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 DEFINE_MUTEX(uuid_mutex);
  54 static LIST_HEAD(fs_uuids);
  55 struct list_head *btrfs_get_fs_uuids(void)
  56 {
  57         return &fs_uuids;
  58 }
  59
  60 static struct btrfs_fs_devices *__alloc_fs_devices(void)
  61 {
  62         struct btrfs_fs_devices *fs_devs;
  63
  64         fs_devs = kzalloc(sizeof(*fs_devs), GFP_NOFS);
  65         if (!fs_devs)
  66                 return ERR_PTR(-ENOMEM);
  67
  68         mutex_init(&fs_devs->device_list_mutex);
  69
  70         INIT_LIST_HEAD(&fs_devs->devices);
  71         INIT_LIST_HEAD(&fs_devs->resized_devices);
  72         INIT_LIST_HEAD(&fs_devs->alloc_list);
  73         INIT_LIST_HEAD(&fs_devs->list);
  74
  75         return fs_devs;
  76 }
  77
  78 /**
  79  * alloc_fs_devices - allocate struct btrfs_fs_devices
  80  * @fsid:       a pointer to UUID for this FS.  If NULL a new UUID is
  81  *              generated.
  82  *
  83  * Return: a pointer to a new &struct btrfs_fs_devices on success;
  84  * ERR_PTR() on error.  Returned struct is not linked onto any lists and
  85  * can be destroyed with kfree() right away.
  86  */
  87 static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid)
  88 {
  89         struct btrfs_fs_devices *fs_devs;
  90
  91         fs_devs = __alloc_fs_devices();
  92         if (IS_ERR(fs_devs))
  93                 return fs_devs;
  94
  95         if (fsid)
  96                 memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE);
  97         else
  98                 generate_random_uuid(fs_devs->fsid);
  99
 100         return fs_devs;
 101 }
 102
 103 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
 104 {
 105         struct btrfs_device *device;
 106         WARN_ON(fs_devices->opened);
 107         while (!list_empty(&fs_devices->devices)) {
 108                 device = list_entry(fs_devices->devices.next,
 109                                     struct btrfs_device, dev_list);
 110                 list_del(&device->dev_list);
 111                 rcu_string_free(device->name);
 112                 kfree(device);
 113         }
 114         kfree(fs_devices);
 115 }
 116
 117 static void btrfs_kobject_uevent(struct block_device *bdev,
 118                                  enum kobject_action action)
 119 {
 120         int ret;
 121
 122         ret = kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, action);
 123         if (ret)
 124                 pr_warn("BTRFS: Sending event '%d' to kobject: '%s' (%p): failed\n",
 125                         action,
 126                         kobject_name(&disk_to_dev(bdev->bd_disk)->kobj),
 127                         &disk_to_dev(bdev->bd_disk)->kobj);
 128 }
 129
 130 void btrfs_cleanup_fs_uuids(void)
 131 {
 132         struct btrfs_fs_devices *fs_devices;
 133
 134         while (!list_empty(&fs_uuids)) {
 135                 fs_devices = list_entry(fs_uuids.next,
 136                                         struct btrfs_fs_devices, list);
 137                 list_del(&fs_devices->list);
 138                 free_fs_devices(fs_devices);
 139         }
 140 }
 141
 142 static struct btrfs_device *__alloc_device(void)
 143 {
 144         struct btrfs_device *dev;
 145
 146         dev = kzalloc(sizeof(*dev), GFP_NOFS);
 147         if (!dev)
 148                 return ERR_PTR(-ENOMEM);
 149
 150         INIT_LIST_HEAD(&dev->dev_list);
 151         INIT_LIST_HEAD(&dev->dev_alloc_list);
 152         INIT_LIST_HEAD(&dev->resized_list);
 153
 154         spin_lock_init(&dev->io_lock);
 155
 156         spin_lock_init(&dev->reada_lock);
 157         atomic_set(&dev->reada_in_flight, 0);
 158         atomic_set(&dev->dev_stats_ccnt, 0);
 159         INIT_RADIX_TREE(&dev->reada_zones, GFP_NOFS & ~__GFP_WAIT);
 160         INIT_RADIX_TREE(&dev->reada_extents, GFP_NOFS & ~__GFP_WAIT);
 161
 162         return dev;
 163 }
 164
 165 static noinline struct btrfs_device *__find_device(struct list_head *head,
 166                                                    u64 devid, u8 *uuid)
 167 {
 168         struct btrfs_device *dev;
 169
 170         list_for_each_entry(dev, head, dev_list) {
 171                 if (dev->devid == devid &&
 172                     (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
 173                         return dev;
 174                 }
 175         }
 176         return NULL;
 177 }
 178
 179 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
 180 {
 181         struct btrfs_fs_devices *fs_devices;
 182
 183         list_for_each_entry(fs_devices, &fs_uuids, list) {
 184                 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
 185                         return fs_devices;
 186         }
 187         return NULL;
 188 }
 189
 190 static int
 191 btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder,
 192                       int flush, struct block_device **bdev,
 193                       struct buffer_head **bh)
 194 {
 195         int ret;
 196
 197         *bdev = blkdev_get_by_path(device_path, flags, holder);
 198
 199         if (IS_ERR(*bdev)) {
 200                 ret = PTR_ERR(*bdev);
 201                 printk(KERN_INFO "BTRFS: open %s failed\n", device_path);
 202                 goto error;
 203         }
 204
 205         if (flush)
 206                 filemap_write_and_wait((*bdev)->bd_inode->i_mapping);
 207         ret = set_blocksize(*bdev, 4096);
 208         if (ret) {
 209                 blkdev_put(*bdev, flags);
 210                 goto error;
 211         }
 212         invalidate_bdev(*bdev);
 213         *bh = btrfs_read_dev_super(*bdev);
 214         if (!*bh) {
 215                 ret = -EINVAL;
 216                 blkdev_put(*bdev, flags);
 217                 goto error;
 218         }
 219
 220         return 0;
 221
 222 error:
 223         *bdev = NULL;
 224         *bh = NULL;
 225         return ret;
 226 }
 227
 228 static void requeue_list(struct btrfs_pending_bios *pending_bios,
 229                         struct bio *head, struct bio *tail)
 230 {
 231
 232         struct bio *old_head;
 233
 234         old_head = pending_bios->head;
 235         pending_bios->head = head;
 236         if (pending_bios->tail)
 237                 tail->bi_next = old_head;
 238         else
 239                 pending_bios->tail = tail;
 240 }
 241
 242 /*
 243  * we try to collect pending bios for a device so we don't get a large
 244  * number of procs sending bios down to the same device.  This greatly
 245  * improves the schedulers ability to collect and merge the bios.
 246  *
 247  * But, it also turns into a long list of bios to process and that is sure
 248  * to eventually make the worker thread block.  The solution here is to
 249  * make some progress and then put this work struct back at the end of
 250  * the list if the block device is congested.  This way, multiple devices
 251  * can make progress from a single worker thread.
 252  */
 253 static noinline void run_scheduled_bios(struct btrfs_device *device)
 254 {
 255         struct bio *pending;
 256         struct backing_dev_info *bdi;
 257         struct btrfs_fs_info *fs_info;
 258         struct btrfs_pending_bios *pending_bios;
 259         struct bio *tail;
 260         struct bio *cur;
 261         int again = 0;
 262         unsigned long num_run;
 263         unsigned long batch_run = 0;
 264         unsigned long limit;
 265         unsigned long last_waited = 0;
 266         int force_reg = 0;
 267         int sync_pending = 0;
 268         struct blk_plug plug;
 269
 270         /*
 271          * this function runs all the bios we've collected for
 272          * a particular device.  We don't want to wander off to
 273          * another device without first sending all of these down.
 274          * So, setup a plug here and finish it off before we return
 275          */
 276         blk_start_plug(&plug);
 277
 278         bdi = blk_get_backing_dev_info(device->bdev);
 279         fs_info = device->dev_root->fs_info;
 280         limit = btrfs_async_submit_limit(fs_info);
 281         limit = limit * 2 / 3;
 282
 283 loop:
 284         spin_lock(&device->io_lock);
 285
 286 loop_lock:
 287         num_run = 0;
 288
 289         /* take all the bios off the list at once and process them
 290          * later on (without the lock held).  But, remember the
 291          * tail and other pointers so the bios can be properly reinserted
 292          * into the list if we hit congestion
 293          */
 294         if (!force_reg && device->pending_sync_bios.head) {
 295                 pending_bios = &device->pending_sync_bios;
 296                 force_reg = 1;
 297         } else {
 298                 pending_bios = &device->pending_bios;
 299                 force_reg = 0;
 300         }
 301
 302         pending = pending_bios->head;
 303         tail = pending_bios->tail;
 304         WARN_ON(pending && !tail);
 305
 306         /*
 307          * if pending was null this time around, no bios need processing
 308          * at all and we can stop.  Otherwise it'll loop back up again
 309          * and do an additional check so no bios are missed.
 310          *
 311          * device->running_pending is used to synchronize with the
 312          * schedule_bio code.
 313          */
 314         if (device->pending_sync_bios.head == NULL &&
 315             device->pending_bios.head == NULL) {
 316                 again = 0;
 317                 device->running_pending = 0;
 318         } else {
 319                 again = 1;
 320                 device->running_pending = 1;
 321         }
 322
 323         pending_bios->head = NULL;
 324         pending_bios->tail = NULL;
 325
 326         spin_unlock(&device->io_lock);
 327
 328         while (pending) {
 329
 330                 rmb();
 331                 /* we want to work on both lists, but do more bios on the
 332                  * sync list than the regular list
 333                  */
 334                 if ((num_run > 32 &&
 335                     pending_bios != &device->pending_sync_bios &&
 336                     device->pending_sync_bios.head) ||
 337                    (num_run > 64 && pending_bios == &device->pending_sync_bios &&
 338                     device->pending_bios.head)) {
 339                         spin_lock(&device->io_lock);
 340                         requeue_list(pending_bios, pending, tail);
 341                         goto loop_lock;
 342                 }
 343
 344                 cur = pending;
 345                 pending = pending->bi_next;
 346                 cur->bi_next = NULL;
 347
 348                 if (atomic_dec_return(&fs_info->nr_async_bios) < limit &&
 349                     waitqueue_active(&fs_info->async_submit_wait))
 350                         wake_up(&fs_info->async_submit_wait);
 351
 352                 BUG_ON(atomic_read(&cur->__bi_cnt) == 0);
 353
 354                 /*
 355                  * if we're doing the sync list, record that our
 356                  * plug has some sync requests on it
 357                  *
 358                  * If we're doing the regular list and there are
 359                  * sync requests sitting around, unplug before
 360                  * we add more
 361                  */
 362                 if (pending_bios == &device->pending_sync_bios) {
 363                         sync_pending = 1;
 364                 } else if (sync_pending) {
 365                         blk_finish_plug(&plug);
 366                         blk_start_plug(&plug);
 367                         sync_pending = 0;
 368                 }
 369
 370                 btrfsic_submit_bio(cur->bi_rw, cur);
 371                 num_run++;
 372                 batch_run++;
 373
 374                 cond_resched();
 375
 376                 /*
 377                  * we made progress, there is more work to do and the bdi
 378                  * is now congested.  Back off and let other work structs
 379                  * run instead
 380                  */
 381                 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
 382                     fs_info->fs_devices->open_devices > 1) {
 383                         struct io_context *ioc;
 384
 385                         ioc = current->io_context;
 386
 387                         /*
 388                          * the main goal here is that we don't want to
 389                          * block if we're going to be able to submit
 390                          * more requests without blocking.
 391                          *
 392                          * This code does two great things, it pokes into
 393                          * the elevator code from a filesystem _and_
 394                          * it makes assumptions about how batching works.
 395                          */
 396                         if (ioc && ioc->nr_batch_requests > 0 &&
 397                             time_before(jiffies, ioc->last_waited + HZ/50UL) &&
 398                             (last_waited == 0 ||
 399                              ioc->last_waited == last_waited)) {
 400                                 /*
 401                                  * we want to go through our batch of
 402                                  * requests and stop.  So, we copy out
 403                                  * the ioc->last_waited time and test
 404                                  * against it before looping
 405                                  */
 406                                 last_waited = ioc->last_waited;
 407                                 cond_resched();
 408                                 continue;
 409                         }
 410                         spin_lock(&device->io_lock);
 411                         requeue_list(pending_bios, pending, tail);
 412                         device->running_pending = 1;
 413
 414                         spin_unlock(&device->io_lock);
 415                         btrfs_queue_work(fs_info->submit_workers,
 416                                          &device->work);
 417                         goto done;
 418                 }
 419                 /* unplug every 64 requests just for good measure */
 420                 if (batch_run % 64 == 0) {
 421                         blk_finish_plug(&plug);
 422                         blk_start_plug(&plug);
 423                         sync_pending = 0;
 424                 }
 425         }
 426
 427         cond_resched();
 428         if (again)
 429                 goto loop;
 430
 431         spin_lock(&device->io_lock);
 432         if (device->pending_bios.head || device->pending_sync_bios.head)
 433                 goto loop_lock;
 434         spin_unlock(&device->io_lock);
 435
 436 done:
 437         blk_finish_plug(&plug);
 438 }
 439
 440 static void pending_bios_fn(struct btrfs_work *work)
 441 {
 442         struct btrfs_device *device;
 443
 444         device = container_of(work, struct btrfs_device, work);
 445         run_scheduled_bios(device);
 446 }
 447
 448
 449 void btrfs_free_stale_device(struct btrfs_device *cur_dev)
 450 {
 451         struct btrfs_fs_devices *fs_devs;
 452         struct btrfs_device *dev;
 453
 454         if (!cur_dev->name)
 455                 return;
 456
 457         list_for_each_entry(fs_devs, &fs_uuids, list) {
 458                 int del = 1;
 459
 460                 if (fs_devs->opened)
 461                         continue;
 462                 if (fs_devs->seeding)
 463                         continue;
 464
 465                 list_for_each_entry(dev, &fs_devs->devices, dev_list) {
 466
 467                         if (dev == cur_dev)
 468                                 continue;
 469                         if (!dev->name)
 470                                 continue;
 471
 472                         /*
 473                          * Todo: This won't be enough. What if the same device
 474                          * comes back (with new uuid and) with its mapper path?
 475                          * But for now, this does help as mostly an admin will
 476                          * either use mapper or non mapper path throughout.
 477                          */
 478                         rcu_read_lock();
 479                         del = strcmp(rcu_str_deref(dev->name),
 480                                                 rcu_str_deref(cur_dev->name));
 481                         rcu_read_unlock();
 482                         if (!del)
 483                                 break;
 484                 }
 485
 486                 if (!del) {
 487                         /* delete the stale device */
 488                         if (fs_devs->num_devices == 1) {
 489                                 btrfs_sysfs_remove_fsid(fs_devs);
 490                                 list_del(&fs_devs->list);
 491                                 free_fs_devices(fs_devs);
 492                         } else {
 493                                 fs_devs->num_devices--;
 494                                 list_del(&dev->dev_list);
 495                                 rcu_string_free(dev->name);
 496                                 kfree(dev);
 497                         }
 498                         break;
 499                 }
 500         }
 501 }
 502
 503 /*
 504  * Add new device to list of registered devices
 505  *
 506  * Returns:
 507  * 1   - first time device is seen
 508  * 0   - device already known
 509  * < 0 - error
 510  */
 511 static noinline int device_list_add(const char *path,
 512                            struct btrfs_super_block *disk_super,
 513                            u64 devid, struct btrfs_fs_devices **fs_devices_ret)
 514 {
 515         struct btrfs_device *device;
 516         struct btrfs_fs_devices *fs_devices;
 517         struct rcu_string *name;
 518         int ret = 0;
 519         u64 found_transid = btrfs_super_generation(disk_super);
 520
 521         fs_devices = find_fsid(disk_super->fsid);
 522         if (!fs_devices) {
 523                 fs_devices = alloc_fs_devices(disk_super->fsid);
 524                 if (IS_ERR(fs_devices))
 525                         return PTR_ERR(fs_devices);
 526
 527                 list_add(&fs_devices->list, &fs_uuids);
 528
 529                 device = NULL;
 530         } else {
 531                 device = __find_device(&fs_devices->devices, devid,
 532                                        disk_super->dev_item.uuid);
 533         }
 534
 535         if (!device) {
 536                 if (fs_devices->opened)
 537                         return -EBUSY;
 538
 539                 device = btrfs_alloc_device(NULL, &devid,
 540                                             disk_super->dev_item.uuid);
 541                 if (IS_ERR(device)) {
 542                         /* we can safely leave the fs_devices entry around */
 543                         return PTR_ERR(device);
 544                 }
 545
 546                 name = rcu_string_strdup(path, GFP_NOFS);
 547                 if (!name) {
 548                         kfree(device);
 549                         return -ENOMEM;
 550                 }
 551                 rcu_assign_pointer(device->name, name);
 552
 553                 mutex_lock(&fs_devices->device_list_mutex);
 554                 list_add_rcu(&device->dev_list, &fs_devices->devices);
 555                 fs_devices->num_devices++;
 556                 mutex_unlock(&fs_devices->device_list_mutex);
 557
 558                 ret = 1;
 559                 device->fs_devices = fs_devices;
 560         } else if (!device->name || strcmp(device->name->str, path)) {
 561                 /*
 562                  * When FS is already mounted.
 563                  * 1. If you are here and if the device->name is NULL that
 564                  *    means this device was missing at time of FS mount.
 565                  * 2. If you are here and if the device->name is different
 566                  *    from 'path' that means either
 567                  *      a. The same device disappeared and reappeared with
 568                  *         different name. or
 569                  *      b. The missing-disk-which-was-replaced, has
 570                  *         reappeared now.
 571                  *
 572                  * We must allow 1 and 2a above. But 2b would be a spurious
 573                  * and unintentional.
 574                  *
 575                  * Further in case of 1 and 2a above, the disk at 'path'
 576                  * would have missed some transaction when it was away and
 577                  * in case of 2a the stale bdev has to be updated as well.
 578                  * 2b must not be allowed at all time.
 579                  */
 580
 581                 /*
 582                  * For now, we do allow update to btrfs_fs_device through the
 583                  * btrfs dev scan cli after FS has been mounted.  We're still
 584                  * tracking a problem where systems fail mount by subvolume id
 585                  * when we reject replacement on a mounted FS.
 586                  */
 587                 if (!fs_devices->opened && found_transid < device->generation) {
 588                         /*
 589                          * That is if the FS is _not_ mounted and if you
 590                          * are here, that means there is more than one
 591                          * disk with same uuid and devid.We keep the one
 592                          * with larger generation number or the last-in if
 593                          * generation are equal.
 594                          */
 595                         return -EEXIST;
 596                 }
 597
 598                 name = rcu_string_strdup(path, GFP_NOFS);
 599                 if (!name)
 600                         return -ENOMEM;
 601                 rcu_string_free(device->name);
 602                 rcu_assign_pointer(device->name, name);
 603                 if (device->missing) {
 604                         fs_devices->missing_devices--;
 605                         device->missing = 0;
 606                 }
 607         }
 608
 609         /*
 610          * Unmount does not free the btrfs_device struct but would zero
 611          * generation along with most of the other members. So just update
 612          * it back. We need it to pick the disk with largest generation
 613          * (as above).
 614          */
 615         if (!fs_devices->opened)
 616                 device->generation = found_transid;
 617
 618         /*
 619          * if there is new btrfs on an already registered device,
 620          * then remove the stale device entry.
 621          */
 622         btrfs_free_stale_device(device);
 623
 624         *fs_devices_ret = fs_devices;
 625
 626         return ret;
 627 }
 628
 629 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
 630 {
 631         struct btrfs_fs_devices *fs_devices;
 632         struct btrfs_device *device;
 633         struct btrfs_device *orig_dev;
 634
 635         fs_devices = alloc_fs_devices(orig->fsid);
 636         if (IS_ERR(fs_devices))
 637                 return fs_devices;
 638
 639         mutex_lock(&orig->device_list_mutex);
 640         fs_devices->total_devices = orig->total_devices;
 641
 642         /* We have held the volume lock, it is safe to get the devices. */
 643         list_for_each_entry(orig_dev, &orig->devices, dev_list) {
 644                 struct rcu_string *name;
 645
 646                 device = btrfs_alloc_device(NULL, &orig_dev->devid,
 647                                             orig_dev->uuid);
 648                 if (IS_ERR(device))
 649                         goto error;
 650
 651                 /*
 652                  * This is ok to do without rcu read locked because we hold the
 653                  * uuid mutex so nothing we touch in here is going to disappear.
 654                  */
 655                 if (orig_dev->name) {
 656                         name = rcu_string_strdup(orig_dev->name->str, GFP_NOFS);
 657                         if (!name) {
 658                                 kfree(device);
 659                                 goto error;
 660                         }
 661                         rcu_assign_pointer(device->name, name);
 662                 }
 663
 664                 list_add(&device->dev_list, &fs_devices->devices);
 665                 device->fs_devices = fs_devices;
 666                 fs_devices->num_devices++;
 667         }
 668         mutex_unlock(&orig->device_list_mutex);
 669         return fs_devices;
 670 error:
 671         mutex_unlock(&orig->device_list_mutex);
 672         free_fs_devices(fs_devices);
 673         return ERR_PTR(-ENOMEM);
 674 }
 675
 676 void btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices, int step)
 677 {
 678         struct btrfs_device *device, *next;
 679         struct btrfs_device *latest_dev = NULL;
 680
 681         mutex_lock(&uuid_mutex);
 682 again:
 683         /* This is the initialized path, it is safe to release the devices. */
 684         list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
 685                 if (device->in_fs_metadata) {
 686                         if (!device->is_tgtdev_for_dev_replace &&
 687                             (!latest_dev ||
 688                              device->generation > latest_dev->generation)) {
 689                                 latest_dev = device;
 690                         }
 691                         continue;
 692                 }
 693
 694                 if (device->devid == BTRFS_DEV_REPLACE_DEVID) {
 695                         /*
 696                          * In the first step, keep the device which has
 697                          * the correct fsid and the devid that is used
 698                          * for the dev_replace procedure.
 699                          * In the second step, the dev_replace state is
 700                          * read from the device tree and it is known
 701                          * whether the procedure is really active or
 702                          * not, which means whether this device is
 703                          * used or whether it should be removed.
 704                          */
 705                         if (step == 0 || device->is_tgtdev_for_dev_replace) {
 706                                 continue;
 707                         }
 708                 }
 709                 if (device->bdev) {
 710                         blkdev_put(device->bdev, device->mode);
 711                         device->bdev = NULL;
 712                         fs_devices->open_devices--;
 713                 }
 714                 if (device->writeable) {
 715                         list_del_init(&device->dev_alloc_list);
 716                         device->writeable = 0;
 717                         if (!device->is_tgtdev_for_dev_replace)
 718                                 fs_devices->rw_devices--;
 719                 }
 720                 list_del_init(&device->dev_list);
 721                 fs_devices->num_devices--;
 722                 rcu_string_free(device->name);
 723                 kfree(device);
 724         }
 725
 726         if (fs_devices->seed) {
 727                 fs_devices = fs_devices->seed;
 728                 goto again;
 729         }
 730
 731         fs_devices->latest_bdev = latest_dev->bdev;
 732
 733         mutex_unlock(&uuid_mutex);
 734 }
 735
 736 static void __free_device(struct work_struct *work)
 737 {
 738         struct btrfs_device *device;
 739
 740         device = container_of(work, struct btrfs_device, rcu_work);
 741
 742         if (device->bdev)
 743                 blkdev_put(device->bdev, device->mode);
 744
 745         rcu_string_free(device->name);
 746         kfree(device);
 747 }
 748
 749 static void free_device(struct rcu_head *head)
 750 {
 751         struct btrfs_device *device;
 752
 753         device = container_of(head, struct btrfs_device, rcu);
 754
 755         INIT_WORK(&device->rcu_work, __free_device);
 756         schedule_work(&device->rcu_work);
 757 }
 758
 759 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
 760 {
 761         struct btrfs_device *device, *tmp;
 762
 763         if (--fs_devices->opened > 0)
 764                 return 0;
 765
 766         mutex_lock(&fs_devices->device_list_mutex);
 767         list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list) {
 768                 struct btrfs_device *new_device;
 769                 struct rcu_string *name;
 770
 771                 if (device->bdev)
 772                         fs_devices->open_devices--;
 773
 774                 if (device->writeable &&
 775                     device->devid != BTRFS_DEV_REPLACE_DEVID) {
 776                         list_del_init(&device->dev_alloc_list);
 777                         fs_devices->rw_devices--;
 778                 }
 779
 780                 if (device->missing)
 781                         fs_devices->missing_devices--;
 782
 783                 new_device = btrfs_alloc_device(NULL, &device->devid,
 784                                                 device->uuid);
 785                 BUG_ON(IS_ERR(new_device)); /* -ENOMEM */
 786
 787                 /* Safe because we are under uuid_mutex */
 788                 if (device->name) {
 789                         name = rcu_string_strdup(device->name->str, GFP_NOFS);
 790                         BUG_ON(!name); /* -ENOMEM */
 791                         rcu_assign_pointer(new_device->name, name);
 792                 }
 793
 794                 list_replace_rcu(&device->dev_list, &new_device->dev_list);
 795                 new_device->fs_devices = device->fs_devices;
 796
 797                 call_rcu(&device->rcu, free_device);
 798         }
 799         mutex_unlock(&fs_devices->device_list_mutex);
 800
 801         WARN_ON(fs_devices->open_devices);
 802         WARN_ON(fs_devices->rw_devices);
 803         fs_devices->opened = 0;
 804         fs_devices->seeding = 0;
 805
 806         return 0;
 807 }
 808
 809 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
 810 {
 811         struct btrfs_fs_devices *seed_devices = NULL;
 812         int ret;
 813
 814         mutex_lock(&uuid_mutex);
 815         ret = __btrfs_close_devices(fs_devices);
 816         if (!fs_devices->opened) {
 817                 seed_devices = fs_devices->seed;
 818                 fs_devices->seed = NULL;
 819         }
 820         mutex_unlock(&uuid_mutex);
 821
 822         while (seed_devices) {
 823                 fs_devices = seed_devices;
 824                 seed_devices = fs_devices->seed;
 825                 __btrfs_close_devices(fs_devices);
 826                 free_fs_devices(fs_devices);
 827         }
 828         /*
 829          * Wait for rcu kworkers under __btrfs_close_devices
 830          * to finish all blkdev_puts so device is really
 831          * free when umount is done.
 832          */
 833         rcu_barrier();
 834         return ret;
 835 }
 836
 837 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
 838                                 fmode_t flags, void *holder)
 839 {
 840         struct request_queue *q;
 841         struct block_device *bdev;
 842         struct list_head *head = &fs_devices->devices;
 843         struct btrfs_device *device;
 844         struct btrfs_device *latest_dev = NULL;
 845         struct buffer_head *bh;
 846         struct btrfs_super_block *disk_super;
 847         u64 devid;
 848         int seeding = 1;
 849         int ret = 0;
 850
 851         flags |= FMODE_EXCL;
 852
 853         list_for_each_entry(device, head, dev_list) {
 854                 if (device->bdev)
 855                         continue;
 856                 if (!device->name)
 857                         continue;
 858
 859                 /* Just open everything we can; ignore failures here */
 860                 if (btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1,
 861                                             &bdev, &bh))
 862                         continue;
 863
 864                 disk_super = (struct btrfs_super_block *)bh->b_data;
 865                 devid = btrfs_stack_device_id(&disk_super->dev_item);
 866                 if (devid != device->devid)
 867                         goto error_brelse;
 868
 869                 if (memcmp(device->uuid, disk_super->dev_item.uuid,
 870                            BTRFS_UUID_SIZE))
 871                         goto error_brelse;
 872
 873                 device->generation = btrfs_super_generation(disk_super);
 874                 if (!latest_dev ||
 875                     device->generation > latest_dev->generation)
 876                         latest_dev = device;
 877
 878                 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
 879                         device->writeable = 0;
 880                 } else {
 881                         device->writeable = !bdev_read_only(bdev);
 882                         seeding = 0;
 883                 }
 884
 885                 q = bdev_get_queue(bdev);
 886                 if (blk_queue_discard(q))
 887                         device->can_discard = 1;
 888
 889                 device->bdev = bdev;
 890                 device->in_fs_metadata = 0;
 891                 device->mode = flags;
 892
 893                 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
 894                         fs_devices->rotating = 1;
 895
 896                 fs_devices->open_devices++;
 897                 if (device->writeable &&
 898                     device->devid != BTRFS_DEV_REPLACE_DEVID) {
 899                         fs_devices->rw_devices++;
 900                         list_add(&device->dev_alloc_list,
 901                                  &fs_devices->alloc_list);
 902                 }
 903                 brelse(bh);
 904                 continue;
 905
 906 error_brelse:
 907                 brelse(bh);
 908                 blkdev_put(bdev, flags);
 909                 continue;
 910         }
 911         if (fs_devices->open_devices == 0) {
 912                 ret = -EINVAL;
 913                 goto out;
 914         }
 915         fs_devices->seeding = seeding;
 916         fs_devices->opened = 1;
 917         fs_devices->latest_bdev = latest_dev->bdev;
 918         fs_devices->total_rw_bytes = 0;
 919 out:
 920         return ret;
 921 }
 922
 923 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
 924                        fmode_t flags, void *holder)
 925 {
 926         int ret;
 927
 928         mutex_lock(&uuid_mutex);
 929         if (fs_devices->opened) {
 930                 fs_devices->opened++;
 931                 ret = 0;
 932         } else {
 933                 ret = __btrfs_open_devices(fs_devices, flags, holder);
 934         }
 935         mutex_unlock(&uuid_mutex);
 936         return ret;
 937 }
 938
 939 /*
 940  * Look for a btrfs signature on a device. This may be called out of the mount path
 941  * and we are not allowed to call set_blocksize during the scan. The superblock
 942  * is read via pagecache
 943  */
 944 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
 945                           struct btrfs_fs_devices **fs_devices_ret)
 946 {
 947         struct btrfs_super_block *disk_super;
 948         struct block_device *bdev;
 949         struct page *page;
 950         void *p;
 951         int ret = -EINVAL;
 952         u64 devid;
 953         u64 transid;
 954         u64 total_devices;
 955         u64 bytenr;
 956         pgoff_t index;
 957
 958         /*
 959          * we would like to check all the supers, but that would make
 960          * a btrfs mount succeed after a mkfs from a different FS.
