fs/btrfs/volumes.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18 #include <linux/sched.h>
  19 #include <linux/bio.h>
  20 #include <linux/slab.h>
  21 #include <linux/buffer_head.h>
  22 #include <linux/blkdev.h>
  23 #include <linux/random.h>
  24 #include <linux/iocontext.h>
  25 #include <linux/capability.h>
  26 #include <linux/ratelimit.h>
  27 #include <linux/kthread.h>
  28 #include <linux/raid/pq.h>
  29 #include <linux/semaphore.h>
  30 #include <asm/div64.h>
  31 #include "ctree.h"
  32 #include "extent_map.h"
  33 #include "disk-io.h"
  34 #include "transaction.h"
  35 #include "print-tree.h"
  36 #include "volumes.h"
  37 #include "raid56.h"
  38 #include "async-thread.h"
  39 #include "check-integrity.h"
  40 #include "rcu-string.h"
  41 #include "math.h"
  42 #include "dev-replace.h"
  43 #include "sysfs.h"
  44
  45 const struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = {
  46         [BTRFS_RAID_RAID10] = {
  47                 .sub_stripes    = 2,
  48                 .dev_stripes    = 1,
  49                 .devs_max       = 0,    /* 0 == as many as possible */
  50                 .devs_min       = 4,
  51                 .tolerated_failures = 1,
  52                 .devs_increment = 2,
  53                 .ncopies        = 2,
  54         },
  55         [BTRFS_RAID_RAID1] = {
  56                 .sub_stripes    = 1,
  57                 .dev_stripes    = 1,
  58                 .devs_max       = 2,
  59                 .devs_min       = 2,
  60                 .tolerated_failures = 1,
  61                 .devs_increment = 2,
  62                 .ncopies        = 2,
  63         },
  64         [BTRFS_RAID_DUP] = {
  65                 .sub_stripes    = 1,
  66                 .dev_stripes    = 2,
  67                 .devs_max       = 1,
  68                 .devs_min       = 1,
  69                 .tolerated_failures = 0,
  70                 .devs_increment = 1,
  71                 .ncopies        = 2,
  72         },
  73         [BTRFS_RAID_RAID0] = {
  74                 .sub_stripes    = 1,
  75                 .dev_stripes    = 1,
  76                 .devs_max       = 0,
  77                 .devs_min       = 2,
  78                 .tolerated_failures = 0,
  79                 .devs_increment = 1,
  80                 .ncopies        = 1,
  81         },
  82         [BTRFS_RAID_SINGLE] = {
  83                 .sub_stripes    = 1,
  84                 .dev_stripes    = 1,
  85                 .devs_max       = 1,
  86                 .devs_min       = 1,
  87                 .tolerated_failures = 0,
  88                 .devs_increment = 1,
  89                 .ncopies        = 1,
  90         },
  91         [BTRFS_RAID_RAID5] = {
  92                 .sub_stripes    = 1,
  93                 .dev_stripes    = 1,
  94                 .devs_max       = 0,
  95                 .devs_min       = 2,
  96                 .tolerated_failures = 1,
  97                 .devs_increment = 1,
  98                 .ncopies        = 2,
  99         },
 100         [BTRFS_RAID_RAID6] = {
 101                 .sub_stripes    = 1,
 102                 .dev_stripes    = 1,
 103                 .devs_max       = 0,
 104                 .devs_min       = 3,
 105                 .tolerated_failures = 2,
 106                 .devs_increment = 1,
 107                 .ncopies        = 3,
 108         },
 109 };
 110
 111 const u64 btrfs_raid_group[BTRFS_NR_RAID_TYPES] = {
 112         [BTRFS_RAID_RAID10] = BTRFS_BLOCK_GROUP_RAID10,
 113         [BTRFS_RAID_RAID1]  = BTRFS_BLOCK_GROUP_RAID1,
 114         [BTRFS_RAID_DUP]    = BTRFS_BLOCK_GROUP_DUP,
 115         [BTRFS_RAID_RAID0]  = BTRFS_BLOCK_GROUP_RAID0,
 116         [BTRFS_RAID_SINGLE] = 0,
 117         [BTRFS_RAID_RAID5]  = BTRFS_BLOCK_GROUP_RAID5,
 118         [BTRFS_RAID_RAID6]  = BTRFS_BLOCK_GROUP_RAID6,
 119 };
 120
 121 /*
 122  * Table to convert BTRFS_RAID_* to the error code if minimum number of devices
 123  * condition is not met. Zero means there's no corresponding
 124  * BTRFS_ERROR_DEV_*_NOT_MET value.
 125  */
 126 const int btrfs_raid_mindev_error[BTRFS_NR_RAID_TYPES] = {
 127         [BTRFS_RAID_RAID10] = BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET,
 128         [BTRFS_RAID_RAID1]  = BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET,
 129         [BTRFS_RAID_DUP]    = 0,
 130         [BTRFS_RAID_RAID0]  = 0,
 131         [BTRFS_RAID_SINGLE] = 0,
 132         [BTRFS_RAID_RAID5]  = BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET,
 133         [BTRFS_RAID_RAID6]  = BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET,
 134 };
 135
 136 static int init_first_rw_device(struct btrfs_trans_handle *trans,
 137                                 struct btrfs_root *root,
 138                                 struct btrfs_device *device);
 139 static int btrfs_relocate_sys_chunks(struct btrfs_root *root);
 140 static void __btrfs_reset_dev_stats(struct btrfs_device *dev);
 141 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev);
 142 static void btrfs_dev_stat_print_on_load(struct btrfs_device *device);
 143 static void btrfs_close_one_device(struct btrfs_device *device);
 144
 145 DEFINE_MUTEX(uuid_mutex);
 146 static LIST_HEAD(fs_uuids);
 147 struct list_head *btrfs_get_fs_uuids(void)
 148 {
 149         return &fs_uuids;
 150 }
 151
 152 static struct btrfs_fs_devices *__alloc_fs_devices(void)
 153 {
 154         struct btrfs_fs_devices *fs_devs;
 155
 156         fs_devs = kzalloc(sizeof(*fs_devs), GFP_KERNEL);
 157         if (!fs_devs)
 158                 return ERR_PTR(-ENOMEM);
 159
 160         mutex_init(&fs_devs->device_list_mutex);
 161
 162         INIT_LIST_HEAD(&fs_devs->devices);
 163         INIT_LIST_HEAD(&fs_devs->resized_devices);
 164         INIT_LIST_HEAD(&fs_devs->alloc_list);
 165         INIT_LIST_HEAD(&fs_devs->list);
 166
 167         return fs_devs;
 168 }
 169
 170 /**
 171  * alloc_fs_devices - allocate struct btrfs_fs_devices
 172  * @fsid:       a pointer to UUID for this FS.  If NULL a new UUID is
 173  *              generated.
 174  *
 175  * Return: a pointer to a new &struct btrfs_fs_devices on success;
 176  * ERR_PTR() on error.  Returned struct is not linked onto any lists and
 177  * can be destroyed with kfree() right away.
 178  */
 179 static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid)
 180 {
 181         struct btrfs_fs_devices *fs_devs;
 182
 183         fs_devs = __alloc_fs_devices();
 184         if (IS_ERR(fs_devs))
 185                 return fs_devs;
 186
 187         if (fsid)
 188                 memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE);
 189         else
 190                 generate_random_uuid(fs_devs->fsid);
 191
 192         return fs_devs;
 193 }
 194
 195 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
 196 {
 197         struct btrfs_device *device;
 198         WARN_ON(fs_devices->opened);
 199         while (!list_empty(&fs_devices->devices)) {
 200                 device = list_entry(fs_devices->devices.next,
 201                                     struct btrfs_device, dev_list);
 202                 list_del(&device->dev_list);
 203                 rcu_string_free(device->name);
 204                 kfree(device);
 205         }
 206         kfree(fs_devices);
 207 }
 208
 209 static void btrfs_kobject_uevent(struct block_device *bdev,
 210                                  enum kobject_action action)
 211 {
 212         int ret;
 213
 214         ret = kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, action);
 215         if (ret)
 216                 pr_warn("BTRFS: Sending event '%d' to kobject: '%s' (%p): failed\n",
 217                         action,
 218                         kobject_name(&disk_to_dev(bdev->bd_disk)->kobj),
 219                         &disk_to_dev(bdev->bd_disk)->kobj);
 220 }
 221
 222 void btrfs_cleanup_fs_uuids(void)
 223 {
 224         struct btrfs_fs_devices *fs_devices;
 225
 226         while (!list_empty(&fs_uuids)) {
 227                 fs_devices = list_entry(fs_uuids.next,
 228                                         struct btrfs_fs_devices, list);
 229                 list_del(&fs_devices->list);
 230                 free_fs_devices(fs_devices);
 231         }
 232 }
 233
 234 static struct btrfs_device *__alloc_device(void)
 235 {
 236         struct btrfs_device *dev;
 237
 238         dev = kzalloc(sizeof(*dev), GFP_KERNEL);
 239         if (!dev)
 240                 return ERR_PTR(-ENOMEM);
 241
 242         INIT_LIST_HEAD(&dev->dev_list);
 243         INIT_LIST_HEAD(&dev->dev_alloc_list);
 244         INIT_LIST_HEAD(&dev->resized_list);
 245
 246         spin_lock_init(&dev->io_lock);
 247
 248         spin_lock_init(&dev->reada_lock);
 249         atomic_set(&dev->reada_in_flight, 0);
 250         atomic_set(&dev->dev_stats_ccnt, 0);
 251         btrfs_device_data_ordered_init(dev);
 252         INIT_RADIX_TREE(&dev->reada_zones, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
 253         INIT_RADIX_TREE(&dev->reada_extents, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
 254
 255         return dev;
 256 }
 257
 258 static noinline struct btrfs_device *__find_device(struct list_head *head,
 259                                                    u64 devid, u8 *uuid)
 260 {
 261         struct btrfs_device *dev;
 262
 263         list_for_each_entry(dev, head, dev_list) {
 264                 if (dev->devid == devid &&
 265                     (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
 266                         return dev;
 267                 }
 268         }
 269         return NULL;
 270 }
 271
 272 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
 273 {
 274         struct btrfs_fs_devices *fs_devices;
 275
 276         list_for_each_entry(fs_devices, &fs_uuids, list) {
 277                 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
 278                         return fs_devices;
 279         }
 280         return NULL;
 281 }
 282
 283 static int
 284 btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder,
 285                       int flush, struct block_device **bdev,
 286                       struct buffer_head **bh)
 287 {
 288         int ret;
 289
 290         *bdev = blkdev_get_by_path(device_path, flags, holder);
 291
 292         if (IS_ERR(*bdev)) {
 293                 ret = PTR_ERR(*bdev);
 294                 goto error;
 295         }
 296
 297         if (flush)
 298                 filemap_write_and_wait((*bdev)->bd_inode->i_mapping);
 299         ret = set_blocksize(*bdev, 4096);
 300         if (ret) {
 301                 blkdev_put(*bdev, flags);
 302                 goto error;
 303         }
 304         invalidate_bdev(*bdev);
 305         *bh = btrfs_read_dev_super(*bdev);
 306         if (IS_ERR(*bh)) {
 307                 ret = PTR_ERR(*bh);
 308                 blkdev_put(*bdev, flags);
 309                 goto error;
 310         }
 311
 312         return 0;
 313
 314 error:
 315         *bdev = NULL;
 316         *bh = NULL;
 317         return ret;
 318 }
 319
 320 static void requeue_list(struct btrfs_pending_bios *pending_bios,
 321                         struct bio *head, struct bio *tail)
 322 {
 323
 324         struct bio *old_head;
 325
 326         old_head = pending_bios->head;
 327         pending_bios->head = head;
 328         if (pending_bios->tail)
 329                 tail->bi_next = old_head;
 330         else
 331                 pending_bios->tail = tail;
 332 }
 333
 334 /*
 335  * we try to collect pending bios for a device so we don't get a large
 336  * number of procs sending bios down to the same device.  This greatly
 337  * improves the schedulers ability to collect and merge the bios.
 338  *
 339  * But, it also turns into a long list of bios to process and that is sure
 340  * to eventually make the worker thread block.  The solution here is to
 341  * make some progress and then put this work struct back at the end of
 342  * the list if the block device is congested.  This way, multiple devices
 343  * can make progress from a single worker thread.
 344  */
 345 static noinline void run_scheduled_bios(struct btrfs_device *device)
 346 {
 347         struct bio *pending;
 348         struct backing_dev_info *bdi;
 349         struct btrfs_fs_info *fs_info;
 350         struct btrfs_pending_bios *pending_bios;
 351         struct bio *tail;
 352         struct bio *cur;
 353         int again = 0;
 354         unsigned long num_run;
 355         unsigned long batch_run = 0;
 356         unsigned long limit;
 357         unsigned long last_waited = 0;
 358         int force_reg = 0;
 359         int sync_pending = 0;
 360         struct blk_plug plug;
 361
 362         /*
 363          * this function runs all the bios we've collected for
 364          * a particular device.  We don't want to wander off to
 365          * another device without first sending all of these down.
 366          * So, setup a plug here and finish it off before we return
 367          */
 368         blk_start_plug(&plug);
 369
 370         bdi = blk_get_backing_dev_info(device->bdev);
 371         fs_info = device->dev_root->fs_info;
 372         limit = btrfs_async_submit_limit(fs_info);
 373         limit = limit * 2 / 3;
 374
 375 loop:
 376         spin_lock(&device->io_lock);
 377
 378 loop_lock:
 379         num_run = 0;
 380
 381         /* take all the bios off the list at once and process them
 382          * later on (without the lock held).  But, remember the
 383          * tail and other pointers so the bios can be properly reinserted
 384          * into the list if we hit congestion
 385          */
 386         if (!force_reg && device->pending_sync_bios.head) {
 387                 pending_bios = &device->pending_sync_bios;
 388                 force_reg = 1;
 389         } else {
 390                 pending_bios = &device->pending_bios;
 391                 force_reg = 0;
 392         }
 393
 394         pending = pending_bios->head;
 395         tail = pending_bios->tail;
 396         WARN_ON(pending && !tail);
 397
 398         /*
 399          * if pending was null this time around, no bios need processing
 400          * at all and we can stop.  Otherwise it'll loop back up again
 401          * and do an additional check so no bios are missed.
 402          *
 403          * device->running_pending is used to synchronize with the
 404          * schedule_bio code.
 405          */
 406         if (device->pending_sync_bios.head == NULL &&
 407             device->pending_bios.head == NULL) {
 408                 again = 0;
 409                 device->running_pending = 0;
 410         } else {
 411                 again = 1;
 412                 device->running_pending = 1;
 413         }
 414
 415         pending_bios->head = NULL;
 416         pending_bios->tail = NULL;
 417
 418         spin_unlock(&device->io_lock);
 419
 420         while (pending) {
 421
 422                 rmb();
 423                 /* we want to work on both lists, but do more bios on the
 424                  * sync list than the regular list
 425                  */
 426                 if ((num_run > 32 &&
 427                     pending_bios != &device->pending_sync_bios &&
 428                     device->pending_sync_bios.head) ||
 429                    (num_run > 64 && pending_bios == &device->pending_sync_bios &&
 430                     device->pending_bios.head)) {
 431                         spin_lock(&device->io_lock);
 432                         requeue_list(pending_bios, pending, tail);
 433                         goto loop_lock;
 434                 }
 435
 436                 cur = pending;
 437                 pending = pending->bi_next;
 438                 cur->bi_next = NULL;
 439
 440                 /*
 441                  * atomic_dec_return implies a barrier for waitqueue_active
 442                  */
 443                 if (atomic_dec_return(&fs_info->nr_async_bios) < limit &&
 444                     waitqueue_active(&fs_info->async_submit_wait))
 445                         wake_up(&fs_info->async_submit_wait);
 446
 447                 BUG_ON(atomic_read(&cur->__bi_cnt) == 0);
 448
 449                 /*
 450                  * if we're doing the sync list, record that our
 451                  * plug has some sync requests on it
 452                  *
 453                  * If we're doing the regular list and there are
 454                  * sync requests sitting around, unplug before
 455                  * we add more
 456                  */
 457                 if (pending_bios == &device->pending_sync_bios) {
 458                         sync_pending = 1;
 459                 } else if (sync_pending) {
 460                         blk_finish_plug(&plug);
 461                         blk_start_plug(&plug);
 462                         sync_pending = 0;
 463                 }
 464
 465                 btrfsic_submit_bio(cur->bi_rw, cur);
 466                 num_run++;
 467                 batch_run++;
 468
 469                 cond_resched();
 470
 471                 /*
 472                  * we made progress, there is more work to do and the bdi
 473                  * is now congested.  Back off and let other work structs
 474                  * run instead
 475                  */
 476                 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
 477                     fs_info->fs_devices->open_devices > 1) {
 478                         struct io_context *ioc;
 479
 480                         ioc = current->io_context;
 481
 482                         /*
 483                          * the main goal here is that we don't want to
 484                          * block if we're going to be able to submit
 485                          * more requests without blocking.
 486                          *
 487                          * This code does two great things, it pokes into
 488                          * the elevator code from a filesystem _and_
 489                          * it makes assumptions about how batching works.
 490                          */
 491                         if (ioc && ioc->nr_batch_requests > 0 &&
 492                             time_before(jiffies, ioc->last_waited + HZ/50UL) &&
 493                             (last_waited == 0 ||
 494                              ioc->last_waited == last_waited)) {
 495                                 /*
 496                                  * we want to go through our batch of
 497                                  * requests and stop.  So, we copy out
 498                                  * the ioc->last_waited time and test
 499                                  * against it before looping
 500                                  */
 501                                 last_waited = ioc->last_waited;
 502                                 cond_resched();
 503                                 continue;
 504                         }
 505                         spin_lock(&device->io_lock);
 506                         requeue_list(pending_bios, pending, tail);
 507                         device->running_pending = 1;
 508
 509                         spin_unlock(&device->io_lock);
 510                         btrfs_queue_work(fs_info->submit_workers,
 511                                          &device->work);
 512                         goto done;
 513                 }
 514                 /* unplug every 64 requests just for good measure */
 515                 if (batch_run % 64 == 0) {
 516                         blk_finish_plug(&plug);
 517                         blk_start_plug(&plug);
 518                         sync_pending = 0;
 519                 }
 520         }
 521
 522         cond_resched();
 523         if (again)
 524                 goto loop;
 525
 526         spin_lock(&device->io_lock);
 527         if (device->pending_bios.head || device->pending_sync_bios.head)
 528                 goto loop_lock;
 529         spin_unlock(&device->io_lock);
 530
 531 done:
 532         blk_finish_plug(&plug);
 533 }
 534
 535 static void pending_bios_fn(struct btrfs_work *work)
 536 {
 537         struct btrfs_device *device;
 538
 539         device = container_of(work, struct btrfs_device, work);
 540         run_scheduled_bios(device);
 541 }
 542
 543
 544 void btrfs_free_stale_device(struct btrfs_device *cur_dev)
 545 {
 546         struct btrfs_fs_devices *fs_devs;
 547         struct btrfs_device *dev;
 548
 549         if (!cur_dev->name)
 550                 return;
 551
 552         list_for_each_entry(fs_devs, &fs_uuids, list) {
 553                 int del = 1;
 554
 555                 if (fs_devs->opened)
 556                         continue;
 557                 if (fs_devs->seeding)
 558                         continue;
 559
 560                 list_for_each_entry(dev, &fs_devs->devices, dev_list) {
 561
 562                         if (dev == cur_dev)
 563                                 continue;
 564                         if (!dev->name)
 565                                 continue;
 566
 567                         /*
 568                          * Todo: This won't be enough. What if the same device
 569                          * comes back (with new uuid and) with its mapper path?
 570                          * But for now, this does help as mostly an admin will
 571                          * either use mapper or non mapper path throughout.
 572                          */
 573                         rcu_read_lock();
 574                         del = strcmp(rcu_str_deref(dev->name),
 575                                                 rcu_str_deref(cur_dev->name));
 576                         rcu_read_unlock();
 577                         if (!del)
 578                                 break;
 579                 }
 580
 581                 if (!del) {
 582                         /* delete the stale device */
 583                         if (fs_devs->num_devices == 1) {
 584                                 btrfs_sysfs_remove_fsid(fs_devs);
 585                                 list_del(&fs_devs->list);
 586                                 free_fs_devices(fs_devs);
 587                         } else {
 588                                 fs_devs->num_devices--;
 589                                 list_del(&dev->dev_list);
 590                                 rcu_string_free(dev->name);
 591                                 kfree(dev);
 592                         }
 593                         break;
 594                 }
 595         }
 596 }
 597
 598 /*
 599  * Add new device to list of registered devices
 600  *
 601  * Returns:
 602  * 1   - first time device is seen
 603  * 0   - device already known
 604  * < 0 - error
 605  */
 606 static noinline int device_list_add(const char *path,
 607                            struct btrfs_super_block *disk_super,
 608                            u64 devid, struct btrfs_fs_devices **fs_devices_ret)
 609 {
 610         struct btrfs_device *device;
 611         struct btrfs_fs_devices *fs_devices;
 612         struct rcu_string *name;
 613         int ret = 0;
 614         u64 found_transid = btrfs_super_generation(disk_super);
 615
 616         fs_devices = find_fsid(disk_super->fsid);
 617         if (!fs_devices) {
 618                 fs_devices = alloc_fs_devices(disk_super->fsid);
 619                 if (IS_ERR(fs_devices))
 620                         return PTR_ERR(fs_devices);
 621
 622                 list_add(&fs_devices->list, &fs_uuids);
 623
 624                 device = NULL;
 625         } else {
 626                 device = __find_device(&fs_devices->devices, devid,
 627                                        disk_super->dev_item.uuid);
 628         }
 629
 630         if (!device) {
 631                 if (fs_devices->opened)
 632                         return -EBUSY;
 633
 634                 device = btrfs_alloc_device(NULL, &devid,
 635                                             disk_super->dev_item.uuid);
 636                 if (IS_ERR(device)) {
 637                         /* we can safely leave the fs_devices entry around */
 638                         return PTR_ERR(device);
 639                 }
 640
 641                 name = rcu_string_strdup(path, GFP_NOFS);
 642                 if (!name) {
 643                         kfree(device);
 644                         return -ENOMEM;
 645                 }
 646                 rcu_assign_pointer(device->name, name);
 647
 648                 mutex_lock(&fs_devices->device_list_mutex);
 649                 list_add_rcu(&device->dev_list, &fs_devices->devices);
 650                 fs_devices->num_devices++;
 651                 mutex_unlock(&fs_devices->device_list_mutex);
 652
 653                 ret = 1;
 654                 device->fs_devices = fs_devices;
 655         } else if (!device->name || strcmp(device->name->str, path)) {
 656                 /*
 657                  * When FS is already mounted.
 658                  * 1. If you are here and if the device->name is NULL that
 659                  *    means this device was missing at time of FS mount.
