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 const struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = {
  46         [BTRFS_RAID_RAID10] = {
  47                 .sub_stripes    = 2,
  48                 .dev_stripes    = 1,
  49                 .devs_max       = 0,    /* 0 == as many as possible */
  50                 .devs_min       = 4,
  51                 .tolerated_failures = 1,
  52                 .devs_increment = 2,
  53                 .ncopies        = 2,
  54         },
  55         [BTRFS_RAID_RAID1] = {
  56                 .sub_stripes    = 1,
  57                 .dev_stripes    = 1,
  58                 .devs_max       = 2,
  59                 .devs_min       = 2,
  60                 .tolerated_failures = 1,
  61                 .devs_increment = 2,
  62                 .ncopies        = 2,
  63         },
  64         [BTRFS_RAID_DUP] = {
  65                 .sub_stripes    = 1,
  66                 .dev_stripes    = 2,
  67                 .devs_max       = 1,
  68                 .devs_min       = 1,
  69                 .tolerated_failures = 0,
  70                 .devs_increment = 1,
  71                 .ncopies        = 2,
  72         },
  73         [BTRFS_RAID_RAID0] = {
  74                 .sub_stripes    = 1,
  75                 .dev_stripes    = 1,
  76                 .devs_max       = 0,
  77                 .devs_min       = 2,
  78                 .tolerated_failures = 0,
  79                 .devs_increment = 1,
  80                 .ncopies        = 1,
  81         },
  82         [BTRFS_RAID_SINGLE] = {
  83                 .sub_stripes    = 1,
  84                 .dev_stripes    = 1,
  85                 .devs_max       = 1,
  86                 .devs_min       = 1,
  87                 .tolerated_failures = 0,
  88                 .devs_increment = 1,
  89                 .ncopies        = 1,
  90         },
  91         [BTRFS_RAID_RAID5] = {
  92                 .sub_stripes    = 1,
  93                 .dev_stripes    = 1,
  94                 .devs_max       = 0,
  95                 .devs_min       = 2,
  96                 .tolerated_failures = 1,
  97                 .devs_increment = 1,
  98                 .ncopies        = 2,
  99         },
 100         [BTRFS_RAID_RAID6] = {
 101                 .sub_stripes    = 1,
 102                 .dev_stripes    = 1,
 103                 .devs_max       = 0,
 104                 .devs_min       = 3,
 105                 .tolerated_failures = 2,
 106                 .devs_increment = 1,
 107                 .ncopies        = 3,
 108         },
 109 };
 110
 111 const u64 btrfs_raid_group[BTRFS_NR_RAID_TYPES] = {
 112         [BTRFS_RAID_RAID10] = BTRFS_BLOCK_GROUP_RAID10,
 113         [BTRFS_RAID_RAID1]  = BTRFS_BLOCK_GROUP_RAID1,
 114         [BTRFS_RAID_DUP]    = BTRFS_BLOCK_GROUP_DUP,
 115         [BTRFS_RAID_RAID0]  = BTRFS_BLOCK_GROUP_RAID0,
 116         [BTRFS_RAID_SINGLE] = 0,
 117         [BTRFS_RAID_RAID5]  = BTRFS_BLOCK_GROUP_RAID5,
 118         [BTRFS_RAID_RAID6]  = BTRFS_BLOCK_GROUP_RAID6,
 119 };
 120
 121 static int init_first_rw_device(struct btrfs_trans_handle *trans,
 122                                 struct btrfs_root *root,
 123                                 struct btrfs_device *device);
 124 static int btrfs_relocate_sys_chunks(struct btrfs_root *root);
 125 static void __btrfs_reset_dev_stats(struct btrfs_device *dev);
 126 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev);
 127 static void btrfs_dev_stat_print_on_load(struct btrfs_device *device);
 128 static void btrfs_close_one_device(struct btrfs_device *device);
 129
 130 DEFINE_MUTEX(uuid_mutex);
 131 static LIST_HEAD(fs_uuids);
 132 struct list_head *btrfs_get_fs_uuids(void)
 133 {
 134         return &fs_uuids;
 135 }
 136
 137 static struct btrfs_fs_devices *__alloc_fs_devices(void)
 138 {
 139         struct btrfs_fs_devices *fs_devs;
 140
 141         fs_devs = kzalloc(sizeof(*fs_devs), GFP_NOFS);
 142         if (!fs_devs)
 143                 return ERR_PTR(-ENOMEM);
 144
 145         mutex_init(&fs_devs->device_list_mutex);
 146
 147         INIT_LIST_HEAD(&fs_devs->devices);
 148         INIT_LIST_HEAD(&fs_devs->resized_devices);
 149         INIT_LIST_HEAD(&fs_devs->alloc_list);
 150         INIT_LIST_HEAD(&fs_devs->list);
 151
 152         return fs_devs;
 153 }
 154
 155 /**
 156  * alloc_fs_devices - allocate struct btrfs_fs_devices
 157  * @fsid:       a pointer to UUID for this FS.  If NULL a new UUID is
 158  *              generated.
 159  *
 160  * Return: a pointer to a new &struct btrfs_fs_devices on success;
 161  * ERR_PTR() on error.  Returned struct is not linked onto any lists and
 162  * can be destroyed with kfree() right away.
 163  */
 164 static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid)
 165 {
 166         struct btrfs_fs_devices *fs_devs;
 167
 168         fs_devs = __alloc_fs_devices();
 169         if (IS_ERR(fs_devs))
 170                 return fs_devs;
 171
 172         if (fsid)
 173                 memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE);
 174         else
 175                 generate_random_uuid(fs_devs->fsid);
 176
 177         return fs_devs;
 178 }
 179
 180 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
 181 {
 182         struct btrfs_device *device;
 183         WARN_ON(fs_devices->opened);
 184         while (!list_empty(&fs_devices->devices)) {
 185                 device = list_entry(fs_devices->devices.next,
 186                                     struct btrfs_device, dev_list);
 187                 list_del(&device->dev_list);
 188                 rcu_string_free(device->name);
 189                 kfree(device);
 190         }
 191         kfree(fs_devices);
 192 }
 193
 194 static void btrfs_kobject_uevent(struct block_device *bdev,
 195                                  enum kobject_action action)
 196 {
 197         int ret;
 198
 199         ret = kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, action);
 200         if (ret)
 201                 pr_warn("BTRFS: Sending event '%d' to kobject: '%s' (%p): failed\n",
 202                         action,
 203                         kobject_name(&disk_to_dev(bdev->bd_disk)->kobj),
 204                         &disk_to_dev(bdev->bd_disk)->kobj);
 205 }
 206
 207 void btrfs_cleanup_fs_uuids(void)
 208 {
 209         struct btrfs_fs_devices *fs_devices;
 210
 211         while (!list_empty(&fs_uuids)) {
 212                 fs_devices = list_entry(fs_uuids.next,
 213                                         struct btrfs_fs_devices, list);
 214                 list_del(&fs_devices->list);
 215                 free_fs_devices(fs_devices);
 216         }
 217 }
 218
 219 static struct btrfs_device *__alloc_device(void)
 220 {
 221         struct btrfs_device *dev;
 222
 223         dev = kzalloc(sizeof(*dev), GFP_NOFS);
 224         if (!dev)
 225                 return ERR_PTR(-ENOMEM);
 226
 227         INIT_LIST_HEAD(&dev->dev_list);
 228         INIT_LIST_HEAD(&dev->dev_alloc_list);
 229         INIT_LIST_HEAD(&dev->resized_list);
 230
 231         spin_lock_init(&dev->io_lock);
 232
 233         spin_lock_init(&dev->reada_lock);
 234         atomic_set(&dev->reada_in_flight, 0);
 235         atomic_set(&dev->dev_stats_ccnt, 0);
 236         btrfs_device_data_ordered_init(dev);
 237         INIT_RADIX_TREE(&dev->reada_zones, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
 238         INIT_RADIX_TREE(&dev->reada_extents, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
 239
 240         return dev;
 241 }
 242
 243 static noinline struct btrfs_device *__find_device(struct list_head *head,
 244                                                    u64 devid, u8 *uuid)
 245 {
 246         struct btrfs_device *dev;
 247
 248         list_for_each_entry(dev, head, dev_list) {
 249                 if (dev->devid == devid &&
 250                     (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
 251                         return dev;
 252                 }
 253         }
 254         return NULL;
 255 }
 256
 257 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
 258 {
 259         struct btrfs_fs_devices *fs_devices;
 260
 261         list_for_each_entry(fs_devices, &fs_uuids, list) {
 262                 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
 263                         return fs_devices;
 264         }
 265         return NULL;
 266 }
 267
 268 static int
 269 btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder,
 270                       int flush, struct block_device **bdev,
 271                       struct buffer_head **bh)
 272 {
 273         int ret;
 274
 275         *bdev = blkdev_get_by_path(device_path, flags, holder);
 276
 277         if (IS_ERR(*bdev)) {
 278                 ret = PTR_ERR(*bdev);
 279                 goto error;
 280         }
 281
 282         if (flush)
 283                 filemap_write_and_wait((*bdev)->bd_inode->i_mapping);
 284         ret = set_blocksize(*bdev, 4096);
 285         if (ret) {
 286                 blkdev_put(*bdev, flags);
 287                 goto error;
 288         }
 289         invalidate_bdev(*bdev);
 290         *bh = btrfs_read_dev_super(*bdev);
 291         if (IS_ERR(*bh)) {
 292                 ret = PTR_ERR(*bh);
 293                 blkdev_put(*bdev, flags);
 294                 goto error;
 295         }
 296
 297         return 0;
 298
 299 error:
 300         *bdev = NULL;
 301         *bh = NULL;
 302         return ret;
 303 }
 304
 305 static void requeue_list(struct btrfs_pending_bios *pending_bios,
 306                         struct bio *head, struct bio *tail)
 307 {
 308
 309         struct bio *old_head;
 310
 311         old_head = pending_bios->head;
 312         pending_bios->head = head;
 313         if (pending_bios->tail)
 314                 tail->bi_next = old_head;
 315         else
 316                 pending_bios->tail = tail;
 317 }
 318
 319 /*
 320  * we try to collect pending bios for a device so we don't get a large
 321  * number of procs sending bios down to the same device.  This greatly
 322  * improves the schedulers ability to collect and merge the bios.
 323  *
 324  * But, it also turns into a long list of bios to process and that is sure
 325  * to eventually make the worker thread block.  The solution here is to
 326  * make some progress and then put this work struct back at the end of
 327  * the list if the block device is congested.  This way, multiple devices
 328  * can make progress from a single worker thread.
 329  */
 330 static noinline void run_scheduled_bios(struct btrfs_device *device)
 331 {
 332         struct bio *pending;
 333         struct backing_dev_info *bdi;
 334         struct btrfs_fs_info *fs_info;
 335         struct btrfs_pending_bios *pending_bios;
 336         struct bio *tail;
 337         struct bio *cur;
 338         int again = 0;
 339         unsigned long num_run;
 340         unsigned long batch_run = 0;
 341         unsigned long limit;
 342         unsigned long last_waited = 0;
 343         int force_reg = 0;
 344         int sync_pending = 0;
 345         struct blk_plug plug;
 346
 347         /*
 348          * this function runs all the bios we've collected for
 349          * a particular device.  We don't want to wander off to
 350          * another device without first sending all of these down.
 351          * So, setup a plug here and finish it off before we return
 352          */
 353         blk_start_plug(&plug);
 354
 355         bdi = blk_get_backing_dev_info(device->bdev);
 356         fs_info = device->dev_root->fs_info;
 357         limit = btrfs_async_submit_limit(fs_info);
 358         limit = limit * 2 / 3;
 359
 360 loop:
 361         spin_lock(&device->io_lock);
 362
 363 loop_lock:
 364         num_run = 0;
 365
 366         /* take all the bios off the list at once and process them
 367          * later on (without the lock held).  But, remember the
 368          * tail and other pointers so the bios can be properly reinserted
 369          * into the list if we hit congestion
 370          */
 371         if (!force_reg && device->pending_sync_bios.head) {
 372                 pending_bios = &device->pending_sync_bios;
 373                 force_reg = 1;
 374         } else {
 375                 pending_bios = &device->pending_bios;
 376                 force_reg = 0;
 377         }
 378
 379         pending = pending_bios->head;
 380         tail = pending_bios->tail;
 381         WARN_ON(pending && !tail);
 382
 383         /*
 384          * if pending was null this time around, no bios need processing
 385          * at all and we can stop.  Otherwise it'll loop back up again
 386          * and do an additional check so no bios are missed.
 387          *
 388          * device->running_pending is used to synchronize with the
 389          * schedule_bio code.
 390          */
 391         if (device->pending_sync_bios.head == NULL &&
 392             device->pending_bios.head == NULL) {
 393                 again = 0;
 394                 device->running_pending = 0;
 395         } else {
 396                 again = 1;
 397                 device->running_pending = 1;
 398         }
 399
 400         pending_bios->head = NULL;
 401         pending_bios->tail = NULL;
 402
 403         spin_unlock(&device->io_lock);
 404
 405         while (pending) {
 406
 407                 rmb();
 408                 /* we want to work on both lists, but do more bios on the
 409                  * sync list than the regular list
 410                  */
 411                 if ((num_run > 32 &&
 412                     pending_bios != &device->pending_sync_bios &&
 413                     device->pending_sync_bios.head) ||
 414                    (num_run > 64 && pending_bios == &device->pending_sync_bios &&
 415                     device->pending_bios.head)) {
 416                         spin_lock(&device->io_lock);
 417                         requeue_list(pending_bios, pending, tail);
 418                         goto loop_lock;
 419                 }
 420
 421                 cur = pending;
 422                 pending = pending->bi_next;
 423                 cur->bi_next = NULL;
 424
 425                 /*
 426                  * atomic_dec_return implies a barrier for waitqueue_active
 427                  */
 428                 if (atomic_dec_return(&fs_info->nr_async_bios) < limit &&
 429                     waitqueue_active(&fs_info->async_submit_wait))
 430                         wake_up(&fs_info->async_submit_wait);
 431
 432                 BUG_ON(atomic_read(&cur->__bi_cnt) == 0);
 433
 434                 /*
 435                  * if we're doing the sync list, record that our
 436                  * plug has some sync requests on it
 437                  *
 438                  * If we're doing the regular list and there are
 439                  * sync requests sitting around, unplug before
 440                  * we add more
 441                  */
 442                 if (pending_bios == &device->pending_sync_bios) {
 443                         sync_pending = 1;
 444                 } else if (sync_pending) {
 445                         blk_finish_plug(&plug);
 446                         blk_start_plug(&plug);
 447                         sync_pending = 0;
 448                 }
 449
 450                 btrfsic_submit_bio(cur->bi_rw, cur);
 451                 num_run++;
 452                 batch_run++;
 453
 454                 cond_resched();
 455
 456                 /*
 457                  * we made progress, there is more work to do and the bdi
 458                  * is now congested.  Back off and let other work structs
 459                  * run instead
 460                  */
 461                 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
 462                     fs_info->fs_devices->open_devices > 1) {
 463                         struct io_context *ioc;
 464
 465                         ioc = current->io_context;
 466
 467                         /*
 468                          * the main goal here is that we don't want to
 469                          * block if we're going to be able to submit
 470                          * more requests without blocking.
 471                          *
 472                          * This code does two great things, it pokes into
 473                          * the elevator code from a filesystem _and_
 474                          * it makes assumptions about how batching works.
 475                          */
 476                         if (ioc && ioc->nr_batch_requests > 0 &&
 477                             time_before(jiffies, ioc->last_waited + HZ/50UL) &&
 478                             (last_waited == 0 ||
 479                              ioc->last_waited == last_waited)) {
 480                                 /*
 481                                  * we want to go through our batch of
 482                                  * requests and stop.  So, we copy out
 483                                  * the ioc->last_waited time and test
 484                                  * against it before looping
 485                                  */
 486                                 last_waited = ioc->last_waited;
 487                                 cond_resched();
 488                                 continue;
 489                         }
 490                         spin_lock(&device->io_lock);
 491                         requeue_list(pending_bios, pending, tail);
 492                         device->running_pending = 1;
 493
 494                         spin_unlock(&device->io_lock);
 495                         btrfs_queue_work(fs_info->submit_workers,
 496                                          &device->work);
 497                         goto done;
 498                 }
 499                 /* unplug every 64 requests just for good measure */
 500                 if (batch_run % 64 == 0) {
 501                         blk_finish_plug(&plug);
 502                         blk_start_plug(&plug);
 503                         sync_pending = 0;
 504                 }
 505         }
 506
 507         cond_resched();
 508         if (again)
 509                 goto loop;
 510
 511         spin_lock(&device->io_lock);
 512         if (device->pending_bios.head || device->pending_sync_bios.head)
 513                 goto loop_lock;
 514         spin_unlock(&device->io_lock);
 515
 516 done:
 517         blk_finish_plug(&plug);
 518 }
 519
 520 static void pending_bios_fn(struct btrfs_work *work)
 521 {
 522         struct btrfs_device *device;
 523
 524         device = container_of(work, struct btrfs_device, work);
 525         run_scheduled_bios(device);
 526 }
 527
 528
 529 void btrfs_free_stale_device(struct btrfs_device *cur_dev)
 530 {
 531         struct btrfs_fs_devices *fs_devs;
 532         struct btrfs_device *dev;
 533
 534         if (!cur_dev->name)
 535                 return;
 536
 537         list_for_each_entry(fs_devs, &fs_uuids, list) {
 538                 int del = 1;
 539
 540                 if (fs_devs->opened)
 541                         continue;
 542                 if (fs_devs->seeding)
 543                         continue;
 544
 545                 list_for_each_entry(dev, &fs_devs->devices, dev_list) {
 546
 547                         if (dev == cur_dev)
 548                                 continue;
 549                         if (!dev->name)
 550                                 continue;
 551
 552                         /*
 553                          * Todo: This won't be enough. What if the same device
 554                          * comes back (with new uuid and) with its mapper path?
 555                          * But for now, this does help as mostly an admin will
 556                          * either use mapper or non mapper path throughout.
 557                          */
 558                         rcu_read_lock();
 559                         del = strcmp(rcu_str_deref(dev->name),
 560                                                 rcu_str_deref(cur_dev->name));
 561                         rcu_read_unlock();
 562                         if (!del)
 563                                 break;
 564                 }
 565
 566                 if (!del) {
 567                         /* delete the stale device */
 568                         if (fs_devs->num_devices == 1) {
 569                                 btrfs_sysfs_remove_fsid(fs_devs);
 570                                 list_del(&fs_devs->list);
 571                                 free_fs_devices(fs_devs);
 572                         } else {
 573                                 fs_devs->num_devices--;
 574                                 list_del(&dev->dev_list);
 575                                 rcu_string_free(dev->name);
 576                                 kfree(dev);
 577                         }
 578                         break;
 579                 }
 580         }
 581 }
 582
 583 /*
 584  * Add new device to list of registered devices
 585  *
 586  * Returns:
 587  * 1   - first time device is seen
 588  * 0   - device already known
 589  * < 0 - error
 590  */
 591 static noinline int device_list_add(const char *path,
 592                            struct btrfs_super_block *disk_super,
 593                            u64 devid, struct btrfs_fs_devices **fs_devices_ret)
 594 {
 595         struct btrfs_device *device;
 596         struct btrfs_fs_devices *fs_devices;
 597         struct rcu_string *name;
 598         int ret = 0;
 599         u64 found_transid = btrfs_super_generation(disk_super);
 600
 601         fs_devices = find_fsid(disk_super->fsid);
 602         if (!fs_devices) {
 603                 fs_devices = alloc_fs_devices(disk_super->fsid);
 604                 if (IS_ERR(fs_devices))
 605                         return PTR_ERR(fs_devices);
 606
 607                 list_add(&fs_devices->list, &fs_uuids);
 608
 609                 device = NULL;
 610         } else {
 611                 device = __find_device(&fs_devices->devices, devid,
 612                                        disk_super->dev_item.uuid);
 613         }
 614
 615         if (!device) {
 616                 if (fs_devices->opened)
 617                         return -EBUSY;
 618
 619                 device = btrfs_alloc_device(NULL, &devid,
 620                                             disk_super->dev_item.uuid);
 621                 if (IS_ERR(device)) {
 622                         /* we can safely leave the fs_devices entry around */
 623                         return PTR_ERR(device);
 624                 }
 625
 626                 name = rcu_string_strdup(path, GFP_NOFS);
 627                 if (!name) {
 628                         kfree(device);
 629                         return -ENOMEM;
 630                 }
 631                 rcu_assign_pointer(device->name, name);
 632
 633                 mutex_lock(&fs_devices->device_list_mutex);
 634                 list_add_rcu(&device->dev_list, &fs_devices->devices);
 635                 fs_devices->num_devices++;
 636                 mutex_unlock(&fs_devices->device_list_mutex);
 637
 638                 ret = 1;
 639                 device->fs_devices = fs_devices;
 640         } else if (!device->name || strcmp(device->name->str, path)) {
 641                 /*
 642                  * When FS is already mounted.
 643                  * 1. If you are here and if the device->name is NULL that
 644                  *    means this device was missing at time of FS mount.
 645                  * 2. If you are here and if the device->name is different
 646                  *    from 'path' that means either
 647                  *      a. The same device disappeared and reappeared with
 648                  *         different name. or
 649                  *      b. The missing-disk-which-was-replaced, has
 650                  *         reappeared now.
 651                  *
 652                  * We must allow 1 and 2a above. But 2b would be a spurious
 653                  * and unintentional.
 654                  *
 655                  * Further in case of 1 and 2a above, the disk at 'path'
 656                  * would have missed some transaction when it was away and
 657                  * in case of 2a the stale bdev has to be updated as well.
 658                  * 2b must not be allowed at all time.
 659                  */
 660
 661                 /*
 662                  * For now, we do allow update to btrfs_fs_device through the
 663                  * btrfs dev scan cli after FS has been mounted.  We're still
 664                  * tracking a problem where systems fail mount by subvolume id
 665                  * when we reject replacement on a mounted FS.
 666                  */
 667                 if (!fs_devices->opened && found_transid < device->generation) {
 668                         /*
 669                          * That is if the FS is _not_ mounted and if you
 670                          * are here, that means there is more than one
 671                          * disk with same uuid and devid.We keep the one
 672                          * with larger generation number or the last-in if
 673                          * generation are equal.
 674                          */
 675                         return -EEXIST;
 676                 }
 677
 678                 name = rcu_string_strdup(path, GFP_NOFS);
 679                 if (!name)
 680                         return -ENOMEM;
 681                 rcu_string_free(device->name);
 682                 rcu_assign_pointer(device->name, name);
 683                 if (device->missing) {
 684                         fs_devices->missing_devices--;
 685                         device->missing = 0;
 686                 }
 687         }
 688
 689         /*
 690          * Unmount does not free the btrfs_device struct but would zero
 691          * generation along with most of the other members. So just update
 692          * it back. We need it to pick the disk with largest generation
 693          * (as above).
 694          */
 695         if (!fs_devices->opened)
 696                 device->generation = found_transid;
 697
 698         /*
 699          * if there is new btrfs on an already registered device,
 700          * then remove the stale device entry.
 701          */
 702         btrfs_free_stale_device(device);
 703
 704         *fs_devices_ret = fs_devices;
 705
 706         return ret;
 707 }
 708
 709 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
 710 {
 711         struct btrfs_fs_devices *fs_devices;
 712         struct btrfs_device *device;
 713         struct btrfs_device *orig_dev;
 714
 715         fs_devices = alloc_fs_devices(orig->fsid);
 716         if (IS_ERR(fs_devices))
 717                 return fs_devices;
 718
 719         mutex_lock(&orig->device_list_mutex);
 720         fs_devices->total_devices = orig->total_devices;
 721
 722         /* We have held the volume lock, it is safe to get the devices. */
 723         list_for_each_entry(orig_dev, &orig->devices, dev_list) {
 724                 struct rcu_string *name;
 725
 726                 device = btrfs_alloc_device(NULL, &orig_dev->devid,
 727                                             orig_dev->uuid);
 728                 if (IS_ERR(device))
 729                         goto error;
 730
 731                 /*
 732                  * This is ok to do without rcu read locked because we hold the
 733                  * uuid mutex so nothing we touch in here is going to disappear.
 734                  */
 735                 if (orig_dev->name) {
 736                         name = rcu_string_strdup(orig_dev->name->str, GFP_NOFS);
 737                         if (!name) {
 738                                 kfree(device);
 739                                 goto error;
 740                         }
 741                         rcu_assign_pointer(device->name, name);
 742                 }
 743
 744                 list_add(&device->dev_list, &fs_devices->devices);
 745                 device->fs_devices = fs_devices;
 746                 fs_devices->num_devices++;
 747         }
 748         mutex_unlock(&orig->device_list_mutex);
 749         return fs_devices;
 750 error:
 751         mutex_unlock(&orig->device_list_mutex);
 752         free_fs_devices(fs_devices);
 753         return ERR_PTR(-ENOMEM);
 754 }
 755
 756 void btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices, int step)
 757 {
 758         struct btrfs_device *device, *next;
 759         struct btrfs_device *latest_dev = NULL;
 760
 761         mutex_lock(&uuid_mutex);
 762 again:
 763         /* This is the initialized path, it is safe to release the devices. */
 764         list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
 765                 if (device->in_fs_metadata) {
 766                         if (!device->is_tgtdev_for_dev_replace &&
 767                             (!latest_dev ||
 768                              device->generation > latest_dev->generation)) {
 769                                 latest_dev = device;
 770                         }
 771                         continue;
 772                 }
 773
 774                 if (device->devid == BTRFS_DEV_REPLACE_DEVID) {
 775                         /*
 776                          * In the first step, keep the device which has
 777                          * the correct fsid and the devid that is used
 778                          * for the dev_replace procedure.
 779                          * In the second step, the dev_replace state is
 780                          * read from the device tree and it is known
 781                          * whether the procedure is really active or
 782                          * not, which means whether this device is
 783                          * used or whether it should be removed.
 784                          */
 785                         if (step == 0 || device->is_tgtdev_for_dev_replace) {
 786                                 continue;
 787                         }
 788                 }
 789                 if (device->bdev) {
 790                         blkdev_put(device->bdev, device->mode);
 791                         device->bdev = NULL;
 792                         fs_devices->open_devices--;
 793                 }
 794                 if (device->writeable) {
 795                         list_del_init(&device->dev_alloc_list);
 796                         device->writeable = 0;
 797                         if (!device->is_tgtdev_for_dev_replace)
 798                                 fs_devices->rw_devices--;
 799                 }
 800                 list_del_init(&device->dev_list);
 801                 fs_devices->num_devices--;
 802                 rcu_string_free(device->name);
 803                 kfree(device);
 804         }
 805
 806         if (fs_devices->seed) {
 807                 fs_devices = fs_devices->seed;
 808                 goto again;
 809         }
 810
 811         fs_devices->latest_bdev = latest_dev->bdev;
 812
 813         mutex_unlock(&uuid_mutex);
 814 }
 815
 816 static void __free_device(struct work_struct *work)
 817 {
 818         struct btrfs_device *device;
 819
 820         device = container_of(work, struct btrfs_device, rcu_work);
 821
 822         if (device->bdev)
 823                 blkdev_put(device->bdev, device->mode);
 824
 825         rcu_string_free(device->name);
 826         kfree(device);
 827 }
 828
 829 static void free_device(struct rcu_head *head)
 830 {
 831         struct btrfs_device *device;
 832
 833         device = container_of(head, struct btrfs_device, rcu);
 834
 835         INIT_WORK(&device->rcu_work, __free_device);
 836         schedule_work(&device->rcu_work);
 837 }
 838
 839 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
 840 {
 841         struct btrfs_device *device, *tmp;
 842
 843         if (--fs_devices->opened > 0)
 844                 return 0;
 845
 846         mutex_lock(&fs_devices->device_list_mutex);
 847         list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list) {
 848                 btrfs_close_one_device(device);
 849         }
 850         mutex_unlock(&fs_devices->device_list_mutex);
 851
 852         WARN_ON(fs_devices->open_devices);
 853         WARN_ON(fs_devices->rw_devices);
 854         fs_devices->opened = 0;
 855         fs_devices->seeding = 0;
 856
 857         return 0;
 858 }
 859
 860 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
 861 {
 862         struct btrfs_fs_devices *seed_devices = NULL;
 863         int ret;
 864
 865         mutex_lock(&uuid_mutex);
 866         ret = __btrfs_close_devices(fs_devices);
 867         if (!fs_devices->opened) {
 868                 seed_devices = fs_devices->seed;
 869                 fs_devices->seed = NULL;
 870         }
 871         mutex_unlock(&uuid_mutex);
 872
 873         while (seed_devices) {
 874                 fs_devices = seed_devices;
 875                 seed_devices = fs_devices->seed;
 876                 __btrfs_close_devices(fs_devices);
 877                 free_fs_devices(fs_devices);
 878         }
 879         /*
 880          * Wait for rcu kworkers under __btrfs_close_devices
 881          * to finish all blkdev_puts so device is really
 882          * free when umount is done.