 961          * So, we need to add a special mount option to scan for
 962          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
 963          */
 964         bytenr = btrfs_sb_offset(0);
 965         flags |= FMODE_EXCL;
 966         mutex_lock(&uuid_mutex);
 967
 968         bdev = blkdev_get_by_path(path, flags, holder);
 969
 970         if (IS_ERR(bdev)) {
 971                 ret = PTR_ERR(bdev);
 972                 goto error;
 973         }
 974
 975         /* make sure our super fits in the device */
 976         if (bytenr + PAGE_CACHE_SIZE >= i_size_read(bdev->bd_inode))
 977                 goto error_bdev_put;
 978
 979         /* make sure our super fits in the page */
 980         if (sizeof(*disk_super) > PAGE_CACHE_SIZE)
 981                 goto error_bdev_put;
 982
 983         /* make sure our super doesn't straddle pages on disk */
 984         index = bytenr >> PAGE_CACHE_SHIFT;
 985         if ((bytenr + sizeof(*disk_super) - 1) >> PAGE_CACHE_SHIFT != index)
 986                 goto error_bdev_put;
 987
 988         /* pull in the page with our super */
 989         page = read_cache_page_gfp(bdev->bd_inode->i_mapping,
 990                                    index, GFP_NOFS);
 991
 992         if (IS_ERR_OR_NULL(page))
 993                 goto error_bdev_put;
 994
 995         p = kmap(page);
 996
 997         /* align our pointer to the offset of the super block */
 998         disk_super = p + (bytenr & ~PAGE_CACHE_MASK);
 999
1000         if (btrfs_super_bytenr(disk_super) != bytenr ||
1001             btrfs_super_magic(disk_super) != BTRFS_MAGIC)
1002                 goto error_unmap;
1003
1004         devid = btrfs_stack_device_id(&disk_super->dev_item);
1005         transid = btrfs_super_generation(disk_super);
1006         total_devices = btrfs_super_num_devices(disk_super);
1007
1008         ret = device_list_add(path, disk_super, devid, fs_devices_ret);
1009         if (ret > 0) {
1010                 if (disk_super->label[0]) {
1011                         if (disk_super->label[BTRFS_LABEL_SIZE - 1])
1012                                 disk_super->label[BTRFS_LABEL_SIZE - 1] = '\0';
1013                         printk(KERN_INFO "BTRFS: device label %s ", disk_super->label);
1014                 } else {
1015                         printk(KERN_INFO "BTRFS: device fsid %pU ", disk_super->fsid);
1016                 }
1017
1018                 printk(KERN_CONT "devid %llu transid %llu %s\n", devid, transid, path);
1019                 ret = 0;
1020         }
1021         if (!ret && fs_devices_ret)
1022                 (*fs_devices_ret)->total_devices = total_devices;
1023
1024 error_unmap:
1025         kunmap(page);
1026         page_cache_release(page);
1027
1028 error_bdev_put:
1029         blkdev_put(bdev, flags);
1030 error:
1031         mutex_unlock(&uuid_mutex);
1032         return ret;
1033 }
1034
1035 /* helper to account the used device space in the range */
1036 int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
1037                                    u64 end, u64 *length)
1038 {
1039         struct btrfs_key key;
1040         struct btrfs_root *root = device->dev_root;
1041         struct btrfs_dev_extent *dev_extent;
1042         struct btrfs_path *path;
1043         u64 extent_end;
1044         int ret;
1045         int slot;
1046         struct extent_buffer *l;
1047
1048         *length = 0;
1049
1050         if (start >= device->total_bytes || device->is_tgtdev_for_dev_replace)
1051                 return 0;
1052
1053         path = btrfs_alloc_path();
1054         if (!path)
1055                 return -ENOMEM;
1056         path->reada = 2;
1057
1058         key.objectid = device->devid;
1059         key.offset = start;
1060         key.type = BTRFS_DEV_EXTENT_KEY;
1061
1062         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1063         if (ret < 0)
1064                 goto out;
1065         if (ret > 0) {
1066                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1067                 if (ret < 0)
1068                         goto out;
1069         }
1070
1071         while (1) {
1072                 l = path->nodes[0];
1073                 slot = path->slots[0];
1074                 if (slot >= btrfs_header_nritems(l)) {
1075                         ret = btrfs_next_leaf(root, path);
1076                         if (ret == 0)
1077                                 continue;
1078                         if (ret < 0)
1079                                 goto out;
1080
1081                         break;
1082                 }
1083                 btrfs_item_key_to_cpu(l, &key, slot);
1084
1085                 if (key.objectid < device->devid)
1086                         goto next;
1087
1088                 if (key.objectid > device->devid)
1089                         break;
1090
1091                 if (key.type != BTRFS_DEV_EXTENT_KEY)
1092                         goto next;
1093
1094                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1095                 extent_end = key.offset + btrfs_dev_extent_length(l,
1096                                                                   dev_extent);
1097                 if (key.offset <= start && extent_end > end) {
1098                         *length = end - start + 1;
1099                         break;
1100                 } else if (key.offset <= start && extent_end > start)
1101                         *length += extent_end - start;
1102                 else if (key.offset > start && extent_end <= end)
1103                         *length += extent_end - key.offset;
1104                 else if (key.offset > start && key.offset <= end) {
1105                         *length += end - key.offset + 1;
1106                         break;
1107                 } else if (key.offset > end)
1108                         break;
1109
1110 next:
1111                 path->slots[0]++;
1112         }
1113         ret = 0;
1114 out:
1115         btrfs_free_path(path);
1116         return ret;
1117 }
1118
1119 static int contains_pending_extent(struct btrfs_transaction *transaction,
1120                                    struct btrfs_device *device,
1121                                    u64 *start, u64 len)
1122 {
1123         struct btrfs_fs_info *fs_info = device->dev_root->fs_info;
1124         struct extent_map *em;
1125         struct list_head *search_list = &fs_info->pinned_chunks;
1126         int ret = 0;
1127         u64 physical_start = *start;
1128
1129         if (transaction)
1130                 search_list = &transaction->pending_chunks;
1131 again:
1132         list_for_each_entry(em, search_list, list) {
1133                 struct map_lookup *map;
1134                 int i;
1135
1136                 map = (struct map_lookup *)em->bdev;
1137                 for (i = 0; i < map->num_stripes; i++) {
1138                         u64 end;
1139
1140                         if (map->stripes[i].dev != device)
1141                                 continue;
1142                         if (map->stripes[i].physical >= physical_start + len ||
1143                             map->stripes[i].physical + em->orig_block_len <=
1144                             physical_start)
1145                                 continue;
1146                         /*
1147                          * Make sure that while processing the pinned list we do
1148                          * not override our *start with a lower value, because
1149                          * we can have pinned chunks that fall within this
1150                          * device hole and that have lower physical addresses
1151                          * than the pending chunks we processed before. If we
1152                          * do not take this special care we can end up getting
1153                          * 2 pending chunks that start at the same physical
1154                          * device offsets because the end offset of a pinned
1155                          * chunk can be equal to the start offset of some
1156                          * pending chunk.
1157                          */
1158                         end = map->stripes[i].physical + em->orig_block_len;
1159                         if (end > *start) {
1160                                 *start = end;
1161                                 ret = 1;
1162                         }
1163                 }
1164         }
1165         if (search_list != &fs_info->pinned_chunks) {
1166                 search_list = &fs_info->pinned_chunks;
1167                 goto again;
1168         }
1169
1170         return ret;
1171 }
1172
1173
1174 /*
1175  * find_free_dev_extent_start - find free space in the specified device
1176  * @device:       the device which we search the free space in
1177  * @num_bytes:    the size of the free space that we need
1178  * @search_start: the position from which to begin the search
1179  * @start:        store the start of the free space.
1180  * @len:          the size of the free space. that we find, or the size
1181  *                of the max free space if we don't find suitable free space
1182  *
1183  * this uses a pretty simple search, the expectation is that it is
1184  * called very infrequently and that a given device has a small number
1185  * of extents
1186  *
1187  * @start is used to store the start of the free space if we find. But if we
1188  * don't find suitable free space, it will be used to store the start position
1189  * of the max free space.
1190  *
1191  * @len is used to store the size of the free space that we find.
1192  * But if we don't find suitable free space, it is used to store the size of
1193  * the max free space.
1194  */
1195 int find_free_dev_extent_start(struct btrfs_transaction *transaction,
1196                                struct btrfs_device *device, u64 num_bytes,
1197                                u64 search_start, u64 *start, u64 *len)
1198 {
1199         struct btrfs_key key;
1200         struct btrfs_root *root = device->dev_root;
1201         struct btrfs_dev_extent *dev_extent;
1202         struct btrfs_path *path;
1203         u64 hole_size;
1204         u64 max_hole_start;
1205         u64 max_hole_size;
1206         u64 extent_end;
1207         u64 search_end = device->total_bytes;
1208         int ret;
1209         int slot;
1210         struct extent_buffer *l;
1211
1212         path = btrfs_alloc_path();
1213         if (!path)
1214                 return -ENOMEM;
1215
1216         max_hole_start = search_start;
1217         max_hole_size = 0;
1218
1219 again:
1220         if (search_start >= search_end || device->is_tgtdev_for_dev_replace) {
1221                 ret = -ENOSPC;
1222                 goto out;
1223         }
1224
1225         path->reada = 2;
1226         path->search_commit_root = 1;
1227         path->skip_locking = 1;
1228
1229         key.objectid = device->devid;
1230         key.offset = search_start;
1231         key.type = BTRFS_DEV_EXTENT_KEY;
1232
1233         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1234         if (ret < 0)
1235                 goto out;
1236         if (ret > 0) {
1237                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1238                 if (ret < 0)
1239                         goto out;
1240         }
1241
1242         while (1) {
1243                 l = path->nodes[0];
1244                 slot = path->slots[0];
1245                 if (slot >= btrfs_header_nritems(l)) {
1246                         ret = btrfs_next_leaf(root, path);
1247                         if (ret == 0)
1248                                 continue;
1249                         if (ret < 0)
1250                                 goto out;
1251
1252                         break;
1253                 }
1254                 btrfs_item_key_to_cpu(l, &key, slot);
1255
1256                 if (key.objectid < device->devid)
1257                         goto next;
1258
1259                 if (key.objectid > device->devid)
1260                         break;
1261
1262                 if (key.type != BTRFS_DEV_EXTENT_KEY)
1263                         goto next;
1264
1265                 if (key.offset > search_start) {
1266                         hole_size = key.offset - search_start;
1267
1268                         /*
1269                          * Have to check before we set max_hole_start, otherwise
1270                          * we could end up sending back this offset anyway.
1271                          */
1272                         if (contains_pending_extent(transaction, device,
1273                                                     &search_start,
1274                                                     hole_size)) {
1275                                 if (key.offset >= search_start) {
1276                                         hole_size = key.offset - search_start;
1277                                 } else {
1278                                         WARN_ON_ONCE(1);
1279                                         hole_size = 0;
1280                                 }
1281                         }
1282
1283                         if (hole_size > max_hole_size) {
1284                                 max_hole_start = search_start;
1285                                 max_hole_size = hole_size;
1286                         }
1287
1288                         /*
1289                          * If this free space is greater than which we need,
1290                          * it must be the max free space that we have found
1291                          * until now, so max_hole_start must point to the start
1292                          * of this free space and the length of this free space
1293                          * is stored in max_hole_size. Thus, we return
1294                          * max_hole_start and max_hole_size and go back to the
1295                          * caller.
1296                          */
1297                         if (hole_size >= num_bytes) {
1298                                 ret = 0;
1299                                 goto out;
1300                         }
1301                 }
1302
1303                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1304                 extent_end = key.offset + btrfs_dev_extent_length(l,
1305                                                                   dev_extent);
1306                 if (extent_end > search_start)
1307                         search_start = extent_end;
1308 next:
1309                 path->slots[0]++;
1310                 cond_resched();
1311         }
1312
1313         /*
1314          * At this point, search_start should be the end of
1315          * allocated dev extents, and when shrinking the device,
1316          * search_end may be smaller than search_start.
1317          */
1318         if (search_end > search_start) {
1319                 hole_size = search_end - search_start;
1320
1321                 if (contains_pending_extent(transaction, device, &search_start,
1322                                             hole_size)) {
1323                         btrfs_release_path(path);
1324                         goto again;
1325                 }
1326
1327                 if (hole_size > max_hole_size) {
1328                         max_hole_start = search_start;
1329                         max_hole_size = hole_size;
1330                 }
1331         }
1332
1333         /* See above. */
1334         if (max_hole_size < num_bytes)
1335                 ret = -ENOSPC;
1336         else
1337                 ret = 0;
1338
1339 out:
1340         btrfs_free_path(path);
1341         *start = max_hole_start;
1342         if (len)
1343                 *len = max_hole_size;
1344         return ret;
1345 }
1346
1347 int find_free_dev_extent(struct btrfs_trans_handle *trans,
1348                          struct btrfs_device *device, u64 num_bytes,
1349                          u64 *start, u64 *len)
1350 {
1351         struct btrfs_root *root = device->dev_root;
1352         u64 search_start;
1353
1354         /* FIXME use last free of some kind */
1355
1356         /*
1357          * we don't want to overwrite the superblock on the drive,
1358          * so we make sure to start at an offset of at least 1MB
1359          */
1360         search_start = max(root->fs_info->alloc_start, 1024ull * 1024);
1361         return find_free_dev_extent_start(trans->transaction, device,
1362                                           num_bytes, search_start, start, len);
1363 }
1364
1365 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1366                           struct btrfs_device *device,
1367                           u64 start, u64 *dev_extent_len)
1368 {
1369         int ret;
1370         struct btrfs_path *path;
1371         struct btrfs_root *root = device->dev_root;
1372         struct btrfs_key key;
1373         struct btrfs_key found_key;
1374         struct extent_buffer *leaf = NULL;
1375         struct btrfs_dev_extent *extent = NULL;
1376
1377         path = btrfs_alloc_path();
1378         if (!path)
1379                 return -ENOMEM;
1380
1381         key.objectid = device->devid;
1382         key.offset = start;
1383         key.type = BTRFS_DEV_EXTENT_KEY;
1384 again:
1385         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1386         if (ret > 0) {
1387                 ret = btrfs_previous_item(root, path, key.objectid,
1388                                           BTRFS_DEV_EXTENT_KEY);
1389                 if (ret)
1390                         goto out;
1391                 leaf = path->nodes[0];
1392                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1393                 extent = btrfs_item_ptr(leaf, path->slots[0],
1394                                         struct btrfs_dev_extent);
1395                 BUG_ON(found_key.offset > start || found_key.offset +
1396                        btrfs_dev_extent_length(leaf, extent) < start);
1397                 key = found_key;
1398                 btrfs_release_path(path);
1399                 goto again;
1400         } else if (ret == 0) {
1401                 leaf = path->nodes[0];
1402                 extent = btrfs_item_ptr(leaf, path->slots[0],
1403                                         struct btrfs_dev_extent);
1404         } else {
1405                 btrfs_error(root->fs_info, ret, "Slot search failed");
1406                 goto out;
1407         }
1408
1409         *dev_extent_len = btrfs_dev_extent_length(leaf, extent);
1410
1411         ret = btrfs_del_item(trans, root, path);
1412         if (ret) {
1413                 btrfs_error(root->fs_info, ret,
1414                             "Failed to remove dev extent item");
1415         } else {
1416                 trans->transaction->have_free_bgs = 1;
1417         }
1418 out:
1419         btrfs_free_path(path);
1420         return ret;
1421 }
1422
1423 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1424                                   struct btrfs_device *device,
1425                                   u64 chunk_tree, u64 chunk_objectid,
1426                                   u64 chunk_offset, u64 start, u64 num_bytes)
1427 {
1428         int ret;
1429         struct btrfs_path *path;
1430         struct btrfs_root *root = device->dev_root;
1431         struct btrfs_dev_extent *extent;
1432         struct extent_buffer *leaf;
1433         struct btrfs_key key;
1434
1435         WARN_ON(!device->in_fs_metadata);
1436         WARN_ON(device->is_tgtdev_for_dev_replace);
1437         path = btrfs_alloc_path();
1438         if (!path)
1439                 return -ENOMEM;
1440
1441         key.objectid = device->devid;
1442         key.offset = start;
1443         key.type = BTRFS_DEV_EXTENT_KEY;
1444         ret = btrfs_insert_empty_item(trans, root, path, &key,
1445                                       sizeof(*extent));
1446         if (ret)
1447                 goto out;
1448
1449         leaf = path->nodes[0];
1450         extent = btrfs_item_ptr(leaf, path->slots[0],
1451                                 struct btrfs_dev_extent);
1452         btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
1453         btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
1454         btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1455
1456         write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
1457                     btrfs_dev_extent_chunk_tree_uuid(extent), BTRFS_UUID_SIZE);
1458
1459         btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1460         btrfs_mark_buffer_dirty(leaf);
1461 out:
1462         btrfs_free_path(path);
1463         return ret;
1464 }
1465
1466 static u64 find_next_chunk(struct btrfs_fs_info *fs_info)
1467 {
1468         struct extent_map_tree *em_tree;
1469         struct extent_map *em;
1470         struct rb_node *n;
1471         u64 ret = 0;
1472
1473         em_tree = &fs_info->mapping_tree.map_tree;
1474         read_lock(&em_tree->lock);
1475         n = rb_last(&em_tree->map);
1476         if (n) {
1477                 em = rb_entry(n, struct extent_map, rb_node);
1478                 ret = em->start + em->len;
1479         }
1480         read_unlock(&em_tree->lock);
1481
1482         return ret;
1483 }
1484
1485 static noinline int find_next_devid(struct btrfs_fs_info *fs_info,
1486                                     u64 *devid_ret)
1487 {
1488         int ret;
1489         struct btrfs_key key;
1490         struct btrfs_key found_key;
1491         struct btrfs_path *path;
1492
1493         path = btrfs_alloc_path();
1494         if (!path)
1495                 return -ENOMEM;
1496
1497         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1498         key.type = BTRFS_DEV_ITEM_KEY;
1499         key.offset = (u64)-1;
1500
1501         ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
1502         if (ret < 0)
1503                 goto error;
1504
1505         BUG_ON(ret == 0); /* Corruption */
1506
1507         ret = btrfs_previous_item(fs_info->chunk_root, path,
1508                                   BTRFS_DEV_ITEMS_OBJECTID,
1509                                   BTRFS_DEV_ITEM_KEY);
1510         if (ret) {
1511                 *devid_ret = 1;
1512         } else {
1513                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1514                                       path->slots[0]);
1515                 *devid_ret = found_key.offset + 1;
1516         }
1517         ret = 0;
1518 error:
1519         btrfs_free_path(path);
1520         return ret;
1521 }
1522
1523 /*
1524  * the device information is stored in the chunk root
1525  * the btrfs_device struct should be fully filled in
1526  */
1527 static int btrfs_add_device(struct btrfs_trans_handle *trans,
1528                             struct btrfs_root *root,
1529                             struct btrfs_device *device)
1530 {
1531         int ret;
1532         struct btrfs_path *path;
1533         struct btrfs_dev_item *dev_item;
1534         struct extent_buffer *leaf;
1535         struct btrfs_key key;
1536         unsigned long ptr;
1537
1538         root = root->fs_info->chunk_root;
1539
1540         path = btrfs_alloc_path();
1541         if (!path)
1542                 return -ENOMEM;
1543
1544         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1545         key.type = BTRFS_DEV_ITEM_KEY;
1546         key.offset = device->devid;
1547
1548         ret = btrfs_insert_empty_item(trans, root, path, &key,
1549                                       sizeof(*dev_item));
1550         if (ret)
1551                 goto out;
1552
1553         leaf = path->nodes[0];
1554         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1555
1556         btrfs_set_device_id(leaf, dev_item, device->devid);
1557         btrfs_set_device_generation(leaf, dev_item, 0);
1558         btrfs_set_device_type(leaf, dev_item, device->type);
1559         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1560         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1561         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1562         btrfs_set_device_total_bytes(leaf, dev_item,
1563                                      btrfs_device_get_disk_total_bytes(device));
1564         btrfs_set_device_bytes_used(leaf, dev_item,
1565                                     btrfs_device_get_bytes_used(device));
1566         btrfs_set_device_group(leaf, dev_item, 0);
1567         btrfs_set_device_seek_speed(leaf, dev_item, 0);
1568         btrfs_set_device_bandwidth(leaf, dev_item, 0);
1569         btrfs_set_device_start_offset(leaf, dev_item, 0);
1570
1571         ptr = btrfs_device_uuid(dev_item);
1572         write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1573         ptr = btrfs_device_fsid(dev_item);
1574         write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1575         btrfs_mark_buffer_dirty(leaf);
1576
1577         ret = 0;
1578 out:
1579         btrfs_free_path(path);
1580         return ret;
1581 }
1582
1583 /*
1584  * Function to update ctime/mtime for a given device path.
1585  * Mainly used for ctime/mtime based probe like libblkid.
1586  */
1587 static void update_dev_time(char *path_name)
1588 {
1589         struct file *filp;
1590
1591         filp = filp_open(path_name, O_RDWR, 0);
1592         if (IS_ERR(filp))
1593                 return;
1594         file_update_time(filp);
1595         filp_close(filp, NULL);
1596         return;
1597 }
1598
1599 static int btrfs_rm_dev_item(struct btrfs_root *root,
1600                              struct btrfs_device *device)
1601 {
1602         int ret;
1603         struct btrfs_path *path;
1604         struct btrfs_key key;
1605         struct btrfs_trans_handle *trans;
1606
1607         root = root->fs_info->chunk_root;
1608
1609         path = btrfs_alloc_path();
1610         if (!path)
1611                 return -ENOMEM;
1612
1613         trans = btrfs_start_transaction(root, 0);
1614         if (IS_ERR(trans)) {
1615                 btrfs_free_path(path);
1616                 return PTR_ERR(trans);
1617         }
1618         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1619         key.type = BTRFS_DEV_ITEM_KEY;
1620         key.offset = device->devid;
1621
1622         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1623         if (ret < 0)
1624                 goto out;
1625
1626         if (ret > 0) {
1627                 ret = -ENOENT;
1628                 goto out;
1629         }
1630
1631         ret = btrfs_del_item(trans, root, path);
1632         if (ret)
1633                 goto out;
1634 out:
1635         btrfs_free_path(path);
1636         btrfs_commit_transaction(trans, root);
1637         return ret;
1638 }
1639
1640 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
1641 {
1642         struct btrfs_device *device;
1643         struct btrfs_device *next_device;
1644         struct block_device *bdev;
1645         struct buffer_head *bh = NULL;
1646         struct btrfs_super_block *disk_super;
1647         struct btrfs_fs_devices *cur_devices;
1648         u64 all_avail;
1649         u64 devid;
1650         u64 num_devices;
1651         u8 *dev_uuid;
1652         unsigned seq;
1653         int ret = 0;
1654         bool clear_super = false;
1655
1656         mutex_lock(&uuid_mutex);
1657
1658         do {
1659                 seq = read_seqbegin(&root->fs_info->profiles_lock);
1660
1661                 all_avail = root->fs_info->avail_data_alloc_bits |
1662                             root->fs_info->avail_system_alloc_bits |
1663                             root->fs_info->avail_metadata_alloc_bits;
1664         } while (read_seqretry(&root->fs_info->profiles_lock, seq));
1665
1666         num_devices = root->fs_info->fs_devices->num_devices;
1667         btrfs_dev_replace_lock(&root->fs_info->dev_replace);
1668         if (btrfs_dev_replace_is_ongoing(&root->fs_info->dev_replace)) {
1669                 WARN_ON(num_devices < 1);
1670                 num_devices--;
1671         }
1672         btrfs_dev_replace_unlock(&root->fs_info->dev_replace);
1673
1674         if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) && num_devices <= 4) {
1675                 ret = BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET;
1676                 goto out;
1677         }
1678
1679         if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) && num_devices <= 2) {
1680                 ret = BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET;
1681                 goto out;
1682         }
1683
1684         if ((all_avail & BTRFS_BLOCK_GROUP_RAID5) &&
1685             root->fs_info->fs_devices->rw_devices <= 2) {
1686                 ret = BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET;
1687                 goto out;
1688         }
1689         if ((all_avail & BTRFS_BLOCK_GROUP_RAID6) &&
1690             root->fs_info->fs_devices->rw_devices <= 3) {
1691                 ret = BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET;
1692                 goto out;
1693         }
1694
1695         if (strcmp(device_path, "missing") == 0) {
1696                 struct list_head *devices;
1697                 struct btrfs_device *tmp;
1698
1699                 device = NULL;
1700                 devices = &root->fs_info->fs_devices->devices;
1701                 /*
1702                  * It is safe to read the devices since the volume_mutex
1703                  * is held.
1704                  */
1705                 list_for_each_entry(tmp, devices, dev_list) {
1706                         if (tmp->in_fs_metadata &&
1707                             !tmp->is_tgtdev_for_dev_replace &&
1708                             !tmp->bdev) {
1709                                 device = tmp;
1710                                 break;
1711                         }
1712                 }
1713                 bdev = NULL;
1714                 bh = NULL;
1715                 disk_super = NULL;
1716                 if (!device) {
1717                         ret = BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
1718                         goto out;
1719                 }
1720         } else {
1721                 ret = btrfs_get_bdev_and_sb(device_path,
1722                                             FMODE_WRITE | FMODE_EXCL,
1723                                             root->fs_info->bdev_holder, 0,
1724                                             &bdev, &bh);
1725                 if (ret)
1726                         goto out;
1727                 disk_super = (struct btrfs_super_block *)bh->b_data;
1728                 devid = btrfs_stack_device_id(&disk_super->dev_item);
1729                 dev_uuid = disk_super->dev_item.uuid;
1730                 device = btrfs_find_device(root->fs_info, devid, dev_uuid,
1731                                            disk_super->fsid);
1732                 if (!device) {
1733                         ret = -ENOENT;
1734                         goto error_brelse;
1735                 }
1736         }
1737
1738         if (device->is_tgtdev_for_dev_replace) {
1739                 ret = BTRFS_ERROR_DEV_TGT_REPLACE;
1740                 goto error_brelse;
1741         }
1742
1743         if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1744                 ret = BTRFS_ERROR_DEV_ONLY_WRITABLE;
1745                 goto error_brelse;
1746         }
1747
1748         if (device->writeable) {
1749                 lock_chunks(root);
1750                 list_del_init(&device->dev_alloc_list);
1751                 device->fs_devices->rw_devices--;
1752                 unlock_chunks(root);
1753                 clear_super = true;
1754         }
1755
1756         mutex_unlock(&uuid_mutex);
1757         ret = btrfs_shrink_device(device, 0);
1758         mutex_lock(&uuid_mutex);
1759         if (ret)
1760                 goto error_undo;
1761
1762         /*
1763          * TODO: the superblock still includes this device in its num_devices
1764          * counter although write_all_supers() is not locked out. This
1765          * could give a filesystem state which requires a degraded mount.
1766          */
1767         ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1768         if (ret)
1769                 goto error_undo;
1770
1771         device->in_fs_metadata = 0;
1772         btrfs_scrub_cancel_dev(root->fs_info, device);
1773
1774         /*
1775          * the device list mutex makes sure that we don't change
1776          * the device list while someone else is writing out all
1777          * the device supers. Whoever is writing all supers, should
1778          * lock the device list mutex before getting the number of
1779          * devices in the super block (super_copy). Conversely,
1780          * whoever updates the number of devices in the super block
1781          * (super_copy) should hold the device list mutex.
