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