, ret,
2493                             "Failed to delete chunk item.");
2494 out:
2495         btrfs_free_path(path);
2496         return ret;
2497 }
2498
2499 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
2500                         chunk_offset)
2501 {
2502         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
2503         struct btrfs_disk_key *disk_key;
2504         struct btrfs_chunk *chunk;
2505         u8 *ptr;
2506         int ret = 0;
2507         u32 num_stripes;
2508         u32 array_size;
2509         u32 len = 0;
2510         u32 cur;
2511         struct btrfs_key key;
2512
2513         array_size = btrfs_super_sys_array_size(super_copy);
2514
2515         ptr = super_copy->sys_chunk_array;
2516         cur = 0;
2517
2518         while (cur < array_size) {
2519                 disk_key = (struct btrfs_disk_key *)ptr;
2520                 btrfs_disk_key_to_cpu(&key, disk_key);
2521
2522                 len = sizeof(*disk_key);
2523
2524                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2525                         chunk = (struct btrfs_chunk *)(ptr + len);
2526                         num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2527                         len += btrfs_chunk_item_size(num_stripes);
2528                 } else {
2529                         ret = -EIO;
2530                         break;
2531                 }
2532                 if (key.objectid == chunk_objectid &&
2533                     key.offset == chunk_offset) {
2534                         memmove(ptr, ptr + len, array_size - (cur + len));
2535                         array_size -= len;
2536                         btrfs_set_super_sys_array_size(super_copy, array_size);
2537                 } else {
2538                         ptr += len;
2539                         cur += len;
2540                 }
2541         }
2542         return ret;
2543 }
2544
2545 static int btrfs_relocate_chunk(struct btrfs_root *root,
2546                          u64 chunk_tree, u64 chunk_objectid,
2547                          u64 chunk_offset)
2548 {
2549         struct extent_map_tree *em_tree;
2550         struct btrfs_root *extent_root;
2551         struct btrfs_trans_handle *trans;
2552         struct extent_map *em;
2553         struct map_lookup *map;
2554         int ret;
2555         int i;
2556
2557         root = root->fs_info->chunk_root;
2558         extent_root = root->fs_info->extent_root;
2559         em_tree = &root->fs_info->mapping_tree.map_tree;
2560
2561         ret = btrfs_can_relocate(extent_root, chunk_offset);
2562         if (ret)
2563                 return -ENOSPC;
2564
2565         /* step one, relocate all the extents inside this chunk */
2566         ret = btrfs_relocate_block_group(extent_root, chunk_offset);
2567         if (ret)
2568                 return ret;
2569
2570         trans = btrfs_start_transaction(root, 0);
2571         if (IS_ERR(trans)) {
2572                 ret = PTR_ERR(trans);
2573                 btrfs_std_error(root->fs_info, ret);
2574                 return ret;
2575         }
2576
2577         lock_chunks(root);
2578
2579         /*
2580          * step two, delete the device extents and the
2581          * chunk tree entries
2582          */
2583         read_lock(&em_tree->lock);
2584         em = lookup_extent_mapping(em_tree, chunk_offset, 1);
2585         read_unlock(&em_tree->lock);
2586
2587         BUG_ON(!em || em->start > chunk_offset ||
2588                em->start + em->len < chunk_offset);
2589         map = (struct map_lookup *)em->bdev;
2590
2591         for (i = 0; i < map->num_stripes; i++) {
2592                 ret = btrfs_free_dev_extent(trans, map->stripes[i].dev,
2593                                             map->stripes[i].physical);
2594                 BUG_ON(ret);
2595
2596                 if (map->stripes[i].dev) {
2597                         ret = btrfs_update_device(trans, map->stripes[i].dev);
2598                         BUG_ON(ret);
2599                 }
2600         }
2601         ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
2602                                chunk_offset);
2603
2604         BUG_ON(ret);
2605
2606         trace_btrfs_chunk_free(root, map, chunk_offset, em->len);
2607
2608         if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2609                 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
2610                 BUG_ON(ret);
2611         }
2612
2613         ret = btrfs_remove_block_group(trans, extent_root, chunk_offset);
2614         BUG_ON(ret);
2615
2616         write_lock(&em_tree->lock);
2617         remove_extent_mapping(em_tree, em);
2618         write_unlock(&em_tree->lock);
2619
2620         /* once for the tree */
2621         free_extent_map(em);
2622         /* once for us */
2623         free_extent_map(em);
2624
2625         unlock_chunks(root);
2626         btrfs_end_transaction(trans, root);
2627         return 0;
2628 }
2629
2630 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
2631 {
2632         struct btrfs_root *chunk_root = root->fs_info->chunk_root;
2633         struct btrfs_path *path;
2634         struct extent_buffer *leaf;
2635         struct btrfs_chunk *chunk;
2636         struct btrfs_key key;
2637         struct btrfs_key found_key;
2638         u64 chunk_tree = chunk_root->root_key.objectid;
2639         u64 chunk_type;
2640         bool retried = false;
2641         int failed = 0;
2642         int ret;
2643
2644         path = btrfs_alloc_path();
2645         if (!path)
2646                 return -ENOMEM;
2647
2648 again:
2649         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2650         key.offset = (u64)-1;
2651         key.type = BTRFS_CHUNK_ITEM_KEY;
2652
2653         while (1) {
2654                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2655                 if (ret < 0)
2656                         goto error;
2657                 BUG_ON(ret == 0); /* Corruption */
2658
2659                 ret = btrfs_previous_item(chunk_root, path, key.objectid,
2660                                           key.type);
2661                 if (ret < 0)
2662                         goto error;
2663                 if (ret > 0)
2664                         break;
2665
2666                 leaf = path->nodes[0];
2667                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2668
2669                 chunk = btrfs_item_ptr(leaf, path->slots[0],
2670                                        struct btrfs_chunk);
2671                 chunk_type = btrfs_chunk_type(leaf, chunk);
2672                 btrfs_release_path(path);
2673
2674                 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
2675                         ret = btrfs_relocate_chunk(chunk_root, chunk_tree,
2676                                                    found_key.objectid,
2677                                                    found_key.offset);
2678                         if (ret == -ENOSPC)
2679                                 failed++;
2680                         else if (ret)
2681                                 BUG();
2682                 }
2683
2684                 if (found_key.offset == 0)
2685                         break;
2686                 key.offset = found_key.offset - 1;
2687         }
2688         ret = 0;
2689         if (failed && !retried) {
2690                 failed = 0;
2691                 retried = true;
2692                 goto again;
2693         } else if (WARN_ON(failed && retried)) {
2694                 ret = -ENOSPC;
2695         }
2696 error:
2697         btrfs_free_path(path);
2698         return ret;
2699 }
2700
2701 static int insert_balance_item(struct btrfs_root *root,
2702                                struct btrfs_balance_control *bctl)
2703 {
2704         struct btrfs_trans_handle *trans;
2705         struct btrfs_balance_item *item;
2706         struct btrfs_disk_balance_args disk_bargs;
2707         struct btrfs_path *path;
2708         struct extent_buffer *leaf;
2709         struct btrfs_key key;
2710         int ret, err;
2711
2712         path = btrfs_alloc_path();
2713         if (!path)
2714                 return -ENOMEM;
2715
2716         trans = btrfs_start_transaction(root, 0);
2717         if (IS_ERR(trans)) {
2718                 btrfs_free_path(path);
2719                 return PTR_ERR(trans);
2720         }
2721
2722         key.objectid = BTRFS_BALANCE_OBJECTID;
2723         key.type = BTRFS_BALANCE_ITEM_KEY;
2724         key.offset = 0;
2725
2726         ret = btrfs_insert_empty_item(trans, root, path, &key,
2727                                       sizeof(*item));
2728         if (ret)
2729                 goto out;
2730
2731         leaf = path->nodes[0];
2732         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
2733
2734         memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
2735
2736         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
2737         btrfs_set_balance_data(leaf, item, &disk_bargs);
2738         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
2739         btrfs_set_balance_meta(leaf, item, &disk_bargs);
2740         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
2741         btrfs_set_balance_sys(leaf, item, &disk_bargs);
2742
2743         btrfs_set_balance_flags(leaf, item, bctl->flags);
2744
2745         btrfs_mark_buffer_dirty(leaf);
2746 out:
2747         btrfs_free_path(path);
2748         err = btrfs_commit_transaction(trans, root);
2749         if (err && !ret)
2750                 ret = err;
2751         return ret;
2752 }
2753
2754 static int del_balance_item(struct btrfs_root *root)
2755 {
2756         struct btrfs_trans_handle *trans;
2757         struct btrfs_path *path;
2758         struct btrfs_key key;
2759         int ret, err;
2760
2761         path = btrfs_alloc_path();
2762         if (!path)
2763                 return -ENOMEM;
2764
2765         trans = btrfs_start_transaction(root, 0);
2766         if (IS_ERR(trans)) {
2767                 btrfs_free_path(path);
2768                 return PTR_ERR(trans);
2769         }
2770
2771         key.objectid = BTRFS_BALANCE_OBJECTID;
2772         key.type = BTRFS_BALANCE_ITEM_KEY;
2773         key.offset = 0;
2774
2775         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2776         if (ret < 0)
2777                 goto out;
2778         if (ret > 0) {
2779                 ret = -ENOENT;
2780                 goto out;
2781         }
2782
2783         ret = btrfs_del_item(trans, root, path);
2784 out:
2785         btrfs_free_path(path);
2786         err = btrfs_commit_transaction(trans, root);
2787         if (err && !ret)
2788                 ret = err;
2789         return ret;
2790 }
2791
2792 /*
2793  * This is a heuristic used to reduce the number of chunks balanced on
2794  * resume after balance was interrupted.
2795  */
2796 static void update_balance_args(struct btrfs_balance_control *bctl)
2797 {
2798         /*
2799          * Turn on soft mode for chunk types that were being converted.
2800          */
2801         if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
2802                 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
2803         if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
2804                 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
2805         if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
2806                 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
2807
2808         /*
2809          * Turn on usage filter if is not already used.  The idea is
2810          * that chunks that we have already balanced should be
2811          * reasonably full.  Don't do it for chunks that are being
2812          * converted - that will keep us from relocating unconverted
2813          * (albeit full) chunks.
2814          */
2815         if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2816             !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2817                 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
2818                 bctl->data.usage = 90;
2819         }
2820         if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2821             !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2822                 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
2823                 bctl->sys.usage = 90;
2824         }
2825         if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2826             !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2827                 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
2828                 bctl->meta.usage = 90;
2829         }
2830 }
2831
2832 /*
2833  * Should be called with both balance and volume mutexes held to
2834  * serialize other volume operations (add_dev/rm_dev/resize) with
2835  * restriper.  Same goes for unset_balance_control.
2836  */
2837 static void set_balance_control(struct btrfs_balance_control *bctl)
2838 {
2839         struct btrfs_fs_info *fs_info = bctl->fs_info;
2840
2841         BUG_ON(fs_info->balance_ctl);
2842
2843         spin_lock(&fs_info->balance_lock);
2844         fs_info->balance_ctl = bctl;
2845         spin_unlock(&fs_info->balance_lock);
2846 }
2847
2848 static void unset_balance_control(struct btrfs_fs_info *fs_info)
2849 {
2850         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
2851
2852         BUG_ON(!fs_info->balance_ctl);
2853
2854         spin_lock(&fs_info->balance_lock);
2855         fs_info->balance_ctl = NULL;
2856         spin_unlock(&fs_info->balance_lock);
2857
2858         kfree(bctl);
2859 }
2860
2861 /*
2862  * Balance filters.  Return 1 if chunk should be filtered out
2863  * (should not be balanced).
2864  */
2865 static int chunk_profiles_filter(u64 chunk_type,
2866                                  struct btrfs_balance_args *bargs)
2867 {
2868         chunk_type = chunk_to_extended(chunk_type) &
2869                                 BTRFS_EXTENDED_PROFILE_MASK;
2870
2871         if (bargs->profiles & chunk_type)
2872                 return 0;
2873
2874         return 1;
2875 }
2876
2877 static int chunk_usage_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
2878                               struct btrfs_balance_args *bargs)
2879 {
2880         struct btrfs_block_group_cache *cache;
2881         u64 chunk_used, user_thresh;
2882         int ret = 1;
2883
2884         cache = btrfs_lookup_block_group(fs_info, chunk_offset);
2885         chunk_used = btrfs_block_group_used(&cache->item);
2886
2887         if (bargs->usage == 0)
2888                 user_thresh = 1;
2889         else if (bargs->usage > 100)
2890                 user_thresh = cache->key.offset;
2891         else
2892                 user_thresh = div_factor_fine(cache->key.offset,
2893                                               bargs->usage);
2894
2895         if (chunk_used < user_thresh)
2896                 ret = 0;
2897
2898         btrfs_put_block_group(cache);
2899         return ret;
2900 }
2901
2902 static int chunk_devid_filter(struct extent_buffer *leaf,
2903                               struct btrfs_chunk *chunk,
2904                               struct btrfs_balance_args *bargs)
2905 {
2906         struct btrfs_stripe *stripe;
2907         int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2908         int i;
2909
2910         for (i = 0; i < num_stripes; i++) {
2911                 stripe = btrfs_stripe_nr(chunk, i);
2912                 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
2913                         return 0;
2914         }
2915
2916         return 1;
2917 }
2918
2919 /* [pstart, pend) */
2920 static int chunk_drange_filter(struct extent_buffer *leaf,
2921                                struct btrfs_chunk *chunk,
2922                                u64 chunk_offset,
2923                                struct btrfs_balance_args *bargs)
2924 {
2925         struct btrfs_stripe *stripe;
2926         int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2927         u64 stripe_offset;
2928         u64 stripe_length;
2929         int factor;
2930         int i;
2931
2932         if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
2933                 return 0;
2934
2935         if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP |
2936              BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) {
2937                 factor = num_stripes / 2;
2938         } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) {
2939                 factor = num_stripes - 1;
2940         } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) {
2941                 factor = num_stripes - 2;
2942         } else {
2943                 factor = num_stripes;
2944         }
2945
2946         for (i = 0; i < num_stripes; i++) {
2947                 stripe = btrfs_stripe_nr(chunk, i);
2948                 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
2949                         continue;
2950
2951                 stripe_offset = btrfs_stripe_offset(leaf, stripe);
2952                 stripe_length = btrfs_chunk_length(leaf, chunk);
2953                 do_div(stripe_length, factor);
2954
2955                 if (stripe_offset < bargs->pend &&
2956                     stripe_offset + stripe_length > bargs->pstart)
2957                         return 0;
2958         }
2959
2960         return 1;
2961 }
2962
2963 /* [vstart, vend) */
2964 static int chunk_vrange_filter(struct extent_buffer *leaf,
2965                                struct btrfs_chunk *chunk,
2966                                u64 chunk_offset,
2967                                struct btrfs_balance_args *bargs)
2968 {
2969         if (chunk_offset < bargs->vend &&
2970             chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
2971                 /* at least part of the chunk is inside this vrange */
2972                 return 0;
2973
2974         return 1;
2975 }
2976
2977 static int chunk_soft_convert_filter(u64 chunk_type,
2978                                      struct btrfs_balance_args *bargs)
2979 {
2980         if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
2981                 return 0;
2982
2983         chunk_type = chunk_to_extended(chunk_type) &
2984                                 BTRFS_EXTENDED_PROFILE_MASK;
2985
2986         if (bargs->target == chunk_type)
2987                 return 1;
2988
2989         return 0;
2990 }
2991
2992 static int should_balance_chunk(struct btrfs_root *root,
2993                                 struct extent_buffer *leaf,
2994                                 struct btrfs_chunk *chunk, u64 chunk_offset)
2995 {
2996         struct btrfs_balance_control *bctl = root->fs_info->balance_ctl;
2997         struct btrfs_balance_args *bargs = NULL;
2998         u64 chunk_type = btrfs_chunk_type(leaf, chunk);
2999
3000         /* type filter */
3001         if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
3002               (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
3003                 return 0;
3004         }
3005
3006         if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3007                 bargs = &bctl->data;
3008         else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3009                 bargs = &bctl->sys;
3010         else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3011                 bargs = &bctl->meta;
3012
3013         /* profiles filter */
3014         if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
3015             chunk_profiles_filter(chunk_type, bargs)) {
3016                 return 0;
3017         }
3018
3019         /* usage filter */
3020         if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
3021             chunk_usage_filter(bctl->fs_info, chunk_offset, bargs)) {
3022                 return 0;
3023         }
3024
3025         /* devid filter */
3026         if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
3027             chunk_devid_filter(leaf, chunk, bargs)) {
3028                 return 0;
3029         }
3030
3031         /* drange filter, makes sense only with devid filter */
3032         if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