1782          */
1783
1784         cur_devices = device->fs_devices;
1785         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1786         list_del_rcu(&device->dev_list);
1787
1788         device->fs_devices->num_devices--;
1789         device->fs_devices->total_devices--;
1790
1791         if (device->missing)
1792                 device->fs_devices->missing_devices--;
1793
1794         next_device = list_entry(root->fs_info->fs_devices->devices.next,
1795                                  struct btrfs_device, dev_list);
1796         if (device->bdev == root->fs_info->sb->s_bdev)
1797                 root->fs_info->sb->s_bdev = next_device->bdev;
1798         if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1799                 root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1800
1801         if (device->bdev) {
1802                 device->fs_devices->open_devices--;
1803                 /* remove sysfs entry */
1804                 btrfs_kobj_rm_device(root->fs_info->fs_devices, device);
1805         }
1806
1807         call_rcu(&device->rcu, free_device);
1808
1809         num_devices = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
1810         btrfs_set_super_num_devices(root->fs_info->super_copy, num_devices);
1811         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1812
1813         if (cur_devices->open_devices == 0) {
1814                 struct btrfs_fs_devices *fs_devices;
1815                 fs_devices = root->fs_info->fs_devices;
1816                 while (fs_devices) {
1817                         if (fs_devices->seed == cur_devices) {
1818                                 fs_devices->seed = cur_devices->seed;
1819                                 break;
1820                         }
1821                         fs_devices = fs_devices->seed;
1822                 }
1823                 cur_devices->seed = NULL;
1824                 __btrfs_close_devices(cur_devices);
1825                 free_fs_devices(cur_devices);
1826         }
1827
1828         root->fs_info->num_tolerated_disk_barrier_failures =
1829                 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
1830
1831         /*
1832          * at this point, the device is zero sized.  We want to
1833          * remove it from the devices list and zero out the old super
1834          */
1835         if (clear_super && disk_super) {
1836                 u64 bytenr;
1837                 int i;
1838
1839                 /* make sure this device isn't detected as part of
1840                  * the FS anymore
1841                  */
1842                 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1843                 set_buffer_dirty(bh);
1844                 sync_dirty_buffer(bh);
1845
1846                 /* clear the mirror copies of super block on the disk
1847                  * being removed, 0th copy is been taken care above and
1848                  * the below would take of the rest
1849                  */
1850                 for (i = 1; i < BTRFS_SUPER_MIRROR_MAX; i++) {
1851                         bytenr = btrfs_sb_offset(i);
1852                         if (bytenr + BTRFS_SUPER_INFO_SIZE >=
1853                                         i_size_read(bdev->bd_inode))
1854                                 break;
1855
1856                         brelse(bh);
1857                         bh = __bread(bdev, bytenr / 4096,
1858                                         BTRFS_SUPER_INFO_SIZE);
1859                         if (!bh)
1860                                 continue;
1861
1862                         disk_super = (struct btrfs_super_block *)bh->b_data;
1863
1864                         if (btrfs_super_bytenr(disk_super) != bytenr ||
1865                                 btrfs_super_magic(disk_super) != BTRFS_MAGIC) {
1866                                 continue;
1867                         }
1868                         memset(&disk_super->magic, 0,
1869                                                 sizeof(disk_super->magic));
1870                         set_buffer_dirty(bh);
1871                         sync_dirty_buffer(bh);
1872                 }
1873         }
1874
1875         ret = 0;
1876
1877         if (bdev) {
1878                 /* Notify udev that device has changed */
1879                 btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
1880
1881                 /* Update ctime/mtime for device path for libblkid */
1882                 update_dev_time(device_path);
1883         }
1884
1885 error_brelse:
1886         brelse(bh);
1887         if (bdev)
1888                 blkdev_put(bdev, FMODE_READ | FMODE_EXCL);
1889 out:
1890         mutex_unlock(&uuid_mutex);
1891         return ret;
1892 error_undo:
1893         if (device->writeable) {
1894                 lock_chunks(root);
1895                 list_add(&device->dev_alloc_list,
1896                          &root->fs_info->fs_devices->alloc_list);
1897                 device->fs_devices->rw_devices++;
1898                 unlock_chunks(root);
1899         }
1900         goto error_brelse;
1901 }
1902
1903 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_fs_info *fs_info,
1904                                         struct btrfs_device *srcdev)
1905 {
1906         struct btrfs_fs_devices *fs_devices;
1907
1908         WARN_ON(!mutex_is_locked(&fs_info->fs_devices->device_list_mutex));
1909
1910         /*
1911          * in case of fs with no seed, srcdev->fs_devices will point
1912          * to fs_devices of fs_info. However when the dev being replaced is
1913          * a seed dev it will point to the seed's local fs_devices. In short
1914          * srcdev will have its correct fs_devices in both the cases.
1915          */
1916         fs_devices = srcdev->fs_devices;
1917
1918         list_del_rcu(&srcdev->dev_list);
1919         list_del_rcu(&srcdev->dev_alloc_list);
1920         fs_devices->num_devices--;
1921         if (srcdev->missing)
1922                 fs_devices->missing_devices--;
1923
1924         if (srcdev->writeable) {
1925                 fs_devices->rw_devices--;
1926                 /* zero out the old super if it is writable */
1927                 btrfs_scratch_superblock(srcdev);
1928         }
1929
1930         if (srcdev->bdev)
1931                 fs_devices->open_devices--;
1932 }
1933
1934 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_fs_info *fs_info,
1935                                       struct btrfs_device *srcdev)
1936 {
1937         struct btrfs_fs_devices *fs_devices = srcdev->fs_devices;
1938
1939         call_rcu(&srcdev->rcu, free_device);
1940
1941         /*
1942          * unless fs_devices is seed fs, num_devices shouldn't go
1943          * zero
1944          */
1945         BUG_ON(!fs_devices->num_devices && !fs_devices->seeding);
1946
1947         /* if this is no devs we rather delete the fs_devices */
1948         if (!fs_devices->num_devices) {
1949                 struct btrfs_fs_devices *tmp_fs_devices;
1950
1951                 tmp_fs_devices = fs_info->fs_devices;
1952                 while (tmp_fs_devices) {
1953                         if (tmp_fs_devices->seed == fs_devices) {
1954                                 tmp_fs_devices->seed = fs_devices->seed;
1955                                 break;
1956                         }
1957                         tmp_fs_devices = tmp_fs_devices->seed;
1958                 }
1959                 fs_devices->seed = NULL;
1960                 __btrfs_close_devices(fs_devices);
1961                 free_fs_devices(fs_devices);
1962         }
1963 }
1964
1965 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
1966                                       struct btrfs_device *tgtdev)
1967 {
1968         struct btrfs_device *next_device;
1969
1970         mutex_lock(&uuid_mutex);
1971         WARN_ON(!tgtdev);
1972         mutex_lock(&fs_info->fs_devices->device_list_mutex);
1973
1974         btrfs_kobj_rm_device(fs_info->fs_devices, tgtdev);
1975
1976         if (tgtdev->bdev) {
1977                 btrfs_scratch_superblock(tgtdev);
1978                 fs_info->fs_devices->open_devices--;
1979         }
1980         fs_info->fs_devices->num_devices--;
1981
1982         next_device = list_entry(fs_info->fs_devices->devices.next,
1983                                  struct btrfs_device, dev_list);
1984         if (tgtdev->bdev == fs_info->sb->s_bdev)
1985                 fs_info->sb->s_bdev = next_device->bdev;
1986         if (tgtdev->bdev == fs_info->fs_devices->latest_bdev)
1987                 fs_info->fs_devices->latest_bdev = next_device->bdev;
1988         list_del_rcu(&tgtdev->dev_list);
1989
1990         call_rcu(&tgtdev->rcu, free_device);
1991
1992         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
1993         mutex_unlock(&uuid_mutex);
1994 }
1995
1996 static int btrfs_find_device_by_path(struct btrfs_root *root, char *device_path,
1997                                      struct btrfs_device **device)
1998 {
1999         int ret = 0;
2000         struct btrfs_super_block *disk_super;
2001         u64 devid;
2002         u8 *dev_uuid;
2003         struct block_device *bdev;
2004         struct buffer_head *bh;
2005
2006         *device = NULL;
2007         ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
2008                                     root->fs_info->bdev_holder, 0, &bdev, &bh);
2009         if (ret)
2010                 return ret;
2011         disk_super = (struct btrfs_super_block *)bh->b_data;
2012         devid = btrfs_stack_device_id(&disk_super->dev_item);
2013         dev_uuid = disk_super->dev_item.uuid;
2014         *device = btrfs_find_device(root->fs_info, devid, dev_uuid,
2015                                     disk_super->fsid);
2016         brelse(bh);
2017         if (!*device)
2018                 ret = -ENOENT;
2019         blkdev_put(bdev, FMODE_READ);
2020         return ret;
2021 }
2022
2023 int btrfs_find_device_missing_or_by_path(struct btrfs_root *root,
2024                                          char *device_path,
2025                                          struct btrfs_device **device)
2026 {
2027         *device = NULL;
2028         if (strcmp(device_path, "missing") == 0) {
2029                 struct list_head *devices;
2030                 struct btrfs_device *tmp;
2031
2032                 devices = &root->fs_info->fs_devices->devices;
2033                 /*
2034                  * It is safe to read the devices since the volume_mutex
2035                  * is held by the caller.
2036                  */
2037                 list_for_each_entry(tmp, devices, dev_list) {
2038                         if (tmp->in_fs_metadata && !tmp->bdev) {
2039                                 *device = tmp;
2040                                 break;
2041                         }
2042                 }
2043
2044                 if (!*device) {
2045                         btrfs_err(root->fs_info, "no missing device found");
2046                         return -ENOENT;
2047                 }
2048
2049                 return 0;
2050         } else {
2051                 return btrfs_find_device_by_path(root, device_path, device);
2052         }
2053 }
2054
2055 /*
2056  * does all the dirty work required for changing file system's UUID.
2057  */
2058 static int btrfs_prepare_sprout(struct btrfs_root *root)
2059 {
2060         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2061         struct btrfs_fs_devices *old_devices;
2062         struct btrfs_fs_devices *seed_devices;
2063         struct btrfs_super_block *disk_super = root->fs_info->super_copy;
2064         struct btrfs_device *device;
2065         u64 super_flags;
2066
2067         BUG_ON(!mutex_is_locked(&uuid_mutex));
2068         if (!fs_devices->seeding)
2069                 return -EINVAL;
2070
2071         seed_devices = __alloc_fs_devices();
2072         if (IS_ERR(seed_devices))
2073                 return PTR_ERR(seed_devices);
2074
2075         old_devices = clone_fs_devices(fs_devices);
2076         if (IS_ERR(old_devices)) {
2077                 kfree(seed_devices);
2078                 return PTR_ERR(old_devices);
2079         }
2080
2081         list_add(&old_devices->list, &fs_uuids);
2082
2083         memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
2084         seed_devices->opened = 1;
2085         INIT_LIST_HEAD(&seed_devices->devices);
2086         INIT_LIST_HEAD(&seed_devices->alloc_list);
2087         mutex_init(&seed_devices->device_list_mutex);
2088
2089         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2090         list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
2091                               synchronize_rcu);
2092         list_for_each_entry(device, &seed_devices->devices, dev_list)
2093                 device->fs_devices = seed_devices;
2094
2095         lock_chunks(root);
2096         list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
2097         unlock_chunks(root);
2098
2099         fs_devices->seeding = 0;
2100         fs_devices->num_devices = 0;
2101         fs_devices->open_devices = 0;
2102         fs_devices->missing_devices = 0;
2103         fs_devices->rotating = 0;
2104         fs_devices->seed = seed_devices;
2105
2106         generate_random_uuid(fs_devices->fsid);
2107         memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2108         memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2109         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2110
2111         super_flags = btrfs_super_flags(disk_super) &
2112                       ~BTRFS_SUPER_FLAG_SEEDING;
2113         btrfs_set_super_flags(disk_super, super_flags);
2114
2115         return 0;
2116 }
2117
2118 /*
2119  * strore the expected generation for seed devices in device items.
2120  */
2121 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
2122                                struct btrfs_root *root)
2123 {
2124         struct btrfs_path *path;
2125         struct extent_buffer *leaf;
2126         struct btrfs_dev_item *dev_item;
2127         struct btrfs_device *device;
2128         struct btrfs_key key;
2129         u8 fs_uuid[BTRFS_UUID_SIZE];
2130         u8 dev_uuid[BTRFS_UUID_SIZE];
2131         u64 devid;
2132         int ret;
2133
2134         path = btrfs_alloc_path();
2135         if (!path)
2136                 return -ENOMEM;
2137
2138         root = root->fs_info->chunk_root;
2139         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2140         key.offset = 0;
2141         key.type = BTRFS_DEV_ITEM_KEY;
2142
2143         while (1) {
2144                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2145                 if (ret < 0)
2146                         goto error;
2147
2148                 leaf = path->nodes[0];
2149 next_slot:
2150                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2151                         ret = btrfs_next_leaf(root, path);
2152                         if (ret > 0)
2153                                 break;
2154                         if (ret < 0)
2155                                 goto error;
2156                         leaf = path->nodes[0];
2157                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2158                         btrfs_release_path(path);
2159                         continue;
2160                 }
2161
2162                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2163                 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
2164                     key.type != BTRFS_DEV_ITEM_KEY)
2165                         break;
2166
2167                 dev_item = btrfs_item_ptr(leaf, path->slots[0],
2168                                           struct btrfs_dev_item);
2169                 devid = btrfs_device_id(leaf, dev_item);
2170                 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
2171                                    BTRFS_UUID_SIZE);
2172                 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
2173                                    BTRFS_UUID_SIZE);
2174                 device = btrfs_find_device(root->fs_info, devid, dev_uuid,
2175                                            fs_uuid);
2176                 BUG_ON(!device); /* Logic error */
2177
2178                 if (device->fs_devices->seeding) {
2179                         btrfs_set_device_generation(leaf, dev_item,
2180                                                     device->generation);
2181                         btrfs_mark_buffer_dirty(leaf);
2182                 }
2183
2184                 path->slots[0]++;
2185                 goto next_slot;
2186         }
2187         ret = 0;
2188 error:
2189         btrfs_free_path(path);
2190         return ret;
2191 }
2192
2193 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
2194 {
2195         struct request_queue *q;
2196         struct btrfs_trans_handle *trans;
2197         struct btrfs_device *device;
2198         struct block_device *bdev;
2199         struct list_head *devices;
2200         struct super_block *sb = root->fs_info->sb;
2201         struct rcu_string *name;
2202         u64 tmp;
2203         int seeding_dev = 0;
2204         int ret = 0;
2205
2206         if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
2207                 return -EROFS;
2208
2209         bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2210                                   root->fs_info->bdev_holder);
2211         if (IS_ERR(bdev))
2212                 return PTR_ERR(bdev);
2213
2214         if (root->fs_info->fs_devices->seeding) {
2215                 seeding_dev = 1;
2216                 down_write(&sb->s_umount);
2217                 mutex_lock(&uuid_mutex);
2218         }
2219
2220         filemap_write_and_wait(bdev->bd_inode->i_mapping);
2221
2222         devices = &root->fs_info->fs_devices->devices;
2223
2224         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2225         list_for_each_entry(device, devices, dev_list) {
2226                 if (device->bdev == bdev) {
2227                         ret = -EEXIST;
2228                         mutex_unlock(
2229                                 &root->fs_info->fs_devices->device_list_mutex);
2230                         goto error;
2231                 }
2232         }
2233         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2234
2235         device = btrfs_alloc_device(root->fs_info, NULL, NULL);
2236         if (IS_ERR(device)) {
2237                 /* we can safely leave the fs_devices entry around */
2238                 ret = PTR_ERR(device);
2239                 goto error;
2240         }
2241
2242         name = rcu_string_strdup(device_path, GFP_NOFS);
2243         if (!name) {
2244                 kfree(device);
2245                 ret = -ENOMEM;
2246                 goto error;
2247         }
2248         rcu_assign_pointer(device->name, name);
2249
2250         trans = btrfs_start_transaction(root, 0);
2251         if (IS_ERR(trans)) {
2252                 rcu_string_free(device->name);
2253                 kfree(device);
2254                 ret = PTR_ERR(trans);
2255                 goto error;
2256         }
2257
2258         q = bdev_get_queue(bdev);
2259         if (blk_queue_discard(q))
2260                 device->can_discard = 1;
2261         device->writeable = 1;
2262         device->generation = trans->transid;
2263         device->io_width = root->sectorsize;
2264         device->io_align = root->sectorsize;
2265         device->sector_size = root->sectorsize;
2266         device->total_bytes = i_size_read(bdev->bd_inode);
2267         device->disk_total_bytes = device->total_bytes;
2268         device->commit_total_bytes = device->total_bytes;
2269         device->dev_root = root->fs_info->dev_root;
2270         device->bdev = bdev;
2271         device->in_fs_metadata = 1;
2272         device->is_tgtdev_for_dev_replace = 0;
2273         device->mode = FMODE_EXCL;
2274         device->dev_stats_valid = 1;
2275         set_blocksize(device->bdev, 4096);
2276
2277         if (seeding_dev) {
2278                 sb->s_flags &= ~MS_RDONLY;
2279                 ret = btrfs_prepare_sprout(root);
2280                 BUG_ON(ret); /* -ENOMEM */
2281         }
2282
2283         device->fs_devices = root->fs_info->fs_devices;
2284
2285         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2286         lock_chunks(root);
2287         list_add_rcu(&device->dev_list, &root->fs_info->fs_devices->devices);
2288         list_add(&device->dev_alloc_list,
2289                  &root->fs_info->fs_devices->alloc_list);
2290         root->fs_info->fs_devices->num_devices++;
2291         root->fs_info->fs_devices->open_devices++;
2292         root->fs_info->fs_devices->rw_devices++;
2293         root->fs_info->fs_devices->total_devices++;
2294         root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
2295
2296         spin_lock(&root->fs_info->free_chunk_lock);
2297         root->fs_info->free_chunk_space += device->total_bytes;
2298         spin_unlock(&root->fs_info->free_chunk_lock);
2299
2300         if (!blk_queue_nonrot(bdev_get_queue(bdev)))
2301                 root->fs_info->fs_devices->rotating = 1;
2302
2303         tmp = btrfs_super_total_bytes(root->fs_info->super_copy);
2304         btrfs_set_super_total_bytes(root->fs_info->super_copy,
2305                                     tmp + device->total_bytes);
2306
2307         tmp = btrfs_super_num_devices(root->fs_info->super_copy);
2308         btrfs_set_super_num_devices(root->fs_info->super_copy,
2309                                     tmp + 1);
2310
2311         /* add sysfs device entry */
2312         btrfs_kobj_add_device(root->fs_info->fs_devices, device);
2313
2314         /*
2315          * we've got more storage, clear any full flags on the space
2316          * infos
2317          */
2318         btrfs_clear_space_info_full(root->fs_info);
2319
2320         unlock_chunks(root);
2321         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2322
2323         if (seeding_dev) {
2324                 lock_chunks(root);
2325                 ret = init_first_rw_device(trans, root, device);
2326                 unlock_chunks(root);
2327                 if (ret) {
2328                         btrfs_abort_transaction(trans, root, ret);
2329                         goto error_trans;
2330                 }
2331         }
2332
2333         ret = btrfs_add_device(trans, root, device);
2334         if (ret) {
2335                 btrfs_abort_transaction(trans, root, ret);
2336                 goto error_trans;
2337         }
2338
2339         if (seeding_dev) {
2340                 char fsid_buf[BTRFS_UUID_UNPARSED_SIZE];
2341
2342                 ret = btrfs_finish_sprout(trans, root);
2343                 if (ret) {
2344                         btrfs_abort_transaction(trans, root, ret);
2345                         goto error_trans;
2346                 }
2347
2348                 /* Sprouting would change fsid of the mounted root,
2349                  * so rename the fsid on the sysfs
2350                  */
2351                 snprintf(fsid_buf, BTRFS_UUID_UNPARSED_SIZE, "%pU",
2352                                                 root->fs_info->fsid);
2353                 if (kobject_rename(&root->fs_info->fs_devices->super_kobj,
2354                                                                 fsid_buf))
2355                         pr_warn("BTRFS: sysfs: failed to create fsid for sprout\n");
2356         }
2357
2358         root->fs_info->num_tolerated_disk_barrier_failures =
2359                 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
2360         ret = btrfs_commit_transaction(trans, root);
2361
2362         if (seeding_dev) {
2363                 mutex_unlock(&uuid_mutex);
2364                 up_write(&sb->s_umount);
2365
2366                 if (ret) /* transaction commit */
2367                         return ret;
2368
2369                 ret = btrfs_relocate_sys_chunks(root);
2370                 if (ret < 0)
2371                         btrfs_error(root->fs_info, ret,
2372                                     "Failed to relocate sys chunks after "
2373                                     "device initialization. This can be fixed "
2374                                     "using the \"btrfs balance\" command.");
2375                 trans = btrfs_attach_transaction(root);
2376                 if (IS_ERR(trans)) {
2377                         if (PTR_ERR(trans) == -ENOENT)
2378                                 return 0;
2379                         return PTR_ERR(trans);
2380                 }
2381                 ret = btrfs_commit_transaction(trans, root);
2382         }
2383
2384         /* Update ctime/mtime for libblkid */
2385         update_dev_time(device_path);
2386         return ret;
2387
2388 error_trans:
2389         btrfs_end_transaction(trans, root);
2390         rcu_string_free(device->name);
2391         btrfs_kobj_rm_device(root->fs_info->fs_devices, device);
2392         kfree(device);
2393 error:
2394         blkdev_put(bdev, FMODE_EXCL);
2395         if (seeding_dev) {
2396                 mutex_unlock(&uuid_mutex);
2397                 up_write(&sb->s_umount);
2398         }
2399         return ret;
2400 }
2401
2402 int btrfs_init_dev_replace_tgtdev(struct btrfs_root *root, char *device_path,
2403                                   struct btrfs_device *srcdev,
2404                                   struct btrfs_device **device_out)
2405 {
2406         struct request_queue *q;
2407         struct btrfs_device *device;
2408         struct block_device *bdev;
2409         struct btrfs_fs_info *fs_info = root->fs_info;
2410         struct list_head *devices;
2411         struct rcu_string *name;
2412         u64 devid = BTRFS_DEV_REPLACE_DEVID;
2413         int ret = 0;
2414
2415         *device_out = NULL;
2416         if (fs_info->fs_devices->seeding) {
2417                 btrfs_err(fs_info, "the filesystem is a seed filesystem!");
2418                 return -EINVAL;
2419         }
2420
2421         bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2422                                   fs_info->bdev_holder);
2423         if (IS_ERR(bdev)) {
2424                 btrfs_err(fs_info, "target device %s is invalid!", device_path);
2425                 return PTR_ERR(bdev);
2426         }
2427
2428         filemap_write_and_wait(bdev->bd_inode->i_mapping);
2429
2430         devices = &fs_info->fs_devices->devices;
2431         list_for_each_entry(device, devices, dev_list) {
2432                 if (device->bdev == bdev) {
2433                         btrfs_err(fs_info, "target device is in the filesystem!");
2434                         ret = -EEXIST;
2435                         goto error;
2436                 }
2437         }
2438
2439
2440         if (i_size_read(bdev->bd_inode) <
2441             btrfs_device_get_total_bytes(srcdev)) {
2442                 btrfs_err(fs_info, "target device is smaller than source device!");
2443                 ret = -EINVAL;
2444                 goto error;
2445         }
2446
2447
2448         device = btrfs_alloc_device(NULL, &devid, NULL);
2449         if (IS_ERR(device)) {
2450                 ret = PTR_ERR(device);
2451                 goto error;
2452         }
2453
2454         name = rcu_string_strdup(device_path, GFP_NOFS);
2455         if (!name) {
2456                 kfree(device);
2457                 ret = -ENOMEM;
2458                 goto error;
2459         }
2460         rcu_assign_pointer(device->name, name);
2461
2462         q = bdev_get_queue(bdev);
2463         if (blk_queue_discard(q))
2464                 device->can_discard = 1;
2465         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2466         device->writeable = 1;
2467         device->generation = 0;
2468         device->io_width = root->sectorsize;
2469         device->io_align = root->sectorsize;
2470         device->sector_size = root->sectorsize;
2471         device->total_bytes = btrfs_device_get_total_bytes(srcdev);
2472         device->disk_total_bytes = btrfs_device_get_disk_total_bytes(srcdev);
2473         device->bytes_used = btrfs_device_get_bytes_used(srcdev);
2474         ASSERT(list_empty(&srcdev->resized_list));
2475         device->commit_total_bytes = srcdev->commit_total_bytes;
2476         device->commit_bytes_used = device->bytes_used;
2477         device->dev_root = fs_info->dev_root;
2478         device->bdev = bdev;
2479         device->in_fs_metadata = 1;
2480         device->is_tgtdev_for_dev_replace = 1;
2481         device->mode = FMODE_EXCL;
2482         device->dev_stats_valid = 1;
2483         set_blocksize(device->bdev, 4096);
2484         device->fs_devices = fs_info->fs_devices;
2485         list_add(&device->dev_list, &fs_info->fs_devices->devices);
2486         fs_info->fs_devices->num_devices++;
2487         fs_info->fs_devices->open_devices++;
2488         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2489
2490         *device_out = device;
2491         return ret;
2492
2493 error:
2494         blkdev_put(bdev, FMODE_EXCL);
2495         return ret;
2496 }
2497
2498 void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info *fs_info,
2499                                               struct btrfs_device *tgtdev)
2500 {
2501         WARN_ON(fs_info->fs_devices->rw_devices == 0);
2502         tgtdev->io_width = fs_info->dev_root->sectorsize;
2503         tgtdev->io_align = fs_info->dev_root->sectorsize;
2504         tgtdev->sector_size = fs_info->dev_root->sectorsize;
2505         tgtdev->dev_root = fs_info->dev_root;
2506         tgtdev->in_fs_metadata = 1;
2507 }
2508
2509 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2510                                         struct btrfs_device *device)
2511 {
2512         int ret;
2513         struct btrfs_path *path;
2514         struct btrfs_root *root;
2515         struct btrfs_dev_item *dev_item;
2516         struct extent_buffer *leaf;
2517         struct btrfs_key key;
2518
2519         root = device->dev_root->fs_info->chunk_root;
2520
2521         path = btrfs_alloc_path();
2522         if (!path)
2523                 return -ENOMEM;
2524
2525         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2526         key.type = BTRFS_DEV_ITEM_KEY;
2527         key.offset = device->devid;
2528
2529         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2530         if (ret < 0)
2531                 goto out;
2532
2533         if (ret > 0) {
2534                 ret = -ENOENT;
2535                 goto out;
2536         }
2537
2538         leaf = path->nodes[0];
2539         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2540
2541         btrfs_set_device_id(leaf, dev_item, device->devid);
2542         btrfs_set_device_type(leaf, dev_item, device->type);
2543         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2544         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2545         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2546         btrfs_set_device_total_bytes(leaf, dev_item,
2547                                      btrfs_device_get_disk_total_bytes(device));
2548         btrfs_set_device_bytes_used(leaf, dev_item,
2549                                     btrfs_device_get_bytes_used(device));
2550         btrfs_mark_buffer_dirty(leaf);
2551
2552 out:
2553         btrfs_free_path(path);
2554         return ret;
2555 }
2556
2557 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2558                       struct btrfs_device *device, u64 new_size)
2559 {
2560         struct btrfs_super_block *super_copy =
2561                 device->dev_root->fs_info->super_copy;
2562         struct btrfs_fs_devices *fs_devices;
2563         u64 old_total;
2564         u64 diff;
2565
2566         if (!device->writeable)
2567                 return -EACCES;
2568
2569         lock_chunks(device->dev_root);
2570         old_total = btrfs_super_total_bytes(super_copy);
2571         diff = new_size - device->total_bytes;
2572
2573         if (new_size <= device->total_bytes ||
2574             device->is_tgtdev_for_dev_replace) {
2575                 unlock_chunks(device->dev_root);
2576                 return -EINVAL;
2577         }
2578
2579         fs_devices = device->dev_root->fs_info->fs_devices;
2580
2581         btrfs_set_super_total_bytes(super_copy, old_total + diff);
2582         device->fs_devices->total_rw_bytes += diff;
2583
2584         btrfs_device_set_total_bytes(device, new_size);
2585         btrfs_device_set_disk_total_bytes(device, new_size);
2586         btrfs_clear_space_info_full(device->dev_root->fs_info);
2587         if (list_empty(&device->resized_list))
2588                 list_add_tail(&device->resized_list,
2589                               &fs_devices->resized_devices);
2590         unlock_chunks(device->dev_root);
2591
2592         return btrfs_update_device(trans, device);
2593 }
2594
2595 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
2596                             struct btrfs_root *root, u64 chunk_objectid,
2597                             u64 chunk_offset)
2598 {
2599         int ret;
2600         struct btrfs_path *path;
2601         struct btrfs_key key;
2602
2603         root = root->fs_info->chunk_root;
2604         path = btrfs_alloc_path();
2605         if (!path)
2606                 return -ENOMEM;
2607
2608         key.objectid = chunk_objectid;
2609         key.offset = chunk_offset;
2610         key.type = BTRFS_CHUNK_ITEM_KEY;
2611
2612         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2613         if (ret < 0)
2614                 goto out;
2615         else if (ret > 0) { /* Logic error or corruption */
2616                 btrfs_error(root->fs_info, -ENOENT,
2617                             "Failed lookup while freeing chunk.");
2618                 ret = -ENOENT;
2619                 goto out;
2620         }
2621
2622         ret = btrfs_del_item(trans, root, path);
2623         if (ret < 0)
2624                 btrfs_error(root->fs_info, ret,
2625                             "Failed to delete chunk item.");
2626 out:
2627         btrfs_free_path(path);
2628         return ret;
2629 }
2630
2631 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
2632                         chunk_offset)
2633 {
2634         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
2635         struct btrfs_disk_key *disk_key;
2636         struct btrfs_chunk *chunk;
2637         u8 *ptr;
2638         int ret = 0;
2639         u32 num_stripes;
2640         u32 array_size;
2641         u32 len = 0;
2642         u32 cur;
2643         struct btrfs_key key;
2644
2645         lock_chunks(root);
2646         array_size = btrfs_super_sys_array_size(super_copy);
2647
2648         ptr = super_copy->sys_chunk_array;
2649         cur = 0;
2650
2651         while (cur < array_size) {
2652                 disk_key = (struct btrfs_disk_key *)ptr;
2653                 btrfs_disk_key_to_cpu(&key, disk_key);
2654
2655                 len = sizeof(*disk_key);
2656
2657                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2658                         chunk = (struct btrfs_chunk *)(ptr + len);
2659                         num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2660                         len += btrfs_chunk_item_size(num_stripes);
2661                 } else {
2662                         ret = -EIO;
2663                         break;
2664                 }
2665                 if (key.objectid == chunk_objectid &&
2666                     key.offset == chunk_offset) {
2667                         memmove(ptr, ptr + len, array_size - (cur + len));
2668                         array_size -= len;
2669                         btrfs_set_super_sys_array_size(super_copy, array_size);
2670                 } else {
2671                         ptr += len;
2672                         cur += len;
2673                 }
2674         }
2675         unlock_chunks(root);
2676         return ret;
2677 }
2678
2679 int btrfs_remove_chunk(struct btrfs_trans_handle *trans,
2680                        struct btrfs_root *root, u64 chunk_offset)
2681 {
2682         struct extent_map_tree *em_tree;
2683         struct extent_map *em;
2684         struct btrfs_root *extent_root = root->fs_info->extent_root;
2685         struct map_lookup *map;
2686         u64 dev_extent_len = 0;
2687         u64 chunk_objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2688         int i, ret = 0;
2689
2690         /* Just in case */
2691         root = root->fs_info->chunk_root;
2692         em_tree = &root->fs_info->mapping_tree.map_tree;
2693
2694         read_lock(&em_tree->lock);
2695         em = lookup_extent_mapping(em_tree, chunk_offset, 1);
2696         read_unlock(&em_tree->lock);
2697
2698         if (!em || em->start > chunk_offset ||
2699             em->start + em->len < chunk_offset) {
2700                 /*
2701                  * This is a logic error, but we don't want to just rely on the
2702                  * user having built with ASSERT enabled, so if ASSERT doens't
2703                  * do anything we still error out.