 883          */
 884         rcu_barrier();
 885         return ret;
 886 }
 887
 888 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
 889                                 fmode_t flags, void *holder)
 890 {
 891         struct request_queue *q;
 892         struct block_device *bdev;
 893         struct list_head *head = &fs_devices->devices;
 894         struct btrfs_device *device;
 895         struct btrfs_device *latest_dev = NULL;
 896         struct buffer_head *bh;
 897         struct btrfs_super_block *disk_super;
 898         u64 devid;
 899         int seeding = 1;
 900         int ret = 0;
 901
 902         flags |= FMODE_EXCL;
 903
 904         list_for_each_entry(device, head, dev_list) {
 905                 if (device->bdev)
 906                         continue;
 907                 if (!device->name)
 908                         continue;
 909
 910                 /* Just open everything we can; ignore failures here */
 911                 if (btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1,
 912                                             &bdev, &bh))
 913                         continue;
 914
 915                 disk_super = (struct btrfs_super_block *)bh->b_data;
 916                 devid = btrfs_stack_device_id(&disk_super->dev_item);
 917                 if (devid != device->devid)
 918                         goto error_brelse;
 919
 920                 if (memcmp(device->uuid, disk_super->dev_item.uuid,
 921                            BTRFS_UUID_SIZE))
 922                         goto error_brelse;
 923
 924                 device->generation = btrfs_super_generation(disk_super);
 925                 if (!latest_dev ||
 926                     device->generation > latest_dev->generation)
 927                         latest_dev = device;
 928
 929                 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
 930                         device->writeable = 0;
 931                 } else {
 932                         device->writeable = !bdev_read_only(bdev);
 933                         seeding = 0;
 934                 }
 935
 936                 q = bdev_get_queue(bdev);
 937                 if (blk_queue_discard(q))
 938                         device->can_discard = 1;
 939
 940                 device->bdev = bdev;
 941                 device->in_fs_metadata = 0;
 942                 device->mode = flags;
 943
 944                 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
 945                         fs_devices->rotating = 1;
 946
 947                 fs_devices->open_devices++;
 948                 if (device->writeable &&
 949                     device->devid != BTRFS_DEV_REPLACE_DEVID) {
 950                         fs_devices->rw_devices++;
 951                         list_add(&device->dev_alloc_list,
 952                                  &fs_devices->alloc_list);
 953                 }
 954                 brelse(bh);
 955                 continue;
 956
 957 error_brelse:
 958                 brelse(bh);
 959                 blkdev_put(bdev, flags);
 960                 continue;
 961         }
 962         if (fs_devices->open_devices == 0) {
 963                 ret = -EINVAL;
 964                 goto out;
 965         }
 966         fs_devices->seeding = seeding;
 967         fs_devices->opened = 1;
 968         fs_devices->latest_bdev = latest_dev->bdev;
 969         fs_devices->total_rw_bytes = 0;
 970 out:
 971         return ret;
 972 }
 973
 974 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
 975                        fmode_t flags, void *holder)
 976 {
 977         int ret;
 978
 979         mutex_lock(&uuid_mutex);
 980         if (fs_devices->opened) {
 981                 fs_devices->opened++;
 982                 ret = 0;
 983         } else {
 984                 ret = __btrfs_open_devices(fs_devices, flags, holder);
 985         }
 986         mutex_unlock(&uuid_mutex);
 987         return ret;
 988 }
 989
 990 /*
 991  * Look for a btrfs signature on a device. This may be called out of the mount path
 992  * and we are not allowed to call set_blocksize during the scan. The superblock
 993  * is read via pagecache
 994  */
 995 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
 996                           struct btrfs_fs_devices **fs_devices_ret)
 997 {
 998         struct btrfs_super_block *disk_super;
 999         struct block_device *bdev;
1000         struct page *page;
1001         void *p;
1002         int ret = -EINVAL;
1003         u64 devid;
1004         u64 transid;
1005         u64 total_devices;
1006         u64 bytenr;
1007         pgoff_t index;
1008
1009         /*
1010          * we would like to check all the supers, but that would make
1011          * a btrfs mount succeed after a mkfs from a different FS.
1012          * So, we need to add a special mount option to scan for
1013          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
1014          */
1015         bytenr = btrfs_sb_offset(0);
1016         flags |= FMODE_EXCL;
1017         mutex_lock(&uuid_mutex);
1018
1019         bdev = blkdev_get_by_path(path, flags, holder);
1020
1021         if (IS_ERR(bdev)) {
1022                 ret = PTR_ERR(bdev);
1023                 goto error;
1024         }
1025
1026         /* make sure our super fits in the device */
1027         if (bytenr + PAGE_CACHE_SIZE >= i_size_read(bdev->bd_inode))
1028                 goto error_bdev_put;
1029
1030         /* make sure our super fits in the page */
1031         if (sizeof(*disk_super) > PAGE_CACHE_SIZE)
1032                 goto error_bdev_put;
1033
1034         /* make sure our super doesn't straddle pages on disk */
1035         index = bytenr >> PAGE_CACHE_SHIFT;
1036         if ((bytenr + sizeof(*disk_super) - 1) >> PAGE_CACHE_SHIFT != index)
1037                 goto error_bdev_put;
1038
1039         /* pull in the page with our super */
1040         page = read_cache_page_gfp(bdev->bd_inode->i_mapping,
1041                                    index, GFP_NOFS);
1042
1043         if (IS_ERR_OR_NULL(page))
1044                 goto error_bdev_put;
1045
1046         p = kmap(page);
1047
1048         /* align our pointer to the offset of the super block */
1049         disk_super = p + (bytenr & ~PAGE_CACHE_MASK);
1050
1051         if (btrfs_super_bytenr(disk_super) != bytenr ||
1052             btrfs_super_magic(disk_super) != BTRFS_MAGIC)
1053                 goto error_unmap;
1054
1055         devid = btrfs_stack_device_id(&disk_super->dev_item);
1056         transid = btrfs_super_generation(disk_super);
1057         total_devices = btrfs_super_num_devices(disk_super);
1058
1059         ret = device_list_add(path, disk_super, devid, fs_devices_ret);
1060         if (ret > 0) {
1061                 if (disk_super->label[0]) {
1062                         if (disk_super->label[BTRFS_LABEL_SIZE - 1])
1063                                 disk_super->label[BTRFS_LABEL_SIZE - 1] = '\0';
1064                         printk(KERN_INFO "BTRFS: device label %s ", disk_super->label);
1065                 } else {
1066                         printk(KERN_INFO "BTRFS: device fsid %pU ", disk_super->fsid);
1067                 }
1068
1069                 printk(KERN_CONT "devid %llu transid %llu %s\n", devid, transid, path);
1070                 ret = 0;
1071         }
1072         if (!ret && fs_devices_ret)
1073                 (*fs_devices_ret)->total_devices = total_devices;
1074
1075 error_unmap:
1076         kunmap(page);
1077         page_cache_release(page);
1078
1079 error_bdev_put:
1080         blkdev_put(bdev, flags);
1081 error:
1082         mutex_unlock(&uuid_mutex);
1083         return ret;
1084 }
1085
1086 /* helper to account the used device space in the range */
1087 int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
1088                                    u64 end, u64 *length)
1089 {
1090         struct btrfs_key key;
1091         struct btrfs_root *root = device->dev_root;
1092         struct btrfs_dev_extent *dev_extent;
1093         struct btrfs_path *path;
1094         u64 extent_end;
1095         int ret;
1096         int slot;
1097         struct extent_buffer *l;
1098
1099         *length = 0;
1100
1101         if (start >= device->total_bytes || device->is_tgtdev_for_dev_replace)
1102                 return 0;
1103
1104         path = btrfs_alloc_path();
1105         if (!path)
1106                 return -ENOMEM;
1107         path->reada = READA_FORWARD;
1108
1109         key.objectid = device->devid;
1110         key.offset = start;
1111         key.type = BTRFS_DEV_EXTENT_KEY;
1112
1113         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1114         if (ret < 0)
1115                 goto out;
1116         if (ret > 0) {
1117                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1118                 if (ret < 0)
1119                         goto out;
1120         }
1121
1122         while (1) {
1123                 l = path->nodes[0];
1124                 slot = path->slots[0];
1125                 if (slot >= btrfs_header_nritems(l)) {
1126                         ret = btrfs_next_leaf(root, path);
1127                         if (ret == 0)
1128                                 continue;
1129                         if (ret < 0)
1130                                 goto out;
1131
1132                         break;
1133                 }
1134                 btrfs_item_key_to_cpu(l, &key, slot);
1135
1136                 if (key.objectid < device->devid)
1137                         goto next;
1138
1139                 if (key.objectid > device->devid)
1140                         break;
1141
1142                 if (key.type != BTRFS_DEV_EXTENT_KEY)
1143                         goto next;
1144
1145                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1146                 extent_end = key.offset + btrfs_dev_extent_length(l,
1147                                                                   dev_extent);
1148                 if (key.offset <= start && extent_end > end) {
1149                         *length = end - start + 1;
1150                         break;
1151                 } else if (key.offset <= start && extent_end > start)
1152                         *length += extent_end - start;
1153                 else if (key.offset > start && extent_end <= end)
1154                         *length += extent_end - key.offset;
1155                 else if (key.offset > start && key.offset <= end) {
1156                         *length += end - key.offset + 1;
1157                         break;
1158                 } else if (key.offset > end)
1159                         break;
1160
1161 next:
1162                 path->slots[0]++;
1163         }
1164         ret = 0;
1165 out:
1166         btrfs_free_path(path);
1167         return ret;
1168 }
1169
1170 static int contains_pending_extent(struct btrfs_transaction *transaction,
1171                                    struct btrfs_device *device,
1172                                    u64 *start, u64 len)
1173 {
1174         struct btrfs_fs_info *fs_info = device->dev_root->fs_info;
1175         struct extent_map *em;
1176         struct list_head *search_list = &fs_info->pinned_chunks;
1177         int ret = 0;
1178         u64 physical_start = *start;
1179
1180         if (transaction)
1181                 search_list = &transaction->pending_chunks;
1182 again:
1183         list_for_each_entry(em, search_list, list) {
1184                 struct map_lookup *map;
1185                 int i;
1186
1187                 map = em->map_lookup;
1188                 for (i = 0; i < map->num_stripes; i++) {
1189                         u64 end;
1190
1191                         if (map->stripes[i].dev != device)
1192                                 continue;
1193                         if (map->stripes[i].physical >= physical_start + len ||
1194                             map->stripes[i].physical + em->orig_block_len <=
1195                             physical_start)
1196                                 continue;
1197                         /*
1198                          * Make sure that while processing the pinned list we do
1199                          * not override our *start with a lower value, because
1200                          * we can have pinned chunks that fall within this
1201                          * device hole and that have lower physical addresses
1202                          * than the pending chunks we processed before. If we
1203                          * do not take this special care we can end up getting
1204                          * 2 pending chunks that start at the same physical
1205                          * device offsets because the end offset of a pinned
1206                          * chunk can be equal to the start offset of some
1207                          * pending chunk.
1208                          */
1209                         end = map->stripes[i].physical + em->orig_block_len;
1210                         if (end > *start) {
1211                                 *start = end;
1212                                 ret = 1;
1213                         }
1214                 }
1215         }
1216         if (search_list != &fs_info->pinned_chunks) {
1217                 search_list = &fs_info->pinned_chunks;
1218                 goto again;
1219         }
1220
1221         return ret;
1222 }
1223
1224
1225 /*
1226  * find_free_dev_extent_start - find free space in the specified device
1227  * @device:       the device which we search the free space in
1228  * @num_bytes:    the size of the free space that we need
1229  * @search_start: the position from which to begin the search
1230  * @start:        store the start of the free space.
1231  * @len:          the size of the free space. that we find, or the size
1232  *                of the max free space if we don't find suitable free space
1233  *
1234  * this uses a pretty simple search, the expectation is that it is
1235  * called very infrequently and that a given device has a small number
1236  * of extents
1237  *
1238  * @start is used to store the start of the free space if we find. But if we
1239  * don't find suitable free space, it will be used to store the start position
1240  * of the max free space.
1241  *
1242  * @len is used to store the size of the free space that we find.
1243  * But if we don't find suitable free space, it is used to store the size of
1244  * the max free space.
1245  */
1246 int find_free_dev_extent_start(struct btrfs_transaction *transaction,
1247                                struct btrfs_device *device, u64 num_bytes,
1248                                u64 search_start, u64 *start, u64 *len)
1249 {
1250         struct btrfs_key key;
1251         struct btrfs_root *root = device->dev_root;
1252         struct btrfs_dev_extent *dev_extent;
1253         struct btrfs_path *path;
1254         u64 hole_size;
1255         u64 max_hole_start;
1256         u64 max_hole_size;
1257         u64 extent_end;
1258         u64 search_end = device->total_bytes;
1259         int ret;
1260         int slot;
1261         struct extent_buffer *l;
1262         u64 min_search_start;
1263
1264         /*
1265          * We don't want to overwrite the superblock on the drive nor any area
1266          * used by the boot loader (grub for example), so we make sure to start
1267          * at an offset of at least 1MB.
1268          */
1269         min_search_start = max(root->fs_info->alloc_start, 1024ull * 1024);
1270         search_start = max(search_start, min_search_start);
1271
1272         path = btrfs_alloc_path();
1273         if (!path)
1274                 return -ENOMEM;
1275
1276         max_hole_start = search_start;
1277         max_hole_size = 0;
1278
1279 again:
1280         if (search_start >= search_end || device->is_tgtdev_for_dev_replace) {
1281                 ret = -ENOSPC;
1282                 goto out;
1283         }
1284
1285         path->reada = READA_FORWARD;
1286         path->search_commit_root = 1;
1287         path->skip_locking = 1;
1288
1289         key.objectid = device->devid;
1290         key.offset = search_start;
1291         key.type = BTRFS_DEV_EXTENT_KEY;
1292
1293         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1294         if (ret < 0)
1295                 goto out;
1296         if (ret > 0) {
1297                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1298                 if (ret < 0)
1299                         goto out;
1300         }
1301
1302         while (1) {
1303                 l = path->nodes[0];
1304                 slot = path->slots[0];
1305                 if (slot >= btrfs_header_nritems(l)) {
1306                         ret = btrfs_next_leaf(root, path);
1307                         if (ret == 0)
1308                                 continue;
1309                         if (ret < 0)
1310                                 goto out;
1311
1312                         break;
1313                 }
1314                 btrfs_item_key_to_cpu(l, &key, slot);
1315
1316                 if (key.objectid < device->devid)
1317                         goto next;
1318
1319                 if (key.objectid > device->devid)
1320                         break;
1321
1322                 if (key.type != BTRFS_DEV_EXTENT_KEY)
1323                         goto next;
1324
1325                 if (key.offset > search_start) {
1326                         hole_size = key.offset - search_start;
1327
1328                         /*
1329                          * Have to check before we set max_hole_start, otherwise
1330                          * we could end up sending back this offset anyway.
1331                          */
1332                         if (contains_pending_extent(transaction, device,
1333                                                     &search_start,
1334                                                     hole_size)) {
1335                                 if (key.offset >= search_start) {
1336                                         hole_size = key.offset - search_start;
1337                                 } else {
1338                                         WARN_ON_ONCE(1);
1339                                         hole_size = 0;
1340                                 }
1341                         }
1342
1343                         if (hole_size > max_hole_size) {
1344                                 max_hole_start = search_start;
1345                                 max_hole_size = hole_size;
1346                         }
1347
1348                         /*
1349                          * If this free space is greater than which we need,
1350                          * it must be the max free space that we have found
1351                          * until now, so max_hole_start must point to the start
1352                          * of this free space and the length of this free space
1353                          * is stored in max_hole_size. Thus, we return
1354                          * max_hole_start and max_hole_size and go back to the
1355                          * caller.
1356                          */
1357                         if (hole_size >= num_bytes) {
1358                                 ret = 0;
1359                                 goto out;
1360                         }
1361                 }
1362
1363                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1364                 extent_end = key.offset + btrfs_dev_extent_length(l,
1365                                                                   dev_extent);
1366                 if (extent_end > search_start)
1367                         search_start = extent_end;
1368 next:
1369                 path->slots[0]++;
1370                 cond_resched();
1371         }
1372
1373         /*
1374          * At this point, search_start should be the end of
1375          * allocated dev extents, and when shrinking the device,
1376          * search_end may be smaller than search_start.
1377          */
1378         if (search_end > search_start) {
1379                 hole_size = search_end - search_start;
1380
1381                 if (contains_pending_extent(transaction, device, &search_start,
1382                                             hole_size)) {
1383                         btrfs_release_path(path);
1384                         goto again;
1385                 }
1386
1387                 if (hole_size > max_hole_size) {
1388                         max_hole_start = search_start;
1389                         max_hole_size = hole_size;
1390                 }
1391         }
1392
1393         /* See above. */
1394         if (max_hole_size < num_bytes)
1395                 ret = -ENOSPC;
1396         else
1397                 ret = 0;
1398
1399 out:
1400         btrfs_free_path(path);
1401         *start = max_hole_start;
1402         if (len)
1403                 *len = max_hole_size;
1404         return ret;
1405 }
1406
1407 int find_free_dev_extent(struct btrfs_trans_handle *trans,
1408                          struct btrfs_device *device, u64 num_bytes,
1409                          u64 *start, u64 *len)
1410 {
1411         /* FIXME use last free of some kind */
1412         return find_free_dev_extent_start(trans->transaction, device,
1413                                           num_bytes, 0, start, len);
1414 }
1415
1416 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1417                           struct btrfs_device *device,
1418                           u64 start, u64 *dev_extent_len)
1419 {
1420         int ret;
1421         struct btrfs_path *path;
1422         struct btrfs_root *root = device->dev_root;
1423         struct btrfs_key key;
1424         struct btrfs_key found_key;
1425         struct extent_buffer *leaf = NULL;
1426         struct btrfs_dev_extent *extent = NULL;
1427
1428         path = btrfs_alloc_path();
1429         if (!path)
1430                 return -ENOMEM;
1431
1432         key.objectid = device->devid;
1433         key.offset = start;
1434         key.type = BTRFS_DEV_EXTENT_KEY;
1435 again:
1436         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1437         if (ret > 0) {
1438                 ret = btrfs_previous_item(root, path, key.objectid,
1439                                           BTRFS_DEV_EXTENT_KEY);
1440                 if (ret)
1441                         goto out;
1442                 leaf = path->nodes[0];
1443                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1444                 extent = btrfs_item_ptr(leaf, path->slots[0],
1445                                         struct btrfs_dev_extent);
1446                 BUG_ON(found_key.offset > start || found_key.offset +
1447                        btrfs_dev_extent_length(leaf, extent) < start);
1448                 key = found_key;
1449                 btrfs_release_path(path);
1450                 goto again;
1451         } else if (ret == 0) {
1452                 leaf = path->nodes[0];
1453                 extent = btrfs_item_ptr(leaf, path->slots[0],
1454                                         struct btrfs_dev_extent);
1455         } else {
1456                 btrfs_std_error(root->fs_info, ret, "Slot search failed");
1457                 goto out;
1458         }
1459
1460         *dev_extent_len = btrfs_dev_extent_length(leaf, extent);
1461
1462         ret = btrfs_del_item(trans, root, path);
1463         if (ret) {
1464                 btrfs_std_error(root->fs_info, ret,
1465                             "Failed to remove dev extent item");
1466         } else {
1467                 set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags);
1468         }
1469 out:
1470         btrfs_free_path(path);
1471         return ret;
1472 }
1473
1474 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1475                                   struct btrfs_device *device,
1476                                   u64 chunk_tree, u64 chunk_objectid,
1477                                   u64 chunk_offset, u64 start, u64 num_bytes)
1478 {
1479         int ret;
1480         struct btrfs_path *path;
1481         struct btrfs_root *root = device->dev_root;
1482         struct btrfs_dev_extent *extent;
1483         struct extent_buffer *leaf;
1484         struct btrfs_key key;
1485
1486         WARN_ON(!device->in_fs_metadata);
1487         WARN_ON(device->is_tgtdev_for_dev_replace);
1488         path = btrfs_alloc_path();
1489         if (!path)
1490                 return -ENOMEM;
1491
1492         key.objectid = device->devid;
1493         key.offset = start;
1494         key.type = BTRFS_DEV_EXTENT_KEY;
1495         ret = btrfs_insert_empty_item(trans, root, path, &key,
1496                                       sizeof(*extent));
1497         if (ret)
1498                 goto out;
1499
1500         leaf = path->nodes[0];
1501         extent = btrfs_item_ptr(leaf, path->slots[0],
1502                                 struct btrfs_dev_extent);
1503         btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
1504         btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
1505         btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1506
1507         write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
1508                     btrfs_dev_extent_chunk_tree_uuid(extent), BTRFS_UUID_SIZE);
1509
1510         btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1511         btrfs_mark_buffer_dirty(leaf);
1512 out:
1513         btrfs_free_path(path);
1514         return ret;
1515 }
1516
1517 static u64 find_next_chunk(struct btrfs_fs_info *fs_info)
1518 {
1519         struct extent_map_tree *em_tree;
1520         struct extent_map *em;
1521         struct rb_node *n;
1522         u64 ret = 0;
1523
1524         em_tree = &fs_info->mapping_tree.map_tree;
1525         read_lock(&em_tree->lock);
1526         n = rb_last(&em_tree->map);
1527         if (n) {
1528                 em = rb_entry(n, struct extent_map, rb_node);
1529                 ret = em->start + em->len;
1530         }
1531         read_unlock(&em_tree->lock);
1532
1533         return ret;
1534 }
1535
1536 static noinline int find_next_devid(struct btrfs_fs_info *fs_info,
1537                                     u64 *devid_ret)
1538 {
1539         int ret;
1540         struct btrfs_key key;
1541         struct btrfs_key found_key;
1542         struct btrfs_path *path;
1543
1544         path = btrfs_alloc_path();
1545         if (!path)
1546                 return -ENOMEM;
1547
1548         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1549         key.type = BTRFS_DEV_ITEM_KEY;
1550         key.offset = (u64)-1;
1551
1552         ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
1553         if (ret < 0)
1554                 goto error;
1555
1556         BUG_ON(ret == 0); /* Corruption */
1557
1558         ret = btrfs_previous_item(fs_info->chunk_root, path,
1559                                   BTRFS_DEV_ITEMS_OBJECTID,
1560                                   BTRFS_DEV_ITEM_KEY);
1561         if (ret) {
1562                 *devid_ret = 1;
1563         } else {
1564                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1565                                       path->slots[0]);
1566                 *devid_ret = found_key.offset + 1;
1567         }
1568         ret = 0;
1569 error:
1570         btrfs_free_path(path);
1571         return ret;
1572 }
1573
1574 /*
1575  * the device information is stored in the chunk root
1576  * the btrfs_device struct should be fully filled in
1577  */
1578 static int btrfs_add_device(struct btrfs_trans_handle *trans,
1579                             struct btrfs_root *root,
1580                             struct btrfs_device *device)
1581 {
1582         int ret;
1583         struct btrfs_path *path;
1584         struct btrfs_dev_item *dev_item;
1585         struct extent_buffer *leaf;
1586         struct btrfs_key key;
1587         unsigned long ptr;
1588
1589         root = root->fs_info->chunk_root;
1590
1591         path = btrfs_alloc_path();
1592         if (!path)
1593                 return -ENOMEM;
1594
1595         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1596         key.type = BTRFS_DEV_ITEM_KEY;
1597         key.offset = device->devid;
1598
1599         ret = btrfs_insert_empty_item(trans, root, path, &key,
1600                                       sizeof(*dev_item));
1601         if (ret)
1602                 goto out;
1603
1604         leaf = path->nodes[0];
1605         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1606
1607         btrfs_set_device_id(leaf, dev_item, device->devid);
1608         btrfs_set_device_generation(leaf, dev_item, 0);
1609         btrfs_set_device_type(leaf, dev_item, device->type);
1610         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1611         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1612         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1613         btrfs_set_device_total_bytes(leaf, dev_item,
1614                                      btrfs_device_get_disk_total_bytes(device));
1615         btrfs_set_device_bytes_used(leaf, dev_item,
1616                                     btrfs_device_get_bytes_used(device));
1617         btrfs_set_device_group(leaf, dev_item, 0);
1618         btrfs_set_device_seek_speed(leaf, dev_item, 0);
1619         btrfs_set_device_bandwidth(leaf, dev_item, 0);
1620         btrfs_set_device_start_offset(leaf, dev_item, 0);
1621
1622         ptr = btrfs_device_uuid(dev_item);
1623         write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1624         ptr = btrfs_device_fsid(dev_item);
1625         write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1626         btrfs_mark_buffer_dirty(leaf);
1627
1628         ret = 0;
1629 out:
1630         btrfs_free_path(path);
1631         return ret;
1632 }
1633
1634 /*
1635  * Function to update ctime/mtime for a given device path.
1636  * Mainly used for ctime/mtime based probe like libblkid.
1637  */
1638 static void update_dev_time(char *path_name)
1639 {
1640         struct file *filp;
1641
1642         filp = filp_open(path_name, O_RDWR, 0);
1643         if (IS_ERR(filp))
1644                 return;
1645         file_update_time(filp);
1646         filp_close(filp, NULL);
1647 }
1648
1649 static int btrfs_rm_dev_item(struct btrfs_root *root,
1650                              struct btrfs_device *device)
1651 {
1652         int ret;
1653         struct btrfs_path *path;
1654         struct btrfs_key key;
1655         struct btrfs_trans_handle *trans;
1656
1657         root = root->fs_info->chunk_root;
1658
1659         path = btrfs_alloc_path();
1660         if (!path)
1661                 return -ENOMEM;
1662
1663         trans = btrfs_start_transaction(root, 0);
1664         if (IS_ERR(trans)) {
1665                 btrfs_free_path(path);
1666                 return PTR_ERR(trans);
1667         }
1668         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1669         key.type = BTRFS_DEV_ITEM_KEY;
1670         key.offset = device->devid;
1671
1672         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1673         if (ret < 0)
1674                 goto out;
1675
1676         if (ret > 0) {
1677                 ret = -ENOENT;
1678                 goto out;
1679         }
1680
1681         ret = btrfs_del_item(trans, root, path);
1682         if (ret)
1683                 goto out;
1684 out:
1685         btrfs_free_path(path);
1686         btrfs_commit_transaction(trans, root);
1687         return ret;
1688 }
1689
1690 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
1691 {
1692         struct btrfs_device *device;
1693         struct btrfs_device *next_device;
1694         struct block_device *bdev;
1695         struct buffer_head *bh = NULL;
1696         struct btrfs_super_block *disk_super;
1697         struct btrfs_fs_devices *cur_devices;
1698         u64 all_avail;
1699         u64 devid;
1700         u64 num_devices;
1701         u8 *dev_uuid;
1702         unsigned seq;
1703         int ret = 0;
1704         bool clear_super = false;
1705
1706         mutex_lock(&uuid_mutex);
1707
1708         do {
1709                 seq = read_seqbegin(&root->fs_info->profiles_lock);
1710
1711                 all_avail = root->fs_info->avail_data_alloc_bits |
1712                             root->fs_info->avail_system_alloc_bits |
1713                             root->fs_info->avail_metadata_alloc_bits;
1714         } while (read_seqretry(&root->fs_info->profiles_lock, seq));
1715
1716         num_devices = root->fs_info->fs_devices->num_devices;
1717         btrfs_dev_replace_lock(&root->fs_info->dev_replace);
1718         if (btrfs_dev_replace_is_ongoing(&root->fs_info->dev_replace)) {
1719                 WARN_ON(num_devices < 1);
1720                 num_devices--;
1721         }
1722         btrfs_dev_replace_unlock(&root->fs_info->dev_replace);
1723
1724         if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) && num_devices <= 4) {
1725                 ret = BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET;
1726                 goto out;
1727         }
1728
1729         if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) && num_devices <= 2) {
1730                 ret = BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET;
1731                 goto out;
1732         }
1733
1734         if ((all_avail & BTRFS_BLOCK_GROUP_RAID5) &&
1735             root->fs_info->fs_devices->rw_devices <= 2) {
1736                 ret = BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET;
1737                 goto out;
1738         }
1739         if ((all_avail & BTRFS_BLOCK_GROUP_RAID6) &&
1740             root->fs_info->fs_devices->rw_devices <= 3) {
1741                 ret = BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET;
1742                 goto out;
1743         }
1744
1745         if (strcmp(device_path, "missing") == 0) {
1746                 struct list_head *devices;
1747                 struct btrfs_device *tmp;
1748
1749                 device = NULL;
1750                 devices = &root->fs_info->fs_devices->devices;
1751                 /*
1752                  * It is safe to read the devices since the volume_mutex
1753                  * is held.
1754                  */
1755                 list_for_each_entry(tmp, devices, dev_list) {
1756                         if (tmp->in_fs_metadata &&
1757                             !tmp->is_tgtdev_for_dev_replace &&
1758                             !tmp->bdev) {
1759                                 device = tmp;
1760                                 break;
1761                         }
1762                 }
1763                 bdev = NULL;
1764                 bh = NULL;
1765                 disk_super = NULL;
1766                 if (!device) {
1767                         ret = BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
1768                         goto out;
1769                 }
1770         } else {
1771                 ret = btrfs_get_bdev_and_sb(device_path,
1772                                             FMODE_WRITE | FMODE_EXCL,
1773                                             root->fs_info->bdev_holder, 0,
1774                                             &bdev, &bh);
1775                 if (ret)
1776                         goto out;
1777                 disk_super = (struct btrfs_super_block *)bh->b_data;
1778                 devid = btrfs_stack_device_id(&disk_super->dev_item);
1779                 dev_uuid = disk_super->dev_item.uuid;
1780                 device = btrfs_find_device(root->fs_info, devid, dev_uuid,
1781                                            disk_super->fsid);
1782                 if (!device) {
1783                         ret = -ENOENT;
1784                         goto error_brelse;
1785                 }
1786         }
1787
1788         if (device->is_tgtdev_for_dev_replace) {
1789                 ret = BTRFS_ERROR_DEV_TGT_REPLACE;
1790                 goto error_brelse;
1791         }
1792
1793         if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1794                 ret = BTRFS_ERROR_DEV_ONLY_WRITABLE;
1795                 goto error_brelse;
1796         }
1797
1798         if (device->writeable) {
1799                 lock_chunks(root);
1800                 list_del_init(&device->dev_alloc_list);
1801                 device->fs_devices->rw_devices--;
1802                 unlock_chunks(root);
1803                 clear_super = true;
1804         }
1805
1806         mutex_unlock(&uuid_mutex);
1807         ret = btrfs_shrink_device(device, 0);
1808         mutex_lock(&uuid_mutex);
1809         if (ret)
1810                 goto error_undo;
1811
1812         /*
1813          * TODO: the superblock still includes this device in its num_devices
1814          * counter although write_all_supers() is not locked out. This
1815          * could give a filesystem state which requires a degraded mount.
1816          */
1817         ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1818         if (ret)
1819                 goto error_undo;
1820
1821         device->in_fs_metadata = 0;
1822         btrfs_scrub_cancel_dev(root->fs_info, device);
1823
1824         /*
1825          * the device list mutex makes sure that we don't change
1826          * the device list while someone else is writing out all
1827          * the device supers. Whoever is writing all supers, should
1828          * lock the device list mutex before getting the number of
1829          * devices in the super block (super_copy). Conversely,
1830          * whoever updates the number of devices in the super block
1831          * (super_copy) should hold the device list mutex.