3033             chunk_drange_filter(leaf, chunk, chunk_offset, bargs)) {
3034                 return 0;
3035         }
3036
3037         /* vrange filter */
3038         if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
3039             chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
3040                 return 0;
3041         }
3042
3043         /* soft profile changing mode */
3044         if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
3045             chunk_soft_convert_filter(chunk_type, bargs)) {
3046                 return 0;
3047         }
3048
3049         /*
3050          * limited by count, must be the last filter
3051          */
3052         if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) {
3053                 if (bargs->limit == 0)
3054                         return 0;
3055                 else
3056                         bargs->limit--;
3057         }
3058
3059         return 1;
3060 }
3061
3062 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
3063 {
3064         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3065         struct btrfs_root *chunk_root = fs_info->chunk_root;
3066         struct btrfs_root *dev_root = fs_info->dev_root;
3067         struct list_head *devices;
3068         struct btrfs_device *device;
3069         u64 old_size;
3070         u64 size_to_free;
3071         struct btrfs_chunk *chunk;
3072         struct btrfs_path *path;
3073         struct btrfs_key key;
3074         struct btrfs_key found_key;
3075         struct btrfs_trans_handle *trans;
3076         struct extent_buffer *leaf;
3077         int slot;
3078         int ret;
3079         int enospc_errors = 0;
3080         bool counting = true;
3081         u64 limit_data = bctl->data.limit;
3082         u64 limit_meta = bctl->meta.limit;
3083         u64 limit_sys = bctl->sys.limit;
3084
3085         /* step one make some room on all the devices */
3086         devices = &fs_info->fs_devices->devices;
3087         list_for_each_entry(device, devices, dev_list) {
3088                 old_size = device->total_bytes;
3089                 size_to_free = div_factor(old_size, 1);
3090                 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
3091                 if (!device->writeable ||
3092                     device->total_bytes - device->bytes_used > size_to_free ||
3093                     device->is_tgtdev_for_dev_replace)
3094                         continue;
3095
3096                 ret = btrfs_shrink_device(device, old_size - size_to_free);
3097                 if (ret == -ENOSPC)
3098                         break;
3099                 BUG_ON(ret);
3100
3101                 trans = btrfs_start_transaction(dev_root, 0);
3102                 BUG_ON(IS_ERR(trans));
3103
3104                 ret = btrfs_grow_device(trans, device, old_size);
3105                 BUG_ON(ret);
3106
3107                 btrfs_end_transaction(trans, dev_root);
3108         }
3109
3110         /* step two, relocate all the chunks */
3111         path = btrfs_alloc_path();
3112         if (!path) {
3113                 ret = -ENOMEM;
3114                 goto error;
3115         }
3116
3117         /* zero out stat counters */
3118         spin_lock(&fs_info->balance_lock);
3119         memset(&bctl->stat, 0, sizeof(bctl->stat));
3120         spin_unlock(&fs_info->balance_lock);
3121 again:
3122         if (!counting) {
3123                 bctl->data.limit = limit_data;
3124                 bctl->meta.limit = limit_meta;
3125                 bctl->sys.limit = limit_sys;
3126         }
3127         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3128         key.offset = (u64)-1;
3129         key.type = BTRFS_CHUNK_ITEM_KEY;
3130
3131         while (1) {
3132                 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
3133                     atomic_read(&fs_info->balance_cancel_req)) {
3134                         ret = -ECANCELED;
3135                         goto error;
3136                 }
3137
3138                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3139                 if (ret < 0)
3140                         goto error;
3141
3142                 /*
3143                  * this shouldn't happen, it means the last relocate
3144                  * failed
3145                  */
3146                 if (ret == 0)
3147                         BUG(); /* FIXME break ? */
3148
3149                 ret = btrfs_previous_item(chunk_root, path, 0,
3150                                           BTRFS_CHUNK_ITEM_KEY);
3151                 if (ret) {
3152                         ret = 0;
3153                         break;
3154                 }
3155
3156                 leaf = path->nodes[0];
3157                 slot = path->slots[0];
3158                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3159
3160                 if (found_key.objectid != key.objectid)
3161                         break;
3162
3163                 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3164
3165                 if (!counting) {
3166                         spin_lock(&fs_info->balance_lock);
3167                         bctl->stat.considered++;
3168                         spin_unlock(&fs_info->balance_lock);
3169                 }
3170
3171                 ret = should_balance_chunk(chunk_root, leaf, chunk,
3172                                            found_key.offset);
3173                 btrfs_release_path(path);
3174                 if (!ret)
3175                         goto loop;
3176
3177                 if (counting) {
3178                         spin_lock(&fs_info->balance_lock);
3179                         bctl->stat.expected++;
3180                         spin_unlock(&fs_info->balance_lock);
3181                         goto loop;
3182                 }
3183
3184                 ret = btrfs_relocate_chunk(chunk_root,
3185                                            chunk_root->root_key.objectid,
3186                                            found_key.objectid,
3187                                            found_key.offset);
3188                 if (ret && ret != -ENOSPC)
3189                         goto error;
3190                 if (ret == -ENOSPC) {
3191                         enospc_errors++;
3192                 } else {
3193                         spin_lock(&fs_info->balance_lock);
3194                         bctl->stat.completed++;
3195                         spin_unlock(&fs_info->balance_lock);
3196                 }
3197 loop:
3198                 if (found_key.offset == 0)
3199                         break;
3200                 key.offset = found_key.offset - 1;
3201         }
3202
3203         if (counting) {
3204                 btrfs_release_path(path);
3205                 counting = false;
3206                 goto again;
3207         }
3208 error:
3209         btrfs_free_path(path);
3210         if (enospc_errors) {
3211                 btrfs_info(fs_info, "%d enospc errors during balance",
3212                        enospc_errors);
3213                 if (!ret)
3214                         ret = -ENOSPC;
3215         }
3216
3217         return ret;
3218 }
3219
3220 /**
3221  * alloc_profile_is_valid - see if a given profile is valid and reduced
3222  * @flags: profile to validate
3223  * @extended: if true @flags is treated as an extended profile
3224  */
3225 static int alloc_profile_is_valid(u64 flags, int extended)
3226 {
3227         u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
3228                                BTRFS_BLOCK_GROUP_PROFILE_MASK);
3229
3230         flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
3231
3232         /* 1) check that all other bits are zeroed */
3233         if (flags & ~mask)
3234                 return 0;
3235
3236         /* 2) see if profile is reduced */
3237         if (flags == 0)
3238                 return !extended; /* "0" is valid for usual profiles */
3239
3240         /* true if exactly one bit set */
3241         return (flags & (flags - 1)) == 0;
3242 }
3243
3244 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
3245 {
3246         /* cancel requested || normal exit path */
3247         return atomic_read(&fs_info->balance_cancel_req) ||
3248                 (atomic_read(&fs_info->balance_pause_req) == 0 &&
3249                  atomic_read(&fs_info->balance_cancel_req) == 0);
3250 }
3251
3252 static void __cancel_balance(struct btrfs_fs_info *fs_info)
3253 {
3254         int ret;
3255
3256         unset_balance_control(fs_info);
3257         ret = del_balance_item(fs_info->tree_root);
3258         if (ret)
3259                 btrfs_std_error(fs_info, ret);
3260
3261         atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3262 }
3263
3264 /*
3265  * Should be called with both balance and volume mutexes held
3266  */
3267 int btrfs_balance(struct btrfs_balance_control *bctl,
3268                   struct btrfs_ioctl_balance_args *bargs)
3269 {
3270         struct btrfs_fs_info *fs_info = bctl->fs_info;
3271         u64 allowed;
3272         int mixed = 0;
3273         int ret;
3274         u64 num_devices;
3275         unsigned seq;
3276
3277         if (btrfs_fs_closing(fs_info) ||
3278             atomic_read(&fs_info->balance_pause_req) ||
3279             atomic_read(&fs_info->balance_cancel_req)) {
3280                 ret = -EINVAL;
3281                 goto out;
3282         }
3283
3284         allowed = btrfs_super_incompat_flags(fs_info->super_copy);
3285         if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
3286                 mixed = 1;
3287
3288         /*
3289          * In case of mixed groups both data and meta should be picked,
3290          * and identical options should be given for both of them.
3291          */
3292         allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
3293         if (mixed && (bctl->flags & allowed)) {
3294                 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
3295                     !(bctl->flags & BTRFS_BALANCE_METADATA) ||
3296                     memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
3297                         btrfs_err(fs_info, "with mixed groups data and "
3298                                    "metadata balance options must be the same");
3299                         ret = -EINVAL;
3300                         goto out;
3301                 }
3302         }
3303
3304         num_devices = fs_info->fs_devices->num_devices;
3305         btrfs_dev_replace_lock(&fs_info->dev_replace);
3306         if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
3307                 BUG_ON(num_devices < 1);
3308                 num_devices--;
3309         }
3310         btrfs_dev_replace_unlock(&fs_info->dev_replace);
3311         allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE;
3312         if (num_devices == 1)
3313                 allowed |= BTRFS_BLOCK_GROUP_DUP;
3314         else if (num_devices > 1)
3315                 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1);
3316         if (num_devices > 2)
3317                 allowed |= BTRFS_BLOCK_GROUP_RAID5;
3318         if (num_devices > 3)
3319                 allowed |= (BTRFS_BLOCK_GROUP_RAID10 |
3320                             BTRFS_BLOCK_GROUP_RAID6);
3321         if ((bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3322             (!alloc_profile_is_valid(bctl->data.target, 1) ||
3323              (bctl->data.target & ~allowed))) {
3324                 btrfs_err(fs_info, "unable to start balance with target "
3325                            "data profile %llu",
3326                        bctl->data.target);
3327                 ret = -EINVAL;
3328                 goto out;
3329         }
3330         if ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3331             (!alloc_profile_is_valid(bctl->meta.target, 1) ||
3332              (bctl->meta.target & ~allowed))) {
3333                 btrfs_err(fs_info,
3334                            "unable to start balance with target metadata profile %llu",
3335                        bctl->meta.target);
3336                 ret = -EINVAL;
3337                 goto out;
3338         }
3339         if ((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3340             (!alloc_profile_is_valid(bctl->sys.target, 1) ||
3341              (bctl->sys.target & ~allowed))) {
3342                 btrfs_err(fs_info,
3343                            "unable to start balance with target system profile %llu",
3344                        bctl->sys.target);
3345                 ret = -EINVAL;
3346                 goto out;
3347         }
3348
3349         /* allow dup'ed data chunks only in mixed mode */
3350         if (!mixed && (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3351             (bctl->data.target & BTRFS_BLOCK_GROUP_DUP)) {
3352                 btrfs_err(fs_info, "dup for data is not allowed");
3353                 ret = -EINVAL;
3354                 goto out;
3355         }
3356
3357         /* allow to reduce meta or sys integrity only if force set */
3358         allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3359                         BTRFS_BLOCK_GROUP_RAID10 |
3360                         BTRFS_BLOCK_GROUP_RAID5 |
3361                         BTRFS_BLOCK_GROUP_RAID6;
3362         do {
3363                 seq = read_seqbegin(&fs_info->profiles_lock);
3364
3365                 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3366                      (fs_info->avail_system_alloc_bits & allowed) &&
3367                      !(bctl->sys.target & allowed)) ||
3368                     ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3369                      (fs_info->avail_metadata_alloc_bits & allowed) &&
3370                      !(bctl->meta.target & allowed))) {
3371                         if (bctl->flags & BTRFS_BALANCE_FORCE) {
3372                                 btrfs_info(fs_info, "force reducing metadata integrity");
3373                         } else {
3374                                 btrfs_err(fs_info, "balance will reduce metadata "
3375                                            "integrity, use force if you want this");
3376                                 ret = -EINVAL;
3377                                 goto out;
3378                         }
3379                 }
3380         } while (read_seqretry(&fs_info->profiles_lock, seq));
3381
3382         if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3383                 int num_tolerated_disk_barrier_failures;
3384                 u64 target = bctl->sys.target;
3385
3386                 num_tolerated_disk_barrier_failures =
3387                         btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3388                 if (num_tolerated_disk_barrier_failures > 0 &&
3389                     (target &
3390                      (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID0 |
3391                       BTRFS_AVAIL_ALLOC_BIT_SINGLE)))
3392                         num_tolerated_disk_barrier_failures = 0;
3393                 else if (num_tolerated_disk_barrier_failures > 1 &&
3394                          (target &
3395                           (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)))
3396                         num_tolerated_disk_barrier_failures = 1;
3397
3398                 fs_info->num_tolerated_disk_barrier_failures =
3399                         num_tolerated_disk_barrier_failures;
3400         }
3401
3402         ret = insert_balance_item(fs_info->tree_root, bctl);
3403         if (ret && ret != -EEXIST)
3404                 goto out;
3405
3406         if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
3407                 BUG_ON(ret == -EEXIST);
3408                 set_balance_control(bctl);
3409         } else {
3410                 BUG_ON(ret != -EEXIST);
3411                 spin_lock(&fs_info->balance_lock);
3412                 update_balance_args(bctl);
3413                 spin_unlock(&fs_info->balance_lock);
3414         }
3415
3416         atomic_inc(&fs_info->balance_running);
3417         mutex_unlock(&fs_info->balance_mutex);
3418
3419         ret = __btrfs_balance(fs_info);
3420
3421         mutex_lock(&fs_info->balance_mutex);
3422         atomic_dec(&fs_info->balance_running);
3423
3424         if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3425                 fs_info->num_tolerated_disk_barrier_failures =
3426                         btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3427         }
3428
3429         if (bargs) {
3430                 memset(bargs, 0, sizeof(*bargs));
3431                 update_ioctl_balance_args(fs_info, 0, bargs);
3432         }
3433
3434         if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
3435             balance_need_close(fs_info)) {
3436                 __cancel_balance(fs_info);
3437         }
3438
3439         wake_up(&fs_info->balance_wait_q);
3440
3441         return ret;
3442 out:
3443         if (bctl->flags & BTRFS_BALANCE_RESUME)
3444                 __cancel_balance(fs_info);
3445         else {
3446                 kfree(bctl);
3447                 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3448         }
3449         return ret;
3450 }
3451
3452 static int balance_kthread(void *data)
3453 {
3454         struct btrfs_fs_info *fs_info = data;
3455         int ret = 0;
3456
3457         mutex_lock(&fs_info->volume_mutex);
3458         mutex_lock(&fs_info->balance_mutex);
3459
3460         if (fs_info->balance_ctl) {
3461                 btrfs_info(fs_info, "continuing balance");
3462                 ret = btrfs_balance(fs_info->balance_ctl, NULL);
3463         }
3464
3465         mutex_unlock(&fs_info->balance_mutex);
3466         mutex_unlock(&fs_info->volume_mutex);
3467
3468         return ret;
3469 }
3470
3471 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
3472 {
3473         struct task_struct *tsk;
3474
3475         spin_lock(&fs_info->balance_lock);
3476         if (!fs_info->balance_ctl) {
3477                 spin_unlock(&fs_info->balance_lock);
3478                 return 0;
3479         }
3480         spin_unlock(&fs_info->balance_lock);
3481
3482         if (btrfs_test_opt(fs_info->tree_root, SKIP_BALANCE)) {
3483                 btrfs_info(fs_info, "force skipping balance");
3484                 return 0;
3485         }
3486
3487         tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
3488         return PTR_ERR_OR_ZERO(tsk);
3489 }
3490
3491 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
3492 {
3493         struct btrfs_balance_control *bctl;
3494         struct btrfs_balance_item *item;
3495         struct btrfs_disk_balance_args disk_bargs;
3496         struct btrfs_path *path;
3497         struct extent_buffer *leaf;
3498         struct btrfs_key key;
3499         int ret;
3500
3501         path = btrfs_alloc_path();
3502         if (!path)
3503                 return -ENOMEM;
3504
3505         key.objectid = BTRFS_BALANCE_OBJECTID;
3506         key.type = BTRFS_BALANCE_ITEM_KEY;
3507         key.offset = 0;
3508
3509         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
3510         if (ret < 0)
3511                 goto out;
3512         if (ret > 0) { /* ret = -ENOENT; */
3513                 ret = 0;
3514                 goto out;
3515         }
3516
3517         bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
3518         if (!bctl) {
3519                 ret = -ENOMEM;
3520                 goto out;
3521         }
3522
3523         leaf = path->nodes[0];
3524         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3525
3526         bctl->fs_info = fs_info;
3527         bctl->flags = btrfs_balance_flags(leaf, item);
3528         bctl->flags |= BTRFS_BALANCE_RESUME;
3529
3530         btrfs_balance_data(leaf, item, &disk_bargs);
3531         btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
3532         btrfs_balance_meta(leaf, item, &disk_bargs);
3533         btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
3534         btrfs_balance_sys(leaf, item, &disk_bargs);
3535         btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