2704                  */
2705                 ASSERT(0);
2706                 if (em)
2707                         free_extent_map(em);
2708                 return -EINVAL;
2709         }
2710         map = (struct map_lookup *)em->bdev;
2711         lock_chunks(root->fs_info->chunk_root);
2712         check_system_chunk(trans, extent_root, map->type);
2713         unlock_chunks(root->fs_info->chunk_root);
2714
2715         for (i = 0; i < map->num_stripes; i++) {
2716                 struct btrfs_device *device = map->stripes[i].dev;
2717                 ret = btrfs_free_dev_extent(trans, device,
2718                                             map->stripes[i].physical,
2719                                             &dev_extent_len);
2720                 if (ret) {
2721                         btrfs_abort_transaction(trans, root, ret);
2722                         goto out;
2723                 }
2724
2725                 if (device->bytes_used > 0) {
2726                         lock_chunks(root);
2727                         btrfs_device_set_bytes_used(device,
2728                                         device->bytes_used - dev_extent_len);
2729                         spin_lock(&root->fs_info->free_chunk_lock);
2730                         root->fs_info->free_chunk_space += dev_extent_len;
2731                         spin_unlock(&root->fs_info->free_chunk_lock);
2732                         btrfs_clear_space_info_full(root->fs_info);
2733                         unlock_chunks(root);
2734                 }
2735
2736                 if (map->stripes[i].dev) {
2737                         ret = btrfs_update_device(trans, map->stripes[i].dev);
2738                         if (ret) {
2739                                 btrfs_abort_transaction(trans, root, ret);
2740                                 goto out;
2741                         }
2742                 }
2743         }
2744         ret = btrfs_free_chunk(trans, root, chunk_objectid, chunk_offset);
2745         if (ret) {
2746                 btrfs_abort_transaction(trans, root, ret);
2747                 goto out;
2748         }
2749
2750         trace_btrfs_chunk_free(root, map, chunk_offset, em->len);
2751
2752         if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2753                 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
2754                 if (ret) {
2755                         btrfs_abort_transaction(trans, root, ret);
2756                         goto out;
2757                 }
2758         }
2759
2760         ret = btrfs_remove_block_group(trans, extent_root, chunk_offset, em);
2761         if (ret) {
2762                 btrfs_abort_transaction(trans, extent_root, ret);
2763                 goto out;
2764         }
2765
2766 out:
2767         /* once for us */
2768         free_extent_map(em);
2769         return ret;
2770 }
2771
2772 static int btrfs_relocate_chunk(struct btrfs_root *root, u64 chunk_offset)
2773 {
2774         struct btrfs_root *extent_root;
2775         struct btrfs_trans_handle *trans;
2776         int ret;
2777
2778         root = root->fs_info->chunk_root;
2779         extent_root = root->fs_info->extent_root;
2780
2781         /*
2782          * Prevent races with automatic removal of unused block groups.
2783          * After we relocate and before we remove the chunk with offset
2784          * chunk_offset, automatic removal of the block group can kick in,
2785          * resulting in a failure when calling btrfs_remove_chunk() below.
2786          *
2787          * Make sure to acquire this mutex before doing a tree search (dev
2788          * or chunk trees) to find chunks. Otherwise the cleaner kthread might
2789          * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after
2790          * we release the path used to search the chunk/dev tree and before
2791          * the current task acquires this mutex and calls us.
2792          */
2793         ASSERT(mutex_is_locked(&root->fs_info->delete_unused_bgs_mutex));
2794
2795         ret = btrfs_can_relocate(extent_root, chunk_offset);
2796         if (ret)
2797                 return -ENOSPC;
2798
2799         /* step one, relocate all the extents inside this chunk */
2800         btrfs_scrub_pause(root);
2801         ret = btrfs_relocate_block_group(extent_root, chunk_offset);
2802         btrfs_scrub_continue(root);
2803         if (ret)
2804                 return ret;
2805
2806         trans = btrfs_start_transaction(root, 0);
2807         if (IS_ERR(trans)) {
2808                 ret = PTR_ERR(trans);
2809                 btrfs_std_error(root->fs_info, ret);
2810                 return ret;
2811         }
2812
2813         /*
2814          * step two, delete the device extents and the
2815          * chunk tree entries
2816          */
2817         ret = btrfs_remove_chunk(trans, root, chunk_offset);
2818         btrfs_end_transaction(trans, root);
2819         return ret;
2820 }
2821
2822 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
2823 {
2824         struct btrfs_root *chunk_root = root->fs_info->chunk_root;
2825         struct btrfs_path *path;
2826         struct extent_buffer *leaf;
2827         struct btrfs_chunk *chunk;
2828         struct btrfs_key key;
2829         struct btrfs_key found_key;
2830         u64 chunk_type;
2831         bool retried = false;
2832         int failed = 0;
2833         int ret;
2834
2835         path = btrfs_alloc_path();
2836         if (!path)
2837                 return -ENOMEM;
2838
2839 again:
2840         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2841         key.offset = (u64)-1;
2842         key.type = BTRFS_CHUNK_ITEM_KEY;
2843
2844         while (1) {
2845                 mutex_lock(&root->fs_info->delete_unused_bgs_mutex);
2846                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2847                 if (ret < 0) {
2848                         mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
2849                         goto error;
2850                 }
2851                 BUG_ON(ret == 0); /* Corruption */
2852
2853                 ret = btrfs_previous_item(chunk_root, path, key.objectid,
2854                                           key.type);
2855                 if (ret)
2856                         mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
2857                 if (ret < 0)
2858                         goto error;
2859                 if (ret > 0)
2860                         break;
2861
2862                 leaf = path->nodes[0];
2863                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2864
2865                 chunk = btrfs_item_ptr(leaf, path->slots[0],
2866                                        struct btrfs_chunk);
2867                 chunk_type = btrfs_chunk_type(leaf, chunk);
2868                 btrfs_release_path(path);
2869
2870                 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
2871                         ret = btrfs_relocate_chunk(chunk_root,
2872                                                    found_key.offset);
2873                         if (ret == -ENOSPC)
2874                                 failed++;
2875                         else
2876                                 BUG_ON(ret);
2877                 }
2878                 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
2879
2880                 if (found_key.offset == 0)
2881                         break;
2882                 key.offset = found_key.offset - 1;
2883         }
2884         ret = 0;
2885         if (failed && !retried) {
2886                 failed = 0;
2887                 retried = true;
2888                 goto again;
2889         } else if (WARN_ON(failed && retried)) {
2890                 ret = -ENOSPC;
2891         }
2892 error:
2893         btrfs_free_path(path);
2894         return ret;
2895 }
2896
2897 static int insert_balance_item(struct btrfs_root *root,
2898                                struct btrfs_balance_control *bctl)
2899 {
2900         struct btrfs_trans_handle *trans;
2901         struct btrfs_balance_item *item;
2902         struct btrfs_disk_balance_args disk_bargs;
2903         struct btrfs_path *path;
2904         struct extent_buffer *leaf;
2905         struct btrfs_key key;
2906         int ret, err;
2907
2908         path = btrfs_alloc_path();
2909         if (!path)
2910                 return -ENOMEM;
2911
2912         trans = btrfs_start_transaction(root, 0);
2913         if (IS_ERR(trans)) {
2914                 btrfs_free_path(path);
2915                 return PTR_ERR(trans);
2916         }
2917
2918         key.objectid = BTRFS_BALANCE_OBJECTID;
2919         key.type = BTRFS_BALANCE_ITEM_KEY;
2920         key.offset = 0;
2921
2922         ret = btrfs_insert_empty_item(trans, root, path, &key,
2923                                       sizeof(*item));
2924         if (ret)
2925                 goto out;
2926
2927         leaf = path->nodes[0];
2928         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
2929
2930         memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
2931
2932         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
2933         btrfs_set_balance_data(leaf, item, &disk_bargs);
2934         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
2935         btrfs_set_balance_meta(leaf, item, &disk_bargs);
2936         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
2937         btrfs_set_balance_sys(leaf, item, &disk_bargs);
2938
2939         btrfs_set_balance_flags(leaf, item, bctl->flags);
2940
2941         btrfs_mark_buffer_dirty(leaf);
2942 out:
2943         btrfs_free_path(path);
2944         err = btrfs_commit_transaction(trans, root);
2945         if (err && !ret)
2946                 ret = err;
2947         return ret;
2948 }
2949
2950 static int del_balance_item(struct btrfs_root *root)
2951 {
2952         struct btrfs_trans_handle *trans;
2953         struct btrfs_path *path;
2954         struct btrfs_key key;
2955         int ret, err;
2956
2957         path = btrfs_alloc_path();
2958         if (!path)
2959                 return -ENOMEM;
2960
2961         trans = btrfs_start_transaction(root, 0);
2962         if (IS_ERR(trans)) {
2963                 btrfs_free_path(path);
2964                 return PTR_ERR(trans);
2965         }
2966
2967         key.objectid = BTRFS_BALANCE_OBJECTID;
2968         key.type = BTRFS_BALANCE_ITEM_KEY;
2969         key.offset = 0;
2970
2971         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2972         if (ret < 0)
2973                 goto out;
2974         if (ret > 0) {
2975                 ret = -ENOENT;
2976                 goto out;
2977         }
2978
2979         ret = btrfs_del_item(trans, root, path);
2980 out:
2981         btrfs_free_path(path);
2982         err = btrfs_commit_transaction(trans, root);
2983         if (err && !ret)
2984                 ret = err;
2985         return ret;
2986 }
2987
2988 /*
2989  * This is a heuristic used to reduce the number of chunks balanced on
2990  * resume after balance was interrupted.
2991  */
2992 static void update_balance_args(struct btrfs_balance_control *bctl)
2993 {
2994         /*
2995          * Turn on soft mode for chunk types that were being converted.
2996          */
2997         if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
2998                 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
2999         if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
3000                 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
3001         if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
3002                 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
3003
3004         /*
3005          * Turn on usage filter if is not already used.  The idea is
3006          * that chunks that we have already balanced should be
3007          * reasonably full.  Don't do it for chunks that are being
3008          * converted - that will keep us from relocating unconverted
3009          * (albeit full) chunks.
3010          */
3011         if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3012             !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3013                 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
3014                 bctl->data.usage = 90;
3015         }
3016         if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3017             !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3018                 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
3019                 bctl->sys.usage = 90;
3020         }
3021         if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3022             !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3023                 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
3024                 bctl->meta.usage = 90;
3025         }
3026 }
3027
3028 /*
3029  * Should be called with both balance and volume mutexes held to
3030  * serialize other volume operations (add_dev/rm_dev/resize) with
3031  * restriper.  Same goes for unset_balance_control.
3032  */
3033 static void set_balance_control(struct btrfs_balance_control *bctl)
3034 {
3035         struct btrfs_fs_info *fs_info = bctl->fs_info;
3036
3037         BUG_ON(fs_info->balance_ctl);
3038
3039         spin_lock(&fs_info->balance_lock);
3040         fs_info->balance_ctl = bctl;
3041         spin_unlock(&fs_info->balance_lock);
3042 }
3043
3044 static void unset_balance_control(struct btrfs_fs_info *fs_info)
3045 {
3046         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3047
3048         BUG_ON(!fs_info->balance_ctl);
3049
3050         spin_lock(&fs_info->balance_lock);
3051         fs_info->balance_ctl = NULL;
3052         spin_unlock(&fs_info->balance_lock);
3053
3054         kfree(bctl);
3055 }
3056
3057 /*
3058  * Balance filters.  Return 1 if chunk should be filtered out
3059  * (should not be balanced).
3060  */
3061 static int chunk_profiles_filter(u64 chunk_type,
3062                                  struct btrfs_balance_args *bargs)
3063 {
3064         chunk_type = chunk_to_extended(chunk_type) &
3065                                 BTRFS_EXTENDED_PROFILE_MASK;
3066
3067         if (bargs->profiles & chunk_type)
3068                 return 0;
3069
3070         return 1;
3071 }
3072
3073 static int chunk_usage_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
3074                               struct btrfs_balance_args *bargs)
3075 {
3076         struct btrfs_block_group_cache *cache;
3077         u64 chunk_used, user_thresh;
3078         int ret = 1;
3079
3080         cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3081         chunk_used = btrfs_block_group_used(&cache->item);
3082
3083         if (bargs->usage == 0)
3084                 user_thresh = 1;
3085         else if (bargs->usage > 100)
3086                 user_thresh = cache->key.offset;
3087         else
3088                 user_thresh = div_factor_fine(cache->key.offset,
3089                                               bargs->usage);
3090
3091         if (chunk_used < user_thresh)
3092                 ret = 0;
3093
3094         btrfs_put_block_group(cache);
3095         return ret;
3096 }
3097
3098 static int chunk_devid_filter(struct extent_buffer *leaf,
3099                               struct btrfs_chunk *chunk,
3100                               struct btrfs_balance_args *bargs)
3101 {
3102         struct btrfs_stripe *stripe;
3103         int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3104         int i;
3105
3106         for (i = 0; i < num_stripes; i++) {
3107                 stripe = btrfs_stripe_nr(chunk, i);
3108                 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
3109                         return 0;
3110         }
3111
3112         return 1;
3113 }
3114
3115 /* [pstart, pend) */
3116 static int chunk_drange_filter(struct extent_buffer *leaf,
3117                                struct btrfs_chunk *chunk,
3118                                u64 chunk_offset,
3119                                struct btrfs_balance_args *bargs)
3120 {
3121         struct btrfs_stripe *stripe;
3122         int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3123         u64 stripe_offset;
3124         u64 stripe_length;
3125         int factor;
3126         int i;
3127
3128         if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
3129                 return 0;
3130
3131         if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP |
3132              BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) {
3133                 factor = num_stripes / 2;
3134         } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) {
3135                 factor = num_stripes - 1;
3136         } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) {
3137                 factor = num_stripes - 2;
3138         } else {
3139                 factor = num_stripes;
3140         }
3141
3142         for (i = 0; i < num_stripes; i++) {
3143                 stripe = btrfs_stripe_nr(chunk, i);
3144                 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
3145                         continue;
3146
3147                 stripe_offset = btrfs_stripe_offset(leaf, stripe);
3148                 stripe_length = btrfs_chunk_length(leaf, chunk);
3149                 stripe_length = div_u64(stripe_length, factor);
3150
3151                 if (stripe_offset < bargs->pend &&
3152                     stripe_offset + stripe_length > bargs->pstart)
3153                         return 0;
3154         }
3155
3156         return 1;
3157 }
3158
3159 /* [vstart, vend) */
3160 static int chunk_vrange_filter(struct extent_buffer *leaf,
3161                                struct btrfs_chunk *chunk,
3162                                u64 chunk_offset,
3163                                struct btrfs_balance_args *bargs)
3164 {
3165         if (chunk_offset < bargs->vend &&
3166             chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
3167                 /* at least part of the chunk is inside this vrange */
3168                 return 0;
3169
3170         return 1;
3171 }
3172
3173 static int chunk_soft_convert_filter(u64 chunk_type,
3174                                      struct btrfs_balance_args *bargs)
3175 {
3176         if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
3177                 return 0;
3178
3179         chunk_type = chunk_to_extended(chunk_type) &
3180                                 BTRFS_EXTENDED_PROFILE_MASK;
3181
3182         if (bargs->target == chunk_type)
3183                 return 1;
3184
3185         return 0;
3186 }
3187
3188 static int should_balance_chunk(struct btrfs_root *root,
3189                                 struct extent_buffer *leaf,
3190                                 struct btrfs_chunk *chunk, u64 chunk_offset)
3191 {
3192         struct btrfs_balance_control *bctl = root->fs_info->balance_ctl;
3193         struct btrfs_balance_args *bargs = NULL;
3194         u64 chunk_type = btrfs_chunk_type(leaf, chunk);
3195
3196         /* type filter */
3197         if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
3198               (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
3199                 return 0;
3200         }
3201
3202         if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3203                 bargs = &bctl->data;
3204         else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3205                 bargs = &bctl->sys;
3206         else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3207                 bargs = &bctl->meta;
3208
3209         /* profiles filter */
3210         if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
3211             chunk_profiles_filter(chunk_type, bargs)) {
3212                 return 0;
3213         }
3214
3215         /* usage filter */
3216         if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
3217             chunk_usage_filter(bctl->fs_info, chunk_offset, bargs)) {
3218                 return 0;
3219         }
3220
3221         /* devid filter */
3222         if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
3223             chunk_devid_filter(leaf, chunk, bargs)) {
3224                 return 0;
3225         }
3226
3227         /* drange filter, makes sense only with devid filter */
3228         if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
3229             chunk_drange_filter(leaf, chunk, chunk_offset, bargs)) {
3230                 return 0;
3231         }
3232
3233         /* vrange filter */
3234         if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
3235             chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
3236                 return 0;
3237         }
3238
3239         /* soft profile changing mode */
3240         if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
3241             chunk_soft_convert_filter(chunk_type, bargs)) {
3242                 return 0;
3243         }
3244
3245         /*
3246          * limited by count, must be the last filter
3247          */
3248         if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) {
3249                 if (bargs->limit == 0)
3250                         return 0;
3251                 else
3252                         bargs->limit--;
3253         }
3254
3255         return 1;
3256 }
3257
3258 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
3259 {
3260         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3261         struct btrfs_root *chunk_root = fs_info->chunk_root;
3262         struct btrfs_root *dev_root = fs_info->dev_root;
3263         struct list_head *devices;
3264         struct btrfs_device *device;
3265         u64 old_size;
3266         u64 size_to_free;
3267         struct btrfs_chunk *chunk;
3268         struct btrfs_path *path;
3269         struct btrfs_key key;
3270         struct btrfs_key found_key;
3271         struct btrfs_trans_handle *trans;
3272         struct extent_buffer *leaf;
3273         int slot;
3274         int ret;
3275         int enospc_errors = 0;
3276         bool counting = true;
3277         u64 limit_data = bctl->data.limit;
3278         u64 limit_meta = bctl->meta.limit;
3279         u64 limit_sys = bctl->sys.limit;
3280
3281         /* step one make some room on all the devices */
3282         devices = &fs_info->fs_devices->devices;
3283         list_for_each_entry(device, devices, dev_list) {
3284                 old_size = btrfs_device_get_total_bytes(device);
3285                 size_to_free = div_factor(old_size, 1);
3286                 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
3287                 if (!device->writeable ||
3288                     btrfs_device_get_total_bytes(device) -
3289                     btrfs_device_get_bytes_used(device) > size_to_free ||
3290                     device->is_tgtdev_for_dev_replace)
3291                         continue;
3292
3293                 ret = btrfs_shrink_device(device, old_size - size_to_free);
3294                 if (ret == -ENOSPC)
3295                         break;
3296                 BUG_ON(ret);
3297
3298                 trans = btrfs_start_transaction(dev_root, 0);
3299                 BUG_ON(IS_ERR(trans));
3300
3301                 ret = btrfs_grow_device(trans, device, old_size);
3302                 BUG_ON(ret);
3303
3304                 btrfs_end_transaction(trans, dev_root);
3305         }
3306
3307         /* step two, relocate all the chunks */
3308         path = btrfs_alloc_path();
3309         if (!path) {
3310                 ret = -ENOMEM;
3311                 goto error;
3312         }
3313
3314         /* zero out stat counters */
3315         spin_lock(&fs_info->balance_lock);
3316         memset(&bctl->stat, 0, sizeof(bctl->stat));
3317         spin_unlock(&fs_info->balance_lock);
3318 again:
3319         if (!counting) {
3320                 bctl->data.limit = limit_data;
3321                 bctl->meta.limit = limit_meta;
3322                 bctl->sys.limit = limit_sys;
3323         }
3324         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3325         key.offset = (u64)-1;
3326         key.type = BTRFS_CHUNK_ITEM_KEY;
3327
3328         while (1) {
3329                 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
3330                     atomic_read(&fs_info->balance_cancel_req)) {
3331                         ret = -ECANCELED;
3332                         goto error;
3333                 }
3334
3335                 mutex_lock(&fs_info->delete_unused_bgs_mutex);
3336                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3337                 if (ret < 0) {
3338                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3339                         goto error;
3340                 }
3341
3342                 /*
3343                  * this shouldn't happen, it means the last relocate
3344                  * failed
3345                  */
3346                 if (ret == 0)
3347                         BUG(); /* FIXME break ? */
3348
3349                 ret = btrfs_previous_item(chunk_root, path, 0,
3350                                           BTRFS_CHUNK_ITEM_KEY);
3351                 if (ret) {
3352                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3353                         ret = 0;
3354                         break;
3355                 }
3356
3357                 leaf = path->nodes[0];
3358                 slot = path->slots[0];
3359                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3360
3361                 if (found_key.objectid != key.objectid) {
3362                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3363                         break;
3364                 }
3365
3366                 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3367
3368                 if (!counting) {
3369                         spin_lock(&fs_info->balance_lock);
3370                         bctl->stat.considered++;
3371                         spin_unlock(&fs_info->balance_lock);
3372                 }
3373
3374                 ret = should_balance_chunk(chunk_root, leaf, chunk,
3375                                            found_key.offset);
3376                 btrfs_release_path(path);
3377                 if (!ret) {
3378                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3379                         goto loop;
3380                 }
3381
3382                 if (counting) {
3383                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3384                         spin_lock(&fs_info->balance_lock);
3385                         bctl->stat.expected++;
3386                         spin_unlock(&fs_info->balance_lock);
3387                         goto loop;
3388                 }
3389
3390                 ret = btrfs_relocate_chunk(chunk_root,
3391                                            found_key.offset);
3392                 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3393                 if (ret && ret != -ENOSPC)
3394                         goto error;
3395                 if (ret == -ENOSPC) {
3396                         enospc_errors++;
3397                 } else {
3398                         spin_lock(&fs_info->balance_lock);
3399                         bctl->stat.completed++;
3400                         spin_unlock(&fs_info->balance_lock);
3401                 }
3402 loop:
3403                 if (found_key.offset == 0)
3404                         break;
3405                 key.offset = found_key.offset - 1;
3406         }
3407
3408         if (counting) {
3409                 btrfs_release_path(path);
3410                 counting = false;
3411                 goto again;
3412         }
3413 error:
3414         btrfs_free_path(path);
3415         if (enospc_errors) {
3416                 btrfs_info(fs_info, "%d enospc errors during balance",
3417                        enospc_errors);
3418                 if (!ret)
3419                         ret = -ENOSPC;
3420         }
3421
3422         return ret;
3423 }
3424
3425 /**
3426  * alloc_profile_is_valid - see if a given profile is valid and reduced
3427  * @flags: profile to validate
3428  * @extended: if true @flags is treated as an extended profile
3429  */
3430 static int alloc_profile_is_valid(u64 flags, int extended)
3431 {
3432         u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
3433                                BTRFS_BLOCK_GROUP_PROFILE_MASK);
3434
3435         flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
3436
3437         /* 1) check that all other bits are zeroed */
3438         if (flags & ~mask)
3439                 return 0;
3440
3441         /* 2) see if profile is reduced */
3442         if (flags == 0)
3443                 return !extended; /* "0" is valid for usual profiles */
3444
3445         /* true if exactly one bit set */
3446         return (flags & (flags - 1)) == 0;
3447 }
3448
3449 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
3450 {
3451         /* cancel requested || normal exit path */
3452         return atomic_read(&fs_info->balance_cancel_req) ||
3453                 (atomic_read(&fs_info->balance_pause_req) == 0 &&
3454                  atomic_read(&fs_info->balance_cancel_req) == 0);
3455 }
3456
3457 static void __cancel_balance(struct btrfs_fs_info *fs_info)
3458 {
3459         int ret;
3460
3461         unset_balance_control(fs_info);
3462         ret = del_balance_item(fs_info->tree_root);
3463         if (ret)
3464                 btrfs_std_error(fs_info, ret);
3465
3466         atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3467 }
3468
3469 /*
3470  * Should be called with both balance and volume mutexes held
3471  */
3472 int btrfs_balance(struct btrfs_balance_control *bctl,
3473                   struct btrfs_ioctl_balance_args *bargs)
3474 {
3475         struct btrfs_fs_info *fs_info = bctl->fs_info;
3476         u64 allowed;
3477         int mixed = 0;
3478         int ret;
3479         u64 num_devices;
3480         unsigned seq;
3481
3482         if (btrfs_fs_closing(fs_info) ||
3483             atomic_read(&fs_info->balance_pause_req) ||
3484             atomic_read(&fs_info->balance_cancel_req)) {
3485                 ret = -EINVAL;
3486                 goto out;
3487         }
3488
3489         allowed = btrfs_super_incompat_flags(fs_info->super_copy);
3490         if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
3491                 mixed = 1;
3492
3493         /*
3494          * In case of mixed groups both data and meta should be picked,
3495          * and identical options should be given for both of them.