1832          */
1833
1834         cur_devices = device->fs_devices;
1835         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1836         list_del_rcu(&device->dev_list);
1837
1838         device->fs_devices->num_devices--;
1839         device->fs_devices->total_devices--;
1840
1841         if (device->missing)
1842                 device->fs_devices->missing_devices--;
1843
1844         next_device = list_entry(root->fs_info->fs_devices->devices.next,
1845                                  struct btrfs_device, dev_list);
1846         if (device->bdev == root->fs_info->sb->s_bdev)
1847                 root->fs_info->sb->s_bdev = next_device->bdev;
1848         if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1849                 root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1850
1851         if (device->bdev) {
1852                 device->fs_devices->open_devices--;
1853                 /* remove sysfs entry */
1854                 btrfs_sysfs_rm_device_link(root->fs_info->fs_devices, device);
1855         }
1856
1857         call_rcu(&device->rcu, free_device);
1858
1859         num_devices = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
1860         btrfs_set_super_num_devices(root->fs_info->super_copy, num_devices);
1861         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1862
1863         if (cur_devices->open_devices == 0) {
1864                 struct btrfs_fs_devices *fs_devices;
1865                 fs_devices = root->fs_info->fs_devices;
1866                 while (fs_devices) {
1867                         if (fs_devices->seed == cur_devices) {
1868                                 fs_devices->seed = cur_devices->seed;
1869                                 break;
1870                         }
1871                         fs_devices = fs_devices->seed;
1872                 }
1873                 cur_devices->seed = NULL;
1874                 __btrfs_close_devices(cur_devices);
1875                 free_fs_devices(cur_devices);
1876         }
1877
1878         root->fs_info->num_tolerated_disk_barrier_failures =
1879                 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
1880
1881         /*
1882          * at this point, the device is zero sized.  We want to
1883          * remove it from the devices list and zero out the old super
1884          */
1885         if (clear_super && disk_super) {
1886                 u64 bytenr;
1887                 int i;
1888
1889                 /* make sure this device isn't detected as part of
1890                  * the FS anymore
1891                  */
1892                 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1893                 set_buffer_dirty(bh);
1894                 sync_dirty_buffer(bh);
1895
1896                 /* clear the mirror copies of super block on the disk
1897                  * being removed, 0th copy is been taken care above and
1898                  * the below would take of the rest
1899                  */
1900                 for (i = 1; i < BTRFS_SUPER_MIRROR_MAX; i++) {
1901                         bytenr = btrfs_sb_offset(i);
1902                         if (bytenr + BTRFS_SUPER_INFO_SIZE >=
1903                                         i_size_read(bdev->bd_inode))
1904                                 break;
1905
1906                         brelse(bh);
1907                         bh = __bread(bdev, bytenr / 4096,
1908                                         BTRFS_SUPER_INFO_SIZE);
1909                         if (!bh)
1910                                 continue;
1911
1912                         disk_super = (struct btrfs_super_block *)bh->b_data;
1913
1914                         if (btrfs_super_bytenr(disk_super) != bytenr ||
1915                                 btrfs_super_magic(disk_super) != BTRFS_MAGIC) {
1916                                 continue;
1917                         }
1918                         memset(&disk_super->magic, 0,
1919                                                 sizeof(disk_super->magic));
1920                         set_buffer_dirty(bh);
1921                         sync_dirty_buffer(bh);
1922                 }
1923         }
1924
1925         ret = 0;
1926
1927         if (bdev) {
1928                 /* Notify udev that device has changed */
1929                 btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
1930
1931                 /* Update ctime/mtime for device path for libblkid */
1932                 update_dev_time(device_path);
1933         }
1934
1935 error_brelse:
1936         brelse(bh);
1937         if (bdev)
1938                 blkdev_put(bdev, FMODE_READ | FMODE_EXCL);
1939 out:
1940         mutex_unlock(&uuid_mutex);
1941         return ret;
1942 error_undo:
1943         if (device->writeable) {
1944                 lock_chunks(root);
1945                 list_add(&device->dev_alloc_list,
1946                          &root->fs_info->fs_devices->alloc_list);
1947                 device->fs_devices->rw_devices++;
1948                 unlock_chunks(root);
1949         }
1950         goto error_brelse;
1951 }
1952
1953 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_fs_info *fs_info,
1954                                         struct btrfs_device *srcdev)
1955 {
1956         struct btrfs_fs_devices *fs_devices;
1957
1958         WARN_ON(!mutex_is_locked(&fs_info->fs_devices->device_list_mutex));
1959
1960         /*
1961          * in case of fs with no seed, srcdev->fs_devices will point
1962          * to fs_devices of fs_info. However when the dev being replaced is
1963          * a seed dev it will point to the seed's local fs_devices. In short
1964          * srcdev will have its correct fs_devices in both the cases.
1965          */
1966         fs_devices = srcdev->fs_devices;
1967
1968         list_del_rcu(&srcdev->dev_list);
1969         list_del_rcu(&srcdev->dev_alloc_list);
1970         fs_devices->num_devices--;
1971         if (srcdev->missing)
1972                 fs_devices->missing_devices--;
1973
1974         if (srcdev->writeable) {
1975                 fs_devices->rw_devices--;
1976                 /* zero out the old super if it is writable */
1977                 btrfs_scratch_superblocks(srcdev->bdev, srcdev->name->str);
1978         }
1979
1980         if (srcdev->bdev)
1981                 fs_devices->open_devices--;
1982 }
1983
1984 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_fs_info *fs_info,
1985                                       struct btrfs_device *srcdev)
1986 {
1987         struct btrfs_fs_devices *fs_devices = srcdev->fs_devices;
1988
1989         call_rcu(&srcdev->rcu, free_device);
1990
1991         /*
1992          * unless fs_devices is seed fs, num_devices shouldn't go
1993          * zero
1994          */
1995         BUG_ON(!fs_devices->num_devices && !fs_devices->seeding);
1996
1997         /* if this is no devs we rather delete the fs_devices */
1998         if (!fs_devices->num_devices) {
1999                 struct btrfs_fs_devices *tmp_fs_devices;
2000
2001                 tmp_fs_devices = fs_info->fs_devices;
2002                 while (tmp_fs_devices) {
2003                         if (tmp_fs_devices->seed == fs_devices) {
2004                                 tmp_fs_devices->seed = fs_devices->seed;
2005                                 break;
2006                         }
2007                         tmp_fs_devices = tmp_fs_devices->seed;
2008                 }
2009                 fs_devices->seed = NULL;
2010                 __btrfs_close_devices(fs_devices);
2011                 free_fs_devices(fs_devices);
2012         }
2013 }
2014
2015 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
2016                                       struct btrfs_device *tgtdev)
2017 {
2018         struct btrfs_device *next_device;
2019
2020         mutex_lock(&uuid_mutex);
2021         WARN_ON(!tgtdev);
2022         mutex_lock(&fs_info->fs_devices->device_list_mutex);
2023
2024         btrfs_sysfs_rm_device_link(fs_info->fs_devices, tgtdev);
2025
2026         if (tgtdev->bdev) {
2027                 btrfs_scratch_superblocks(tgtdev->bdev, tgtdev->name->str);
2028                 fs_info->fs_devices->open_devices--;
2029         }
2030         fs_info->fs_devices->num_devices--;
2031
2032         next_device = list_entry(fs_info->fs_devices->devices.next,
2033                                  struct btrfs_device, dev_list);
2034         if (tgtdev->bdev == fs_info->sb->s_bdev)
2035                 fs_info->sb->s_bdev = next_device->bdev;
2036         if (tgtdev->bdev == fs_info->fs_devices->latest_bdev)
2037                 fs_info->fs_devices->latest_bdev = next_device->bdev;
2038         list_del_rcu(&tgtdev->dev_list);
2039
2040         call_rcu(&tgtdev->rcu, free_device);
2041
2042         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2043         mutex_unlock(&uuid_mutex);
2044 }
2045
2046 static int btrfs_find_device_by_path(struct btrfs_root *root, char *device_path,
2047                                      struct btrfs_device **device)
2048 {
2049         int ret = 0;
2050         struct btrfs_super_block *disk_super;
2051         u64 devid;
2052         u8 *dev_uuid;
2053         struct block_device *bdev;
2054         struct buffer_head *bh;
2055
2056         *device = NULL;
2057         ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
2058                                     root->fs_info->bdev_holder, 0, &bdev, &bh);
2059         if (ret)
2060                 return ret;
2061         disk_super = (struct btrfs_super_block *)bh->b_data;
2062         devid = btrfs_stack_device_id(&disk_super->dev_item);
2063         dev_uuid = disk_super->dev_item.uuid;
2064         *device = btrfs_find_device(root->fs_info, devid, dev_uuid,
2065                                     disk_super->fsid);
2066         brelse(bh);
2067         if (!*device)
2068                 ret = -ENOENT;
2069         blkdev_put(bdev, FMODE_READ);
2070         return ret;
2071 }
2072
2073 int btrfs_find_device_missing_or_by_path(struct btrfs_root *root,
2074                                          char *device_path,
2075                                          struct btrfs_device **device)
2076 {
2077         *device = NULL;
2078         if (strcmp(device_path, "missing") == 0) {
2079                 struct list_head *devices;
2080                 struct btrfs_device *tmp;
2081
2082                 devices = &root->fs_info->fs_devices->devices;
2083                 /*
2084                  * It is safe to read the devices since the volume_mutex
2085                  * is held by the caller.
2086                  */
2087                 list_for_each_entry(tmp, devices, dev_list) {
2088                         if (tmp->in_fs_metadata && !tmp->bdev) {
2089                                 *device = tmp;
2090                                 break;
2091                         }
2092                 }
2093
2094                 if (!*device)
2095                         return BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
2096
2097                 return 0;
2098         } else {
2099                 return btrfs_find_device_by_path(root, device_path, device);
2100         }
2101 }
2102
2103 /*
2104  * does all the dirty work required for changing file system's UUID.
2105  */
2106 static int btrfs_prepare_sprout(struct btrfs_root *root)
2107 {
2108         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2109         struct btrfs_fs_devices *old_devices;
2110         struct btrfs_fs_devices *seed_devices;
2111         struct btrfs_super_block *disk_super = root->fs_info->super_copy;
2112         struct btrfs_device *device;
2113         u64 super_flags;
2114
2115         BUG_ON(!mutex_is_locked(&uuid_mutex));
2116         if (!fs_devices->seeding)
2117                 return -EINVAL;
2118
2119         seed_devices = __alloc_fs_devices();
2120         if (IS_ERR(seed_devices))
2121                 return PTR_ERR(seed_devices);
2122
2123         old_devices = clone_fs_devices(fs_devices);
2124         if (IS_ERR(old_devices)) {
2125                 kfree(seed_devices);
2126                 return PTR_ERR(old_devices);
2127         }
2128
2129         list_add(&old_devices->list, &fs_uuids);
2130
2131         memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
2132         seed_devices->opened = 1;
2133         INIT_LIST_HEAD(&seed_devices->devices);
2134         INIT_LIST_HEAD(&seed_devices->alloc_list);
2135         mutex_init(&seed_devices->device_list_mutex);
2136
2137         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2138         list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
2139                               synchronize_rcu);
2140         list_for_each_entry(device, &seed_devices->devices, dev_list)
2141                 device->fs_devices = seed_devices;
2142
2143         lock_chunks(root);
2144         list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
2145         unlock_chunks(root);
2146
2147         fs_devices->seeding = 0;
2148         fs_devices->num_devices = 0;
2149         fs_devices->open_devices = 0;
2150         fs_devices->missing_devices = 0;
2151         fs_devices->rotating = 0;
2152         fs_devices->seed = seed_devices;
2153
2154         generate_random_uuid(fs_devices->fsid);
2155         memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2156         memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2157         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2158
2159         super_flags = btrfs_super_flags(disk_super) &
2160                       ~BTRFS_SUPER_FLAG_SEEDING;
2161         btrfs_set_super_flags(disk_super, super_flags);
2162
2163         return 0;
2164 }
2165
2166 /*
2167  * strore the expected generation for seed devices in device items.
2168  */
2169 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
2170                                struct btrfs_root *root)
2171 {
2172         struct btrfs_path *path;
2173         struct extent_buffer *leaf;
2174         struct btrfs_dev_item *dev_item;
2175         struct btrfs_device *device;
2176         struct btrfs_key key;
2177         u8 fs_uuid[BTRFS_UUID_SIZE];
2178         u8 dev_uuid[BTRFS_UUID_SIZE];
2179         u64 devid;
2180         int ret;
2181
2182         path = btrfs_alloc_path();
2183         if (!path)
2184                 return -ENOMEM;
2185
2186         root = root->fs_info->chunk_root;
2187         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2188         key.offset = 0;
2189         key.type = BTRFS_DEV_ITEM_KEY;
2190
2191         while (1) {
2192                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2193                 if (ret < 0)
2194                         goto error;
2195
2196                 leaf = path->nodes[0];
2197 next_slot:
2198                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2199                         ret = btrfs_next_leaf(root, path);
2200                         if (ret > 0)
2201                                 break;
2202                         if (ret < 0)
2203                                 goto error;
2204                         leaf = path->nodes[0];
2205                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2206                         btrfs_release_path(path);
2207                         continue;
2208                 }
2209
2210                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2211                 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
2212                     key.type != BTRFS_DEV_ITEM_KEY)
2213                         break;
2214
2215                 dev_item = btrfs_item_ptr(leaf, path->slots[0],
2216                                           struct btrfs_dev_item);
2217                 devid = btrfs_device_id(leaf, dev_item);
2218                 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
2219                                    BTRFS_UUID_SIZE);
2220                 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
2221                                    BTRFS_UUID_SIZE);
2222                 device = btrfs_find_device(root->fs_info, devid, dev_uuid,
2223                                            fs_uuid);
2224                 BUG_ON(!device); /* Logic error */
2225
2226                 if (device->fs_devices->seeding) {
2227                         btrfs_set_device_generation(leaf, dev_item,
2228                                                     device->generation);
2229                         btrfs_mark_buffer_dirty(leaf);
2230                 }
2231
2232                 path->slots[0]++;
2233                 goto next_slot;
2234         }
2235         ret = 0;
2236 error:
2237         btrfs_free_path(path);
2238         return ret;
2239 }
2240
2241 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
2242 {
2243         struct request_queue *q;
2244         struct btrfs_trans_handle *trans;
2245         struct btrfs_device *device;
2246         struct block_device *bdev;
2247         struct list_head *devices;
2248         struct super_block *sb = root->fs_info->sb;
2249         struct rcu_string *name;
2250         u64 tmp;
2251         int seeding_dev = 0;
2252         int ret = 0;
2253
2254         if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
2255                 return -EROFS;
2256
2257         bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2258                                   root->fs_info->bdev_holder);
2259         if (IS_ERR(bdev))
2260                 return PTR_ERR(bdev);
2261
2262         if (root->fs_info->fs_devices->seeding) {
2263                 seeding_dev = 1;
2264                 down_write(&sb->s_umount);
2265                 mutex_lock(&uuid_mutex);
2266         }
2267
2268         filemap_write_and_wait(bdev->bd_inode->i_mapping);
2269
2270         devices = &root->fs_info->fs_devices->devices;
2271
2272         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2273         list_for_each_entry(device, devices, dev_list) {
2274                 if (device->bdev == bdev) {
2275                         ret = -EEXIST;
2276                         mutex_unlock(
2277                                 &root->fs_info->fs_devices->device_list_mutex);
2278                         goto error;
2279                 }
2280         }
2281         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2282
2283         device = btrfs_alloc_device(root->fs_info, NULL, NULL);
2284         if (IS_ERR(device)) {
2285                 /* we can safely leave the fs_devices entry around */
2286                 ret = PTR_ERR(device);
2287                 goto error;
2288         }
2289
2290         name = rcu_string_strdup(device_path, GFP_NOFS);
2291         if (!name) {
2292                 kfree(device);
2293                 ret = -ENOMEM;
2294                 goto error;
2295         }
2296         rcu_assign_pointer(device->name, name);
2297
2298         trans = btrfs_start_transaction(root, 0);
2299         if (IS_ERR(trans)) {
2300                 rcu_string_free(device->name);
2301                 kfree(device);
2302                 ret = PTR_ERR(trans);
2303                 goto error;
2304         }
2305
2306         q = bdev_get_queue(bdev);
2307         if (blk_queue_discard(q))
2308                 device->can_discard = 1;
2309         device->writeable = 1;
2310         device->generation = trans->transid;
2311         device->io_width = root->sectorsize;
2312         device->io_align = root->sectorsize;
2313         device->sector_size = root->sectorsize;
2314         device->total_bytes = i_size_read(bdev->bd_inode);
2315         device->disk_total_bytes = device->total_bytes;
2316         device->commit_total_bytes = device->total_bytes;
2317         device->dev_root = root->fs_info->dev_root;
2318         device->bdev = bdev;
2319         device->in_fs_metadata = 1;
2320         device->is_tgtdev_for_dev_replace = 0;
2321         device->mode = FMODE_EXCL;
2322         device->dev_stats_valid = 1;
2323         set_blocksize(device->bdev, 4096);
2324
2325         if (seeding_dev) {
2326                 sb->s_flags &= ~MS_RDONLY;
2327                 ret = btrfs_prepare_sprout(root);
2328                 BUG_ON(ret); /* -ENOMEM */
2329         }
2330
2331         device->fs_devices = root->fs_info->fs_devices;
2332
2333         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2334         lock_chunks(root);
2335         list_add_rcu(&device->dev_list, &root->fs_info->fs_devices->devices);
2336         list_add(&device->dev_alloc_list,
2337                  &root->fs_info->fs_devices->alloc_list);
2338         root->fs_info->fs_devices->num_devices++;
2339         root->fs_info->fs_devices->open_devices++;
2340         root->fs_info->fs_devices->rw_devices++;
2341         root->fs_info->fs_devices->total_devices++;
2342         root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
2343
2344         spin_lock(&root->fs_info->free_chunk_lock);
2345         root->fs_info->free_chunk_space += device->total_bytes;
2346         spin_unlock(&root->fs_info->free_chunk_lock);
2347
2348         if (!blk_queue_nonrot(bdev_get_queue(bdev)))
2349                 root->fs_info->fs_devices->rotating = 1;
2350
2351         tmp = btrfs_super_total_bytes(root->fs_info->super_copy);
2352         btrfs_set_super_total_bytes(root->fs_info->super_copy,
2353                                     tmp + device->total_bytes);
2354
2355         tmp = btrfs_super_num_devices(root->fs_info->super_copy);
2356         btrfs_set_super_num_devices(root->fs_info->super_copy,
2357                                     tmp + 1);
2358
2359         /* add sysfs device entry */
2360         btrfs_sysfs_add_device_link(root->fs_info->fs_devices, device);
2361
2362         /*
2363          * we've got more storage, clear any full flags on the space
2364          * infos
2365          */
2366         btrfs_clear_space_info_full(root->fs_info);
2367
2368         unlock_chunks(root);
2369         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2370
2371         if (seeding_dev) {
2372                 lock_chunks(root);
2373                 ret = init_first_rw_device(trans, root, device);
2374                 unlock_chunks(root);
2375                 if (ret) {
2376                         btrfs_abort_transaction(trans, root, ret);
2377                         goto error_trans;
2378                 }
2379         }
2380
2381         ret = btrfs_add_device(trans, root, device);
2382         if (ret) {
2383                 btrfs_abort_transaction(trans, root, ret);
2384                 goto error_trans;
2385         }
2386
2387         if (seeding_dev) {
2388                 char fsid_buf[BTRFS_UUID_UNPARSED_SIZE];
2389
2390                 ret = btrfs_finish_sprout(trans, root);
2391                 if (ret) {
2392                         btrfs_abort_transaction(trans, root, ret);
2393                         goto error_trans;
2394                 }
2395
2396                 /* Sprouting would change fsid of the mounted root,
2397                  * so rename the fsid on the sysfs
2398                  */
2399                 snprintf(fsid_buf, BTRFS_UUID_UNPARSED_SIZE, "%pU",
2400                                                 root->fs_info->fsid);
2401                 if (kobject_rename(&root->fs_info->fs_devices->fsid_kobj,
2402                                                                 fsid_buf))
2403                         btrfs_warn(root->fs_info,
2404                                 "sysfs: failed to create fsid for sprout");
2405         }
2406
2407         root->fs_info->num_tolerated_disk_barrier_failures =
2408                 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
2409         ret = btrfs_commit_transaction(trans, root);
2410
2411         if (seeding_dev) {
2412                 mutex_unlock(&uuid_mutex);
2413                 up_write(&sb->s_umount);
2414
2415                 if (ret) /* transaction commit */
2416                         return ret;
2417
2418                 ret = btrfs_relocate_sys_chunks(root);
2419                 if (ret < 0)
2420                         btrfs_std_error(root->fs_info, ret,
2421                                     "Failed to relocate sys chunks after "
2422                                     "device initialization. This can be fixed "
2423                                     "using the \"btrfs balance\" command.");
2424                 trans = btrfs_attach_transaction(root);
2425                 if (IS_ERR(trans)) {
2426                         if (PTR_ERR(trans) == -ENOENT)
2427                                 return 0;
2428                         return PTR_ERR(trans);
2429                 }
2430                 ret = btrfs_commit_transaction(trans, root);
2431         }
2432
2433         /* Update ctime/mtime for libblkid */
2434         update_dev_time(device_path);
2435         return ret;
2436
2437 error_trans:
2438         btrfs_end_transaction(trans, root);
2439         rcu_string_free(device->name);
2440         btrfs_sysfs_rm_device_link(root->fs_info->fs_devices, device);
2441         kfree(device);
2442 error:
2443         blkdev_put(bdev, FMODE_EXCL);
2444         if (seeding_dev) {
2445                 mutex_unlock(&uuid_mutex);
2446                 up_write(&sb->s_umount);
2447         }
2448         return ret;
2449 }
2450
2451 int btrfs_init_dev_replace_tgtdev(struct btrfs_root *root, char *device_path,
2452                                   struct btrfs_device *srcdev,
2453                                   struct btrfs_device **device_out)
2454 {
2455         struct request_queue *q;
2456         struct btrfs_device *device;
2457         struct block_device *bdev;
2458         struct btrfs_fs_info *fs_info = root->fs_info;
2459         struct list_head *devices;
2460         struct rcu_string *name;
2461         u64 devid = BTRFS_DEV_REPLACE_DEVID;
2462         int ret = 0;
2463
2464         *device_out = NULL;
2465         if (fs_info->fs_devices->seeding) {
2466                 btrfs_err(fs_info, "the filesystem is a seed filesystem!");
2467                 return -EINVAL;
2468         }
2469
2470         bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2471                                   fs_info->bdev_holder);
2472         if (IS_ERR(bdev)) {
2473                 btrfs_err(fs_info, "target device %s is invalid!", device_path);
2474                 return PTR_ERR(bdev);
2475         }
2476
2477         filemap_write_and_wait(bdev->bd_inode->i_mapping);
2478
2479         devices = &fs_info->fs_devices->devices;
2480         list_for_each_entry(device, devices, dev_list) {
2481                 if (device->bdev == bdev) {
2482                         btrfs_err(fs_info, "target device is in the filesystem!");
2483                         ret = -EEXIST;
2484                         goto error;
2485                 }
2486         }
2487
2488
2489         if (i_size_read(bdev->bd_inode) <
2490             btrfs_device_get_total_bytes(srcdev)) {
2491                 btrfs_err(fs_info, "target device is smaller than source device!");
2492                 ret = -EINVAL;
2493                 goto error;
2494         }
2495
2496
2497         device = btrfs_alloc_device(NULL, &devid, NULL);
2498         if (IS_ERR(device)) {
2499                 ret = PTR_ERR(device);
2500                 goto error;
2501         }
2502
2503         name = rcu_string_strdup(device_path, GFP_NOFS);
2504         if (!name) {
2505                 kfree(device);
2506                 ret = -ENOMEM;
2507                 goto error;
2508         }
2509         rcu_assign_pointer(device->name, name);
2510
2511         q = bdev_get_queue(bdev);
2512         if (blk_queue_discard(q))
2513                 device->can_discard = 1;
2514         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2515         device->writeable = 1;
2516         device->generation = 0;
2517         device->io_width = root->sectorsize;
2518         device->io_align = root->sectorsize;
2519         device->sector_size = root->sectorsize;
2520         device->total_bytes = btrfs_device_get_total_bytes(srcdev);
2521         device->disk_total_bytes = btrfs_device_get_disk_total_bytes(srcdev);
2522         device->bytes_used = btrfs_device_get_bytes_used(srcdev);
2523         ASSERT(list_empty(&srcdev->resized_list));
2524         device->commit_total_bytes = srcdev->commit_total_bytes;
2525         device->commit_bytes_used = device->bytes_used;
2526         device->dev_root = fs_info->dev_root;
2527         device->bdev = bdev;
2528         device->in_fs_metadata = 1;
2529         device->is_tgtdev_for_dev_replace = 1;
2530         device->mode = FMODE_EXCL;
2531         device->dev_stats_valid = 1;
2532         set_blocksize(device->bdev, 4096);
2533         device->fs_devices = fs_info->fs_devices;
2534         list_add(&device->dev_list, &fs_info->fs_devices->devices);
2535         fs_info->fs_devices->num_devices++;
2536         fs_info->fs_devices->open_devices++;
2537         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2538
2539         *device_out = device;
2540         return ret;
2541
2542 error:
2543         blkdev_put(bdev, FMODE_EXCL);
2544         return ret;
2545 }
2546
2547 void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info *fs_info,
2548                                               struct btrfs_device *tgtdev)
2549 {
2550         WARN_ON(fs_info->fs_devices->rw_devices == 0);
2551         tgtdev->io_width = fs_info->dev_root->sectorsize;
2552         tgtdev->io_align = fs_info->dev_root->sectorsize;
2553         tgtdev->sector_size = fs_info->dev_root->sectorsize;
2554         tgtdev->dev_root = fs_info->dev_root;
2555         tgtdev->in_fs_metadata = 1;
2556 }
2557
2558 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2559                                         struct btrfs_device *device)
2560 {
2561         int ret;
2562         struct btrfs_path *path;
2563         struct btrfs_root *root;
2564         struct btrfs_dev_item *dev_item;
2565         struct extent_buffer *leaf;
2566         struct btrfs_key key;
2567
2568         root = device->dev_root->fs_info->chunk_root;
2569
2570         path = btrfs_alloc_path();
2571         if (!path)
2572                 return -ENOMEM;
2573
2574         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2575         key.type = BTRFS_DEV_ITEM_KEY;
2576         key.offset = device->devid;
2577
2578         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2579         if (ret < 0)
2580                 goto out;
2581
2582         if (ret > 0) {
2583                 ret = -ENOENT;
2584                 goto out;
2585         }
2586
2587         leaf = path->nodes[0];
2588         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2589
2590         btrfs_set_device_id(leaf, dev_item, device->devid);
2591         btrfs_set_device_type(leaf, dev_item, device->type);
2592         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2593         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2594         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2595         btrfs_set_device_total_bytes(leaf, dev_item,
2596                                      btrfs_device_get_disk_total_bytes(device));
2597         btrfs_set_device_bytes_used(leaf, dev_item,
2598                                     btrfs_device_get_bytes_used(device));
2599         btrfs_mark_buffer_dirty(leaf);
2600
2601 out:
2602         btrfs_free_path(path);
2603         return ret;
2604 }
2605
2606 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2607                       struct btrfs_device *device, u64 new_size)
2608 {
2609         struct btrfs_super_block *super_copy =
2610                 device->dev_root->fs_info->super_copy;
2611         struct btrfs_fs_devices *fs_devices;
2612         u64 old_total;
2613         u64 diff;
2614
2615         if (!device->writeable)
2616                 return -EACCES;
2617
2618         lock_chunks(device->dev_root);
2619         old_total = btrfs_super_total_bytes(super_copy);
2620         diff = new_size - device->total_bytes;
2621
2622         if (new_size <= device->total_bytes ||
2623             device->is_tgtdev_for_dev_replace) {
2624                 unlock_chunks(device->dev_root);
2625                 return -EINVAL;
2626         }
2627
2628         fs_devices = device->dev_root->fs_info->fs_devices;
2629
2630         btrfs_set_super_total_bytes(super_copy, old_total + diff);
2631         device->fs_devices->total_rw_bytes += diff;
2632
2633         btrfs_device_set_total_bytes(device, new_size);
2634         btrfs_device_set_disk_total_bytes(device, new_size);
2635         btrfs_clear_space_info_full(device->dev_root->fs_info);
2636         if (list_empty(&device->resized_list))
2637                 list_add_tail(&device->resized_list,
2638                               &fs_devices->resized_devices);
2639         unlock_chunks(device->dev_root);
2640
2641         return btrfs_update_device(trans, device);
2642 }
2643
2644 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
2645                             struct btrfs_root *root, u64 chunk_objectid,
2646                             u64 chunk_offset)
2647 {
2648         int ret;
2649         struct btrfs_path *path;
2650         struct btrfs_key key;
2651
2652         root = root->fs_info->chunk_root;
2653         path = btrfs_alloc_path();
2654         if (!path)
2655                 return -ENOMEM;
2656
2657         key.objectid = chunk_objectid;
2658         key.offset = chunk_offset;
2659         key.type = BTRFS_CHUNK_ITEM_KEY;
2660
2661         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2662         if (ret < 0)
2663                 goto out;
2664         else if (ret > 0) { /* Logic error or corruption */
2665                 btrfs_std_error(root->fs_info, -ENOENT,
2666                             "Failed lookup while freeing chunk.");
2667                 ret = -ENOENT;
2668                 goto out;
2669         }
2670
2671         ret = btrfs_del_item(trans, root, path);
2672         if (ret < 0)
2673                 btrfs_std_error(root->fs_info, ret,
2674                             "Failed to delete chunk item.");
2675 out:
2676         btrfs_free_path(path);
2677         return ret;
2678 }
2679
2680 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
2681                         chunk_offset)
2682 {
2683         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
2684         struct btrfs_disk_key *disk_key;
2685         struct btrfs_chunk *chunk;
2686         u8 *ptr;
2687         int ret = 0;
2688         u32 num_stripes;
2689         u32 array_size;
2690         u32 len = 0;
2691         u32 cur;
2692         struct btrfs_key key;
2693
2694         lock_chunks(root);
2695         array_size = btrfs_super_sys_array_size(super_copy);
2696
2697         ptr = super_copy->sys_chunk_array;
2698         cur = 0;
2699
2700         while (cur < array_size) {
2701                 disk_key = (struct btrfs_disk_key *)ptr;
2702                 btrfs_disk_key_to_cpu(&key, disk_key);
2703
2704                 len = sizeof(*disk_key);
2705
2706                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2707                         chunk = (struct btrfs_chunk *)(ptr + len);
2708                         num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2709                         len += btrfs_chunk_item_size(num_stripes);
2710                 } else {
2711                         ret = -EIO;
2712                         break;
2713                 }
2714                 if (key.objectid == chunk_objectid &&
2715                     key.offset == chunk_offset) {
2716                         memmove(ptr, ptr + len, array_size - (cur + len));
2717                         array_size -= len;
2718                         btrfs_set_super_sys_array_size(super_copy, array_size);
2719                 } else {
2720                         ptr += len;
2721                         cur += len;
2722                 }
2723         }
2724         unlock_chunks(root);
2725         return ret;
2726 }
2727
2728 int btrfs_remove_chunk(struct btrfs_trans_handle *trans,
2729                        struct btrfs_root *root, u64 chunk_offset)
2730 {
2731         struct extent_map_tree *em_tree;
2732         struct extent_map *em;
2733         struct btrfs_root *extent_root = root->fs_info->extent_root;
2734         struct map_lookup *map;
2735         u64 dev_extent_len = 0;
2736         u64 chunk_objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2737         int i, ret = 0;
2738
2739         /* Just in case */
2740         root = root->fs_info->chunk_root;
2741         em_tree = &root->fs_info->mapping_tree.map_tree;
2742
2743         read_lock(&em_tree->lock);
2744         em = lookup_extent_mapping(em_tree, chunk_offset, 1);
2745         read_unlock(&em_tree->lock);
2746
2747         if (!em || em->start > chunk_offset ||
2748             em->start + em->len < chunk_offset) {
2749                 /*
2750                  * This is a logic error, but we don't want to just rely on the
2751                  * user having built with ASSERT enabled, so if ASSERT doens't
2752                  * do anything we still error out.