3536
3537         WARN_ON(atomic_xchg(&fs_info->mutually_exclusive_operation_running, 1));
3538
3539         mutex_lock(&fs_info->volume_mutex);
3540         mutex_lock(&fs_info->balance_mutex);
3541
3542         set_balance_control(bctl);
3543
3544         mutex_unlock(&fs_info->balance_mutex);
3545         mutex_unlock(&fs_info->volume_mutex);
3546 out:
3547         btrfs_free_path(path);
3548         return ret;
3549 }
3550
3551 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
3552 {
3553         int ret = 0;
3554
3555         mutex_lock(&fs_info->balance_mutex);
3556         if (!fs_info->balance_ctl) {
3557                 mutex_unlock(&fs_info->balance_mutex);
3558                 return -ENOTCONN;
3559         }
3560
3561         if (atomic_read(&fs_info->balance_running)) {
3562                 atomic_inc(&fs_info->balance_pause_req);
3563                 mutex_unlock(&fs_info->balance_mutex);
3564
3565                 wait_event(fs_info->balance_wait_q,
3566                            atomic_read(&fs_info->balance_running) == 0);
3567
3568                 mutex_lock(&fs_info->balance_mutex);
3569                 /* we are good with balance_ctl ripped off from under us */
3570                 BUG_ON(atomic_read(&fs_info->balance_running));
3571                 atomic_dec(&fs_info->balance_pause_req);
3572         } else {
3573                 ret = -ENOTCONN;
3574         }
3575
3576         mutex_unlock(&fs_info->balance_mutex);
3577         return ret;
3578 }
3579
3580 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
3581 {
3582         if (fs_info->sb->s_flags & MS_RDONLY)
3583                 return -EROFS;
3584
3585         mutex_lock(&fs_info->balance_mutex);
3586         if (!fs_info->balance_ctl) {
3587                 mutex_unlock(&fs_info->balance_mutex);
3588                 return -ENOTCONN;
3589         }
3590
3591         atomic_inc(&fs_info->balance_cancel_req);
3592         /*
3593          * if we are running just wait and return, balance item is
3594          * deleted in btrfs_balance in this case
3595          */
3596         if (atomic_read(&fs_info->balance_running)) {
3597                 mutex_unlock(&fs_info->balance_mutex);
3598                 wait_event(fs_info->balance_wait_q,
3599                            atomic_read(&fs_info->balance_running) == 0);
3600                 mutex_lock(&fs_info->balance_mutex);
3601         } else {
3602                 /* __cancel_balance needs volume_mutex */
3603                 mutex_unlock(&fs_info->balance_mutex);
3604                 mutex_lock(&fs_info->volume_mutex);
3605                 mutex_lock(&fs_info->balance_mutex);
3606
3607                 if (fs_info->balance_ctl)
3608                         __cancel_balance(fs_info);
3609
3610                 mutex_unlock(&fs_info->volume_mutex);
3611         }
3612
3613         BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running));
3614         atomic_dec(&fs_info->balance_cancel_req);
3615         mutex_unlock(&fs_info->balance_mutex);
3616         return 0;
3617 }
3618
3619 static int btrfs_uuid_scan_kthread(void *data)
3620 {
3621         struct btrfs_fs_info *fs_info = data;
3622         struct btrfs_root *root = fs_info->tree_root;
3623         struct btrfs_key key;
3624         struct btrfs_key max_key;
3625         struct btrfs_path *path = NULL;
3626         int ret = 0;
3627         struct extent_buffer *eb;
3628         int slot;
3629         struct btrfs_root_item root_item;
3630         u32 item_size;
3631         struct btrfs_trans_handle *trans = NULL;
3632
3633         path = btrfs_alloc_path();
3634         if (!path) {
3635                 ret = -ENOMEM;
3636                 goto out;
3637         }
3638
3639         key.objectid = 0;
3640         key.type = BTRFS_ROOT_ITEM_KEY;
3641         key.offset = 0;
3642
3643         max_key.objectid = (u64)-1;
3644         max_key.type = BTRFS_ROOT_ITEM_KEY;
3645         max_key.offset = (u64)-1;
3646
3647         path->keep_locks = 1;
3648
3649         while (1) {
3650                 ret = btrfs_search_forward(root, &key, path, 0);
3651                 if (ret) {
3652                         if (ret > 0)
3653                                 ret = 0;
3654                         break;
3655                 }
3656
3657                 if (key.type != BTRFS_ROOT_ITEM_KEY ||
3658                     (key.objectid < BTRFS_FIRST_FREE_OBJECTID &&
3659                      key.objectid != BTRFS_FS_TREE_OBJECTID) ||
3660                     key.objectid > BTRFS_LAST_FREE_OBJECTID)
3661                         goto skip;
3662
3663                 eb = path->nodes[0];
3664                 slot = path->slots[0];
3665                 item_size = btrfs_item_size_nr(eb, slot);
3666                 if (item_size < sizeof(root_item))
3667                         goto skip;
3668
3669                 read_extent_buffer(eb, &root_item,
3670                                    btrfs_item_ptr_offset(eb, slot),
3671                                    (int)sizeof(root_item));
3672                 if (btrfs_root_refs(&root_item) == 0)
3673                         goto skip;
3674
3675                 if (!btrfs_is_empty_uuid(root_item.uuid) ||
3676                     !btrfs_is_empty_uuid(root_item.received_uuid)) {
3677                         if (trans)
3678                                 goto update_tree;
3679
3680                         btrfs_release_path(path);
3681                         /*
3682                          * 1 - subvol uuid item
3683                          * 1 - received_subvol uuid item
3684                          */
3685                         trans = btrfs_start_transaction(fs_info->uuid_root, 2);
3686                         if (IS_ERR(trans)) {
3687                                 ret = PTR_ERR(trans);
3688                                 break;
3689                         }
3690                         continue;
3691                 } else {
3692                         goto skip;
3693                 }
3694 update_tree:
3695                 if (!btrfs_is_empty_uuid(root_item.uuid)) {
3696                         ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
3697                                                   root_item.uuid,
3698                                                   BTRFS_UUID_KEY_SUBVOL,
3699                                                   key.objectid);
3700                         if (ret < 0) {
3701                                 btrfs_warn(fs_info, "uuid_tree_add failed %d",
3702                                         ret);
3703                                 break;
3704                         }
3705                 }
3706
3707                 if (!btrfs_is_empty_uuid(root_item.received_uuid)) {
3708                         ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
3709                                                   root_item.received_uuid,
3710                                                  BTRFS_UUID_KEY_RECEIVED_SUBVOL,
3711                                                   key.objectid);
3712                         if (ret < 0) {
3713                                 btrfs_warn(fs_info, "uuid_tree_add failed %d",
3714                                         ret);
3715                                 break;
3716                         }
3717                 }
3718
3719 skip:
3720                 if (trans) {
3721                         ret = btrfs_end_transaction(trans, fs_info->uuid_root);
3722                         trans = NULL;
3723                         if (ret)
3724                                 break;
3725                 }
3726
3727                 btrfs_release_path(path);
3728                 if (key.offset < (u64)-1) {
3729                         key.offset++;
3730                 } else if (key.type < BTRFS_ROOT_ITEM_KEY) {
3731                         key.offset = 0;
3732                         key.type = BTRFS_ROOT_ITEM_KEY;
3733                 } else if (key.objectid < (u64)-1) {
3734                         key.offset = 0;
3735                         key.type = BTRFS_ROOT_ITEM_KEY;
3736                         key.objectid++;
3737                 } else {
3738                         break;
3739                 }
3740                 cond_resched();
3741         }
3742
3743 out:
3744         btrfs_free_path(path);
3745         if (trans && !IS_ERR(trans))
3746                 btrfs_end_transaction(trans, fs_info->uuid_root);
3747         if (ret)
3748                 btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret);
3749         else
3750                 fs_info->update_uuid_tree_gen = 1;
3751         up(&fs_info->uuid_tree_rescan_sem);
3752         return 0;
3753 }
3754
3755 /*
3756  * Callback for btrfs_uuid_tree_iterate().
3757  * returns:
3758  * 0    check succeeded, the entry is not outdated.
3759  * < 0  if an error occured.
3760  * > 0  if the check failed, which means the caller shall remove the entry.
3761  */
3762 static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info *fs_info,
3763                                        u8 *uuid, u8 type, u64 subid)
3764 {
3765         struct btrfs_key key;
3766         int ret = 0;
3767         struct btrfs_root *subvol_root;
3768
3769         if (type != BTRFS_UUID_KEY_SUBVOL &&
3770             type != BTRFS_UUID_KEY_RECEIVED_SUBVOL)
3771                 goto out;
3772
3773         key.objectid = subid;
3774         key.type = BTRFS_ROOT_ITEM_KEY;
3775         key.offset = (u64)-1;
3776         subvol_root = btrfs_read_fs_root_no_name(fs_info, &key);
3777         if (IS_ERR(subvol_root)) {
3778                 ret = PTR_ERR(subvol_root);
3779                 if (ret == -ENOENT)
3780                         ret = 1;
3781                 goto out;
3782         }
3783
3784         switch (type) {
3785         case BTRFS_UUID_KEY_SUBVOL:
3786                 if (memcmp(uuid, subvol_root->root_item.uuid, BTRFS_UUID_SIZE))
3787                         ret = 1;
3788                 break;
3789         case BTRFS_UUID_KEY_RECEIVED_SUBVOL:
3790                 if (memcmp(uuid, subvol_root->root_item.received_uuid,
3791                            BTRFS_UUID_SIZE))
3792                         ret = 1;
3793                 break;
3794         }
3795
3796 out:
3797         return ret;
3798 }
3799
3800 static int btrfs_uuid_rescan_kthread(void *data)
3801 {
3802         struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data;
3803         int ret;
3804
3805         /*
3806          * 1st step is to iterate through the existing UUID tree and
3807          * to delete all entries that contain outdated data.
3808          * 2nd step is to add all missing entries to the UUID tree.
3809          */
3810         ret = btrfs_uuid_tree_iterate(fs_info, btrfs_check_uuid_tree_entry);
3811         if (ret < 0) {
3812                 btrfs_warn(fs_info, "iterating uuid_tree failed %d", ret);
3813                 up(&fs_info->uuid_tree_rescan_sem);
3814                 return ret;
3815         }
3816         return btrfs_uuid_scan_kthread(data);
3817 }
3818
3819 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info)
3820 {
3821         struct btrfs_trans_handle *trans;
3822         struct btrfs_root *tree_root = fs_info->tree_root;
3823         struct btrfs_root *uuid_root;
3824         struct task_struct *task;
3825         int ret;
3826
3827         /*
3828          * 1 - root node
3829          * 1 - root item
3830          */
3831         trans = btrfs_start_transaction(tree_root, 2);
3832         if (IS_ERR(trans))
3833                 return PTR_ERR(trans);
3834
3835         uuid_root = btrfs_create_tree(trans, fs_info,
3836                                       BTRFS_UUID_TREE_OBJECTID);
3837         if (IS_ERR(uuid_root)) {
3838                 btrfs_abort_transaction(trans, tree_root,
3839                                         PTR_ERR(uuid_root));
3840                 return PTR_ERR(uuid_root);
3841         }
3842
3843         fs_info->uuid_root = uuid_root;
3844
3845         ret = btrfs_commit_transaction(trans, tree_root);
3846         if (ret)
3847                 return ret;
3848
3849         down(&fs_info->uuid_tree_rescan_sem);
3850         task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid");
3851         if (IS_ERR(task)) {
3852                 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
3853                 btrfs_warn(fs_info, "failed to start uuid_scan task");
3854                 up(&fs_info->uuid_tree_rescan_sem);
3855                 return PTR_ERR(task);
3856         }
3857
3858         return 0;
3859 }
3860
3861 int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
3862 {
3863         struct task_struct *task;
3864
3865         down(&fs_info->uuid_tree_rescan_sem);
3866         task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
3867         if (IS_ERR(task)) {
3868                 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
3869                 btrfs_warn(fs_info, "failed to start uuid_rescan task");
3870                 up(&fs_info->uuid_tree_rescan_sem);
3871                 return PTR_ERR(task);
3872         }
3873
3874         return 0;
3875 }
3876
3877 /*
3878  * shrinking a device means finding all of the device extents past
3879  * the new size, and then following the back refs to the chunks.
3880  * The chunk relocation code actually frees the device extent
3881  */
3882 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
3883 {
3884         struct btrfs_trans_handle *trans;
3885         struct btrfs_root *root = device->dev_root;
3886         struct btrfs_dev_extent *dev_extent = NULL;
3887         struct btrfs_path *path;
3888         u64 length;
3889         u64 chunk_tree;
3890         u64 chunk_objectid;
3891         u64 chunk_offset;
3892         int ret;
3893         int slot;
3894         int failed = 0;
3895         bool retried = false;
3896         struct extent_buffer *l;
3897         struct btrfs_key key;
3898         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
3899         u64 old_total = btrfs_super_total_bytes(super_copy);
3900         u64 old_size = device->total_bytes;
3901         u64 diff = device->total_bytes - new_size;
3902
3903         if (device->is_tgtdev_for_dev_replace)
3904                 return -EINVAL;
3905
3906         path = btrfs_alloc_path();
3907         if (!path)
3908                 return -ENOMEM;
3909
3910         path->reada = 2;
3911
3912         lock_chunks(root);
3913
3914         device->total_bytes = new_size;
3915         if (device->writeable) {
3916                 device->fs_devices->total_rw_bytes -= diff;
3917                 spin_lock(&root->fs_info->free_chunk_lock);
3918                 root->fs_info->free_chunk_space -= diff;
3919                 spin_unlock(&root->fs_info->free_chunk_lock);
3920         }
3921         unlock_chunks(root);
3922
3923 again:
3924         key.objectid = device->devid;
3925         key.offset = (u64)-1;
3926         key.type = BTRFS_DEV_EXTENT_KEY;
3927
3928         do {
3929                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3930                 if (ret < 0)
3931                         goto done;
3932
3933                 ret = btrfs_previous_item(root, path, 0, key.type);
3934                 if (ret < 0)
3935                         goto done;
3936                 if (ret) {
3937                         ret = 0;
3938                         btrfs_release_path(path);
3939                         break;
3940                 }
3941
3942                 l = path->nodes[0];
3943                 slot = path->slots[0];
3944                 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
3945
3946                 if (key.objectid != device->devid) {
3947                         btrfs_release_path(path);
3948                         break;
3949                 }
3950
3951                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
3952                 length = btrfs_dev_extent_length(l, dev_extent);
3953
3954                 if (key.offset + length <= new_size) {
3955                         btrfs_release_path(path);
3956                         break;
3957                 }
3958
3959                 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
3960                 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
3961                 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
3962                 btrfs_release_path(path);
3963
3964                 ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
3965                                            chunk_offset);
3966                 if (ret && ret != -ENOSPC)
3967                         goto done;
3968                 if (ret == -ENOSPC)
3969                         failed++;
3970         } while (key.offset-- > 0);
3971
3972         if (failed && !retried) {
3973                 failed = 0;
3974                 retried = true;
3975                 goto again;
3976         } else if (failed && retried) {
3977                 ret = -ENOSPC;
3978                 lock_chunks(root);
3979
3980                 device->total_bytes = old_size;
3981                 if (device->writeable)
3982                         device->fs_devices->total_rw_bytes += diff;
3983                 spin_lock(&root->fs_info->free_chunk_lock);
3984                 root->fs_info->free_chunk_space += diff;
3985                 spin_unlock(&root->fs_info->free_chunk_lock);
3986                 unlock_chunks(root);
3987                 goto done;
3988         }
3989
3990         /* Shrinking succeeded, else we would be at "done". */
3991         trans = btrfs_start_transaction(root, 0);
3992         if (IS_ERR(trans)) {
3993                 ret = PTR_ERR(trans);
3994                 goto done;
3995         }
3996
3997         lock_chunks(root);
3998
3999         device->disk_total_bytes = new_size;
4000         /* Now btrfs_update_device() will change the on-disk size. */
4001         ret = btrfs_update_device(trans, device);
4002         if (ret) {
4003                 unlock_chunks(root);
4004                 btrfs_end_transaction(trans, root);
4005                 goto done;
4006         }
4007         WARN_ON(diff > old_total);
4008         btrfs_set_super_total_bytes(super_copy, old_total - diff);
4009         unlock_chunks(root);
4010         btrfs_end_transaction(trans, root);
4011 done:
4012         btrfs_free_path(path);
4013         return ret;
4014 }
4015
4016 static int btrfs_add_system_chunk(struct btrfs_root *root,
4017                            struct btrfs_key *key,
4018                            struct btrfs_chunk *chunk, int item_size)
4019 {
4020         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
4021         struct btrfs_disk_key disk_key;
4022         u32 array_size;
4023         u8 *ptr;
4024
4025         array_size = btrfs_super_sys_array_size(super_copy);
4026         if (array_size + item_size + sizeof(disk_key)
4027                         > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
4028                 return -EFBIG;
4029
4030         ptr = super_copy->sys_chunk_array + array_size;
4031         btrfs_cpu_key_to_disk(&disk_key, key);
4032         memcpy(ptr, &disk_key, sizeof(disk_key));
4033         ptr += sizeof(disk_key);
4034         memcpy(ptr, chunk, item_size);
4035         item_size += sizeof(disk_key);
4036         btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
4037         return 0;
4038 }
4039
4040 /*
4041  * sort the devices in descending order by max_avail, total_avail
4042  */
4043 static int btrfs_cmp_device_info(const void *a, const void *b)
4044 {
4045         const struct btrfs_device_info *di_a = a;
4046         const struct btrfs_device_info *di_b = b;
4047
4048         if (di_a->max_avail > di_b->max_avail)
4049                 return -1;
4050         if (di_a->max_avail < di_b->max_avail)
4051                 return 1;
4052         if (di_a->total_avail > di_b->total_avail)
4053                 return -1;
4054         if (di_a->total_avail < di_b->total_avail)
4055                 return 1;
4056         return 0;
4057 }
4058
4059 static struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = {
4060         [BTRFS_RAID_RAID10] = {
4061                 .sub_stripes    = 2,
4062                 .dev_stripes    = 1,
4063                 .devs_max       = 0,    /* 0 == as many as possible */
4064                 .devs_min       = 4,
4065                 .devs_increment = 2,
4066                 .ncopies        = 2,
4067         },
4068         [BTRFS_RAID_RAID1] = {
4069                 .sub_stripes    = 1,
4070                 .dev_stripes    = 1,
4071                 .devs_max       = 2,
4072                 .devs_min       = 2,
4073                 .devs_increment = 2,
4074                 .ncopies        = 2,
4075         },
4076         [BTRFS_RAID_DUP] = {
4077                 .sub_stripes    = 1,
4078                 .dev_stripes    = 2,
4079                 .devs_max       = 1,
4080                 .devs_min       = 1,
4081                 .devs_increment = 1,
4082                 .ncopies        = 2,
4083         },
4084         [BTRFS_RAID_RAID0] = {
4085                 .sub_stripes    = 1,
4086                 .dev_stripes    = 1,
4087                 .devs_max       = 0,
4088                 .devs_min       = 2,
4089                 .devs_increment = 1,
4090                 .ncopies        = 1,
4091         },
4092         [BTRFS_RAID_SINGLE] = {
4093                 .sub_stripes    = 1,
4094                 .dev_stripes    = 1,
4095                 .devs_max       = 1,
4096                 .devs_min       = 1,
4097                 .devs_increment = 1,
4098                 .ncopies        = 1,
4099         },
4100         [BTRFS_RAID_RAID5] = {
4101                 .sub_stripes    = 1,
4102                 .dev_stripes    = 1,
4103                 .devs_max       = 0,
4104                 .devs_min       = 2,
4105                 .devs_increment = 1,
4106                 .ncopies        = 2,
4107         },
4108         [BTRFS_RAID_RAID6] = {
4109                 .sub_stripes    = 1,
4110                 .dev_stripes    = 1,
4111                 .devs_max       = 0,
4112                 .devs_min       = 3,
4113                 .devs_increment = 1,
4114                 .ncopies        = 3,
4115         },
4116 };
4117
4118 static u32 find_raid56_stripe_len(u32 data_devices, u32 dev_stripe_target)
4119 {
4120         /* TODO allow them to set a preferred stripe size */
4121         return 64 * 1024;
4122 }
4123
4124 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
4125 {
4126         if (!(type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)))
4127                 return;
4128
4129         btrfs_set_fs_incompat(info, RAID56);
4130 }
4131
4132 #define BTRFS_MAX_DEVS(r) ((BTRFS_LEAF_DATA_SIZE(r)             \
4133                         - sizeof(struct btrfs_item)             \
4134                         - sizeof(struct btrfs_chunk))           \
4135                         / sizeof(struct btrfs_stripe) + 1)
4136
4137 #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE        \
4138                                 - 2 * sizeof(struct btrfs_disk_key)     \
4139                                 - 2 * sizeof(struct btrfs_chunk))       \
4140                                 / sizeof(struct btrfs_stripe) + 1)
4141
4142 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4143                                struct btrfs_root *extent_root, u64 start,
4144                                u64 type)
4145 {
4146         struct btrfs_fs_info *info = extent_root->fs_info;
4147         struct btrfs_fs_devices *fs_devices = info->fs_devices;
4148         struct list_head *cur;
4149         struct map_lookup *map = NULL;
4150         struct extent_map_tree *em_tree;
4151         struct extent_map *em;
4152         struct btrfs_device_info *devices_info = NULL;
4153         u64 total_avail;
4154         int num_stripes;        /* total number of stripes to allocate */
4155         int data_stripes;       /* number of stripes that count for
4156                                    block group size */
4157         int sub_stripes;        /* sub_stripes info for map */
4158         int dev_stripes;        /* stripes per dev */
4159         int devs_max;           /* max devs to use */
4160         int devs_min;           /* min devs needed */
4161         int devs_increment;     /* ndevs has to be a multiple of this */
4162         int ncopies;            /* how many copies to data has */
4163         int ret;
4164         u64 max_stripe_size;
4165         u64 max_chunk_size;
4166         u64 stripe_size;
4167         u64 num_bytes;
4168         u64 raid_stripe_len = BTRFS_STRIPE_LEN;
4169         int ndevs;
4170         int i;
4171         int j;
4172         int index;
4173
4174         BUG_ON(!alloc_profile_is_valid(type, 0));
4175
4176         if (list_empty(&fs_devices->alloc_list))
4177                 return -ENOSPC;
4178
4179         index = __get_raid_index(type);
4180
4181         sub_stripes = btrfs_raid_array[index].sub_stripes;
4182         dev_stripes = btrfs_raid_array[index].dev_stripes;
4183         devs_max = btrfs_raid_array[index].devs_max;
4184         devs_min = btrfs_raid_array[index].devs_min;
4185         devs_increment = btrfs_raid_array[index].devs_increment;
4186         ncopies = btrfs_raid_array[index].ncopies;
4187
4188         if (type & BTRFS_BLOCK_GROUP_DATA) {
4189                 max_stripe_size = 1024 * 1024 * 1024;
4190                 max_chunk_size = 10 * max_stripe_size;
4191                 if (!devs_max)
4192                         devs_max = BTRFS_MAX_DEVS(info->chunk_root);
4193         } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
4194                 /* for larger filesystems, use larger metadata chunks */
4195                 if (fs_devices->total_rw_bytes > 50ULL * 1024 * 1024 * 1024)
4196                         max_stripe_size = 1024 * 1024 * 1024;
4197                 else
4198                         max_stripe_size = 256 * 1024 * 1024;
4199                 max_chunk_size = max_stripe_size;
4200                 if (!devs_max)
4201                         devs_max = BTRFS_MAX_DEVS(info->chunk_root);
4202         } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
4203                 max_stripe_size = 32 * 1024 * 1024;
4204                 max_chunk_size = 2 * max_stripe_size;
4205                 if (!devs_max)
4206                         devs_max = BTRFS_MAX_DEVS_SYS_CHUNK;
4207         } else {
4208                 btrfs_err(info, "invalid chunk type 0x%llx requested",
4209                        type);
4210                 BUG_ON(1);
4211         }
4212
4213         /* we don't want a chunk larger than 10% of writeable space */
4214         max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
4215                              max_chunk_size);
4216
4217         devices_info = kzalloc(sizeof(*devices_info) * fs_devices->rw_devices,
4218                                GFP_NOFS);
4219         if (!devices_info)
4220                 return -ENOMEM;
4221
4222         cur = fs_devices->alloc_list.next;
4223
4224         /*
4225          * in the first pass through the devices list, we gather information
4226          * about the available holes on each device.
4227          */
4228         ndevs = 0;
4229         while (cur != &fs_devices->alloc_list) {
4230                 struct btrfs_device *device;
4231                 u64 max_avail;
4232                 u64 dev_offset;
4233
4234                 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
4235
4236                 cur = cur->next;
4237
4238                 if (!device->writeable) {
4239                         WARN(1, KERN_ERR
4240                                "BTRFS: read-only device in alloc_list\n");
4241                         continue;
4242                 }
4243
4244                 if (!device->in_fs_metadata ||
4245                     device->is_tgtdev_for_dev_replace)
4246                         continue;
4247
4248                 if (device->total_bytes > device->bytes_used)
4249                         total_avail = device->total_bytes - device->bytes_used;
4250                 else
4251                         total_avail = 0;
4252
4253                 /* If there is no space on this device, skip it. */
4254                 if (total_avail == 0)
4255                         continue;
4256
4257                 ret = find_free_dev_extent(trans, device,
4258                                            max_stripe_size * dev_stripes,
4259                                            &dev_offset, &max_avail);
4260                 if (ret && ret != -ENOSPC)
4261                         goto error;
4262
4263                 if (ret == 0)
4264                         max_avail = max_stripe_size * dev_stripes;
4265
4266                 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes)
4267                         continue;
4268
4269                 if (ndevs == fs_devices->rw_devices) {
4270                         WARN(1, "%s: found more than %llu devices\n",
4271                              __func__, fs_devices->rw_devices);
4272                         break;
4273                 }
4274                 devices_info[ndevs].dev_offset = dev_offset;
4275                 devices_info[ndevs].max_avail = max_avail;
4276                 devices_info[ndevs].total_avail = total_avail;
4277                 devices_info[ndevs].dev = device;
4278                 ++ndevs;
4279         }
4280
4281         /*
4282          * now sort the devices by hole size / available space
4283          */
4284         sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
4285              btrfs_cmp_device_info, NULL);
4286
4287         /* round down to number of usable stripes */
4288         ndevs -= ndevs % devs_increment;
4289
4290         if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) {
4291                 ret = -ENOSPC;
4292                 goto error;
4293         }
4294
4295         if (devs_max && ndevs > devs_max)
4296                 ndevs = devs_max;
4297         /*
4298          * the primary goal is to maximize the number of stripes, so use as many
4299          * devices as possible, even if the stripes are not maximum sized.
4300          */
4301         stripe_size = devices_info[ndevs-1].max_avail;
4302         num_stripes = ndevs * dev_stripes;
4303
4304         /*
4305          * this will have to be fixed for RAID1 and RAID10 over
4306          * more drives
4307          */
4308         data_stripes = num_stripes / ncopies;
4309
4310         if (type & BTRFS_BLOCK_GROUP_RAID5) {
4311                 raid_stripe_len = find_raid56_stripe_len(ndevs - 1,
4312                                  btrfs_super_stripesize(info->super_copy));
4313                 data_stripes = num_stripes - 1;
4314         }
4315         if (type & BTRFS_BLOCK_GROUP_RAID6) {
4316                 raid_stripe_len = find_raid56_stripe_len(ndevs - 2,
4317                                  btrfs_super_stripesize(info->super_copy));
4318                 data_stripes = num_stripes - 2;
4319         }
4320
4321         /*
4322          * Use the number of data stripes to figure out how big this chunk
4323          * is really going to be in terms of logical address space,
4324          * and compare that answer with the max chunk size
4325          */
4326         if (stripe_size * data_stripes > max_chunk_size) {
4327                 u64 mask = (1ULL << 24) - 1;
4328                 stripe_size = max_chunk_size;
4329                 do_div(stripe_size, data_stripes);
4330
4331                 /* bump the answer up to a 16MB boundary */
4332                 stripe_size = (stripe_size + mask) & ~mask;
4333
4334                 /* but don't go higher than the limits we found
4335                  * while searching for free extents
4336                  */
4337                 if (stripe_size > devices_info[ndevs-1].max_avail)
4338                         stripe_size = devices_info[ndevs-1].max_avail;
4339         }
4340
4341         do_div(stripe_size, dev_stripes);
4342
4343         /* align to BTRFS_STRIPE_LEN */
4344         do_div(stripe_size, raid_stripe_len);
4345         stripe_size *= raid_stripe_len;
4346
4347         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
4348         if (!map) {
4349                 ret = -ENOMEM;
4350                 goto error;
4351         }
4352         map->num_stripes = num_stripes;
4353
4354         for (i = 0; i < ndevs; ++i) {
4355                 for (j = 0; j < dev_stripes; ++j) {
4356                         int s = i * dev_stripes + j;
4357                         map->stripes[s].dev = devices_info[i].dev;
4358                         map->stripes[s].physical = devices_info[i].dev_offset +
4359                                                    j * stripe_size;
4360                 }
4361         }
4362         map->sector_size = extent_root->sectorsize;
4363         map->stripe_len = raid_stripe_len;
4364         map->io_align = raid_stripe_len;
4365         map->io_width = raid_stripe_len;
4366         map->type = type;
4367         map->sub_stripes = sub_stripes;
4368
4369         num_bytes = stripe_size * data_stripes;
4370
4371         trace_btrfs_chunk_alloc(info->chunk_root, map, start, num_bytes);
4372
4373         em = alloc_extent_map();
4374         if (!em) {
4375                 kfree(map);
4376                 ret = -ENOMEM;
4377                 goto error;
4378         }
4379         set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
4380         em->bdev = (struct block_device *)map;
4381         em->start = start;
4382         em->len = num_bytes;
4383         em->block_start = 0;
4384         em->block_len = em->len;
4385         em->orig_block_len = stripe_size;
4386
4387         em_tree = &extent_root->fs_info->mapping_tree.map_tree;
4388         write_lock(&em_tree->lock);
4389         ret = add_extent_mapping(em_tree, em, 0);
4390         if (!ret) {
4391                 list_add_tail(&em->list, &trans->transaction->pending_chunks);
4392                 atomic_inc(&em->refs);
4393         }
4394         write_unlock(&em_tree->lock);
4395         if (ret) {
4396                 free_extent_map(em);
4397                 goto error;
4398         }
4399
4400         ret = btrfs_make_block_group(trans, extent_root, 0, type,
4401                                      BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4402                                      start, num_bytes);
4403         if (ret)
4404                 goto error_del_extent;
4405
4406         free_extent_map(em);
4407         check_raid56_incompat_flag(extent_root->fs_info, type);
4408
4409         kfree(devices_info);
4410         return 0;
4411
4412 error_del_extent:
4413         write_lock(&em_tree->lock);
4414         remove_extent_mapping(em_tree, em);
4415         write_unlock(&em_tree->lock);
4416
4417         /* One for our allocation */
4418         free_extent_map(em);
4419         /* One for the tree reference */
4420         free_extent_map(em);
4421 error:
4422         kfree(devices_info);
4423         return ret;
4424 }
4425
4426 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans,
4427                                 struct btrfs_root *extent_root,
4428                                 u64 chunk_offset, u64 chunk_size)
4429 {
4430         struct btrfs_key key;
4431         struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
4432         struct btrfs_device *device;
4433         struct btrfs_chunk *chunk;
4434         struct btrfs_stripe *stripe;
4435         struct extent_map_tree *em_tree;
4436         struct extent_map *em;
4437         struct map_lookup *map;
4438         size_t item_size;
4439         u64 dev_offset;
4440         u64 stripe_size;
4441         int i = 0;
4442         int ret;
4443
4444         em_tree = &extent_root->fs_info->mapping_tree.map_tree;
4445         read_lock(&em_tree->lock);
4446         em = lookup_extent_mapping(em_tree, chunk_offset, chunk_size);
4447         read_unlock(&em_tree->lock);
4448
4449         if (!em) {
4450                 btrfs_crit(extent_root->fs_info, "unable to find logical "
4451                            "%Lu len %Lu", chunk_offset, chunk_size);
4452                 return -EINVAL;
4453         }
4454
4455         if (em->start != chunk_offset || em->len != chunk_size) {
4456                 btrfs_crit(extent_root->fs_info, "found a bad mapping, wanted"
4457                           " %Lu-%Lu, found %Lu-%Lu", chunk_offset,
4458                           chunk_size, em->start, em->len);
4459                 free_extent_map(em);
4460                 return -EINVAL;
4461         }
4462
4463         map = (struct map_lookup *)em->bdev;
4464         item_size = btrfs_chunk_item_size(map->num_stripes);
4465         stripe_size = em->orig_block_len;
4466
4467         chunk = kzalloc(item_size, GFP_NOFS);
4468         if (!chunk) {
4469                 ret = -ENOMEM;
4470                 goto out;
4471         }
4472
4473         for (i = 0; i < map->num_stripes; i++) {
4474                 device = map->stripes[i].dev;
4475                 dev_offset = map->stripes[i].physical;
4476
4477                 device->bytes_used += stripe_size;
4478                 ret = btrfs_update_device(trans, device);
4479                 if (ret)
4480                         goto out;
4481                 ret = btrfs_alloc_dev_extent(trans, device,
4482                                              chunk_root->root_key.objectid,
4483                                              BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4484                                              chunk_offset, dev_offset,
4485                                              stripe_size);
4486                 if (ret)
4487                         goto out;
4488         }
4489
4490         spin_lock(&extent_root->fs_info->free_chunk_lock);
4491         extent_root->fs_info->free_chunk_space -= (stripe_size *
4492                                                    map->num_stripes);
4493         spin_unlock(&extent_root->fs_info->free_chunk_lock);
4494
4495         stripe = &chunk->stripe;
4496         for (i = 0; i < map->num_stripes; i++) {
4497                 device = map->stripes[i].dev;
4498                 dev_offset = map->stripes[i].physical;
4499
4500                 btrfs_set_stack_stripe_devid(stripe, device->devid);
4501                 btrfs_set_stack_stripe_offset(stripe, dev_offset);
4502                 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
4503                 stripe++;
4504         }
4505
4506         btrfs_set_stack_chunk_length(chunk, chunk_size);
4507         btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
4508         btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
4509         btrfs_set_stack_chunk_type(chunk, map->type);
4510         btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
4511         btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
4512         btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
4513         btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
4514         btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
4515
4516         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
4517         key.type = BTRFS_CHUNK_ITEM_KEY;
4518         key.offset = chunk_offset;
4519
4520         ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
4521         if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
4522                 /*
4523                  * TODO: Cleanup of inserted chunk root in case of
4524                  * failure.
4525                  */
4526                 ret = btrfs_add_system_chunk(chunk_root, &key, chunk,
4527                                              item_size);
4528         }
4529
4530 out:
4531         kfree(chunk);
4532         free_extent_map(em);
4533         return ret;
4534 }
4535
4536 /*
4537  * Chunk allocation falls into two parts. The first part does works
4538  * that make the new allocated chunk useable, but not do any operation
4539  * that modifies the chunk tree. The second part does the works that
4540  * require modifying the chunk tree. This division is important for the
4541  * bootstrap process of adding storage to a seed btrfs.