3496          */
3497         allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
3498         if (mixed && (bctl->flags & allowed)) {
3499                 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
3500                     !(bctl->flags & BTRFS_BALANCE_METADATA) ||
3501                     memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
3502                         btrfs_err(fs_info, "with mixed groups data and "
3503                                    "metadata balance options must be the same");
3504                         ret = -EINVAL;
3505                         goto out;
3506                 }
3507         }
3508
3509         num_devices = fs_info->fs_devices->num_devices;
3510         btrfs_dev_replace_lock(&fs_info->dev_replace);
3511         if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
3512                 BUG_ON(num_devices < 1);
3513                 num_devices--;
3514         }
3515         btrfs_dev_replace_unlock(&fs_info->dev_replace);
3516         allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE;
3517         if (num_devices == 1)
3518                 allowed |= BTRFS_BLOCK_GROUP_DUP;
3519         else if (num_devices > 1)
3520                 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1);
3521         if (num_devices > 2)
3522                 allowed |= BTRFS_BLOCK_GROUP_RAID5;
3523         if (num_devices > 3)
3524                 allowed |= (BTRFS_BLOCK_GROUP_RAID10 |
3525                             BTRFS_BLOCK_GROUP_RAID6);
3526         if ((bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3527             (!alloc_profile_is_valid(bctl->data.target, 1) ||
3528              (bctl->data.target & ~allowed))) {
3529                 btrfs_err(fs_info, "unable to start balance with target "
3530                            "data profile %llu",
3531                        bctl->data.target);
3532                 ret = -EINVAL;
3533                 goto out;
3534         }
3535         if ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3536             (!alloc_profile_is_valid(bctl->meta.target, 1) ||
3537              (bctl->meta.target & ~allowed))) {
3538                 btrfs_err(fs_info,
3539                            "unable to start balance with target metadata profile %llu",
3540                        bctl->meta.target);
3541                 ret = -EINVAL;
3542                 goto out;
3543         }
3544         if ((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3545             (!alloc_profile_is_valid(bctl->sys.target, 1) ||
3546              (bctl->sys.target & ~allowed))) {
3547                 btrfs_err(fs_info,
3548                            "unable to start balance with target system profile %llu",
3549                        bctl->sys.target);
3550                 ret = -EINVAL;
3551                 goto out;
3552         }
3553
3554         /* allow dup'ed data chunks only in mixed mode */
3555         if (!mixed && (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3556             (bctl->data.target & BTRFS_BLOCK_GROUP_DUP)) {
3557                 btrfs_err(fs_info, "dup for data is not allowed");
3558                 ret = -EINVAL;
3559                 goto out;
3560         }
3561
3562         /* allow to reduce meta or sys integrity only if force set */
3563         allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3564                         BTRFS_BLOCK_GROUP_RAID10 |
3565                         BTRFS_BLOCK_GROUP_RAID5 |
3566                         BTRFS_BLOCK_GROUP_RAID6;
3567         do {
3568                 seq = read_seqbegin(&fs_info->profiles_lock);
3569
3570                 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3571                      (fs_info->avail_system_alloc_bits & allowed) &&
3572                      !(bctl->sys.target & allowed)) ||
3573                     ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3574                      (fs_info->avail_metadata_alloc_bits & allowed) &&
3575                      !(bctl->meta.target & allowed))) {
3576                         if (bctl->flags & BTRFS_BALANCE_FORCE) {
3577                                 btrfs_info(fs_info, "force reducing metadata integrity");
3578                         } else {
3579                                 btrfs_err(fs_info, "balance will reduce metadata "
3580                                            "integrity, use force if you want this");
3581                                 ret = -EINVAL;
3582                                 goto out;
3583                         }
3584                 }
3585         } while (read_seqretry(&fs_info->profiles_lock, seq));
3586
3587         if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3588                 int num_tolerated_disk_barrier_failures;
3589                 u64 target = bctl->sys.target;
3590
3591                 num_tolerated_disk_barrier_failures =
3592                         btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3593                 if (num_tolerated_disk_barrier_failures > 0 &&
3594                     (target &
3595                      (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID0 |
3596                       BTRFS_AVAIL_ALLOC_BIT_SINGLE)))
3597                         num_tolerated_disk_barrier_failures = 0;
3598                 else if (num_tolerated_disk_barrier_failures > 1 &&
3599                          (target &
3600                           (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)))
3601                         num_tolerated_disk_barrier_failures = 1;
3602
3603                 fs_info->num_tolerated_disk_barrier_failures =
3604                         num_tolerated_disk_barrier_failures;
3605         }
3606
3607         ret = insert_balance_item(fs_info->tree_root, bctl);
3608         if (ret && ret != -EEXIST)
3609                 goto out;
3610
3611         if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
3612                 BUG_ON(ret == -EEXIST);
3613                 set_balance_control(bctl);
3614         } else {
3615                 BUG_ON(ret != -EEXIST);
3616                 spin_lock(&fs_info->balance_lock);
3617                 update_balance_args(bctl);
3618                 spin_unlock(&fs_info->balance_lock);
3619         }
3620
3621         atomic_inc(&fs_info->balance_running);
3622         mutex_unlock(&fs_info->balance_mutex);
3623
3624         ret = __btrfs_balance(fs_info);
3625
3626         mutex_lock(&fs_info->balance_mutex);
3627         atomic_dec(&fs_info->balance_running);
3628
3629         if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3630                 fs_info->num_tolerated_disk_barrier_failures =
3631                         btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3632         }
3633
3634         if (bargs) {
3635                 memset(bargs, 0, sizeof(*bargs));
3636                 update_ioctl_balance_args(fs_info, 0, bargs);
3637         }
3638
3639         if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
3640             balance_need_close(fs_info)) {
3641                 __cancel_balance(fs_info);
3642         }
3643
3644         wake_up(&fs_info->balance_wait_q);
3645
3646         return ret;
3647 out:
3648         if (bctl->flags & BTRFS_BALANCE_RESUME)
3649                 __cancel_balance(fs_info);
3650         else {
3651                 kfree(bctl);
3652                 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3653         }
3654         return ret;
3655 }
3656
3657 static int balance_kthread(void *data)
3658 {
3659         struct btrfs_fs_info *fs_info = data;
3660         int ret = 0;
3661
3662         mutex_lock(&fs_info->volume_mutex);
3663         mutex_lock(&fs_info->balance_mutex);
3664
3665         if (fs_info->balance_ctl) {
3666                 btrfs_info(fs_info, "continuing balance");
3667                 ret = btrfs_balance(fs_info->balance_ctl, NULL);
3668         }
3669
3670         mutex_unlock(&fs_info->balance_mutex);
3671         mutex_unlock(&fs_info->volume_mutex);
3672
3673         return ret;
3674 }
3675
3676 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
3677 {
3678         struct task_struct *tsk;
3679
3680         spin_lock(&fs_info->balance_lock);
3681         if (!fs_info->balance_ctl) {
3682                 spin_unlock(&fs_info->balance_lock);
3683                 return 0;
3684         }
3685         spin_unlock(&fs_info->balance_lock);
3686
3687         if (btrfs_test_opt(fs_info->tree_root, SKIP_BALANCE)) {
3688                 btrfs_info(fs_info, "force skipping balance");
3689                 return 0;
3690         }
3691
3692         tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
3693         return PTR_ERR_OR_ZERO(tsk);
3694 }
3695
3696 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
3697 {
3698         struct btrfs_balance_control *bctl;
3699         struct btrfs_balance_item *item;
3700         struct btrfs_disk_balance_args disk_bargs;
3701         struct btrfs_path *path;
3702         struct extent_buffer *leaf;
3703         struct btrfs_key key;
3704         int ret;
3705
3706         path = btrfs_alloc_path();
3707         if (!path)
3708                 return -ENOMEM;
3709
3710         key.objectid = BTRFS_BALANCE_OBJECTID;
3711         key.type = BTRFS_BALANCE_ITEM_KEY;
3712         key.offset = 0;
3713
3714         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
3715         if (ret < 0)
3716                 goto out;
3717         if (ret > 0) { /* ret = -ENOENT; */
3718                 ret = 0;
3719                 goto out;
3720         }
3721
3722         bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
3723         if (!bctl) {
3724                 ret = -ENOMEM;
3725                 goto out;
3726         }
3727
3728         leaf = path->nodes[0];
3729         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3730
3731         bctl->fs_info = fs_info;
3732         bctl->flags = btrfs_balance_flags(leaf, item);
3733         bctl->flags |= BTRFS_BALANCE_RESUME;
3734
3735         btrfs_balance_data(leaf, item, &disk_bargs);
3736         btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
3737         btrfs_balance_meta(leaf, item, &disk_bargs);
3738         btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
3739         btrfs_balance_sys(leaf, item, &disk_bargs);
3740         btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
3741
3742         WARN_ON(atomic_xchg(&fs_info->mutually_exclusive_operation_running, 1));
3743
3744         mutex_lock(&fs_info->volume_mutex);
3745         mutex_lock(&fs_info->balance_mutex);
3746
3747         set_balance_control(bctl);
3748
3749         mutex_unlock(&fs_info->balance_mutex);
3750         mutex_unlock(&fs_info->volume_mutex);
3751 out:
3752         btrfs_free_path(path);
3753         return ret;
3754 }
3755
3756 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
3757 {
3758         int ret = 0;
3759
3760         mutex_lock(&fs_info->balance_mutex);
3761         if (!fs_info->balance_ctl) {
3762                 mutex_unlock(&fs_info->balance_mutex);
3763                 return -ENOTCONN;
3764         }
3765
3766         if (atomic_read(&fs_info->balance_running)) {
3767                 atomic_inc(&fs_info->balance_pause_req);
3768                 mutex_unlock(&fs_info->balance_mutex);
3769
3770                 wait_event(fs_info->balance_wait_q,
3771                            atomic_read(&fs_info->balance_running) == 0);
3772
3773                 mutex_lock(&fs_info->balance_mutex);
3774                 /* we are good with balance_ctl ripped off from under us */
3775                 BUG_ON(atomic_read(&fs_info->balance_running));
3776                 atomic_dec(&fs_info->balance_pause_req);
3777         } else {
3778                 ret = -ENOTCONN;
3779         }
3780
3781         mutex_unlock(&fs_info->balance_mutex);
3782         return ret;
3783 }
3784
3785 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
3786 {
3787         if (fs_info->sb->s_flags & MS_RDONLY)
3788                 return -EROFS;
3789
3790         mutex_lock(&fs_info->balance_mutex);
3791         if (!fs_info->balance_ctl) {
3792                 mutex_unlock(&fs_info->balance_mutex);
3793                 return -ENOTCONN;
3794         }
3795
3796         atomic_inc(&fs_info->balance_cancel_req);
3797         /*
3798          * if we are running just wait and return, balance item is
3799          * deleted in btrfs_balance in this case
3800          */
3801         if (atomic_read(&fs_info->balance_running)) {
3802                 mutex_unlock(&fs_info->balance_mutex);
3803                 wait_event(fs_info->balance_wait_q,
3804                            atomic_read(&fs_info->balance_running) == 0);
3805                 mutex_lock(&fs_info->balance_mutex);
3806         } else {
3807                 /* __cancel_balance needs volume_mutex */
3808                 mutex_unlock(&fs_info->balance_mutex);
3809                 mutex_lock(&fs_info->volume_mutex);
3810                 mutex_lock(&fs_info->balance_mutex);
3811
3812                 if (fs_info->balance_ctl)
3813                         __cancel_balance(fs_info);
3814
3815                 mutex_unlock(&fs_info->volume_mutex);
3816         }
3817
3818         BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running));
3819         atomic_dec(&fs_info->balance_cancel_req);
3820         mutex_unlock(&fs_info->balance_mutex);
3821         return 0;
3822 }
3823
3824 static int btrfs_uuid_scan_kthread(void *data)
3825 {
3826         struct btrfs_fs_info *fs_info = data;
3827         struct btrfs_root *root = fs_info->tree_root;
3828         struct btrfs_key key;
3829         struct btrfs_key max_key;
3830         struct btrfs_path *path = NULL;
3831         int ret = 0;
3832         struct extent_buffer *eb;
3833         int slot;
3834         struct btrfs_root_item root_item;
3835         u32 item_size;
3836         struct btrfs_trans_handle *trans = NULL;
3837
3838         path = btrfs_alloc_path();
3839         if (!path) {
3840                 ret = -ENOMEM;
3841                 goto out;
3842         }
3843
3844         key.objectid = 0;
3845         key.type = BTRFS_ROOT_ITEM_KEY;
3846         key.offset = 0;
3847
3848         max_key.objectid = (u64)-1;
3849         max_key.type = BTRFS_ROOT_ITEM_KEY;
3850         max_key.offset = (u64)-1;
3851
3852         while (1) {
3853                 ret = btrfs_search_forward(root, &key, path, 0);
3854                 if (ret) {
3855                         if (ret > 0)
3856                                 ret = 0;
3857                         break;
3858                 }
3859
3860                 if (key.type != BTRFS_ROOT_ITEM_KEY ||
3861                     (key.objectid < BTRFS_FIRST_FREE_OBJECTID &&
3862                      key.objectid != BTRFS_FS_TREE_OBJECTID) ||
3863                     key.objectid > BTRFS_LAST_FREE_OBJECTID)
3864                         goto skip;
3865
3866                 eb = path->nodes[0];
3867                 slot = path->slots[0];
3868                 item_size = btrfs_item_size_nr(eb, slot);
3869                 if (item_size < sizeof(root_item))
3870                         goto skip;
3871
3872                 read_extent_buffer(eb, &root_item,
3873                                    btrfs_item_ptr_offset(eb, slot),
3874                                    (int)sizeof(root_item));
3875                 if (btrfs_root_refs(&root_item) == 0)
3876                         goto skip;
3877
3878                 if (!btrfs_is_empty_uuid(root_item.uuid) ||
3879                     !btrfs_is_empty_uuid(root_item.received_uuid)) {
3880                         if (trans)
3881                                 goto update_tree;
3882
3883                         btrfs_release_path(path);
3884                         /*
3885                          * 1 - subvol uuid item
3886                          * 1 - received_subvol uuid item
3887                          */
3888                         trans = btrfs_start_transaction(fs_info->uuid_root, 2);
3889                         if (IS_ERR(trans)) {
3890                                 ret = PTR_ERR(trans);
3891                                 break;
3892                         }
3893                         continue;
3894                 } else {
3895                         goto skip;
3896                 }
3897 update_tree:
3898                 if (!btrfs_is_empty_uuid(root_item.uuid)) {
3899                         ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
3900                                                   root_item.uuid,
3901                                                   BTRFS_UUID_KEY_SUBVOL,
3902                                                   key.objectid);
3903                         if (ret < 0) {
3904                                 btrfs_warn(fs_info, "uuid_tree_add failed %d",
3905                                         ret);
3906                                 break;
3907                         }
3908                 }
3909
3910                 if (!btrfs_is_empty_uuid(root_item.received_uuid)) {
3911                         ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
3912                                                   root_item.received_uuid,
3913                                                  BTRFS_UUID_KEY_RECEIVED_SUBVOL,
3914                                                   key.objectid);
3915                         if (ret < 0) {
3916                                 btrfs_warn(fs_info, "uuid_tree_add failed %d",
3917                                         ret);
3918                                 break;
3919                         }
3920                 }
3921
3922 skip:
3923                 if (trans) {
3924                         ret = btrfs_end_transaction(trans, fs_info->uuid_root);
3925                         trans = NULL;
3926                         if (ret)
3927                                 break;
3928                 }
3929
3930                 btrfs_release_path(path);
3931                 if (key.offset < (u64)-1) {
3932                         key.offset++;
3933                 } else if (key.type < BTRFS_ROOT_ITEM_KEY) {
3934                         key.offset = 0;
3935                         key.type = BTRFS_ROOT_ITEM_KEY;
3936                 } else if (key.objectid < (u64)-1) {
3937                         key.offset = 0;
3938                         key.type = BTRFS_ROOT_ITEM_KEY;
3939                         key.objectid++;
3940                 } else {
3941                         break;
3942                 }
3943                 cond_resched();
3944         }
3945
3946 out:
3947         btrfs_free_path(path);
3948         if (trans && !IS_ERR(trans))
3949                 btrfs_end_transaction(trans, fs_info->uuid_root);
3950         if (ret)
3951                 btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret);
3952         else
3953                 fs_info->update_uuid_tree_gen = 1;
3954         up(&fs_info->uuid_tree_rescan_sem);
3955         return 0;
3956 }
3957
3958 /*
3959  * Callback for btrfs_uuid_tree_iterate().
3960  * returns:
3961  * 0    check succeeded, the entry is not outdated.
3962  * < 0  if an error occured.
3963  * > 0  if the check failed, which means the caller shall remove the entry.
3964  */
3965 static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info *fs_info,
3966                                        u8 *uuid, u8 type, u64 subid)
3967 {
3968         struct btrfs_key key;
3969         int ret = 0;
3970         struct btrfs_root *subvol_root;
3971
3972         if (type != BTRFS_UUID_KEY_SUBVOL &&
3973             type != BTRFS_UUID_KEY_RECEIVED_SUBVOL)
3974                 goto out;
3975
3976         key.objectid = subid;
3977         key.type = BTRFS_ROOT_ITEM_KEY;
3978         key.offset = (u64)-1;
3979         subvol_root = btrfs_read_fs_root_no_name(fs_info, &key);
3980         if (IS_ERR(subvol_root)) {
3981                 ret = PTR_ERR(subvol_root);
3982                 if (ret == -ENOENT)
3983                         ret = 1;
3984                 goto out;
3985         }
3986
3987         switch (type) {
3988         case BTRFS_UUID_KEY_SUBVOL:
3989                 if (memcmp(uuid, subvol_root->root_item.uuid, BTRFS_UUID_SIZE))
3990                         ret = 1;
3991                 break;
3992         case BTRFS_UUID_KEY_RECEIVED_SUBVOL:
3993                 if (memcmp(uuid, subvol_root->root_item.received_uuid,
3994                            BTRFS_UUID_SIZE))
3995                         ret = 1;
3996                 break;
3997         }
3998
3999 out:
4000         return ret;
4001 }
4002
4003 static int btrfs_uuid_rescan_kthread(void *data)
4004 {
4005         struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data;
4006         int ret;
4007
4008         /*
4009          * 1st step is to iterate through the existing UUID tree and
4010          * to delete all entries that contain outdated data.
4011          * 2nd step is to add all missing entries to the UUID tree.
4012          */
4013         ret = btrfs_uuid_tree_iterate(fs_info, btrfs_check_uuid_tree_entry);
4014         if (ret < 0) {
4015                 btrfs_warn(fs_info, "iterating uuid_tree failed %d", ret);
4016                 up(&fs_info->uuid_tree_rescan_sem);
4017                 return ret;
4018         }
4019         return btrfs_uuid_scan_kthread(data);
4020 }
4021
4022 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info)
4023 {
4024         struct btrfs_trans_handle *trans;
4025         struct btrfs_root *tree_root = fs_info->tree_root;
4026         struct btrfs_root *uuid_root;
4027         struct task_struct *task;
4028         int ret;
4029
4030         /*
4031          * 1 - root node
4032          * 1 - root item
4033          */
4034         trans = btrfs_start_transaction(tree_root, 2);
4035         if (IS_ERR(trans))
4036                 return PTR_ERR(trans);
4037
4038         uuid_root = btrfs_create_tree(trans, fs_info,
4039                                       BTRFS_UUID_TREE_OBJECTID);
4040         if (IS_ERR(uuid_root)) {
4041                 ret = PTR_ERR(uuid_root);
4042                 btrfs_abort_transaction(trans, tree_root, ret);
4043                 return ret;
4044         }
4045
4046         fs_info->uuid_root = uuid_root;
4047
4048         ret = btrfs_commit_transaction(trans, tree_root);
4049         if (ret)
4050                 return ret;
4051
4052         down(&fs_info->uuid_tree_rescan_sem);
4053         task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid");
4054         if (IS_ERR(task)) {
4055                 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4056                 btrfs_warn(fs_info, "failed to start uuid_scan task");
4057                 up(&fs_info->uuid_tree_rescan_sem);
4058                 return PTR_ERR(task);
4059         }
4060
4061         return 0;
4062 }
4063
4064 int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
4065 {
4066         struct task_struct *task;
4067
4068         down(&fs_info->uuid_tree_rescan_sem);
4069         task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
4070         if (IS_ERR(task)) {
4071                 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4072                 btrfs_warn(fs_info, "failed to start uuid_rescan task");
4073                 up(&fs_info->uuid_tree_rescan_sem);
4074                 return PTR_ERR(task);
4075         }
4076
4077         return 0;
4078 }
4079
4080 /*
4081  * shrinking a device means finding all of the device extents past
4082  * the new size, and then following the back refs to the chunks.
4083  * The chunk relocation code actually frees the device extent
4084  */
4085 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
4086 {
4087         struct btrfs_trans_handle *trans;
4088         struct btrfs_root *root = device->dev_root;
4089         struct btrfs_dev_extent *dev_extent = NULL;
4090         struct btrfs_path *path;
4091         u64 length;
4092         u64 chunk_offset;
4093         int ret;
4094         int slot;
4095         int failed = 0;
4096         bool retried = false;
4097         bool checked_pending_chunks = false;
4098         struct extent_buffer *l;
4099         struct btrfs_key key;
4100         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
4101         u64 old_total = btrfs_super_total_bytes(super_copy);
4102         u64 old_size = btrfs_device_get_total_bytes(device);
4103         u64 diff = old_size - new_size;
4104
4105         if (device->is_tgtdev_for_dev_replace)
4106                 return -EINVAL;
4107
4108         path = btrfs_alloc_path();
4109         if (!path)
4110                 return -ENOMEM;
4111
4112         path->reada = 2;
4113
4114         lock_chunks(root);
4115
4116         btrfs_device_set_total_bytes(device, new_size);
4117         if (device->writeable) {
4118                 device->fs_devices->total_rw_bytes -= diff;
4119                 spin_lock(&root->fs_info->free_chunk_lock);
4120                 root->fs_info->free_chunk_space -= diff;
4121                 spin_unlock(&root->fs_info->free_chunk_lock);
4122         }
4123         unlock_chunks(root);
4124
4125 again:
4126         key.objectid = device->devid;
4127         key.offset = (u64)-1;
4128         key.type = BTRFS_DEV_EXTENT_KEY;
4129
4130         do {
4131                 mutex_lock(&root->fs_info->delete_unused_bgs_mutex);
4132                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4133                 if (ret < 0) {
4134                         mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4135                         goto done;
4136                 }
4137
4138                 ret = btrfs_previous_item(root, path, 0, key.type);
4139                 if (ret)
4140                         mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4141                 if (ret < 0)
4142                         goto done;
4143                 if (ret) {
4144                         ret = 0;
4145                         btrfs_release_path(path);
4146                         break;
4147                 }
4148
4149                 l = path->nodes[0];
4150                 slot = path->slots[0];
4151                 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
4152
4153                 if (key.objectid != device->devid) {
4154                         mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4155                         btrfs_release_path(path);
4156                         break;
4157                 }
4158
4159                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
4160                 length = btrfs_dev_extent_length(l, dev_extent);
4161
4162                 if (key.offset + length <= new_size) {
4163                         mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4164                         btrfs_release_path(path);
4165                         break;
4166                 }
4167
4168                 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
4169                 btrfs_release_path(path);
4170
4171                 ret = btrfs_relocate_chunk(root, chunk_offset);
4172                 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4173                 if (ret && ret != -ENOSPC)
4174                         goto done;
4175                 if (ret == -ENOSPC)
4176                         failed++;
4177         } while (key.offset-- > 0);
4178
4179         if (failed && !retried) {
4180                 failed = 0;
4181                 retried = true;
4182                 goto again;
4183         } else if (failed && retried) {
4184                 ret = -ENOSPC;
4185                 goto done;
4186         }
4187
4188         /* Shrinking succeeded, else we would be at "done". */
4189         trans = btrfs_start_transaction(root, 0);
4190         if (IS_ERR(trans)) {
4191                 ret = PTR_ERR(trans);
4192                 goto done;
4193         }
4194
4195         lock_chunks(root);
4196
4197         /*
4198          * We checked in the above loop all device extents that were already in
4199          * the device tree. However before we have updated the device's
4200          * total_bytes to the new size, we might have had chunk allocations that
4201          * have not complete yet (new block groups attached to transaction
4202          * handles), and therefore their device extents were not yet in the
4203          * device tree and we missed them in the loop above. So if we have any
4204          * pending chunk using a device extent that overlaps the device range
4205          * that we can not use anymore, commit the current transaction and
4206          * repeat the search on the device tree - this way we guarantee we will
4207          * not have chunks using device extents that end beyond 'new_size'.
4208          */
4209         if (!checked_pending_chunks) {
4210                 u64 start = new_size;
4211                 u64 len = old_size - new_size;
4212
4213                 if (contains_pending_extent(trans->transaction, device,
4214                                             &start, len)) {
4215                         unlock_chunks(root);
4216                         checked_pending_chunks = true;
4217                         failed = 0;
4218                         retried = false;
4219                         ret = btrfs_commit_transaction(trans, root);
4220                         if (ret)
4221                                 goto done;
4222                         goto again;
4223                 }
4224         }
4225
4226         btrfs_device_set_disk_total_bytes(device, new_size);
4227         if (list_empty(&device->resized_list))
4228                 list_add_tail(&device->resized_list,
4229                               &root->fs_info->fs_devices->resized_devices);
4230
4231         WARN_ON(diff > old_total);
4232         btrfs_set_super_total_bytes(super_copy, old_total - diff);
4233         unlock_chunks(root);
4234
4235         /* Now btrfs_update_device() will change the on-disk size. */
4236         ret = btrfs_update_device(trans, device);
4237         btrfs_end_transaction(trans, root);
4238 done:
4239         btrfs_free_path(path);
4240         if (ret) {
4241                 lock_chunks(root);
4242                 btrfs_device_set_total_bytes(device, old_size);
4243                 if (device->writeable)
4244                         device->fs_devices->total_rw_bytes += diff;
4245                 spin_lock(&root->fs_info->free_chunk_lock);
4246                 root->fs_info->free_chunk_space += diff;
4247                 spin_unlock(&root->fs_info->free_chunk_lock);
4248                 unlock_chunks(root);
4249         }
4250         return ret;
4251 }
4252
4253 static int btrfs_add_system_chunk(struct btrfs_root *root,
4254                            struct btrfs_key *key,
4255                            struct btrfs_chunk *chunk, int item_size)
4256 {
4257         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
4258         struct btrfs_disk_key disk_key;
4259         u32 array_size;
4260         u8 *ptr;
4261
4262         lock_chunks(root);
4263         array_size = btrfs_super_sys_array_size(super_copy);
4264         if (array_size + item_size + sizeof(disk_key)
4265                         > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
4266                 unlock_chunks(root);
4267                 return -EFBIG;
4268         }
4269
4270         ptr = super_copy->sys_chunk_array + array_size;
4271         btrfs_cpu_key_to_disk(&disk_key, key);
4272         memcpy(ptr, &disk_key, sizeof(disk_key));
4273         ptr += sizeof(disk_key);
4274         memcpy(ptr, chunk, item_size);
4275         item_size += sizeof(disk_key);
4276         btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
4277         unlock_chunks(root);
4278
4279         return 0;
4280 }
4281
4282 /*
4283  * sort the devices in descending order by max_avail, total_avail
4284  */
4285 static int btrfs_cmp_device_info(const void *a, const void *b)
4286 {
4287         const struct btrfs_device_info *di_a = a;
4288         const struct btrfs_device_info *di_b = b;
4289
4290         if (di_a->max_avail > di_b->max_avail)
4291                 return -1;
4292         if (di_a->max_avail < di_b->max_avail)
4293                 return 1;
4294         if (di_a->total_avail > di_b->total_avail)
4295                 return -1;
4296         if (di_a->total_avail < di_b->total_avail)
4297                 return 1;
4298         return 0;
4299 }
4300
4301 static const struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = {
4302         [BTRFS_RAID_RAID10] = {
4303                 .sub_stripes    = 2,
4304                 .dev_stripes    = 1,
4305                 .devs_max       = 0,    /* 0 == as many as possible */
4306                 .devs_min       = 4,
4307                 .devs_increment = 2,
4308                 .ncopies        = 2,
4309         },
4310         [BTRFS_RAID_RAID1] = {
4311                 .sub_stripes    = 1,
4312                 .dev_stripes    = 1,
4313                 .devs_max       = 2,
4314                 .devs_min       = 2,
4315                 .devs_increment = 2,
4316                 .ncopies        = 2,
4317         },
4318         [BTRFS_RAID_DUP] = {
4319                 .sub_stripes    = 1,
4320                 .dev_stripes    = 2,
4321                 .devs_max       = 1,
4322                 .devs_min       = 1,
4323                 .devs_increment = 1,
4324                 .ncopies        = 2,
4325         },
4326         [BTRFS_RAID_RAID0] = {
4327                 .sub_stripes    = 1,
4328                 .dev_stripes    = 1,
4329                 .devs_max       = 0,
4330                 .devs_min       = 2,
4331                 .devs_increment = 1,
4332                 .ncopies        = 1,
4333         },
4334         [BTRFS_RAID_SINGLE] = {
4335                 .sub_stripes    = 1,
4336                 .dev_stripes    = 1,
4337                 .devs_max       = 1,
4338                 .devs_min       = 1,
4339                 .devs_increment = 1,
4340                 .ncopies        = 1,
4341         },
4342         [BTRFS_RAID_RAID5] = {
4343                 .sub_stripes    = 1,
4344                 .dev_stripes    = 1,
4345                 .devs_max       = 0,
4346                 .devs_min       = 2,
4347                 .devs_increment = 1,
4348                 .ncopies        = 2,
4349         },
4350         [BTRFS_RAID_RAID6] = {
4351                 .sub_stripes    = 1,
4352                 .dev_stripes    = 1,
4353                 .devs_max       = 0,
4354                 .devs_min       = 3,
4355                 .devs_increment = 1,
4356                 .ncopies        = 3,
4357         },
4358 };
4359
4360 static u32 find_raid56_stripe_len(u32 data_devices, u32 dev_stripe_target)
4361 {
4362         /* TODO allow them to set a preferred stripe size */
4363         return 64 * 1024;
4364 }
4365
4366 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
4367 {
4368         if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK))
4369                 return;
4370
4371         btrfs_set_fs_incompat(info, RAID56);
4372 }
4373
4374 #define BTRFS_MAX_DEVS(r) ((BTRFS_LEAF_DATA_SIZE(r)             \
4375                         - sizeof(struct btrfs_item)             \
4376                         - sizeof(struct btrfs_chunk))           \
4377                         / sizeof(struct btrfs_stripe) + 1)
4378
4379 #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE        \
4380                                 - 2 * sizeof(struct btrfs_disk_key)     \
4381                                 - 2 * sizeof(struct btrfs_chunk))       \
4382                                 / sizeof(struct btrfs_stripe) + 1)
4383
4384 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4385                                struct btrfs_root *extent_root, u64 start,
4386                                u64 type)
4387 {
4388         struct btrfs_fs_info *info = extent_root->fs_info;
4389         struct btrfs_fs_devices *fs_devices = info->fs_devices;
4390         struct list_head *cur;
4391         struct map_lookup *map = NULL;
4392         struct extent_map_tree *em_tree;
4393         struct extent_map *em;
4394         struct btrfs_device_info *devices_info = NULL;
4395         u64 total_avail;
4396         int num_stripes;        /* total number of stripes to allocate */
4397         int data_stripes;       /* number of stripes that count for
4398                                    block group size */
4399         int sub_stripes;        /* sub_stripes info for map */
4400         int dev_stripes;        /* stripes per dev */
4401         int devs_max;           /* max devs to use */
4402         int devs_min;           /* min devs needed */
4403         int devs_increment;     /* ndevs has to be a multiple of this */
4404         int ncopies;            /* how many copies to data has */
4405         int ret;
4406         u64 max_stripe_size;
4407         u64 max_chunk_size;
4408         u64 stripe_size;
4409         u64 num_bytes;
4410         u64 raid_stripe_len = BTRFS_STRIPE_LEN;
4411         int ndevs;
4412         int i;
4413         int j;
4414         int index;
4415
4416         BUG_ON(!alloc_profile_is_valid(type, 0));
4417
4418         if (list_empty(&fs_devices->alloc_list))
4419                 return -ENOSPC;
4420
4421         index = __get_raid_index(type);
4422
4423         sub_stripes = btrfs_raid_array[index].sub_stripes;
4424         dev_stripes = btrfs_raid_array[index].dev_stripes;
4425         devs_max = btrfs_raid_array[index].devs_max;
4426         devs_min = btrfs_raid_array[index].devs_min;
4427         devs_increment = btrfs_raid_array[index].devs_increment;
4428         ncopies = btrfs_raid_array[index].ncopies;
4429
4430         if (type & BTRFS_BLOCK_GROUP_DATA) {
4431                 max_stripe_size = 1024 * 1024 * 1024;
4432                 max_chunk_size = 10 * max_stripe_size;
4433                 if (!devs_max)
4434                         devs_max = BTRFS_MAX_DEVS(info->chunk_root);
4435         } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
4436                 /* for larger filesystems, use larger metadata chunks */
4437                 if (fs_devices->total_rw_bytes > 50ULL * 1024 * 1024 * 1024)
4438                         max_stripe_size = 1024 * 1024 * 1024;
4439                 else
4440                         max_stripe_size = 256 * 1024 * 1024;
4441                 max_chunk_size = max_stripe_size;
4442                 if (!devs_max)
4443                         devs_max = BTRFS_MAX_DEVS(info->chunk_root);
4444         } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
4445                 max_stripe_size = 32 * 1024 * 1024;
4446                 max_chunk_size = 2 * max_stripe_size;
4447                 if (!devs_max)
4448                         devs_max = BTRFS_MAX_DEVS_SYS_CHUNK;
4449         } else {
4450                 btrfs_err(info, "invalid chunk type 0x%llx requested",
4451                        type);
4452                 BUG_ON(1);
4453         }
4454
4455         /* we don't want a chunk larger than 10% of writeable space */
4456         max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
4457                              max_chunk_size);
4458
4459         devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info),
4460                                GFP_NOFS);
4461         if (!devices_info)
4462                 return -ENOMEM;
4463
4464         cur = fs_devices->alloc_list.next;
4465
4466         /*
4467          * in the first pass through the devices list, we gather information
4468          * about the available holes on each device.