2753                  */
2754                 ASSERT(0);
2755                 if (em)
2756                         free_extent_map(em);
2757                 return -EINVAL;
2758         }
2759         map = em->map_lookup;
2760         lock_chunks(root->fs_info->chunk_root);
2761         check_system_chunk(trans, extent_root, map->type);
2762         unlock_chunks(root->fs_info->chunk_root);
2763
2764         for (i = 0; i < map->num_stripes; i++) {
2765                 struct btrfs_device *device = map->stripes[i].dev;
2766                 ret = btrfs_free_dev_extent(trans, device,
2767                                             map->stripes[i].physical,
2768                                             &dev_extent_len);
2769                 if (ret) {
2770                         btrfs_abort_transaction(trans, root, ret);
2771                         goto out;
2772                 }
2773
2774                 if (device->bytes_used > 0) {
2775                         lock_chunks(root);
2776                         btrfs_device_set_bytes_used(device,
2777                                         device->bytes_used - dev_extent_len);
2778                         spin_lock(&root->fs_info->free_chunk_lock);
2779                         root->fs_info->free_chunk_space += dev_extent_len;
2780                         spin_unlock(&root->fs_info->free_chunk_lock);
2781                         btrfs_clear_space_info_full(root->fs_info);
2782                         unlock_chunks(root);
2783                 }
2784
2785                 if (map->stripes[i].dev) {
2786                         ret = btrfs_update_device(trans, map->stripes[i].dev);
2787                         if (ret) {
2788                                 btrfs_abort_transaction(trans, root, ret);
2789                                 goto out;
2790                         }
2791                 }
2792         }
2793         ret = btrfs_free_chunk(trans, root, chunk_objectid, chunk_offset);
2794         if (ret) {
2795                 btrfs_abort_transaction(trans, root, ret);
2796                 goto out;
2797         }
2798
2799         trace_btrfs_chunk_free(root, map, chunk_offset, em->len);
2800
2801         if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2802                 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
2803                 if (ret) {
2804                         btrfs_abort_transaction(trans, root, ret);
2805                         goto out;
2806                 }
2807         }
2808
2809         ret = btrfs_remove_block_group(trans, extent_root, chunk_offset, em);
2810         if (ret) {
2811                 btrfs_abort_transaction(trans, extent_root, ret);
2812                 goto out;
2813         }
2814
2815 out:
2816         /* once for us */
2817         free_extent_map(em);
2818         return ret;
2819 }
2820
2821 static int btrfs_relocate_chunk(struct btrfs_root *root, u64 chunk_offset)
2822 {
2823         struct btrfs_root *extent_root;
2824         struct btrfs_trans_handle *trans;
2825         int ret;
2826
2827         root = root->fs_info->chunk_root;
2828         extent_root = root->fs_info->extent_root;
2829
2830         /*
2831          * Prevent races with automatic removal of unused block groups.
2832          * After we relocate and before we remove the chunk with offset
2833          * chunk_offset, automatic removal of the block group can kick in,
2834          * resulting in a failure when calling btrfs_remove_chunk() below.
2835          *
2836          * Make sure to acquire this mutex before doing a tree search (dev
2837          * or chunk trees) to find chunks. Otherwise the cleaner kthread might
2838          * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after
2839          * we release the path used to search the chunk/dev tree and before
2840          * the current task acquires this mutex and calls us.
2841          */
2842         ASSERT(mutex_is_locked(&root->fs_info->delete_unused_bgs_mutex));
2843
2844         ret = btrfs_can_relocate(extent_root, chunk_offset);
2845         if (ret)
2846                 return -ENOSPC;
2847
2848         /* step one, relocate all the extents inside this chunk */
2849         btrfs_scrub_pause(root);
2850         ret = btrfs_relocate_block_group(extent_root, chunk_offset);
2851         btrfs_scrub_continue(root);
2852         if (ret)
2853                 return ret;
2854
2855         trans = btrfs_start_trans_remove_block_group(root->fs_info,
2856                                                      chunk_offset);
2857         if (IS_ERR(trans)) {
2858                 ret = PTR_ERR(trans);
2859                 btrfs_std_error(root->fs_info, ret, NULL);
2860                 return ret;
2861         }
2862
2863         /*
2864          * step two, delete the device extents and the
2865          * chunk tree entries
2866          */
2867         ret = btrfs_remove_chunk(trans, root, chunk_offset);
2868         btrfs_end_transaction(trans, root);
2869         return ret;
2870 }
2871
2872 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
2873 {
2874         struct btrfs_root *chunk_root = root->fs_info->chunk_root;
2875         struct btrfs_path *path;
2876         struct extent_buffer *leaf;
2877         struct btrfs_chunk *chunk;
2878         struct btrfs_key key;
2879         struct btrfs_key found_key;
2880         u64 chunk_type;
2881         bool retried = false;
2882         int failed = 0;
2883         int ret;
2884
2885         path = btrfs_alloc_path();
2886         if (!path)
2887                 return -ENOMEM;
2888
2889 again:
2890         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2891         key.offset = (u64)-1;
2892         key.type = BTRFS_CHUNK_ITEM_KEY;
2893
2894         while (1) {
2895                 mutex_lock(&root->fs_info->delete_unused_bgs_mutex);
2896                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2897                 if (ret < 0) {
2898                         mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
2899                         goto error;
2900                 }
2901                 BUG_ON(ret == 0); /* Corruption */
2902
2903                 ret = btrfs_previous_item(chunk_root, path, key.objectid,
2904                                           key.type);
2905                 if (ret)
2906                         mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
2907                 if (ret < 0)
2908                         goto error;
2909                 if (ret > 0)
2910                         break;
2911
2912                 leaf = path->nodes[0];
2913                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2914
2915                 chunk = btrfs_item_ptr(leaf, path->slots[0],
2916                                        struct btrfs_chunk);
2917                 chunk_type = btrfs_chunk_type(leaf, chunk);
2918                 btrfs_release_path(path);
2919
2920                 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
2921                         ret = btrfs_relocate_chunk(chunk_root,
2922                                                    found_key.offset);
2923                         if (ret == -ENOSPC)
2924                                 failed++;
2925                         else
2926                                 BUG_ON(ret);
2927                 }
2928                 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
2929
2930                 if (found_key.offset == 0)
2931                         break;
2932                 key.offset = found_key.offset - 1;
2933         }
2934         ret = 0;
2935         if (failed && !retried) {
2936                 failed = 0;
2937                 retried = true;
2938                 goto again;
2939         } else if (WARN_ON(failed && retried)) {
2940                 ret = -ENOSPC;
2941         }
2942 error:
2943         btrfs_free_path(path);
2944         return ret;
2945 }
2946
2947 static int insert_balance_item(struct btrfs_root *root,
2948                                struct btrfs_balance_control *bctl)
2949 {
2950         struct btrfs_trans_handle *trans;
2951         struct btrfs_balance_item *item;
2952         struct btrfs_disk_balance_args disk_bargs;
2953         struct btrfs_path *path;
2954         struct extent_buffer *leaf;
2955         struct btrfs_key key;
2956         int ret, err;
2957
2958         path = btrfs_alloc_path();
2959         if (!path)
2960                 return -ENOMEM;
2961
2962         trans = btrfs_start_transaction(root, 0);
2963         if (IS_ERR(trans)) {
2964                 btrfs_free_path(path);
2965                 return PTR_ERR(trans);
2966         }
2967
2968         key.objectid = BTRFS_BALANCE_OBJECTID;
2969         key.type = BTRFS_BALANCE_ITEM_KEY;
2970         key.offset = 0;
2971
2972         ret = btrfs_insert_empty_item(trans, root, path, &key,
2973                                       sizeof(*item));
2974         if (ret)
2975                 goto out;
2976
2977         leaf = path->nodes[0];
2978         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
2979
2980         memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
2981
2982         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
2983         btrfs_set_balance_data(leaf, item, &disk_bargs);
2984         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
2985         btrfs_set_balance_meta(leaf, item, &disk_bargs);
2986         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
2987         btrfs_set_balance_sys(leaf, item, &disk_bargs);
2988
2989         btrfs_set_balance_flags(leaf, item, bctl->flags);
2990
2991         btrfs_mark_buffer_dirty(leaf);
2992 out:
2993         btrfs_free_path(path);
2994         err = btrfs_commit_transaction(trans, root);
2995         if (err && !ret)
2996                 ret = err;
2997         return ret;
2998 }
2999
3000 static int del_balance_item(struct btrfs_root *root)
3001 {
3002         struct btrfs_trans_handle *trans;
3003         struct btrfs_path *path;
3004         struct btrfs_key key;
3005         int ret, err;
3006
3007         path = btrfs_alloc_path();
3008         if (!path)
3009                 return -ENOMEM;
3010
3011         trans = btrfs_start_transaction(root, 0);
3012         if (IS_ERR(trans)) {
3013                 btrfs_free_path(path);
3014                 return PTR_ERR(trans);
3015         }
3016
3017         key.objectid = BTRFS_BALANCE_OBJECTID;
3018         key.type = BTRFS_BALANCE_ITEM_KEY;
3019         key.offset = 0;
3020
3021         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3022         if (ret < 0)
3023                 goto out;
3024         if (ret > 0) {
3025                 ret = -ENOENT;
3026                 goto out;
3027         }
3028
3029         ret = btrfs_del_item(trans, root, path);
3030 out:
3031         btrfs_free_path(path);
3032         err = btrfs_commit_transaction(trans, root);
3033         if (err && !ret)
3034                 ret = err;
3035         return ret;
3036 }
3037
3038 /*
3039  * This is a heuristic used to reduce the number of chunks balanced on
3040  * resume after balance was interrupted.
3041  */
3042 static void update_balance_args(struct btrfs_balance_control *bctl)
3043 {
3044         /*
3045          * Turn on soft mode for chunk types that were being converted.
3046          */
3047         if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
3048                 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
3049         if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
3050                 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
3051         if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
3052                 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
3053
3054         /*
3055          * Turn on usage filter if is not already used.  The idea is
3056          * that chunks that we have already balanced should be
3057          * reasonably full.  Don't do it for chunks that are being
3058          * converted - that will keep us from relocating unconverted
3059          * (albeit full) chunks.
3060          */
3061         if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3062             !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3063             !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3064                 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
3065                 bctl->data.usage = 90;
3066         }
3067         if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3068             !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3069             !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3070                 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
3071                 bctl->sys.usage = 90;
3072         }
3073         if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3074             !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3075             !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3076                 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
3077                 bctl->meta.usage = 90;
3078         }
3079 }
3080
3081 /*
3082  * Should be called with both balance and volume mutexes held to
3083  * serialize other volume operations (add_dev/rm_dev/resize) with
3084  * restriper.  Same goes for unset_balance_control.
3085  */
3086 static void set_balance_control(struct btrfs_balance_control *bctl)
3087 {
3088         struct btrfs_fs_info *fs_info = bctl->fs_info;
3089
3090         BUG_ON(fs_info->balance_ctl);
3091
3092         spin_lock(&fs_info->balance_lock);
3093         fs_info->balance_ctl = bctl;
3094         spin_unlock(&fs_info->balance_lock);
3095 }
3096
3097 static void unset_balance_control(struct btrfs_fs_info *fs_info)
3098 {
3099         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3100
3101         BUG_ON(!fs_info->balance_ctl);
3102
3103         spin_lock(&fs_info->balance_lock);
3104         fs_info->balance_ctl = NULL;
3105         spin_unlock(&fs_info->balance_lock);
3106
3107         kfree(bctl);
3108 }
3109
3110 /*
3111  * Balance filters.  Return 1 if chunk should be filtered out
3112  * (should not be balanced).
3113  */
3114 static int chunk_profiles_filter(u64 chunk_type,
3115                                  struct btrfs_balance_args *bargs)
3116 {
3117         chunk_type = chunk_to_extended(chunk_type) &
3118                                 BTRFS_EXTENDED_PROFILE_MASK;
3119
3120         if (bargs->profiles & chunk_type)
3121                 return 0;
3122
3123         return 1;
3124 }
3125
3126 static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
3127                               struct btrfs_balance_args *bargs)
3128 {
3129         struct btrfs_block_group_cache *cache;
3130         u64 chunk_used;
3131         u64 user_thresh_min;
3132         u64 user_thresh_max;
3133         int ret = 1;
3134
3135         cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3136         chunk_used = btrfs_block_group_used(&cache->item);
3137
3138         if (bargs->usage_min == 0)
3139                 user_thresh_min = 0;
3140         else
3141                 user_thresh_min = div_factor_fine(cache->key.offset,
3142                                         bargs->usage_min);
3143
3144         if (bargs->usage_max == 0)
3145                 user_thresh_max = 1;
3146         else if (bargs->usage_max > 100)
3147                 user_thresh_max = cache->key.offset;
3148         else
3149                 user_thresh_max = div_factor_fine(cache->key.offset,
3150                                         bargs->usage_max);
3151
3152         if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max)
3153                 ret = 0;
3154
3155         btrfs_put_block_group(cache);
3156         return ret;
3157 }
3158
3159 static int chunk_usage_filter(struct btrfs_fs_info *fs_info,
3160                 u64 chunk_offset, struct btrfs_balance_args *bargs)
3161 {
3162         struct btrfs_block_group_cache *cache;
3163         u64 chunk_used, user_thresh;
3164         int ret = 1;
3165
3166         cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3167         chunk_used = btrfs_block_group_used(&cache->item);
3168
3169         if (bargs->usage_min == 0)
3170                 user_thresh = 1;
3171         else if (bargs->usage > 100)
3172                 user_thresh = cache->key.offset;
3173         else
3174                 user_thresh = div_factor_fine(cache->key.offset,
3175                                               bargs->usage);
3176
3177         if (chunk_used < user_thresh)
3178                 ret = 0;
3179
3180         btrfs_put_block_group(cache);
3181         return ret;
3182 }
3183
3184 static int chunk_devid_filter(struct extent_buffer *leaf,
3185                               struct btrfs_chunk *chunk,
3186                               struct btrfs_balance_args *bargs)
3187 {
3188         struct btrfs_stripe *stripe;
3189         int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3190         int i;
3191
3192         for (i = 0; i < num_stripes; i++) {
3193                 stripe = btrfs_stripe_nr(chunk, i);
3194                 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
3195                         return 0;
3196         }
3197
3198         return 1;
3199 }
3200
3201 /* [pstart, pend) */
3202 static int chunk_drange_filter(struct extent_buffer *leaf,
3203                                struct btrfs_chunk *chunk,
3204                                u64 chunk_offset,
3205                                struct btrfs_balance_args *bargs)
3206 {
3207         struct btrfs_stripe *stripe;
3208         int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3209         u64 stripe_offset;
3210         u64 stripe_length;
3211         int factor;
3212         int i;
3213
3214         if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
3215                 return 0;
3216
3217         if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP |
3218              BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) {
3219                 factor = num_stripes / 2;
3220         } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) {
3221                 factor = num_stripes - 1;
3222         } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) {
3223                 factor = num_stripes - 2;
3224         } else {
3225                 factor = num_stripes;
3226         }
3227
3228         for (i = 0; i < num_stripes; i++) {
3229                 stripe = btrfs_stripe_nr(chunk, i);
3230                 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
3231                         continue;
3232
3233                 stripe_offset = btrfs_stripe_offset(leaf, stripe);
3234                 stripe_length = btrfs_chunk_length(leaf, chunk);
3235                 stripe_length = div_u64(stripe_length, factor);
3236
3237                 if (stripe_offset < bargs->pend &&
3238                     stripe_offset + stripe_length > bargs->pstart)
3239                         return 0;
3240         }
3241
3242         return 1;
3243 }
3244
3245 /* [vstart, vend) */
3246 static int chunk_vrange_filter(struct extent_buffer *leaf,
3247                                struct btrfs_chunk *chunk,
3248                                u64 chunk_offset,
3249                                struct btrfs_balance_args *bargs)
3250 {
3251         if (chunk_offset < bargs->vend &&
3252             chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
3253                 /* at least part of the chunk is inside this vrange */
3254                 return 0;
3255
3256         return 1;
3257 }
3258
3259 static int chunk_stripes_range_filter(struct extent_buffer *leaf,
3260                                struct btrfs_chunk *chunk,
3261                                struct btrfs_balance_args *bargs)
3262 {
3263         int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3264
3265         if (bargs->stripes_min <= num_stripes
3266                         && num_stripes <= bargs->stripes_max)
3267                 return 0;
3268
3269         return 1;
3270 }
3271
3272 static int chunk_soft_convert_filter(u64 chunk_type,
3273                                      struct btrfs_balance_args *bargs)
3274 {
3275         if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
3276                 return 0;
3277
3278         chunk_type = chunk_to_extended(chunk_type) &
3279                                 BTRFS_EXTENDED_PROFILE_MASK;
3280
3281         if (bargs->target == chunk_type)
3282                 return 1;
3283
3284         return 0;
3285 }
3286
3287 static int should_balance_chunk(struct btrfs_root *root,
3288                                 struct extent_buffer *leaf,
3289                                 struct btrfs_chunk *chunk, u64 chunk_offset)
3290 {
3291         struct btrfs_balance_control *bctl = root->fs_info->balance_ctl;
3292         struct btrfs_balance_args *bargs = NULL;
3293         u64 chunk_type = btrfs_chunk_type(leaf, chunk);
3294
3295         /* type filter */
3296         if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
3297               (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
3298                 return 0;
3299         }
3300
3301         if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3302                 bargs = &bctl->data;
3303         else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3304                 bargs = &bctl->sys;
3305         else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3306                 bargs = &bctl->meta;
3307
3308         /* profiles filter */
3309         if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
3310             chunk_profiles_filter(chunk_type, bargs)) {
3311                 return 0;
3312         }
3313
3314         /* usage filter */
3315         if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
3316             chunk_usage_filter(bctl->fs_info, chunk_offset, bargs)) {
3317                 return 0;
3318         } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3319             chunk_usage_range_filter(bctl->fs_info, chunk_offset, bargs)) {
3320                 return 0;
3321         }
3322
3323         /* devid filter */
3324         if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
3325             chunk_devid_filter(leaf, chunk, bargs)) {
3326                 return 0;
3327         }
3328
3329         /* drange filter, makes sense only with devid filter */
3330         if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
3331             chunk_drange_filter(leaf, chunk, chunk_offset, bargs)) {
3332                 return 0;
3333         }
3334
3335         /* vrange filter */
3336         if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
3337             chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
3338                 return 0;
3339         }
3340
3341         /* stripes filter */
3342         if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) &&
3343             chunk_stripes_range_filter(leaf, chunk, bargs)) {
3344                 return 0;
3345         }
3346
3347         /* soft profile changing mode */
3348         if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
3349             chunk_soft_convert_filter(chunk_type, bargs)) {
3350                 return 0;
3351         }
3352
3353         /*
3354          * limited by count, must be the last filter
3355          */
3356         if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) {
3357                 if (bargs->limit == 0)
3358                         return 0;
3359                 else
3360                         bargs->limit--;
3361         } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) {
3362                 /*
3363                  * Same logic as the 'limit' filter; the minimum cannot be
3364                  * determined here because we do not have the global informatoin
3365                  * about the count of all chunks that satisfy the filters.
3366                  */
3367                 if (bargs->limit_max == 0)
3368                         return 0;
3369                 else
3370                         bargs->limit_max--;
3371         }
3372
3373         return 1;
3374 }
3375
3376 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
3377 {
3378         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3379         struct btrfs_root *chunk_root = fs_info->chunk_root;
3380         struct btrfs_root *dev_root = fs_info->dev_root;
3381         struct list_head *devices;
3382         struct btrfs_device *device;
3383         u64 old_size;
3384         u64 size_to_free;
3385         u64 chunk_type;
3386         struct btrfs_chunk *chunk;
3387         struct btrfs_path *path;
3388         struct btrfs_key key;
3389         struct btrfs_key found_key;
3390         struct btrfs_trans_handle *trans;
3391         struct extent_buffer *leaf;
3392         int slot;
3393         int ret;
3394         int enospc_errors = 0;
3395         bool counting = true;
3396         /* The single value limit and min/max limits use the same bytes in the */
3397         u64 limit_data = bctl->data.limit;
3398         u64 limit_meta = bctl->meta.limit;
3399         u64 limit_sys = bctl->sys.limit;
3400         u32 count_data = 0;
3401         u32 count_meta = 0;
3402         u32 count_sys = 0;
3403         int chunk_reserved = 0;
3404
3405         /* step one make some room on all the devices */
3406         devices = &fs_info->fs_devices->devices;
3407         list_for_each_entry(device, devices, dev_list) {
3408                 old_size = btrfs_device_get_total_bytes(device);
3409                 size_to_free = div_factor(old_size, 1);
3410                 size_to_free = min_t(u64, size_to_free, SZ_1M);
3411                 if (!device->writeable ||
3412                     btrfs_device_get_total_bytes(device) -
3413                     btrfs_device_get_bytes_used(device) > size_to_free ||
3414                     device->is_tgtdev_for_dev_replace)
3415                         continue;
3416
3417                 ret = btrfs_shrink_device(device, old_size - size_to_free);
3418                 if (ret == -ENOSPC)
3419                         break;
3420                 BUG_ON(ret);
3421
3422                 trans = btrfs_start_transaction(dev_root, 0);
3423                 BUG_ON(IS_ERR(trans));
3424
3425                 ret = btrfs_grow_device(trans, device, old_size);
3426                 BUG_ON(ret);
3427
3428                 btrfs_end_transaction(trans, dev_root);
3429         }
3430
3431         /* step two, relocate all the chunks */
3432         path = btrfs_alloc_path();
3433         if (!path) {
3434                 ret = -ENOMEM;
3435                 goto error;
3436         }
3437
3438         /* zero out stat counters */
3439         spin_lock(&fs_info->balance_lock);
3440         memset(&bctl->stat, 0, sizeof(bctl->stat));
3441         spin_unlock(&fs_info->balance_lock);
3442 again:
3443         if (!counting) {
3444                 /*
3445                  * The single value limit and min/max limits use the same bytes
3446                  * in the
3447                  */
3448                 bctl->data.limit = limit_data;
3449                 bctl->meta.limit = limit_meta;
3450                 bctl->sys.limit = limit_sys;
3451         }
3452         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3453         key.offset = (u64)-1;
3454         key.type = BTRFS_CHUNK_ITEM_KEY;
3455
3456         while (1) {
3457                 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
3458                     atomic_read(&fs_info->balance_cancel_req)) {
3459                         ret = -ECANCELED;
3460                         goto error;
3461                 }
3462
3463                 mutex_lock(&fs_info->delete_unused_bgs_mutex);
3464                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3465                 if (ret < 0) {
3466                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3467                         goto error;
3468                 }
3469
3470                 /*
3471                  * this shouldn't happen, it means the last relocate
3472                  * failed
3473                  */
3474                 if (ret == 0)
3475                         BUG(); /* FIXME break ? */
3476
3477                 ret = btrfs_previous_item(chunk_root, path, 0,
3478                                           BTRFS_CHUNK_ITEM_KEY);
3479                 if (ret) {
3480                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3481                         ret = 0;
3482                         break;
3483                 }
3484
3485                 leaf = path->nodes[0];
3486                 slot = path->slots[0];
3487                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3488
3489                 if (found_key.objectid != key.objectid) {
3490                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3491                         break;
3492                 }
3493
3494                 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3495                 chunk_type = btrfs_chunk_type(leaf, chunk);
3496
3497                 if (!counting) {
3498                         spin_lock(&fs_info->balance_lock);
3499                         bctl->stat.considered++;
3500                         spin_unlock(&fs_info->balance_lock);
3501                 }
3502
3503                 ret = should_balance_chunk(chunk_root, leaf, chunk,
3504                                            found_key.offset);
3505
3506                 btrfs_release_path(path);
3507                 if (!ret) {
3508                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3509                         goto loop;
3510                 }
3511
3512                 if (counting) {
3513                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3514                         spin_lock(&fs_info->balance_lock);
3515                         bctl->stat.expected++;
3516                         spin_unlock(&fs_info->balance_lock);
3517
3518                         if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3519                                 count_data++;
3520                         else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3521                                 count_sys++;
3522                         else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3523                                 count_meta++;
3524
3525                         goto loop;
3526                 }
3527
3528                 /*
3529                  * Apply limit_min filter, no need to check if the LIMITS
3530                  * filter is used, limit_min is 0 by default
3531                  */
3532                 if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
3533                                         count_data < bctl->data.limit_min)
3534                                 || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) &&
3535                                         count_meta < bctl->meta.limit_min)
3536                                 || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) &&
3537                                         count_sys < bctl->sys.limit_min)) {
3538                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3539                         goto loop;
3540                 }
3541
3542                 if ((chunk_type & BTRFS_BLOCK_GROUP_DATA) && !chunk_reserved) {
3543                         trans = btrfs_start_transaction(chunk_root, 0);
3544                         if (IS_ERR(trans)) {
3545                                 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3546                                 ret = PTR_ERR(trans);
3547                                 goto error;
3548                         }
3549
3550                         ret = btrfs_force_chunk_alloc(trans, chunk_root,
3551                                                       BTRFS_BLOCK_GROUP_DATA);
3552                         btrfs_end_transaction(trans, chunk_root);
3553                         if (ret < 0) {
3554                                 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3555                                 goto error;
3556                         }
3557                         chunk_reserved = 1;
3558                 }
3559
3560                 ret = btrfs_relocate_chunk(chunk_root,
3561                                            found_key.offset);
3562                 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3563                 if (ret && ret != -ENOSPC)
3564                         goto error;
3565                 if (ret == -ENOSPC) {
3566                         enospc_errors++;
3567                 } else {
3568                         spin_lock(&fs_info->balance_lock);
3569                         bctl->stat.completed++;
3570                         spin_unlock(&fs_info->balance_lock);
3571                 }
3572 loop:
3573                 if (found_key.offset == 0)
3574                         break;
3575                 key.offset = found_key.offset - 1;
3576         }
3577
3578         if (counting) {
3579                 btrfs_release_path(path);
3580                 counting = false;
3581                 goto again;
3582         }
3583 error:
3584         btrfs_free_path(path);
3585         if (enospc_errors) {
3586                 btrfs_info(fs_info, "%d enospc errors during balance",
3587                        enospc_errors);
3588                 if (!ret)
3589                         ret = -ENOSPC;
3590         }
3591
3592         return ret;
3593 }
3594
3595 /**
3596  * alloc_profile_is_valid - see if a given profile is valid and reduced
3597  * @flags: profile to validate
3598  * @extended: if true @flags is treated as an extended profile
3599  */
3600 static int alloc_profile_is_valid(u64 flags, int extended)
3601 {
3602         u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
3603                                BTRFS_BLOCK_GROUP_PROFILE_MASK);
3604
3605         flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
3606
3607         /* 1) check that all other bits are zeroed */
3608         if (flags & ~mask)
3609                 return 0;
3610
3611         /* 2) see if profile is reduced */
3612         if (flags == 0)
3613                 return !extended; /* "0" is valid for usual profiles */
3614
3615         /* true if exactly one bit set */
3616         return (flags & (flags - 1)) == 0;
3617 }
3618
3619 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
3620 {
3621         /* cancel requested || normal exit path */
3622         return atomic_read(&fs_info->balance_cancel_req) ||
3623                 (atomic_read(&fs_info->balance_pause_req) == 0 &&
3624                  atomic_read(&fs_info->balance_cancel_req) == 0);
3625 }
3626
3627 static void __cancel_balance(struct btrfs_fs_info *fs_info)
3628 {
3629         int ret;
3630
3631         unset_balance_control(fs_info);
3632         ret = del_balance_item(fs_info->tree_root);
3633         if (ret)
3634                 btrfs_std_error(fs_info, ret, NULL);
3635
3636         atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3637 }
3638
3639 /* Non-zero return value signifies invalidity */
3640 static inline int validate_convert_profile(struct btrfs_balance_args *bctl_arg,
3641                 u64 allowed)
3642 {
3643         return ((bctl_arg->flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3644                 (!alloc_profile_is_valid(bctl_arg->target, 1) ||
3645                  (bctl_arg->target & ~allowed)));
3646 }
3647
3648 /*
3649  * Should be called with both balance and volume mutexes held
3650  */
3651 int btrfs_balance(struct btrfs_balance_control *bctl,
3652                   struct btrfs_ioctl_balance_args *bargs)
3653 {
3654         struct btrfs_fs_info *fs_info = bctl->fs_info;
3655         u64 allowed;
3656         int mixed = 0;
3657         int ret;
3658         u64 num_devices;
3659         unsigned seq;
3660
3661         if (btrfs_fs_closing(fs_info) ||
3662             atomic_read(&fs_info->balance_pause_req) ||
3663             atomic_read(&fs_info->balance_cancel_req)) {
3664                 ret = -EINVAL;
3665                 goto out;
3666         }
3667
3668         allowed = btrfs_super_incompat_flags(fs_info->super_copy);
3669         if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
3670                 mixed = 1;
3671
3672         /*
3673          * In case of mixed groups both data and meta should be picked,
3674          * and identical options should be given for both of them.