4542  */
4543 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4544                       struct btrfs_root *extent_root, u64 type)
4545 {
4546         u64 chunk_offset;
4547
4548         chunk_offset = find_next_chunk(extent_root->fs_info);
4549         return __btrfs_alloc_chunk(trans, extent_root, chunk_offset, type);
4550 }
4551
4552 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
4553                                          struct btrfs_root *root,
4554                                          struct btrfs_device *device)
4555 {
4556         u64 chunk_offset;
4557         u64 sys_chunk_offset;
4558         u64 alloc_profile;
4559         struct btrfs_fs_info *fs_info = root->fs_info;
4560         struct btrfs_root *extent_root = fs_info->extent_root;
4561         int ret;
4562
4563         chunk_offset = find_next_chunk(fs_info);
4564         alloc_profile = btrfs_get_alloc_profile(extent_root, 0);
4565         ret = __btrfs_alloc_chunk(trans, extent_root, chunk_offset,
4566                                   alloc_profile);
4567         if (ret)
4568                 return ret;
4569
4570         sys_chunk_offset = find_next_chunk(root->fs_info);
4571         alloc_profile = btrfs_get_alloc_profile(fs_info->chunk_root, 0);
4572         ret = __btrfs_alloc_chunk(trans, extent_root, sys_chunk_offset,
4573                                   alloc_profile);
4574         if (ret) {
4575                 btrfs_abort_transaction(trans, root, ret);
4576                 goto out;
4577         }
4578
4579         ret = btrfs_add_device(trans, fs_info->chunk_root, device);
4580         if (ret)
4581                 btrfs_abort_transaction(trans, root, ret);
4582 out:
4583         return ret;
4584 }
4585
4586 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
4587 {
4588         struct extent_map *em;
4589         struct map_lookup *map;
4590         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
4591         int readonly = 0;
4592         int i;
4593
4594         read_lock(&map_tree->map_tree.lock);
4595         em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
4596         read_unlock(&map_tree->map_tree.lock);
4597         if (!em)
4598                 return 1;
4599
4600         if (btrfs_test_opt(root, DEGRADED)) {
4601                 free_extent_map(em);
4602                 return 0;
4603         }
4604
4605         map = (struct map_lookup *)em->bdev;
4606         for (i = 0; i < map->num_stripes; i++) {
4607                 if (!map->stripes[i].dev->writeable) {
4608                         readonly = 1;
4609                         break;
4610                 }
4611         }
4612         free_extent_map(em);
4613         return readonly;
4614 }
4615
4616 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
4617 {
4618         extent_map_tree_init(&tree->map_tree);
4619 }
4620
4621 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
4622 {
4623         struct extent_map *em;
4624
4625         while (1) {
4626                 write_lock(&tree->map_tree.lock);
4627                 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
4628                 if (em)
4629                         remove_extent_mapping(&tree->map_tree, em);
4630                 write_unlock(&tree->map_tree.lock);
4631                 if (!em)
4632                         break;
4633                 /* once for us */
4634                 free_extent_map(em);
4635                 /* once for the tree */
4636                 free_extent_map(em);
4637         }
4638 }
4639
4640 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
4641 {
4642         struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
4643         struct extent_map *em;
4644         struct map_lookup *map;
4645         struct extent_map_tree *em_tree = &map_tree->map_tree;
4646         int ret;
4647
4648         read_lock(&em_tree->lock);
4649         em = lookup_extent_mapping(em_tree, logical, len);
4650         read_unlock(&em_tree->lock);
4651
4652         /*
4653          * We could return errors for these cases, but that could get ugly and
4654          * we'd probably do the same thing which is just not do anything else
4655          * and exit, so return 1 so the callers don't try to use other copies.
4656          */
4657         if (!em) {
4658                 btrfs_crit(fs_info, "No mapping for %Lu-%Lu", logical,
4659                             logical+len);
4660                 return 1;
4661         }
4662
4663         if (em->start > logical || em->start + em->len < logical) {
4664                 btrfs_crit(fs_info, "Invalid mapping for %Lu-%Lu, got "
4665                             "%Lu-%Lu", logical, logical+len, em->start,
4666                             em->start + em->len);
4667                 free_extent_map(em);
4668                 return 1;
4669         }
4670
4671         map = (struct map_lookup *)em->bdev;
4672         if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
4673                 ret = map->num_stripes;
4674         else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
4675                 ret = map->sub_stripes;
4676         else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
4677                 ret = 2;
4678         else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
4679                 ret = 3;
4680         else
4681                 ret = 1;
4682         free_extent_map(em);
4683
4684         btrfs_dev_replace_lock(&fs_info->dev_replace);
4685         if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace))
4686                 ret++;
4687         btrfs_dev_replace_unlock(&fs_info->dev_replace);
4688
4689         return ret;
4690 }
4691
4692 unsigned long btrfs_full_stripe_len(struct btrfs_root *root,
4693                                     struct btrfs_mapping_tree *map_tree,
4694                                     u64 logical)
4695 {
4696         struct extent_map *em;
4697         struct map_lookup *map;
4698         struct extent_map_tree *em_tree = &map_tree->map_tree;
4699         unsigned long len = root->sectorsize;
4700
4701         read_lock(&em_tree->lock);
4702         em = lookup_extent_mapping(em_tree, logical, len);
4703         read_unlock(&em_tree->lock);
4704         BUG_ON(!em);
4705
4706         BUG_ON(em->start > logical || em->start + em->len < logical);
4707         map = (struct map_lookup *)em->bdev;
4708         if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
4709                          BTRFS_BLOCK_GROUP_RAID6)) {
4710                 len = map->stripe_len * nr_data_stripes(map);
4711         }
4712         free_extent_map(em);
4713         return len;
4714 }
4715
4716 int btrfs_is_parity_mirror(struct btrfs_mapping_tree *map_tree,
4717                            u64 logical, u64 len, int mirror_num)
4718 {
4719         struct extent_map *em;
4720         struct map_lookup *map;
4721         struct extent_map_tree *em_tree = &map_tree->map_tree;
4722         int ret = 0;
4723
4724         read_lock(&em_tree->lock);
4725         em = lookup_extent_mapping(em_tree, logical, len);
4726         read_unlock(&em_tree->lock);
4727         BUG_ON(!em);
4728
4729         BUG_ON(em->start > logical || em->start + em->len < logical);
4730         map = (struct map_lookup *)em->bdev;
4731         if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
4732                          BTRFS_BLOCK_GROUP_RAID6))
4733                 ret = 1;
4734         free_extent_map(em);
4735         return ret;
4736 }
4737
4738 static int find_live_mirror(struct btrfs_fs_info *fs_info,
4739                             struct map_lookup *map, int first, int num,
4740                             int optimal, int dev_replace_is_ongoing)
4741 {
4742         int i;
4743         int tolerance;
4744         struct btrfs_device *srcdev;
4745
4746         if (dev_replace_is_ongoing &&
4747             fs_info->dev_replace.cont_reading_from_srcdev_mode ==
4748              BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
4749                 srcdev = fs_info->dev_replace.srcdev;
4750         else
4751                 srcdev = NULL;
4752
4753         /*
4754          * try to avoid the drive that is the source drive for a
4755          * dev-replace procedure, only choose it if no other non-missing
4756          * mirror is available
4757          */
4758         for (tolerance = 0; tolerance < 2; tolerance++) {
4759                 if (map->stripes[optimal].dev->bdev &&
4760                     (tolerance || map->stripes[optimal].dev != srcdev))
4761                         return optimal;
4762                 for (i = first; i < first + num; i++) {
4763                         if (map->stripes[i].dev->bdev &&
4764                             (tolerance || map->stripes[i].dev != srcdev))
4765                                 return i;
4766                 }
4767         }
4768
4769         /* we couldn't find one that doesn't fail.  Just return something
4770          * and the io error handling code will clean up eventually
4771          */
4772         return optimal;
4773 }
4774
4775 static inline int parity_smaller(u64 a, u64 b)
4776 {
4777         return a > b;
4778 }
4779
4780 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
4781 static void sort_parity_stripes(struct btrfs_bio *bbio, u64 *raid_map)
4782 {
4783         struct btrfs_bio_stripe s;
4784         int i;
4785         u64 l;
4786         int again = 1;
4787
4788         while (again) {
4789                 again = 0;
4790                 for (i = 0; i < bbio->num_stripes - 1; i++) {
4791                         if (parity_smaller(raid_map[i], raid_map[i+1])) {
4792                                 s = bbio->stripes[i];
4793                                 l = raid_map[i];
4794                                 bbio->stripes[i] = bbio->stripes[i+1];
4795                                 raid_map[i] = raid_map[i+1];
4796                                 bbio->stripes[i+1] = s;
4797                                 raid_map[i+1] = l;
4798                                 again = 1;
4799                         }
4800                 }
4801         }
4802 }
4803
4804 static int __btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
4805                              u64 logical, u64 *length,
4806                              struct btrfs_bio **bbio_ret,
4807                              int mirror_num, u64 **raid_map_ret)
4808 {
4809         struct extent_map *em;
4810         struct map_lookup *map;
4811         struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
4812         struct extent_map_tree *em_tree = &map_tree->map_tree;
4813         u64 offset;
4814         u64 stripe_offset;
4815         u64 stripe_end_offset;
4816         u64 stripe_nr;
4817         u64 stripe_nr_orig;
4818         u64 stripe_nr_end;
4819         u64 stripe_len;
4820         u64 *raid_map = NULL;
4821         int stripe_index;
4822         int i;
4823         int ret = 0;
4824         int num_stripes;
4825         int max_errors = 0;
4826         struct btrfs_bio *bbio = NULL;
4827         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
4828         int dev_replace_is_ongoing = 0;
4829         int num_alloc_stripes;
4830         int patch_the_first_stripe_for_dev_replace = 0;
4831         u64 physical_to_patch_in_first_stripe = 0;
4832         u64 raid56_full_stripe_start = (u64)-1;
4833
4834         read_lock(&em_tree->lock);
4835         em = lookup_extent_mapping(em_tree, logical, *length);
4836         read_unlock(&em_tree->lock);
4837
4838         if (!em) {
4839                 btrfs_crit(fs_info, "unable to find logical %llu len %llu",
4840                         logical, *length);
4841                 return -EINVAL;
4842         }
4843
4844         if (em->start > logical || em->start + em->len < logical) {
4845                 btrfs_crit(fs_info, "found a bad mapping, wanted %Lu, "
4846                            "found %Lu-%Lu", logical, em->start,
4847                            em->start + em->len);
4848                 free_extent_map(em);
4849                 return -EINVAL;
4850         }
4851
4852         map = (struct map_lookup *)em->bdev;
4853         offset = logical - em->start;
4854
4855         stripe_len = map->stripe_len;
4856         stripe_nr = offset;
4857         /*
4858          * stripe_nr counts the total number of stripes we have to stride
4859          * to get to this block
4860          */
4861         do_div(stripe_nr, stripe_len);
4862
4863         stripe_offset = stripe_nr * stripe_len;
4864         BUG_ON(offset < stripe_offset);
4865
4866         /* stripe_offset is the offset of this block in its stripe*/
4867         stripe_offset = offset - stripe_offset;
4868
4869         /* if we're here for raid56, we need to know the stripe aligned start */
4870         if (map->type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)) {
4871                 unsigned long full_stripe_len = stripe_len * nr_data_stripes(map);
4872                 raid56_full_stripe_start = offset;
4873
4874                 /* allow a write of a full stripe, but make sure we don't
4875                  * allow straddling of stripes
4876                  */
4877                 do_div(raid56_full_stripe_start, full_stripe_len);
4878                 raid56_full_stripe_start *= full_stripe_len;
4879         }
4880
4881         if (rw & REQ_DISCARD) {
4882                 /* we don't discard raid56 yet */
4883                 if (map->type &
4884                     (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)) {
4885                         ret = -EOPNOTSUPP;
4886                         goto out;
4887                 }
4888                 *length = min_t(u64, em->len - offset, *length);
4889         } else if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
4890                 u64 max_len;
4891                 /* For writes to RAID[56], allow a full stripeset across all disks.
4892                    For other RAID types and for RAID[56] reads, just allow a single
4893                    stripe (on a single disk). */
4894                 if (map->type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6) &&
4895                     (rw & REQ_WRITE)) {
4896                         max_len = stripe_len * nr_data_stripes(map) -
4897                                 (offset - raid56_full_stripe_start);
4898                 } else {
4899                         /* we limit the length of each bio to what fits in a stripe */
4900                         max_len = stripe_len - stripe_offset;
4901                 }
4902                 *length = min_t(u64, em->len - offset, max_len);
4903         } else {
4904                 *length = em->len - offset;
4905         }
4906
4907         /* This is for when we're called from btrfs_merge_bio_hook() and all
4908            it cares about is the length */
4909         if (!bbio_ret)
4910                 goto out;
4911
4912         btrfs_dev_replace_lock(dev_replace);
4913         dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
4914         if (!dev_replace_is_ongoing)
4915                 btrfs_dev_replace_unlock(dev_replace);
4916
4917         if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
4918             !(rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) &&
4919             dev_replace->tgtdev != NULL) {
4920                 /*
4921                  * in dev-replace case, for repair case (that's the only
4922                  * case where the mirror is selected explicitly when
4923                  * calling btrfs_map_block), blocks left of the left cursor
4924                  * can also be read from the target drive.
4925                  * For REQ_GET_READ_MIRRORS, the target drive is added as
4926                  * the last one to the array of stripes. For READ, it also
4927                  * needs to be supported using the same mirror number.
4928                  * If the requested block is not left of the left cursor,
4929                  * EIO is returned. This can happen because btrfs_num_copies()
4930                  * returns one more in the dev-replace case.
4931                  */
4932                 u64 tmp_length = *length;
4933                 struct btrfs_bio *tmp_bbio = NULL;
4934                 int tmp_num_stripes;
4935                 u64 srcdev_devid = dev_replace->srcdev->devid;
4936                 int index_srcdev = 0;
4937                 int found = 0;
4938                 u64 physical_of_found = 0;
4939
4940                 ret = __btrfs_map_block(fs_info, REQ_GET_READ_MIRRORS,
4941                              logical, &tmp_length, &tmp_bbio, 0, NULL);
4942                 if (ret) {
4943                         WARN_ON(tmp_bbio != NULL);
4944                         goto out;
4945                 }
4946
4947                 tmp_num_stripes = tmp_bbio->num_stripes;
4948                 if (mirror_num > tmp_num_stripes) {
4949                         /*
4950                          * REQ_GET_READ_MIRRORS does not contain this
4951                          * mirror, that means that the requested area
4952                          * is not left of the left cursor
4953                          */
4954                         ret = -EIO;
4955                         kfree(tmp_bbio);
4956                         goto out;
4957                 }
4958
4959                 /*
4960                  * process the rest of the function using the mirror_num
4961                  * of the source drive. Therefore look it up first.
4962                  * At the end, patch the device pointer to the one of the
4963                  * target drive.