 660                  * 2. If you are here and if the device->name is different
 661                  *    from 'path' that means either
 662                  *      a. The same device disappeared and reappeared with
 663                  *         different name. or
 664                  *      b. The missing-disk-which-was-replaced, has
 665                  *         reappeared now.
 666                  *
 667                  * We must allow 1 and 2a above. But 2b would be a spurious
 668                  * and unintentional.
 669                  *
 670                  * Further in case of 1 and 2a above, the disk at 'path'
 671                  * would have missed some transaction when it was away and
 672                  * in case of 2a the stale bdev has to be updated as well.
 673                  * 2b must not be allowed at all time.
 674                  */
 675
 676                 /*
 677                  * For now, we do allow update to btrfs_fs_device through the
 678                  * btrfs dev scan cli after FS has been mounted.  We're still
 679                  * tracking a problem where systems fail mount by subvolume id
 680                  * when we reject replacement on a mounted FS.
 681                  */
 682                 if (!fs_devices->opened && found_transid < device->generation) {
 683                         /*
 684                          * That is if the FS is _not_ mounted and if you
 685                          * are here, that means there is more than one
 686                          * disk with same uuid and devid.We keep the one
 687                          * with larger generation number or the last-in if
 688                          * generation are equal.
 689                          */
 690                         return -EEXIST;
 691                 }
 692
 693                 name = rcu_string_strdup(path, GFP_NOFS);
 694                 if (!name)
 695                         return -ENOMEM;
 696                 rcu_string_free(device->name);
 697                 rcu_assign_pointer(device->name, name);
 698                 if (device->missing) {
 699                         fs_devices->missing_devices--;
 700                         device->missing = 0;
 701                 }
 702         }
 703
 704         /*
 705          * Unmount does not free the btrfs_device struct but would zero
 706          * generation along with most of the other members. So just update
 707          * it back. We need it to pick the disk with largest generation
 708          * (as above).
 709          */
 710         if (!fs_devices->opened)
 711                 device->generation = found_transid;
 712
 713         /*
 714          * if there is new btrfs on an already registered device,
 715          * then remove the stale device entry.
 716          */
 717         if (ret > 0)
 718                 btrfs_free_stale_device(device);
 719
 720         *fs_devices_ret = fs_devices;
 721
 722         return ret;
 723 }
 724
 725 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
 726 {
 727         struct btrfs_fs_devices *fs_devices;
 728         struct btrfs_device *device;
 729         struct btrfs_device *orig_dev;
 730
 731         fs_devices = alloc_fs_devices(orig->fsid);
 732         if (IS_ERR(fs_devices))
 733                 return fs_devices;
 734
 735         mutex_lock(&orig->device_list_mutex);
 736         fs_devices->total_devices = orig->total_devices;
 737
 738         /* We have held the volume lock, it is safe to get the devices. */
 739         list_for_each_entry(orig_dev, &orig->devices, dev_list) {
 740                 struct rcu_string *name;
 741
 742                 device = btrfs_alloc_device(NULL, &orig_dev->devid,
 743                                             orig_dev->uuid);
 744                 if (IS_ERR(device))
 745                         goto error;
 746
 747                 /*
 748                  * This is ok to do without rcu read locked because we hold the
 749                  * uuid mutex so nothing we touch in here is going to disappear.
 750                  */
 751                 if (orig_dev->name) {
 752                         name = rcu_string_strdup(orig_dev->name->str,
 753                                         GFP_KERNEL);
 754                         if (!name) {
 755                                 kfree(device);
 756                                 goto error;
 757                         }
 758                         rcu_assign_pointer(device->name, name);
 759                 }
 760
 761                 list_add(&device->dev_list, &fs_devices->devices);
 762                 device->fs_devices = fs_devices;
 763                 fs_devices->num_devices++;
 764         }
 765         mutex_unlock(&orig->device_list_mutex);
 766         return fs_devices;
 767 error:
 768         mutex_unlock(&orig->device_list_mutex);
 769         free_fs_devices(fs_devices);
 770         return ERR_PTR(-ENOMEM);
 771 }
 772
 773 void btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices, int step)
 774 {
 775         struct btrfs_device *device, *next;
 776         struct btrfs_device *latest_dev = NULL;
 777
 778         mutex_lock(&uuid_mutex);
 779 again:
 780         /* This is the initialized path, it is safe to release the devices. */
 781         list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
 782                 if (device->in_fs_metadata) {
 783                         if (!device->is_tgtdev_for_dev_replace &&
 784                             (!latest_dev ||
 785                              device->generation > latest_dev->generation)) {
 786                                 latest_dev = device;
 787                         }
 788                         continue;
 789                 }
 790
 791                 if (device->devid == BTRFS_DEV_REPLACE_DEVID) {
 792                         /*
 793                          * In the first step, keep the device which has
 794                          * the correct fsid and the devid that is used
 795                          * for the dev_replace procedure.
 796                          * In the second step, the dev_replace state is
 797                          * read from the device tree and it is known
 798                          * whether the procedure is really active or
 799                          * not, which means whether this device is
 800                          * used or whether it should be removed.
 801                          */
 802                         if (step == 0 || device->is_tgtdev_for_dev_replace) {
 803                                 continue;
 804                         }
 805                 }
 806                 if (device->bdev) {
 807                         blkdev_put(device->bdev, device->mode);
 808                         device->bdev = NULL;
 809                         fs_devices->open_devices--;
 810                 }
 811                 if (device->writeable) {
 812                         list_del_init(&device->dev_alloc_list);
 813                         device->writeable = 0;
 814                         if (!device->is_tgtdev_for_dev_replace)
 815                                 fs_devices->rw_devices--;
 816                 }
 817                 list_del_init(&device->dev_list);
 818                 fs_devices->num_devices--;
 819                 rcu_string_free(device->name);
 820                 kfree(device);
 821         }
 822
 823         if (fs_devices->seed) {
 824                 fs_devices = fs_devices->seed;
 825                 goto again;
 826         }
 827
 828         fs_devices->latest_bdev = latest_dev->bdev;
 829
 830         mutex_unlock(&uuid_mutex);
 831 }
 832
 833 static void __free_device(struct work_struct *work)
 834 {
 835         struct btrfs_device *device;
 836
 837         device = container_of(work, struct btrfs_device, rcu_work);
 838
 839         if (device->bdev)
 840                 blkdev_put(device->bdev, device->mode);
 841
 842         rcu_string_free(device->name);
 843         kfree(device);
 844 }
 845
 846 static void free_device(struct rcu_head *head)
 847 {
 848         struct btrfs_device *device;
 849
 850         device = container_of(head, struct btrfs_device, rcu);
 851
 852         INIT_WORK(&device->rcu_work, __free_device);
 853         schedule_work(&device->rcu_work);
 854 }
 855
 856 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
 857 {
 858         struct btrfs_device *device, *tmp;
 859
 860         if (--fs_devices->opened > 0)
 861                 return 0;
 862
 863         mutex_lock(&fs_devices->device_list_mutex);
 864         list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list) {
 865                 btrfs_close_one_device(device);
 866         }
 867         mutex_unlock(&fs_devices->device_list_mutex);
 868
 869         WARN_ON(fs_devices->open_devices);
 870         WARN_ON(fs_devices->rw_devices);
 871         fs_devices->opened = 0;
 872         fs_devices->seeding = 0;
 873
 874         return 0;
 875 }
 876
 877 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
 878 {
 879         struct btrfs_fs_devices *seed_devices = NULL;
 880         int ret;
 881
 882         mutex_lock(&uuid_mutex);
 883         ret = __btrfs_close_devices(fs_devices);
 884         if (!fs_devices->opened) {
 885                 seed_devices = fs_devices->seed;
 886                 fs_devices->seed = NULL;
 887         }
 888         mutex_unlock(&uuid_mutex);
 889
 890         while (seed_devices) {
 891                 fs_devices = seed_devices;
 892                 seed_devices = fs_devices->seed;
 893                 __btrfs_close_devices(fs_devices);
 894                 free_fs_devices(fs_devices);
 895         }
 896         /*
 897          * Wait for rcu kworkers under __btrfs_close_devices
 898          * to finish all blkdev_puts so device is really
 899          * free when umount is done.
 900          */
 901         rcu_barrier();
 902         return ret;
 903 }
 904
 905 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
 906                                 fmode_t flags, void *holder)
 907 {
 908         struct request_queue *q;
 909         struct block_device *bdev;
 910         struct list_head *head = &fs_devices->devices;
 911         struct btrfs_device *device;
 912         struct btrfs_device *latest_dev = NULL;
 913         struct buffer_head *bh;
 914         struct btrfs_super_block *disk_super;
 915         u64 devid;
 916         int seeding = 1;
 917         int ret = 0;
 918
 919         flags |= FMODE_EXCL;
 920
 921         list_for_each_entry(device, head, dev_list) {
 922                 if (device->bdev)
 923                         continue;
 924                 if (!device->name)
 925                         continue;
 926
 927                 /* Just open everything we can; ignore failures here */
 928                 if (btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1,
 929                                             &bdev, &bh))
 930                         continue;
 931
 932                 disk_super = (struct btrfs_super_block *)bh->b_data;
 933                 devid = btrfs_stack_device_id(&disk_super->dev_item);
 934                 if (devid != device->devid)
 935                         goto error_brelse;
 936
 937                 if (memcmp(device->uuid, disk_super->dev_item.uuid,
 938                            BTRFS_UUID_SIZE))
 939                         goto error_brelse;
 940
 941                 device->generation = btrfs_super_generation(disk_super);
 942                 if (!latest_dev ||
 943                     device->generation > latest_dev->generation)
 944                         latest_dev = device;
 945
 946                 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
 947                         device->writeable = 0;
 948                 } else {
 949                         device->writeable = !bdev_read_only(bdev);
 950                         seeding = 0;
 951                 }
 952
 953                 q = bdev_get_queue(bdev);
 954                 if (blk_queue_discard(q))
 955                         device->can_discard = 1;
 956
 957                 device->bdev = bdev;
 958                 device->in_fs_metadata = 0;
 959                 device->mode = flags;
 960
 961                 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
 962                         fs_devices->rotating = 1;
 963
 964                 fs_devices->open_devices++;
 965                 if (device->writeable &&
 966                     device->devid != BTRFS_DEV_REPLACE_DEVID) {
 967                         fs_devices->rw_devices++;
 968                         list_add(&device->dev_alloc_list,
 969                                  &fs_devices->alloc_list);
 970                 }
 971                 brelse(bh);
 972                 continue;
 973
 974 error_brelse:
 975                 brelse(bh);
 976                 blkdev_put(bdev, flags);
 977                 continue;
 978         }
 979         if (fs_devices->open_devices == 0) {
 980                 ret = -EINVAL;
 981                 goto out;
 982         }
 983         fs_devices->seeding = seeding;
 984         fs_devices->opened = 1;
 985         fs_devices->latest_bdev = latest_dev->bdev;
 986         fs_devices->total_rw_bytes = 0;
 987 out:
 988         return ret;
 989 }
 990
 991 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
 992                        fmode_t flags, void *holder)
 993 {
 994         int ret;
 995
 996         mutex_lock(&uuid_mutex);
 997         if (fs_devices->opened) {
 998                 fs_devices->opened++;
 999                 ret = 0;
1000         } else {
1001                 ret = __btrfs_open_devices(fs_devices, flags, holder);
1002         }
1003         mutex_unlock(&uuid_mutex);
1004         return ret;
1005 }
1006
1007 void btrfs_release_disk_super(struct page *page)
1008 {
1009         kunmap(page);
1010         put_page(page);
1011 }
1012
1013 int btrfs_read_disk_super(struct block_device *bdev, u64 bytenr,
1014                 struct page **page, struct btrfs_super_block **disk_super)
1015 {
1016         void *p;
1017         pgoff_t index;
1018
1019         /* make sure our super fits in the device */
1020         if (bytenr + PAGE_SIZE >= i_size_read(bdev->bd_inode))
1021                 return 1;
1022
1023         /* make sure our super fits in the page */
1024         if (sizeof(**disk_super) > PAGE_SIZE)
1025                 return 1;
1026
1027         /* make sure our super doesn't straddle pages on disk */
1028         index = bytenr >> PAGE_SHIFT;
1029         if ((bytenr + sizeof(**disk_super) - 1) >> PAGE_SHIFT != index)
1030                 return 1;
1031
1032         /* pull in the page with our super */
1033         *page = read_cache_page_gfp(bdev->bd_inode->i_mapping,
1034                                    index, GFP_KERNEL);
1035
1036         if (IS_ERR_OR_NULL(*page))
1037                 return 1;
1038
1039         p = kmap(*page);
1040
1041         /* align our pointer to the offset of the super block */
1042         *disk_super = p + (bytenr & ~PAGE_MASK);
1043
1044         if (btrfs_super_bytenr(*disk_super) != bytenr ||
1045             btrfs_super_magic(*disk_super) != BTRFS_MAGIC) {
1046                 btrfs_release_disk_super(*page);
1047                 return 1;
1048         }
1049
1050         if ((*disk_super)->label[0] &&
1051                 (*disk_super)->label[BTRFS_LABEL_SIZE - 1])
1052                 (*disk_super)->label[BTRFS_LABEL_SIZE - 1] = '\0';
1053
1054         return 0;
1055 }
1056
1057 /*
1058  * Look for a btrfs signature on a device. This may be called out of the mount path
1059  * and we are not allowed to call set_blocksize during the scan. The superblock
1060  * is read via pagecache
1061  */
1062 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
1063                           struct btrfs_fs_devices **fs_devices_ret)
1064 {
1065         struct btrfs_super_block *disk_super;
1066         struct block_device *bdev;
1067         struct page *page;
1068         int ret = -EINVAL;
1069         u64 devid;
1070         u64 transid;
1071         u64 total_devices;
1072         u64 bytenr;
1073
1074         /*
1075          * we would like to check all the supers, but that would make
1076          * a btrfs mount succeed after a mkfs from a different FS.
1077          * So, we need to add a special mount option to scan for
1078          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
1079          */
1080         bytenr = btrfs_sb_offset(0);
1081         flags |= FMODE_EXCL;
1082         mutex_lock(&uuid_mutex);
1083
1084         bdev = blkdev_get_by_path(path, flags, holder);
1085         if (IS_ERR(bdev)) {
1086                 ret = PTR_ERR(bdev);
1087                 goto error;
1088         }
1089
1090         if (btrfs_read_disk_super(bdev, bytenr, &page, &disk_super))
1091                 goto error_bdev_put;
1092
1093         devid = btrfs_stack_device_id(&disk_super->dev_item);
1094         transid = btrfs_super_generation(disk_super);
1095         total_devices = btrfs_super_num_devices(disk_super);
1096
1097         ret = device_list_add(path, disk_super, devid, fs_devices_ret);
1098         if (ret > 0) {
1099                 if (disk_super->label[0]) {
1100                         printk(KERN_INFO "BTRFS: device label %s ", disk_super->label);
1101                 } else {
1102                         printk(KERN_INFO "BTRFS: device fsid %pU ", disk_super->fsid);
1103                 }
1104
1105                 printk(KERN_CONT "devid %llu transid %llu %s\n", devid, transid, path);
1106                 ret = 0;
1107         }
1108         if (!ret && fs_devices_ret)
1109                 (*fs_devices_ret)->total_devices = total_devices;
1110
1111         btrfs_release_disk_super(page);
1112
1113 error_bdev_put:
1114         blkdev_put(bdev, flags);
1115 error:
1116         mutex_unlock(&uuid_mutex);
1117         return ret;
1118 }
1119
1120 /* helper to account the used device space in the range */
1121 int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
1122                                    u64 end, u64 *length)
1123 {
1124         struct btrfs_key key;
1125         struct btrfs_root *root = device->dev_root;
1126         struct btrfs_dev_extent *dev_extent;
1127         struct btrfs_path *path;
1128         u64 extent_end;
1129         int ret;
1130         int slot;
1131         struct extent_buffer *l;
1132
1133         *length = 0;
1134
1135         if (start >= device->total_bytes || device->is_tgtdev_for_dev_replace)
1136                 return 0;
1137
1138         path = btrfs_alloc_path();
1139         if (!path)
1140                 return -ENOMEM;
1141         path->reada = READA_FORWARD;
1142
1143         key.objectid = device->devid;
1144         key.offset = start;
1145         key.type = BTRFS_DEV_EXTENT_KEY;
1146
1147         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1148         if (ret < 0)
1149                 goto out;
1150         if (ret > 0) {
1151                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1152                 if (ret < 0)
1153                         goto out;
1154         }
1155
1156         while (1) {
1157                 l = path->nodes[0];
1158                 slot = path->slots[0];
1159                 if (slot >= btrfs_header_nritems(l)) {
1160                         ret = btrfs_next_leaf(root, path);
1161                         if (ret == 0)
1162                                 continue;
1163                         if (ret < 0)
1164                                 goto out;
1165
1166                         break;
1167                 }
1168                 btrfs_item_key_to_cpu(l, &key, slot);
1169
1170                 if (key.objectid < device->devid)
1171                         goto next;
1172
1173                 if (key.objectid > device->devid)
1174                         break;
1175
1176                 if (key.type != BTRFS_DEV_EXTENT_KEY)
1177                         goto next;
1178
1179                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1180                 extent_end = key.offset + btrfs_dev_extent_length(l,
1181                                                                   dev_extent);
1182                 if (key.offset <= start && extent_end > end) {
1183                         *length = end - start + 1;
1184                         break;
1185                 } else if (key.offset <= start && extent_end > start)
1186                         *length += extent_end - start;
1187                 else if (key.offset > start && extent_end <= end)
1188                         *length += extent_end - key.offset;
1189                 else if (key.offset > start && key.offset <= end) {
1190                         *length += end - key.offset + 1;
1191                         break;
1192                 } else if (key.offset > end)
1193                         break;
1194
1195 next:
1196                 path->slots[0]++;
1197         }
1198         ret = 0;
1199 out:
1200         btrfs_free_path(path);
1201         return ret;
1202 }
1203
1204 static int contains_pending_extent(struct btrfs_transaction *transaction,
1205                                    struct btrfs_device *device,
1206                                    u64 *start, u64 len)
1207 {
1208         struct btrfs_fs_info *fs_info = device->dev_root->fs_info;
1209         struct extent_map *em;
1210         struct list_head *search_list = &fs_info->pinned_chunks;
1211         int ret = 0;
1212         u64 physical_start = *start;
1213
1214         if (transaction)
1215                 search_list = &transaction->pending_chunks;
1216 again:
1217         list_for_each_entry(em, search_list, list) {
1218                 struct map_lookup *map;
1219                 int i;
1220
1221                 map = em->map_lookup;
1222                 for (i = 0; i < map->num_stripes; i++) {
1223                         u64 end;
1224
1225                         if (map->stripes[i].dev != device)
1226                                 continue;
1227                         if (map->stripes[i].physical >= physical_start + len ||
1228                             map->stripes[i].physical + em->orig_block_len <=
1229                             physical_start)
1230                                 continue;
1231                         /*
1232                          * Make sure that while processing the pinned list we do
1233                          * not override our *start with a lower value, because
1234                          * we can have pinned chunks that fall within this
1235                          * device hole and that have lower physical addresses
1236                          * than the pending chunks we processed before. If we
1237                          * do not take this special care we can end up getting
1238                          * 2 pending chunks that start at the same physical
1239                          * device offsets because the end offset of a pinned
1240                          * chunk can be equal to the start offset of some
1241                          * pending chunk.
1242                          */
1243                         end = map->stripes[i].physical + em->orig_block_len;
1244                         if (end > *start) {
1245                                 *start = end;
1246                                 ret = 1;
1247                         }
1248                 }
1249         }
1250         if (search_list != &fs_info->pinned_chunks) {
1251                 search_list = &fs_info->pinned_chunks;
1252                 goto again;
1253         }
1254
1255         return ret;
1256 }
1257
1258
1259 /*
1260  * find_free_dev_extent_start - find free space in the specified device
1261  * @device:       the device which we search the free space in
1262  * @num_bytes:    the size of the free space that we need
1263  * @search_start: the position from which to begin the search
1264  * @start:        store the start of the free space.
1265  * @len:          the size of the free space. that we find, or the size
1266  *                of the max free space if we don't find suitable free space
1267  *
1268  * this uses a pretty simple search, the expectation is that it is
1269  * called very infrequently and that a given device has a small number
1270  * of extents
1271  *
1272  * @start is used to store the start of the free space if we find. But if we
1273  * don't find suitable free space, it will be used to store the start position
1274  * of the max free space.
1275  *
1276  * @len is used to store the size of the free space that we find.
1277  * But if we don't find suitable free space, it is used to store the size of
1278  * the max free space.
1279  */
1280 int find_free_dev_extent_start(struct btrfs_transaction *transaction,
1281                                struct btrfs_device *device, u64 num_bytes,
1282                                u64 search_start, u64 *start, u64 *len)
1283 {
1284         struct btrfs_key key;
1285         struct btrfs_root *root = device->dev_root;
1286         struct btrfs_dev_extent *dev_extent;
1287         struct btrfs_path *path;
1288         u64 hole_size;
1289         u64 max_hole_start;
1290         u64 max_hole_size;
1291         u64 extent_end;
1292         u64 search_end = device->total_bytes;
1293         int ret;
1294         int slot;
1295         struct extent_buffer *l;
1296         u64 min_search_start;
1297
1298         /*
1299          * We don't want to overwrite the superblock on the drive nor any area
1300          * used by the boot loader (grub for example), so we make sure to start
1301          * at an offset of at least 1MB.
1302          */
1303         min_search_start = max(root->fs_info->alloc_start, 1024ull * 1024);
1304         search_start = max(search_start, min_search_start);
1305
1306         path = btrfs_alloc_path();
1307         if (!path)
1308                 return -ENOMEM;
1309
1310         max_hole_start = search_start;
1311         max_hole_size = 0;
1312
1313 again:
1314         if (search_start >= search_end || device->is_tgtdev_for_dev_replace) {
1315                 ret = -ENOSPC;
1316                 goto out;
1317         }
1318
1319         path->reada = READA_FORWARD;
1320         path->search_commit_root = 1;
1321         path->skip_locking = 1;
1322
1323         key.objectid = device->devid;
1324         key.offset = search_start;
1325         key.type = BTRFS_DEV_EXTENT_KEY;
1326
1327         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1328         if (ret < 0)
1329                 goto out;
1330         if (ret > 0) {
1331                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1332                 if (ret < 0)
1333                         goto out;
1334         }
1335
1336         while (1) {
1337                 l = path->nodes[0];
1338                 slot = path->slots[0];
1339                 if (slot >= btrfs_header_nritems(l)) {
1340                         ret = btrfs_next_leaf(root, path);
1341                         if (ret == 0)
1342                                 continue;
1343                         if (ret < 0)
1344                                 goto out;
1345
1346                         break;
1347                 }
1348                 btrfs_item_key_to_cpu(l, &key, slot);
1349
1350                 if (key.objectid < device->devid)
1351                         goto next;
1352
1353                 if (key.objectid > device->devid)
1354                         break;
1355
1356                 if (key.type != BTRFS_DEV_EXTENT_KEY)
1357                         goto next;
1358
1359                 if (key.offset > search_start) {
1360                         hole_size = key.offset - search_start;
1361
1362                         /*
1363                          * Have to check before we set max_hole_start, otherwise
1364                          * we could end up sending back this offset anyway.