3675          */
3676         allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
3677         if (mixed && (bctl->flags & allowed)) {
3678                 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
3679                     !(bctl->flags & BTRFS_BALANCE_METADATA) ||
3680                     memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
3681                         btrfs_err(fs_info, "with mixed groups data and "
3682                                    "metadata balance options must be the same");
3683                         ret = -EINVAL;
3684                         goto out;
3685                 }
3686         }
3687
3688         num_devices = fs_info->fs_devices->num_devices;
3689         btrfs_dev_replace_lock(&fs_info->dev_replace);
3690         if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
3691                 BUG_ON(num_devices < 1);
3692                 num_devices--;
3693         }
3694         btrfs_dev_replace_unlock(&fs_info->dev_replace);
3695         allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE;
3696         if (num_devices == 1)
3697                 allowed |= BTRFS_BLOCK_GROUP_DUP;
3698         else if (num_devices > 1)
3699                 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1);
3700         if (num_devices > 2)
3701                 allowed |= BTRFS_BLOCK_GROUP_RAID5;
3702         if (num_devices > 3)
3703                 allowed |= (BTRFS_BLOCK_GROUP_RAID10 |
3704                             BTRFS_BLOCK_GROUP_RAID6);
3705         if (validate_convert_profile(&bctl->data, allowed)) {
3706                 btrfs_err(fs_info, "unable to start balance with target "
3707                            "data profile %llu",
3708                        bctl->data.target);
3709                 ret = -EINVAL;
3710                 goto out;
3711         }
3712         if (validate_convert_profile(&bctl->meta, allowed)) {
3713                 btrfs_err(fs_info,
3714                            "unable to start balance with target metadata profile %llu",
3715                        bctl->meta.target);
3716                 ret = -EINVAL;
3717                 goto out;
3718         }
3719         if (validate_convert_profile(&bctl->sys, allowed)) {
3720                 btrfs_err(fs_info,
3721                            "unable to start balance with target system profile %llu",
3722                        bctl->sys.target);
3723                 ret = -EINVAL;
3724                 goto out;
3725         }
3726
3727         /* allow to reduce meta or sys integrity only if force set */
3728         allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3729                         BTRFS_BLOCK_GROUP_RAID10 |
3730                         BTRFS_BLOCK_GROUP_RAID5 |
3731                         BTRFS_BLOCK_GROUP_RAID6;
3732         do {
3733                 seq = read_seqbegin(&fs_info->profiles_lock);
3734
3735                 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3736                      (fs_info->avail_system_alloc_bits & allowed) &&
3737                      !(bctl->sys.target & allowed)) ||
3738                     ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3739                      (fs_info->avail_metadata_alloc_bits & allowed) &&
3740                      !(bctl->meta.target & allowed))) {
3741                         if (bctl->flags & BTRFS_BALANCE_FORCE) {
3742                                 btrfs_info(fs_info, "force reducing metadata integrity");
3743                         } else {
3744                                 btrfs_err(fs_info, "balance will reduce metadata "
3745                                            "integrity, use force if you want this");
3746                                 ret = -EINVAL;
3747                                 goto out;
3748                         }
3749                 }
3750         } while (read_seqretry(&fs_info->profiles_lock, seq));
3751
3752         if (btrfs_get_num_tolerated_disk_barrier_failures(bctl->meta.target) <
3753                 btrfs_get_num_tolerated_disk_barrier_failures(bctl->data.target)) {
3754                 btrfs_warn(fs_info,
3755         "metadata profile 0x%llx has lower redundancy than data profile 0x%llx",
3756                         bctl->meta.target, bctl->data.target);
3757         }
3758
3759         if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3760                 fs_info->num_tolerated_disk_barrier_failures = min(
3761                         btrfs_calc_num_tolerated_disk_barrier_failures(fs_info),
3762                         btrfs_get_num_tolerated_disk_barrier_failures(
3763                                 bctl->sys.target));
3764         }
3765
3766         ret = insert_balance_item(fs_info->tree_root, bctl);
3767         if (ret && ret != -EEXIST)
3768                 goto out;
3769
3770         if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
3771                 BUG_ON(ret == -EEXIST);
3772                 set_balance_control(bctl);
3773         } else {
3774                 BUG_ON(ret != -EEXIST);
3775                 spin_lock(&fs_info->balance_lock);
3776                 update_balance_args(bctl);
3777                 spin_unlock(&fs_info->balance_lock);
3778         }
3779
3780         atomic_inc(&fs_info->balance_running);
3781         mutex_unlock(&fs_info->balance_mutex);
3782
3783         ret = __btrfs_balance(fs_info);
3784
3785         mutex_lock(&fs_info->balance_mutex);
3786         atomic_dec(&fs_info->balance_running);
3787
3788         if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3789                 fs_info->num_tolerated_disk_barrier_failures =
3790                         btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3791         }
3792
3793         if (bargs) {
3794                 memset(bargs, 0, sizeof(*bargs));
3795                 update_ioctl_balance_args(fs_info, 0, bargs);
3796         }
3797
3798         if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
3799             balance_need_close(fs_info)) {
3800                 __cancel_balance(fs_info);
3801         }
3802
3803         wake_up(&fs_info->balance_wait_q);
3804
3805         return ret;
3806 out:
3807         if (bctl->flags & BTRFS_BALANCE_RESUME)
3808                 __cancel_balance(fs_info);
3809         else {
3810                 kfree(bctl);
3811                 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3812         }
3813         return ret;
3814 }
3815
3816 static int balance_kthread(void *data)
3817 {
3818         struct btrfs_fs_info *fs_info = data;
3819         int ret = 0;
3820
3821         mutex_lock(&fs_info->volume_mutex);
3822         mutex_lock(&fs_info->balance_mutex);
3823
3824         if (fs_info->balance_ctl) {
3825                 btrfs_info(fs_info, "continuing balance");
3826                 ret = btrfs_balance(fs_info->balance_ctl, NULL);
3827         }
3828
3829         mutex_unlock(&fs_info->balance_mutex);
3830         mutex_unlock(&fs_info->volume_mutex);
3831
3832         return ret;
3833 }
3834
3835 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
3836 {
3837         struct task_struct *tsk;
3838
3839         spin_lock(&fs_info->balance_lock);
3840         if (!fs_info->balance_ctl) {
3841                 spin_unlock(&fs_info->balance_lock);
3842                 return 0;
3843         }
3844         spin_unlock(&fs_info->balance_lock);
3845
3846         if (btrfs_test_opt(fs_info->tree_root, SKIP_BALANCE)) {
3847                 btrfs_info(fs_info, "force skipping balance");
3848                 return 0;
3849         }
3850
3851         tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
3852         return PTR_ERR_OR_ZERO(tsk);
3853 }
3854
3855 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
3856 {
3857         struct btrfs_balance_control *bctl;
3858         struct btrfs_balance_item *item;
3859         struct btrfs_disk_balance_args disk_bargs;
3860         struct btrfs_path *path;
3861         struct extent_buffer *leaf;
3862         struct btrfs_key key;
3863         int ret;
3864
3865         path = btrfs_alloc_path();
3866         if (!path)
3867                 return -ENOMEM;
3868
3869         key.objectid = BTRFS_BALANCE_OBJECTID;
3870         key.type = BTRFS_BALANCE_ITEM_KEY;
3871         key.offset = 0;
3872
3873         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
3874         if (ret < 0)
3875                 goto out;
3876         if (ret > 0) { /* ret = -ENOENT; */
3877                 ret = 0;
3878                 goto out;
3879         }
3880
3881         bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
3882         if (!bctl) {
3883                 ret = -ENOMEM;
3884                 goto out;
3885         }
3886
3887         leaf = path->nodes[0];
3888         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3889
3890         bctl->fs_info = fs_info;
3891         bctl->flags = btrfs_balance_flags(leaf, item);
3892         bctl->flags |= BTRFS_BALANCE_RESUME;
3893
3894         btrfs_balance_data(leaf, item, &disk_bargs);
3895         btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
3896         btrfs_balance_meta(leaf, item, &disk_bargs);
3897         btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
3898         btrfs_balance_sys(leaf, item, &disk_bargs);
3899         btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
3900
3901         WARN_ON(atomic_xchg(&fs_info->mutually_exclusive_operation_running, 1));
3902
3903         mutex_lock(&fs_info->volume_mutex);
3904         mutex_lock(&fs_info->balance_mutex);
3905
3906         set_balance_control(bctl);
3907
3908         mutex_unlock(&fs_info->balance_mutex);
3909         mutex_unlock(&fs_info->volume_mutex);
3910 out:
3911         btrfs_free_path(path);
3912         return ret;
3913 }
3914
3915 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
3916 {
3917         int ret = 0;
3918
3919         mutex_lock(&fs_info->balance_mutex);
3920         if (!fs_info->balance_ctl) {
3921                 mutex_unlock(&fs_info->balance_mutex);
3922                 return -ENOTCONN;
3923         }
3924
3925         if (atomic_read(&fs_info->balance_running)) {
3926                 atomic_inc(&fs_info->balance_pause_req);
3927                 mutex_unlock(&fs_info->balance_mutex);
3928
3929                 wait_event(fs_info->balance_wait_q,
3930                            atomic_read(&fs_info->balance_running) == 0);
3931
3932                 mutex_lock(&fs_info->balance_mutex);
3933                 /* we are good with balance_ctl ripped off from under us */
3934                 BUG_ON(atomic_read(&fs_info->balance_running));
3935                 atomic_dec(&fs_info->balance_pause_req);
3936         } else {
3937                 ret = -ENOTCONN;
3938         }
3939
3940         mutex_unlock(&fs_info->balance_mutex);
3941         return ret;
3942 }
3943
3944 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
3945 {
3946         if (fs_info->sb->s_flags & MS_RDONLY)
3947                 return -EROFS;
3948
3949         mutex_lock(&fs_info->balance_mutex);
3950         if (!fs_info->balance_ctl) {
3951                 mutex_unlock(&fs_info->balance_mutex);
3952                 return -ENOTCONN;
3953         }
3954
3955         atomic_inc(&fs_info->balance_cancel_req);
3956         /*
3957          * if we are running just wait and return, balance item is
3958          * deleted in btrfs_balance in this case
3959          */
3960         if (atomic_read(&fs_info->balance_running)) {
3961                 mutex_unlock(&fs_info->balance_mutex);
3962                 wait_event(fs_info->balance_wait_q,
3963                            atomic_read(&fs_info->balance_running) == 0);
3964                 mutex_lock(&fs_info->balance_mutex);
3965         } else {
3966                 /* __cancel_balance needs volume_mutex */
3967                 mutex_unlock(&fs_info->balance_mutex);
3968                 mutex_lock(&fs_info->volume_mutex);
3969                 mutex_lock(&fs_info->balance_mutex);
3970
3971                 if (fs_info->balance_ctl)
3972                         __cancel_balance(fs_info);
3973
3974                 mutex_unlock(&fs_info->volume_mutex);
3975         }
3976
3977         BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running));
3978         atomic_dec(&fs_info->balance_cancel_req);
3979         mutex_unlock(&fs_info->balance_mutex);
3980         return 0;
3981 }
3982
3983 static int btrfs_uuid_scan_kthread(void *data)
3984 {
3985         struct btrfs_fs_info *fs_info = data;
3986         struct btrfs_root *root = fs_info->tree_root;
3987         struct btrfs_key key;
3988         struct btrfs_key max_key;
3989         struct btrfs_path *path = NULL;
3990         int ret = 0;
3991         struct extent_buffer *eb;
3992         int slot;
3993         struct btrfs_root_item root_item;
3994         u32 item_size;
3995         struct btrfs_trans_handle *trans = NULL;
3996
3997         path = btrfs_alloc_path();
3998         if (!path) {
3999                 ret = -ENOMEM;
4000                 goto out;
4001         }
4002
4003         key.objectid = 0;
4004         key.type = BTRFS_ROOT_ITEM_KEY;
4005         key.offset = 0;
4006
4007         max_key.objectid = (u64)-1;
4008         max_key.type = BTRFS_ROOT_ITEM_KEY;
4009         max_key.offset = (u64)-1;
4010
4011         while (1) {
4012                 ret = btrfs_search_forward(root, &key, path, 0);
4013                 if (ret) {
4014                         if (ret > 0)
4015                                 ret = 0;
4016                         break;
4017                 }
4018
4019                 if (key.type != BTRFS_ROOT_ITEM_KEY ||
4020                     (key.objectid < BTRFS_FIRST_FREE_OBJECTID &&
4021                      key.objectid != BTRFS_FS_TREE_OBJECTID) ||
4022                     key.objectid > BTRFS_LAST_FREE_OBJECTID)
4023                         goto skip;
4024
4025                 eb = path->nodes[0];
4026                 slot = path->slots[0];
4027                 item_size = btrfs_item_size_nr(eb, slot);
4028                 if (item_size < sizeof(root_item))
4029                         goto skip;
4030
4031                 read_extent_buffer(eb, &root_item,
4032                                    btrfs_item_ptr_offset(eb, slot),
4033                                    (int)sizeof(root_item));
4034                 if (btrfs_root_refs(&root_item) == 0)
4035                         goto skip;
4036
4037                 if (!btrfs_is_empty_uuid(root_item.uuid) ||
4038                     !btrfs_is_empty_uuid(root_item.received_uuid)) {
4039                         if (trans)
4040                                 goto update_tree;
4041
4042                         btrfs_release_path(path);
4043                         /*
4044                          * 1 - subvol uuid item
4045                          * 1 - received_subvol uuid item
4046                          */
4047                         trans = btrfs_start_transaction(fs_info->uuid_root, 2);
4048                         if (IS_ERR(trans)) {
4049                                 ret = PTR_ERR(trans);
4050                                 break;
4051                         }
4052                         continue;
4053                 } else {
4054                         goto skip;
4055                 }
4056 update_tree:
4057                 if (!btrfs_is_empty_uuid(root_item.uuid)) {
4058                         ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
4059                                                   root_item.uuid,
4060                                                   BTRFS_UUID_KEY_SUBVOL,
4061                                                   key.objectid);
4062                         if (ret < 0) {
4063                                 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4064                                         ret);
4065                                 break;
4066                         }
4067                 }
4068
4069                 if (!btrfs_is_empty_uuid(root_item.received_uuid)) {
4070                         ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
4071                                                   root_item.received_uuid,
4072                                                  BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4073                                                   key.objectid);
4074                         if (ret < 0) {
4075                                 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4076                                         ret);
4077                                 break;
4078                         }
4079                 }
4080
4081 skip:
4082                 if (trans) {
4083                         ret = btrfs_end_transaction(trans, fs_info->uuid_root);
4084                         trans = NULL;
4085                         if (ret)
4086                                 break;
4087                 }
4088
4089                 btrfs_release_path(path);
4090                 if (key.offset < (u64)-1) {
4091                         key.offset++;
4092                 } else if (key.type < BTRFS_ROOT_ITEM_KEY) {
4093                         key.offset = 0;
4094                         key.type = BTRFS_ROOT_ITEM_KEY;
4095                 } else if (key.objectid < (u64)-1) {
4096                         key.offset = 0;
4097                         key.type = BTRFS_ROOT_ITEM_KEY;
4098                         key.objectid++;
4099                 } else {
4100                         break;
4101                 }
4102                 cond_resched();
4103         }
4104
4105 out:
4106         btrfs_free_path(path);
4107         if (trans && !IS_ERR(trans))
4108                 btrfs_end_transaction(trans, fs_info->uuid_root);
4109         if (ret)
4110                 btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret);
4111         else
4112                 fs_info->update_uuid_tree_gen = 1;
4113         up(&fs_info->uuid_tree_rescan_sem);
4114         return 0;
4115 }
4116
4117 /*
4118  * Callback for btrfs_uuid_tree_iterate().
4119  * returns:
4120  * 0    check succeeded, the entry is not outdated.
4121  * < 0  if an error occured.
4122  * > 0  if the check failed, which means the caller shall remove the entry.
4123  */
4124 static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info *fs_info,
4125                                        u8 *uuid, u8 type, u64 subid)
4126 {
4127         struct btrfs_key key;
4128         int ret = 0;
4129         struct btrfs_root *subvol_root;
4130
4131         if (type != BTRFS_UUID_KEY_SUBVOL &&
4132             type != BTRFS_UUID_KEY_RECEIVED_SUBVOL)
4133                 goto out;
4134
4135         key.objectid = subid;
4136         key.type = BTRFS_ROOT_ITEM_KEY;
4137         key.offset = (u64)-1;
4138         subvol_root = btrfs_read_fs_root_no_name(fs_info, &key);
4139         if (IS_ERR(subvol_root)) {
4140                 ret = PTR_ERR(subvol_root);
4141                 if (ret == -ENOENT)
4142                         ret = 1;
4143                 goto out;
4144         }
4145
4146         switch (type) {
4147         case BTRFS_UUID_KEY_SUBVOL:
4148                 if (memcmp(uuid, subvol_root->root_item.uuid, BTRFS_UUID_SIZE))
4149                         ret = 1;
4150                 break;
4151         case BTRFS_UUID_KEY_RECEIVED_SUBVOL:
4152                 if (memcmp(uuid, subvol_root->root_item.received_uuid,
4153                            BTRFS_UUID_SIZE))
4154                         ret = 1;
4155                 break;
4156         }
4157
4158 out:
4159         return ret;
4160 }
4161
4162 static int btrfs_uuid_rescan_kthread(void *data)
4163 {
4164         struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data;
4165         int ret;
4166
4167         /*
4168          * 1st step is to iterate through the existing UUID tree and
4169          * to delete all entries that contain outdated data.
4170          * 2nd step is to add all missing entries to the UUID tree.
4171          */
4172         ret = btrfs_uuid_tree_iterate(fs_info, btrfs_check_uuid_tree_entry);
4173         if (ret < 0) {
4174                 btrfs_warn(fs_info, "iterating uuid_tree failed %d", ret);
4175                 up(&fs_info->uuid_tree_rescan_sem);
4176                 return ret;
4177         }
4178         return btrfs_uuid_scan_kthread(data);
4179 }
4180
4181 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info)
4182 {
4183         struct btrfs_trans_handle *trans;
4184         struct btrfs_root *tree_root = fs_info->tree_root;
4185         struct btrfs_root *uuid_root;
4186         struct task_struct *task;
4187         int ret;
4188
4189         /*
4190          * 1 - root node
4191          * 1 - root item
4192          */
4193         trans = btrfs_start_transaction(tree_root, 2);
4194         if (IS_ERR(trans))
4195                 return PTR_ERR(trans);
4196
4197         uuid_root = btrfs_create_tree(trans, fs_info,
4198                                       BTRFS_UUID_TREE_OBJECTID);
4199         if (IS_ERR(uuid_root)) {
4200                 ret = PTR_ERR(uuid_root);
4201                 btrfs_abort_transaction(trans, tree_root, ret);
4202                 return ret;
4203         }
4204
4205         fs_info->uuid_root = uuid_root;
4206
4207         ret = btrfs_commit_transaction(trans, tree_root);
4208         if (ret)
4209                 return ret;
4210
4211         down(&fs_info->uuid_tree_rescan_sem);
4212         task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid");
4213         if (IS_ERR(task)) {
4214                 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4215                 btrfs_warn(fs_info, "failed to start uuid_scan task");
4216                 up(&fs_info->uuid_tree_rescan_sem);
4217                 return PTR_ERR(task);
4218         }
4219
4220         return 0;
4221 }
4222
4223 int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
4224 {
4225         struct task_struct *task;
4226
4227         down(&fs_info->uuid_tree_rescan_sem);
4228         task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
4229         if (IS_ERR(task)) {
4230                 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4231                 btrfs_warn(fs_info, "failed to start uuid_rescan task");
4232                 up(&fs_info->uuid_tree_rescan_sem);
4233                 return PTR_ERR(task);
4234         }
4235
4236         return 0;
4237 }
4238
4239 /*
4240  * shrinking a device means finding all of the device extents past
4241  * the new size, and then following the back refs to the chunks.
4242  * The chunk relocation code actually frees the device extent
4243  */
4244 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
4245 {
4246         struct btrfs_trans_handle *trans;
4247         struct btrfs_root *root = device->dev_root;
4248         struct btrfs_dev_extent *dev_extent = NULL;
4249         struct btrfs_path *path;
4250         u64 length;
4251         u64 chunk_offset;
4252         int ret;
4253         int slot;
4254         int failed = 0;
4255         bool retried = false;
4256         bool checked_pending_chunks = false;
4257         struct extent_buffer *l;
4258         struct btrfs_key key;
4259         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
4260         u64 old_total = btrfs_super_total_bytes(super_copy);
4261         u64 old_size = btrfs_device_get_total_bytes(device);
4262         u64 diff = old_size - new_size;
4263
4264         if (device->is_tgtdev_for_dev_replace)
4265                 return -EINVAL;
4266
4267         path = btrfs_alloc_path();
4268         if (!path)
4269                 return -ENOMEM;
4270
4271         path->reada = READA_FORWARD;
4272
4273         lock_chunks(root);
4274
4275         btrfs_device_set_total_bytes(device, new_size);
4276         if (device->writeable) {
4277                 device->fs_devices->total_rw_bytes -= diff;
4278                 spin_lock(&root->fs_info->free_chunk_lock);
4279                 root->fs_info->free_chunk_space -= diff;
4280                 spin_unlock(&root->fs_info->free_chunk_lock);
4281         }
4282         unlock_chunks(root);
4283
4284 again:
4285         key.objectid = device->devid;
4286         key.offset = (u64)-1;
4287         key.type = BTRFS_DEV_EXTENT_KEY;
4288
4289         do {
4290                 mutex_lock(&root->fs_info->delete_unused_bgs_mutex);
4291                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4292                 if (ret < 0) {
4293                         mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4294                         goto done;
4295                 }
4296
4297                 ret = btrfs_previous_item(root, path, 0, key.type);
4298                 if (ret)
4299                         mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4300                 if (ret < 0)
4301                         goto done;
4302                 if (ret) {
4303                         ret = 0;
4304                         btrfs_release_path(path);
4305                         break;
4306                 }
4307
4308                 l = path->nodes[0];
4309                 slot = path->slots[0];
4310                 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
4311
4312                 if (key.objectid != device->devid) {
4313                         mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4314                         btrfs_release_path(path);
4315                         break;
4316                 }
4317
4318                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
4319                 length = btrfs_dev_extent_length(l, dev_extent);
4320
4321                 if (key.offset + length <= new_size) {
4322                         mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4323                         btrfs_release_path(path);
4324                         break;
4325                 }
4326
4327                 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
4328                 btrfs_release_path(path);
4329
4330                 ret = btrfs_relocate_chunk(root, chunk_offset);
4331                 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4332                 if (ret && ret != -ENOSPC)
4333                         goto done;
4334                 if (ret == -ENOSPC)
4335                         failed++;
4336         } while (key.offset-- > 0);
4337
4338         if (failed && !retried) {
4339                 failed = 0;
4340                 retried = true;
4341                 goto again;
4342         } else if (failed && retried) {
4343                 ret = -ENOSPC;
4344                 goto done;
4345         }
4346
4347         /* Shrinking succeeded, else we would be at "done". */
4348         trans = btrfs_start_transaction(root, 0);
4349         if (IS_ERR(trans)) {
4350                 ret = PTR_ERR(trans);
4351                 goto done;
4352         }
4353
4354         lock_chunks(root);
4355
4356         /*
4357          * We checked in the above loop all device extents that were already in
4358          * the device tree. However before we have updated the device's
4359          * total_bytes to the new size, we might have had chunk allocations that
4360          * have not complete yet (new block groups attached to transaction
4361          * handles), and therefore their device extents were not yet in the
4362          * device tree and we missed them in the loop above. So if we have any
4363          * pending chunk using a device extent that overlaps the device range
4364          * that we can not use anymore, commit the current transaction and
4365          * repeat the search on the device tree - this way we guarantee we will
4366          * not have chunks using device extents that end beyond 'new_size'.
4367          */
4368         if (!checked_pending_chunks) {
4369                 u64 start = new_size;
4370                 u64 len = old_size - new_size;
4371
4372                 if (contains_pending_extent(trans->transaction, device,
4373                                             &start, len)) {
4374                         unlock_chunks(root);
4375                         checked_pending_chunks = true;
4376                         failed = 0;
4377                         retried = false;
4378                         ret = btrfs_commit_transaction(trans, root);
4379                         if (ret)
4380                                 goto done;
4381                         goto again;
4382                 }
4383         }
4384
4385         btrfs_device_set_disk_total_bytes(device, new_size);
4386         if (list_empty(&device->resized_list))
4387                 list_add_tail(&device->resized_list,
4388                               &root->fs_info->fs_devices->resized_devices);
4389
4390         WARN_ON(diff > old_total);
4391         btrfs_set_super_total_bytes(super_copy, old_total - diff);
4392         unlock_chunks(root);
4393
4394         /* Now btrfs_update_device() will change the on-disk size. */
4395         ret = btrfs_update_device(trans, device);
4396         btrfs_end_transaction(trans, root);
4397 done:
4398         btrfs_free_path(path);
4399         if (ret) {
4400                 lock_chunks(root);
4401                 btrfs_device_set_total_bytes(device, old_size);
4402                 if (device->writeable)
4403                         device->fs_devices->total_rw_bytes += diff;
4404                 spin_lock(&root->fs_info->free_chunk_lock);
4405                 root->fs_info->free_chunk_space += diff;
4406                 spin_unlock(&root->fs_info->free_chunk_lock);
4407                 unlock_chunks(root);
4408         }
4409         return ret;
4410 }
4411
4412 static int btrfs_add_system_chunk(struct btrfs_root *root,
4413                            struct btrfs_key *key,
4414                            struct btrfs_chunk *chunk, int item_size)
4415 {
4416         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
4417         struct btrfs_disk_key disk_key;
4418         u32 array_size;
4419         u8 *ptr;
4420
4421         lock_chunks(root);
4422         array_size = btrfs_super_sys_array_size(super_copy);
4423         if (array_size + item_size + sizeof(disk_key)
4424                         > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
4425                 unlock_chunks(root);
4426                 return -EFBIG;
4427         }
4428
4429         ptr = super_copy->sys_chunk_array + array_size;
4430         btrfs_cpu_key_to_disk(&disk_key, key);
4431         memcpy(ptr, &disk_key, sizeof(disk_key));
4432         ptr += sizeof(disk_key);
4433         memcpy(ptr, chunk, item_size);
4434         item_size += sizeof(disk_key);
4435         btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
4436         unlock_chunks(root);
4437
4438         return 0;
4439 }
4440
4441 /*
4442  * sort the devices in descending order by max_avail, total_avail
4443  */
4444 static int btrfs_cmp_device_info(const void *a, const void *b)
4445 {
4446         const struct btrfs_device_info *di_a = a;
4447         const struct btrfs_device_info *di_b = b;
4448
4449         if (di_a->max_avail > di_b->max_avail)
4450                 return -1;
4451         if (di_a->max_avail < di_b->max_avail)
4452                 return 1;
4453         if (di_a->total_avail > di_b->total_avail)
4454                 return -1;
4455         if (di_a->total_avail < di_b->total_avail)
4456                 return 1;
4457         return 0;
4458 }
4459
4460 static u32 find_raid56_stripe_len(u32 data_devices, u32 dev_stripe_target)
4461 {
4462         /* TODO allow them to set a preferred stripe size */
4463         return SZ_64K;
4464 }
4465
4466 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
4467 {
4468         if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK))
4469                 return;
4470
4471         btrfs_set_fs_incompat(info, RAID56);
4472 }
4473
4474 #define BTRFS_MAX_DEVS(r) ((BTRFS_LEAF_DATA_SIZE(r)             \
4475                         - sizeof(struct btrfs_item)             \
4476                         - sizeof(struct btrfs_chunk))           \
4477                         / sizeof(struct btrfs_stripe) + 1)
4478
4479 #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE        \
4480                                 - 2 * sizeof(struct btrfs_disk_key)     \
4481                                 - 2 * sizeof(struct btrfs_chunk))       \
4482                                 / sizeof(struct btrfs_stripe) + 1)
4483
4484 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4485                                struct btrfs_root *extent_root, u64 start,
4486                                u64 type)
4487 {
4488         struct btrfs_fs_info *info = extent_root->fs_info;
4489         struct btrfs_fs_devices *fs_devices = info->fs_devices;
4490         struct list_head *cur;
4491         struct map_lookup *map = NULL;
4492         struct extent_map_tree *em_tree;
4493         struct extent_map *em;
4494         struct btrfs_device_info *devices_info = NULL;
4495         u64 total_avail;
4496         int num_stripes;        /* total number of stripes to allocate */
4497         int data_stripes;       /* number of stripes that count for
4498                                    block group size */
4499         int sub_stripes;        /* sub_stripes info for map */
4500         int dev_stripes;        /* stripes per dev */
4501         int devs_max;           /* max devs to use */
4502         int devs_min;           /* min devs needed */
4503         int devs_increment;     /* ndevs has to be a multiple of this */
4504         int ncopies;            /* how many copies to data has */
4505         int ret;
4506         u64 max_stripe_size;
4507         u64 max_chunk_size;
4508         u64 stripe_size;
4509         u64 num_bytes;
4510         u64 raid_stripe_len = BTRFS_STRIPE_LEN;
4511         int ndevs;
4512         int i;
4513         int j;
4514         int index;
4515
4516         BUG_ON(!alloc_profile_is_valid(type, 0));
4517
4518         if (list_empty(&fs_devices->alloc_list))
4519                 return -ENOSPC;
4520
4521         index = __get_raid_index(type);
4522
4523         sub_stripes = btrfs_raid_array[index].sub_stripes;
4524         dev_stripes = btrfs_raid_array[index].dev_stripes;
4525         devs_max = btrfs_raid_array[index].devs_max;
4526         devs_min = btrfs_raid_array[index].devs_min;
4527         devs_increment = btrfs_raid_array[index].devs_increment;
4528         ncopies = btrfs_raid_array[index].ncopies;
4529
4530         if (type & BTRFS_BLOCK_GROUP_DATA) {
4531                 max_stripe_size = SZ_1G;
4532                 max_chunk_size = 10 * max_stripe_size;
4533                 if (!devs_max)
4534                         devs_max = BTRFS_MAX_DEVS(info->chunk_root);
4535         } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
4536                 /* for larger filesystems, use larger metadata chunks */
4537                 if (fs_devices->total_rw_bytes > 50ULL * SZ_1G)
4538                         max_stripe_size = SZ_1G;
4539                 else
4540                         max_stripe_size = SZ_256M;
4541                 max_chunk_size = max_stripe_size;
4542                 if (!devs_max)
4543                         devs_max = BTRFS_MAX_DEVS(info->chunk_root);
4544         } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
4545                 max_stripe_size = SZ_32M;
4546                 max_chunk_size = 2 * max_stripe_size;
4547                 if (!devs_max)
4548                         devs_max = BTRFS_MAX_DEVS_SYS_CHUNK;
4549         } else {
4550                 btrfs_err(info, "invalid chunk type 0x%llx requested",
4551                        type);
4552                 BUG_ON(1);
4553         }
4554
4555         /* we don't want a chunk larger than 10% of writeable space */
4556         max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
4557                              max_chunk_size);
4558
4559         devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info),
4560                                GFP_NOFS);
4561         if (!devices_info)
4562                 return -ENOMEM;
4563
4564         cur = fs_devices->alloc_list.next;
4565
4566         /*
4567          * in the first pass through the devices list, we gather information
4568          * about the available holes on each device.