4964                  */
4965                 for (i = 0; i < tmp_num_stripes; i++) {
4966                         if (tmp_bbio->stripes[i].dev->devid == srcdev_devid) {
4967                                 /*
4968                                  * In case of DUP, in order to keep it
4969                                  * simple, only add the mirror with the
4970                                  * lowest physical address
4971                                  */
4972                                 if (found &&
4973                                     physical_of_found <=
4974                                      tmp_bbio->stripes[i].physical)
4975                                         continue;
4976                                 index_srcdev = i;
4977                                 found = 1;
4978                                 physical_of_found =
4979                                         tmp_bbio->stripes[i].physical;
4980                         }
4981                 }
4982
4983                 if (found) {
4984                         mirror_num = index_srcdev + 1;
4985                         patch_the_first_stripe_for_dev_replace = 1;
4986                         physical_to_patch_in_first_stripe = physical_of_found;
4987                 } else {
4988                         WARN_ON(1);
4989                         ret = -EIO;
4990                         kfree(tmp_bbio);
4991                         goto out;
4992                 }
4993
4994                 kfree(tmp_bbio);
4995         } else if (mirror_num > map->num_stripes) {
4996                 mirror_num = 0;
4997         }
4998
4999         num_stripes = 1;
5000         stripe_index = 0;
5001         stripe_nr_orig = stripe_nr;
5002         stripe_nr_end = ALIGN(offset + *length, map->stripe_len);
5003         do_div(stripe_nr_end, map->stripe_len);
5004         stripe_end_offset = stripe_nr_end * map->stripe_len -
5005                             (offset + *length);
5006
5007         if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5008                 if (rw & REQ_DISCARD)
5009                         num_stripes = min_t(u64, map->num_stripes,
5010                                             stripe_nr_end - stripe_nr_orig);
5011                 stripe_index = do_div(stripe_nr, map->num_stripes);
5012         } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
5013                 if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS))
5014                         num_stripes = map->num_stripes;
5015                 else if (mirror_num)
5016                         stripe_index = mirror_num - 1;
5017                 else {
5018                         stripe_index = find_live_mirror(fs_info, map, 0,
5019                                             map->num_stripes,
5020                                             current->pid % map->num_stripes,
5021                                             dev_replace_is_ongoing);
5022                         mirror_num = stripe_index + 1;
5023                 }
5024
5025         } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
5026                 if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) {
5027                         num_stripes = map->num_stripes;
5028                 } else if (mirror_num) {
5029                         stripe_index = mirror_num - 1;
5030                 } else {
5031                         mirror_num = 1;
5032                 }
5033
5034         } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5035                 int factor = map->num_stripes / map->sub_stripes;
5036
5037                 stripe_index = do_div(stripe_nr, factor);
5038                 stripe_index *= map->sub_stripes;
5039
5040                 if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS))
5041                         num_stripes = map->sub_stripes;
5042                 else if (rw & REQ_DISCARD)
5043                         num_stripes = min_t(u64, map->sub_stripes *
5044                                             (stripe_nr_end - stripe_nr_orig),
5045                                             map->num_stripes);
5046                 else if (mirror_num)
5047                         stripe_index += mirror_num - 1;
5048                 else {
5049                         int old_stripe_index = stripe_index;
5050                         stripe_index = find_live_mirror(fs_info, map,
5051                                               stripe_index,
5052                                               map->sub_stripes, stripe_index +
5053                                               current->pid % map->sub_stripes,
5054                                               dev_replace_is_ongoing);
5055                         mirror_num = stripe_index - old_stripe_index + 1;
5056                 }
5057
5058         } else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
5059                                 BTRFS_BLOCK_GROUP_RAID6)) {
5060                 u64 tmp;
5061
5062                 if (bbio_ret && ((rw & REQ_WRITE) || mirror_num > 1)
5063                     && raid_map_ret) {
5064                         int i, rot;
5065
5066                         /* push stripe_nr back to the start of the full stripe */
5067                         stripe_nr = raid56_full_stripe_start;
5068                         do_div(stripe_nr, stripe_len);
5069
5070                         stripe_index = do_div(stripe_nr, nr_data_stripes(map));
5071
5072                         /* RAID[56] write or recovery. Return all stripes */
5073                         num_stripes = map->num_stripes;
5074                         max_errors = nr_parity_stripes(map);
5075
5076                         raid_map = kmalloc_array(num_stripes, sizeof(u64),
5077                                            GFP_NOFS);
5078                         if (!raid_map) {
5079                                 ret = -ENOMEM;
5080                                 goto out;
5081                         }
5082
5083                         /* Work out the disk rotation on this stripe-set */
5084                         tmp = stripe_nr;
5085                         rot = do_div(tmp, num_stripes);
5086
5087                         /* Fill in the logical address of each stripe */
5088                         tmp = stripe_nr * nr_data_stripes(map);
5089                         for (i = 0; i < nr_data_stripes(map); i++)
5090                                 raid_map[(i+rot) % num_stripes] =
5091                                         em->start + (tmp + i) * map->stripe_len;
5092
5093                         raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE;
5094                         if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5095                                 raid_map[(i+rot+1) % num_stripes] =
5096                                         RAID6_Q_STRIPE;
5097
5098                         *length = map->stripe_len;
5099                         stripe_index = 0;
5100                         stripe_offset = 0;
5101                 } else {
5102                         /*
5103                          * Mirror #0 or #1 means the original data block.
5104                          * Mirror #2 is RAID5 parity block.
5105                          * Mirror #3 is RAID6 Q block.
5106                          */
5107                         stripe_index = do_div(stripe_nr, nr_data_stripes(map));
5108                         if (mirror_num > 1)
5109                                 stripe_index = nr_data_stripes(map) +
5110                                                 mirror_num - 2;
5111
5112                         /* We distribute the parity blocks across stripes */
5113                         tmp = stripe_nr + stripe_index;
5114                         stripe_index = do_div(tmp, map->num_stripes);
5115                 }
5116         } else {
5117                 /*
5118                  * after this do_div call, stripe_nr is the number of stripes
5119                  * on this device we have to walk to find the data, and
5120                  * stripe_index is the number of our device in the stripe array
5121                  */
5122                 stripe_index = do_div(stripe_nr, map->num_stripes);
5123                 mirror_num = stripe_index + 1;
5124         }
5125         BUG_ON(stripe_index >= map->num_stripes);
5126
5127         num_alloc_stripes = num_stripes;
5128         if (dev_replace_is_ongoing) {
5129                 if (rw & (REQ_WRITE | REQ_DISCARD))
5130                         num_alloc_stripes <<= 1;
5131                 if (rw & REQ_GET_READ_MIRRORS)
5132                         num_alloc_stripes++;
5133         }
5134         bbio = kzalloc(btrfs_bio_size(num_alloc_stripes), GFP_NOFS);
5135         if (!bbio) {
5136                 kfree(raid_map);
5137                 ret = -ENOMEM;
5138                 goto out;
5139         }
5140         atomic_set(&bbio->error, 0);
5141
5142         if (rw & REQ_DISCARD) {
5143                 int factor = 0;
5144                 int sub_stripes = 0;
5145                 u64 stripes_per_dev = 0;
5146                 u32 remaining_stripes = 0;
5147                 u32 last_stripe = 0;
5148
5149                 if (map->type &
5150                     (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID10)) {
5151                         if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5152                                 sub_stripes = 1;
5153                         else
5154                                 sub_stripes = map->sub_stripes;
5155
5156                         factor = map->num_stripes / sub_stripes;
5157                         stripes_per_dev = div_u64_rem(stripe_nr_end -
5158                                                       stripe_nr_orig,
5159                                                       factor,
5160                                                       &remaining_stripes);
5161                         div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
5162                         last_stripe *= sub_stripes;
5163                 }
5164
5165                 for (i = 0; i < num_stripes; i++) {
5166                         bbio->stripes[i].physical =
5167                                 map->stripes[stripe_index].physical +
5168                                 stripe_offset + stripe_nr * map->stripe_len;
5169                         bbio->stripes[i].dev = map->stripes[stripe_index].dev;
5170
5171                         if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5172                                          BTRFS_BLOCK_GROUP_RAID10)) {
5173                                 bbio->stripes[i].length = stripes_per_dev *
5174                                                           map->stripe_len;
5175
5176                                 if (i / sub_stripes < remaining_stripes)
5177                                         bbio->stripes[i].length +=
5178                                                 map->stripe_len;
5179
5180                                 /*
5181                                  * Special for the first stripe and
5182                                  * the last stripe:
5183                                  *
5184                                  * |-------|...|-------|
5185                                  *     |----------|
5186                                  *    off     end_off
5187                                  */
5188                                 if (i < sub_stripes)
5189                                         bbio->stripes[i].length -=
5190                                                 stripe_offset;
5191
5192                                 if (stripe_index >= last_stripe &&
5193                                     stripe_index <= (last_stripe +
5194                                                      sub_stripes - 1))
5195                                         bbio->stripes[i].length -=
5196                                                 stripe_end_offset;
5197
5198                                 if (i == sub_stripes - 1)
5199                                         stripe_offset = 0;
5200                         } else
5201                                 bbio->stripes[i].length = *length;
5202
5203                         stripe_index++;
5204                         if (stripe_index == map->num_stripes) {
5205                                 /* This could only happen for RAID0/10 */
5206                                 stripe_index = 0;
5207                                 stripe_nr++;
5208                         }
5209                 }
5210         } else {
5211                 for (i = 0; i < num_stripes; i++) {
5212                         bbio->stripes[i].physical =
5213                                 map->stripes[stripe_index].physical +
5214                                 stripe_offset +
5215                                 stripe_nr * map->stripe_len;
5216                         bbio->stripes[i].dev =
5217                                 map->stripes[stripe_index].dev;
5218                         stripe_index++;
5219                 }
5220         }
5221
5222         if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS)) {
5223                 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
5224                                  BTRFS_BLOCK_GROUP_RAID10 |
5225                                  BTRFS_BLOCK_GROUP_RAID5 |
5226                                  BTRFS_BLOCK_GROUP_DUP)) {
5227                         max_errors = 1;
5228                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID6) {
5229                         max_errors = 2;
5230                 }
5231         }
5232
5233         if (dev_replace_is_ongoing && (rw & (REQ_WRITE | REQ_DISCARD)) &&
5234             dev_replace->tgtdev != NULL) {
5235                 int index_where_to_add;
5236                 u64 srcdev_devid = dev_replace->srcdev->devid;
5237
5238                 /*
5239                  * duplicate the write operations while the dev replace
5240                  * procedure is running. Since the copying of the old disk
5241                  * to the new disk takes place at run time while the
5242                  * filesystem is mounted writable, the regular write
5243                  * operations to the old disk have to be duplicated to go
5244                  * to the new disk as well.
5245                  * Note that device->missing is handled by the caller, and
5246                  * that the write to the old disk is already set up in the
5247                  * stripes array.
5248                  */
5249                 index_where_to_add = num_stripes;
5250                 for (i = 0; i < num_stripes; i++) {
5251                         if (bbio->stripes[i].dev->devid == srcdev_devid) {
5252                                 /* write to new disk, too */
5253                                 struct btrfs_bio_stripe *new =
5254                                         bbio->stripes + index_where_to_add;
5255                                 struct btrfs_bio_stripe *old =
5256                                         bbio->stripes + i;
5257
5258                                 new->physical = old->physical;
5259                                 new->length = old->length;
5260                                 new->dev = dev_replace->tgtdev;
5261                                 index_where_to_add++;
5262                                 max_errors++;
5263                         }
5264                 }
5265                 num_stripes = index_where_to_add;
5266         } else if (dev_replace_is_ongoing && (rw & REQ_GET_READ_MIRRORS) &&
5267                    dev_replace->tgtdev != NULL) {
5268                 u64 srcdev_devid = dev_replace->srcdev->devid;
5269                 int index_srcdev = 0;
5270                 int found = 0;
5271                 u64 physical_of_found = 0;
5272
5273                 /*
5274                  * During the dev-replace procedure, the target drive can
5275                  * also be used to read data in case it is needed to repair
5276                  * a corrupt block elsewhere. This is possible if the
5277                  * requested area is left of the left cursor. In this area,
5278                  * the target drive is a full copy of the source drive.
5279                  */
5280                 for (i = 0; i < num_stripes; i++) {
5281                         if (bbio->stripes[i].dev->devid == srcdev_devid) {
5282                                 /*
5283                                  * In case of DUP, in order to keep it
5284                                  * simple, only add the mirror with the
5285                                  * lowest physical address
5286                                  */
5287                                 if (found &&
5288                                     physical_of_found <=
5289                                      bbio->stripes[i].physical)
5290                                         continue;
5291                                 index_srcdev = i;
5292                                 found = 1;
5293                                 physical_of_found = bbio->stripes[i].physical;
5294                         }
5295                 }
5296                 if (found) {
5297                         u64 length = map->stripe_len;
5298
5299                         if (physical_of_found + length <=
5300                             dev_replace->cursor_left) {
5301                                 struct btrfs_bio_stripe *tgtdev_stripe =
5302                                         bbio->stripes + num_stripes;
5303
5304                                 tgtdev_stripe->physical = physical_of_found;
5305                                 tgtdev_stripe->length =
5306                                         bbio->stripes[index_srcdev].length;
5307                                 tgtdev_stripe->dev = dev_replace->tgtdev;
5308
5309                                 num_stripes++;
5310                         }
5311                 }
5312         }
5313
5314         *bbio_ret = bbio;
5315         bbio->num_stripes = num_stripes;
5316         bbio->max_errors = max_errors;
5317         bbio->mirror_num = mirror_num;
5318
5319         /*
5320          * this is the case that REQ_READ && dev_replace_is_ongoing &&
5321          * mirror_num == num_stripes + 1 && dev_replace target drive is
5322          * available as a mirror
5323          */
5324         if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) {
5325                 WARN_ON(num_stripes > 1);
5326                 bbio->stripes[0].dev = dev_replace->tgtdev;
5327                 bbio->stripes[0].physical = physical_to_patch_in_first_stripe;
5328                 bbio->mirror_num = map->num_stripes + 1;
5329         }
5330         if (raid_map) {
5331                 sort_parity_stripes(bbio, raid_map);
5332                 *raid_map_ret = raid_map;
5333         }
5334 out:
5335         if (dev_replace_is_ongoing)
5336                 btrfs_dev_replace_unlock(dev_replace);
5337         free_extent_map(em);
5338         return ret;
5339 }
5340
5341 int btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
5342                       u64 logical, u64 *length,
5343                       struct btrfs_bio **bbio_ret, int mirror_num)
5344 {
5345         return __btrfs_map_block(fs_info, rw, logical, length, bbio_ret,
5346                                  mirror_num, NULL);
5347 }
5348
5349 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
5350                      u64 chunk_start, u64 physical, u64 devid,
5351                      u64 **logical, int *naddrs, int *stripe_len)
5352 {
5353         struct extent_map_tree *em_tree = &map_tree->map_tree;
5354         struct extent_map *em;
5355         struct map_lookup *map;
5356         u64 *buf;
5357         u64 bytenr;
5358         u64 length;
5359         u64 stripe_nr;
5360         u64 rmap_len;
5361         int i, j, nr = 0;
5362
5363         read_lock(&em_tree->lock);
5364         em = lookup_extent_mapping(em_tree, chunk_start, 1);
5365         read_unlock(&em_tree->lock);
5366
5367         if (!em) {
5368                 printk(KERN_ERR "BTRFS: couldn't find em for chunk %Lu\n",
5369                        chunk_start);
5370                 return -EIO;
5371         }
5372
5373         if (em->start != chunk_start) {
5374                 printk(KERN_ERR "BTRFS: bad chunk start, em=%Lu, wanted=%Lu\n",
5375                        em->start, chunk_start);
5376                 free_extent_map(em);
5377                 return -EIO;
5378         }
5379         map = (struct map_lookup *)em->bdev;
5380
5381         length = em->len;
5382         rmap_len = map->stripe_len;
5383
5384         if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5385                 do_div(length, map->num_stripes / map->sub_stripes);
5386         else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5387                 do_div(length, map->num_stripes);
5388         else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
5389                               BTRFS_BLOCK_GROUP_RAID6)) {
5390                 do_div(length, nr_data_stripes(map));
5391                 rmap_len = map->stripe_len * nr_data_stripes(map);
5392         }
5393
5394         buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
5395         BUG_ON(!buf); /* -ENOMEM */
5396
5397         for (i = 0; i < map->num_stripes; i++) {
5398                 if (devid && map->stripes[i].dev->devid != devid)
5399                         continue;
5400                 if (map->stripes[i].physical > physical ||
5401                     map->stripes[i].physical + length <= physical)
5402                         continue;
5403
5404                 stripe_nr = physical - map->stripes[i].physical;
5405                 do_div(stripe_nr, map->stripe_len);
5406
5407                 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5408                         stripe_nr = stripe_nr * map->num_stripes + i;
5409                         do_div(stripe_nr, map->sub_stripes);
5410                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5411                         stripe_nr = stripe_nr * map->num_stripes + i;
5412                 } /* else if RAID[56], multiply by nr_data_stripes().
5413                    * Alternatively, just use rmap_len below instead of
5414                    * map->stripe_len */
5415
5416                 bytenr = chunk_start + stripe_nr * rmap_len;
5417                 WARN_ON(nr >= map->num_stripes);
5418                 for (j = 0; j < nr; j++) {
5419                         if (buf[j] == bytenr)
5420                                 break;
5421                 }
5422                 if (j == nr) {
5423                         WARN_ON(nr >= map->num_stripes);
5424                         buf[nr++] = bytenr;
5425                 }
5426         }
5427
5428         *logical = buf;
5429         *naddrs = nr;
5430         *stripe_len = rmap_len;
5431
5432         free_extent_map(em);
5433         return 0;
5434 }
5435
5436 static inline void btrfs_end_bbio(struct btrfs_bio *bbio, struct bio *bio, int err)
5437 {
5438         if (likely(bbio->flags & BTRFS_BIO_ORIG_BIO_SUBMITTED))
5439                 bio_endio_nodec(bio, err);
5440         else
5441                 bio_endio(bio, err);
5442         kfree(bbio);
5443 }
5444
5445 static void btrfs_end_bio(struct bio *bio, int err)
5446 {
5447         struct btrfs_bio *bbio = bio->bi_private;
5448         struct btrfs_device *dev = bbio->stripes[0].dev;
5449         int is_orig_bio = 0;
5450
5451         if (err) {
5452                 atomic_inc(&bbio->error);
5453                 if (err == -EIO || err == -EREMOTEIO) {
5454                         unsigned int stripe_index =
5455                                 btrfs_io_bio(bio)->stripe_index;
5456
5457                         BUG_ON(stripe_index >= bbio->num_stripes);
5458                         dev = bbio->stripes[stripe_index].dev;
5459                         if (dev->bdev) {
5460                                 if (bio->bi_rw & WRITE)
5461                                         btrfs_dev_stat_inc(dev,
5462                                                 BTRFS_DEV_STAT_WRITE_ERRS);
5463                                 else
5464                                         btrfs_dev_stat_inc(dev,
5465                                                 BTRFS_DEV_STAT_READ_ERRS);
5466                                 if ((bio->bi_rw & WRITE_FLUSH) == WRITE_FLUSH)
5467                                         btrfs_dev_stat_inc(dev,
5468                                                 BTRFS_DEV_STAT_FLUSH_ERRS);
5469                                 btrfs_dev_stat_print_on_error(dev);
5470                         }
5471                 }
5472         }
5473
5474         if (bio == bbio->orig_bio)
5475                 is_orig_bio = 1;
5476
5477         btrfs_bio_counter_dec(bbio->fs_info);
5478
5479         if (atomic_dec_and_test(&bbio->stripes_pending)) {
5480                 if (!is_orig_bio) {
5481                         bio_put(bio);
5482                         bio = bbio->orig_bio;
5483                 }
5484
5485                 bio->bi_private = bbio->private;
5486                 bio->bi_end_io = bbio->end_io;
5487                 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
5488                 /* only send an error to the higher layers if it is
5489                  * beyond the tolerance of the btrfs bio
5490                  */
5491                 if (atomic_read(&bbio->error) > bbio->max_errors) {
5492                         err = -EIO;
5493                 } else {
5494                         /*
5495                          * this bio is actually up to date, we didn't
5496                          * go over the max number of errors
5497                          */
5498                         set_bit(BIO_UPTODATE, &bio->bi_flags);
5499                         err = 0;
5500                 }
5501
5502                 btrfs_end_bbio(bbio, bio, err);
5503         } else if (!is_orig_bio) {
5504                 bio_put(bio);
5505         }
5506 }
5507
5508 /*
5509  * see run_scheduled_bios for a description of why bios are collected for
5510  * async submit.