4469          */
4470         ndevs = 0;
4471         while (cur != &fs_devices->alloc_list) {
4472                 struct btrfs_device *device;
4473                 u64 max_avail;
4474                 u64 dev_offset;
4475
4476                 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
4477
4478                 cur = cur->next;
4479
4480                 if (!device->writeable) {
4481                         WARN(1, KERN_ERR
4482                                "BTRFS: read-only device in alloc_list\n");
4483                         continue;
4484                 }
4485
4486                 if (!device->in_fs_metadata ||
4487                     device->is_tgtdev_for_dev_replace)
4488                         continue;
4489
4490                 if (device->total_bytes > device->bytes_used)
4491                         total_avail = device->total_bytes - device->bytes_used;
4492                 else
4493                         total_avail = 0;
4494
4495                 /* If there is no space on this device, skip it. */
4496                 if (total_avail == 0)
4497                         continue;
4498
4499                 ret = find_free_dev_extent(trans, device,
4500                                            max_stripe_size * dev_stripes,
4501                                            &dev_offset, &max_avail);
4502                 if (ret && ret != -ENOSPC)
4503                         goto error;
4504
4505                 if (ret == 0)
4506                         max_avail = max_stripe_size * dev_stripes;
4507
4508                 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes)
4509                         continue;
4510
4511                 if (ndevs == fs_devices->rw_devices) {
4512                         WARN(1, "%s: found more than %llu devices\n",
4513                              __func__, fs_devices->rw_devices);
4514                         break;
4515                 }
4516                 devices_info[ndevs].dev_offset = dev_offset;
4517                 devices_info[ndevs].max_avail = max_avail;
4518                 devices_info[ndevs].total_avail = total_avail;
4519                 devices_info[ndevs].dev = device;
4520                 ++ndevs;
4521         }
4522
4523         /*
4524          * now sort the devices by hole size / available space
4525          */
4526         sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
4527              btrfs_cmp_device_info, NULL);
4528
4529         /* round down to number of usable stripes */
4530         ndevs -= ndevs % devs_increment;
4531
4532         if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) {
4533                 ret = -ENOSPC;
4534                 goto error;
4535         }
4536
4537         if (devs_max && ndevs > devs_max)
4538                 ndevs = devs_max;
4539         /*
4540          * the primary goal is to maximize the number of stripes, so use as many
4541          * devices as possible, even if the stripes are not maximum sized.
4542          */
4543         stripe_size = devices_info[ndevs-1].max_avail;
4544         num_stripes = ndevs * dev_stripes;
4545
4546         /*
4547          * this will have to be fixed for RAID1 and RAID10 over
4548          * more drives
4549          */
4550         data_stripes = num_stripes / ncopies;
4551
4552         if (type & BTRFS_BLOCK_GROUP_RAID5) {
4553                 raid_stripe_len = find_raid56_stripe_len(ndevs - 1,
4554                                  btrfs_super_stripesize(info->super_copy));
4555                 data_stripes = num_stripes - 1;
4556         }
4557         if (type & BTRFS_BLOCK_GROUP_RAID6) {
4558                 raid_stripe_len = find_raid56_stripe_len(ndevs - 2,
4559                                  btrfs_super_stripesize(info->super_copy));
4560                 data_stripes = num_stripes - 2;
4561         }
4562
4563         /*
4564          * Use the number of data stripes to figure out how big this chunk
4565          * is really going to be in terms of logical address space,
4566          * and compare that answer with the max chunk size
4567          */
4568         if (stripe_size * data_stripes > max_chunk_size) {
4569                 u64 mask = (1ULL << 24) - 1;
4570
4571                 stripe_size = div_u64(max_chunk_size, data_stripes);
4572
4573                 /* bump the answer up to a 16MB boundary */
4574                 stripe_size = (stripe_size + mask) & ~mask;
4575
4576                 /* but don't go higher than the limits we found
4577                  * while searching for free extents
4578                  */
4579                 if (stripe_size > devices_info[ndevs-1].max_avail)
4580                         stripe_size = devices_info[ndevs-1].max_avail;
4581         }
4582
4583         stripe_size = div_u64(stripe_size, dev_stripes);
4584
4585         /* align to BTRFS_STRIPE_LEN */
4586         stripe_size = div_u64(stripe_size, raid_stripe_len);
4587         stripe_size *= raid_stripe_len;
4588
4589         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
4590         if (!map) {
4591                 ret = -ENOMEM;
4592                 goto error;
4593         }
4594         map->num_stripes = num_stripes;
4595
4596         for (i = 0; i < ndevs; ++i) {
4597                 for (j = 0; j < dev_stripes; ++j) {
4598                         int s = i * dev_stripes + j;
4599                         map->stripes[s].dev = devices_info[i].dev;
4600                         map->stripes[s].physical = devices_info[i].dev_offset +
4601                                                    j * stripe_size;
4602                 }
4603         }
4604         map->sector_size = extent_root->sectorsize;
4605         map->stripe_len = raid_stripe_len;
4606         map->io_align = raid_stripe_len;
4607         map->io_width = raid_stripe_len;
4608         map->type = type;
4609         map->sub_stripes = sub_stripes;
4610
4611         num_bytes = stripe_size * data_stripes;
4612
4613         trace_btrfs_chunk_alloc(info->chunk_root, map, start, num_bytes);
4614
4615         em = alloc_extent_map();
4616         if (!em) {
4617                 kfree(map);
4618                 ret = -ENOMEM;
4619                 goto error;
4620         }
4621         set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
4622         em->bdev = (struct block_device *)map;
4623         em->start = start;
4624         em->len = num_bytes;
4625         em->block_start = 0;
4626         em->block_len = em->len;
4627         em->orig_block_len = stripe_size;
4628
4629         em_tree = &extent_root->fs_info->mapping_tree.map_tree;
4630         write_lock(&em_tree->lock);
4631         ret = add_extent_mapping(em_tree, em, 0);
4632         if (!ret) {
4633                 list_add_tail(&em->list, &trans->transaction->pending_chunks);
4634                 atomic_inc(&em->refs);
4635         }
4636         write_unlock(&em_tree->lock);
4637         if (ret) {
4638                 free_extent_map(em);
4639                 goto error;
4640         }
4641
4642         ret = btrfs_make_block_group(trans, extent_root, 0, type,
4643                                      BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4644                                      start, num_bytes);
4645         if (ret)
4646                 goto error_del_extent;
4647
4648         for (i = 0; i < map->num_stripes; i++) {
4649                 num_bytes = map->stripes[i].dev->bytes_used + stripe_size;
4650                 btrfs_device_set_bytes_used(map->stripes[i].dev, num_bytes);
4651         }
4652
4653         spin_lock(&extent_root->fs_info->free_chunk_lock);
4654         extent_root->fs_info->free_chunk_space -= (stripe_size *
4655                                                    map->num_stripes);
4656         spin_unlock(&extent_root->fs_info->free_chunk_lock);
4657
4658         free_extent_map(em);
4659         check_raid56_incompat_flag(extent_root->fs_info, type);
4660
4661         kfree(devices_info);
4662         return 0;
4663
4664 error_del_extent:
4665         write_lock(&em_tree->lock);
4666         remove_extent_mapping(em_tree, em);
4667         write_unlock(&em_tree->lock);
4668
4669         /* One for our allocation */
4670         free_extent_map(em);
4671         /* One for the tree reference */
4672         free_extent_map(em);
4673         /* One for the pending_chunks list reference */
4674         free_extent_map(em);
4675 error:
4676         kfree(devices_info);
4677         return ret;
4678 }
4679
4680 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans,
4681                                 struct btrfs_root *extent_root,
4682                                 u64 chunk_offset, u64 chunk_size)
4683 {
4684         struct btrfs_key key;
4685         struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
4686         struct btrfs_device *device;
4687         struct btrfs_chunk *chunk;
4688         struct btrfs_stripe *stripe;
4689         struct extent_map_tree *em_tree;
4690         struct extent_map *em;
4691         struct map_lookup *map;
4692         size_t item_size;
4693         u64 dev_offset;
4694         u64 stripe_size;
4695         int i = 0;
4696         int ret;
4697
4698         em_tree = &extent_root->fs_info->mapping_tree.map_tree;
4699         read_lock(&em_tree->lock);
4700         em = lookup_extent_mapping(em_tree, chunk_offset, chunk_size);
4701         read_unlock(&em_tree->lock);
4702
4703         if (!em) {
4704                 btrfs_crit(extent_root->fs_info, "unable to find logical "
4705                            "%Lu len %Lu", chunk_offset, chunk_size);
4706                 return -EINVAL;
4707         }
4708
4709         if (em->start != chunk_offset || em->len != chunk_size) {
4710                 btrfs_crit(extent_root->fs_info, "found a bad mapping, wanted"
4711                           " %Lu-%Lu, found %Lu-%Lu", chunk_offset,
4712                           chunk_size, em->start, em->len);
4713                 free_extent_map(em);
4714                 return -EINVAL;
4715         }
4716
4717         map = (struct map_lookup *)em->bdev;
4718         item_size = btrfs_chunk_item_size(map->num_stripes);
4719         stripe_size = em->orig_block_len;
4720
4721         chunk = kzalloc(item_size, GFP_NOFS);
4722         if (!chunk) {
4723                 ret = -ENOMEM;
4724                 goto out;
4725         }
4726
4727         for (i = 0; i < map->num_stripes; i++) {
4728                 device = map->stripes[i].dev;
4729                 dev_offset = map->stripes[i].physical;
4730
4731                 ret = btrfs_update_device(trans, device);
4732                 if (ret)
4733                         goto out;
4734                 ret = btrfs_alloc_dev_extent(trans, device,
4735                                              chunk_root->root_key.objectid,
4736                                              BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4737                                              chunk_offset, dev_offset,
4738                                              stripe_size);
4739                 if (ret)
4740                         goto out;
4741         }
4742
4743         stripe = &chunk->stripe;
4744         for (i = 0; i < map->num_stripes; i++) {
4745                 device = map->stripes[i].dev;
4746                 dev_offset = map->stripes[i].physical;
4747
4748                 btrfs_set_stack_stripe_devid(stripe, device->devid);
4749                 btrfs_set_stack_stripe_offset(stripe, dev_offset);
4750                 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
4751                 stripe++;
4752         }
4753
4754         btrfs_set_stack_chunk_length(chunk, chunk_size);
4755         btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
4756         btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
4757         btrfs_set_stack_chunk_type(chunk, map->type);
4758         btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
4759         btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
4760         btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
4761         btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
4762         btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
4763
4764         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
4765         key.type = BTRFS_CHUNK_ITEM_KEY;
4766         key.offset = chunk_offset;
4767
4768         ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
4769         if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
4770                 /*
4771                  * TODO: Cleanup of inserted chunk root in case of
4772                  * failure.
4773                  */
4774                 ret = btrfs_add_system_chunk(chunk_root, &key, chunk,
4775                                              item_size);
4776         }
4777
4778 out:
4779         kfree(chunk);
4780         free_extent_map(em);
4781         return ret;
4782 }
4783
4784 /*
4785  * Chunk allocation falls into two parts. The first part does works
4786  * that make the new allocated chunk useable, but not do any operation
4787  * that modifies the chunk tree. The second part does the works that
4788  * require modifying the chunk tree. This division is important for the
4789  * bootstrap process of adding storage to a seed btrfs.
4790  */
4791 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4792                       struct btrfs_root *extent_root, u64 type)
4793 {
4794         u64 chunk_offset;
4795
4796         ASSERT(mutex_is_locked(&extent_root->fs_info->chunk_mutex));
4797         chunk_offset = find_next_chunk(extent_root->fs_info);
4798         return __btrfs_alloc_chunk(trans, extent_root, chunk_offset, type);
4799 }
4800
4801 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
4802                                          struct btrfs_root *root,
4803                                          struct btrfs_device *device)
4804 {
4805         u64 chunk_offset;
4806         u64 sys_chunk_offset;
4807         u64 alloc_profile;
4808         struct btrfs_fs_info *fs_info = root->fs_info;
4809         struct btrfs_root *extent_root = fs_info->extent_root;
4810         int ret;
4811
4812         chunk_offset = find_next_chunk(fs_info);
4813         alloc_profile = btrfs_get_alloc_profile(extent_root, 0);
4814         ret = __btrfs_alloc_chunk(trans, extent_root, chunk_offset,
4815                                   alloc_profile);
4816         if (ret)
4817                 return ret;
4818
4819         sys_chunk_offset = find_next_chunk(root->fs_info);
4820         alloc_profile = btrfs_get_alloc_profile(fs_info->chunk_root, 0);
4821         ret = __btrfs_alloc_chunk(trans, extent_root, sys_chunk_offset,
4822                                   alloc_profile);
4823         return ret;
4824 }
4825
4826 static inline int btrfs_chunk_max_errors(struct map_lookup *map)
4827 {
4828         int max_errors;
4829
4830         if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
4831                          BTRFS_BLOCK_GROUP_RAID10 |
4832                          BTRFS_BLOCK_GROUP_RAID5 |
4833                          BTRFS_BLOCK_GROUP_DUP)) {
4834                 max_errors = 1;
4835         } else if (map->type & BTRFS_BLOCK_GROUP_RAID6) {
4836                 max_errors = 2;
4837         } else {
4838                 max_errors = 0;
4839         }
4840
4841         return max_errors;
4842 }
4843
4844 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
4845 {
4846         struct extent_map *em;
4847         struct map_lookup *map;
4848         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
4849         int readonly = 0;
4850         int miss_ndevs = 0;
4851         int i;
4852
4853         read_lock(&map_tree->map_tree.lock);
4854         em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
4855         read_unlock(&map_tree->map_tree.lock);
4856         if (!em)
4857                 return 1;
4858
4859         map = (struct map_lookup *)em->bdev;
4860         for (i = 0; i < map->num_stripes; i++) {
4861                 if (map->stripes[i].dev->missing) {
4862                         miss_ndevs++;
4863                         continue;
4864                 }
4865
4866                 if (!map->stripes[i].dev->writeable) {
4867                         readonly = 1;
4868                         goto end;
4869                 }
4870         }
4871
4872         /*
4873          * If the number of missing devices is larger than max errors,
4874          * we can not write the data into that chunk successfully, so
4875          * set it readonly.
4876          */
4877         if (miss_ndevs > btrfs_chunk_max_errors(map))
4878                 readonly = 1;
4879 end:
4880         free_extent_map(em);
4881         return readonly;
4882 }
4883
4884 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
4885 {
4886         extent_map_tree_init(&tree->map_tree);
4887 }
4888
4889 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
4890 {
4891         struct extent_map *em;
4892
4893         while (1) {
4894                 write_lock(&tree->map_tree.lock);
4895                 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
4896                 if (em)
4897                         remove_extent_mapping(&tree->map_tree, em);
4898                 write_unlock(&tree->map_tree.lock);
4899                 if (!em)
4900                         break;
4901                 /* once for us */
4902                 free_extent_map(em);
4903                 /* once for the tree */
4904                 free_extent_map(em);
4905         }
4906 }
4907
4908 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
4909 {
4910         struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
4911         struct extent_map *em;
4912         struct map_lookup *map;
4913         struct extent_map_tree *em_tree = &map_tree->map_tree;
4914         int ret;
4915
4916         read_lock(&em_tree->lock);
4917         em = lookup_extent_mapping(em_tree, logical, len);
4918         read_unlock(&em_tree->lock);
4919
4920         /*
4921          * We could return errors for these cases, but that could get ugly and
4922          * we'd probably do the same thing which is just not do anything else
4923          * and exit, so return 1 so the callers don't try to use other copies.
4924          */
4925         if (!em) {
4926                 btrfs_crit(fs_info, "No mapping for %Lu-%Lu", logical,
4927                             logical+len);
4928                 return 1;
4929         }
4930
4931         if (em->start > logical || em->start + em->len < logical) {
4932                 btrfs_crit(fs_info, "Invalid mapping for %Lu-%Lu, got "
4933                             "%Lu-%Lu", logical, logical+len, em->start,
4934                             em->start + em->len);
4935                 free_extent_map(em);
4936                 return 1;
4937         }
4938
4939         map = (struct map_lookup *)em->bdev;
4940         if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
4941                 ret = map->num_stripes;
4942         else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
4943                 ret = map->sub_stripes;
4944         else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
4945                 ret = 2;
4946         else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
4947                 ret = 3;
4948         else
4949                 ret = 1;
4950         free_extent_map(em);
4951
4952         btrfs_dev_replace_lock(&fs_info->dev_replace);
4953         if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace))
4954                 ret++;
4955         btrfs_dev_replace_unlock(&fs_info->dev_replace);
4956
4957         return ret;
4958 }
4959
4960 unsigned long btrfs_full_stripe_len(struct btrfs_root *root,
4961                                     struct btrfs_mapping_tree *map_tree,
4962                                     u64 logical)
4963 {
4964         struct extent_map *em;
4965         struct map_lookup *map;
4966         struct extent_map_tree *em_tree = &map_tree->map_tree;
4967         unsigned long len = root->sectorsize;
4968
4969         read_lock(&em_tree->lock);
4970         em = lookup_extent_mapping(em_tree, logical, len);
4971         read_unlock(&em_tree->lock);
4972         BUG_ON(!em);
4973
4974         BUG_ON(em->start > logical || em->start + em->len < logical);
4975         map = (struct map_lookup *)em->bdev;
4976         if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
4977                 len = map->stripe_len * nr_data_stripes(map);
4978         free_extent_map(em);
4979         return len;
4980 }
4981
4982 int btrfs_is_parity_mirror(struct btrfs_mapping_tree *map_tree,
4983                            u64 logical, u64 len, int mirror_num)
4984 {
4985         struct extent_map *em;
4986         struct map_lookup *map;
4987         struct extent_map_tree *em_tree = &map_tree->map_tree;
4988         int ret = 0;
4989
4990         read_lock(&em_tree->lock);
4991         em = lookup_extent_mapping(em_tree, logical, len);
4992         read_unlock(&em_tree->lock);
4993         BUG_ON(!em);
4994
4995         BUG_ON(em->start > logical || em->start + em->len < logical);
4996         map = (struct map_lookup *)em->bdev;
4997         if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
4998                 ret = 1;
4999         free_extent_map(em);
5000         return ret;
5001 }
5002
5003 static int find_live_mirror(struct btrfs_fs_info *fs_info,
5004                             struct map_lookup *map, int first, int num,
5005                             int optimal, int dev_replace_is_ongoing)
5006 {
5007         int i;
5008         int tolerance;
5009         struct btrfs_device *srcdev;
5010
5011         if (dev_replace_is_ongoing &&
5012             fs_info->dev_replace.cont_reading_from_srcdev_mode ==
5013              BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
5014                 srcdev = fs_info->dev_replace.srcdev;
5015         else
5016                 srcdev = NULL;
5017
5018         /*
5019          * try to avoid the drive that is the source drive for a
5020          * dev-replace procedure, only choose it if no other non-missing
5021          * mirror is available
5022          */
5023         for (tolerance = 0; tolerance < 2; tolerance++) {
5024                 if (map->stripes[optimal].dev->bdev &&
5025                     (tolerance || map->stripes[optimal].dev != srcdev))
5026                         return optimal;
5027                 for (i = first; i < first + num; i++) {
5028                         if (map->stripes[i].dev->bdev &&
5029                             (tolerance || map->stripes[i].dev != srcdev))
5030                                 return i;
5031                 }
5032         }
5033
5034         /* we couldn't find one that doesn't fail.  Just return something
5035          * and the io error handling code will clean up eventually
5036          */
5037         return optimal;
5038 }
5039
5040 static inline int parity_smaller(u64 a, u64 b)
5041 {
5042         return a > b;
5043 }
5044
5045 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
5046 static void sort_parity_stripes(struct btrfs_bio *bbio, int num_stripes)
5047 {
5048         struct btrfs_bio_stripe s;
5049         int i;
5050         u64 l;
5051         int again = 1;
5052
5053         while (again) {
5054                 again = 0;
5055                 for (i = 0; i < num_stripes - 1; i++) {
5056                         if (parity_smaller(bbio->raid_map[i],
5057                                            bbio->raid_map[i+1])) {
5058                                 s = bbio->stripes[i];
5059                                 l = bbio->raid_map[i];
5060                                 bbio->stripes[i] = bbio->stripes[i+1];
5061                                 bbio->raid_map[i] = bbio->raid_map[i+1];
5062                                 bbio->stripes[i+1] = s;
5063                                 bbio->raid_map[i+1] = l;
5064
5065                                 again = 1;
5066                         }
5067                 }
5068         }
5069 }
5070
5071 static struct btrfs_bio *alloc_btrfs_bio(int total_stripes, int real_stripes)
5072 {
5073         struct btrfs_bio *bbio = kzalloc(
5074                  /* the size of the btrfs_bio */
5075                 sizeof(struct btrfs_bio) +
5076                 /* plus the variable array for the stripes */
5077                 sizeof(struct btrfs_bio_stripe) * (total_stripes) +
5078                 /* plus the variable array for the tgt dev */
5079                 sizeof(int) * (real_stripes) +
5080                 /*
5081                  * plus the raid_map, which includes both the tgt dev
5082                  * and the stripes
5083                  */
5084                 sizeof(u64) * (total_stripes),
5085                 GFP_NOFS|__GFP_NOFAIL);
5086
5087         atomic_set(&bbio->error, 0);
5088         atomic_set(&bbio->refs, 1);
5089
5090         return bbio;
5091 }
5092
5093 void btrfs_get_bbio(struct btrfs_bio *bbio)
5094 {
5095         WARN_ON(!atomic_read(&bbio->refs));
5096         atomic_inc(&bbio->refs);
5097 }
5098
5099 void btrfs_put_bbio(struct btrfs_bio *bbio)
5100 {
5101         if (!bbio)
5102                 return;
5103         if (atomic_dec_and_test(&bbio->refs))
5104                 kfree(bbio);
5105 }
5106
5107 static int __btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
5108                              u64 logical, u64 *length,
5109                              struct btrfs_bio **bbio_ret,
5110                              int mirror_num, int need_raid_map)
5111 {
5112         struct extent_map *em;
5113         struct map_lookup *map;
5114         struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
5115         struct extent_map_tree *em_tree = &map_tree->map_tree;
5116         u64 offset;
5117         u64 stripe_offset;
5118         u64 stripe_end_offset;
5119         u64 stripe_nr;
5120         u64 stripe_nr_orig;
5121         u64 stripe_nr_end;
5122         u64 stripe_len;
5123         u32 stripe_index;
5124         int i;
5125         int ret = 0;
5126         int num_stripes;
5127         int max_errors = 0;
5128         int tgtdev_indexes = 0;
5129         struct btrfs_bio *bbio = NULL;
5130         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
5131         int dev_replace_is_ongoing = 0;
5132         int num_alloc_stripes;
5133         int patch_the_first_stripe_for_dev_replace = 0;
5134         u64 physical_to_patch_in_first_stripe = 0;
5135         u64 raid56_full_stripe_start = (u64)-1;
5136
5137         read_lock(&em_tree->lock);
5138         em = lookup_extent_mapping(em_tree, logical, *length);
5139         read_unlock(&em_tree->lock);
5140
5141         if (!em) {
5142                 btrfs_crit(fs_info, "unable to find logical %llu len %llu",
5143                         logical, *length);
5144                 return -EINVAL;
5145         }
5146
5147         if (em->start > logical || em->start + em->len < logical) {
5148                 btrfs_crit(fs_info, "found a bad mapping, wanted %Lu, "
5149                            "found %Lu-%Lu", logical, em->start,
5150                            em->start + em->len);
5151                 free_extent_map(em);
5152                 return -EINVAL;
5153         }
5154
5155         map = (struct map_lookup *)em->bdev;
5156         offset = logical - em->start;
5157
5158         stripe_len = map->stripe_len;
5159         stripe_nr = offset;
5160         /*
5161          * stripe_nr counts the total number of stripes we have to stride
5162          * to get to this block
5163          */
5164         stripe_nr = div64_u64(stripe_nr, stripe_len);
5165
5166         stripe_offset = stripe_nr * stripe_len;
5167         BUG_ON(offset < stripe_offset);
5168
5169         /* stripe_offset is the offset of this block in its stripe*/
5170         stripe_offset = offset - stripe_offset;
5171
5172         /* if we're here for raid56, we need to know the stripe aligned start */
5173         if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5174                 unsigned long full_stripe_len = stripe_len * nr_data_stripes(map);
5175                 raid56_full_stripe_start = offset;
5176
5177                 /* allow a write of a full stripe, but make sure we don't
5178                  * allow straddling of stripes
5179                  */
5180                 raid56_full_stripe_start = div64_u64(raid56_full_stripe_start,
5181                                 full_stripe_len);
5182                 raid56_full_stripe_start *= full_stripe_len;
5183         }
5184
5185         if (rw & REQ_DISCARD) {
5186                 /* we don't discard raid56 yet */
5187                 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5188                         ret = -EOPNOTSUPP;
5189                         goto out;
5190                 }
5191                 *length = min_t(u64, em->len - offset, *length);
5192         } else if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
5193                 u64 max_len;
5194                 /* For writes to RAID[56], allow a full stripeset across all disks.