4569          */
4570         ndevs = 0;
4571         while (cur != &fs_devices->alloc_list) {
4572                 struct btrfs_device *device;
4573                 u64 max_avail;
4574                 u64 dev_offset;
4575
4576                 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
4577
4578                 cur = cur->next;
4579
4580                 if (!device->writeable) {
4581                         WARN(1, KERN_ERR
4582                                "BTRFS: read-only device in alloc_list\n");
4583                         continue;
4584                 }
4585
4586                 if (!device->in_fs_metadata ||
4587                     device->is_tgtdev_for_dev_replace)
4588                         continue;
4589
4590                 if (device->total_bytes > device->bytes_used)
4591                         total_avail = device->total_bytes - device->bytes_used;
4592                 else
4593                         total_avail = 0;
4594
4595                 /* If there is no space on this device, skip it. */
4596                 if (total_avail == 0)
4597                         continue;
4598
4599                 ret = find_free_dev_extent(trans, device,
4600                                            max_stripe_size * dev_stripes,
4601                                            &dev_offset, &max_avail);
4602                 if (ret && ret != -ENOSPC)
4603                         goto error;
4604
4605                 if (ret == 0)
4606                         max_avail = max_stripe_size * dev_stripes;
4607
4608                 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes)
4609                         continue;
4610
4611                 if (ndevs == fs_devices->rw_devices) {
4612                         WARN(1, "%s: found more than %llu devices\n",
4613                              __func__, fs_devices->rw_devices);
4614                         break;
4615                 }
4616                 devices_info[ndevs].dev_offset = dev_offset;
4617                 devices_info[ndevs].max_avail = max_avail;
4618                 devices_info[ndevs].total_avail = total_avail;
4619                 devices_info[ndevs].dev = device;
4620                 ++ndevs;
4621         }
4622
4623         /*
4624          * now sort the devices by hole size / available space
4625          */
4626         sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
4627              btrfs_cmp_device_info, NULL);
4628
4629         /* round down to number of usable stripes */
4630         ndevs -= ndevs % devs_increment;
4631
4632         if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) {
4633                 ret = -ENOSPC;
4634                 goto error;
4635         }
4636
4637         if (devs_max && ndevs > devs_max)
4638                 ndevs = devs_max;
4639         /*
4640          * the primary goal is to maximize the number of stripes, so use as many
4641          * devices as possible, even if the stripes are not maximum sized.
4642          */
4643         stripe_size = devices_info[ndevs-1].max_avail;
4644         num_stripes = ndevs * dev_stripes;
4645
4646         /*
4647          * this will have to be fixed for RAID1 and RAID10 over
4648          * more drives
4649          */
4650         data_stripes = num_stripes / ncopies;
4651
4652         if (type & BTRFS_BLOCK_GROUP_RAID5) {
4653                 raid_stripe_len = find_raid56_stripe_len(ndevs - 1,
4654                                  btrfs_super_stripesize(info->super_copy));
4655                 data_stripes = num_stripes - 1;
4656         }
4657         if (type & BTRFS_BLOCK_GROUP_RAID6) {
4658                 raid_stripe_len = find_raid56_stripe_len(ndevs - 2,
4659                                  btrfs_super_stripesize(info->super_copy));
4660                 data_stripes = num_stripes - 2;
4661         }
4662
4663         /*
4664          * Use the number of data stripes to figure out how big this chunk
4665          * is really going to be in terms of logical address space,
4666          * and compare that answer with the max chunk size
4667          */
4668         if (stripe_size * data_stripes > max_chunk_size) {
4669                 u64 mask = (1ULL << 24) - 1;
4670
4671                 stripe_size = div_u64(max_chunk_size, data_stripes);
4672
4673                 /* bump the answer up to a 16MB boundary */
4674                 stripe_size = (stripe_size + mask) & ~mask;
4675
4676                 /* but don't go higher than the limits we found
4677                  * while searching for free extents
4678                  */
4679                 if (stripe_size > devices_info[ndevs-1].max_avail)
4680                         stripe_size = devices_info[ndevs-1].max_avail;
4681         }
4682
4683         stripe_size = div_u64(stripe_size, dev_stripes);
4684
4685         /* align to BTRFS_STRIPE_LEN */
4686         stripe_size = div_u64(stripe_size, raid_stripe_len);
4687         stripe_size *= raid_stripe_len;
4688
4689         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
4690         if (!map) {
4691                 ret = -ENOMEM;
4692                 goto error;
4693         }
4694         map->num_stripes = num_stripes;
4695
4696         for (i = 0; i < ndevs; ++i) {
4697                 for (j = 0; j < dev_stripes; ++j) {
4698                         int s = i * dev_stripes + j;
4699                         map->stripes[s].dev = devices_info[i].dev;
4700                         map->stripes[s].physical = devices_info[i].dev_offset +
4701                                                    j * stripe_size;
4702                 }
4703         }
4704         map->sector_size = extent_root->sectorsize;
4705         map->stripe_len = raid_stripe_len;
4706         map->io_align = raid_stripe_len;
4707         map->io_width = raid_stripe_len;
4708         map->type = type;
4709         map->sub_stripes = sub_stripes;
4710
4711         num_bytes = stripe_size * data_stripes;
4712
4713         trace_btrfs_chunk_alloc(info->chunk_root, map, start, num_bytes);
4714
4715         em = alloc_extent_map();
4716         if (!em) {
4717                 kfree(map);
4718                 ret = -ENOMEM;
4719                 goto error;
4720         }
4721         set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
4722         em->map_lookup = map;
4723         em->start = start;
4724         em->len = num_bytes;
4725         em->block_start = 0;
4726         em->block_len = em->len;
4727         em->orig_block_len = stripe_size;
4728
4729         em_tree = &extent_root->fs_info->mapping_tree.map_tree;
4730         write_lock(&em_tree->lock);
4731         ret = add_extent_mapping(em_tree, em, 0);
4732         if (!ret) {
4733                 list_add_tail(&em->list, &trans->transaction->pending_chunks);
4734                 atomic_inc(&em->refs);
4735         }
4736         write_unlock(&em_tree->lock);
4737         if (ret) {
4738                 free_extent_map(em);
4739                 goto error;
4740         }
4741
4742         ret = btrfs_make_block_group(trans, extent_root, 0, type,
4743                                      BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4744                                      start, num_bytes);
4745         if (ret)
4746                 goto error_del_extent;
4747
4748         for (i = 0; i < map->num_stripes; i++) {
4749                 num_bytes = map->stripes[i].dev->bytes_used + stripe_size;
4750                 btrfs_device_set_bytes_used(map->stripes[i].dev, num_bytes);
4751         }
4752
4753         spin_lock(&extent_root->fs_info->free_chunk_lock);
4754         extent_root->fs_info->free_chunk_space -= (stripe_size *
4755                                                    map->num_stripes);
4756         spin_unlock(&extent_root->fs_info->free_chunk_lock);
4757
4758         free_extent_map(em);
4759         check_raid56_incompat_flag(extent_root->fs_info, type);
4760
4761         kfree(devices_info);
4762         return 0;
4763
4764 error_del_extent:
4765         write_lock(&em_tree->lock);
4766         remove_extent_mapping(em_tree, em);
4767         write_unlock(&em_tree->lock);
4768
4769         /* One for our allocation */
4770         free_extent_map(em);
4771         /* One for the tree reference */
4772         free_extent_map(em);
4773         /* One for the pending_chunks list reference */
4774         free_extent_map(em);
4775 error:
4776         kfree(devices_info);
4777         return ret;
4778 }
4779
4780 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans,
4781                                 struct btrfs_root *extent_root,
4782                                 u64 chunk_offset, u64 chunk_size)
4783 {
4784         struct btrfs_key key;
4785         struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
4786         struct btrfs_device *device;
4787         struct btrfs_chunk *chunk;
4788         struct btrfs_stripe *stripe;
4789         struct extent_map_tree *em_tree;
4790         struct extent_map *em;
4791         struct map_lookup *map;
4792         size_t item_size;
4793         u64 dev_offset;
4794         u64 stripe_size;
4795         int i = 0;
4796         int ret = 0;
4797
4798         em_tree = &extent_root->fs_info->mapping_tree.map_tree;
4799         read_lock(&em_tree->lock);
4800         em = lookup_extent_mapping(em_tree, chunk_offset, chunk_size);
4801         read_unlock(&em_tree->lock);
4802
4803         if (!em) {
4804                 btrfs_crit(extent_root->fs_info, "unable to find logical "
4805                            "%Lu len %Lu", chunk_offset, chunk_size);
4806                 return -EINVAL;
4807         }
4808
4809         if (em->start != chunk_offset || em->len != chunk_size) {
4810                 btrfs_crit(extent_root->fs_info, "found a bad mapping, wanted"
4811                           " %Lu-%Lu, found %Lu-%Lu", chunk_offset,
4812                           chunk_size, em->start, em->len);
4813                 free_extent_map(em);
4814                 return -EINVAL;
4815         }
4816
4817         map = em->map_lookup;
4818         item_size = btrfs_chunk_item_size(map->num_stripes);
4819         stripe_size = em->orig_block_len;
4820
4821         chunk = kzalloc(item_size, GFP_NOFS);
4822         if (!chunk) {
4823                 ret = -ENOMEM;
4824                 goto out;
4825         }
4826
4827         /*
4828          * Take the device list mutex to prevent races with the final phase of
4829          * a device replace operation that replaces the device object associated
4830          * with the map's stripes, because the device object's id can change
4831          * at any time during that final phase of the device replace operation
4832          * (dev-replace.c:btrfs_dev_replace_finishing()).
4833          */
4834         mutex_lock(&chunk_root->fs_info->fs_devices->device_list_mutex);
4835         for (i = 0; i < map->num_stripes; i++) {
4836                 device = map->stripes[i].dev;
4837                 dev_offset = map->stripes[i].physical;
4838
4839                 ret = btrfs_update_device(trans, device);
4840                 if (ret)
4841                         break;
4842                 ret = btrfs_alloc_dev_extent(trans, device,
4843                                              chunk_root->root_key.objectid,
4844                                              BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4845                                              chunk_offset, dev_offset,
4846                                              stripe_size);
4847                 if (ret)
4848                         break;
4849         }
4850         if (ret) {
4851                 mutex_unlock(&chunk_root->fs_info->fs_devices->device_list_mutex);
4852                 goto out;
4853         }
4854
4855         stripe = &chunk->stripe;
4856         for (i = 0; i < map->num_stripes; i++) {
4857                 device = map->stripes[i].dev;
4858                 dev_offset = map->stripes[i].physical;
4859
4860                 btrfs_set_stack_stripe_devid(stripe, device->devid);
4861                 btrfs_set_stack_stripe_offset(stripe, dev_offset);
4862                 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
4863                 stripe++;
4864         }
4865         mutex_unlock(&chunk_root->fs_info->fs_devices->device_list_mutex);
4866
4867         btrfs_set_stack_chunk_length(chunk, chunk_size);
4868         btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
4869         btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
4870         btrfs_set_stack_chunk_type(chunk, map->type);
4871         btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
4872         btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
4873         btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
4874         btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
4875         btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
4876
4877         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
4878         key.type = BTRFS_CHUNK_ITEM_KEY;
4879         key.offset = chunk_offset;
4880
4881         ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
4882         if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
4883                 /*
4884                  * TODO: Cleanup of inserted chunk root in case of
4885                  * failure.
4886                  */
4887                 ret = btrfs_add_system_chunk(chunk_root, &key, chunk,
4888                                              item_size);
4889         }
4890
4891 out:
4892         kfree(chunk);
4893         free_extent_map(em);
4894         return ret;
4895 }
4896
4897 /*
4898  * Chunk allocation falls into two parts. The first part does works
4899  * that make the new allocated chunk useable, but not do any operation
4900  * that modifies the chunk tree. The second part does the works that
4901  * require modifying the chunk tree. This division is important for the
4902  * bootstrap process of adding storage to a seed btrfs.
4903  */
4904 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4905                       struct btrfs_root *extent_root, u64 type)
4906 {
4907         u64 chunk_offset;
4908
4909         ASSERT(mutex_is_locked(&extent_root->fs_info->chunk_mutex));
4910         chunk_offset = find_next_chunk(extent_root->fs_info);
4911         return __btrfs_alloc_chunk(trans, extent_root, chunk_offset, type);
4912 }
4913
4914 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
4915                                          struct btrfs_root *root,
4916                                          struct btrfs_device *device)
4917 {
4918         u64 chunk_offset;
4919         u64 sys_chunk_offset;
4920         u64 alloc_profile;
4921         struct btrfs_fs_info *fs_info = root->fs_info;
4922         struct btrfs_root *extent_root = fs_info->extent_root;
4923         int ret;
4924
4925         chunk_offset = find_next_chunk(fs_info);
4926         alloc_profile = btrfs_get_alloc_profile(extent_root, 0);
4927         ret = __btrfs_alloc_chunk(trans, extent_root, chunk_offset,
4928                                   alloc_profile);
4929         if (ret)
4930                 return ret;
4931
4932         sys_chunk_offset = find_next_chunk(root->fs_info);
4933         alloc_profile = btrfs_get_alloc_profile(fs_info->chunk_root, 0);
4934         ret = __btrfs_alloc_chunk(trans, extent_root, sys_chunk_offset,
4935                                   alloc_profile);
4936         return ret;
4937 }
4938
4939 static inline int btrfs_chunk_max_errors(struct map_lookup *map)
4940 {
4941         int max_errors;
4942
4943         if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
4944                          BTRFS_BLOCK_GROUP_RAID10 |
4945                          BTRFS_BLOCK_GROUP_RAID5 |
4946                          BTRFS_BLOCK_GROUP_DUP)) {
4947                 max_errors = 1;
4948         } else if (map->type & BTRFS_BLOCK_GROUP_RAID6) {
4949                 max_errors = 2;
4950         } else {
4951                 max_errors = 0;
4952         }
4953
4954         return max_errors;
4955 }
4956
4957 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
4958 {
4959         struct extent_map *em;
4960         struct map_lookup *map;
4961         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
4962         int readonly = 0;
4963         int miss_ndevs = 0;
4964         int i;
4965
4966         read_lock(&map_tree->map_tree.lock);
4967         em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
4968         read_unlock(&map_tree->map_tree.lock);
4969         if (!em)
4970                 return 1;
4971
4972         map = em->map_lookup;
4973         for (i = 0; i < map->num_stripes; i++) {
4974                 if (map->stripes[i].dev->missing) {
4975                         miss_ndevs++;
4976                         continue;
4977                 }
4978
4979                 if (!map->stripes[i].dev->writeable) {
4980                         readonly = 1;
4981                         goto end;
4982                 }
4983         }
4984
4985         /*
4986          * If the number of missing devices is larger than max errors,
4987          * we can not write the data into that chunk successfully, so
4988          * set it readonly.
4989          */
4990         if (miss_ndevs > btrfs_chunk_max_errors(map))
4991                 readonly = 1;
4992 end:
4993         free_extent_map(em);
4994         return readonly;
4995 }
4996
4997 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
4998 {
4999         extent_map_tree_init(&tree->map_tree);
5000 }
5001
5002 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
5003 {
5004         struct extent_map *em;
5005
5006         while (1) {
5007                 write_lock(&tree->map_tree.lock);
5008                 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
5009                 if (em)
5010                         remove_extent_mapping(&tree->map_tree, em);
5011                 write_unlock(&tree->map_tree.lock);
5012                 if (!em)
5013                         break;
5014                 /* once for us */
5015                 free_extent_map(em);
5016                 /* once for the tree */
5017                 free_extent_map(em);
5018         }
5019 }
5020
5021 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5022 {
5023         struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
5024         struct extent_map *em;
5025         struct map_lookup *map;
5026         struct extent_map_tree *em_tree = &map_tree->map_tree;
5027         int ret;
5028
5029         read_lock(&em_tree->lock);
5030         em = lookup_extent_mapping(em_tree, logical, len);
5031         read_unlock(&em_tree->lock);
5032
5033         /*
5034          * We could return errors for these cases, but that could get ugly and
5035          * we'd probably do the same thing which is just not do anything else
5036          * and exit, so return 1 so the callers don't try to use other copies.
5037          */
5038         if (!em) {
5039                 btrfs_crit(fs_info, "No mapping for %Lu-%Lu", logical,
5040                             logical+len);
5041                 return 1;
5042         }
5043
5044         if (em->start > logical || em->start + em->len < logical) {
5045                 btrfs_crit(fs_info, "Invalid mapping for %Lu-%Lu, got "
5046                             "%Lu-%Lu", logical, logical+len, em->start,
5047                             em->start + em->len);
5048                 free_extent_map(em);
5049                 return 1;
5050         }
5051
5052         map = em->map_lookup;
5053         if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
5054                 ret = map->num_stripes;
5055         else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5056                 ret = map->sub_stripes;
5057         else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
5058                 ret = 2;
5059         else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5060                 ret = 3;
5061         else
5062                 ret = 1;
5063         free_extent_map(em);
5064
5065         btrfs_dev_replace_lock(&fs_info->dev_replace);
5066         if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace))
5067                 ret++;
5068         btrfs_dev_replace_unlock(&fs_info->dev_replace);
5069
5070         return ret;
5071 }
5072
5073 unsigned long btrfs_full_stripe_len(struct btrfs_root *root,
5074                                     struct btrfs_mapping_tree *map_tree,
5075                                     u64 logical)
5076 {
5077         struct extent_map *em;
5078         struct map_lookup *map;
5079         struct extent_map_tree *em_tree = &map_tree->map_tree;
5080         unsigned long len = root->sectorsize;
5081
5082         read_lock(&em_tree->lock);
5083         em = lookup_extent_mapping(em_tree, logical, len);
5084         read_unlock(&em_tree->lock);
5085         BUG_ON(!em);
5086
5087         BUG_ON(em->start > logical || em->start + em->len < logical);
5088         map = em->map_lookup;
5089         if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5090                 len = map->stripe_len * nr_data_stripes(map);
5091         free_extent_map(em);
5092         return len;
5093 }
5094
5095 int btrfs_is_parity_mirror(struct btrfs_mapping_tree *map_tree,
5096                            u64 logical, u64 len, int mirror_num)
5097 {
5098         struct extent_map *em;
5099         struct map_lookup *map;
5100         struct extent_map_tree *em_tree = &map_tree->map_tree;
5101         int ret = 0;
5102
5103         read_lock(&em_tree->lock);
5104         em = lookup_extent_mapping(em_tree, logical, len);
5105         read_unlock(&em_tree->lock);
5106         BUG_ON(!em);
5107
5108         BUG_ON(em->start > logical || em->start + em->len < logical);
5109         map = em->map_lookup;
5110         if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5111                 ret = 1;
5112         free_extent_map(em);
5113         return ret;
5114 }
5115
5116 static int find_live_mirror(struct btrfs_fs_info *fs_info,
5117                             struct map_lookup *map, int first, int num,
5118                             int optimal, int dev_replace_is_ongoing)
5119 {
5120         int i;
5121         int tolerance;
5122         struct btrfs_device *srcdev;
5123
5124         if (dev_replace_is_ongoing &&
5125             fs_info->dev_replace.cont_reading_from_srcdev_mode ==
5126              BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
5127                 srcdev = fs_info->dev_replace.srcdev;
5128         else
5129                 srcdev = NULL;
5130
5131         /*
5132          * try to avoid the drive that is the source drive for a
5133          * dev-replace procedure, only choose it if no other non-missing
5134          * mirror is available
5135          */
5136         for (tolerance = 0; tolerance < 2; tolerance++) {
5137                 if (map->stripes[optimal].dev->bdev &&
5138                     (tolerance || map->stripes[optimal].dev != srcdev))
5139                         return optimal;
5140                 for (i = first; i < first + num; i++) {
5141                         if (map->stripes[i].dev->bdev &&
5142                             (tolerance || map->stripes[i].dev != srcdev))
5143                                 return i;
5144                 }
5145         }
5146
5147         /* we couldn't find one that doesn't fail.  Just return something
5148          * and the io error handling code will clean up eventually
5149          */
5150         return optimal;
5151 }
5152
5153 static inline int parity_smaller(u64 a, u64 b)
5154 {
5155         return a > b;
5156 }
5157
5158 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
5159 static void sort_parity_stripes(struct btrfs_bio *bbio, int num_stripes)
5160 {
5161         struct btrfs_bio_stripe s;
5162         int i;
5163         u64 l;
5164         int again = 1;
5165
5166         while (again) {
5167                 again = 0;
5168                 for (i = 0; i < num_stripes - 1; i++) {
5169                         if (parity_smaller(bbio->raid_map[i],
5170                                            bbio->raid_map[i+1])) {
5171                                 s = bbio->stripes[i];
5172                                 l = bbio->raid_map[i];
5173                                 bbio->stripes[i] = bbio->stripes[i+1];
5174                                 bbio->raid_map[i] = bbio->raid_map[i+1];
5175                                 bbio->stripes[i+1] = s;
5176                                 bbio->raid_map[i+1] = l;
5177
5178                                 again = 1;
5179                         }
5180                 }
5181         }
5182 }
5183
5184 static struct btrfs_bio *alloc_btrfs_bio(int total_stripes, int real_stripes)
5185 {
5186         struct btrfs_bio *bbio = kzalloc(
5187                  /* the size of the btrfs_bio */
5188                 sizeof(struct btrfs_bio) +
5189                 /* plus the variable array for the stripes */
5190                 sizeof(struct btrfs_bio_stripe) * (total_stripes) +
5191                 /* plus the variable array for the tgt dev */
5192                 sizeof(int) * (real_stripes) +
5193                 /*
5194                  * plus the raid_map, which includes both the tgt dev
5195                  * and the stripes
5196                  */
5197                 sizeof(u64) * (total_stripes),
5198                 GFP_NOFS|__GFP_NOFAIL);
5199
5200         atomic_set(&bbio->error, 0);
5201         atomic_set(&bbio->refs, 1);
5202
5203         return bbio;
5204 }
5205
5206 void btrfs_get_bbio(struct btrfs_bio *bbio)
5207 {
5208         WARN_ON(!atomic_read(&bbio->refs));
5209         atomic_inc(&bbio->refs);
5210 }
5211
5212 void btrfs_put_bbio(struct btrfs_bio *bbio)
5213 {
5214         if (!bbio)
5215                 return;
5216         if (atomic_dec_and_test(&bbio->refs))
5217                 kfree(bbio);
5218 }
5219
5220 static int __btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
5221                              u64 logical, u64 *length,
5222                              struct btrfs_bio **bbio_ret,
5223                              int mirror_num, int need_raid_map)
5224 {
5225         struct extent_map *em;
5226         struct map_lookup *map;
5227         struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
5228         struct extent_map_tree *em_tree = &map_tree->map_tree;
5229         u64 offset;
5230         u64 stripe_offset;
5231         u64 stripe_end_offset;
5232         u64 stripe_nr;
5233         u64 stripe_nr_orig;
5234         u64 stripe_nr_end;
5235         u64 stripe_len;
5236         u32 stripe_index;
5237         int i;
5238         int ret = 0;
5239         int num_stripes;
5240         int max_errors = 0;
5241         int tgtdev_indexes = 0;
5242         struct btrfs_bio *bbio = NULL;
5243         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
5244         int dev_replace_is_ongoing = 0;
5245         int num_alloc_stripes;
5246         int patch_the_first_stripe_for_dev_replace = 0;
5247         u64 physical_to_patch_in_first_stripe = 0;
5248         u64 raid56_full_stripe_start = (u64)-1;
5249
5250         read_lock(&em_tree->lock);
5251         em = lookup_extent_mapping(em_tree, logical, *length);
5252         read_unlock(&em_tree->lock);
5253
5254         if (!em) {
5255                 btrfs_crit(fs_info, "unable to find logical %llu len %llu",
5256                         logical, *length);
5257                 return -EINVAL;
5258         }
5259
5260         if (em->start > logical || em->start + em->len < logical) {
5261                 btrfs_crit(fs_info, "found a bad mapping, wanted %Lu, "
5262                            "found %Lu-%Lu", logical, em->start,
5263                            em->start + em->len);
5264                 free_extent_map(em);
5265                 return -EINVAL;
5266         }
5267
5268         map = em->map_lookup;
5269         offset = logical - em->start;
5270
5271         stripe_len = map->stripe_len;
5272         stripe_nr = offset;
5273         /*
5274          * stripe_nr counts the total number of stripes we have to stride
5275          * to get to this block
5276          */
5277         stripe_nr = div64_u64(stripe_nr, stripe_len);
5278
5279         stripe_offset = stripe_nr * stripe_len;
5280         BUG_ON(offset < stripe_offset);
5281
5282         /* stripe_offset is the offset of this block in its stripe*/
5283         stripe_offset = offset - stripe_offset;
5284
5285         /* if we're here for raid56, we need to know the stripe aligned start */
5286         if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5287                 unsigned long full_stripe_len = stripe_len * nr_data_stripes(map);
5288                 raid56_full_stripe_start = offset;
5289
5290                 /* allow a write of a full stripe, but make sure we don't
5291                  * allow straddling of stripes
5292                  */
5293                 raid56_full_stripe_start = div64_u64(raid56_full_stripe_start,
5294                                 full_stripe_len);
5295                 raid56_full_stripe_start *= full_stripe_len;
5296         }
5297
5298         if (rw & REQ_DISCARD) {
5299                 /* we don't discard raid56 yet */
5300                 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5301                         ret = -EOPNOTSUPP;
5302                         goto out;
5303                 }
5304                 *length = min_t(u64, em->len - offset, *length);
5305         } else if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
5306                 u64 max_len;
5307                 /* For writes to RAID[56], allow a full stripeset across all disks.
5308                    For other RAID types and for RAID[56] reads, just allow a single
5309                    stripe (on a single disk). */
5310                 if ((map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
5311                     (rw & REQ_WRITE)) {
5312                         max_len = stripe_len * nr_data_stripes(map) -
5313                                 (offset - raid56_full_stripe_start);
5314                 } else {
5315                         /* we limit the length of each bio to what fits in a stripe */
5316                         max_len = stripe_len - stripe_offset;
5317                 }
5318                 *length = min_t(u64, em->len - offset, max_len);
5319         } else {
5320                 *length = em->len - offset;
5321         }
5322
5323         /* This is for when we're called from btrfs_merge_bio_hook() and all
5324            it cares about is the length */
5325         if (!bbio_ret)
5326                 goto out;
5327
5328         btrfs_dev_replace_lock(dev_replace);
5329         dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
5330         if (!dev_replace_is_ongoing)
5331                 btrfs_dev_replace_unlock(dev_replace);
5332
5333         if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
5334             !(rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) &&
5335             dev_replace->tgtdev != NULL) {
5336                 /*
5337                  * in dev-replace case, for repair case (that's the only
5338                  * case where the mirror is selected explicitly when
5339                  * calling btrfs_map_block), blocks left of the left cursor
5340                  * can also be read from the target drive.
5341                  * For REQ_GET_READ_MIRRORS, the target drive is added as
5342                  * the last one to the array of stripes. For READ, it also
5343                  * needs to be supported using the same mirror number.
5344                  * If the requested block is not left of the left cursor,
5345                  * EIO is returned. This can happen because btrfs_num_copies()
5346                  * returns one more in the dev-replace case.