1365                          */
1366                         if (contains_pending_extent(transaction, device,
1367                                                     &search_start,
1368                                                     hole_size)) {
1369                                 if (key.offset >= search_start) {
1370                                         hole_size = key.offset - search_start;
1371                                 } else {
1372                                         WARN_ON_ONCE(1);
1373                                         hole_size = 0;
1374                                 }
1375                         }
1376
1377                         if (hole_size > max_hole_size) {
1378                                 max_hole_start = search_start;
1379                                 max_hole_size = hole_size;
1380                         }
1381
1382                         /*
1383                          * If this free space is greater than which we need,
1384                          * it must be the max free space that we have found
1385                          * until now, so max_hole_start must point to the start
1386                          * of this free space and the length of this free space
1387                          * is stored in max_hole_size. Thus, we return
1388                          * max_hole_start and max_hole_size and go back to the
1389                          * caller.
1390                          */
1391                         if (hole_size >= num_bytes) {
1392                                 ret = 0;
1393                                 goto out;
1394                         }
1395                 }
1396
1397                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1398                 extent_end = key.offset + btrfs_dev_extent_length(l,
1399                                                                   dev_extent);
1400                 if (extent_end > search_start)
1401                         search_start = extent_end;
1402 next:
1403                 path->slots[0]++;
1404                 cond_resched();
1405         }
1406
1407         /*
1408          * At this point, search_start should be the end of
1409          * allocated dev extents, and when shrinking the device,
1410          * search_end may be smaller than search_start.
1411          */
1412         if (search_end > search_start) {
1413                 hole_size = search_end - search_start;
1414
1415                 if (contains_pending_extent(transaction, device, &search_start,
1416                                             hole_size)) {
1417                         btrfs_release_path(path);
1418                         goto again;
1419                 }
1420
1421                 if (hole_size > max_hole_size) {
1422                         max_hole_start = search_start;
1423                         max_hole_size = hole_size;
1424                 }
1425         }
1426
1427         /* See above. */
1428         if (max_hole_size < num_bytes)
1429                 ret = -ENOSPC;
1430         else
1431                 ret = 0;
1432
1433 out:
1434         btrfs_free_path(path);
1435         *start = max_hole_start;
1436         if (len)
1437                 *len = max_hole_size;
1438         return ret;
1439 }
1440
1441 int find_free_dev_extent(struct btrfs_trans_handle *trans,
1442                          struct btrfs_device *device, u64 num_bytes,
1443                          u64 *start, u64 *len)
1444 {
1445         /* FIXME use last free of some kind */
1446         return find_free_dev_extent_start(trans->transaction, device,
1447                                           num_bytes, 0, start, len);
1448 }
1449
1450 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1451                           struct btrfs_device *device,
1452                           u64 start, u64 *dev_extent_len)
1453 {
1454         int ret;
1455         struct btrfs_path *path;
1456         struct btrfs_root *root = device->dev_root;
1457         struct btrfs_key key;
1458         struct btrfs_key found_key;
1459         struct extent_buffer *leaf = NULL;
1460         struct btrfs_dev_extent *extent = NULL;
1461
1462         path = btrfs_alloc_path();
1463         if (!path)
1464                 return -ENOMEM;
1465
1466         key.objectid = device->devid;
1467         key.offset = start;
1468         key.type = BTRFS_DEV_EXTENT_KEY;
1469 again:
1470         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1471         if (ret > 0) {
1472                 ret = btrfs_previous_item(root, path, key.objectid,
1473                                           BTRFS_DEV_EXTENT_KEY);
1474                 if (ret)
1475                         goto out;
1476                 leaf = path->nodes[0];
1477                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1478                 extent = btrfs_item_ptr(leaf, path->slots[0],
1479                                         struct btrfs_dev_extent);
1480                 BUG_ON(found_key.offset > start || found_key.offset +
1481                        btrfs_dev_extent_length(leaf, extent) < start);
1482                 key = found_key;
1483                 btrfs_release_path(path);
1484                 goto again;
1485         } else if (ret == 0) {
1486                 leaf = path->nodes[0];
1487                 extent = btrfs_item_ptr(leaf, path->slots[0],
1488                                         struct btrfs_dev_extent);
1489         } else {
1490                 btrfs_std_error(root->fs_info, ret, "Slot search failed");
1491                 goto out;
1492         }
1493
1494         *dev_extent_len = btrfs_dev_extent_length(leaf, extent);
1495
1496         ret = btrfs_del_item(trans, root, path);
1497         if (ret) {
1498                 btrfs_std_error(root->fs_info, ret,
1499                             "Failed to remove dev extent item");
1500         } else {
1501                 set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags);
1502         }
1503 out:
1504         btrfs_free_path(path);
1505         return ret;
1506 }
1507
1508 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1509                                   struct btrfs_device *device,
1510                                   u64 chunk_tree, u64 chunk_objectid,
1511                                   u64 chunk_offset, u64 start, u64 num_bytes)
1512 {
1513         int ret;
1514         struct btrfs_path *path;
1515         struct btrfs_root *root = device->dev_root;
1516         struct btrfs_dev_extent *extent;
1517         struct extent_buffer *leaf;
1518         struct btrfs_key key;
1519
1520         WARN_ON(!device->in_fs_metadata);
1521         WARN_ON(device->is_tgtdev_for_dev_replace);
1522         path = btrfs_alloc_path();
1523         if (!path)
1524                 return -ENOMEM;
1525
1526         key.objectid = device->devid;
1527         key.offset = start;
1528         key.type = BTRFS_DEV_EXTENT_KEY;
1529         ret = btrfs_insert_empty_item(trans, root, path, &key,
1530                                       sizeof(*extent));
1531         if (ret)
1532                 goto out;
1533
1534         leaf = path->nodes[0];
1535         extent = btrfs_item_ptr(leaf, path->slots[0],
1536                                 struct btrfs_dev_extent);
1537         btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
1538         btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
1539         btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1540
1541         write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
1542                     btrfs_dev_extent_chunk_tree_uuid(extent), BTRFS_UUID_SIZE);
1543
1544         btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1545         btrfs_mark_buffer_dirty(leaf);
1546 out:
1547         btrfs_free_path(path);
1548         return ret;
1549 }
1550
1551 static u64 find_next_chunk(struct btrfs_fs_info *fs_info)
1552 {
1553         struct extent_map_tree *em_tree;
1554         struct extent_map *em;
1555         struct rb_node *n;
1556         u64 ret = 0;
1557
1558         em_tree = &fs_info->mapping_tree.map_tree;
1559         read_lock(&em_tree->lock);
1560         n = rb_last(&em_tree->map);
1561         if (n) {
1562                 em = rb_entry(n, struct extent_map, rb_node);
1563                 ret = em->start + em->len;
1564         }
1565         read_unlock(&em_tree->lock);
1566
1567         return ret;
1568 }
1569
1570 static noinline int find_next_devid(struct btrfs_fs_info *fs_info,
1571                                     u64 *devid_ret)
1572 {
1573         int ret;
1574         struct btrfs_key key;
1575         struct btrfs_key found_key;
1576         struct btrfs_path *path;
1577
1578         path = btrfs_alloc_path();
1579         if (!path)
1580                 return -ENOMEM;
1581
1582         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1583         key.type = BTRFS_DEV_ITEM_KEY;
1584         key.offset = (u64)-1;
1585
1586         ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
1587         if (ret < 0)
1588                 goto error;
1589
1590         BUG_ON(ret == 0); /* Corruption */
1591
1592         ret = btrfs_previous_item(fs_info->chunk_root, path,
1593                                   BTRFS_DEV_ITEMS_OBJECTID,
1594                                   BTRFS_DEV_ITEM_KEY);
1595         if (ret) {
1596                 *devid_ret = 1;
1597         } else {
1598                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1599                                       path->slots[0]);
1600                 *devid_ret = found_key.offset + 1;
1601         }
1602         ret = 0;
1603 error:
1604         btrfs_free_path(path);
1605         return ret;
1606 }
1607
1608 /*
1609  * the device information is stored in the chunk root
1610  * the btrfs_device struct should be fully filled in
1611  */
1612 static int btrfs_add_device(struct btrfs_trans_handle *trans,
1613                             struct btrfs_root *root,
1614                             struct btrfs_device *device)
1615 {
1616         int ret;
1617         struct btrfs_path *path;
1618         struct btrfs_dev_item *dev_item;
1619         struct extent_buffer *leaf;
1620         struct btrfs_key key;
1621         unsigned long ptr;
1622
1623         root = root->fs_info->chunk_root;
1624
1625         path = btrfs_alloc_path();
1626         if (!path)
1627                 return -ENOMEM;
1628
1629         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1630         key.type = BTRFS_DEV_ITEM_KEY;
1631         key.offset = device->devid;
1632
1633         ret = btrfs_insert_empty_item(trans, root, path, &key,
1634                                       sizeof(*dev_item));
1635         if (ret)
1636                 goto out;
1637
1638         leaf = path->nodes[0];
1639         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1640
1641         btrfs_set_device_id(leaf, dev_item, device->devid);
1642         btrfs_set_device_generation(leaf, dev_item, 0);
1643         btrfs_set_device_type(leaf, dev_item, device->type);
1644         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1645         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1646         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1647         btrfs_set_device_total_bytes(leaf, dev_item,
1648                                      btrfs_device_get_disk_total_bytes(device));
1649         btrfs_set_device_bytes_used(leaf, dev_item,
1650                                     btrfs_device_get_bytes_used(device));
1651         btrfs_set_device_group(leaf, dev_item, 0);
1652         btrfs_set_device_seek_speed(leaf, dev_item, 0);
1653         btrfs_set_device_bandwidth(leaf, dev_item, 0);
1654         btrfs_set_device_start_offset(leaf, dev_item, 0);
1655
1656         ptr = btrfs_device_uuid(dev_item);
1657         write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1658         ptr = btrfs_device_fsid(dev_item);
1659         write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1660         btrfs_mark_buffer_dirty(leaf);
1661
1662         ret = 0;
1663 out:
1664         btrfs_free_path(path);
1665         return ret;
1666 }
1667
1668 /*
1669  * Function to update ctime/mtime for a given device path.
1670  * Mainly used for ctime/mtime based probe like libblkid.
1671  */
1672 static void update_dev_time(char *path_name)
1673 {
1674         struct file *filp;
1675
1676         filp = filp_open(path_name, O_RDWR, 0);
1677         if (IS_ERR(filp))
1678                 return;
1679         file_update_time(filp);
1680         filp_close(filp, NULL);
1681 }
1682
1683 static int btrfs_rm_dev_item(struct btrfs_root *root,
1684                              struct btrfs_device *device)
1685 {
1686         int ret;
1687         struct btrfs_path *path;
1688         struct btrfs_key key;
1689         struct btrfs_trans_handle *trans;
1690
1691         root = root->fs_info->chunk_root;
1692
1693         path = btrfs_alloc_path();
1694         if (!path)
1695                 return -ENOMEM;
1696
1697         trans = btrfs_start_transaction(root, 0);
1698         if (IS_ERR(trans)) {
1699                 btrfs_free_path(path);
1700                 return PTR_ERR(trans);
1701         }
1702         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1703         key.type = BTRFS_DEV_ITEM_KEY;
1704         key.offset = device->devid;
1705
1706         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1707         if (ret < 0)
1708                 goto out;
1709
1710         if (ret > 0) {
1711                 ret = -ENOENT;
1712                 goto out;
1713         }
1714
1715         ret = btrfs_del_item(trans, root, path);
1716         if (ret)
1717                 goto out;
1718 out:
1719         btrfs_free_path(path);
1720         btrfs_commit_transaction(trans, root);
1721         return ret;
1722 }
1723
1724 /*
1725  * Verify that @num_devices satisfies the RAID profile constraints in the whole
1726  * filesystem. It's up to the caller to adjust that number regarding eg. device
1727  * replace.
1728  */
1729 static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info,
1730                 u64 num_devices)
1731 {
1732         u64 all_avail;
1733         unsigned seq;
1734
1735         do {
1736                 seq = read_seqbegin(&fs_info->profiles_lock);
1737
1738                 all_avail = fs_info->avail_data_alloc_bits |
1739                             fs_info->avail_system_alloc_bits |
1740                             fs_info->avail_metadata_alloc_bits;
1741         } while (read_seqretry(&fs_info->profiles_lock, seq));
1742
1743         if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) && num_devices < 4) {
1744                 return BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET;
1745         }
1746
1747         if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) && num_devices < 2) {
1748                 return BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET;
1749         }
1750
1751         if ((all_avail & BTRFS_BLOCK_GROUP_RAID5) &&
1752             fs_info->fs_devices->rw_devices < 2) {
1753                 return BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET;
1754         }
1755
1756         if ((all_avail & BTRFS_BLOCK_GROUP_RAID6) &&
1757             fs_info->fs_devices->rw_devices < 3) {
1758                 return BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET;
1759         }
1760
1761         return 0;
1762 }
1763
1764 int btrfs_rm_device(struct btrfs_root *root, char *device_path, u64 devid)
1765 {
1766         struct btrfs_device *device;
1767         struct btrfs_device *next_device;
1768         struct btrfs_fs_devices *cur_devices;
1769         u64 num_devices;
1770         int ret = 0;
1771         bool clear_super = false;
1772         char *dev_name = NULL;
1773
1774         mutex_lock(&uuid_mutex);
1775
1776         num_devices = root->fs_info->fs_devices->num_devices;
1777         btrfs_dev_replace_lock(&root->fs_info->dev_replace, 0);
1778         if (btrfs_dev_replace_is_ongoing(&root->fs_info->dev_replace)) {
1779                 WARN_ON(num_devices < 1);
1780                 num_devices--;
1781         }
1782         btrfs_dev_replace_unlock(&root->fs_info->dev_replace, 0);
1783
1784         ret = btrfs_check_raid_min_devices(root->fs_info, num_devices - 1);
1785         if (ret)
1786                 goto out;
1787
1788         ret = btrfs_find_device_by_user_input(root, devid, device_path,
1789                                 &device);
1790         if (ret)
1791                 goto out;
1792
1793         if (device->is_tgtdev_for_dev_replace) {
1794                 ret = BTRFS_ERROR_DEV_TGT_REPLACE;
1795                 goto out;
1796         }
1797
1798         if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1799                 ret = BTRFS_ERROR_DEV_ONLY_WRITABLE;
1800                 goto out;
1801         }
1802
1803         if (device->writeable) {
1804                 lock_chunks(root);
1805                 list_del_init(&device->dev_alloc_list);
1806                 device->fs_devices->rw_devices--;
1807                 unlock_chunks(root);
1808                 dev_name = kstrdup(device->name->str, GFP_KERNEL);
1809                 if (!dev_name) {
1810                         ret = -ENOMEM;
1811                         goto error_undo;
1812                 }
1813                 clear_super = true;
1814         }
1815
1816         mutex_unlock(&uuid_mutex);
1817         ret = btrfs_shrink_device(device, 0);
1818         mutex_lock(&uuid_mutex);
1819         if (ret)
1820                 goto error_undo;
1821
1822         /*
1823          * TODO: the superblock still includes this device in its num_devices
1824          * counter although write_all_supers() is not locked out. This
1825          * could give a filesystem state which requires a degraded mount.
1826          */
1827         ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1828         if (ret)
1829                 goto error_undo;
1830
1831         device->in_fs_metadata = 0;
1832         btrfs_scrub_cancel_dev(root->fs_info, device);
1833
1834         /*
1835          * the device list mutex makes sure that we don't change
1836          * the device list while someone else is writing out all
1837          * the device supers. Whoever is writing all supers, should
1838          * lock the device list mutex before getting the number of
1839          * devices in the super block (super_copy). Conversely,
1840          * whoever updates the number of devices in the super block
1841          * (super_copy) should hold the device list mutex.
1842          */
1843
1844         cur_devices = device->fs_devices;
1845         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1846         list_del_rcu(&device->dev_list);
1847
1848         device->fs_devices->num_devices--;
1849         device->fs_devices->total_devices--;
1850
1851         if (device->missing)
1852                 device->fs_devices->missing_devices--;
1853
1854         next_device = list_entry(root->fs_info->fs_devices->devices.next,
1855                                  struct btrfs_device, dev_list);
1856         if (device->bdev == root->fs_info->sb->s_bdev)
1857                 root->fs_info->sb->s_bdev = next_device->bdev;
1858         if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1859                 root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1860
1861         if (device->bdev) {
1862                 device->fs_devices->open_devices--;
1863                 /* remove sysfs entry */
1864                 btrfs_sysfs_rm_device_link(root->fs_info->fs_devices, device);
1865         }
1866
1867         call_rcu(&device->rcu, free_device);
1868
1869         num_devices = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
1870         btrfs_set_super_num_devices(root->fs_info->super_copy, num_devices);
1871         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1872
1873         if (cur_devices->open_devices == 0) {
1874                 struct btrfs_fs_devices *fs_devices;
1875                 fs_devices = root->fs_info->fs_devices;
1876                 while (fs_devices) {
1877                         if (fs_devices->seed == cur_devices) {
1878                                 fs_devices->seed = cur_devices->seed;
1879                                 break;
1880                         }
1881                         fs_devices = fs_devices->seed;
1882                 }
1883                 cur_devices->seed = NULL;
1884                 __btrfs_close_devices(cur_devices);
1885                 free_fs_devices(cur_devices);
1886         }
1887
1888         root->fs_info->num_tolerated_disk_barrier_failures =
1889                 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
1890
1891         /*
1892          * at this point, the device is zero sized.  We want to
1893          * remove it from the devices list and zero out the old super
1894          */
1895         if (clear_super) {
1896                 struct block_device *bdev;
1897
1898                 bdev = blkdev_get_by_path(dev_name, FMODE_READ | FMODE_EXCL,
1899                                                 root->fs_info->bdev_holder);
1900                 if (!IS_ERR(bdev)) {
1901                         btrfs_scratch_superblocks(bdev, dev_name);
1902                         blkdev_put(bdev, FMODE_READ | FMODE_EXCL);
1903                 }
1904         }
1905
1906 out:
1907         kfree(dev_name);
1908
1909         mutex_unlock(&uuid_mutex);
1910         return ret;
1911
1912 error_undo:
1913         if (device->writeable) {
1914                 lock_chunks(root);
1915                 list_add(&device->dev_alloc_list,
1916                          &root->fs_info->fs_devices->alloc_list);
1917                 device->fs_devices->rw_devices++;
1918                 unlock_chunks(root);
1919         }
1920         goto out;
1921 }
1922
1923 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_fs_info *fs_info,
1924                                         struct btrfs_device *srcdev)
1925 {
1926         struct btrfs_fs_devices *fs_devices;
1927
1928         WARN_ON(!mutex_is_locked(&fs_info->fs_devices->device_list_mutex));
1929
1930         /*
1931          * in case of fs with no seed, srcdev->fs_devices will point
1932          * to fs_devices of fs_info. However when the dev being replaced is
1933          * a seed dev it will point to the seed's local fs_devices. In short
1934          * srcdev will have its correct fs_devices in both the cases.
1935          */
1936         fs_devices = srcdev->fs_devices;
1937
1938         list_del_rcu(&srcdev->dev_list);
1939         list_del_rcu(&srcdev->dev_alloc_list);
1940         fs_devices->num_devices--;
1941         if (srcdev->missing)
1942                 fs_devices->missing_devices--;
1943
1944         if (srcdev->writeable) {
1945                 fs_devices->rw_devices--;
1946                 /* zero out the old super if it is writable */
1947                 btrfs_scratch_superblocks(srcdev->bdev, srcdev->name->str);
1948         }
1949
1950         if (srcdev->bdev)
1951                 fs_devices->open_devices--;
1952 }
1953
1954 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_fs_info *fs_info,
1955                                       struct btrfs_device *srcdev)
1956 {
1957         struct btrfs_fs_devices *fs_devices = srcdev->fs_devices;
1958
1959         call_rcu(&srcdev->rcu, free_device);
1960
1961         /*
1962          * unless fs_devices is seed fs, num_devices shouldn't go
1963          * zero
1964          */
1965         BUG_ON(!fs_devices->num_devices && !fs_devices->seeding);
1966
1967         /* if this is no devs we rather delete the fs_devices */
1968         if (!fs_devices->num_devices) {
1969                 struct btrfs_fs_devices *tmp_fs_devices;
1970
1971                 tmp_fs_devices = fs_info->fs_devices;
1972                 while (tmp_fs_devices) {
1973                         if (tmp_fs_devices->seed == fs_devices) {
1974                                 tmp_fs_devices->seed = fs_devices->seed;
1975                                 break;
1976                         }
1977                         tmp_fs_devices = tmp_fs_devices->seed;
1978                 }
1979                 fs_devices->seed = NULL;
1980                 __btrfs_close_devices(fs_devices);
1981                 free_fs_devices(fs_devices);
1982         }
1983 }
1984
1985 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
1986                                       struct btrfs_device *tgtdev)
1987 {
1988         struct btrfs_device *next_device;
1989
1990         mutex_lock(&uuid_mutex);
1991         WARN_ON(!tgtdev);
1992         mutex_lock(&fs_info->fs_devices->device_list_mutex);
1993
1994         btrfs_sysfs_rm_device_link(fs_info->fs_devices, tgtdev);
1995
1996         if (tgtdev->bdev) {
1997                 btrfs_scratch_superblocks(tgtdev->bdev, tgtdev->name->str);
1998                 fs_info->fs_devices->open_devices--;
1999         }
2000         fs_info->fs_devices->num_devices--;
2001
2002         next_device = list_entry(fs_info->fs_devices->devices.next,
2003                                  struct btrfs_device, dev_list);
2004         if (tgtdev->bdev == fs_info->sb->s_bdev)
2005                 fs_info->sb->s_bdev = next_device->bdev;
2006         if (tgtdev->bdev == fs_info->fs_devices->latest_bdev)
2007                 fs_info->fs_devices->latest_bdev = next_device->bdev;
2008         list_del_rcu(&tgtdev->dev_list);
2009
2010         call_rcu(&tgtdev->rcu, free_device);
2011
2012         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2013         mutex_unlock(&uuid_mutex);
2014 }
2015
2016 static int btrfs_find_device_by_path(struct btrfs_root *root, char *device_path,
2017                                      struct btrfs_device **device)
2018 {
2019         int ret = 0;
2020         struct btrfs_super_block *disk_super;
2021         u64 devid;
2022         u8 *dev_uuid;
2023         struct block_device *bdev;
2024         struct buffer_head *bh;
2025
2026         *device = NULL;
2027         ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
2028                                     root->fs_info->bdev_holder, 0, &bdev, &bh);
2029         if (ret)
2030                 return ret;
2031         disk_super = (struct btrfs_super_block *)bh->b_data;
2032         devid = btrfs_stack_device_id(&disk_super->dev_item);
2033         dev_uuid = disk_super->dev_item.uuid;
2034         *device = btrfs_find_device(root->fs_info, devid, dev_uuid,
2035                                     disk_super->fsid);
2036         brelse(bh);
2037         if (!*device)
2038                 ret = -ENOENT;
2039         blkdev_put(bdev, FMODE_READ);
2040         return ret;
2041 }
2042
2043 int btrfs_find_device_missing_or_by_path(struct btrfs_root *root,
2044                                          char *device_path,
2045                                          struct btrfs_device **device)
2046 {
2047         *device = NULL;
2048         if (strcmp(device_path, "missing") == 0) {
2049                 struct list_head *devices;
2050                 struct btrfs_device *tmp;
2051
2052                 devices = &root->fs_info->fs_devices->devices;
2053                 /*
2054                  * It is safe to read the devices since the volume_mutex
2055                  * is held by the caller.