5195                    For other RAID types and for RAID[56] reads, just allow a single
5196                    stripe (on a single disk). */
5197                 if ((map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
5198                     (rw & REQ_WRITE)) {
5199                         max_len = stripe_len * nr_data_stripes(map) -
5200                                 (offset - raid56_full_stripe_start);
5201                 } else {
5202                         /* we limit the length of each bio to what fits in a stripe */
5203                         max_len = stripe_len - stripe_offset;
5204                 }
5205                 *length = min_t(u64, em->len - offset, max_len);
5206         } else {
5207                 *length = em->len - offset;
5208         }
5209
5210         /* This is for when we're called from btrfs_merge_bio_hook() and all
5211            it cares about is the length */
5212         if (!bbio_ret)
5213                 goto out;
5214
5215         btrfs_dev_replace_lock(dev_replace);
5216         dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
5217         if (!dev_replace_is_ongoing)
5218                 btrfs_dev_replace_unlock(dev_replace);
5219
5220         if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
5221             !(rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) &&
5222             dev_replace->tgtdev != NULL) {
5223                 /*
5224                  * in dev-replace case, for repair case (that's the only
5225                  * case where the mirror is selected explicitly when
5226                  * calling btrfs_map_block), blocks left of the left cursor
5227                  * can also be read from the target drive.
5228                  * For REQ_GET_READ_MIRRORS, the target drive is added as
5229                  * the last one to the array of stripes. For READ, it also
5230                  * needs to be supported using the same mirror number.
5231                  * If the requested block is not left of the left cursor,
5232                  * EIO is returned. This can happen because btrfs_num_copies()
5233                  * returns one more in the dev-replace case.
5234                  */
5235                 u64 tmp_length = *length;
5236                 struct btrfs_bio *tmp_bbio = NULL;
5237                 int tmp_num_stripes;
5238                 u64 srcdev_devid = dev_replace->srcdev->devid;
5239                 int index_srcdev = 0;
5240                 int found = 0;
5241                 u64 physical_of_found = 0;
5242
5243                 ret = __btrfs_map_block(fs_info, REQ_GET_READ_MIRRORS,
5244                              logical, &tmp_length, &tmp_bbio, 0, 0);
5245                 if (ret) {
5246                         WARN_ON(tmp_bbio != NULL);
5247                         goto out;
5248                 }
5249
5250                 tmp_num_stripes = tmp_bbio->num_stripes;
5251                 if (mirror_num > tmp_num_stripes) {
5252                         /*
5253                          * REQ_GET_READ_MIRRORS does not contain this
5254                          * mirror, that means that the requested area
5255                          * is not left of the left cursor
5256                          */
5257                         ret = -EIO;
5258                         btrfs_put_bbio(tmp_bbio);
5259                         goto out;
5260                 }
5261
5262                 /*
5263                  * process the rest of the function using the mirror_num
5264                  * of the source drive. Therefore look it up first.
5265                  * At the end, patch the device pointer to the one of the
5266                  * target drive.
5267                  */
5268                 for (i = 0; i < tmp_num_stripes; i++) {
5269                         if (tmp_bbio->stripes[i].dev->devid == srcdev_devid) {
5270                                 /*
5271                                  * In case of DUP, in order to keep it
5272                                  * simple, only add the mirror with the
5273                                  * lowest physical address
5274                                  */
5275                                 if (found &&
5276                                     physical_of_found <=
5277                                      tmp_bbio->stripes[i].physical)
5278                                         continue;
5279                                 index_srcdev = i;
5280                                 found = 1;
5281                                 physical_of_found =
5282                                         tmp_bbio->stripes[i].physical;
5283                         }
5284                 }
5285
5286                 if (found) {
5287                         mirror_num = index_srcdev + 1;
5288                         patch_the_first_stripe_for_dev_replace = 1;
5289                         physical_to_patch_in_first_stripe = physical_of_found;
5290                 } else {
5291                         WARN_ON(1);
5292                         ret = -EIO;
5293                         btrfs_put_bbio(tmp_bbio);
5294                         goto out;
5295                 }
5296
5297                 btrfs_put_bbio(tmp_bbio);
5298         } else if (mirror_num > map->num_stripes) {
5299                 mirror_num = 0;
5300         }
5301
5302         num_stripes = 1;
5303         stripe_index = 0;
5304         stripe_nr_orig = stripe_nr;
5305         stripe_nr_end = ALIGN(offset + *length, map->stripe_len);
5306         stripe_nr_end = div_u64(stripe_nr_end, map->stripe_len);
5307         stripe_end_offset = stripe_nr_end * map->stripe_len -
5308                             (offset + *length);
5309
5310         if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5311                 if (rw & REQ_DISCARD)
5312                         num_stripes = min_t(u64, map->num_stripes,
5313                                             stripe_nr_end - stripe_nr_orig);
5314                 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5315                                 &stripe_index);
5316                 if (!(rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)))
5317                         mirror_num = 1;
5318         } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
5319                 if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS))
5320                         num_stripes = map->num_stripes;
5321                 else if (mirror_num)
5322                         stripe_index = mirror_num - 1;
5323                 else {
5324                         stripe_index = find_live_mirror(fs_info, map, 0,
5325                                             map->num_stripes,
5326                                             current->pid % map->num_stripes,
5327                                             dev_replace_is_ongoing);
5328                         mirror_num = stripe_index + 1;
5329                 }
5330
5331         } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
5332                 if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) {
5333                         num_stripes = map->num_stripes;
5334                 } else if (mirror_num) {
5335                         stripe_index = mirror_num - 1;
5336                 } else {
5337                         mirror_num = 1;
5338                 }
5339
5340         } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5341                 u32 factor = map->num_stripes / map->sub_stripes;
5342
5343                 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
5344                 stripe_index *= map->sub_stripes;
5345
5346                 if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS))
5347                         num_stripes = map->sub_stripes;
5348                 else if (rw & REQ_DISCARD)
5349                         num_stripes = min_t(u64, map->sub_stripes *
5350                                             (stripe_nr_end - stripe_nr_orig),
5351                                             map->num_stripes);
5352                 else if (mirror_num)
5353                         stripe_index += mirror_num - 1;
5354                 else {
5355                         int old_stripe_index = stripe_index;
5356                         stripe_index = find_live_mirror(fs_info, map,
5357                                               stripe_index,
5358                                               map->sub_stripes, stripe_index +
5359                                               current->pid % map->sub_stripes,
5360                                               dev_replace_is_ongoing);
5361                         mirror_num = stripe_index - old_stripe_index + 1;
5362                 }
5363
5364         } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5365                 if (need_raid_map &&
5366                     ((rw & (REQ_WRITE | REQ_GET_READ_MIRRORS)) ||
5367                      mirror_num > 1)) {
5368                         /* push stripe_nr back to the start of the full stripe */
5369                         stripe_nr = div_u64(raid56_full_stripe_start,
5370                                         stripe_len * nr_data_stripes(map));
5371
5372                         /* RAID[56] write or recovery. Return all stripes */
5373                         num_stripes = map->num_stripes;
5374                         max_errors = nr_parity_stripes(map);
5375
5376                         *length = map->stripe_len;
5377                         stripe_index = 0;
5378                         stripe_offset = 0;
5379                 } else {
5380                         /*
5381                          * Mirror #0 or #1 means the original data block.
5382                          * Mirror #2 is RAID5 parity block.
5383                          * Mirror #3 is RAID6 Q block.
5384                          */
5385                         stripe_nr = div_u64_rem(stripe_nr,
5386                                         nr_data_stripes(map), &stripe_index);
5387                         if (mirror_num > 1)
5388                                 stripe_index = nr_data_stripes(map) +
5389                                                 mirror_num - 2;
5390
5391                         /* We distribute the parity blocks across stripes */
5392                         div_u64_rem(stripe_nr + stripe_index, map->num_stripes,
5393                                         &stripe_index);
5394                         if (!(rw & (REQ_WRITE | REQ_DISCARD |
5395                                     REQ_GET_READ_MIRRORS)) && mirror_num <= 1)
5396                                 mirror_num = 1;
5397                 }
5398         } else {
5399                 /*
5400                  * after this, stripe_nr is the number of stripes on this
5401                  * device we have to walk to find the data, and stripe_index is
5402                  * the number of our device in the stripe array
5403                  */
5404                 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5405                                 &stripe_index);
5406                 mirror_num = stripe_index + 1;
5407         }
5408         BUG_ON(stripe_index >= map->num_stripes);
5409
5410         num_alloc_stripes = num_stripes;
5411         if (dev_replace_is_ongoing) {
5412                 if (rw & (REQ_WRITE | REQ_DISCARD))
5413                         num_alloc_stripes <<= 1;
5414                 if (rw & REQ_GET_READ_MIRRORS)
5415                         num_alloc_stripes++;
5416                 tgtdev_indexes = num_stripes;
5417         }
5418
5419         bbio = alloc_btrfs_bio(num_alloc_stripes, tgtdev_indexes);
5420         if (!bbio) {
5421                 ret = -ENOMEM;
5422                 goto out;
5423         }
5424         if (dev_replace_is_ongoing)
5425                 bbio->tgtdev_map = (int *)(bbio->stripes + num_alloc_stripes);
5426
5427         /* build raid_map */
5428         if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK &&
5429             need_raid_map && ((rw & (REQ_WRITE | REQ_GET_READ_MIRRORS)) ||
5430             mirror_num > 1)) {
5431                 u64 tmp;
5432                 unsigned rot;
5433
5434                 bbio->raid_map = (u64 *)((void *)bbio->stripes +
5435                                  sizeof(struct btrfs_bio_stripe) *
5436                                  num_alloc_stripes +
5437                                  sizeof(int) * tgtdev_indexes);
5438
5439                 /* Work out the disk rotation on this stripe-set */
5440                 div_u64_rem(stripe_nr, num_stripes, &rot);
5441
5442                 /* Fill in the logical address of each stripe */
5443                 tmp = stripe_nr * nr_data_stripes(map);
5444                 for (i = 0; i < nr_data_stripes(map); i++)
5445                         bbio->raid_map[(i+rot) % num_stripes] =
5446                                 em->start + (tmp + i) * map->stripe_len;
5447
5448                 bbio->raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE;
5449                 if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5450                         bbio->raid_map[(i+rot+1) % num_stripes] =
5451                                 RAID6_Q_STRIPE;
5452         }
5453
5454         if (rw & REQ_DISCARD) {
5455                 u32 factor = 0;
5456                 u32 sub_stripes = 0;
5457                 u64 stripes_per_dev = 0;
5458                 u32 remaining_stripes = 0;
5459                 u32 last_stripe = 0;
5460
5461                 if (map->type &
5462                     (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID10)) {
5463                         if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5464                                 sub_stripes = 1;
5465                         else
5466                                 sub_stripes = map->sub_stripes;
5467
5468                         factor = map->num_stripes / sub_stripes;
5469                         stripes_per_dev = div_u64_rem(stripe_nr_end -
5470                                                       stripe_nr_orig,
5471                                                       factor,
5472                                                       &remaining_stripes);
5473                         div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
5474                         last_stripe *= sub_stripes;
5475                 }
5476
5477                 for (i = 0; i < num_stripes; i++) {
5478                         bbio->stripes[i].physical =
5479                                 map->stripes[stripe_index].physical +
5480                                 stripe_offset + stripe_nr * map->stripe_len;
5481                         bbio->stripes[i].dev = map->stripes[stripe_index].dev;
5482
5483                         if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5484                                          BTRFS_BLOCK_GROUP_RAID10)) {
5485                                 bbio->stripes[i].length = stripes_per_dev *
5486                                                           map->stripe_len;
5487
5488                                 if (i / sub_stripes < remaining_stripes)
5489                                         bbio->stripes[i].length +=
5490                                                 map->stripe_len;
5491
5492                                 /*
5493                                  * Special for the first stripe and
5494                                  * the last stripe:
5495                                  *
5496                                  * |-------|...|-------|
5497                                  *     |----------|
5498                                  *    off     end_off
5499                                  */
5500                                 if (i < sub_stripes)
5501                                         bbio->stripes[i].length -=
5502                                                 stripe_offset;
5503
5504                                 if (stripe_index >= last_stripe &&
5505                                     stripe_index <= (last_stripe +
5506                                                      sub_stripes - 1))
5507                                         bbio->stripes[i].length -=
5508                                                 stripe_end_offset;
5509
5510                                 if (i == sub_stripes - 1)
5511                                         stripe_offset = 0;
5512                         } else
5513                                 bbio->stripes[i].length = *length;
5514
5515                         stripe_index++;
5516                         if (stripe_index == map->num_stripes) {
5517                                 /* This could only happen for RAID0/10 */
5518                                 stripe_index = 0;
5519                                 stripe_nr++;
5520                         }
5521                 }
5522         } else {
5523                 for (i = 0; i < num_stripes; i++) {
5524                         bbio->stripes[i].physical =
5525                                 map->stripes[stripe_index].physical +
5526                                 stripe_offset +
5527                                 stripe_nr * map->stripe_len;
5528                         bbio->stripes[i].dev =
5529                                 map->stripes[stripe_index].dev;
5530                         stripe_index++;
5531                 }
5532         }
5533
5534         if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS))
5535                 max_errors = btrfs_chunk_max_errors(map);
5536
5537         if (bbio->raid_map)
5538                 sort_parity_stripes(bbio, num_stripes);
5539
5540         tgtdev_indexes = 0;
5541         if (dev_replace_is_ongoing && (rw & (REQ_WRITE | REQ_DISCARD)) &&
5542             dev_replace->tgtdev != NULL) {
5543                 int index_where_to_add;
5544                 u64 srcdev_devid = dev_replace->srcdev->devid;
5545
5546                 /*
5547                  * duplicate the write operations while the dev replace
5548                  * procedure is running. Since the copying of the old disk
5549                  * to the new disk takes place at run time while the
5550                  * filesystem is mounted writable, the regular write
5551                  * operations to the old disk have to be duplicated to go
5552                  * to the new disk as well.
5553                  * Note that device->missing is handled by the caller, and
5554                  * that the write to the old disk is already set up in the
5555                  * stripes array.
5556                  */
5557                 index_where_to_add = num_stripes;
5558                 for (i = 0; i < num_stripes; i++) {
5559                         if (bbio->stripes[i].dev->devid == srcdev_devid) {
5560                                 /* write to new disk, too */
5561                                 struct btrfs_bio_stripe *new =
5562                                         bbio->stripes + index_where_to_add;
5563                                 struct btrfs_bio_stripe *old =
5564                                         bbio->stripes + i;
5565
5566                                 new->physical = old->physical;
5567                                 new->length = old->length;
5568                                 new->dev = dev_replace->tgtdev;
5569                                 bbio->tgtdev_map[i] = index_where_to_add;
5570                                 index_where_to_add++;
5571                                 max_errors++;
5572                                 tgtdev_indexes++;
5573                         }
5574                 }
5575                 num_stripes = index_where_to_add;
5576         } else if (dev_replace_is_ongoing && (rw & REQ_GET_READ_MIRRORS) &&
5577                    dev_replace->tgtdev != NULL) {
5578                 u64 srcdev_devid = dev_replace->srcdev->devid;
5579                 int index_srcdev = 0;
5580                 int found = 0;
5581                 u64 physical_of_found = 0;
5582
5583                 /*
5584                  * During the dev-replace procedure, the target drive can
5585                  * also be used to read data in case it is needed to repair
5586                  * a corrupt block elsewhere. This is possible if the
5587                  * requested area is left of the left cursor. In this area,
5588                  * the target drive is a full copy of the source drive.
5589                  */
5590                 for (i = 0; i < num_stripes; i++) {
5591                         if (bbio->stripes[i].dev->devid == srcdev_devid) {
5592                                 /*
5593                                  * In case of DUP, in order to keep it
5594                                  * simple, only add the mirror with the
5595                                  * lowest physical address
5596                                  */
5597                                 if (found &&
5598                                     physical_of_found <=
5599                                      bbio->stripes[i].physical)
5600                                         continue;
5601                                 index_srcdev = i;
5602                                 found = 1;
5603                                 physical_of_found = bbio->stripes[i].physical;
5604                         }
5605                 }
5606                 if (found) {
5607                         if (physical_of_found + map->stripe_len <=
5608                             dev_replace->cursor_left) {
5609                                 struct btrfs_bio_stripe *tgtdev_stripe =
5610                                         bbio->stripes + num_stripes;
5611
5612                                 tgtdev_stripe->physical = physical_of_found;
5613                                 tgtdev_stripe->length =
5614                                         bbio->stripes[index_srcdev].length;
5615                                 tgtdev_stripe->dev = dev_replace->tgtdev;
5616                                 bbio->tgtdev_map[index_srcdev] = num_stripes;
5617
5618                                 tgtdev_indexes++;
5619                                 num_stripes++;
5620                         }
5621                 }
5622         }
5623
5624         *bbio_ret = bbio;
5625         bbio->map_type = map->type;
5626         bbio->num_stripes = num_stripes;
5627         bbio->max_errors = max_errors;
5628         bbio->mirror_num = mirror_num;
5629         bbio->num_tgtdevs = tgtdev_indexes;
5630
5631         /*
5632          * this is the case that REQ_READ && dev_replace_is_ongoing &&
5633          * mirror_num == num_stripes + 1 && dev_replace target drive is
5634          * available as a mirror
5635          */
5636         if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) {
5637                 WARN_ON(num_stripes > 1);
5638                 bbio->stripes[0].dev = dev_replace->tgtdev;
5639                 bbio->stripes[0].physical = physical_to_patch_in_first_stripe;
5640                 bbio->mirror_num = map->num_stripes + 1;
5641         }
5642 out:
5643         if (dev_replace_is_ongoing)
5644                 btrfs_dev_replace_unlock(dev_replace);
5645         free_extent_map(em);
5646         return ret;
5647 }
5648
5649 int btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
5650                       u64 logical, u64 *length,
5651                       struct btrfs_bio **bbio_ret, int mirror_num)
5652 {
5653         return __btrfs_map_block(fs_info, rw, logical, length, bbio_ret,
5654                                  mirror_num, 0);
5655 }
5656
5657 /* For Scrub/replace */
5658 int btrfs_map_sblock(struct btrfs_fs_info *fs_info, int rw,
5659                      u64 logical, u64 *length,
5660                      struct btrfs_bio **bbio_ret, int mirror_num,
5661                      int need_raid_map)
5662 {
5663         return __btrfs_map_block(fs_info, rw, logical, length, bbio_ret,
5664                                  mirror_num, need_raid_map);
5665 }
5666
5667 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
5668                      u64 chunk_start, u64 physical, u64 devid,
5669                      u64 **logical, int *naddrs, int *stripe_len)
5670 {
5671         struct extent_map_tree *em_tree = &map_tree->map_tree;
5672         struct extent_map *em;
5673         struct map_lookup *map;
5674         u64 *buf;
5675         u64 bytenr;
5676         u64 length;
5677         u64 stripe_nr;
5678         u64 rmap_len;
5679         int i, j, nr = 0;
5680
5681         read_lock(&em_tree->lock);
5682         em = lookup_extent_mapping(em_tree, chunk_start, 1);
5683         read_unlock(&em_tree->lock);
5684
5685         if (!em) {
5686                 printk(KERN_ERR "BTRFS: couldn't find em for chunk %Lu\n",
5687                        chunk_start);
5688                 return -EIO;
5689         }
5690
5691         if (em->start != chunk_start) {
5692                 printk(KERN_ERR "BTRFS: bad chunk start, em=%Lu, wanted=%Lu\n",
5693                        em->start, chunk_start);
5694                 free_extent_map(em);
5695                 return -EIO;
5696         }
5697         map = (struct map_lookup *)em->bdev;
5698
5699         length = em->len;
5700         rmap_len = map->stripe_len;
5701
5702         if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5703                 length = div_u64(length, map->num_stripes / map->sub_stripes);
5704         else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5705                 length = div_u64(length, map->num_stripes);
5706         else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5707                 length = div_u64(length, nr_data_stripes(map));
5708                 rmap_len = map->stripe_len * nr_data_stripes(map);
5709         }
5710
5711         buf = kcalloc(map->num_stripes, sizeof(u64), GFP_NOFS);
5712         BUG_ON(!buf); /* -ENOMEM */
5713
5714         for (i = 0; i < map->num_stripes; i++) {
5715                 if (devid && map->stripes[i].dev->devid != devid)
5716                         continue;
5717                 if (map->stripes[i].physical > physical ||
5718                     map->stripes[i].physical + length <= physical)
5719                         continue;
5720
5721                 stripe_nr = physical - map->stripes[i].physical;
5722                 stripe_nr = div_u64(stripe_nr, map->stripe_len);
5723
5724                 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5725                         stripe_nr = stripe_nr * map->num_stripes + i;
5726                         stripe_nr = div_u64(stripe_nr, map->sub_stripes);
5727                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5728                         stripe_nr = stripe_nr * map->num_stripes + i;
5729                 } /* else if RAID[56], multiply by nr_data_stripes().
5730                    * Alternatively, just use rmap_len below instead of
5731                    * map->stripe_len */
5732
5733                 bytenr = chunk_start + stripe_nr * rmap_len;
5734                 WARN_ON(nr >= map->num_stripes);
5735                 for (j = 0; j < nr; j++) {
5736                         if (buf[j] == bytenr)
5737                                 break;
5738                 }
5739                 if (j == nr) {
5740                         WARN_ON(nr >= map->num_stripes);
5741                         buf[nr++] = bytenr;
5742                 }
5743         }
5744
5745         *logical = buf;
5746         *naddrs = nr;
5747         *stripe_len = rmap_len;
5748
5749         free_extent_map(em);
5750         return 0;
5751 }
5752
5753 static inline void btrfs_end_bbio(struct btrfs_bio *bbio, struct bio *bio)
5754 {
5755         bio->bi_private = bbio->private;
5756         bio->bi_end_io = bbio->end_io;
5757         bio_endio(bio);
5758
5759         btrfs_put_bbio(bbio);
5760 }
5761
5762 static void btrfs_end_bio(struct bio *bio)
5763 {
5764         struct btrfs_bio *bbio = bio->bi_private;
5765         int is_orig_bio = 0;
5766
5767         if (bio->bi_error) {
5768                 atomic_inc(&bbio->error);
5769                 if (bio->bi_error == -EIO || bio->bi_error == -EREMOTEIO) {
5770                         unsigned int stripe_index =
5771                                 btrfs_io_bio(bio)->stripe_index;
5772                         struct btrfs_device *dev;
5773
5774                         BUG_ON(stripe_index >= bbio->num_stripes);
5775                         dev = bbio->stripes[stripe_index].dev;
5776                         if (dev->bdev) {
5777                                 if (bio->bi_rw & WRITE)
5778                                         btrfs_dev_stat_inc(dev,
5779                                                 BTRFS_DEV_STAT_WRITE_ERRS);
5780                                 else
5781                                         btrfs_dev_stat_inc(dev,
5782                                                 BTRFS_DEV_STAT_READ_ERRS);
5783                                 if ((bio->bi_rw & WRITE_FLUSH) == WRITE_FLUSH)
5784                                         btrfs_dev_stat_inc(dev,
5785                                                 BTRFS_DEV_STAT_FLUSH_ERRS);
5786                                 btrfs_dev_stat_print_on_error(dev);
5787                         }
5788                 }
5789         }
5790
5791         if (bio == bbio->orig_bio)
5792                 is_orig_bio = 1;
5793
5794         btrfs_bio_counter_dec(bbio->fs_info);
5795
5796         if (atomic_dec_and_test(&bbio->stripes_pending)) {
5797                 if (!is_orig_bio) {
5798                         bio_put(bio);
5799                         bio = bbio->orig_bio;
5800                 }
5801
5802                 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
5803                 /* only send an error to the higher layers if it is
5804                  * beyond the tolerance of the btrfs bio
5805                  */
5806                 if (atomic_read(&bbio->error) > bbio->max_errors) {
5807                         bio->bi_error = -EIO;
5808                 } else {
5809                         /*
5810                          * this bio is actually up to date, we didn't
5811                          * go over the max number of errors
5812                          */
5813                         bio->bi_error = 0;
5814                 }
5815
5816                 btrfs_end_bbio(bbio, bio);
5817         } else if (!is_orig_bio) {
5818                 bio_put(bio);
5819         }
5820 }
5821
5822 /*
5823  * see run_scheduled_bios for a description of why bios are collected for
5824  * async submit.
5825  *
5826  * This will add one bio to the pending list for a device and make sure
5827  * the work struct is scheduled.
5828  */
5829 static noinline void btrfs_schedule_bio(struct btrfs_root *root,
5830                                         struct btrfs_device *device,
5831                                         int rw, struct bio *bio)
5832 {
5833         int should_queue = 1;
5834         struct btrfs_pending_bios *pending_bios;
5835
5836         if (device->missing || !device->bdev) {
5837                 bio_io_error(bio);
5838                 return;
5839         }
5840
5841         /* don't bother with additional async steps for reads, right now */
5842         if (!(rw & REQ_WRITE)) {
5843                 bio_get(bio);
5844                 btrfsic_submit_bio(rw, bio);
5845                 bio_put(bio);
5846                 return;
5847         }
5848
5849         /*
5850          * nr_async_bios allows us to reliably return congestion to the
5851          * higher layers.  Otherwise, the async bio makes it appear we have
5852          * made progress against dirty pages when we've really just put it
5853          * on a queue for later
5854          */
5855         atomic_inc(&root->fs_info->nr_async_bios);
5856         WARN_ON(bio->bi_next);
5857         bio->bi_next = NULL;
5858         bio->bi_rw |= rw;
5859
5860         spin_lock(&device->io_lock);
5861         if (bio->bi_rw & REQ_SYNC)
5862                 pending_bios = &device->pending_sync_bios;
5863         else
5864                 pending_bios = &device->pending_bios;
5865
5866         if (pending_bios->tail)
5867                 pending_bios->tail->bi_next = bio;
5868
5869         pending_bios->tail = bio;
5870         if (!pending_bios->head)
5871                 pending_bios->head = bio;
5872         if (device->running_pending)
5873                 should_queue = 0;
5874
5875         spin_unlock(&device->io_lock);
5876
5877         if (should_queue)
5878                 btrfs_queue_work(root->fs_info->submit_workers,
5879                                  &device->work);
5880 }
5881
5882 static void submit_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio,
5883                               struct bio *bio, u64 physical, int dev_nr,
5884                               int rw, int async)
5885 {
5886         struct btrfs_device *dev = bbio->stripes[dev_nr].dev;
5887
5888         bio->bi_private = bbio;
5889         btrfs_io_bio(bio)->stripe_index = dev_nr;
5890         bio->bi_end_io = btrfs_end_bio;
5891         bio->bi_iter.bi_sector = physical >> 9;
5892 #ifdef DEBUG
5893         {
5894                 struct rcu_string *name;
5895
5896                 rcu_read_lock();
5897                 name = rcu_dereference(dev->name);
5898                 pr_debug("btrfs_map_bio: rw %d, sector=%llu, dev=%lu "
5899                          "(%s id %llu), size=%u\n", rw,
5900                          (u64)bio->bi_iter.bi_sector, (u_long)dev->bdev->bd_dev,
5901                          name->str, dev->devid, bio->bi_iter.bi_size);
5902                 rcu_read_unlock();
5903         }
5904 #endif
5905         bio->bi_bdev = dev->bdev;
5906
5907         btrfs_bio_counter_inc_noblocked(root->fs_info);
5908
5909         if (async)
5910                 btrfs_schedule_bio(root, dev, rw, bio);
5911         else
5912                 btrfsic_submit_bio(rw, bio);
5913 }
5914
5915 static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical)
5916 {
5917         atomic_inc(&bbio->error);
5918         if (atomic_dec_and_test(&bbio->stripes_pending)) {
5919                 /* Shoud be the original bio. */
5920                 WARN_ON(bio != bbio->orig_bio);
5921
5922                 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
5923                 bio->bi_iter.bi_sector = logical >> 9;
5924                 bio->bi_error = -EIO;
5925                 btrfs_end_bbio(bbio, bio);
5926         }
5927 }
5928
5929 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
5930                   int mirror_num, int async_submit)
5931 {
5932         struct btrfs_device *dev;
5933         struct bio *first_bio = bio;
5934         u64 logical = (u64)bio->bi_iter.bi_sector << 9;
5935         u64 length = 0;
5936         u64 map_length;
5937         int ret;
5938         int dev_nr;
5939         int total_devs;
5940         struct btrfs_bio *bbio = NULL;
5941
5942         length = bio->bi_iter.bi_size;
5943         map_length = length;
5944
5945         btrfs_bio_counter_inc_blocked(root->fs_info);
5946         ret = __btrfs_map_block(root->fs_info, rw, logical, &map_length, &bbio,
5947                               mirror_num, 1);
5948         if (ret) {
5949                 btrfs_bio_counter_dec(root->fs_info);
5950                 return ret;
5951         }
5952
5953         total_devs = bbio->num_stripes;
5954         bbio->orig_bio = first_bio;
5955         bbio->private = first_bio->bi_private;
5956         bbio->end_io = first_bio->bi_end_io;
5957         bbio->fs_info = root->fs_info;
5958         atomic_set(&bbio->stripes_pending, bbio->num_stripes);
5959
5960         if (bbio->raid_map) {
5961                 /* In this case, map_length has been set to the length of
5962                    a single stripe; not the whole write */
5963                 if (rw & WRITE) {
5964                         ret = raid56_parity_write(root, bio, bbio, map_length);
5965                 } else {
5966                         ret = raid56_parity_recover(root, bio, bbio, map_length,
5967                                                     mirror_num, 1);
5968                 }
5969
5970                 btrfs_bio_counter_dec(root->fs_info);
5971                 return ret;
5972         }
5973
5974         if (map_length < length) {
5975                 btrfs_crit(root->fs_info, "mapping failed logical %llu bio len %llu len %llu",
5976                         logical, length, map_length);
5977                 BUG();
5978         }
5979
5980         for (dev_nr = 0; dev_nr < total_devs; dev_nr++) {
5981                 dev = bbio->stripes[dev_nr].dev;
5982                 if (!dev || !dev->bdev || (rw & WRITE && !dev->writeable)) {
5983                         bbio_error(bbio, first_bio, logical);
5984                         continue;
5985                 }
5986
5987                 if (dev_nr < total_devs - 1) {
5988                         bio = btrfs_bio_clone(first_bio, GFP_NOFS);
5989                         BUG_ON(!bio); /* -ENOMEM */
5990                 } else
5991                         bio = first_bio;
5992
5993                 submit_stripe_bio(root, bbio, bio,
5994                                   bbio->stripes[dev_nr].physical, dev_nr, rw,
5995                                   async_submit);
5996         }
5997         btrfs_bio_counter_dec(root->fs_info);
5998         return 0;
5999 }
6000
6001 struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid,
6002                                        u8 *uuid, u8 *fsid)
6003 {
6004         struct btrfs_device *device;
6005         struct btrfs_fs_devices *cur_devices;
6006
6007         cur_devices = fs_info->fs_devices;
6008         while (cur_devices) {
6009                 if (!fsid ||
6010                     !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
6011                         device = __find_device(&cur_devices->devices,
6012                                                devid, uuid);
6013                         if (device)
6014                                 return device;
6015                 }
6016                 cur_devices = cur_devices->seed;
6017         }
6018         return NULL;
6019 }
6020
6021 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
6022                                             struct btrfs_fs_devices *fs_devices,
6023                                             u64 devid, u8 *dev_uuid)
6024 {
6025         struct btrfs_device *device;
6026
6027         device = btrfs_alloc_device(NULL, &devid, dev_uuid);
6028         if (IS_ERR(device))
6029                 return NULL;
6030
6031         list_add(&device->dev_list, &fs_devices->devices);
6032         device->fs_devices = fs_devices;
6033         fs_devices->num_devices++;
6034
6035         device->missing = 1;
6036         fs_devices->missing_devices++;
6037
6038         return device;
6039 }
6040
6041 /**
6042  * btrfs_alloc_device - allocate struct btrfs_device
6043  * @fs_info:    used only for generating a new devid, can be NULL if
6044  *              devid is provided (i.e. @devid != NULL).