5511  *
5512  * This will add one bio to the pending list for a device and make sure
5513  * the work struct is scheduled.
5514  */
5515 static noinline void btrfs_schedule_bio(struct btrfs_root *root,
5516                                         struct btrfs_device *device,
5517                                         int rw, struct bio *bio)
5518 {
5519         int should_queue = 1;
5520         struct btrfs_pending_bios *pending_bios;
5521
5522         if (device->missing || !device->bdev) {
5523                 bio_endio(bio, -EIO);
5524                 return;
5525         }
5526
5527         /* don't bother with additional async steps for reads, right now */
5528         if (!(rw & REQ_WRITE)) {
5529                 bio_get(bio);
5530                 btrfsic_submit_bio(rw, bio);
5531                 bio_put(bio);
5532                 return;
5533         }
5534
5535         /*
5536          * nr_async_bios allows us to reliably return congestion to the
5537          * higher layers.  Otherwise, the async bio makes it appear we have
5538          * made progress against dirty pages when we've really just put it
5539          * on a queue for later
5540          */
5541         atomic_inc(&root->fs_info->nr_async_bios);
5542         WARN_ON(bio->bi_next);
5543         bio->bi_next = NULL;
5544         bio->bi_rw |= rw;
5545
5546         spin_lock(&device->io_lock);
5547         if (bio->bi_rw & REQ_SYNC)
5548                 pending_bios = &device->pending_sync_bios;
5549         else
5550                 pending_bios = &device->pending_bios;
5551
5552         if (pending_bios->tail)
5553                 pending_bios->tail->bi_next = bio;
5554
5555         pending_bios->tail = bio;
5556         if (!pending_bios->head)
5557                 pending_bios->head = bio;
5558         if (device->running_pending)
5559                 should_queue = 0;
5560
5561         spin_unlock(&device->io_lock);
5562
5563         if (should_queue)
5564                 btrfs_queue_work(root->fs_info->submit_workers,
5565                                  &device->work);
5566 }
5567
5568 static int bio_size_ok(struct block_device *bdev, struct bio *bio,
5569                        sector_t sector)
5570 {
5571         struct bio_vec *prev;
5572         struct request_queue *q = bdev_get_queue(bdev);
5573         unsigned int max_sectors = queue_max_sectors(q);
5574         struct bvec_merge_data bvm = {
5575                 .bi_bdev = bdev,
5576                 .bi_sector = sector,
5577                 .bi_rw = bio->bi_rw,
5578         };
5579
5580         if (WARN_ON(bio->bi_vcnt == 0))
5581                 return 1;
5582
5583         prev = &bio->bi_io_vec[bio->bi_vcnt - 1];
5584         if (bio_sectors(bio) > max_sectors)
5585                 return 0;
5586
5587         if (!q->merge_bvec_fn)
5588                 return 1;
5589
5590         bvm.bi_size = bio->bi_iter.bi_size - prev->bv_len;
5591         if (q->merge_bvec_fn(q, &bvm, prev) < prev->bv_len)
5592                 return 0;
5593         return 1;
5594 }
5595
5596 static void submit_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio,
5597                               struct bio *bio, u64 physical, int dev_nr,
5598                               int rw, int async)
5599 {
5600         struct btrfs_device *dev = bbio->stripes[dev_nr].dev;
5601
5602         bio->bi_private = bbio;
5603         btrfs_io_bio(bio)->stripe_index = dev_nr;
5604         bio->bi_end_io = btrfs_end_bio;
5605         bio->bi_iter.bi_sector = physical >> 9;
5606 #ifdef DEBUG
5607         {
5608                 struct rcu_string *name;
5609
5610                 rcu_read_lock();
5611                 name = rcu_dereference(dev->name);
5612                 pr_debug("btrfs_map_bio: rw %d, sector=%llu, dev=%lu "
5613                          "(%s id %llu), size=%u\n", rw,
5614                          (u64)bio->bi_sector, (u_long)dev->bdev->bd_dev,
5615                          name->str, dev->devid, bio->bi_size);
5616                 rcu_read_unlock();
5617         }
5618 #endif
5619         bio->bi_bdev = dev->bdev;
5620
5621         btrfs_bio_counter_inc_noblocked(root->fs_info);
5622
5623         if (async)
5624                 btrfs_schedule_bio(root, dev, rw, bio);
5625         else
5626                 btrfsic_submit_bio(rw, bio);
5627 }
5628
5629 static int breakup_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio,
5630                               struct bio *first_bio, struct btrfs_device *dev,
5631                               int dev_nr, int rw, int async)
5632 {
5633         struct bio_vec *bvec = first_bio->bi_io_vec;
5634         struct bio *bio;
5635         int nr_vecs = bio_get_nr_vecs(dev->bdev);
5636         u64 physical = bbio->stripes[dev_nr].physical;
5637
5638 again:
5639         bio = btrfs_bio_alloc(dev->bdev, physical >> 9, nr_vecs, GFP_NOFS);
5640         if (!bio)
5641                 return -ENOMEM;
5642
5643         while (bvec <= (first_bio->bi_io_vec + first_bio->bi_vcnt - 1)) {
5644                 if (bio_add_page(bio, bvec->bv_page, bvec->bv_len,
5645                                  bvec->bv_offset) < bvec->bv_len) {
5646                         u64 len = bio->bi_iter.bi_size;
5647
5648                         atomic_inc(&bbio->stripes_pending);
5649                         submit_stripe_bio(root, bbio, bio, physical, dev_nr,
5650                                           rw, async);
5651                         physical += len;
5652                         goto again;
5653                 }
5654                 bvec++;
5655         }
5656
5657         submit_stripe_bio(root, bbio, bio, physical, dev_nr, rw, async);
5658         return 0;
5659 }
5660
5661 static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical)
5662 {
5663         atomic_inc(&bbio->error);
5664         if (atomic_dec_and_test(&bbio->stripes_pending)) {
5665                 /* Shoud be the original bio. */
5666                 WARN_ON(bio != bbio->orig_bio);
5667
5668                 bio->bi_private = bbio->private;
5669                 bio->bi_end_io = bbio->end_io;
5670                 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
5671                 bio->bi_iter.bi_sector = logical >> 9;
5672
5673                 btrfs_end_bbio(bbio, bio, -EIO);
5674         }
5675 }
5676
5677 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
5678                   int mirror_num, int async_submit)
5679 {
5680         struct btrfs_device *dev;
5681         struct bio *first_bio = bio;
5682         u64 logical = (u64)bio->bi_iter.bi_sector << 9;
5683         u64 length = 0;
5684         u64 map_length;
5685         u64 *raid_map = NULL;
5686         int ret;
5687         int dev_nr = 0;
5688         int total_devs = 1;
5689         struct btrfs_bio *bbio = NULL;
5690
5691         length = bio->bi_iter.bi_size;
5692         map_length = length;
5693
5694         btrfs_bio_counter_inc_blocked(root->fs_info);
5695         ret = __btrfs_map_block(root->fs_info, rw, logical, &map_length, &bbio,
5696                               mirror_num, &raid_map);
5697         if (ret) {
5698                 btrfs_bio_counter_dec(root->fs_info);
5699                 return ret;
5700         }
5701
5702         total_devs = bbio->num_stripes;
5703         bbio->orig_bio = first_bio;
5704         bbio->private = first_bio->bi_private;
5705         bbio->end_io = first_bio->bi_end_io;
5706         bbio->fs_info = root->fs_info;
5707         atomic_set(&bbio->stripes_pending, bbio->num_stripes);
5708
5709         if (raid_map) {
5710                 /* In this case, map_length has been set to the length of
5711                    a single stripe; not the whole write */
5712                 if (rw & WRITE) {
5713                         ret = raid56_parity_write(root, bio, bbio,
5714                                                   raid_map, map_length);
5715                 } else {
5716                         ret = raid56_parity_recover(root, bio, bbio,
5717                                                     raid_map, map_length,
5718                                                     mirror_num);
5719                 }
5720                 /*
5721                  * FIXME, replace dosen't support raid56 yet, please fix
5722                  * it in the future.
5723                  */
5724                 btrfs_bio_counter_dec(root->fs_info);
5725                 return ret;
5726         }
5727
5728         if (map_length < length) {
5729                 btrfs_crit(root->fs_info, "mapping failed logical %llu bio len %llu len %llu",
5730                         logical, length, map_length);
5731                 BUG();
5732         }
5733
5734         while (dev_nr < total_devs) {
5735                 dev = bbio->stripes[dev_nr].dev;
5736                 if (!dev || !dev->bdev || (rw & WRITE && !dev->writeable)) {
5737                         bbio_error(bbio, first_bio, logical);
5738                         dev_nr++;
5739                         continue;
5740                 }
5741
5742                 /*
5743                  * Check and see if we're ok with this bio based on it's size
5744                  * and offset with the given device.
5745                  */
5746                 if (!bio_size_ok(dev->bdev, first_bio,
5747                                  bbio->stripes[dev_nr].physical >> 9)) {
5748                         ret = breakup_stripe_bio(root, bbio, first_bio, dev,
5749                                                  dev_nr, rw, async_submit);
5750                         BUG_ON(ret);
5751                         dev_nr++;
5752                         continue;
5753                 }
5754
5755                 if (dev_nr < total_devs - 1) {
5756                         bio = btrfs_bio_clone(first_bio, GFP_NOFS);
5757                         BUG_ON(!bio); /* -ENOMEM */
5758                 } else {
5759                         bio = first_bio;
5760                         bbio->flags |= BTRFS_BIO_ORIG_BIO_SUBMITTED;
5761                 }
5762
5763                 submit_stripe_bio(root, bbio, bio,
5764                                   bbio->stripes[dev_nr].physical, dev_nr, rw,
5765                                   async_submit);
5766                 dev_nr++;
5767         }
5768         btrfs_bio_counter_dec(root->fs_info);
5769         return 0;
5770 }
5771
5772 struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid,
5773                                        u8 *uuid, u8 *fsid)
5774 {
5775         struct btrfs_device *device;
5776         struct btrfs_fs_devices *cur_devices;
5777
5778         cur_devices = fs_info->fs_devices;
5779         while (cur_devices) {
5780                 if (!fsid ||
5781                     !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
5782                         device = __find_device(&cur_devices->devices,
5783                                                devid, uuid);
5784                         if (device)
5785                                 return device;
5786                 }
5787                 cur_devices = cur_devices->seed;
5788         }
5789         return NULL;
5790 }
5791
5792 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
5793                                             u64 devid, u8 *dev_uuid)
5794 {
5795         struct btrfs_device *device;
5796         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
5797
5798         device = btrfs_alloc_device(NULL, &devid, dev_uuid);
5799         if (IS_ERR(device))
5800                 return NULL;
5801
5802         list_add(&device->dev_list, &fs_devices->devices);
5803         device->fs_devices = fs_devices;
5804         fs_devices->num_devices++;
5805
5806         device->missing = 1;
5807         fs_devices->missing_devices++;
5808
5809         return device;
5810 }
5811
5812 /**
5813  * btrfs_alloc_device - allocate struct btrfs_device
5814  * @fs_info:    used only for generating a new devid, can be NULL if
5815  *              devid is provided (i.e. @devid != NULL).
5816  * @devid:      a pointer to devid for this device.  If NULL a new devid
5817  *              is generated.
5818  * @uuid:       a pointer to UUID for this device.  If NULL a new UUID
5819  *              is generated.
5820  *
5821  * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
5822  * on error.  Returned struct is not linked onto any lists and can be
5823  * destroyed with kfree() right away.
5824  */
5825 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
5826                                         const u64 *devid,
5827                                         const u8 *uuid)
5828 {
5829         struct btrfs_device *dev;
5830         u64 tmp;
5831
5832         if (WARN_ON(!devid && !fs_info))
5833                 return ERR_PTR(-EINVAL);
5834
5835         dev = __alloc_device();
5836         if (IS_ERR(dev))
5837                 return dev;
5838
5839         if (devid)
5840                 tmp = *devid;
5841         else {
5842                 int ret;
5843
5844                 ret = find_next_devid(fs_info, &tmp);
5845                 if (ret) {
5846                         kfree(dev);
5847                         return ERR_PTR(ret);
5848                 }
5849         }
5850         dev->devid = tmp;
5851
5852         if (uuid)
5853                 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE);
5854         else
5855                 generate_random_uuid(dev->uuid);
5856
5857         btrfs_init_work(&dev->work, btrfs_submit_helper,
5858                         pending_bios_fn, NULL, NULL);
5859
5860         return dev;
5861 }
5862
5863 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
5864                           struct extent_buffer *leaf,
5865                           struct btrfs_chunk *chunk)
5866 {
5867         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
5868         struct map_lookup *map;
5869         struct extent_map *em;
5870         u64 logical;
5871         u64 length;
5872         u64 devid;
5873         u8 uuid[BTRFS_UUID_SIZE];
5874         int num_stripes;
5875         int ret;
5876         int i;
5877
5878         logical = key->offset;
5879         length = btrfs_chunk_length(leaf, chunk);
5880
5881         read_lock(&map_tree->map_tree.lock);
5882         em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
5883         read_unlock(&map_tree->map_tree.lock);
5884
5885         /* already mapped? */
5886         if (em && em->start <= logical && em->start + em->len > logical) {
5887                 free_extent_map(em);
5888                 return 0;
5889         } else if (em) {
5890                 free_extent_map(em);
5891         }
5892
5893         em = alloc_extent_map();
5894         if (!em)
5895                 return -ENOMEM;
5896         num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
5897         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
5898         if (!map) {
5899                 free_extent_map(em);
5900                 return -ENOMEM;
5901         }
5902
5903         set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
5904         em->bdev = (struct block_device *)map;
5905         em->start = logical;
5906         em->len = length;
5907         em->orig_start = 0;
5908         em->block_start = 0;
5909         em->block_len = em->len;
5910
5911         map->num_stripes = num_stripes;
5912         map->io_width = btrfs_chunk_io_width(leaf, chunk);
5913         map->io_align = btrfs_chunk_io_align(leaf, chunk);
5914         map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
5915         map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
5916         map->type = btrfs_chunk_type(leaf, chunk);
5917         map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
5918         for (i = 0; i < num_stripes; i++) {
5919                 map->stripes[i].physical =
5920                         btrfs_stripe_offset_nr(leaf, chunk, i);
5921                 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
5922                 read_extent_buffer(leaf, uuid, (unsigned long)
5923                                    btrfs_stripe_dev_uuid_nr(chunk, i),
5924                                    BTRFS_UUID_SIZE);
5925                 map->stripes[i].dev = btrfs_find_device(root->fs_info, devid,
5926                                                         uuid, NULL);
5927                 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
5928                         free_extent_map(em);
5929                         return -EIO;
5930                 }
5931                 if (!map->stripes[i].dev) {
5932                         map->stripes[i].dev =
5933                                 add_missing_dev(root, devid, uuid);
5934                         if (!map->stripes[i].dev) {
5935                                 free_extent_map(em);
5936                                 return -EIO;
5937                         }
5938                 }
5939                 map->stripes[i].dev->in_fs_metadata = 1;
5940         }
5941
5942         write_lock(&map_tree->map_tree.lock);
5943         ret = add_extent_mapping(&map_tree->map_tree, em, 0);
5944         write_unlock(&map_tree->map_tree.lock);
5945         BUG_ON(ret); /* Tree corruption */
5946         free_extent_map(em);
5947
5948         return 0;
5949 }
5950
5951 static void fill_device_from_item(struct extent_buffer *leaf,
5952                                  struct btrfs_dev_item *dev_item,
5953                                  struct btrfs_device *device)
5954 {
5955         unsigned long ptr;
5956
5957         device->devid = btrfs_device_id(leaf, dev_item);
5958         device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
5959         device->total_bytes = device->disk_total_bytes;
5960         device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
5961         device->type = btrfs_device_type(leaf, dev_item);
5962         device->io_align = btrfs_device_io_align(leaf, dev_item);
5963         device->io_width = btrfs_device_io_width(leaf, dev_item);
5964         device->sector_size = btrfs_device_sector_size(leaf, dev_item);
5965         WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
5966         device->is_tgtdev_for_dev_replace = 0;
5967
5968         ptr = btrfs_device_uuid(dev_item);
5969         read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
5970 }
5971
5972 static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
5973 {
5974         struct btrfs_fs_devices *fs_devices;
5975         int ret;
5976
5977         BUG_ON(!mutex_is_locked(&uuid_mutex));
5978
5979         fs_devices = root->fs_info->fs_devices->seed;
5980         while (fs_devices) {
5981                 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
5982                         ret = 0;
5983                         goto out;
5984                 }
5985                 fs_devices = fs_devices->seed;
5986         }
5987
5988         fs_devices = find_fsid(fsid);
5989         if (!fs_devices) {
5990                 ret = -ENOENT;
5991                 goto out;
5992         }
5993
5994         fs_devices = clone_fs_devices(fs_devices);
5995         if (IS_ERR(fs_devices)) {
5996                 ret = PTR_ERR(fs_devices);
5997                 goto out;
5998         }
5999
6000         ret = __btrfs_open_devices(fs_devices, FMODE_READ,
6001                                    root->fs_info->bdev_holder);
6002         if (ret) {
6003                 free_fs_devices(fs_devices);
6004                 goto out;
6005         }
6006
6007         if (!fs_devices->seeding) {
6008                 __btrfs_close_devices(fs_devices);
6009                 free_fs_devices(fs_devices);
6010                 ret = -EINVAL;
6011                 goto out;
6012         }
6013
6014         fs_devices->seed = root->fs_info->fs_devices->seed;
6015         root->fs_info->fs_devices->seed = fs_devices;
6016 out:
6017         return ret;
6018 }
6019
6020 static int read_one_dev(struct btrfs_root *root,
6021                         struct extent_buffer *leaf,
6022                         struct btrfs_dev_item *dev_item)
6023 {
6024         struct btrfs_device *device;
6025         u64 devid;
6026         int ret;
6027         u8 fs_uuid[BTRFS_UUID_SIZE];
6028         u8 dev_uuid[BTRFS_UUID_SIZE];
6029
6030         devid = btrfs_device_id(leaf, dev_item);
6031         read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
6032                            BTRFS_UUID_SIZE);
6033         read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
6034                            BTRFS_UUID_SIZE);
6035
6036         if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
6037                 ret = open_seed_devices(root, fs_uuid);
6038                 if (ret && !btrfs_test_opt(root, DEGRADED))
6039                         return ret;
6040         }
6041
6042         device = btrfs_find_device(root->fs_info, devid, dev_uuid, fs_uuid);
6043         if (!device || !device->bdev) {
6044                 if (!btrfs_test_opt(root, DEGRADED))
6045                         return -EIO;
6046
6047                 if (!device) {
6048                         btrfs_warn(root->fs_info, "devid %llu missing", devid);
6049                         device = add_missing_dev(root, devid, dev_uuid);
6050                         if (!device)
6051                                 return -ENOMEM;
6052                 } else if (!device->missing) {
6053                         /*
6054                          * this happens when a device that was properly setup
6055                          * in the device info lists suddenly goes bad.