2056                  */
2057                 list_for_each_entry(tmp, devices, dev_list) {
2058                         if (tmp->in_fs_metadata && !tmp->bdev) {
2059                                 *device = tmp;
2060                                 break;
2061                         }
2062                 }
2063
2064                 if (!*device)
2065                         return BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
2066
2067                 return 0;
2068         } else {
2069                 return btrfs_find_device_by_path(root, device_path, device);
2070         }
2071 }
2072
2073 int btrfs_find_device_by_user_input(struct btrfs_root *root, u64 srcdevid,
2074                                          char *srcdev_name,
2075                                          struct btrfs_device **device)
2076 {
2077         int ret;
2078
2079         if (srcdevid) {
2080                 ret = 0;
2081                 *device = btrfs_find_device(root->fs_info, srcdevid, NULL,
2082                                             NULL);
2083                 if (!*device)
2084                         ret = -ENOENT;
2085         } else {
2086                 if (!srcdev_name || !srcdev_name[0])
2087                         return -EINVAL;
2088
2089                 ret = btrfs_find_device_missing_or_by_path(root, srcdev_name,
2090                                                            device);
2091         }
2092         return ret;
2093 }
2094
2095 /*
2096  * does all the dirty work required for changing file system's UUID.
2097  */
2098 static int btrfs_prepare_sprout(struct btrfs_root *root)
2099 {
2100         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2101         struct btrfs_fs_devices *old_devices;
2102         struct btrfs_fs_devices *seed_devices;
2103         struct btrfs_super_block *disk_super = root->fs_info->super_copy;
2104         struct btrfs_device *device;
2105         u64 super_flags;
2106
2107         BUG_ON(!mutex_is_locked(&uuid_mutex));
2108         if (!fs_devices->seeding)
2109                 return -EINVAL;
2110
2111         seed_devices = __alloc_fs_devices();
2112         if (IS_ERR(seed_devices))
2113                 return PTR_ERR(seed_devices);
2114
2115         old_devices = clone_fs_devices(fs_devices);
2116         if (IS_ERR(old_devices)) {
2117                 kfree(seed_devices);
2118                 return PTR_ERR(old_devices);
2119         }
2120
2121         list_add(&old_devices->list, &fs_uuids);
2122
2123         memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
2124         seed_devices->opened = 1;
2125         INIT_LIST_HEAD(&seed_devices->devices);
2126         INIT_LIST_HEAD(&seed_devices->alloc_list);
2127         mutex_init(&seed_devices->device_list_mutex);
2128
2129         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2130         list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
2131                               synchronize_rcu);
2132         list_for_each_entry(device, &seed_devices->devices, dev_list)
2133                 device->fs_devices = seed_devices;
2134
2135         lock_chunks(root);
2136         list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
2137         unlock_chunks(root);
2138
2139         fs_devices->seeding = 0;
2140         fs_devices->num_devices = 0;
2141         fs_devices->open_devices = 0;
2142         fs_devices->missing_devices = 0;
2143         fs_devices->rotating = 0;
2144         fs_devices->seed = seed_devices;
2145
2146         generate_random_uuid(fs_devices->fsid);
2147         memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2148         memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2149         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2150
2151         super_flags = btrfs_super_flags(disk_super) &
2152                       ~BTRFS_SUPER_FLAG_SEEDING;
2153         btrfs_set_super_flags(disk_super, super_flags);
2154
2155         return 0;
2156 }
2157
2158 /*
2159  * strore the expected generation for seed devices in device items.
2160  */
2161 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
2162                                struct btrfs_root *root)
2163 {
2164         struct btrfs_path *path;
2165         struct extent_buffer *leaf;
2166         struct btrfs_dev_item *dev_item;
2167         struct btrfs_device *device;
2168         struct btrfs_key key;
2169         u8 fs_uuid[BTRFS_UUID_SIZE];
2170         u8 dev_uuid[BTRFS_UUID_SIZE];
2171         u64 devid;
2172         int ret;
2173
2174         path = btrfs_alloc_path();
2175         if (!path)
2176                 return -ENOMEM;
2177
2178         root = root->fs_info->chunk_root;
2179         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2180         key.offset = 0;
2181         key.type = BTRFS_DEV_ITEM_KEY;
2182
2183         while (1) {
2184                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2185                 if (ret < 0)
2186                         goto error;
2187
2188                 leaf = path->nodes[0];
2189 next_slot:
2190                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2191                         ret = btrfs_next_leaf(root, path);
2192                         if (ret > 0)
2193                                 break;
2194                         if (ret < 0)
2195                                 goto error;
2196                         leaf = path->nodes[0];
2197                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2198                         btrfs_release_path(path);
2199                         continue;
2200                 }
2201
2202                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2203                 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
2204                     key.type != BTRFS_DEV_ITEM_KEY)
2205                         break;
2206
2207                 dev_item = btrfs_item_ptr(leaf, path->slots[0],
2208                                           struct btrfs_dev_item);
2209                 devid = btrfs_device_id(leaf, dev_item);
2210                 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
2211                                    BTRFS_UUID_SIZE);
2212                 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
2213                                    BTRFS_UUID_SIZE);
2214                 device = btrfs_find_device(root->fs_info, devid, dev_uuid,
2215                                            fs_uuid);
2216                 BUG_ON(!device); /* Logic error */
2217
2218                 if (device->fs_devices->seeding) {
2219                         btrfs_set_device_generation(leaf, dev_item,
2220                                                     device->generation);
2221                         btrfs_mark_buffer_dirty(leaf);
2222                 }
2223
2224                 path->slots[0]++;
2225                 goto next_slot;
2226         }
2227         ret = 0;
2228 error:
2229         btrfs_free_path(path);
2230         return ret;
2231 }
2232
2233 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
2234 {
2235         struct request_queue *q;
2236         struct btrfs_trans_handle *trans;
2237         struct btrfs_device *device;
2238         struct block_device *bdev;
2239         struct list_head *devices;
2240         struct super_block *sb = root->fs_info->sb;
2241         struct rcu_string *name;
2242         u64 tmp;
2243         int seeding_dev = 0;
2244         int ret = 0;
2245
2246         if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
2247                 return -EROFS;
2248
2249         bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2250                                   root->fs_info->bdev_holder);
2251         if (IS_ERR(bdev))
2252                 return PTR_ERR(bdev);
2253
2254         if (root->fs_info->fs_devices->seeding) {
2255                 seeding_dev = 1;
2256                 down_write(&sb->s_umount);
2257                 mutex_lock(&uuid_mutex);
2258         }
2259
2260         filemap_write_and_wait(bdev->bd_inode->i_mapping);
2261
2262         devices = &root->fs_info->fs_devices->devices;
2263
2264         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2265         list_for_each_entry(device, devices, dev_list) {
2266                 if (device->bdev == bdev) {
2267                         ret = -EEXIST;
2268                         mutex_unlock(
2269                                 &root->fs_info->fs_devices->device_list_mutex);
2270                         goto error;
2271                 }
2272         }
2273         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2274
2275         device = btrfs_alloc_device(root->fs_info, NULL, NULL);
2276         if (IS_ERR(device)) {
2277                 /* we can safely leave the fs_devices entry around */
2278                 ret = PTR_ERR(device);
2279                 goto error;
2280         }
2281
2282         name = rcu_string_strdup(device_path, GFP_KERNEL);
2283         if (!name) {
2284                 kfree(device);
2285                 ret = -ENOMEM;
2286                 goto error;
2287         }
2288         rcu_assign_pointer(device->name, name);
2289
2290         trans = btrfs_start_transaction(root, 0);
2291         if (IS_ERR(trans)) {
2292                 rcu_string_free(device->name);
2293                 kfree(device);
2294                 ret = PTR_ERR(trans);
2295                 goto error;
2296         }
2297
2298         q = bdev_get_queue(bdev);
2299         if (blk_queue_discard(q))
2300                 device->can_discard = 1;
2301         device->writeable = 1;
2302         device->generation = trans->transid;
2303         device->io_width = root->sectorsize;
2304         device->io_align = root->sectorsize;
2305         device->sector_size = root->sectorsize;
2306         device->total_bytes = i_size_read(bdev->bd_inode);
2307         device->disk_total_bytes = device->total_bytes;
2308         device->commit_total_bytes = device->total_bytes;
2309         device->dev_root = root->fs_info->dev_root;
2310         device->bdev = bdev;
2311         device->in_fs_metadata = 1;
2312         device->is_tgtdev_for_dev_replace = 0;
2313         device->mode = FMODE_EXCL;
2314         device->dev_stats_valid = 1;
2315         set_blocksize(device->bdev, 4096);
2316
2317         if (seeding_dev) {
2318                 sb->s_flags &= ~MS_RDONLY;
2319                 ret = btrfs_prepare_sprout(root);
2320                 BUG_ON(ret); /* -ENOMEM */
2321         }
2322
2323         device->fs_devices = root->fs_info->fs_devices;
2324
2325         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2326         lock_chunks(root);
2327         list_add_rcu(&device->dev_list, &root->fs_info->fs_devices->devices);
2328         list_add(&device->dev_alloc_list,
2329                  &root->fs_info->fs_devices->alloc_list);
2330         root->fs_info->fs_devices->num_devices++;
2331         root->fs_info->fs_devices->open_devices++;
2332         root->fs_info->fs_devices->rw_devices++;
2333         root->fs_info->fs_devices->total_devices++;
2334         root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
2335
2336         spin_lock(&root->fs_info->free_chunk_lock);
2337         root->fs_info->free_chunk_space += device->total_bytes;
2338         spin_unlock(&root->fs_info->free_chunk_lock);
2339
2340         if (!blk_queue_nonrot(bdev_get_queue(bdev)))
2341                 root->fs_info->fs_devices->rotating = 1;
2342
2343         tmp = btrfs_super_total_bytes(root->fs_info->super_copy);
2344         btrfs_set_super_total_bytes(root->fs_info->super_copy,
2345                                     tmp + device->total_bytes);
2346
2347         tmp = btrfs_super_num_devices(root->fs_info->super_copy);
2348         btrfs_set_super_num_devices(root->fs_info->super_copy,
2349                                     tmp + 1);
2350
2351         /* add sysfs device entry */
2352         btrfs_sysfs_add_device_link(root->fs_info->fs_devices, device);
2353
2354         /*
2355          * we've got more storage, clear any full flags on the space
2356          * infos
2357          */
2358         btrfs_clear_space_info_full(root->fs_info);
2359
2360         unlock_chunks(root);
2361         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2362
2363         if (seeding_dev) {
2364                 lock_chunks(root);
2365                 ret = init_first_rw_device(trans, root, device);
2366                 unlock_chunks(root);
2367                 if (ret) {
2368                         btrfs_abort_transaction(trans, root, ret);
2369                         goto error_trans;
2370                 }
2371         }
2372
2373         ret = btrfs_add_device(trans, root, device);
2374         if (ret) {
2375                 btrfs_abort_transaction(trans, root, ret);
2376                 goto error_trans;
2377         }
2378
2379         if (seeding_dev) {
2380                 char fsid_buf[BTRFS_UUID_UNPARSED_SIZE];
2381
2382                 ret = btrfs_finish_sprout(trans, root);
2383                 if (ret) {
2384                         btrfs_abort_transaction(trans, root, ret);
2385                         goto error_trans;
2386                 }
2387
2388                 /* Sprouting would change fsid of the mounted root,
2389                  * so rename the fsid on the sysfs
2390                  */
2391                 snprintf(fsid_buf, BTRFS_UUID_UNPARSED_SIZE, "%pU",
2392                                                 root->fs_info->fsid);
2393                 if (kobject_rename(&root->fs_info->fs_devices->fsid_kobj,
2394                                                                 fsid_buf))
2395                         btrfs_warn(root->fs_info,
2396                                 "sysfs: failed to create fsid for sprout");
2397         }
2398
2399         root->fs_info->num_tolerated_disk_barrier_failures =
2400                 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
2401         ret = btrfs_commit_transaction(trans, root);
2402
2403         if (seeding_dev) {
2404                 mutex_unlock(&uuid_mutex);
2405                 up_write(&sb->s_umount);
2406
2407                 if (ret) /* transaction commit */
2408                         return ret;
2409
2410                 ret = btrfs_relocate_sys_chunks(root);
2411                 if (ret < 0)
2412                         btrfs_std_error(root->fs_info, ret,
2413                                     "Failed to relocate sys chunks after "
2414                                     "device initialization. This can be fixed "
2415                                     "using the \"btrfs balance\" command.");
2416                 trans = btrfs_attach_transaction(root);
2417                 if (IS_ERR(trans)) {
2418                         if (PTR_ERR(trans) == -ENOENT)
2419                                 return 0;
2420                         return PTR_ERR(trans);
2421                 }
2422                 ret = btrfs_commit_transaction(trans, root);
2423         }
2424
2425         /* Update ctime/mtime for libblkid */
2426         update_dev_time(device_path);
2427         return ret;
2428
2429 error_trans:
2430         btrfs_end_transaction(trans, root);
2431         rcu_string_free(device->name);
2432         btrfs_sysfs_rm_device_link(root->fs_info->fs_devices, device);
2433         kfree(device);
2434 error:
2435         blkdev_put(bdev, FMODE_EXCL);
2436         if (seeding_dev) {
2437                 mutex_unlock(&uuid_mutex);
2438                 up_write(&sb->s_umount);
2439         }
2440         return ret;
2441 }
2442
2443 int btrfs_init_dev_replace_tgtdev(struct btrfs_root *root, char *device_path,
2444                                   struct btrfs_device *srcdev,
2445                                   struct btrfs_device **device_out)
2446 {
2447         struct request_queue *q;
2448         struct btrfs_device *device;
2449         struct block_device *bdev;
2450         struct btrfs_fs_info *fs_info = root->fs_info;
2451         struct list_head *devices;
2452         struct rcu_string *name;
2453         u64 devid = BTRFS_DEV_REPLACE_DEVID;
2454         int ret = 0;
2455
2456         *device_out = NULL;
2457         if (fs_info->fs_devices->seeding) {
2458                 btrfs_err(fs_info, "the filesystem is a seed filesystem!");
2459                 return -EINVAL;
2460         }
2461
2462         bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2463                                   fs_info->bdev_holder);
2464         if (IS_ERR(bdev)) {
2465                 btrfs_err(fs_info, "target device %s is invalid!", device_path);
2466                 return PTR_ERR(bdev);
2467         }
2468
2469         filemap_write_and_wait(bdev->bd_inode->i_mapping);
2470
2471         devices = &fs_info->fs_devices->devices;
2472         list_for_each_entry(device, devices, dev_list) {
2473                 if (device->bdev == bdev) {
2474                         btrfs_err(fs_info, "target device is in the filesystem!");
2475                         ret = -EEXIST;
2476                         goto error;
2477                 }
2478         }
2479
2480
2481         if (i_size_read(bdev->bd_inode) <
2482             btrfs_device_get_total_bytes(srcdev)) {
2483                 btrfs_err(fs_info, "target device is smaller than source device!");
2484                 ret = -EINVAL;
2485                 goto error;
2486         }
2487
2488
2489         device = btrfs_alloc_device(NULL, &devid, NULL);
2490         if (IS_ERR(device)) {
2491                 ret = PTR_ERR(device);
2492                 goto error;
2493         }
2494
2495         name = rcu_string_strdup(device_path, GFP_NOFS);
2496         if (!name) {
2497                 kfree(device);
2498                 ret = -ENOMEM;
2499                 goto error;
2500         }
2501         rcu_assign_pointer(device->name, name);
2502
2503         q = bdev_get_queue(bdev);
2504         if (blk_queue_discard(q))
2505                 device->can_discard = 1;
2506         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2507         device->writeable = 1;
2508         device->generation = 0;
2509         device->io_width = root->sectorsize;
2510         device->io_align = root->sectorsize;
2511         device->sector_size = root->sectorsize;
2512         device->total_bytes = btrfs_device_get_total_bytes(srcdev);
2513         device->disk_total_bytes = btrfs_device_get_disk_total_bytes(srcdev);
2514         device->bytes_used = btrfs_device_get_bytes_used(srcdev);
2515         ASSERT(list_empty(&srcdev->resized_list));
2516         device->commit_total_bytes = srcdev->commit_total_bytes;
2517         device->commit_bytes_used = device->bytes_used;
2518         device->dev_root = fs_info->dev_root;
2519         device->bdev = bdev;
2520         device->in_fs_metadata = 1;
2521         device->is_tgtdev_for_dev_replace = 1;
2522         device->mode = FMODE_EXCL;
2523         device->dev_stats_valid = 1;
2524         set_blocksize(device->bdev, 4096);
2525         device->fs_devices = fs_info->fs_devices;
2526         list_add(&device->dev_list, &fs_info->fs_devices->devices);
2527         fs_info->fs_devices->num_devices++;
2528         fs_info->fs_devices->open_devices++;
2529         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2530
2531         *device_out = device;
2532         return ret;
2533
2534 error:
2535         blkdev_put(bdev, FMODE_EXCL);
2536         return ret;
2537 }
2538
2539 void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info *fs_info,
2540                                               struct btrfs_device *tgtdev)
2541 {
2542         WARN_ON(fs_info->fs_devices->rw_devices == 0);
2543         tgtdev->io_width = fs_info->dev_root->sectorsize;
2544         tgtdev->io_align = fs_info->dev_root->sectorsize;
2545         tgtdev->sector_size = fs_info->dev_root->sectorsize;
2546         tgtdev->dev_root = fs_info->dev_root;
2547         tgtdev->in_fs_metadata = 1;
2548 }
2549
2550 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2551                                         struct btrfs_device *device)
2552 {
2553         int ret;
2554         struct btrfs_path *path;
2555         struct btrfs_root *root;
2556         struct btrfs_dev_item *dev_item;
2557         struct extent_buffer *leaf;
2558         struct btrfs_key key;
2559
2560         root = device->dev_root->fs_info->chunk_root;
2561
2562         path = btrfs_alloc_path();
2563         if (!path)
2564                 return -ENOMEM;
2565
2566         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2567         key.type = BTRFS_DEV_ITEM_KEY;
2568         key.offset = device->devid;
2569
2570         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2571         if (ret < 0)
2572                 goto out;
2573
2574         if (ret > 0) {
2575                 ret = -ENOENT;
2576                 goto out;
2577         }
2578
2579         leaf = path->nodes[0];
2580         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2581
2582         btrfs_set_device_id(leaf, dev_item, device->devid);
2583         btrfs_set_device_type(leaf, dev_item, device->type);
2584         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2585         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2586         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2587         btrfs_set_device_total_bytes(leaf, dev_item,
2588                                      btrfs_device_get_disk_total_bytes(device));
2589         btrfs_set_device_bytes_used(leaf, dev_item,
2590                                     btrfs_device_get_bytes_used(device));
2591         btrfs_mark_buffer_dirty(leaf);
2592
2593 out:
2594         btrfs_free_path(path);
2595         return ret;
2596 }
2597
2598 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2599                       struct btrfs_device *device, u64 new_size)
2600 {
2601         struct btrfs_super_block *super_copy =
2602                 device->dev_root->fs_info->super_copy;
2603         struct btrfs_fs_devices *fs_devices;
2604         u64 old_total;
2605         u64 diff;
2606
2607         if (!device->writeable)
2608                 return -EACCES;
2609
2610         lock_chunks(device->dev_root);
2611         old_total = btrfs_super_total_bytes(super_copy);
2612         diff = new_size - device->total_bytes;
2613
2614         if (new_size <= device->total_bytes ||
2615             device->is_tgtdev_for_dev_replace) {
2616                 unlock_chunks(device->dev_root);
2617                 return -EINVAL;
2618         }
2619
2620         fs_devices = device->dev_root->fs_info->fs_devices;
2621
2622         btrfs_set_super_total_bytes(super_copy, old_total + diff);
2623         device->fs_devices->total_rw_bytes += diff;
2624
2625         btrfs_device_set_total_bytes(device, new_size);
2626         btrfs_device_set_disk_total_bytes(device, new_size);
2627         btrfs_clear_space_info_full(device->dev_root->fs_info);
2628         if (list_empty(&device->resized_list))
2629                 list_add_tail(&device->resized_list,
2630                               &fs_devices->resized_devices);
2631         unlock_chunks(device->dev_root);
2632
2633         return btrfs_update_device(trans, device);
2634 }
2635
2636 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
2637                             struct btrfs_root *root, u64 chunk_objectid,
2638                             u64 chunk_offset)
2639 {
2640         int ret;
2641         struct btrfs_path *path;
2642         struct btrfs_key key;
2643
2644         root = root->fs_info->chunk_root;
2645         path = btrfs_alloc_path();
2646         if (!path)
2647                 return -ENOMEM;
2648
2649         key.objectid = chunk_objectid;
2650         key.offset = chunk_offset;
2651         key.type = BTRFS_CHUNK_ITEM_KEY;
2652
2653         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2654         if (ret < 0)
2655                 goto out;
2656         else if (ret > 0) { /* Logic error or corruption */
2657                 btrfs_std_error(root->fs_info, -ENOENT,
2658                             "Failed lookup while freeing chunk.");
2659                 ret = -ENOENT;
2660                 goto out;
2661         }
2662
2663         ret = btrfs_del_item(trans, root, path);
2664         if (ret < 0)
2665                 btrfs_std_error(root->fs_info, ret,
2666                             "Failed to delete chunk item.");
2667 out:
2668         btrfs_free_path(path);
2669         return ret;
2670 }
2671
2672 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
2673                         chunk_offset)
2674 {
2675         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
2676         struct btrfs_disk_key *disk_key;
2677         struct btrfs_chunk *chunk;
2678         u8 *ptr;
2679         int ret = 0;
2680         u32 num_stripes;
2681         u32 array_size;
2682         u32 len = 0;
2683         u32 cur;
2684         struct btrfs_key key;
2685
2686         lock_chunks(root);
2687         array_size = btrfs_super_sys_array_size(super_copy);
2688
2689         ptr = super_copy->sys_chunk_array;
2690         cur = 0;
2691
2692         while (cur < array_size) {
2693                 disk_key = (struct btrfs_disk_key *)ptr;
2694                 btrfs_disk_key_to_cpu(&key, disk_key);
2695
2696                 len = sizeof(*disk_key);
2697
2698                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2699                         chunk = (struct btrfs_chunk *)(ptr + len);
2700                         num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2701                         len += btrfs_chunk_item_size(num_stripes);
2702                 } else {
2703                         ret = -EIO;
2704                         break;
2705                 }
2706                 if (key.objectid == chunk_objectid &&
2707                     key.offset == chunk_offset) {
2708                         memmove(ptr, ptr + len, array_size - (cur + len));
2709                         array_size -= len;
2710                         btrfs_set_super_sys_array_size(super_copy, array_size);
2711                 } else {
2712                         ptr += len;
2713                         cur += len;
2714                 }
2715         }
2716         unlock_chunks(root);
2717         return ret;
2718 }
2719
2720 int btrfs_remove_chunk(struct btrfs_trans_handle *trans,
2721                        struct btrfs_root *root, u64 chunk_offset)
2722 {
2723         struct extent_map_tree *em_tree;
2724         struct extent_map *em;
2725         struct btrfs_root *extent_root = root->fs_info->extent_root;
2726         struct map_lookup *map;
2727         u64 dev_extent_len = 0;
2728         u64 chunk_objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2729         int i, ret = 0;
2730
2731         /* Just in case */
2732         root = root->fs_info->chunk_root;
2733         em_tree = &root->fs_info->mapping_tree.map_tree;
2734
2735         read_lock(&em_tree->lock);
2736         em = lookup_extent_mapping(em_tree, chunk_offset, 1);
2737         read_unlock(&em_tree->lock);
2738
2739         if (!em || em->start > chunk_offset ||
2740             em->start + em->len < chunk_offset) {
2741                 /*
2742                  * This is a logic error, but we don't want to just rely on the
2743                  * user having built with ASSERT enabled, so if ASSERT doesn't
2744                  * do anything we still error out.