6045  * @devid:      a pointer to devid for this device.  If NULL a new devid
6046  *              is generated.
6047  * @uuid:       a pointer to UUID for this device.  If NULL a new UUID
6048  *              is generated.
6049  *
6050  * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
6051  * on error.  Returned struct is not linked onto any lists and can be
6052  * destroyed with kfree() right away.
6053  */
6054 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
6055                                         const u64 *devid,
6056                                         const u8 *uuid)
6057 {
6058         struct btrfs_device *dev;
6059         u64 tmp;
6060
6061         if (WARN_ON(!devid && !fs_info))
6062                 return ERR_PTR(-EINVAL);
6063
6064         dev = __alloc_device();
6065         if (IS_ERR(dev))
6066                 return dev;
6067
6068         if (devid)
6069                 tmp = *devid;
6070         else {
6071                 int ret;
6072
6073                 ret = find_next_devid(fs_info, &tmp);
6074                 if (ret) {
6075                         kfree(dev);
6076                         return ERR_PTR(ret);
6077                 }
6078         }
6079         dev->devid = tmp;
6080
6081         if (uuid)
6082                 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE);
6083         else
6084                 generate_random_uuid(dev->uuid);
6085
6086         btrfs_init_work(&dev->work, btrfs_submit_helper,
6087                         pending_bios_fn, NULL, NULL);
6088
6089         return dev;
6090 }
6091
6092 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
6093                           struct extent_buffer *leaf,
6094                           struct btrfs_chunk *chunk)
6095 {
6096         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
6097         struct map_lookup *map;
6098         struct extent_map *em;
6099         u64 logical;
6100         u64 length;
6101         u64 devid;
6102         u8 uuid[BTRFS_UUID_SIZE];
6103         int num_stripes;
6104         int ret;
6105         int i;
6106
6107         logical = key->offset;
6108         length = btrfs_chunk_length(leaf, chunk);
6109
6110         read_lock(&map_tree->map_tree.lock);
6111         em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
6112         read_unlock(&map_tree->map_tree.lock);
6113
6114         /* already mapped? */
6115         if (em && em->start <= logical && em->start + em->len > logical) {
6116                 free_extent_map(em);
6117                 return 0;
6118         } else if (em) {
6119                 free_extent_map(em);
6120         }
6121
6122         em = alloc_extent_map();
6123         if (!em)
6124                 return -ENOMEM;
6125         num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6126         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
6127         if (!map) {
6128                 free_extent_map(em);
6129                 return -ENOMEM;
6130         }
6131
6132         set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
6133         em->bdev = (struct block_device *)map;
6134         em->start = logical;
6135         em->len = length;
6136         em->orig_start = 0;
6137         em->block_start = 0;
6138         em->block_len = em->len;
6139
6140         map->num_stripes = num_stripes;
6141         map->io_width = btrfs_chunk_io_width(leaf, chunk);
6142         map->io_align = btrfs_chunk_io_align(leaf, chunk);
6143         map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
6144         map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6145         map->type = btrfs_chunk_type(leaf, chunk);
6146         map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
6147         for (i = 0; i < num_stripes; i++) {
6148                 map->stripes[i].physical =
6149                         btrfs_stripe_offset_nr(leaf, chunk, i);
6150                 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
6151                 read_extent_buffer(leaf, uuid, (unsigned long)
6152                                    btrfs_stripe_dev_uuid_nr(chunk, i),
6153                                    BTRFS_UUID_SIZE);
6154                 map->stripes[i].dev = btrfs_find_device(root->fs_info, devid,
6155                                                         uuid, NULL);
6156                 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
6157                         free_extent_map(em);
6158                         return -EIO;
6159                 }
6160                 if (!map->stripes[i].dev) {
6161                         map->stripes[i].dev =
6162                                 add_missing_dev(root, root->fs_info->fs_devices,
6163                                                 devid, uuid);
6164                         if (!map->stripes[i].dev) {
6165                                 free_extent_map(em);
6166                                 return -EIO;
6167                         }
6168                         btrfs_warn(root->fs_info, "devid %llu uuid %pU is missing",
6169                                                 devid, uuid);
6170                 }
6171                 map->stripes[i].dev->in_fs_metadata = 1;
6172         }
6173
6174         write_lock(&map_tree->map_tree.lock);
6175         ret = add_extent_mapping(&map_tree->map_tree, em, 0);
6176         write_unlock(&map_tree->map_tree.lock);
6177         BUG_ON(ret); /* Tree corruption */
6178         free_extent_map(em);
6179
6180         return 0;
6181 }
6182
6183 static void fill_device_from_item(struct extent_buffer *leaf,
6184                                  struct btrfs_dev_item *dev_item,
6185                                  struct btrfs_device *device)
6186 {
6187         unsigned long ptr;
6188
6189         device->devid = btrfs_device_id(leaf, dev_item);
6190         device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
6191         device->total_bytes = device->disk_total_bytes;
6192         device->commit_total_bytes = device->disk_total_bytes;
6193         device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
6194         device->commit_bytes_used = device->bytes_used;
6195         device->type = btrfs_device_type(leaf, dev_item);
6196         device->io_align = btrfs_device_io_align(leaf, dev_item);
6197         device->io_width = btrfs_device_io_width(leaf, dev_item);
6198         device->sector_size = btrfs_device_sector_size(leaf, dev_item);
6199         WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
6200         device->is_tgtdev_for_dev_replace = 0;
6201
6202         ptr = btrfs_device_uuid(dev_item);
6203         read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
6204 }
6205
6206 static struct btrfs_fs_devices *open_seed_devices(struct btrfs_root *root,
6207                                                   u8 *fsid)
6208 {
6209         struct btrfs_fs_devices *fs_devices;
6210         int ret;
6211
6212         BUG_ON(!mutex_is_locked(&uuid_mutex));
6213
6214         fs_devices = root->fs_info->fs_devices->seed;
6215         while (fs_devices) {
6216                 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE))
6217                         return fs_devices;
6218
6219                 fs_devices = fs_devices->seed;
6220         }
6221
6222         fs_devices = find_fsid(fsid);
6223         if (!fs_devices) {
6224                 if (!btrfs_test_opt(root, DEGRADED))
6225                         return ERR_PTR(-ENOENT);
6226
6227                 fs_devices = alloc_fs_devices(fsid);
6228                 if (IS_ERR(fs_devices))
6229                         return fs_devices;
6230
6231                 fs_devices->seeding = 1;
6232                 fs_devices->opened = 1;
6233                 return fs_devices;
6234         }
6235
6236         fs_devices = clone_fs_devices(fs_devices);
6237         if (IS_ERR(fs_devices))
6238                 return fs_devices;
6239
6240         ret = __btrfs_open_devices(fs_devices, FMODE_READ,
6241                                    root->fs_info->bdev_holder);
6242         if (ret) {
6243                 free_fs_devices(fs_devices);
6244                 fs_devices = ERR_PTR(ret);
6245                 goto out;
6246         }
6247
6248         if (!fs_devices->seeding) {
6249                 __btrfs_close_devices(fs_devices);
6250                 free_fs_devices(fs_devices);
6251                 fs_devices = ERR_PTR(-EINVAL);
6252                 goto out;
6253         }
6254
6255         fs_devices->seed = root->fs_info->fs_devices->seed;
6256         root->fs_info->fs_devices->seed = fs_devices;
6257 out:
6258         return fs_devices;
6259 }
6260
6261 static int read_one_dev(struct btrfs_root *root,
6262                         struct extent_buffer *leaf,
6263                         struct btrfs_dev_item *dev_item)
6264 {
6265         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
6266         struct btrfs_device *device;
6267         u64 devid;
6268         int ret;
6269         u8 fs_uuid[BTRFS_UUID_SIZE];
6270         u8 dev_uuid[BTRFS_UUID_SIZE];
6271
6272         devid = btrfs_device_id(leaf, dev_item);
6273         read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
6274                            BTRFS_UUID_SIZE);
6275         read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
6276                            BTRFS_UUID_SIZE);
6277
6278         if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
6279                 fs_devices = open_seed_devices(root, fs_uuid);
6280                 if (IS_ERR(fs_devices))
6281                         return PTR_ERR(fs_devices);
6282         }
6283
6284         device = btrfs_find_device(root->fs_info, devid, dev_uuid, fs_uuid);
6285         if (!device) {
6286                 if (!btrfs_test_opt(root, DEGRADED))
6287                         return -EIO;
6288
6289                 device = add_missing_dev(root, fs_devices, devid, dev_uuid);
6290                 if (!device)
6291                         return -ENOMEM;
6292                 btrfs_warn(root->fs_info, "devid %llu uuid %pU missing",
6293                                 devid, dev_uuid);
6294         } else {
6295                 if (!device->bdev && !btrfs_test_opt(root, DEGRADED))
6296                         return -EIO;
6297
6298                 if(!device->bdev && !device->missing) {
6299                         /*
6300                          * this happens when a device that was properly setup
6301                          * in the device info lists suddenly goes bad.
6302                          * device->bdev is NULL, and so we have to set
6303                          * device->missing to one here
6304                          */
6305                         device->fs_devices->missing_devices++;
6306                         device->missing = 1;
6307                 }
6308
6309                 /* Move the device to its own fs_devices */
6310                 if (device->fs_devices != fs_devices) {
6311                         ASSERT(device->missing);
6312
6313                         list_move(&device->dev_list, &fs_devices->devices);
6314                         device->fs_devices->num_devices--;
6315                         fs_devices->num_devices++;
6316
6317                         device->fs_devices->missing_devices--;
6318                         fs_devices->missing_devices++;
6319
6320                         device->fs_devices = fs_devices;
6321                 }
6322         }
6323
6324         if (device->fs_devices != root->fs_info->fs_devices) {
6325                 BUG_ON(device->writeable);
6326                 if (device->generation !=
6327                     btrfs_device_generation(leaf, dev_item))
6328                         return -EINVAL;
6329         }
6330
6331         fill_device_from_item(leaf, dev_item, device);
6332         device->in_fs_metadata = 1;
6333         if (device->writeable && !device->is_tgtdev_for_dev_replace) {
6334                 device->fs_devices->total_rw_bytes += device->total_bytes;
6335                 spin_lock(&root->fs_info->free_chunk_lock);
6336                 root->fs_info->free_chunk_space += device->total_bytes -
6337                         device->bytes_used;
6338                 spin_unlock(&root->fs_info->free_chunk_lock);
6339         }
6340         ret = 0;
6341         return ret;
6342 }
6343
6344 int btrfs_read_sys_array(struct btrfs_root *root)
6345 {
6346         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
6347         struct extent_buffer *sb;
6348         struct btrfs_disk_key *disk_key;
6349         struct btrfs_chunk *chunk;
6350         u8 *array_ptr;
6351         unsigned long sb_array_offset;
6352         int ret = 0;
6353         u32 num_stripes;
6354         u32 array_size;
6355         u32 len = 0;
6356         u32 cur_offset;
6357         struct btrfs_key key;
6358
6359         ASSERT(BTRFS_SUPER_INFO_SIZE <= root->nodesize);
6360         /*
6361          * This will create extent buffer of nodesize, superblock size is
6362          * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will
6363          * overallocate but we can keep it as-is, only the first page is used.
6364          */
6365         sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET);
6366         if (!sb)
6367                 return -ENOMEM;
6368         btrfs_set_buffer_uptodate(sb);
6369         btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
6370         /*
6371          * The sb extent buffer is artifical and just used to read the system array.
6372          * btrfs_set_buffer_uptodate() call does not properly mark all it's
6373          * pages up-to-date when the page is larger: extent does not cover the
6374          * whole page and consequently check_page_uptodate does not find all
6375          * the page's extents up-to-date (the hole beyond sb),
6376          * write_extent_buffer then triggers a WARN_ON.
6377          *
6378          * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
6379          * but sb spans only this function. Add an explicit SetPageUptodate call
6380          * to silence the warning eg. on PowerPC 64.
6381          */
6382         if (PAGE_CACHE_SIZE > BTRFS_SUPER_INFO_SIZE)
6383                 SetPageUptodate(sb->pages[0]);
6384
6385         write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
6386         array_size = btrfs_super_sys_array_size(super_copy);
6387
6388         array_ptr = super_copy->sys_chunk_array;
6389         sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array);
6390         cur_offset = 0;
6391
6392         while (cur_offset < array_size) {
6393                 disk_key = (struct btrfs_disk_key *)array_ptr;
6394                 len = sizeof(*disk_key);
6395                 if (cur_offset + len > array_size)
6396                         goto out_short_read;
6397
6398                 btrfs_disk_key_to_cpu(&key, disk_key);
6399
6400                 array_ptr += len;
6401                 sb_array_offset += len;
6402                 cur_offset += len;
6403
6404                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
6405                         chunk = (struct btrfs_chunk *)sb_array_offset;
6406                         /*
6407                          * At least one btrfs_chunk with one stripe must be
6408                          * present, exact stripe count check comes afterwards
6409                          */
6410                         len = btrfs_chunk_item_size(1);
6411                         if (cur_offset + len > array_size)
6412                                 goto out_short_read;
6413
6414                         num_stripes = btrfs_chunk_num_stripes(sb, chunk);
6415                         len = btrfs_chunk_item_size(num_stripes);
6416                         if (cur_offset + len > array_size)
6417                                 goto out_short_read;
6418
6419                         ret = read_one_chunk(root, &key, sb, chunk);
6420                         if (ret)
6421                                 break;
6422                 } else {
6423                         ret = -EIO;
6424                         break;
6425                 }
6426                 array_ptr += len;
6427                 sb_array_offset += len;
6428                 cur_offset += len;
6429         }
6430         free_extent_buffer(sb);
6431         return ret;
6432
6433 out_short_read:
6434         printk(KERN_ERR "BTRFS: sys_array too short to read %u bytes at offset %u\n",
6435                         len, cur_offset);
6436         free_extent_buffer(sb);
6437         return -EIO;
6438 }
6439
6440 int btrfs_read_chunk_tree(struct btrfs_root *root)
6441 {
6442         struct btrfs_path *path;
6443         struct extent_buffer *leaf;
6444         struct btrfs_key key;
6445         struct btrfs_key found_key;
6446         int ret;
6447         int slot;
6448
6449         root = root->fs_info->chunk_root;
6450
6451         path = btrfs_alloc_path();
6452         if (!path)
6453                 return -ENOMEM;
6454
6455         mutex_lock(&uuid_mutex);
6456         lock_chunks(root);
6457
6458         /*
6459          * Read all device items, and then all the chunk items. All
6460          * device items are found before any chunk item (their object id
6461          * is smaller than the lowest possible object id for a chunk
6462          * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
6463          */
6464         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
6465         key.offset = 0;
6466         key.type = 0;
6467         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6468         if (ret < 0)
6469                 goto error;
6470         while (1) {
6471                 leaf = path->nodes[0];
6472                 slot = path->slots[0];
6473                 if (slot >= btrfs_header_nritems(leaf)) {
6474                         ret = btrfs_next_leaf(root, path);
6475                         if (ret == 0)
6476                                 continue;
6477                         if (ret < 0)
6478                                 goto error;
6479                         break;
6480                 }
6481                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6482                 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
6483                         struct btrfs_dev_item *dev_item;
6484                         dev_item = btrfs_item_ptr(leaf, slot,
6485                                                   struct btrfs_dev_item);
6486                         ret = read_one_dev(root, leaf, dev_item);
6487                         if (ret)
6488                                 goto error;
6489                 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
6490                         struct btrfs_chunk *chunk;
6491                         chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
6492                         ret = read_one_chunk(root, &found_key, leaf, chunk);
6493                         if (ret)
6494                                 goto error;
6495                 }
6496                 path->slots[0]++;
6497         }
6498         ret = 0;
6499 error:
6500         unlock_chunks(root);
6501         mutex_unlock(&uuid_mutex);
6502
6503         btrfs_free_path(path);
6504         return ret;
6505 }
6506
6507 void btrfs_init_devices_late(struct btrfs_fs_info *fs_info)
6508 {
6509         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6510         struct btrfs_device *device;
6511
6512         while (fs_devices) {
6513                 mutex_lock(&fs_devices->device_list_mutex);
6514                 list_for_each_entry(device, &fs_devices->devices, dev_list)
6515                         device->dev_root = fs_info->dev_root;
6516                 mutex_unlock(&fs_devices->device_list_mutex);
6517
6518                 fs_devices = fs_devices->seed;
6519         }
6520 }
6521
6522 static void __btrfs_reset_dev_stats(struct btrfs_device *dev)
6523 {
6524         int i;
6525
6526         for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6527                 btrfs_dev_stat_reset(dev, i);
6528 }
6529
6530 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
6531 {
6532         struct btrfs_key key;
6533         struct btrfs_key found_key;
6534         struct btrfs_root *dev_root = fs_info->dev_root;
6535         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6536         struct extent_buffer *eb;
6537         int slot;
6538         int ret = 0;
6539         struct btrfs_device *device;
6540         struct btrfs_path *path = NULL;
6541         int i;
6542
6543         path = btrfs_alloc_path();
6544         if (!path) {
6545                 ret = -ENOMEM;
6546                 goto out;
6547         }
6548
6549         mutex_lock(&fs_devices->device_list_mutex);
6550         list_for_each_entry(device, &fs_devices->devices, dev_list) {
6551                 int item_size;
6552                 struct btrfs_dev_stats_item *ptr;
6553
6554                 key.objectid = 0;
6555                 key.type = BTRFS_DEV_STATS_KEY;
6556                 key.offset = device->devid;
6557                 ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
6558                 if (ret) {
6559                         __btrfs_reset_dev_stats(device);
6560                         device->dev_stats_valid = 1;
6561                         btrfs_release_path(path);
6562                         continue;
6563                 }
6564                 slot = path->slots[0];
6565                 eb = path->nodes[0];
6566                 btrfs_item_key_to_cpu(eb, &found_key, slot);
6567                 item_size = btrfs_item_size_nr(eb, slot);
6568
6569                 ptr = btrfs_item_ptr(eb, slot,
6570                                      struct btrfs_dev_stats_item);
6571
6572                 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
6573                         if (item_size >= (1 + i) * sizeof(__le64))
6574                                 btrfs_dev_stat_set(device, i,
6575                                         btrfs_dev_stats_value(eb, ptr, i));
6576                         else
6577                                 btrfs_dev_stat_reset(device, i);
6578                 }
6579
6580                 device->dev_stats_valid = 1;
6581                 btrfs_dev_stat_print_on_load(device);
6582                 btrfs_release_path(path);
6583         }
6584         mutex_unlock(&fs_devices->device_list_mutex);
6585
6586 out:
6587         btrfs_free_path(path);
6588         return ret < 0 ? ret : 0;
6589 }
6590
6591 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
6592                                 struct btrfs_root *dev_root,
6593                                 struct btrfs_device *device)
6594 {
6595         struct btrfs_path *path;
6596         struct btrfs_key key;
6597         struct extent_buffer *eb;
6598         struct btrfs_dev_stats_item *ptr;
6599         int ret;
6600         int i;
6601
6602         key.objectid = 0;
6603         key.type = BTRFS_DEV_STATS_KEY;
6604         key.offset = device->devid;
6605
6606         path = btrfs_alloc_path();
6607         BUG_ON(!path);
6608         ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
6609         if (ret < 0) {
6610                 printk_in_rcu(KERN_WARNING "BTRFS: "
6611                         "error %d while searching for dev_stats item for device %s!\n",
6612                               ret, rcu_str_deref(device->name));
6613                 goto out;
6614         }
6615
6616         if (ret == 0 &&
6617             btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
6618                 /* need to delete old one and insert a new one */
6619                 ret = btrfs_del_item(trans, dev_root, path);
6620                 if (ret != 0) {
6621                         printk_in_rcu(KERN_WARNING "BTRFS: "
6622                                 "delete too small dev_stats item for device %s failed %d!\n",
6623                                       rcu_str_deref(device->name), ret);
6624                         goto out;
6625                 }
6626                 ret = 1;
6627         }
6628
6629         if (ret == 1) {
6630                 /* need to insert a new item */
6631                 btrfs_release_path(path);
6632                 ret = btrfs_insert_empty_item(trans, dev_root, path,
6633                                               &key, sizeof(*ptr));
6634                 if (ret < 0) {
6635                         printk_in_rcu(KERN_WARNING "BTRFS: "
6636                                           "insert dev_stats item for device %s failed %d!\n",
6637                                       rcu_str_deref(device->name), ret);
6638                         goto out;
6639                 }
6640         }
6641
6642         eb = path->nodes[0];
6643         ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
6644         for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6645                 btrfs_set_dev_stats_value(eb, ptr, i,
6646                                           btrfs_dev_stat_read(device, i));
6647         btrfs_mark_buffer_dirty(eb);
6648
6649 out:
6650         btrfs_free_path(path);
6651         return ret;
6652 }
6653
6654 /*
6655  * called from commit_transaction. Writes all changed device stats to disk.
6656  */
6657 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans,
6658                         struct btrfs_fs_info *fs_info)
6659 {
6660         struct btrfs_root *dev_root = fs_info->dev_root;
6661         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6662         struct btrfs_device *device;
6663         int stats_cnt;
6664         int ret = 0;
6665
6666         mutex_lock(&fs_devices->device_list_mutex);
6667         list_for_each_entry(device, &fs_devices->devices, dev_list) {
6668                 if (!device->dev_stats_valid || !btrfs_dev_stats_dirty(device))
6669                         continue;
6670
6671                 stats_cnt = atomic_read(&device->dev_stats_ccnt);
6672                 ret = update_dev_stat_item(trans, dev_root, device);
6673                 if (!ret)
6674                         atomic_sub(stats_cnt, &device->dev_stats_ccnt);
6675         }
6676         mutex_unlock(&fs_devices->device_list_mutex);
6677
6678         return ret;
6679 }
6680
6681 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
6682 {
6683         btrfs_dev_stat_inc(dev, index);
6684         btrfs_dev_stat_print_on_error(dev);
6685 }
6686
6687 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
6688 {
6689         if (!dev->dev_stats_valid)
6690                 return;
6691         printk_ratelimited_in_rcu(KERN_ERR "BTRFS: "
6692                            "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
6693                            rcu_str_deref(dev->name),
6694                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
6695                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
6696                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
6697                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
6698                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
6699 }
6700
6701 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
6702 {
6703         int i;
6704
6705         for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6706                 if (btrfs_dev_stat_read(dev, i) != 0)
6707                         break;
6708         if (i == BTRFS_DEV_STAT_VALUES_MAX)
6709                 return; /* all values == 0, suppress message */
6710
6711         printk_in_rcu(KERN_INFO "BTRFS: "
6712                    "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
6713                rcu_str_deref(dev->name),
6714                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
6715                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
6716                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
6717                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
6718                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
6719 }
6720
6721 int btrfs_get_dev_stats(struct btrfs_root *root,
6722                         struct btrfs_ioctl_get_dev_stats *stats)
6723 {
6724         struct btrfs_device *dev;
6725         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
6726         int i;
6727
6728         mutex_lock(&fs_devices->device_list_mutex);
6729         dev = btrfs_find_device(root->fs_info, stats->devid, NULL, NULL);
6730         mutex_unlock(&fs_devices->device_list_mutex);
6731
6732         if (!dev) {
6733                 btrfs_warn(root->fs_info, "get dev_stats failed, device not found");
6734                 return -ENODEV;
6735         } else if (!dev->dev_stats_valid) {
6736                 btrfs_warn(root->fs_info, "get dev_stats failed, not yet valid");
6737                 return -ENODEV;
6738         } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
6739                 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
6740                         if (stats->nr_items > i)
6741                                 stats->values[i] =
6742                                         btrfs_dev_stat_read_and_reset(dev, i);
6743                         else
6744                                 btrfs_dev_stat_reset(dev, i);
6745                 }
6746         } else {
6747                 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6748                         if (stats->nr_items > i)
6749                                 stats->values[i] = btrfs_dev_stat_read(dev, i);
6750         }
6751         if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
6752                 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
6753         return 0;
6754 }
6755
6756 int btrfs_scratch_superblock(struct btrfs_device *device)
6757 {
6758         struct buffer_head *bh;
6759         struct btrfs_super_block *disk_super;
6760
6761         bh = btrfs_read_dev_super(device->bdev);
6762         if (!bh)
6763                 return -EINVAL;
6764         disk_super = (struct btrfs_super_block *)bh->b_data;
6765
6766         memset(&disk_super->magic, 0, sizeof(disk_super->magic));
6767         set_buffer_dirty(bh);
6768         sync_dirty_buffer(bh);
6769         brelse(bh);
6770
6771         return 0;
6772 }
6773
6774 /*
6775  * Update the size of all devices, which is used for writing out the
6776  * super blocks.
6777  */
6778 void btrfs_update_commit_device_size(struct btrfs_fs_info *fs_info)
6779 {
6780         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6781         struct btrfs_device *curr, *next;
6782
6783         if (list_empty(&fs_devices->resized_devices))
6784                 return;
6785
6786         mutex_lock(&fs_devices->device_list_mutex);
6787         lock_chunks(fs_info->dev_root);
6788         list_for_each_entry_safe(curr, next, &fs_devices->resized_devices,
6789                                  resized_list) {
6790                 list_del_init(&curr->resized_list);
6791                 curr->commit_total_bytes = curr->disk_total_bytes;
6792         }
6793         unlock_chunks(fs_info->dev_root);
6794         mutex_unlock(&fs_devices->device_list_mutex);
6795 }
6796
6797 /* Must be invoked during the transaction commit */
6798 void btrfs_update_commit_device_bytes_used(struct btrfs_root *root,
6799                                         struct btrfs_transaction *transaction)
6800 {
6801         struct extent_map *em;
6802         struct map_lookup *map;
6803         struct btrfs_device *dev;
6804         int i;
6805
6806         if (list_empty(&transaction->pending_chunks))
6807                 return;
6808
6809         /* In order to kick the device replace finish process */
6810         lock_chunks(root);
6811         list_for_each_entry(em, &transaction->pending_chunks, list) {
6812                 map = (struct map_lookup *)em->bdev;
6813
6814                 for (i = 0; i < map->num_stripes; i++) {
6815                         dev = map->stripes[i].dev;
6816                         dev->commit_bytes_used = dev->bytes_used;
6817                 }
6818         }
6819         unlock_chunks(root);
6820 }
6821
6822 void btrfs_set_fs_info_ptr(struct btrfs_fs_info *fs_info)
6823 {
6824         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6825         while (fs_devices) {
6826                 fs_devices->fs_info = fs_info;
6827                 fs_devices = fs_devices->seed;
6828         }
6829 }
6830
6831 void btrfs_reset_fs_info_ptr(struct btrfs_fs_info *fs_info)
6832 {
6833         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6834         while (fs_devices) {
6835                 fs_devices->fs_info = NULL;
6836                 fs_devices = fs_devices->seed;
6837         }
6838 }