5347                  */
5348                 u64 tmp_length = *length;
5349                 struct btrfs_bio *tmp_bbio = NULL;
5350                 int tmp_num_stripes;
5351                 u64 srcdev_devid = dev_replace->srcdev->devid;
5352                 int index_srcdev = 0;
5353                 int found = 0;
5354                 u64 physical_of_found = 0;
5355
5356                 ret = __btrfs_map_block(fs_info, REQ_GET_READ_MIRRORS,
5357                              logical, &tmp_length, &tmp_bbio, 0, 0);
5358                 if (ret) {
5359                         WARN_ON(tmp_bbio != NULL);
5360                         goto out;
5361                 }
5362
5363                 tmp_num_stripes = tmp_bbio->num_stripes;
5364                 if (mirror_num > tmp_num_stripes) {
5365                         /*
5366                          * REQ_GET_READ_MIRRORS does not contain this
5367                          * mirror, that means that the requested area
5368                          * is not left of the left cursor
5369                          */
5370                         ret = -EIO;
5371                         btrfs_put_bbio(tmp_bbio);
5372                         goto out;
5373                 }
5374
5375                 /*
5376                  * process the rest of the function using the mirror_num
5377                  * of the source drive. Therefore look it up first.
5378                  * At the end, patch the device pointer to the one of the
5379                  * target drive.
5380                  */
5381                 for (i = 0; i < tmp_num_stripes; i++) {
5382                         if (tmp_bbio->stripes[i].dev->devid != srcdev_devid)
5383                                 continue;
5384
5385                         /*
5386                          * In case of DUP, in order to keep it simple, only add
5387                          * the mirror with the lowest physical address
5388                          */
5389                         if (found &&
5390                             physical_of_found <= tmp_bbio->stripes[i].physical)
5391                                 continue;
5392
5393                         index_srcdev = i;
5394                         found = 1;
5395                         physical_of_found = tmp_bbio->stripes[i].physical;
5396                 }
5397
5398                 btrfs_put_bbio(tmp_bbio);
5399
5400                 if (!found) {
5401                         WARN_ON(1);
5402                         ret = -EIO;
5403                         goto out;
5404                 }
5405
5406                 mirror_num = index_srcdev + 1;
5407                 patch_the_first_stripe_for_dev_replace = 1;
5408                 physical_to_patch_in_first_stripe = physical_of_found;
5409         } else if (mirror_num > map->num_stripes) {
5410                 mirror_num = 0;
5411         }
5412
5413         num_stripes = 1;
5414         stripe_index = 0;
5415         stripe_nr_orig = stripe_nr;
5416         stripe_nr_end = ALIGN(offset + *length, map->stripe_len);
5417         stripe_nr_end = div_u64(stripe_nr_end, map->stripe_len);
5418         stripe_end_offset = stripe_nr_end * map->stripe_len -
5419                             (offset + *length);
5420
5421         if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5422                 if (rw & REQ_DISCARD)
5423                         num_stripes = min_t(u64, map->num_stripes,
5424                                             stripe_nr_end - stripe_nr_orig);
5425                 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5426                                 &stripe_index);
5427                 if (!(rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)))
5428                         mirror_num = 1;
5429         } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
5430                 if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS))
5431                         num_stripes = map->num_stripes;
5432                 else if (mirror_num)
5433                         stripe_index = mirror_num - 1;
5434                 else {
5435                         stripe_index = find_live_mirror(fs_info, map, 0,
5436                                             map->num_stripes,
5437                                             current->pid % map->num_stripes,
5438                                             dev_replace_is_ongoing);
5439                         mirror_num = stripe_index + 1;
5440                 }
5441
5442         } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
5443                 if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) {
5444                         num_stripes = map->num_stripes;
5445                 } else if (mirror_num) {
5446                         stripe_index = mirror_num - 1;
5447                 } else {
5448                         mirror_num = 1;
5449                 }
5450
5451         } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5452                 u32 factor = map->num_stripes / map->sub_stripes;
5453
5454                 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
5455                 stripe_index *= map->sub_stripes;
5456
5457                 if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS))
5458                         num_stripes = map->sub_stripes;
5459                 else if (rw & REQ_DISCARD)
5460                         num_stripes = min_t(u64, map->sub_stripes *
5461                                             (stripe_nr_end - stripe_nr_orig),
5462                                             map->num_stripes);
5463                 else if (mirror_num)
5464                         stripe_index += mirror_num - 1;
5465                 else {
5466                         int old_stripe_index = stripe_index;
5467                         stripe_index = find_live_mirror(fs_info, map,
5468                                               stripe_index,
5469                                               map->sub_stripes, stripe_index +
5470                                               current->pid % map->sub_stripes,
5471                                               dev_replace_is_ongoing);
5472                         mirror_num = stripe_index - old_stripe_index + 1;
5473                 }
5474
5475         } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5476                 if (need_raid_map &&
5477                     ((rw & (REQ_WRITE | REQ_GET_READ_MIRRORS)) ||
5478                      mirror_num > 1)) {
5479                         /* push stripe_nr back to the start of the full stripe */
5480                         stripe_nr = div_u64(raid56_full_stripe_start,
5481                                         stripe_len * nr_data_stripes(map));
5482
5483                         /* RAID[56] write or recovery. Return all stripes */
5484                         num_stripes = map->num_stripes;
5485                         max_errors = nr_parity_stripes(map);
5486
5487                         *length = map->stripe_len;
5488                         stripe_index = 0;
5489                         stripe_offset = 0;
5490                 } else {
5491                         /*
5492                          * Mirror #0 or #1 means the original data block.
5493                          * Mirror #2 is RAID5 parity block.
5494                          * Mirror #3 is RAID6 Q block.
5495                          */
5496                         stripe_nr = div_u64_rem(stripe_nr,
5497                                         nr_data_stripes(map), &stripe_index);
5498                         if (mirror_num > 1)
5499                                 stripe_index = nr_data_stripes(map) +
5500                                                 mirror_num - 2;
5501
5502                         /* We distribute the parity blocks across stripes */
5503                         div_u64_rem(stripe_nr + stripe_index, map->num_stripes,
5504                                         &stripe_index);
5505                         if (!(rw & (REQ_WRITE | REQ_DISCARD |
5506                                     REQ_GET_READ_MIRRORS)) && mirror_num <= 1)
5507                                 mirror_num = 1;
5508                 }
5509         } else {
5510                 /*
5511                  * after this, stripe_nr is the number of stripes on this
5512                  * device we have to walk to find the data, and stripe_index is
5513                  * the number of our device in the stripe array
5514                  */
5515                 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5516                                 &stripe_index);
5517                 mirror_num = stripe_index + 1;
5518         }
5519         BUG_ON(stripe_index >= map->num_stripes);
5520
5521         num_alloc_stripes = num_stripes;
5522         if (dev_replace_is_ongoing) {
5523                 if (rw & (REQ_WRITE | REQ_DISCARD))
5524                         num_alloc_stripes <<= 1;
5525                 if (rw & REQ_GET_READ_MIRRORS)
5526                         num_alloc_stripes++;
5527                 tgtdev_indexes = num_stripes;
5528         }
5529
5530         bbio = alloc_btrfs_bio(num_alloc_stripes, tgtdev_indexes);
5531         if (!bbio) {
5532                 ret = -ENOMEM;
5533                 goto out;
5534         }
5535         if (dev_replace_is_ongoing)
5536                 bbio->tgtdev_map = (int *)(bbio->stripes + num_alloc_stripes);
5537
5538         /* build raid_map */
5539         if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK &&
5540             need_raid_map && ((rw & (REQ_WRITE | REQ_GET_READ_MIRRORS)) ||
5541             mirror_num > 1)) {
5542                 u64 tmp;
5543                 unsigned rot;
5544
5545                 bbio->raid_map = (u64 *)((void *)bbio->stripes +
5546                                  sizeof(struct btrfs_bio_stripe) *
5547                                  num_alloc_stripes +
5548                                  sizeof(int) * tgtdev_indexes);
5549
5550                 /* Work out the disk rotation on this stripe-set */
5551                 div_u64_rem(stripe_nr, num_stripes, &rot);
5552
5553                 /* Fill in the logical address of each stripe */
5554                 tmp = stripe_nr * nr_data_stripes(map);
5555                 for (i = 0; i < nr_data_stripes(map); i++)
5556                         bbio->raid_map[(i+rot) % num_stripes] =
5557                                 em->start + (tmp + i) * map->stripe_len;
5558
5559                 bbio->raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE;
5560                 if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5561                         bbio->raid_map[(i+rot+1) % num_stripes] =
5562                                 RAID6_Q_STRIPE;
5563         }
5564
5565         if (rw & REQ_DISCARD) {
5566                 u32 factor = 0;
5567                 u32 sub_stripes = 0;
5568                 u64 stripes_per_dev = 0;
5569                 u32 remaining_stripes = 0;
5570                 u32 last_stripe = 0;
5571
5572                 if (map->type &
5573                     (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID10)) {
5574                         if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5575                                 sub_stripes = 1;
5576                         else
5577                                 sub_stripes = map->sub_stripes;
5578
5579                         factor = map->num_stripes / sub_stripes;
5580                         stripes_per_dev = div_u64_rem(stripe_nr_end -
5581                                                       stripe_nr_orig,
5582                                                       factor,
5583                                                       &remaining_stripes);
5584                         div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
5585                         last_stripe *= sub_stripes;
5586                 }
5587
5588                 for (i = 0; i < num_stripes; i++) {
5589                         bbio->stripes[i].physical =
5590                                 map->stripes[stripe_index].physical +
5591                                 stripe_offset + stripe_nr * map->stripe_len;
5592                         bbio->stripes[i].dev = map->stripes[stripe_index].dev;
5593
5594                         if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5595                                          BTRFS_BLOCK_GROUP_RAID10)) {
5596                                 bbio->stripes[i].length = stripes_per_dev *
5597                                                           map->stripe_len;
5598
5599                                 if (i / sub_stripes < remaining_stripes)
5600                                         bbio->stripes[i].length +=
5601                                                 map->stripe_len;
5602
5603                                 /*
5604                                  * Special for the first stripe and
5605                                  * the last stripe:
5606                                  *
5607                                  * |-------|...|-------|
5608                                  *     |----------|
5609                                  *    off     end_off
5610                                  */
5611                                 if (i < sub_stripes)
5612                                         bbio->stripes[i].length -=
5613                                                 stripe_offset;
5614
5615                                 if (stripe_index >= last_stripe &&
5616                                     stripe_index <= (last_stripe +
5617                                                      sub_stripes - 1))
5618                                         bbio->stripes[i].length -=
5619                                                 stripe_end_offset;
5620
5621                                 if (i == sub_stripes - 1)
5622                                         stripe_offset = 0;
5623                         } else
5624                                 bbio->stripes[i].length = *length;
5625
5626                         stripe_index++;
5627                         if (stripe_index == map->num_stripes) {
5628                                 /* This could only happen for RAID0/10 */
5629                                 stripe_index = 0;
5630                                 stripe_nr++;
5631                         }
5632                 }
5633         } else {
5634                 for (i = 0; i < num_stripes; i++) {
5635                         bbio->stripes[i].physical =
5636                                 map->stripes[stripe_index].physical +
5637                                 stripe_offset +
5638                                 stripe_nr * map->stripe_len;
5639                         bbio->stripes[i].dev =
5640                                 map->stripes[stripe_index].dev;
5641                         stripe_index++;
5642                 }
5643         }
5644
5645         if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS))
5646                 max_errors = btrfs_chunk_max_errors(map);
5647
5648         if (bbio->raid_map)
5649                 sort_parity_stripes(bbio, num_stripes);
5650
5651         tgtdev_indexes = 0;
5652         if (dev_replace_is_ongoing && (rw & (REQ_WRITE | REQ_DISCARD)) &&
5653             dev_replace->tgtdev != NULL) {
5654                 int index_where_to_add;
5655                 u64 srcdev_devid = dev_replace->srcdev->devid;
5656
5657                 /*
5658                  * duplicate the write operations while the dev replace
5659                  * procedure is running. Since the copying of the old disk
5660                  * to the new disk takes place at run time while the
5661                  * filesystem is mounted writable, the regular write
5662                  * operations to the old disk have to be duplicated to go
5663                  * to the new disk as well.
5664                  * Note that device->missing is handled by the caller, and
5665                  * that the write to the old disk is already set up in the
5666                  * stripes array.
5667                  */
5668                 index_where_to_add = num_stripes;
5669                 for (i = 0; i < num_stripes; i++) {
5670                         if (bbio->stripes[i].dev->devid == srcdev_devid) {
5671                                 /* write to new disk, too */
5672                                 struct btrfs_bio_stripe *new =
5673                                         bbio->stripes + index_where_to_add;
5674                                 struct btrfs_bio_stripe *old =
5675                                         bbio->stripes + i;
5676
5677                                 new->physical = old->physical;
5678                                 new->length = old->length;
5679                                 new->dev = dev_replace->tgtdev;
5680                                 bbio->tgtdev_map[i] = index_where_to_add;
5681                                 index_where_to_add++;
5682                                 max_errors++;
5683                                 tgtdev_indexes++;
5684                         }
5685                 }
5686                 num_stripes = index_where_to_add;
5687         } else if (dev_replace_is_ongoing && (rw & REQ_GET_READ_MIRRORS) &&
5688                    dev_replace->tgtdev != NULL) {
5689                 u64 srcdev_devid = dev_replace->srcdev->devid;
5690                 int index_srcdev = 0;
5691                 int found = 0;
5692                 u64 physical_of_found = 0;
5693
5694                 /*
5695                  * During the dev-replace procedure, the target drive can
5696                  * also be used to read data in case it is needed to repair
5697                  * a corrupt block elsewhere. This is possible if the
5698                  * requested area is left of the left cursor. In this area,
5699                  * the target drive is a full copy of the source drive.
5700                  */
5701                 for (i = 0; i < num_stripes; i++) {
5702                         if (bbio->stripes[i].dev->devid == srcdev_devid) {
5703                                 /*
5704                                  * In case of DUP, in order to keep it
5705                                  * simple, only add the mirror with the
5706                                  * lowest physical address
5707                                  */
5708                                 if (found &&
5709                                     physical_of_found <=
5710                                      bbio->stripes[i].physical)
5711                                         continue;
5712                                 index_srcdev = i;
5713                                 found = 1;
5714                                 physical_of_found = bbio->stripes[i].physical;
5715                         }
5716                 }
5717                 if (found) {
5718                         if (physical_of_found + map->stripe_len <=
5719                             dev_replace->cursor_left) {
5720                                 struct btrfs_bio_stripe *tgtdev_stripe =
5721                                         bbio->stripes + num_stripes;
5722
5723                                 tgtdev_stripe->physical = physical_of_found;
5724                                 tgtdev_stripe->length =
5725                                         bbio->stripes[index_srcdev].length;
5726                                 tgtdev_stripe->dev = dev_replace->tgtdev;
5727                                 bbio->tgtdev_map[index_srcdev] = num_stripes;
5728
5729                                 tgtdev_indexes++;
5730                                 num_stripes++;
5731                         }
5732                 }
5733         }
5734
5735         *bbio_ret = bbio;
5736         bbio->map_type = map->type;
5737         bbio->num_stripes = num_stripes;
5738         bbio->max_errors = max_errors;
5739         bbio->mirror_num = mirror_num;
5740         bbio->num_tgtdevs = tgtdev_indexes;
5741
5742         /*
5743          * this is the case that REQ_READ && dev_replace_is_ongoing &&
5744          * mirror_num == num_stripes + 1 && dev_replace target drive is
5745          * available as a mirror
5746          */
5747         if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) {
5748                 WARN_ON(num_stripes > 1);
5749                 bbio->stripes[0].dev = dev_replace->tgtdev;
5750                 bbio->stripes[0].physical = physical_to_patch_in_first_stripe;
5751                 bbio->mirror_num = map->num_stripes + 1;
5752         }
5753 out:
5754         if (dev_replace_is_ongoing)
5755                 btrfs_dev_replace_unlock(dev_replace);
5756         free_extent_map(em);
5757         return ret;
5758 }
5759
5760 int btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
5761                       u64 logical, u64 *length,
5762                       struct btrfs_bio **bbio_ret, int mirror_num)
5763 {
5764         return __btrfs_map_block(fs_info, rw, logical, length, bbio_ret,
5765                                  mirror_num, 0);
5766 }
5767
5768 /* For Scrub/replace */
5769 int btrfs_map_sblock(struct btrfs_fs_info *fs_info, int rw,
5770                      u64 logical, u64 *length,
5771                      struct btrfs_bio **bbio_ret, int mirror_num,
5772                      int need_raid_map)
5773 {
5774         return __btrfs_map_block(fs_info, rw, logical, length, bbio_ret,
5775                                  mirror_num, need_raid_map);
5776 }
5777
5778 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
5779                      u64 chunk_start, u64 physical, u64 devid,
5780                      u64 **logical, int *naddrs, int *stripe_len)
5781 {
5782         struct extent_map_tree *em_tree = &map_tree->map_tree;
5783         struct extent_map *em;
5784         struct map_lookup *map;
5785         u64 *buf;
5786         u64 bytenr;
5787         u64 length;
5788         u64 stripe_nr;
5789         u64 rmap_len;
5790         int i, j, nr = 0;
5791
5792         read_lock(&em_tree->lock);
5793         em = lookup_extent_mapping(em_tree, chunk_start, 1);
5794         read_unlock(&em_tree->lock);
5795
5796         if (!em) {
5797                 printk(KERN_ERR "BTRFS: couldn't find em for chunk %Lu\n",
5798                        chunk_start);
5799                 return -EIO;
5800         }
5801
5802         if (em->start != chunk_start) {
5803                 printk(KERN_ERR "BTRFS: bad chunk start, em=%Lu, wanted=%Lu\n",
5804                        em->start, chunk_start);
5805                 free_extent_map(em);
5806                 return -EIO;
5807         }
5808         map = em->map_lookup;
5809
5810         length = em->len;
5811         rmap_len = map->stripe_len;
5812
5813         if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5814                 length = div_u64(length, map->num_stripes / map->sub_stripes);
5815         else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5816                 length = div_u64(length, map->num_stripes);
5817         else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5818                 length = div_u64(length, nr_data_stripes(map));
5819                 rmap_len = map->stripe_len * nr_data_stripes(map);
5820         }
5821
5822         buf = kcalloc(map->num_stripes, sizeof(u64), GFP_NOFS);
5823         BUG_ON(!buf); /* -ENOMEM */
5824
5825         for (i = 0; i < map->num_stripes; i++) {
5826                 if (devid && map->stripes[i].dev->devid != devid)
5827                         continue;
5828                 if (map->stripes[i].physical > physical ||
5829                     map->stripes[i].physical + length <= physical)
5830                         continue;
5831
5832                 stripe_nr = physical - map->stripes[i].physical;
5833                 stripe_nr = div_u64(stripe_nr, map->stripe_len);
5834
5835                 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5836                         stripe_nr = stripe_nr * map->num_stripes + i;
5837                         stripe_nr = div_u64(stripe_nr, map->sub_stripes);
5838                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5839                         stripe_nr = stripe_nr * map->num_stripes + i;
5840                 } /* else if RAID[56], multiply by nr_data_stripes().
5841                    * Alternatively, just use rmap_len below instead of
5842                    * map->stripe_len */
5843
5844                 bytenr = chunk_start + stripe_nr * rmap_len;
5845                 WARN_ON(nr >= map->num_stripes);
5846                 for (j = 0; j < nr; j++) {
5847                         if (buf[j] == bytenr)
5848                                 break;
5849                 }
5850                 if (j == nr) {
5851                         WARN_ON(nr >= map->num_stripes);
5852                         buf[nr++] = bytenr;
5853                 }
5854         }
5855
5856         *logical = buf;
5857         *naddrs = nr;
5858         *stripe_len = rmap_len;
5859
5860         free_extent_map(em);
5861         return 0;
5862 }
5863
5864 static inline void btrfs_end_bbio(struct btrfs_bio *bbio, struct bio *bio)
5865 {
5866         bio->bi_private = bbio->private;
5867         bio->bi_end_io = bbio->end_io;
5868         bio_endio(bio);
5869
5870         btrfs_put_bbio(bbio);
5871 }
5872
5873 static void btrfs_end_bio(struct bio *bio)
5874 {
5875         struct btrfs_bio *bbio = bio->bi_private;
5876         int is_orig_bio = 0;
5877
5878         if (bio->bi_error) {
5879                 atomic_inc(&bbio->error);
5880                 if (bio->bi_error == -EIO || bio->bi_error == -EREMOTEIO) {
5881                         unsigned int stripe_index =
5882                                 btrfs_io_bio(bio)->stripe_index;
5883                         struct btrfs_device *dev;
5884
5885                         BUG_ON(stripe_index >= bbio->num_stripes);
5886                         dev = bbio->stripes[stripe_index].dev;
5887                         if (dev->bdev) {
5888                                 if (bio->bi_rw & WRITE)
5889                                         btrfs_dev_stat_inc(dev,
5890                                                 BTRFS_DEV_STAT_WRITE_ERRS);
5891                                 else
5892                                         btrfs_dev_stat_inc(dev,
5893                                                 BTRFS_DEV_STAT_READ_ERRS);
5894                                 if ((bio->bi_rw & WRITE_FLUSH) == WRITE_FLUSH)
5895                                         btrfs_dev_stat_inc(dev,
5896                                                 BTRFS_DEV_STAT_FLUSH_ERRS);
5897                                 btrfs_dev_stat_print_on_error(dev);
5898                         }
5899                 }
5900         }
5901
5902         if (bio == bbio->orig_bio)
5903                 is_orig_bio = 1;
5904
5905         btrfs_bio_counter_dec(bbio->fs_info);
5906
5907         if (atomic_dec_and_test(&bbio->stripes_pending)) {
5908                 if (!is_orig_bio) {
5909                         bio_put(bio);
5910                         bio = bbio->orig_bio;
5911                 }
5912
5913                 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
5914                 /* only send an error to the higher layers if it is
5915                  * beyond the tolerance of the btrfs bio
5916                  */
5917                 if (atomic_read(&bbio->error) > bbio->max_errors) {
5918                         bio->bi_error = -EIO;
5919                 } else {
5920                         /*
5921                          * this bio is actually up to date, we didn't
5922                          * go over the max number of errors
5923                          */
5924                         bio->bi_error = 0;
5925                 }
5926
5927                 btrfs_end_bbio(bbio, bio);
5928         } else if (!is_orig_bio) {
5929                 bio_put(bio);
5930         }
5931 }
5932
5933 /*
5934  * see run_scheduled_bios for a description of why bios are collected for
5935  * async submit.
5936  *
5937  * This will add one bio to the pending list for a device and make sure
5938  * the work struct is scheduled.
5939  */
5940 static noinline void btrfs_schedule_bio(struct btrfs_root *root,
5941                                         struct btrfs_device *device,
5942                                         int rw, struct bio *bio)
5943 {
5944         int should_queue = 1;
5945         struct btrfs_pending_bios *pending_bios;
5946
5947         if (device->missing || !device->bdev) {
5948                 bio_io_error(bio);
5949                 return;
5950         }
5951
5952         /* don't bother with additional async steps for reads, right now */
5953         if (!(rw & REQ_WRITE)) {
5954                 bio_get(bio);
5955                 btrfsic_submit_bio(rw, bio);
5956                 bio_put(bio);
5957                 return;
5958         }
5959
5960         /*
5961          * nr_async_bios allows us to reliably return congestion to the
5962          * higher layers.  Otherwise, the async bio makes it appear we have
5963          * made progress against dirty pages when we've really just put it
5964          * on a queue for later
5965          */
5966         atomic_inc(&root->fs_info->nr_async_bios);
5967         WARN_ON(bio->bi_next);
5968         bio->bi_next = NULL;
5969         bio->bi_rw |= rw;
5970
5971         spin_lock(&device->io_lock);
5972         if (bio->bi_rw & REQ_SYNC)
5973                 pending_bios = &device->pending_sync_bios;
5974         else
5975                 pending_bios = &device->pending_bios;
5976
5977         if (pending_bios->tail)
5978                 pending_bios->tail->bi_next = bio;
5979
5980         pending_bios->tail = bio;
5981         if (!pending_bios->head)
5982                 pending_bios->head = bio;
5983         if (device->running_pending)
5984                 should_queue = 0;
5985
5986         spin_unlock(&device->io_lock);
5987
5988         if (should_queue)
5989                 btrfs_queue_work(root->fs_info->submit_workers,
5990                                  &device->work);
5991 }
5992
5993 static void submit_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio,
5994                               struct bio *bio, u64 physical, int dev_nr,
5995                               int rw, int async)
5996 {
5997         struct btrfs_device *dev = bbio->stripes[dev_nr].dev;
5998
5999         bio->bi_private = bbio;
6000         btrfs_io_bio(bio)->stripe_index = dev_nr;
6001         bio->bi_end_io = btrfs_end_bio;
6002         bio->bi_iter.bi_sector = physical >> 9;
6003 #ifdef DEBUG
6004         {
6005                 struct rcu_string *name;
6006
6007                 rcu_read_lock();
6008                 name = rcu_dereference(dev->name);
6009                 pr_debug("btrfs_map_bio: rw %d, sector=%llu, dev=%lu "
6010                          "(%s id %llu), size=%u\n", rw,
6011                          (u64)bio->bi_iter.bi_sector, (u_long)dev->bdev->bd_dev,
6012                          name->str, dev->devid, bio->bi_iter.bi_size);
6013                 rcu_read_unlock();
6014         }
6015 #endif
6016         bio->bi_bdev = dev->bdev;
6017
6018         btrfs_bio_counter_inc_noblocked(root->fs_info);
6019
6020         if (async)
6021                 btrfs_schedule_bio(root, dev, rw, bio);
6022         else
6023                 btrfsic_submit_bio(rw, bio);
6024 }
6025
6026 static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical)
6027 {
6028         atomic_inc(&bbio->error);
6029         if (atomic_dec_and_test(&bbio->stripes_pending)) {
6030                 /* Shoud be the original bio. */
6031                 WARN_ON(bio != bbio->orig_bio);
6032
6033                 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
6034                 bio->bi_iter.bi_sector = logical >> 9;
6035                 bio->bi_error = -EIO;
6036                 btrfs_end_bbio(bbio, bio);
6037         }
6038 }
6039
6040 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
6041                   int mirror_num, int async_submit)
6042 {
6043         struct btrfs_device *dev;
6044         struct bio *first_bio = bio;
6045         u64 logical = (u64)bio->bi_iter.bi_sector << 9;
6046         u64 length = 0;
6047         u64 map_length;
6048         int ret;
6049         int dev_nr;
6050         int total_devs;
6051         struct btrfs_bio *bbio = NULL;
6052
6053         length = bio->bi_iter.bi_size;
6054         map_length = length;
6055
6056         btrfs_bio_counter_inc_blocked(root->fs_info);
6057         ret = __btrfs_map_block(root->fs_info, rw, logical, &map_length, &bbio,
6058                               mirror_num, 1);
6059         if (ret) {
6060                 btrfs_bio_counter_dec(root->fs_info);
6061                 return ret;
6062         }
6063
6064         total_devs = bbio->num_stripes;
6065         bbio->orig_bio = first_bio;
6066         bbio->private = first_bio->bi_private;
6067         bbio->end_io = first_bio->bi_end_io;
6068         bbio->fs_info = root->fs_info;
6069         atomic_set(&bbio->stripes_pending, bbio->num_stripes);
6070
6071         if ((bbio->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
6072             ((rw & WRITE) || (mirror_num > 1))) {
6073                 /* In this case, map_length has been set to the length of
6074                    a single stripe; not the whole write */
6075                 if (rw & WRITE) {
6076                         ret = raid56_parity_write(root, bio, bbio, map_length);
6077                 } else {
6078                         ret = raid56_parity_recover(root, bio, bbio, map_length,
6079                                                     mirror_num, 1);
6080                 }
6081
6082                 btrfs_bio_counter_dec(root->fs_info);
6083                 return ret;
6084         }
6085
6086         if (map_length < length) {
6087                 btrfs_crit(root->fs_info, "mapping failed logical %llu bio len %llu len %llu",
6088                         logical, length, map_length);
6089                 BUG();
6090         }
6091
6092         for (dev_nr = 0; dev_nr < total_devs; dev_nr++) {
6093                 dev = bbio->stripes[dev_nr].dev;
6094                 if (!dev || !dev->bdev || (rw & WRITE && !dev->writeable)) {
6095                         bbio_error(bbio, first_bio, logical);
6096                         continue;
6097                 }
6098
6099                 if (dev_nr < total_devs - 1) {
6100                         bio = btrfs_bio_clone(first_bio, GFP_NOFS);
6101                         BUG_ON(!bio); /* -ENOMEM */
6102                 } else
6103                         bio = first_bio;
6104
6105                 submit_stripe_bio(root, bbio, bio,
6106                                   bbio->stripes[dev_nr].physical, dev_nr, rw,
6107                                   async_submit);
6108         }
6109         btrfs_bio_counter_dec(root->fs_info);
6110         return 0;
6111 }
6112
6113 struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid,
6114                                        u8 *uuid, u8 *fsid)
6115 {
6116         struct btrfs_device *device;
6117         struct btrfs_fs_devices *cur_devices;
6118
6119         cur_devices = fs_info->fs_devices;
6120         while (cur_devices) {
6121                 if (!fsid ||
6122                     !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
6123                         device = __find_device(&cur_devices->devices,
6124                                                devid, uuid);
6125                         if (device)
6126                                 return device;
6127                 }
6128                 cur_devices = cur_devices->seed;
6129         }
6130         return NULL;
6131 }
6132
6133 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
6134                                             struct btrfs_fs_devices *fs_devices,
6135                                             u64 devid, u8 *dev_uuid)
6136 {
6137         struct btrfs_device *device;
6138
6139         device = btrfs_alloc_device(NULL, &devid, dev_uuid);
6140         if (IS_ERR(device))
6141                 return NULL;
6142
6143         list_add(&device->dev_list, &fs_devices->devices);
6144         device->fs_devices = fs_devices;
6145         fs_devices->num_devices++;
6146
6147         device->missing = 1;
6148         fs_devices->missing_devices++;
6149
6150         return device;
6151 }
6152
6153 /**
6154  * btrfs_alloc_device - allocate struct btrfs_device
6155  * @fs_info:    used only for generating a new devid, can be NULL if
6156  *              devid is provided (i.e. @devid != NULL).