2745                  */
2746                 ASSERT(0);
2747                 if (em)
2748                         free_extent_map(em);
2749                 return -EINVAL;
2750         }
2751         map = em->map_lookup;
2752         lock_chunks(root->fs_info->chunk_root);
2753         check_system_chunk(trans, extent_root, map->type);
2754         unlock_chunks(root->fs_info->chunk_root);
2755
2756         for (i = 0; i < map->num_stripes; i++) {
2757                 struct btrfs_device *device = map->stripes[i].dev;
2758                 ret = btrfs_free_dev_extent(trans, device,
2759                                             map->stripes[i].physical,
2760                                             &dev_extent_len);
2761                 if (ret) {
2762                         btrfs_abort_transaction(trans, root, ret);
2763                         goto out;
2764                 }
2765
2766                 if (device->bytes_used > 0) {
2767                         lock_chunks(root);
2768                         btrfs_device_set_bytes_used(device,
2769                                         device->bytes_used - dev_extent_len);
2770                         spin_lock(&root->fs_info->free_chunk_lock);
2771                         root->fs_info->free_chunk_space += dev_extent_len;
2772                         spin_unlock(&root->fs_info->free_chunk_lock);
2773                         btrfs_clear_space_info_full(root->fs_info);
2774                         unlock_chunks(root);
2775                 }
2776
2777                 if (map->stripes[i].dev) {
2778                         ret = btrfs_update_device(trans, map->stripes[i].dev);
2779                         if (ret) {
2780                                 btrfs_abort_transaction(trans, root, ret);
2781                                 goto out;
2782                         }
2783                 }
2784         }
2785         ret = btrfs_free_chunk(trans, root, chunk_objectid, chunk_offset);
2786         if (ret) {
2787                 btrfs_abort_transaction(trans, root, ret);
2788                 goto out;
2789         }
2790
2791         trace_btrfs_chunk_free(root, map, chunk_offset, em->len);
2792
2793         if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2794                 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
2795                 if (ret) {
2796                         btrfs_abort_transaction(trans, root, ret);
2797                         goto out;
2798                 }
2799         }
2800
2801         ret = btrfs_remove_block_group(trans, extent_root, chunk_offset, em);
2802         if (ret) {
2803                 btrfs_abort_transaction(trans, extent_root, ret);
2804                 goto out;
2805         }
2806
2807 out:
2808         /* once for us */
2809         free_extent_map(em);
2810         return ret;
2811 }
2812
2813 static int btrfs_relocate_chunk(struct btrfs_root *root, u64 chunk_offset)
2814 {
2815         struct btrfs_root *extent_root;
2816         struct btrfs_trans_handle *trans;
2817         int ret;
2818
2819         root = root->fs_info->chunk_root;
2820         extent_root = root->fs_info->extent_root;
2821
2822         /*
2823          * Prevent races with automatic removal of unused block groups.
2824          * After we relocate and before we remove the chunk with offset
2825          * chunk_offset, automatic removal of the block group can kick in,
2826          * resulting in a failure when calling btrfs_remove_chunk() below.
2827          *
2828          * Make sure to acquire this mutex before doing a tree search (dev
2829          * or chunk trees) to find chunks. Otherwise the cleaner kthread might
2830          * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after
2831          * we release the path used to search the chunk/dev tree and before
2832          * the current task acquires this mutex and calls us.
2833          */
2834         ASSERT(mutex_is_locked(&root->fs_info->delete_unused_bgs_mutex));
2835
2836         ret = btrfs_can_relocate(extent_root, chunk_offset);
2837         if (ret)
2838                 return -ENOSPC;
2839
2840         /* step one, relocate all the extents inside this chunk */
2841         btrfs_scrub_pause(root);
2842         ret = btrfs_relocate_block_group(extent_root, chunk_offset);
2843         btrfs_scrub_continue(root);
2844         if (ret)
2845                 return ret;
2846
2847         trans = btrfs_start_trans_remove_block_group(root->fs_info,
2848                                                      chunk_offset);
2849         if (IS_ERR(trans)) {
2850                 ret = PTR_ERR(trans);
2851                 btrfs_std_error(root->fs_info, ret, NULL);
2852                 return ret;
2853         }
2854
2855         /*
2856          * step two, delete the device extents and the
2857          * chunk tree entries
2858          */
2859         ret = btrfs_remove_chunk(trans, root, chunk_offset);
2860         btrfs_end_transaction(trans, root);
2861         return ret;
2862 }
2863
2864 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
2865 {
2866         struct btrfs_root *chunk_root = root->fs_info->chunk_root;
2867         struct btrfs_path *path;
2868         struct extent_buffer *leaf;
2869         struct btrfs_chunk *chunk;
2870         struct btrfs_key key;
2871         struct btrfs_key found_key;
2872         u64 chunk_type;
2873         bool retried = false;
2874         int failed = 0;
2875         int ret;
2876
2877         path = btrfs_alloc_path();
2878         if (!path)
2879                 return -ENOMEM;
2880
2881 again:
2882         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2883         key.offset = (u64)-1;
2884         key.type = BTRFS_CHUNK_ITEM_KEY;
2885
2886         while (1) {
2887                 mutex_lock(&root->fs_info->delete_unused_bgs_mutex);
2888                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2889                 if (ret < 0) {
2890                         mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
2891                         goto error;
2892                 }
2893                 BUG_ON(ret == 0); /* Corruption */
2894
2895                 ret = btrfs_previous_item(chunk_root, path, key.objectid,
2896                                           key.type);
2897                 if (ret)
2898                         mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
2899                 if (ret < 0)
2900                         goto error;
2901                 if (ret > 0)
2902                         break;
2903
2904                 leaf = path->nodes[0];
2905                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2906
2907                 chunk = btrfs_item_ptr(leaf, path->slots[0],
2908                                        struct btrfs_chunk);
2909                 chunk_type = btrfs_chunk_type(leaf, chunk);
2910                 btrfs_release_path(path);
2911
2912                 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
2913                         ret = btrfs_relocate_chunk(chunk_root,
2914                                                    found_key.offset);
2915                         if (ret == -ENOSPC)
2916                                 failed++;
2917                         else
2918                                 BUG_ON(ret);
2919                 }
2920                 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
2921
2922                 if (found_key.offset == 0)
2923                         break;
2924                 key.offset = found_key.offset - 1;
2925         }
2926         ret = 0;
2927         if (failed && !retried) {
2928                 failed = 0;
2929                 retried = true;
2930                 goto again;
2931         } else if (WARN_ON(failed && retried)) {
2932                 ret = -ENOSPC;
2933         }
2934 error:
2935         btrfs_free_path(path);
2936         return ret;
2937 }
2938
2939 static int insert_balance_item(struct btrfs_root *root,
2940                                struct btrfs_balance_control *bctl)
2941 {
2942         struct btrfs_trans_handle *trans;
2943         struct btrfs_balance_item *item;
2944         struct btrfs_disk_balance_args disk_bargs;
2945         struct btrfs_path *path;
2946         struct extent_buffer *leaf;
2947         struct btrfs_key key;
2948         int ret, err;
2949
2950         path = btrfs_alloc_path();
2951         if (!path)
2952                 return -ENOMEM;
2953
2954         trans = btrfs_start_transaction(root, 0);
2955         if (IS_ERR(trans)) {
2956                 btrfs_free_path(path);
2957                 return PTR_ERR(trans);
2958         }
2959
2960         key.objectid = BTRFS_BALANCE_OBJECTID;
2961         key.type = BTRFS_TEMPORARY_ITEM_KEY;
2962         key.offset = 0;
2963
2964         ret = btrfs_insert_empty_item(trans, root, path, &key,
2965                                       sizeof(*item));
2966         if (ret)
2967                 goto out;
2968
2969         leaf = path->nodes[0];
2970         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
2971
2972         memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
2973
2974         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
2975         btrfs_set_balance_data(leaf, item, &disk_bargs);
2976         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
2977         btrfs_set_balance_meta(leaf, item, &disk_bargs);
2978         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
2979         btrfs_set_balance_sys(leaf, item, &disk_bargs);
2980
2981         btrfs_set_balance_flags(leaf, item, bctl->flags);
2982
2983         btrfs_mark_buffer_dirty(leaf);
2984 out:
2985         btrfs_free_path(path);
2986         err = btrfs_commit_transaction(trans, root);
2987         if (err && !ret)
2988                 ret = err;
2989         return ret;
2990 }
2991
2992 static int del_balance_item(struct btrfs_root *root)
2993 {
2994         struct btrfs_trans_handle *trans;
2995         struct btrfs_path *path;
2996         struct btrfs_key key;
2997         int ret, err;
2998
2999         path = btrfs_alloc_path();
3000         if (!path)
3001                 return -ENOMEM;
3002
3003         trans = btrfs_start_transaction(root, 0);
3004         if (IS_ERR(trans)) {
3005                 btrfs_free_path(path);
3006                 return PTR_ERR(trans);
3007         }
3008
3009         key.objectid = BTRFS_BALANCE_OBJECTID;
3010         key.type = BTRFS_TEMPORARY_ITEM_KEY;
3011         key.offset = 0;
3012
3013         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3014         if (ret < 0)
3015                 goto out;
3016         if (ret > 0) {
3017                 ret = -ENOENT;
3018                 goto out;
3019         }
3020
3021         ret = btrfs_del_item(trans, root, path);
3022 out:
3023         btrfs_free_path(path);
3024         err = btrfs_commit_transaction(trans, root);
3025         if (err && !ret)
3026                 ret = err;
3027         return ret;
3028 }
3029
3030 /*
3031  * This is a heuristic used to reduce the number of chunks balanced on
3032  * resume after balance was interrupted.
3033  */
3034 static void update_balance_args(struct btrfs_balance_control *bctl)
3035 {
3036         /*
3037          * Turn on soft mode for chunk types that were being converted.
3038          */
3039         if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
3040                 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
3041         if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
3042                 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
3043         if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
3044                 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
3045
3046         /*
3047          * Turn on usage filter if is not already used.  The idea is
3048          * that chunks that we have already balanced should be
3049          * reasonably full.  Don't do it for chunks that are being
3050          * converted - that will keep us from relocating unconverted
3051          * (albeit full) chunks.
3052          */
3053         if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3054             !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3055             !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3056                 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
3057                 bctl->data.usage = 90;
3058         }
3059         if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3060             !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3061             !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3062                 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
3063                 bctl->sys.usage = 90;
3064         }
3065         if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3066             !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3067             !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3068                 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
3069                 bctl->meta.usage = 90;
3070         }
3071 }
3072
3073 /*
3074  * Should be called with both balance and volume mutexes held to
3075  * serialize other volume operations (add_dev/rm_dev/resize) with
3076  * restriper.  Same goes for unset_balance_control.
3077  */
3078 static void set_balance_control(struct btrfs_balance_control *bctl)
3079 {
3080         struct btrfs_fs_info *fs_info = bctl->fs_info;
3081
3082         BUG_ON(fs_info->balance_ctl);
3083
3084         spin_lock(&fs_info->balance_lock);
3085         fs_info->balance_ctl = bctl;
3086         spin_unlock(&fs_info->balance_lock);
3087 }
3088
3089 static void unset_balance_control(struct btrfs_fs_info *fs_info)
3090 {
3091         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3092
3093         BUG_ON(!fs_info->balance_ctl);
3094
3095         spin_lock(&fs_info->balance_lock);
3096         fs_info->balance_ctl = NULL;
3097         spin_unlock(&fs_info->balance_lock);
3098
3099         kfree(bctl);
3100 }
3101
3102 /*
3103  * Balance filters.  Return 1 if chunk should be filtered out
3104  * (should not be balanced).
3105  */
3106 static int chunk_profiles_filter(u64 chunk_type,
3107                                  struct btrfs_balance_args *bargs)
3108 {
3109         chunk_type = chunk_to_extended(chunk_type) &
3110                                 BTRFS_EXTENDED_PROFILE_MASK;
3111
3112         if (bargs->profiles & chunk_type)
3113                 return 0;
3114
3115         return 1;
3116 }
3117
3118 static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
3119                               struct btrfs_balance_args *bargs)
3120 {
3121         struct btrfs_block_group_cache *cache;
3122         u64 chunk_used;
3123         u64 user_thresh_min;
3124         u64 user_thresh_max;
3125         int ret = 1;
3126
3127         cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3128         chunk_used = btrfs_block_group_used(&cache->item);
3129
3130         if (bargs->usage_min == 0)
3131                 user_thresh_min = 0;
3132         else
3133                 user_thresh_min = div_factor_fine(cache->key.offset,
3134                                         bargs->usage_min);
3135
3136         if (bargs->usage_max == 0)
3137                 user_thresh_max = 1;
3138         else if (bargs->usage_max > 100)
3139                 user_thresh_max = cache->key.offset;
3140         else
3141                 user_thresh_max = div_factor_fine(cache->key.offset,
3142                                         bargs->usage_max);
3143
3144         if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max)
3145                 ret = 0;
3146
3147         btrfs_put_block_group(cache);
3148         return ret;
3149 }
3150
3151 static int chunk_usage_filter(struct btrfs_fs_info *fs_info,
3152                 u64 chunk_offset, struct btrfs_balance_args *bargs)
3153 {
3154         struct btrfs_block_group_cache *cache;
3155         u64 chunk_used, user_thresh;
3156         int ret = 1;
3157
3158         cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3159         chunk_used = btrfs_block_group_used(&cache->item);
3160
3161         if (bargs->usage_min == 0)
3162                 user_thresh = 1;
3163         else if (bargs->usage > 100)
3164                 user_thresh = cache->key.offset;
3165         else
3166                 user_thresh = div_factor_fine(cache->key.offset,
3167                                               bargs->usage);
3168
3169         if (chunk_used < user_thresh)
3170                 ret = 0;
3171
3172         btrfs_put_block_group(cache);
3173         return ret;
3174 }
3175
3176 static int chunk_devid_filter(struct extent_buffer *leaf,
3177                               struct btrfs_chunk *chunk,
3178                               struct btrfs_balance_args *bargs)
3179 {
3180         struct btrfs_stripe *stripe;
3181         int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3182         int i;
3183
3184         for (i = 0; i < num_stripes; i++) {
3185                 stripe = btrfs_stripe_nr(chunk, i);
3186                 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
3187                         return 0;
3188         }
3189
3190         return 1;
3191 }
3192
3193 /* [pstart, pend) */
3194 static int chunk_drange_filter(struct extent_buffer *leaf,
3195                                struct btrfs_chunk *chunk,
3196                                u64 chunk_offset,
3197                                struct btrfs_balance_args *bargs)
3198 {
3199         struct btrfs_stripe *stripe;
3200         int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3201         u64 stripe_offset;
3202         u64 stripe_length;
3203         int factor;
3204         int i;
3205
3206         if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
3207                 return 0;
3208
3209         if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP |
3210              BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) {
3211                 factor = num_stripes / 2;
3212         } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) {
3213                 factor = num_stripes - 1;
3214         } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) {
3215                 factor = num_stripes - 2;
3216         } else {
3217                 factor = num_stripes;
3218         }
3219
3220         for (i = 0; i < num_stripes; i++) {
3221                 stripe = btrfs_stripe_nr(chunk, i);
3222                 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
3223                         continue;
3224
3225                 stripe_offset = btrfs_stripe_offset(leaf, stripe);
3226                 stripe_length = btrfs_chunk_length(leaf, chunk);
3227                 stripe_length = div_u64(stripe_length, factor);
3228
3229                 if (stripe_offset < bargs->pend &&
3230                     stripe_offset + stripe_length > bargs->pstart)
3231                         return 0;
3232         }
3233
3234         return 1;
3235 }
3236
3237 /* [vstart, vend) */
3238 static int chunk_vrange_filter(struct extent_buffer *leaf,
3239                                struct btrfs_chunk *chunk,
3240                                u64 chunk_offset,
3241                                struct btrfs_balance_args *bargs)
3242 {
3243         if (chunk_offset < bargs->vend &&
3244             chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
3245                 /* at least part of the chunk is inside this vrange */
3246                 return 0;
3247
3248         return 1;
3249 }
3250
3251 static int chunk_stripes_range_filter(struct extent_buffer *leaf,
3252                                struct btrfs_chunk *chunk,
3253                                struct btrfs_balance_args *bargs)
3254 {
3255         int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3256
3257         if (bargs->stripes_min <= num_stripes
3258                         && num_stripes <= bargs->stripes_max)
3259                 return 0;
3260
3261         return 1;
3262 }
3263
3264 static int chunk_soft_convert_filter(u64 chunk_type,
3265                                      struct btrfs_balance_args *bargs)
3266 {
3267         if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
3268                 return 0;
3269
3270         chunk_type = chunk_to_extended(chunk_type) &
3271                                 BTRFS_EXTENDED_PROFILE_MASK;
3272
3273         if (bargs->target == chunk_type)
3274                 return 1;
3275
3276         return 0;
3277 }
3278
3279 static int should_balance_chunk(struct btrfs_root *root,
3280                                 struct extent_buffer *leaf,
3281                                 struct btrfs_chunk *chunk, u64 chunk_offset)
3282 {
3283         struct btrfs_balance_control *bctl = root->fs_info->balance_ctl;
3284         struct btrfs_balance_args *bargs = NULL;
3285         u64 chunk_type = btrfs_chunk_type(leaf, chunk);
3286
3287         /* type filter */
3288         if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
3289               (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
3290                 return 0;
3291         }
3292
3293         if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3294                 bargs = &bctl->data;
3295         else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3296                 bargs = &bctl->sys;
3297         else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3298                 bargs = &bctl->meta;
3299
3300         /* profiles filter */
3301         if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
3302             chunk_profiles_filter(chunk_type, bargs)) {
3303                 return 0;
3304         }
3305
3306         /* usage filter */
3307         if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
3308             chunk_usage_filter(bctl->fs_info, chunk_offset, bargs)) {
3309                 return 0;
3310         } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3311             chunk_usage_range_filter(bctl->fs_info, chunk_offset, bargs)) {
3312                 return 0;
3313         }
3314
3315         /* devid filter */
3316         if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
3317             chunk_devid_filter(leaf, chunk, bargs)) {
3318                 return 0;
3319         }
3320
3321         /* drange filter, makes sense only with devid filter */
3322         if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
3323             chunk_drange_filter(leaf, chunk, chunk_offset, bargs)) {
3324                 return 0;
3325         }
3326
3327         /* vrange filter */
3328         if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
3329             chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
3330                 return 0;
3331         }
3332
3333         /* stripes filter */
3334         if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) &&
3335             chunk_stripes_range_filter(leaf, chunk, bargs)) {
3336                 return 0;
3337         }
3338
3339         /* soft profile changing mode */
3340         if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
3341             chunk_soft_convert_filter(chunk_type, bargs)) {
3342                 return 0;
3343         }
3344
3345         /*
3346          * limited by count, must be the last filter
3347          */
3348         if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) {
3349                 if (bargs->limit == 0)
3350                         return 0;
3351                 else
3352                         bargs->limit--;
3353         } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) {
3354                 /*
3355                  * Same logic as the 'limit' filter; the minimum cannot be
3356                  * determined here because we do not have the global informatoin
3357                  * about the count of all chunks that satisfy the filters.