6056                          * device->bdev is NULL, and so we have to set
6057                          * device->missing to one here
6058                          */
6059                         root->fs_info->fs_devices->missing_devices++;
6060                         device->missing = 1;
6061                 }
6062         }
6063
6064         if (device->fs_devices != root->fs_info->fs_devices) {
6065                 BUG_ON(device->writeable);
6066                 if (device->generation !=
6067                     btrfs_device_generation(leaf, dev_item))
6068                         return -EINVAL;
6069         }
6070
6071         fill_device_from_item(leaf, dev_item, device);
6072         device->in_fs_metadata = 1;
6073         if (device->writeable && !device->is_tgtdev_for_dev_replace) {
6074                 device->fs_devices->total_rw_bytes += device->total_bytes;
6075                 spin_lock(&root->fs_info->free_chunk_lock);
6076                 root->fs_info->free_chunk_space += device->total_bytes -
6077                         device->bytes_used;
6078                 spin_unlock(&root->fs_info->free_chunk_lock);
6079         }
6080         ret = 0;
6081         return ret;
6082 }
6083
6084 int btrfs_read_sys_array(struct btrfs_root *root)
6085 {
6086         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
6087         struct extent_buffer *sb;
6088         struct btrfs_disk_key *disk_key;
6089         struct btrfs_chunk *chunk;
6090         u8 *ptr;
6091         unsigned long sb_ptr;
6092         int ret = 0;
6093         u32 num_stripes;
6094         u32 array_size;
6095         u32 len = 0;
6096         u32 cur;
6097         struct btrfs_key key;
6098
6099         sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
6100                                           BTRFS_SUPER_INFO_SIZE);
6101         if (!sb)
6102                 return -ENOMEM;
6103         btrfs_set_buffer_uptodate(sb);
6104         btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
6105         /*
6106          * The sb extent buffer is artifical and just used to read the system array.
6107          * btrfs_set_buffer_uptodate() call does not properly mark all it's
6108          * pages up-to-date when the page is larger: extent does not cover the
6109          * whole page and consequently check_page_uptodate does not find all
6110          * the page's extents up-to-date (the hole beyond sb),
6111          * write_extent_buffer then triggers a WARN_ON.
6112          *
6113          * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
6114          * but sb spans only this function. Add an explicit SetPageUptodate call
6115          * to silence the warning eg. on PowerPC 64.
6116          */
6117         if (PAGE_CACHE_SIZE > BTRFS_SUPER_INFO_SIZE)
6118                 SetPageUptodate(sb->pages[0]);
6119
6120         write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
6121         array_size = btrfs_super_sys_array_size(super_copy);
6122
6123         ptr = super_copy->sys_chunk_array;
6124         sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
6125         cur = 0;
6126
6127         while (cur < array_size) {
6128                 disk_key = (struct btrfs_disk_key *)ptr;
6129                 btrfs_disk_key_to_cpu(&key, disk_key);
6130
6131                 len = sizeof(*disk_key); ptr += len;
6132                 sb_ptr += len;
6133                 cur += len;
6134
6135                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
6136                         chunk = (struct btrfs_chunk *)sb_ptr;
6137                         ret = read_one_chunk(root, &key, sb, chunk);
6138                         if (ret)
6139                                 break;
6140                         num_stripes = btrfs_chunk_num_stripes(sb, chunk);
6141                         len = btrfs_chunk_item_size(num_stripes);
6142                 } else {
6143                         ret = -EIO;
6144                         break;
6145                 }
6146                 ptr += len;
6147                 sb_ptr += len;
6148                 cur += len;
6149         }
6150         free_extent_buffer(sb);
6151         return ret;
6152 }
6153
6154 int btrfs_read_chunk_tree(struct btrfs_root *root)
6155 {
6156         struct btrfs_path *path;
6157         struct extent_buffer *leaf;
6158         struct btrfs_key key;
6159         struct btrfs_key found_key;
6160         int ret;
6161         int slot;
6162
6163         root = root->fs_info->chunk_root;
6164
6165         path = btrfs_alloc_path();
6166         if (!path)
6167                 return -ENOMEM;
6168
6169         mutex_lock(&uuid_mutex);
6170         lock_chunks(root);
6171
6172         /*
6173          * Read all device items, and then all the chunk items. All
6174          * device items are found before any chunk item (their object id
6175          * is smaller than the lowest possible object id for a chunk
6176          * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
6177          */
6178         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
6179         key.offset = 0;
6180         key.type = 0;
6181         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6182         if (ret < 0)
6183                 goto error;
6184         while (1) {
6185                 leaf = path->nodes[0];
6186                 slot = path->slots[0];
6187                 if (slot >= btrfs_header_nritems(leaf)) {
6188                         ret = btrfs_next_leaf(root, path);
6189                         if (ret == 0)
6190                                 continue;
6191                         if (ret < 0)
6192                                 goto error;
6193                         break;
6194                 }
6195                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6196                 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
6197                         struct btrfs_dev_item *dev_item;
6198                         dev_item = btrfs_item_ptr(leaf, slot,
6199                                                   struct btrfs_dev_item);
6200                         ret = read_one_dev(root, leaf, dev_item);
6201                         if (ret)
6202                                 goto error;
6203                 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
6204                         struct btrfs_chunk *chunk;
6205                         chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
6206                         ret = read_one_chunk(root, &found_key, leaf, chunk);
6207                         if (ret)
6208                                 goto error;
6209                 }
6210                 path->slots[0]++;
6211         }
6212         ret = 0;
6213 error:
6214         unlock_chunks(root);
6215         mutex_unlock(&uuid_mutex);
6216
6217         btrfs_free_path(path);
6218         return ret;
6219 }
6220
6221 void btrfs_init_devices_late(struct btrfs_fs_info *fs_info)
6222 {
6223         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6224         struct btrfs_device *device;
6225
6226         while (fs_devices) {
6227                 mutex_lock(&fs_devices->device_list_mutex);
6228                 list_for_each_entry(device, &fs_devices->devices, dev_list)
6229                         device->dev_root = fs_info->dev_root;
6230                 mutex_unlock(&fs_devices->device_list_mutex);
6231
6232                 fs_devices = fs_devices->seed;
6233         }
6234 }
6235
6236 static void __btrfs_reset_dev_stats(struct btrfs_device *dev)
6237 {
6238         int i;
6239
6240         for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6241                 btrfs_dev_stat_reset(dev, i);
6242 }
6243
6244 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
6245 {
6246         struct btrfs_key key;
6247         struct btrfs_key found_key;
6248         struct btrfs_root *dev_root = fs_info->dev_root;
6249         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6250         struct extent_buffer *eb;
6251         int slot;
6252         int ret = 0;
6253         struct btrfs_device *device;
6254         struct btrfs_path *path = NULL;
6255         int i;
6256
6257         path = btrfs_alloc_path();
6258         if (!path) {
6259                 ret = -ENOMEM;
6260                 goto out;
6261         }
6262
6263         mutex_lock(&fs_devices->device_list_mutex);
6264         list_for_each_entry(device, &fs_devices->devices, dev_list) {
6265                 int item_size;
6266                 struct btrfs_dev_stats_item *ptr;
6267
6268                 key.objectid = 0;
6269                 key.type = BTRFS_DEV_STATS_KEY;
6270                 key.offset = device->devid;
6271                 ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
6272                 if (ret) {
6273                         __btrfs_reset_dev_stats(device);
6274                         device->dev_stats_valid = 1;
6275                         btrfs_release_path(path);
6276                         continue;
6277                 }
6278                 slot = path->slots[0];
6279                 eb = path->nodes[0];
6280                 btrfs_item_key_to_cpu(eb, &found_key, slot);
6281                 item_size = btrfs_item_size_nr(eb, slot);
6282
6283                 ptr = btrfs_item_ptr(eb, slot,
6284                                      struct btrfs_dev_stats_item);
6285
6286                 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
6287                         if (item_size >= (1 + i) * sizeof(__le64))
6288                                 btrfs_dev_stat_set(device, i,
6289                                         btrfs_dev_stats_value(eb, ptr, i));
6290                         else
6291                                 btrfs_dev_stat_reset(device, i);
6292                 }
6293
6294                 device->dev_stats_valid = 1;
6295                 btrfs_dev_stat_print_on_load(device);
6296                 btrfs_release_path(path);
6297         }
6298         mutex_unlock(&fs_devices->device_list_mutex);
6299
6300 out:
6301         btrfs_free_path(path);
6302         return ret < 0 ? ret : 0;
6303 }
6304
6305 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
6306                                 struct btrfs_root *dev_root,
6307                                 struct btrfs_device *device)
6308 {
6309         struct btrfs_path *path;
6310         struct btrfs_key key;
6311         struct extent_buffer *eb;
6312         struct btrfs_dev_stats_item *ptr;
6313         int ret;
6314         int i;
6315
6316         key.objectid = 0;
6317         key.type = BTRFS_DEV_STATS_KEY;
6318         key.offset = device->devid;
6319
6320         path = btrfs_alloc_path();
6321         BUG_ON(!path);
6322         ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
6323         if (ret < 0) {
6324                 printk_in_rcu(KERN_WARNING "BTRFS: "
6325                         "error %d while searching for dev_stats item for device %s!\n",
6326                               ret, rcu_str_deref(device->name));
6327                 goto out;
6328         }
6329
6330         if (ret == 0 &&
6331             btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
6332                 /* need to delete old one and insert a new one */
6333                 ret = btrfs_del_item(trans, dev_root, path);
6334                 if (ret != 0) {
6335                         printk_in_rcu(KERN_WARNING "BTRFS: "
6336                                 "delete too small dev_stats item for device %s failed %d!\n",
6337                                       rcu_str_deref(device->name), ret);
6338                         goto out;
6339                 }
6340                 ret = 1;
6341         }
6342
6343         if (ret == 1) {
6344                 /* need to insert a new item */
6345                 btrfs_release_path(path);
6346                 ret = btrfs_insert_empty_item(trans, dev_root, path,
6347                                               &key, sizeof(*ptr));
6348                 if (ret < 0) {
6349                         printk_in_rcu(KERN_WARNING "BTRFS: "
6350                                           "insert dev_stats item for device %s failed %d!\n",
6351                                       rcu_str_deref(device->name), ret);
6352                         goto out;
6353                 }
6354         }
6355
6356         eb = path->nodes[0];
6357         ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
6358         for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6359                 btrfs_set_dev_stats_value(eb, ptr, i,
6360                                           btrfs_dev_stat_read(device, i));
6361         btrfs_mark_buffer_dirty(eb);
6362
6363 out:
6364         btrfs_free_path(path);
6365         return ret;
6366 }
6367
6368 /*
6369  * called from commit_transaction. Writes all changed device stats to disk.
6370  */
6371 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans,
6372                         struct btrfs_fs_info *fs_info)
6373 {
6374         struct btrfs_root *dev_root = fs_info->dev_root;
6375         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6376         struct btrfs_device *device;
6377         int ret = 0;
6378
6379         mutex_lock(&fs_devices->device_list_mutex);
6380         list_for_each_entry(device, &fs_devices->devices, dev_list) {
6381                 if (!device->dev_stats_valid || !device->dev_stats_dirty)
6382                         continue;
6383
6384                 ret = update_dev_stat_item(trans, dev_root, device);
6385                 if (!ret)
6386                         device->dev_stats_dirty = 0;
6387         }
6388         mutex_unlock(&fs_devices->device_list_mutex);
6389
6390         return ret;
6391 }
6392
6393 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
6394 {
6395         btrfs_dev_stat_inc(dev, index);
6396         btrfs_dev_stat_print_on_error(dev);
6397 }
6398
6399 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
6400 {
6401         if (!dev->dev_stats_valid)
6402                 return;
6403         printk_ratelimited_in_rcu(KERN_ERR "BTRFS: "
6404                            "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
6405                            rcu_str_deref(dev->name),
6406                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
6407                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
6408                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
6409                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
6410                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
6411 }
6412
6413 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
6414 {
6415         int i;
6416
6417         for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6418                 if (btrfs_dev_stat_read(dev, i) != 0)
6419                         break;
6420         if (i == BTRFS_DEV_STAT_VALUES_MAX)
6421                 return; /* all values == 0, suppress message */
6422
6423         printk_in_rcu(KERN_INFO "BTRFS: "
6424                    "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
6425                rcu_str_deref(dev->name),
6426                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
6427                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
6428                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
6429                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
6430                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
6431 }
6432
6433 int btrfs_get_dev_stats(struct btrfs_root *root,
6434                         struct btrfs_ioctl_get_dev_stats *stats)
6435 {
6436         struct btrfs_device *dev;
6437         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
6438         int i;
6439
6440         mutex_lock(&fs_devices->device_list_mutex);
6441         dev = btrfs_find_device(root->fs_info, stats->devid, NULL, NULL);
6442         mutex_unlock(&fs_devices->device_list_mutex);
6443
6444         if (!dev) {
6445                 btrfs_warn(root->fs_info, "get dev_stats failed, device not found");
6446                 return -ENODEV;
6447         } else if (!dev->dev_stats_valid) {
6448                 btrfs_warn(root->fs_info, "get dev_stats failed, not yet valid");
6449                 return -ENODEV;
6450         } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
6451                 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
6452                         if (stats->nr_items > i)
6453                                 stats->values[i] =
6454                                         btrfs_dev_stat_read_and_reset(dev, i);
6455                         else
6456                                 btrfs_dev_stat_reset(dev, i);
6457                 }
6458         } else {
6459                 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6460                         if (stats->nr_items > i)
6461                                 stats->values[i] = btrfs_dev_stat_read(dev, i);
6462         }
6463         if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
6464                 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
6465         return 0;
6466 }
6467
6468 int btrfs_scratch_superblock(struct btrfs_device *device)
6469 {
6470         struct buffer_head *bh;
6471         struct btrfs_super_block *disk_super;
6472
6473         bh = btrfs_read_dev_super(device->bdev);
6474         if (!bh)
6475                 return -EINVAL;
6476         disk_super = (struct btrfs_super_block *)bh->b_data;
6477
6478         memset(&disk_super->magic, 0, sizeof(disk_super->magic));
6479         set_buffer_dirty(bh);
6480         sync_dirty_buffer(bh);
6481         brelse(bh);
6482
6483         return 0;
6484 }