6157  * @devid:      a pointer to devid for this device.  If NULL a new devid
6158  *              is generated.
6159  * @uuid:       a pointer to UUID for this device.  If NULL a new UUID
6160  *              is generated.
6161  *
6162  * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
6163  * on error.  Returned struct is not linked onto any lists and can be
6164  * destroyed with kfree() right away.
6165  */
6166 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
6167                                         const u64 *devid,
6168                                         const u8 *uuid)
6169 {
6170         struct btrfs_device *dev;
6171         u64 tmp;
6172
6173         if (WARN_ON(!devid && !fs_info))
6174                 return ERR_PTR(-EINVAL);
6175
6176         dev = __alloc_device();
6177         if (IS_ERR(dev))
6178                 return dev;
6179
6180         if (devid)
6181                 tmp = *devid;
6182         else {
6183                 int ret;
6184
6185                 ret = find_next_devid(fs_info, &tmp);
6186                 if (ret) {
6187                         kfree(dev);
6188                         return ERR_PTR(ret);
6189                 }
6190         }
6191         dev->devid = tmp;
6192
6193         if (uuid)
6194                 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE);
6195         else
6196                 generate_random_uuid(dev->uuid);
6197
6198         btrfs_init_work(&dev->work, btrfs_submit_helper,
6199                         pending_bios_fn, NULL, NULL);
6200
6201         return dev;
6202 }
6203
6204 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
6205                           struct extent_buffer *leaf,
6206                           struct btrfs_chunk *chunk)
6207 {
6208         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
6209         struct map_lookup *map;
6210         struct extent_map *em;
6211         u64 logical;
6212         u64 length;
6213         u64 stripe_len;
6214         u64 devid;
6215         u8 uuid[BTRFS_UUID_SIZE];
6216         int num_stripes;
6217         int ret;
6218         int i;
6219
6220         logical = key->offset;
6221         length = btrfs_chunk_length(leaf, chunk);
6222         stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6223         num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6224         /* Validation check */
6225         if (!num_stripes) {
6226                 btrfs_err(root->fs_info, "invalid chunk num_stripes: %u",
6227                           num_stripes);
6228                 return -EIO;
6229         }
6230         if (!IS_ALIGNED(logical, root->sectorsize)) {
6231                 btrfs_err(root->fs_info,
6232                           "invalid chunk logical %llu", logical);
6233                 return -EIO;
6234         }
6235         if (!length || !IS_ALIGNED(length, root->sectorsize)) {
6236                 btrfs_err(root->fs_info,
6237                         "invalid chunk length %llu", length);
6238                 return -EIO;
6239         }
6240         if (!is_power_of_2(stripe_len)) {
6241                 btrfs_err(root->fs_info, "invalid chunk stripe length: %llu",
6242                           stripe_len);
6243                 return -EIO;
6244         }
6245         if (~(BTRFS_BLOCK_GROUP_TYPE_MASK | BTRFS_BLOCK_GROUP_PROFILE_MASK) &
6246             btrfs_chunk_type(leaf, chunk)) {
6247                 btrfs_err(root->fs_info, "unrecognized chunk type: %llu",
6248                           ~(BTRFS_BLOCK_GROUP_TYPE_MASK |
6249                             BTRFS_BLOCK_GROUP_PROFILE_MASK) &
6250                           btrfs_chunk_type(leaf, chunk));
6251                 return -EIO;
6252         }
6253
6254         read_lock(&map_tree->map_tree.lock);
6255         em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
6256         read_unlock(&map_tree->map_tree.lock);
6257
6258         /* already mapped? */
6259         if (em && em->start <= logical && em->start + em->len > logical) {
6260                 free_extent_map(em);
6261                 return 0;
6262         } else if (em) {
6263                 free_extent_map(em);
6264         }
6265
6266         em = alloc_extent_map();
6267         if (!em)
6268                 return -ENOMEM;
6269         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
6270         if (!map) {
6271                 free_extent_map(em);
6272                 return -ENOMEM;
6273         }
6274
6275         set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
6276         em->map_lookup = map;
6277         em->start = logical;
6278         em->len = length;
6279         em->orig_start = 0;
6280         em->block_start = 0;
6281         em->block_len = em->len;
6282
6283         map->num_stripes = num_stripes;
6284         map->io_width = btrfs_chunk_io_width(leaf, chunk);
6285         map->io_align = btrfs_chunk_io_align(leaf, chunk);
6286         map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
6287         map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6288         map->type = btrfs_chunk_type(leaf, chunk);
6289         map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
6290         for (i = 0; i < num_stripes; i++) {
6291                 map->stripes[i].physical =
6292                         btrfs_stripe_offset_nr(leaf, chunk, i);
6293                 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
6294                 read_extent_buffer(leaf, uuid, (unsigned long)
6295                                    btrfs_stripe_dev_uuid_nr(chunk, i),
6296                                    BTRFS_UUID_SIZE);
6297                 map->stripes[i].dev = btrfs_find_device(root->fs_info, devid,
6298                                                         uuid, NULL);
6299                 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
6300                         free_extent_map(em);
6301                         return -EIO;
6302                 }
6303                 if (!map->stripes[i].dev) {
6304                         map->stripes[i].dev =
6305                                 add_missing_dev(root, root->fs_info->fs_devices,
6306                                                 devid, uuid);
6307                         if (!map->stripes[i].dev) {
6308                                 free_extent_map(em);
6309                                 return -EIO;
6310                         }
6311                         btrfs_warn(root->fs_info, "devid %llu uuid %pU is missing",
6312                                                 devid, uuid);
6313                 }
6314                 map->stripes[i].dev->in_fs_metadata = 1;
6315         }
6316
6317         write_lock(&map_tree->map_tree.lock);
6318         ret = add_extent_mapping(&map_tree->map_tree, em, 0);
6319         write_unlock(&map_tree->map_tree.lock);
6320         BUG_ON(ret); /* Tree corruption */
6321         free_extent_map(em);
6322
6323         return 0;
6324 }
6325
6326 static void fill_device_from_item(struct extent_buffer *leaf,
6327                                  struct btrfs_dev_item *dev_item,
6328                                  struct btrfs_device *device)
6329 {
6330         unsigned long ptr;
6331
6332         device->devid = btrfs_device_id(leaf, dev_item);
6333         device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
6334         device->total_bytes = device->disk_total_bytes;
6335         device->commit_total_bytes = device->disk_total_bytes;
6336         device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
6337         device->commit_bytes_used = device->bytes_used;
6338         device->type = btrfs_device_type(leaf, dev_item);
6339         device->io_align = btrfs_device_io_align(leaf, dev_item);
6340         device->io_width = btrfs_device_io_width(leaf, dev_item);
6341         device->sector_size = btrfs_device_sector_size(leaf, dev_item);
6342         WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
6343         device->is_tgtdev_for_dev_replace = 0;
6344
6345         ptr = btrfs_device_uuid(dev_item);
6346         read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
6347 }
6348
6349 static struct btrfs_fs_devices *open_seed_devices(struct btrfs_root *root,
6350                                                   u8 *fsid)
6351 {
6352         struct btrfs_fs_devices *fs_devices;
6353         int ret;
6354
6355         BUG_ON(!mutex_is_locked(&uuid_mutex));
6356
6357         fs_devices = root->fs_info->fs_devices->seed;
6358         while (fs_devices) {
6359                 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE))
6360                         return fs_devices;
6361
6362                 fs_devices = fs_devices->seed;
6363         }
6364
6365         fs_devices = find_fsid(fsid);
6366         if (!fs_devices) {
6367                 if (!btrfs_test_opt(root, DEGRADED))
6368                         return ERR_PTR(-ENOENT);
6369
6370                 fs_devices = alloc_fs_devices(fsid);
6371                 if (IS_ERR(fs_devices))
6372                         return fs_devices;
6373
6374                 fs_devices->seeding = 1;
6375                 fs_devices->opened = 1;
6376                 return fs_devices;
6377         }
6378
6379         fs_devices = clone_fs_devices(fs_devices);
6380         if (IS_ERR(fs_devices))
6381                 return fs_devices;
6382
6383         ret = __btrfs_open_devices(fs_devices, FMODE_READ,
6384                                    root->fs_info->bdev_holder);
6385         if (ret) {
6386                 free_fs_devices(fs_devices);
6387                 fs_devices = ERR_PTR(ret);
6388                 goto out;
6389         }
6390
6391         if (!fs_devices->seeding) {
6392                 __btrfs_close_devices(fs_devices);
6393                 free_fs_devices(fs_devices);
6394                 fs_devices = ERR_PTR(-EINVAL);
6395                 goto out;
6396         }
6397
6398         fs_devices->seed = root->fs_info->fs_devices->seed;
6399         root->fs_info->fs_devices->seed = fs_devices;
6400 out:
6401         return fs_devices;
6402 }
6403
6404 static int read_one_dev(struct btrfs_root *root,
6405                         struct extent_buffer *leaf,
6406                         struct btrfs_dev_item *dev_item)
6407 {
6408         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
6409         struct btrfs_device *device;
6410         u64 devid;
6411         int ret;
6412         u8 fs_uuid[BTRFS_UUID_SIZE];
6413         u8 dev_uuid[BTRFS_UUID_SIZE];
6414
6415         devid = btrfs_device_id(leaf, dev_item);
6416         read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
6417                            BTRFS_UUID_SIZE);
6418         read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
6419                            BTRFS_UUID_SIZE);
6420
6421         if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
6422                 fs_devices = open_seed_devices(root, fs_uuid);
6423                 if (IS_ERR(fs_devices))
6424                         return PTR_ERR(fs_devices);
6425         }
6426
6427         device = btrfs_find_device(root->fs_info, devid, dev_uuid, fs_uuid);
6428         if (!device) {
6429                 if (!btrfs_test_opt(root, DEGRADED))
6430                         return -EIO;
6431
6432                 device = add_missing_dev(root, fs_devices, devid, dev_uuid);
6433                 if (!device)
6434                         return -ENOMEM;
6435                 btrfs_warn(root->fs_info, "devid %llu uuid %pU missing",
6436                                 devid, dev_uuid);
6437         } else {
6438                 if (!device->bdev && !btrfs_test_opt(root, DEGRADED))
6439                         return -EIO;
6440
6441                 if(!device->bdev && !device->missing) {
6442                         /*
6443                          * this happens when a device that was properly setup
6444                          * in the device info lists suddenly goes bad.
6445                          * device->bdev is NULL, and so we have to set
6446                          * device->missing to one here
6447                          */
6448                         device->fs_devices->missing_devices++;
6449                         device->missing = 1;
6450                 }
6451
6452                 /* Move the device to its own fs_devices */
6453                 if (device->fs_devices != fs_devices) {
6454                         ASSERT(device->missing);
6455
6456                         list_move(&device->dev_list, &fs_devices->devices);
6457                         device->fs_devices->num_devices--;
6458                         fs_devices->num_devices++;
6459
6460                         device->fs_devices->missing_devices--;
6461                         fs_devices->missing_devices++;
6462
6463                         device->fs_devices = fs_devices;
6464                 }
6465         }
6466
6467         if (device->fs_devices != root->fs_info->fs_devices) {
6468                 BUG_ON(device->writeable);
6469                 if (device->generation !=
6470                     btrfs_device_generation(leaf, dev_item))
6471                         return -EINVAL;
6472         }
6473
6474         fill_device_from_item(leaf, dev_item, device);
6475         device->in_fs_metadata = 1;
6476         if (device->writeable && !device->is_tgtdev_for_dev_replace) {
6477                 device->fs_devices->total_rw_bytes += device->total_bytes;
6478                 spin_lock(&root->fs_info->free_chunk_lock);
6479                 root->fs_info->free_chunk_space += device->total_bytes -
6480                         device->bytes_used;
6481                 spin_unlock(&root->fs_info->free_chunk_lock);
6482         }
6483         ret = 0;
6484         return ret;
6485 }
6486
6487 int btrfs_read_sys_array(struct btrfs_root *root)
6488 {
6489         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
6490         struct extent_buffer *sb;
6491         struct btrfs_disk_key *disk_key;
6492         struct btrfs_chunk *chunk;
6493         u8 *array_ptr;
6494         unsigned long sb_array_offset;
6495         int ret = 0;
6496         u32 num_stripes;
6497         u32 array_size;
6498         u32 len = 0;
6499         u32 cur_offset;
6500         struct btrfs_key key;
6501
6502         ASSERT(BTRFS_SUPER_INFO_SIZE <= root->nodesize);
6503         /*
6504          * This will create extent buffer of nodesize, superblock size is
6505          * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will
6506          * overallocate but we can keep it as-is, only the first page is used.
6507          */
6508         sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET);
6509         if (!sb)
6510                 return -ENOMEM;
6511         set_extent_buffer_uptodate(sb);
6512         btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
6513         /*
6514          * The sb extent buffer is artifical and just used to read the system array.
6515          * set_extent_buffer_uptodate() call does not properly mark all it's
6516          * pages up-to-date when the page is larger: extent does not cover the
6517          * whole page and consequently check_page_uptodate does not find all
6518          * the page's extents up-to-date (the hole beyond sb),
6519          * write_extent_buffer then triggers a WARN_ON.
6520          *
6521          * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
6522          * but sb spans only this function. Add an explicit SetPageUptodate call
6523          * to silence the warning eg. on PowerPC 64.
6524          */
6525         if (PAGE_CACHE_SIZE > BTRFS_SUPER_INFO_SIZE)
6526                 SetPageUptodate(sb->pages[0]);
6527
6528         write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
6529         array_size = btrfs_super_sys_array_size(super_copy);
6530
6531         array_ptr = super_copy->sys_chunk_array;
6532         sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array);
6533         cur_offset = 0;
6534
6535         while (cur_offset < array_size) {
6536                 disk_key = (struct btrfs_disk_key *)array_ptr;
6537                 len = sizeof(*disk_key);
6538                 if (cur_offset + len > array_size)
6539                         goto out_short_read;
6540
6541                 btrfs_disk_key_to_cpu(&key, disk_key);
6542
6543                 array_ptr += len;
6544                 sb_array_offset += len;
6545                 cur_offset += len;
6546
6547                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
6548                         chunk = (struct btrfs_chunk *)sb_array_offset;
6549                         /*
6550                          * At least one btrfs_chunk with one stripe must be
6551                          * present, exact stripe count check comes afterwards
6552                          */
6553                         len = btrfs_chunk_item_size(1);
6554                         if (cur_offset + len > array_size)
6555                                 goto out_short_read;
6556
6557                         num_stripes = btrfs_chunk_num_stripes(sb, chunk);
6558                         if (!num_stripes) {
6559                                 printk(KERN_ERR
6560             "BTRFS: invalid number of stripes %u in sys_array at offset %u\n",
6561                                         num_stripes, cur_offset);
6562                                 ret = -EIO;
6563                                 break;
6564                         }
6565
6566                         len = btrfs_chunk_item_size(num_stripes);
6567                         if (cur_offset + len > array_size)
6568                                 goto out_short_read;
6569
6570                         ret = read_one_chunk(root, &key, sb, chunk);
6571                         if (ret)
6572                                 break;
6573                 } else {
6574                         printk(KERN_ERR
6575                 "BTRFS: unexpected item type %u in sys_array at offset %u\n",
6576                                 (u32)key.type, cur_offset);
6577                         ret = -EIO;
6578                         break;
6579                 }
6580                 array_ptr += len;
6581                 sb_array_offset += len;
6582                 cur_offset += len;
6583         }
6584         free_extent_buffer(sb);
6585         return ret;
6586
6587 out_short_read:
6588         printk(KERN_ERR "BTRFS: sys_array too short to read %u bytes at offset %u\n",
6589                         len, cur_offset);
6590         free_extent_buffer(sb);
6591         return -EIO;
6592 }
6593
6594 int btrfs_read_chunk_tree(struct btrfs_root *root)
6595 {
6596         struct btrfs_path *path;
6597         struct extent_buffer *leaf;
6598         struct btrfs_key key;
6599         struct btrfs_key found_key;
6600         int ret;
6601         int slot;
6602
6603         root = root->fs_info->chunk_root;
6604
6605         path = btrfs_alloc_path();
6606         if (!path)
6607                 return -ENOMEM;
6608
6609         mutex_lock(&uuid_mutex);
6610         lock_chunks(root);
6611
6612         /*
6613          * Read all device items, and then all the chunk items. All
6614          * device items are found before any chunk item (their object id
6615          * is smaller than the lowest possible object id for a chunk
6616          * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
6617          */
6618         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
6619         key.offset = 0;
6620         key.type = 0;
6621         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6622         if (ret < 0)
6623                 goto error;
6624         while (1) {
6625                 leaf = path->nodes[0];
6626                 slot = path->slots[0];
6627                 if (slot >= btrfs_header_nritems(leaf)) {
6628                         ret = btrfs_next_leaf(root, path);
6629                         if (ret == 0)
6630                                 continue;
6631                         if (ret < 0)
6632                                 goto error;
6633                         break;
6634                 }
6635                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6636                 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
6637                         struct btrfs_dev_item *dev_item;
6638                         dev_item = btrfs_item_ptr(leaf, slot,
6639                                                   struct btrfs_dev_item);
6640                         ret = read_one_dev(root, leaf, dev_item);
6641                         if (ret)
6642                                 goto error;
6643                 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
6644                         struct btrfs_chunk *chunk;
6645                         chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
6646                         ret = read_one_chunk(root, &found_key, leaf, chunk);
6647                         if (ret)
6648                                 goto error;
6649                 }
6650                 path->slots[0]++;
6651         }
6652         ret = 0;
6653 error:
6654         unlock_chunks(root);
6655         mutex_unlock(&uuid_mutex);
6656
6657         btrfs_free_path(path);
6658         return ret;
6659 }
6660
6661 void btrfs_init_devices_late(struct btrfs_fs_info *fs_info)
6662 {
6663         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6664         struct btrfs_device *device;
6665
6666         while (fs_devices) {
6667                 mutex_lock(&fs_devices->device_list_mutex);
6668                 list_for_each_entry(device, &fs_devices->devices, dev_list)
6669                         device->dev_root = fs_info->dev_root;
6670                 mutex_unlock(&fs_devices->device_list_mutex);
6671
6672                 fs_devices = fs_devices->seed;
6673         }
6674 }
6675
6676 static void __btrfs_reset_dev_stats(struct btrfs_device *dev)
6677 {
6678         int i;
6679
6680         for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6681                 btrfs_dev_stat_reset(dev, i);
6682 }
6683
6684 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
6685 {
6686         struct btrfs_key key;
6687         struct btrfs_key found_key;
6688         struct btrfs_root *dev_root = fs_info->dev_root;
6689         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6690         struct extent_buffer *eb;
6691         int slot;
6692         int ret = 0;
6693         struct btrfs_device *device;
6694         struct btrfs_path *path = NULL;
6695         int i;
6696
6697         path = btrfs_alloc_path();
6698         if (!path) {
6699                 ret = -ENOMEM;
6700                 goto out;
6701         }
6702
6703         mutex_lock(&fs_devices->device_list_mutex);
6704         list_for_each_entry(device, &fs_devices->devices, dev_list) {
6705                 int item_size;
6706                 struct btrfs_dev_stats_item *ptr;
6707
6708                 key.objectid = 0;
6709                 key.type = BTRFS_DEV_STATS_KEY;
6710                 key.offset = device->devid;
6711                 ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
6712                 if (ret) {
6713                         __btrfs_reset_dev_stats(device);
6714                         device->dev_stats_valid = 1;
6715                         btrfs_release_path(path);
6716                         continue;
6717                 }
6718                 slot = path->slots[0];
6719                 eb = path->nodes[0];
6720                 btrfs_item_key_to_cpu(eb, &found_key, slot);
6721                 item_size = btrfs_item_size_nr(eb, slot);
6722
6723                 ptr = btrfs_item_ptr(eb, slot,
6724                                      struct btrfs_dev_stats_item);
6725
6726                 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
6727                         if (item_size >= (1 + i) * sizeof(__le64))
6728                                 btrfs_dev_stat_set(device, i,
6729                                         btrfs_dev_stats_value(eb, ptr, i));
6730                         else
6731                                 btrfs_dev_stat_reset(device, i);
6732                 }
6733
6734                 device->dev_stats_valid = 1;
6735                 btrfs_dev_stat_print_on_load(device);
6736                 btrfs_release_path(path);
6737         }
6738         mutex_unlock(&fs_devices->device_list_mutex);
6739
6740 out:
6741         btrfs_free_path(path);
6742         return ret < 0 ? ret : 0;
6743 }
6744
6745 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
6746                                 struct btrfs_root *dev_root,
6747                                 struct btrfs_device *device)
6748 {
6749         struct btrfs_path *path;
6750         struct btrfs_key key;
6751         struct extent_buffer *eb;
6752         struct btrfs_dev_stats_item *ptr;
6753         int ret;
6754         int i;
6755
6756         key.objectid = 0;
6757         key.type = BTRFS_DEV_STATS_KEY;
6758         key.offset = device->devid;
6759
6760         path = btrfs_alloc_path();
6761         BUG_ON(!path);
6762         ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
6763         if (ret < 0) {
6764                 btrfs_warn_in_rcu(dev_root->fs_info,
6765                         "error %d while searching for dev_stats item for device %s",
6766                               ret, rcu_str_deref(device->name));
6767                 goto out;
6768         }
6769
6770         if (ret == 0 &&
6771             btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
6772                 /* need to delete old one and insert a new one */
6773                 ret = btrfs_del_item(trans, dev_root, path);
6774                 if (ret != 0) {
6775                         btrfs_warn_in_rcu(dev_root->fs_info,
6776                                 "delete too small dev_stats item for device %s failed %d",
6777                                       rcu_str_deref(device->name), ret);
6778                         goto out;
6779                 }
6780                 ret = 1;
6781         }
6782
6783         if (ret == 1) {
6784                 /* need to insert a new item */
6785                 btrfs_release_path(path);
6786                 ret = btrfs_insert_empty_item(trans, dev_root, path,
6787                                               &key, sizeof(*ptr));
6788                 if (ret < 0) {
6789                         btrfs_warn_in_rcu(dev_root->fs_info,
6790                                 "insert dev_stats item for device %s failed %d",
6791                                 rcu_str_deref(device->name), ret);
6792                         goto out;
6793                 }
6794         }
6795
6796         eb = path->nodes[0];
6797         ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
6798         for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6799                 btrfs_set_dev_stats_value(eb, ptr, i,
6800                                           btrfs_dev_stat_read(device, i));
6801         btrfs_mark_buffer_dirty(eb);
6802
6803 out:
6804         btrfs_free_path(path);
6805         return ret;
6806 }
6807
6808 /*
6809  * called from commit_transaction. Writes all changed device stats to disk.
6810  */
6811 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans,
6812                         struct btrfs_fs_info *fs_info)
6813 {
6814         struct btrfs_root *dev_root = fs_info->dev_root;
6815         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6816         struct btrfs_device *device;
6817         int stats_cnt;
6818         int ret = 0;
6819
6820         mutex_lock(&fs_devices->device_list_mutex);
6821         list_for_each_entry(device, &fs_devices->devices, dev_list) {
6822                 if (!device->dev_stats_valid || !btrfs_dev_stats_dirty(device))
6823                         continue;
6824
6825                 stats_cnt = atomic_read(&device->dev_stats_ccnt);
6826                 ret = update_dev_stat_item(trans, dev_root, device);
6827                 if (!ret)
6828                         atomic_sub(stats_cnt, &device->dev_stats_ccnt);
6829         }
6830         mutex_unlock(&fs_devices->device_list_mutex);
6831
6832         return ret;
6833 }
6834
6835 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
6836 {
6837         btrfs_dev_stat_inc(dev, index);
6838         btrfs_dev_stat_print_on_error(dev);
6839 }
6840
6841 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
6842 {
6843         if (!dev->dev_stats_valid)
6844                 return;
6845         btrfs_err_rl_in_rcu(dev->dev_root->fs_info,
6846                 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
6847                            rcu_str_deref(dev->name),
6848                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
6849                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
6850                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
6851                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
6852                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
6853 }
6854
6855 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
6856 {
6857         int i;
6858
6859         for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6860                 if (btrfs_dev_stat_read(dev, i) != 0)
6861                         break;
6862         if (i == BTRFS_DEV_STAT_VALUES_MAX)
6863                 return; /* all values == 0, suppress message */
6864
6865         btrfs_info_in_rcu(dev->dev_root->fs_info,
6866                 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
6867                rcu_str_deref(dev->name),
6868                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
6869                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
6870                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
6871                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
6872                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
6873 }
6874
6875 int btrfs_get_dev_stats(struct btrfs_root *root,
6876                         struct btrfs_ioctl_get_dev_stats *stats)
6877 {
6878         struct btrfs_device *dev;
6879         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
6880         int i;
6881
6882         mutex_lock(&fs_devices->device_list_mutex);
6883         dev = btrfs_find_device(root->fs_info, stats->devid, NULL, NULL);
6884         mutex_unlock(&fs_devices->device_list_mutex);
6885
6886         if (!dev) {
6887                 btrfs_warn(root->fs_info, "get dev_stats failed, device not found");
6888                 return -ENODEV;
6889         } else if (!dev->dev_stats_valid) {
6890                 btrfs_warn(root->fs_info, "get dev_stats failed, not yet valid");
6891                 return -ENODEV;
6892         } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
6893                 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
6894                         if (stats->nr_items > i)
6895                                 stats->values[i] =
6896                                         btrfs_dev_stat_read_and_reset(dev, i);
6897                         else
6898                                 btrfs_dev_stat_reset(dev, i);
6899                 }
6900         } else {
6901                 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6902                         if (stats->nr_items > i)
6903                                 stats->values[i] = btrfs_dev_stat_read(dev, i);
6904         }
6905         if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
6906                 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
6907         return 0;
6908 }
6909
6910 void btrfs_scratch_superblocks(struct block_device *bdev, char *device_path)
6911 {
6912         struct buffer_head *bh;
6913         struct btrfs_super_block *disk_super;
6914         int copy_num;
6915
6916         if (!bdev)
6917                 return;
6918
6919         for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX;
6920                 copy_num++) {
6921
6922                 if (btrfs_read_dev_one_super(bdev, copy_num, &bh))
6923                         continue;
6924
6925                 disk_super = (struct btrfs_super_block *)bh->b_data;
6926
6927                 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
6928                 set_buffer_dirty(bh);
6929                 sync_dirty_buffer(bh);
6930                 brelse(bh);
6931         }
6932
6933         /* Notify udev that device has changed */
6934         btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
6935
6936         /* Update ctime/mtime for device path for libblkid */
6937         update_dev_time(device_path);
6938 }
6939
6940 /*
6941  * Update the size of all devices, which is used for writing out the
6942  * super blocks.
6943  */
6944 void btrfs_update_commit_device_size(struct btrfs_fs_info *fs_info)
6945 {
6946         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6947         struct btrfs_device *curr, *next;
6948
6949         if (list_empty(&fs_devices->resized_devices))
6950                 return;
6951
6952         mutex_lock(&fs_devices->device_list_mutex);
6953         lock_chunks(fs_info->dev_root);
6954         list_for_each_entry_safe(curr, next, &fs_devices->resized_devices,
6955                                  resized_list) {
6956                 list_del_init(&curr->resized_list);
6957                 curr->commit_total_bytes = curr->disk_total_bytes;
6958         }
6959         unlock_chunks(fs_info->dev_root);
6960         mutex_unlock(&fs_devices->device_list_mutex);
6961 }
6962
6963 /* Must be invoked during the transaction commit */
6964 void btrfs_update_commit_device_bytes_used(struct btrfs_root *root,
6965                                         struct btrfs_transaction *transaction)
6966 {
6967         struct extent_map *em;
6968         struct map_lookup *map;
6969         struct btrfs_device *dev;
6970         int i;
6971
6972         if (list_empty(&transaction->pending_chunks))
6973                 return;
6974
6975         /* In order to kick the device replace finish process */
6976         lock_chunks(root);
6977         list_for_each_entry(em, &transaction->pending_chunks, list) {
6978                 map = em->map_lookup;
6979
6980                 for (i = 0; i < map->num_stripes; i++) {
6981                         dev = map->stripes[i].dev;
6982                         dev->commit_bytes_used = dev->bytes_used;
6983                 }
6984         }
6985         unlock_chunks(root);
6986 }
6987
6988 void btrfs_set_fs_info_ptr(struct btrfs_fs_info *fs_info)
6989 {
6990         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6991         while (fs_devices) {
6992                 fs_devices->fs_info = fs_info;
6993                 fs_devices = fs_devices->seed;
6994         }
6995 }
6996
6997 void btrfs_reset_fs_info_ptr(struct btrfs_fs_info *fs_info)
6998 {
6999         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7000         while (fs_devices) {
7001                 fs_devices->fs_info = NULL;
7002                 fs_devices = fs_devices->seed;
7003         }
7004 }
7005
7006 static void btrfs_close_one_device(struct btrfs_device *device)
7007 {
7008         struct btrfs_fs_devices *fs_devices = device->fs_devices;
7009         struct btrfs_device *new_device;
7010         struct rcu_string *name;
7011
7012         if (device->bdev)
7013                 fs_devices->open_devices--;
7014
7015         if (device->writeable &&
7016             device->devid != BTRFS_DEV_REPLACE_DEVID) {
7017                 list_del_init(&device->dev_alloc_list);
7018                 fs_devices->rw_devices--;
7019         }
7020
7021         if (device->missing)
7022                 fs_devices->missing_devices--;
7023
7024         new_device = btrfs_alloc_device(NULL, &device->devid,
7025                                         device->uuid);
7026         BUG_ON(IS_ERR(new_device)); /* -ENOMEM */
7027
7028         /* Safe because we are under uuid_mutex */
7029         if (device->name) {
7030                 name = rcu_string_strdup(device->name->str, GFP_NOFS);
7031                 BUG_ON(!name); /* -ENOMEM */
7032                 rcu_assign_pointer(new_device->name, name);
7033         }
7034
7035         list_replace_rcu(&device->dev_list, &new_device->dev_list);
7036         new_device->fs_devices = device->fs_devices;
7037
7038         call_rcu(&device->rcu, free_device);
7039 }