3358                  */
3359                 if (bargs->limit_max == 0)
3360                         return 0;
3361                 else
3362                         bargs->limit_max--;
3363         }
3364
3365         return 1;
3366 }
3367
3368 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
3369 {
3370         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3371         struct btrfs_root *chunk_root = fs_info->chunk_root;
3372         struct btrfs_root *dev_root = fs_info->dev_root;
3373         struct list_head *devices;
3374         struct btrfs_device *device;
3375         u64 old_size;
3376         u64 size_to_free;
3377         u64 chunk_type;
3378         struct btrfs_chunk *chunk;
3379         struct btrfs_path *path;
3380         struct btrfs_key key;
3381         struct btrfs_key found_key;
3382         struct btrfs_trans_handle *trans;
3383         struct extent_buffer *leaf;
3384         int slot;
3385         int ret;
3386         int enospc_errors = 0;
3387         bool counting = true;
3388         /* The single value limit and min/max limits use the same bytes in the */
3389         u64 limit_data = bctl->data.limit;
3390         u64 limit_meta = bctl->meta.limit;
3391         u64 limit_sys = bctl->sys.limit;
3392         u32 count_data = 0;
3393         u32 count_meta = 0;
3394         u32 count_sys = 0;
3395         int chunk_reserved = 0;
3396
3397         /* step one make some room on all the devices */
3398         devices = &fs_info->fs_devices->devices;
3399         list_for_each_entry(device, devices, dev_list) {
3400                 old_size = btrfs_device_get_total_bytes(device);
3401                 size_to_free = div_factor(old_size, 1);
3402                 size_to_free = min_t(u64, size_to_free, SZ_1M);
3403                 if (!device->writeable ||
3404                     btrfs_device_get_total_bytes(device) -
3405                     btrfs_device_get_bytes_used(device) > size_to_free ||
3406                     device->is_tgtdev_for_dev_replace)
3407                         continue;
3408
3409                 ret = btrfs_shrink_device(device, old_size - size_to_free);
3410                 if (ret == -ENOSPC)
3411                         break;
3412                 BUG_ON(ret);
3413
3414                 trans = btrfs_start_transaction(dev_root, 0);
3415                 BUG_ON(IS_ERR(trans));
3416
3417                 ret = btrfs_grow_device(trans, device, old_size);
3418                 BUG_ON(ret);
3419
3420                 btrfs_end_transaction(trans, dev_root);
3421         }
3422
3423         /* step two, relocate all the chunks */
3424         path = btrfs_alloc_path();
3425         if (!path) {
3426                 ret = -ENOMEM;
3427                 goto error;
3428         }
3429
3430         /* zero out stat counters */
3431         spin_lock(&fs_info->balance_lock);
3432         memset(&bctl->stat, 0, sizeof(bctl->stat));
3433         spin_unlock(&fs_info->balance_lock);
3434 again:
3435         if (!counting) {
3436                 /*
3437                  * The single value limit and min/max limits use the same bytes
3438                  * in the
3439                  */
3440                 bctl->data.limit = limit_data;
3441                 bctl->meta.limit = limit_meta;
3442                 bctl->sys.limit = limit_sys;
3443         }
3444         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3445         key.offset = (u64)-1;
3446         key.type = BTRFS_CHUNK_ITEM_KEY;
3447
3448         while (1) {
3449                 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
3450                     atomic_read(&fs_info->balance_cancel_req)) {
3451                         ret = -ECANCELED;
3452                         goto error;
3453                 }
3454
3455                 mutex_lock(&fs_info->delete_unused_bgs_mutex);
3456                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3457                 if (ret < 0) {
3458                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3459                         goto error;
3460                 }
3461
3462                 /*
3463                  * this shouldn't happen, it means the last relocate
3464                  * failed
3465                  */
3466                 if (ret == 0)
3467                         BUG(); /* FIXME break ? */
3468
3469                 ret = btrfs_previous_item(chunk_root, path, 0,
3470                                           BTRFS_CHUNK_ITEM_KEY);
3471                 if (ret) {
3472                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3473                         ret = 0;
3474                         break;
3475                 }
3476
3477                 leaf = path->nodes[0];
3478                 slot = path->slots[0];
3479                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3480
3481                 if (found_key.objectid != key.objectid) {
3482                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3483                         break;
3484                 }
3485
3486                 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3487                 chunk_type = btrfs_chunk_type(leaf, chunk);
3488
3489                 if (!counting) {
3490                         spin_lock(&fs_info->balance_lock);
3491                         bctl->stat.considered++;
3492                         spin_unlock(&fs_info->balance_lock);
3493                 }
3494
3495                 ret = should_balance_chunk(chunk_root, leaf, chunk,
3496                                            found_key.offset);
3497
3498                 btrfs_release_path(path);
3499                 if (!ret) {
3500                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3501                         goto loop;
3502                 }
3503
3504                 if (counting) {
3505                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3506                         spin_lock(&fs_info->balance_lock);
3507                         bctl->stat.expected++;
3508                         spin_unlock(&fs_info->balance_lock);
3509
3510                         if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3511                                 count_data++;
3512                         else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3513                                 count_sys++;
3514                         else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3515                                 count_meta++;
3516
3517                         goto loop;
3518                 }
3519
3520                 /*
3521                  * Apply limit_min filter, no need to check if the LIMITS
3522                  * filter is used, limit_min is 0 by default
3523                  */
3524                 if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
3525                                         count_data < bctl->data.limit_min)
3526                                 || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) &&
3527                                         count_meta < bctl->meta.limit_min)
3528                                 || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) &&
3529                                         count_sys < bctl->sys.limit_min)) {
3530                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3531                         goto loop;
3532                 }
3533
3534                 if ((chunk_type & BTRFS_BLOCK_GROUP_DATA) && !chunk_reserved) {
3535                         trans = btrfs_start_transaction(chunk_root, 0);
3536                         if (IS_ERR(trans)) {
3537                                 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3538                                 ret = PTR_ERR(trans);
3539                                 goto error;
3540                         }
3541
3542                         ret = btrfs_force_chunk_alloc(trans, chunk_root,
3543                                                       BTRFS_BLOCK_GROUP_DATA);
3544                         btrfs_end_transaction(trans, chunk_root);
3545                         if (ret < 0) {
3546                                 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3547                                 goto error;
3548                         }
3549                         chunk_reserved = 1;
3550                 }
3551
3552                 ret = btrfs_relocate_chunk(chunk_root,
3553                                            found_key.offset);
3554                 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3555                 if (ret && ret != -ENOSPC)
3556                         goto error;
3557                 if (ret == -ENOSPC) {
3558                         enospc_errors++;
3559                 } else {
3560                         spin_lock(&fs_info->balance_lock);
3561                         bctl->stat.completed++;
3562                         spin_unlock(&fs_info->balance_lock);
3563                 }
3564 loop:
3565                 if (found_key.offset == 0)
3566                         break;
3567                 key.offset = found_key.offset - 1;
3568         }
3569
3570         if (counting) {
3571                 btrfs_release_path(path);
3572                 counting = false;
3573                 goto again;
3574         }
3575 error:
3576         btrfs_free_path(path);
3577         if (enospc_errors) {
3578                 btrfs_info(fs_info, "%d enospc errors during balance",
3579                        enospc_errors);
3580                 if (!ret)
3581                         ret = -ENOSPC;
3582         }
3583
3584         return ret;
3585 }
3586
3587 /**
3588  * alloc_profile_is_valid - see if a given profile is valid and reduced
3589  * @flags: profile to validate
3590  * @extended: if true @flags is treated as an extended profile
3591  */
3592 static int alloc_profile_is_valid(u64 flags, int extended)
3593 {
3594         u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
3595                                BTRFS_BLOCK_GROUP_PROFILE_MASK);
3596
3597         flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
3598
3599         /* 1) check that all other bits are zeroed */
3600         if (flags & ~mask)
3601                 return 0;
3602
3603         /* 2) see if profile is reduced */
3604         if (flags == 0)
3605                 return !extended; /* "0" is valid for usual profiles */
3606
3607         /* true if exactly one bit set */
3608         return (flags & (flags - 1)) == 0;
3609 }
3610
3611 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
3612 {
3613         /* cancel requested || normal exit path */
3614         return atomic_read(&fs_info->balance_cancel_req) ||
3615                 (atomic_read(&fs_info->balance_pause_req) == 0 &&
3616                  atomic_read(&fs_info->balance_cancel_req) == 0);
3617 }
3618
3619 static void __cancel_balance(struct btrfs_fs_info *fs_info)
3620 {
3621         int ret;
3622
3623         unset_balance_control(fs_info);
3624         ret = del_balance_item(fs_info->tree_root);
3625         if (ret)
3626                 btrfs_std_error(fs_info, ret, NULL);
3627
3628         atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3629 }
3630
3631 /* Non-zero return value signifies invalidity */
3632 static inline int validate_convert_profile(struct btrfs_balance_args *bctl_arg,
3633                 u64 allowed)
3634 {
3635         return ((bctl_arg->flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3636                 (!alloc_profile_is_valid(bctl_arg->target, 1) ||
3637                  (bctl_arg->target & ~allowed)));
3638 }
3639
3640 /*
3641  * Should be called with both balance and volume mutexes held
3642  */
3643 int btrfs_balance(struct btrfs_balance_control *bctl,
3644                   struct btrfs_ioctl_balance_args *bargs)
3645 {
3646         struct btrfs_fs_info *fs_info = bctl->fs_info;
3647         u64 allowed;
3648         int mixed = 0;
3649         int ret;
3650         u64 num_devices;
3651         unsigned seq;
3652
3653         if (btrfs_fs_closing(fs_info) ||
3654             atomic_read(&fs_info->balance_pause_req) ||
3655             atomic_read(&fs_info->balance_cancel_req)) {
3656                 ret = -EINVAL;
3657                 goto out;
3658         }
3659
3660         allowed = btrfs_super_incompat_flags(fs_info->super_copy);
3661         if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
3662                 mixed = 1;
3663
3664         /*
3665          * In case of mixed groups both data and meta should be picked,
3666          * and identical options should be given for both of them.
3667          */
3668         allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
3669         if (mixed && (bctl->flags & allowed)) {
3670                 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
3671                     !(bctl->flags & BTRFS_BALANCE_METADATA) ||
3672                     memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
3673                         btrfs_err(fs_info, "with mixed groups data and "
3674                                    "metadata balance options must be the same");
3675                         ret = -EINVAL;
3676                         goto out;
3677                 }
3678         }
3679
3680         num_devices = fs_info->fs_devices->num_devices;
3681         btrfs_dev_replace_lock(&fs_info->dev_replace, 0);
3682         if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
3683                 BUG_ON(num_devices < 1);
3684                 num_devices--;
3685         }
3686         btrfs_dev_replace_unlock(&fs_info->dev_replace, 0);
3687         allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE;
3688         if (num_devices == 1)
3689                 allowed |= BTRFS_BLOCK_GROUP_DUP;
3690         else if (num_devices > 1)
3691                 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1);
3692         if (num_devices > 2)
3693                 allowed |= BTRFS_BLOCK_GROUP_RAID5;
3694         if (num_devices > 3)
3695                 allowed |= (BTRFS_BLOCK_GROUP_RAID10 |
3696                             BTRFS_BLOCK_GROUP_RAID6);
3697         if (validate_convert_profile(&bctl->data, allowed)) {
3698                 btrfs_err(fs_info, "unable to start balance with target "
3699                            "data profile %llu",
3700                        bctl->data.target);
3701                 ret = -EINVAL;
3702                 goto out;
3703         }
3704         if (validate_convert_profile(&bctl->meta, allowed)) {
3705                 btrfs_err(fs_info,
3706                            "unable to start balance with target metadata profile %llu",
3707                        bctl->meta.target);
3708                 ret = -EINVAL;
3709                 goto out;
3710         }
3711         if (validate_convert_profile(&bctl->sys, allowed)) {
3712                 btrfs_err(fs_info,
3713                            "unable to start balance with target system profile %llu",
3714                        bctl->sys.target);
3715                 ret = -EINVAL;
3716                 goto out;
3717         }
3718
3719         /* allow to reduce meta or sys integrity only if force set */
3720         allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3721                         BTRFS_BLOCK_GROUP_RAID10 |
3722                         BTRFS_BLOCK_GROUP_RAID5 |
3723                         BTRFS_BLOCK_GROUP_RAID6;
3724         do {
3725                 seq = read_seqbegin(&fs_info->profiles_lock);
3726
3727                 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3728                      (fs_info->avail_system_alloc_bits & allowed) &&
3729                      !(bctl->sys.target & allowed)) ||
3730                     ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3731                      (fs_info->avail_metadata_alloc_bits & allowed) &&
3732                      !(bctl->meta.target & allowed))) {
3733                         if (bctl->flags & BTRFS_BALANCE_FORCE) {
3734                                 btrfs_info(fs_info, "force reducing metadata integrity");
3735                         } else {
3736                                 btrfs_err(fs_info, "balance will reduce metadata "
3737                                            "integrity, use force if you want this");
3738                                 ret = -EINVAL;
3739                                 goto out;
3740                         }
3741                 }
3742         } while (read_seqretry(&fs_info->profiles_lock, seq));
3743
3744         if (btrfs_get_num_tolerated_disk_barrier_failures(bctl->meta.target) <
3745                 btrfs_get_num_tolerated_disk_barrier_failures(bctl->data.target)) {
3746                 btrfs_warn(fs_info,
3747         "metadata profile 0x%llx has lower redundancy than data profile 0x%llx",
3748                         bctl->meta.target, bctl->data.target);
3749         }
3750
3751         if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3752                 fs_info->num_tolerated_disk_barrier_failures = min(
3753                         btrfs_calc_num_tolerated_disk_barrier_failures(fs_info),
3754                         btrfs_get_num_tolerated_disk_barrier_failures(
3755                                 bctl->sys.target));
3756         }
3757
3758         ret = insert_balance_item(fs_info->tree_root, bctl);
3759         if (ret && ret != -EEXIST)
3760                 goto out;
3761
3762         if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
3763                 BUG_ON(ret == -EEXIST);
3764                 set_balance_control(bctl);
3765         } else {
3766                 BUG_ON(ret != -EEXIST);
3767                 spin_lock(&fs_info->balance_lock);
3768                 update_balance_args(bctl);
3769                 spin_unlock(&fs_info->balance_lock);
3770         }
3771
3772         atomic_inc(&fs_info->balance_running);
3773         mutex_unlock(&fs_info->balance_mutex);
3774
3775         ret = __btrfs_balance(fs_info);
3776
3777         mutex_lock(&fs_info->balance_mutex);
3778         atomic_dec(&fs_info->balance_running);
3779
3780         if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3781                 fs_info->num_tolerated_disk_barrier_failures =
3782                         btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3783         }
3784
3785         if (bargs) {
3786                 memset(bargs, 0, sizeof(*bargs));
3787                 update_ioctl_balance_args(fs_info, 0, bargs);
3788         }
3789
3790         if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
3791             balance_need_close(fs_info)) {
3792                 __cancel_balance(fs_info);
3793         }
3794
3795         wake_up(&fs_info->balance_wait_q);
3796
3797         return ret;
3798 out:
3799         if (bctl->flags & BTRFS_BALANCE_RESUME)
3800                 __cancel_balance(fs_info);
3801         else {
3802                 kfree(bctl);
3803                 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3804         }
3805         return ret;
3806 }
3807
3808 static int balance_kthread(void *data)
3809 {
3810         struct btrfs_fs_info *fs_info = data;
3811         int ret = 0;
3812
3813         mutex_lock(&fs_info->volume_mutex);
3814         mutex_lock(&fs_info->balance_mutex);
3815
3816         if (fs_info->balance_ctl) {
3817                 btrfs_info(fs_info, "continuing balance");
3818                 ret = btrfs_balance(fs_info->balance_ctl, NULL);
3819         }
3820
3821         mutex_unlock(&fs_info->balance_mutex);
3822         mutex_unlock(&fs_info->volume_mutex);
3823
3824         return ret;
3825 }
3826
3827 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
3828 {
3829         struct task_struct *tsk;
3830
3831         spin_lock(&fs_info->balance_lock);
3832         if (!fs_info->balance_ctl) {
3833                 spin_unlock(&fs_info->balance_lock);
3834                 return 0;
3835         }
3836         spin_unlock(&fs_info->balance_lock);
3837
3838         if (btrfs_test_opt(fs_info->tree_root, SKIP_BALANCE)) {
3839                 btrfs_info(fs_info, "force skipping balance");
3840                 return 0;
3841         }
3842
3843         tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
3844         return PTR_ERR_OR_ZERO(tsk);
3845 }
3846
3847 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
3848 {
3849         struct btrfs_balance_control *bctl;
3850         struct btrfs_balance_item *item;
3851         struct btrfs_disk_balance_args disk_bargs;
3852         struct btrfs_path *path;
3853         struct extent_buffer *leaf;
3854         struct btrfs_key key;
3855         int ret;
3856
3857         path = btrfs_alloc_path();
3858         if (!path)
3859                 return -ENOMEM;
3860
3861         key.objectid = BTRFS_BALANCE_OBJECTID;
3862         key.type = BTRFS_TEMPORARY_ITEM_KEY;
3863         key.offset = 0;
3864
3865         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
3866         if (ret < 0)
3867                 goto out;
3868         if (ret > 0) { /* ret = -ENOENT; */
3869                 ret = 0;
3870                 goto out;
3871         }
3872
3873         bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
3874         if (!bctl) {
3875                 ret = -ENOMEM;
3876                 goto out;
3877         }
3878
3879         leaf = path->nodes[0];
3880         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3881
3882         bctl->fs_info = fs_info;
3883         bctl->flags = btrfs_balance_flags(leaf, item);
3884         bctl->flags |= BTRFS_BALANCE_RESUME;
3885
3886         btrfs_balance_data(leaf, item, &disk_bargs);
3887         btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
3888         btrfs_balance_meta(leaf, item, &disk_bargs);
3889         btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
3890         btrfs_balance_sys(leaf, item, &disk_bargs);
3891         btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
3892
3893         WARN_ON(atomic_xchg(&fs_info->mutually_exclusive_operation_running, 1));
3894
3895         mutex_lock(&fs_info->volume_mutex);
3896         mutex_lock(&fs_info->balance_mutex);
3897
3898         set_balance_control(bctl);
3899
3900         mutex_unlock(&fs_info->balance_mutex);
3901         mutex_unlock(&fs_info->volume_mutex);
3902 out:
3903         btrfs_free_path(path);
3904         return ret;
3905 }
3906
3907 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
3908 {
3909         int ret = 0;
3910
3911         mutex_lock(&fs_info->balance_mutex);
3912         if (!fs_info->balance_ctl) {
3913                 mutex_unlock(&fs_info->balance_mutex);
3914                 return -ENOTCONN;
3915         }
3916
3917         if (atomic_read(&fs_info->balance_running)) {
3918                 atomic_inc(&fs_info->balance_pause_req);
3919                 mutex_unlock(&fs_info->balance_mutex);
3920
3921                 wait_event(fs_info->balance_wait_q,
3922                            atomic_read(&fs_info->balance_running) == 0);
3923
3924                 mutex_lock(&fs_info->balance_mutex);
3925                 /* we are good with balance_ctl ripped off from under us */
3926                 BUG_ON(atomic_read(&fs_info->balance_running));
3927                 atomic_dec(&fs_info->balance_pause_req);
3928         } else {
3929                 ret = -ENOTCONN;
3930         }
3931
3932         mutex_unlock(&fs_info->balance_mutex);
3933         return ret;
3934 }
3935
3936 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
3937 {
3938         if (fs_info->sb->s_flags & MS_RDONLY)
3939                 return -EROFS;
3940
3941         mutex_lock(&fs_info->balance_mutex);
3942         if (!fs_info->balance_ctl) {
3943                 mutex_unlock(&fs_info->balance_mutex);
3944                 return -ENOTCONN;
3945         }
3946
3947         atomic_inc(&fs_info->balance_cancel_req);
3948         /*
3949          * if we are running just wait and return, balance item is
3950          * deleted in btrfs_balance in this case
3951          */
3952         if (atomic_read(&fs_info->balance_running)) {
3953                 mutex_unlock(&fs_info->balance_mutex);
3954                 wait_event(fs_info->balance_wait_q,
3955                            atomic_read(&fs_info->balance_running) == 0);
3956                 mutex_lock(&fs_info->balance_mutex);
3957         } else {
3958                 /* __cancel_balance needs volume_mutex */
3959                 mutex_unlock(&fs_info->balance_mutex);
3960                 mutex_lock(&fs_info->volume_mutex);
3961                 mutex_lock(&fs_info->balance_mutex);
3962
3963                 if (fs_info->balance_ctl)
3964                         __cancel_balance(fs_info);
3965
3966                 mutex_unlock(&fs_info->volume_mutex);
3967         }
3968
3969         BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running));
3970         atomic_dec(&fs_info->balance_cancel_req);
3971         mutex_unlock(&fs_info->balance_mutex);
3972         return 0;
3973 }
3974
3975 static int btrfs_uuid_scan_kthread(void *data)
3976 {
3977         struct btrfs_fs_info *fs_info = data;
3978         struct btrfs_root *root = fs_info->tree_root;
3979         struct btrfs_key key;
3980         struct btrfs_key max_key;
3981         struct btrfs_path *path = NULL;
3982         int ret = 0;
3983         struct extent_buffer *eb;
3984         int slot;
3985         struct btrfs_root_item root_item;
3986         u32 item_size;
3987         struct btrfs_trans_handle *trans = NULL;
3988
3989         path = btrfs_alloc_path();
3990         if (!path) {
3991                 ret = -ENOMEM;
3992                 goto out;
3993         }
3994
3995         key.objectid = 0;
3996         key.type = BTRFS_ROOT_ITEM_KEY;
3997         key.offset = 0;
3998
3999         max_key.objectid = (u64)-1;
4000         max_key.type = BTRFS_ROOT_ITEM_KEY;
4001         max_key.offset = (u64)-1;
4002
4003         while (1) {
4004                 ret = btrfs_search_forward(root, &key, path, 0);
4005                 if (ret) {
4006                         if (ret > 0)
4007                                 ret = 0;
4008                         break;
4009                 }
4010
4011                 if (key.type != BTRFS_ROOT_ITEM_KEY ||
4012                     (key.objectid < BTRFS_FIRST_FREE_OBJECTID &&
4013                      key.objectid != BTRFS_FS_TREE_OBJECTID) ||
4014                     key.objectid > BTRFS_LAST_FREE_OBJECTID)
4015                         goto skip;
4016
4017                 eb = path->nodes[0];
4018                 slot = path->slots[0];
4019                 item_size = btrfs_item_size_nr(eb, slot);
4020                 if (item_size < sizeof(root_item))
4021                         goto skip;
4022
4023                 read_extent_buffer(eb, &root_item,
4024                                    btrfs_item_ptr_offset(eb, slot),
4025                                    (int)sizeof(root_item));
4026                 if (btrfs_root_refs(&root_item) == 0)
4027                         goto skip;
4028
4029                 if (!btrfs_is_empty_uuid(root_item.uuid) ||
4030                     !btrfs_is_empty_uuid(root_item.received_uuid)) {
4031                         if (trans)
4032                                 goto update_tree;
4033
4034                         btrfs_release_path(path);
4035                         /*
4036                          * 1 - subvol uuid item
4037                          * 1 - received_subvol uuid item
4038                          */
4039                         trans = btrfs_start_transaction(fs_info->uuid_root, 2);
4040                         if (IS_ERR(trans)) {
4041                                 ret = PTR_ERR(trans);
4042                                 break;
4043                         }
4044                         continue;
4045                 } else {
4046                         goto skip;
4047                 }
4048 update_tree:
4049                 if (!btrfs_is_empty_uuid(root_item.uuid)) {
4050                         ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
4051                                                   root_item.uuid,
4052                                                   BTRFS_UUID_KEY_SUBVOL,
4053                                                   key.objectid);
4054                         if (ret < 0) {
4055                                 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4056                                         ret);
4057                                 break;
4058                         }
4059                 }
4060
4061                 if (!btrfs_is_empty_uuid(root_item.received_uuid)) {
4062                         ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
4063                                                   root_item.received_uuid,
4064                                                  BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4065                                                   key.objectid);
4066                         if (ret < 0) {
4067                                 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4068                                         ret);
4069                                 break;
4070                         }
4071                 }
4072
4073 skip:
4074                 if (trans) {
4075                         ret = btrfs_end_transaction(trans, fs_info->uuid_root);
4076                         trans = NULL;
4077                         if (ret)
4078                                 break;
4079                 }
4080
4081                 btrfs_release_path(path);
4082                 if (key.offset < (u64)-1) {
4083                         key.offset++;
4084                 } else if (key.type < BTRFS_ROOT_ITEM_KEY) {
4085                         key.offset = 0;
4086                         key.type = BTRFS_ROOT_ITEM_KEY;
4087                 } else if (key.objectid < (u64)-1) {
4088                         key.offset = 0;
4089                         key.type = BTRFS_ROOT_ITEM_KEY;
4090                         key.objectid++;
4091                 } else {
4092                         break;
4093                 }
4094                 cond_resched();
4095         }
4096
4097 out:
4098         btrfs_free_path(path);
4099         if (trans && !IS_ERR(trans))
4100                 btrfs_end_transaction(trans, fs_info->uuid_root);
4101         if (ret)
4102                 btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret);
4103         else
4104                 fs_info->update_uuid_tree_gen = 1;
4105         up(&fs_info->uuid_tree_rescan_sem);
4106         return 0;
4107 }
4108
4109 /*
4110  * Callback for btrfs_uuid_tree_iterate().
4111  * returns:
4112  * 0    check succeeded, the entry is not outdated.
4113  * < 0  if an error occurred.
4114  * > 0  if the check failed, which means the caller shall remove the entry.
4115  */
4116 static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info *fs_info,
4117                                        u8 *uuid, u8 type, u64 subid)
4118 {
4119         struct btrfs_key key;
4120         int ret = 0;
4121         struct btrfs_root *subvol_root;
4122
4123         if (type != BTRFS_UUID_KEY_SUBVOL &&
4124             type != BTRFS_UUID_KEY_RECEIVED_SUBVOL)
4125                 goto out;
4126
4127         key.objectid = subid;
4128         key.type = BTRFS_ROOT_ITEM_KEY;
4129         key.offset = (u64)-1;
4130         subvol_root = btrfs_read_fs_root_no_name(fs_info, &key);
4131         if (IS_ERR(subvol_root)) {
4132                 ret = PTR_ERR(subvol_root);
4133                 if (ret == -ENOENT)
4134                         ret = 1;
4135                 goto out;
4136         }
4137
4138         switch (type) {
4139         case BTRFS_UUID_KEY_SUBVOL:
4140                 if (memcmp(uuid, subvol_root->root_item.uuid, BTRFS_UUID_SIZE))
4141                         ret = 1;
4142                 break;
4143         case BTRFS_UUID_KEY_RECEIVED_SUBVOL:
4144                 if (memcmp(uuid, subvol_root->root_item.received_uuid,
4145                            BTRFS_UUID_SIZE))
4146                         ret = 1;
4147                 break;
4148         }
4149
4150 out:
4151         return ret;
4152 }
4153
4154 static int btrfs_uuid_rescan_kthread(void *data)
4155 {
4156         struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data;
4157         int ret;
4158
4159         /*
4160          * 1st step is to iterate through the existing UUID tree and
4161          * to delete all entries that contain outdated data.
4162          * 2nd step is to add all missing entries to the UUID tree.
4163          */
4164         ret = btrfs_uuid_tree_iterate(fs_info, btrfs_check_uuid_tree_entry);
4165         if (ret < 0) {
4166                 btrfs_warn(fs_info, "iterating uuid_tree failed %d", ret);
4167                 up(&fs_info->uuid_tree_rescan_sem);
4168                 return ret;
4169         }
4170         return btrfs_uuid_scan_kthread(data);
4171 }
4172
4173 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info)
4174 {
4175         struct btrfs_trans_handle *trans;
4176         struct btrfs_root *tree_root = fs_info->tree_root;
4177         struct btrfs_root *uuid_root;
4178         struct task_struct *task;
4179         int ret;
4180
4181         /*
4182          * 1 - root node
4183          * 1 - root item
4184          */
4185         trans = btrfs_start_transaction(tree_root, 2);
4186         if (IS_ERR(trans))
4187                 return PTR_ERR(trans);
4188
4189         uuid_root = btrfs_create_tree(trans, fs_info,
4190                                       BTRFS_UUID_TREE_OBJECTID);
4191         if (IS_ERR(uuid_root)) {
4192                 ret = PTR_ERR(uuid_root);
4193                 btrfs_abort_transaction(trans, tree_root, ret);
4194                 return ret;
4195         }
4196
4197         fs_info->uuid_root = uuid_root;
4198
4199         ret = btrfs_commit_transaction(trans, tree_root);
4200         if (ret)
4201                 return ret;
4202
4203         down(&fs_info->uuid_tree_rescan_sem);
4204         task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid");
4205         if (IS_ERR(task)) {
4206                 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4207                 btrfs_warn(fs_info, "failed to start uuid_scan task");
4208                 up(&fs_info->uuid_tree_rescan_sem);
4209                 return PTR_ERR(task);
4210         }
4211
4212         return 0;
4213 }
4214
4215 int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
4216 {
4217         struct task_struct *task;
4218
4219         down(&fs_info->uuid_tree_rescan_sem);
4220         task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
4221         if (IS_ERR(task)) {
4222                 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4223                 btrfs_warn(fs_info, "failed to start uuid_rescan task");
4224                 up(&fs_info->uuid_tree_rescan_sem);
4225                 return PTR_ERR(task);
4226         }
4227
4228         return 0;
4229 }
4230
4231 /*
4232  * shrinking a device means finding all of the device extents past
4233  * the new size, and then following the back refs to the chunks.
4234  * The chunk relocation code actually frees the device extent
4235  */
4236 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
4237 {
4238         struct btrfs_trans_handle *trans;
4239         struct btrfs_root *root = device->dev_root;
4240         struct btrfs_dev_extent *dev_extent = NULL;
4241         struct btrfs_path *path;
4242         u64 length;
4243         u64 chunk_offset;
4244         int ret;
4245         int slot;
4246         int failed = 0;
4247         bool retried = false;
4248         bool checked_pending_chunks = false;
4249         struct extent_buffer *l;
4250         struct btrfs_key key;
4251         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
4252         u64 old_total = btrfs_super_total_bytes(super_copy);
4253         u64 old_size = btrfs_device_get_total_bytes(device);
4254         u64 diff = old_size - new_size;
4255
4256         if (device->is_tgtdev_for_dev_replace)
4257                 return -EINVAL;
4258
4259         path = btrfs_alloc_path();
4260         if (!path)
4261                 return -ENOMEM;
4262
4263         path->reada = READA_FORWARD;
4264
4265         lock_chunks(root);
4266
4267         btrfs_device_set_total_bytes(device, new_size);
4268         if (device->writeable) {
4269                 device->fs_devices->total_rw_bytes -= diff;
4270                 spin_lock(&root->fs_info->free_chunk_lock);
4271                 root->fs_info->free_chunk_space -= diff;
4272                 spin_unlock(&root->fs_info->free_chunk_lock);
4273         }
4274         unlock_chunks(root);
4275
4276 again:
4277         key.objectid = device->devid;
4278         key.offset = (u64)-1;
4279         key.type = BTRFS_DEV_EXTENT_KEY;
4280
4281         do {
4282                 mutex_lock(&root->fs_info->delete_unused_bgs_mutex);
4283                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4284                 if (ret < 0) {
4285                         mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4286                         goto done;
4287                 }
4288
4289                 ret = btrfs_previous_item(root, path, 0, key.type);
4290                 if (ret)
4291                         mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4292                 if (ret < 0)
4293                         goto done;
4294                 if (ret) {
4295                         ret = 0;
4296                         btrfs_release_path(path);
4297                         break;
4298                 }
4299
4300                 l = path->nodes[0];
4301                 slot = path->slots[0];
4302                 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
4303
4304                 if (key.objectid != device->devid) {
4305                         mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4306                         btrfs_release_path(path);
4307                         break;
4308                 }
4309
4310                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
4311                 length = btrfs_dev_extent_length(l, dev_extent);
4312
4313                 if (key.offset + length <= new_size) {
4314                         mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4315                         btrfs_release_path(path);
4316                         break;
4317                 }
4318
4319                 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
4320                 btrfs_release_path(path);
4321
4322                 ret = btrfs_relocate_chunk(root, chunk_offset);
4323                 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4324                 if (ret && ret != -ENOSPC)
4325                         goto done;
4326                 if (ret == -ENOSPC)
4327                         failed++;
4328         } while (key.offset-- > 0);
4329
4330         if (failed && !retried) {
4331                 failed = 0;
4332                 retried = true;
4333                 goto again;
4334         } else if (failed && retried) {
4335                 ret = -ENOSPC;
4336                 goto done;
4337         }
4338
4339         /* Shrinking succeeded, else we would be at "done". */
4340         trans = btrfs_start_transaction(root, 0);
4341         if (IS_ERR(trans)) {
4342                 ret = PTR_ERR(trans);
4343                 goto done;
4344         }
4345
4346         lock_chunks(root);
4347
4348         /*
4349          * We checked in the above loop all device extents that were already in
4350          * the device tree. However before we have updated the device's
4351          * total_bytes to the new size, we might have had chunk allocations that
4352          * have not complete yet (new block groups attached to transaction
4353          * handles), and therefore their device extents were not yet in the
4354          * device tree and we missed them in the loop above. So if we have any
4355          * pending chunk using a device extent that overlaps the device range
4356          * that we can not use anymore, commit the current transaction and
4357          * repeat the search on the device tree - this way we guarantee we will
4358          * not have chunks using device extents that end beyond 'new_size'.
4359          */
4360         if (!checked_pending_chunks) {
4361                 u64 start = new_size;
4362                 u64 len = old_size - new_size;
4363
4364                 if (contains_pending_extent(trans->transaction, device,
4365                                             &start, len)) {
4366                         unlock_chunks(root);
4367                         checked_pending_chunks = true;
4368                         failed = 0;
4369                         retried = false;
4370                         ret = btrfs_commit_transaction(trans, root);
4371                         if (ret)
4372                                 goto done;
4373                         goto again;
4374                 }
4375         }
4376
4377         btrfs_device_set_disk_total_bytes(device, new_size);
4378         if (list_empty(&device->resized_list))
4379                 list_add_tail(&device->resized_list,
4380                               &root->fs_info->fs_devices->resized_devices);
4381
4382         WARN_ON(diff > old_total);
4383         btrfs_set_super_total_bytes(super_copy, old_total - diff);
4384         unlock_chunks(root);
4385
4386         /* Now btrfs_update_device() will change the on-disk size. */
4387         ret = btrfs_update_device(trans, device);
4388         btrfs_end_transaction(trans, root);
4389 done:
4390         btrfs_free_path(path);
4391         if (ret) {
4392                 lock_chunks(root);
4393                 btrfs_device_set_total_bytes(device, old_size);
4394                 if (device->writeable)
4395                         device->fs_devices->total_rw_bytes += diff;
4396                 spin_lock(&root->fs_info->free_chunk_lock);
4397                 root->fs_info->free_chunk_space += diff;
4398                 spin_unlock(&root->fs_info->free_chunk_lock);
4399                 unlock_chunks(root);
4400         }
4401         return ret;
4402 }
4403
4404 static int btrfs_add_system_chunk(struct btrfs_root *root,
4405                            struct btrfs_key *key,
4406                            struct btrfs_chunk *chunk, int item_size)
4407 {
4408         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
4409         struct btrfs_disk_key disk_key;
4410         u32 array_size;
4411         u8 *ptr;
4412
4413         lock_chunks(root);
4414         array_size = btrfs_super_sys_array_size(super_copy);
4415         if (array_size + item_size + sizeof(disk_key)
4416                         > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
4417                 unlock_chunks(root);
4418                 return -EFBIG;
4419         }
4420
4421         ptr = super_copy->sys_chunk_array + array_size;
4422         btrfs_cpu_key_to_disk(&disk_key, key);
4423         memcpy(ptr, &disk_key, sizeof(disk_key));
4424         ptr += sizeof(disk_key);
4425         memcpy(ptr, chunk, item_size);
4426         item_size += sizeof(disk_key);
4427         btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
4428         unlock_chunks(root);
4429
4430         return 0;
4431 }
4432
4433 /*
4434  * sort the devices in descending order by max_avail, total_avail
4435  */
4436 static int btrfs_cmp_device_info(const void *a, const void *b)
4437 {
4438         const struct btrfs_device_info *di_a = a;
4439         const struct btrfs_device_info *di_b = b;
4440
4441         if (di_a->max_avail > di_b->max_avail)
4442                 return -1;
4443         if (di_a->max_avail < di_b->max_avail)
4444                 return 1;
4445         if (di_a->total_avail > di_b->total_avail)
4446                 return -1;
4447         if (di_a->total_avail < di_b->total_avail)
4448                 return 1;
4449         return 0;
4450 }
4451
4452 static u32 find_raid56_stripe_len(u32 data_devices, u32 dev_stripe_target)
4453 {
4454         /* TODO allow them to set a preferred stripe size */
4455         return SZ_64K;
4456 }
4457
4458 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
4459 {
4460         if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK))
4461                 return;
4462
4463         btrfs_set_fs_incompat(info, RAID56);
4464 }
4465
4466 #define BTRFS_MAX_DEVS(r) ((BTRFS_LEAF_DATA_SIZE(r)             \
4467                         - sizeof(struct btrfs_item)             \
4468                         - sizeof(struct btrfs_chunk))           \
4469                         / sizeof(struct btrfs_stripe) + 1)
4470
4471 #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE        \
4472                                 - 2 * sizeof(struct btrfs_disk_key)     \
4473                                 - 2 * sizeof(struct btrfs_chunk))       \
4474                                 / sizeof(struct btrfs_stripe) + 1)
4475
4476 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4477                                struct btrfs_root *extent_root, u64 start,
4478                                u64 type)
4479 {
4480         struct btrfs_fs_info *info = extent_root->fs_info;
4481         struct btrfs_fs_devices *fs_devices = info->fs_devices;
4482         struct list_head *cur;
4483         struct map_lookup *map = NULL;
4484         struct extent_map_tree *em_tree;
4485         struct extent_map *em;
4486         struct btrfs_device_info *devices_info = NULL;
4487         u64 total_avail;
4488         int num_stripes;        /* total number of stripes to allocate */
4489         int data_stripes;       /* number of stripes that count for
4490                                    block group size */
4491         int sub_stripes;        /* sub_stripes info for map */
4492         int dev_stripes;        /* stripes per dev */
4493         int devs_max;           /* max devs to use */
4494         int devs_min;           /* min devs needed */
4495         int devs_increment;     /* ndevs has to be a multiple of this */
4496         int ncopies;            /* how many copies to data has */
4497         int ret;
4498         u64 max_stripe_size;
4499         u64 max_chunk_size;
4500         u64 stripe_size;
4501         u64 num_bytes;
4502         u64 raid_stripe_len = BTRFS_STRIPE_LEN;
4503         int ndevs;
4504         int i;
4505         int j;
4506         int index;
4507
4508         BUG_ON(!alloc_profile_is_valid(type, 0));
4509
4510         if (list_empty(&fs_devices->alloc_list))
4511                 return -ENOSPC;
4512
4513         index = __get_raid_index(type);
4514
4515         sub_stripes = btrfs_raid_array[index].sub_stripes;
4516         dev_stripes = btrfs_raid_array[index].dev_stripes;
4517         devs_max = btrfs_raid_array[index].devs_max;
4518         devs_min = btrfs_raid_array[index].devs_min;
4519         devs_increment = btrfs_raid_array[index].devs_increment;
4520         ncopies = btrfs_raid_array[index].ncopies;
4521
4522         if (type & BTRFS_BLOCK_GROUP_DATA) {
4523                 max_stripe_size = SZ_1G;
4524                 max_chunk_size = 10 * max_stripe_size;
4525                 if (!devs_max)
4526                         devs_max = BTRFS_MAX_DEVS(info->chunk_root);
4527         } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
4528                 /* for larger filesystems, use larger metadata chunks */
4529                 if (fs_devices->total_rw_bytes > 50ULL * SZ_1G)
4530                         max_stripe_size = SZ_1G;
4531                 else
4532                         max_stripe_size = SZ_256M;
4533                 max_chunk_size = max_stripe_size;
4534                 if (!devs_max)
4535                         devs_max = BTRFS_MAX_DEVS(info->chunk_root);
4536         } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
4537                 max_stripe_size = SZ_32M;
4538                 max_chunk_size = 2 * max_stripe_size;
4539                 if (!devs_max)
4540                         devs_max = BTRFS_MAX_DEVS_SYS_CHUNK;
4541         } else {
4542                 btrfs_err(info, "invalid chunk type 0x%llx requested",
4543                        type);
4544                 BUG_ON(1);
4545         }
4546
4547         /* we don't want a chunk larger than 10% of writeable space */
4548         max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
4549                              max_chunk_size);
4550
4551         devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info),
4552                                GFP_NOFS);
4553         if (!devices_info)
4554                 return -ENOMEM;
4555
4556         cur = fs_devices->alloc_list.next;
4557
4558         /*
4559          * in the first pass through the devices list, we gather information
4560          * about the available holes on each device.
4561          */
4562         ndevs = 0;
4563         while (cur != &fs_devices->alloc_list) {
4564                 struct btrfs_device *device;
4565                 u64 max_avail;
4566                 u64 dev_offset;
4567
4568                 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
4569
4570                 cur = cur->next;
4571
4572                 if (!device->writeable) {
4573                         WARN(1, KERN_ERR
4574                                "BTRFS: read-only device in alloc_list\n");
4575                         continue;
4576                 }
4577
4578                 if (!device->in_fs_metadata ||
4579                     device->is_tgtdev_for_dev_replace)
4580                         continue;
4581
4582                 if (device->total_bytes > device->bytes_used)
4583                         total_avail = device->total_bytes - device->bytes_used;
4584                 else
4585                         total_avail = 0;
4586
4587                 /* If there is no space on this device, skip it. */
4588                 if (total_avail == 0)
4589                         continue;
4590
4591                 ret = find_free_dev_extent(trans, device,
4592                                            max_stripe_size * dev_stripes,
4593                                            &dev_offset, &max_avail);
4594                 if (ret && ret != -ENOSPC)
4595                         goto error;
4596
4597                 if (ret == 0)
4598                         max_avail = max_stripe_size * dev_stripes;
4599
4600                 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes)
4601                         continue;
4602
4603                 if (ndevs == fs_devices->rw_devices) {
4604                         WARN(1, "%s: found more than %llu devices\n",
4605                              __func__, fs_devices->rw_devices);
4606                         break;
4607                 }
4608                 devices_info[ndevs].dev_offset = dev_offset;
4609                 devices_info[ndevs].max_avail = max_avail;
4610                 devices_info[ndevs].total_avail = total_avail;
4611                 devices_info[ndevs].dev = device;
4612                 ++ndevs;
4613         }
4614
4615         /*
4616          * now sort the devices by hole size / available space
4617          */
4618         sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
4619              btrfs_cmp_device_info, NULL);
4620
4621         /* round down to number of usable stripes */
4622         ndevs -= ndevs % devs_increment;
4623
4624         if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) {
4625                 ret = -ENOSPC;
4626                 goto error;
4627         }
4628
4629         if (devs_max && ndevs > devs_max)
4630                 ndevs = devs_max;
4631         /*
4632          * the primary goal is to maximize the number of stripes, so use as many
4633          * devices as possible, even if the stripes are not maximum sized.
4634          */
4635         stripe_size = devices_info[ndevs-1].max_avail;
4636         num_stripes = ndevs * dev_stripes;
4637
4638         /*
4639          * this will have to be fixed for RAID1 and RAID10 over
4640          * more drives
4641          */
4642         data_stripes = num_stripes / ncopies;
4643
4644         if (type & BTRFS_BLOCK_GROUP_RAID5) {
4645                 raid_stripe_len = find_raid56_stripe_len(ndevs - 1,
4646                                  btrfs_super_stripesize(info->super_copy));
4647                 data_stripes = num_stripes - 1;
4648         }
4649         if (type & BTRFS_BLOCK_GROUP_RAID6) {
4650                 raid_stripe_len = find_raid56_stripe_len(ndevs - 2,
4651                                  btrfs_super_stripesize(info->super_copy));
4652                 data_stripes = num_stripes - 2;
4653         }
4654
4655         /*
4656          * Use the number of data stripes to figure out how big this chunk
4657          * is really going to be in terms of logical address space,
4658          * and compare that answer with the max chunk size
4659          */
4660         if (stripe_size * data_stripes > max_chunk_size) {
4661                 u64 mask = (1ULL << 24) - 1;
4662
4663                 stripe_size = div_u64(max_chunk_size, data_stripes);
4664
4665                 /* bump the answer up to a 16MB boundary */
4666                 stripe_size = (stripe_size + mask) & ~mask;
4667
4668                 /* but don't go higher than the limits we found
4669                  * while searching for free extents
4670                  */
4671                 if (stripe_size > devices_info[ndevs-1].max_avail)
4672                         stripe_size = devices_info[ndevs-1].max_avail;
4673         }
4674
4675         stripe_size = div_u64(stripe_size, dev_stripes);
4676
4677         /* align to BTRFS_STRIPE_LEN */
4678         stripe_size = div_u64(stripe_size, raid_stripe_len);
4679         stripe_size *= raid_stripe_len;
4680
4681         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
4682         if (!map) {
4683                 ret = -ENOMEM;
4684                 goto error;
4685         }
4686         map->num_stripes = num_stripes;
4687
4688         for (i = 0; i < ndevs; ++i) {
4689                 for (j = 0; j < dev_stripes; ++j) {
4690                         int s = i * dev_stripes + j;
4691                         map->stripes[s].dev = devices_info[i].dev;
4692                         map->stripes[s].physical = devices_info[i].dev_offset +
4693                                                    j * stripe_size;
4694                 }
4695         }
4696         map->sector_size = extent_root->sectorsize;
4697         map->stripe_len = raid_stripe_len;
4698         map->io_align = raid_stripe_len;
4699         map->io_width = raid_stripe_len;
4700         map->type = type;
4701         map->sub_stripes = sub_stripes;
4702
4703         num_bytes = stripe_size * data_stripes;
4704
4705         trace_btrfs_chunk_alloc(info->chunk_root, map, start, num_bytes);
4706
4707         em = alloc_extent_map();
4708         if (!em) {
4709                 kfree(map);
4710                 ret = -ENOMEM;
4711                 goto error;
4712         }
4713         set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
4714         em->map_lookup = map;
4715         em->start = start;
4716         em->len = num_bytes;
4717         em->block_start = 0;
4718         em->block_len = em->len;
4719         em->orig_block_len = stripe_size;
4720
4721         em_tree = &extent_root->fs_info->mapping_tree.map_tree;
4722         write_lock(&em_tree->lock);
4723         ret = add_extent_mapping(em_tree, em, 0);
4724         if (!ret) {
4725                 list_add_tail(&em->list, &trans->transaction->pending_chunks);
4726                 atomic_inc(&em->refs);
4727         }
4728         write_unlock(&em_tree->lock);
4729         if (ret) {
4730                 free_extent_map(em);
4731                 goto error;
4732         }
4733
4734         ret = btrfs_make_block_group(trans, extent_root, 0, type,
4735                                      BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4736                                      start, num_bytes);
4737         if (ret)
4738                 goto error_del_extent;
4739
4740         for (i = 0; i < map->num_stripes; i++) {
4741                 num_bytes = map->stripes[i].dev->bytes_used + stripe_size;
4742                 btrfs_device_set_bytes_used(map->stripes[i].dev, num_bytes);
4743         }
4744
4745         spin_lock(&extent_root->fs_info->free_chunk_lock);
4746         extent_root->fs_info->free_chunk_space -= (stripe_size *
4747                                                    map->num_stripes);
4748         spin_unlock(&extent_root->fs_info->free_chunk_lock);
4749
4750         free_extent_map(em);
4751         check_raid56_incompat_flag(extent_root->fs_info, type);
4752
4753         kfree(devices_info);
4754         return 0;
4755
4756 error_del_extent:
4757         write_lock(&em_tree->lock);
4758         remove_extent_mapping(em_tree, em);
4759         write_unlock(&em_tree->lock);
4760
4761         /* One for our allocation */
4762         free_extent_map(em);
4763         /* One for the tree reference */
4764         free_extent_map(em);
4765         /* One for the pending_chunks list reference */
4766         free_extent_map(em);
4767 error:
4768         kfree(devices_info);
4769         return ret;
4770 }
4771
4772 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans,
4773                                 struct btrfs_root *extent_root,
4774                                 u64 chunk_offset, u64 chunk_size)
4775 {
4776         struct btrfs_key key;
4777         struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
4778         struct btrfs_device *device;
4779         struct btrfs_chunk *chunk;
4780         struct btrfs_stripe *stripe;
4781         struct extent_map_tree *em_tree;
4782         struct extent_map *em;
4783         struct map_lookup *map;
4784         size_t item_size;
4785         u64 dev_offset;
4786         u64 stripe_size;
4787         int i = 0;
4788         int ret = 0;
4789
4790         em_tree = &extent_root->fs_info->mapping_tree.map_tree;
4791         read_lock(&em_tree->lock);
4792         em = lookup_extent_mapping(em_tree, chunk_offset, chunk_size);
4793         read_unlock(&em_tree->lock);
4794
4795         if (!em) {
4796                 btrfs_crit(extent_root->fs_info, "unable to find logical "
4797                            "%Lu len %Lu", chunk_offset, chunk_size);
4798                 return -EINVAL;
4799         }
4800
4801         if (em->start != chunk_offset || em->len != chunk_size) {
4802                 btrfs_crit(extent_root->fs_info, "found a bad mapping, wanted"
4803                           " %Lu-%Lu, found %Lu-%Lu", chunk_offset,
4804                           chunk_size, em->start, em->len);
4805                 free_extent_map(em);
4806                 return -EINVAL;
4807         }
4808
4809         map = em->map_lookup;
4810         item_size = btrfs_chunk_item_size(map->num_stripes);
4811         stripe_size = em->orig_block_len;
4812
4813         chunk = kzalloc(item_size, GFP_NOFS);
4814         if (!chunk) {
4815                 ret = -ENOMEM;
4816                 goto out;
4817         }
4818
4819         /*
4820          * Take the device list mutex to prevent races with the final phase of
4821          * a device replace operation that replaces the device object associated
4822          * with the map's stripes, because the device object's id can change
4823          * at any time during that final phase of the device replace operation
4824          * (dev-replace.c:btrfs_dev_replace_finishing()).
4825          */
4826         mutex_lock(&chunk_root->fs_info->fs_devices->device_list_mutex);
4827         for (i = 0; i < map->num_stripes; i++) {
4828                 device = map->stripes[i].dev;
4829                 dev_offset = map->stripes[i].physical;
4830
4831                 ret = btrfs_update_device(trans, device);
4832                 if (ret)
4833                         break;
4834                 ret = btrfs_alloc_dev_extent(trans, device,
4835                                              chunk_root->root_key.objectid,
4836                                              BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4837                                              chunk_offset, dev_offset,
4838                                              stripe_size);
4839                 if (ret)
4840                         break;
4841         }
4842         if (ret) {
4843                 mutex_unlock(&chunk_root->fs_info->fs_devices->device_list_mutex);
4844                 goto out;
4845         }
4846
4847         stripe = &chunk->stripe;
4848         for (i = 0; i < map->num_stripes; i++) {
4849                 device = map->stripes[i].dev;
4850                 dev_offset = map->stripes[i].physical;
4851
4852                 btrfs_set_stack_stripe_devid(stripe, device->devid);
4853                 btrfs_set_stack_stripe_offset(stripe, dev_offset);
4854                 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
4855                 stripe++;
4856         }
4857         mutex_unlock(&chunk_root->fs_info->fs_devices->device_list_mutex);
4858
4859         btrfs_set_stack_chunk_length(chunk, chunk_size);
4860         btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
4861         btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
4862         btrfs_set_stack_chunk_type(chunk, map->type);
4863         btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
4864         btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
4865         btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
4866         btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
4867         btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
4868
4869         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
4870         key.type = BTRFS_CHUNK_ITEM_KEY;
4871         key.offset = chunk_offset;
4872
4873         ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
4874         if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
4875                 /*
4876                  * TODO: Cleanup of inserted chunk root in case of
4877                  * failure.
4878                  */
4879                 ret = btrfs_add_system_chunk(chunk_root, &key, chunk,
4880                                              item_size);
4881         }
4882
4883 out:
4884         kfree(chunk);
4885         free_extent_map(em);
4886         return ret;
4887 }
4888
4889 /*
4890  * Chunk allocation falls into two parts. The first part does works
4891  * that make the new allocated chunk useable, but not do any operation
4892  * that modifies the chunk tree. The second part does the works that
4893  * require modifying the chunk tree. This division is important for the
4894  * bootstrap process of adding storage to a seed btrfs.
4895  */
4896 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4897                       struct btrfs_root *extent_root, u64 type)
4898 {
4899         u64 chunk_offset;
4900
4901         ASSERT(mutex_is_locked(&extent_root->fs_info->chunk_mutex));
4902         chunk_offset = find_next_chunk(extent_root->fs_info);
4903         return __btrfs_alloc_chunk(trans, extent_root, chunk_offset, type);
4904 }
4905
4906 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
4907                                          struct btrfs_root *root,
4908                                          struct btrfs_device *device)
4909 {
4910         u64 chunk_offset;
4911         u64 sys_chunk_offset;
4912         u64 alloc_profile;
4913         struct btrfs_fs_info *fs_info = root->fs_info;
4914         struct btrfs_root *extent_root = fs_info->extent_root;
4915         int ret;
4916
4917         chunk_offset = find_next_chunk(fs_info);
4918         alloc_profile = btrfs_get_alloc_profile(extent_root, 0);
4919         ret = __btrfs_alloc_chunk(trans, extent_root, chunk_offset,
4920                                   alloc_profile);
4921         if (ret)
4922                 return ret;
4923
4924         sys_chunk_offset = find_next_chunk(root->fs_info);
4925         alloc_profile = btrfs_get_alloc_profile(fs_info->chunk_root, 0);
4926         ret = __btrfs_alloc_chunk(trans, extent_root, sys_chunk_offset,
4927                                   alloc_profile);
4928         return ret;
4929 }
4930
4931 static inline int btrfs_chunk_max_errors(struct map_lookup *map)
4932 {
4933         int max_errors;
4934
4935         if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
4936                          BTRFS_BLOCK_GROUP_RAID10 |
4937                          BTRFS_BLOCK_GROUP_RAID5 |
4938                          BTRFS_BLOCK_GROUP_DUP)) {
4939                 max_errors = 1;
4940         } else if (map->type & BTRFS_BLOCK_GROUP_RAID6) {
4941                 max_errors = 2;
4942         } else {
4943                 max_errors = 0;
4944         }
4945
4946         return max_errors;
4947 }
4948
4949 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
4950 {
4951         struct extent_map *em;
4952         struct map_lookup *map;
4953         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
4954         int readonly = 0;
4955         int miss_ndevs = 0;
4956         int i;
4957
4958         read_lock(&map_tree->map_tree.lock);
4959         em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
4960         read_unlock(&map_tree->map_tree.lock);
4961         if (!em)
4962                 return 1;
4963
4964         map = em->map_lookup;
4965         for (i = 0; i < map->num_stripes; i++) {
4966                 if (map->stripes[i].dev->missing) {
4967                         miss_ndevs++;
4968                         continue;
4969                 }
4970
4971                 if (!map->stripes[i].dev->writeable) {
4972                         readonly = 1;
4973                         goto end;
4974                 }
4975         }
4976
4977         /*
4978          * If the number of missing devices is larger than max errors,
4979          * we can not write the data into that chunk successfully, so
4980          * set it readonly.
4981          */
4982         if (miss_ndevs > btrfs_chunk_max_errors(map))
4983                 readonly = 1;
4984 end:
4985         free_extent_map(em);
4986         return readonly;
4987 }
4988
4989 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
4990 {
4991         extent_map_tree_init(&tree->map_tree);
4992 }
4993
4994 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
4995 {
4996         struct extent_map *em;
4997
4998         while (1) {
4999                 write_lock(&tree->map_tree.lock);
5000                 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
5001                 if (em)
5002                         remove_extent_mapping(&tree->map_tree, em);
5003                 write_unlock(&tree->map_tree.lock);
5004                 if (!em)
5005                         break;
5006                 /* once for us */
5007                 free_extent_map(em);
5008                 /* once for the tree */
5009                 free_extent_map(em);
5010         }
5011 }
5012
5013 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5014 {
5015         struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
5016         struct extent_map *em;
5017         struct map_lookup *map;
5018         struct extent_map_tree *em_tree = &map_tree->map_tree;
5019         int ret;
5020
5021         read_lock(&em_tree->lock);
5022         em = lookup_extent_mapping(em_tree, logical, len);
5023         read_unlock(&em_tree->lock);
5024
5025         /*
5026          * We could return errors for these cases, but that could get ugly and
5027          * we'd probably do the same thing which is just not do anything else
5028          * and exit, so return 1 so the callers don't try to use other copies.
5029          */
5030         if (!em) {
5031                 btrfs_crit(fs_info, "No mapping for %Lu-%Lu", logical,
5032                             logical+len);
5033                 return 1;
5034         }
5035
5036         if (em->start > logical || em->start + em->len < logical) {
5037                 btrfs_crit(fs_info, "Invalid mapping for %Lu-%Lu, got "
5038                             "%Lu-%Lu", logical, logical+len, em->start,
5039                             em->start + em->len);
5040                 free_extent_map(em);
5041                 return 1;
5042         }
5043
5044         map = em->map_lookup;
5045         if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
5046                 ret = map->num_stripes;
5047         else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5048                 ret = map->sub_stripes;
5049         else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
5050                 ret = 2;
5051         else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5052                 ret = 3;
5053         else
5054                 ret = 1;
5055         free_extent_map(em);
5056
5057         btrfs_dev_replace_lock(&fs_info->dev_replace, 0);
5058         if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace))
5059                 ret++;
5060         btrfs_dev_replace_unlock(&fs_info->dev_replace, 0);
5061
5062         return ret;
5063 }
5064
5065 unsigned long btrfs_full_stripe_len(struct btrfs_root *root,
5066                                     struct btrfs_mapping_tree *map_tree,
5067                                     u64 logical)
5068 {
5069         struct extent_map *em;
5070         struct map_lookup *map;
5071         struct extent_map_tree *em_tree = &map_tree->map_tree;
5072         unsigned long len = root->sectorsize;
5073
5074         read_lock(&em_tree->lock);
5075         em = lookup_extent_mapping(em_tree, logical, len);
5076         read_unlock(&em_tree->lock);
5077         BUG_ON(!em);
5078
5079         BUG_ON(em->start > logical || em->start + em->len < logical);
5080         map = em->map_lookup;
5081         if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5082                 len = map->stripe_len * nr_data_stripes(map);
5083         free_extent_map(em);
5084         return len;
5085 }
5086
5087 int btrfs_is_parity_mirror(struct btrfs_mapping_tree *map_tree,
5088                            u64 logical, u64 len, int mirror_num)
5089 {
5090         struct extent_map *em;
5091         struct map_lookup *map;
5092         struct extent_map_tree *em_tree = &map_tree->map_tree;
5093         int ret = 0;
5094
5095         read_lock(&em_tree->lock);
5096         em = lookup_extent_mapping(em_tree, logical, len);
5097         read_unlock(&em_tree->lock);
5098         BUG_ON(!em);
5099
5100         BUG_ON(em->start > logical || em->start + em->len < logical);
5101         map = em->map_lookup;
5102         if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5103                 ret = 1;
5104         free_extent_map(em);
5105         return ret;
5106 }
5107
5108 static int find_live_mirror(struct btrfs_fs_info *fs_info,
5109                             struct map_lookup *map, int first, int num,
5110                             int optimal, int dev_replace_is_ongoing)
5111 {
5112         int i;
5113         int tolerance;
5114         struct btrfs_device *srcdev;
5115
5116         if (dev_replace_is_ongoing &&
5117             fs_info->dev_replace.cont_reading_from_srcdev_mode ==
5118              BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
5119                 srcdev = fs_info->dev_replace.srcdev;
5120         else
5121                 srcdev = NULL;
5122
5123         /*
5124          * try to avoid the drive that is the source drive for a
5125          * dev-replace procedure, only choose it if no other non-missing
5126          * mirror is available
5127          */
5128         for (tolerance = 0; tolerance < 2; tolerance++) {
5129                 if (map->stripes[optimal].dev->bdev &&
5130                     (tolerance || map->stripes[optimal].dev != srcdev))
5131                         return optimal;
5132                 for (i = first; i < first + num; i++) {
5133                         if (map->stripes[i].dev->bdev &&
5134                             (tolerance || map->stripes[i].dev != srcdev))
5135                                 return i;
5136                 }
5137         }
5138
5139         /* we couldn't find one that doesn't fail.  Just return something
5140          * and the io error handling code will clean up eventually
5141          */
5142         return optimal;
5143 }
5144
5145 static inline int parity_smaller(u64 a, u64 b)
5146 {
5147         return a > b;
5148 }
5149
5150 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
5151 static void sort_parity_stripes(struct btrfs_bio *bbio, int num_stripes)
5152 {
5153         struct btrfs_bio_stripe s;
5154         int i;
5155         u64 l;
5156         int again = 1;
5157
5158         while (again) {
5159                 again = 0;
5160                 for (i = 0; i < num_stripes - 1; i++) {
5161                         if (parity_smaller(bbio->raid_map[i],
5162                                            bbio->raid_map[i+1])) {
5163                                 s = bbio->stripes[i];
5164                                 l = bbio->raid_map[i];
5165                                 bbio->stripes[i] = bbio->stripes[i+1];
5166                                 bbio->raid_map[i] = bbio->raid_map[i+1];
5167                                 bbio->stripes[i+1] = s;
5168                                 bbio->raid_map[i+1] = l;
5169
5170                                 again = 1;
5171                         }
5172                 }
5173         }
5174 }
5175
5176 static struct btrfs_bio *alloc_btrfs_bio(int total_stripes, int real_stripes)
5177 {
5178         struct btrfs_bio *bbio = kzalloc(
5179                  /* the size of the btrfs_bio */
5180                 sizeof(struct btrfs_bio) +
5181                 /* plus the variable array for the stripes */
5182                 sizeof(struct btrfs_bio_stripe) * (total_stripes) +
5183                 /* plus the variable array for the tgt dev */
5184                 sizeof(int) * (real_stripes) +
5185                 /*
5186                  * plus the raid_map, which includes both the tgt dev
5187                  * and the stripes
5188                  */
5189                 sizeof(u64) * (total_stripes),
5190                 GFP_NOFS|__GFP_NOFAIL);
5191
5192         atomic_set(&bbio->error, 0);
5193         atomic_set(&bbio->refs, 1);
5194
5195         return bbio;
5196 }
5197
5198 void btrfs_get_bbio(struct btrfs_bio *bbio)
5199 {
5200         WARN_ON(!atomic_read(&bbio->refs));
5201         atomic_inc(&bbio->refs);
5202 }
5203
5204 void btrfs_put_bbio(struct btrfs_bio *bbio)
5205 {
5206         if (!bbio)
5207                 return;
5208         if (atomic_dec_and_test(&bbio->refs))
5209                 kfree(bbio);
5210 }
5211
5212 static int __btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
5213                              u64 logical, u64 *length,
5214                              struct btrfs_bio **bbio_ret,
5215                              int mirror_num, int need_raid_map)
5216 {
5217         struct extent_map *em;
5218         struct map_lookup *map;
5219         struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
5220         struct extent_map_tree *em_tree = &map_tree->map_tree;
5221         u64 offset;
5222         u64 stripe_offset;
5223         u64 stripe_end_offset;
5224         u64 stripe_nr;
5225         u64 stripe_nr_orig;
5226         u64 stripe_nr_end;
5227         u64 stripe_len;
5228         u32 stripe_index;
5229         int i;
5230         int ret = 0;
5231         int num_stripes;
5232         int max_errors = 0;
5233         int tgtdev_indexes = 0;
5234         struct btrfs_bio *bbio = NULL;
5235         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
5236         int dev_replace_is_ongoing = 0;
5237         int num_alloc_stripes;
5238         int patch_the_first_stripe_for_dev_replace = 0;
5239         u64 physical_to_patch_in_first_stripe = 0;
5240         u64 raid56_full_stripe_start = (u64)-1;
5241
5242         read_lock(&em_tree->lock);
5243         em = lookup_extent_mapping(em_tree, logical, *length);
5244         read_unlock(&em_tree->lock);
5245
5246         if (!em) {
5247                 btrfs_crit(fs_info, "unable to find logical %llu len %llu",
5248                         logical, *length);
5249                 return -EINVAL;
5250         }
5251
5252         if (em->start > logical || em->start + em->len < logical) {
5253                 btrfs_crit(fs_info, "found a bad mapping, wanted %Lu, "
5254                            "found %Lu-%Lu", logical, em->start,
5255                            em->start + em->len);
5256                 free_extent_map(em);
5257                 return -EINVAL;
5258         }
5259
5260         map = em->map_lookup;
5261         offset = logical - em->start;
5262
5263         stripe_len = map->stripe_len;
5264         stripe_nr = offset;
5265         /*
5266          * stripe_nr counts the total number of stripes we have to stride
5267          * to get to this block
5268          */
5269         stripe_nr = div64_u64(stripe_nr, stripe_len);
5270
5271         stripe_offset = stripe_nr * stripe_len;
5272         BUG_ON(offset < stripe_offset);
5273
5274         /* stripe_offset is the offset of this block in its stripe*/
5275         stripe_offset = offset - stripe_offset;
5276
5277         /* if we're here for raid56, we need to know the stripe aligned start */
5278         if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5279                 unsigned long full_stripe_len = stripe_len * nr_data_stripes(map);
5280                 raid56_full_stripe_start = offset;
5281
5282                 /* allow a write of a full stripe, but make sure we don't
5283                  * allow straddling of stripes
5284                  */
5285                 raid56_full_stripe_start = div64_u64(raid56_full_stripe_start,
5286                                 full_stripe_len);
5287                 raid56_full_stripe_start *= full_stripe_len;
5288         }
5289
5290         if (rw & REQ_DISCARD) {
5291                 /* we don't discard raid56 yet */
5292                 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5293                         ret = -EOPNOTSUPP;
5294                         goto out;
5295                 }
5296                 *length = min_t(u64, em->len - offset, *length);
5297         } else if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
5298                 u64 max_len;
5299                 /* For writes to RAID[56], allow a full stripeset across all disks.
5300                    For other RAID types and for RAID[56] reads, just allow a single
5301                    stripe (on a single disk). */
5302                 if ((map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
5303                     (rw & REQ_WRITE)) {
5304                         max_len = stripe_len * nr_data_stripes(map) -
5305                                 (offset - raid56_full_stripe_start);
5306                 } else {
5307                         /* we limit the length of each bio to what fits in a stripe */
5308                         max_len = stripe_len - stripe_offset;
5309                 }
5310                 *length = min_t(u64, em->len - offset, max_len);
5311         } else {
5312                 *length = em->len - offset;
5313         }
5314
5315         /* This is for when we're called from btrfs_merge_bio_hook() and all
5316            it cares about is the length */
5317         if (!bbio_ret)
5318                 goto out;
5319
5320         btrfs_dev_replace_lock(dev_replace, 0);
5321         dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
5322         if (!dev_replace_is_ongoing)
5323                 btrfs_dev_replace_unlock(dev_replace, 0);
5324         else
5325                 btrfs_dev_replace_set_lock_blocking(dev_replace);
5326
5327         if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
5328             !(rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) &&
5329             dev_replace->tgtdev != NULL) {
5330                 /*
5331                  * in dev-replace case, for repair case (that's the only
5332                  * case where the mirror is selected explicitly when
5333                  * calling btrfs_map_block), blocks left of the left cursor
5334                  * can also be read from the target drive.
5335                  * For REQ_GET_READ_MIRRORS, the target drive is added as
5336                  * the last one to the array of stripes. For READ, it also
5337                  * needs to be supported using the same mirror number.
5338                  * If the requested block is not left of the left cursor,
5339                  * EIO is returned. This can happen because btrfs_num_copies()
5340                  * returns one more in the dev-replace case.
5341                  */
5342                 u64 tmp_length = *length;
5343                 struct btrfs_bio *tmp_bbio = NULL;
5344                 int tmp_num_stripes;
5345                 u64 srcdev_devid = dev_replace->srcdev->devid;
5346                 int index_srcdev = 0;
5347                 int found = 0;
5348                 u64 physical_of_found = 0;
5349
5350                 ret = __btrfs_map_block(fs_info, REQ_GET_READ_MIRRORS,
5351                              logical, &tmp_length, &tmp_bbio, 0, 0);
5352                 if (ret) {
5353                         WARN_ON(tmp_bbio != NULL);
5354                         goto out;
5355                 }
5356
5357                 tmp_num_stripes = tmp_bbio->num_stripes;
5358                 if (mirror_num > tmp_num_stripes) {
5359                         /*
5360                          * REQ_GET_READ_MIRRORS does not contain this
5361                          * mirror, that means that the requested area
5362                          * is not left of the left cursor
5363                          */
5364                         ret = -EIO;
5365                         btrfs_put_bbio(tmp_bbio);
5366                         goto out;
5367                 }
5368
5369                 /*
5370                  * process the rest of the function using the mirror_num
5371                  * of the source drive. Therefore look it up first.
5372                  * At the end, patch the device pointer to the one of the
5373                  * target drive.
5374                  */
5375                 for (i = 0; i < tmp_num_stripes; i++) {
5376                         if (tmp_bbio->stripes[i].dev->devid != srcdev_devid)
5377                                 continue;
5378
5379                         /*
5380                          * In case of DUP, in order to keep it simple, only add
5381                          * the mirror with the lowest physical address
5382                          */
5383                         if (found &&
5384                             physical_of_found <= tmp_bbio->stripes[i].physical)
5385                                 continue;
5386
5387                         index_srcdev = i;
5388                         found = 1;
5389                         physical_of_found = tmp_bbio->stripes[i].physical;
5390                 }
5391
5392                 btrfs_put_bbio(tmp_bbio);
5393
5394                 if (!found) {
5395                         WARN_ON(1);
5396                         ret = -EIO;
5397                         goto out;
5398                 }
5399
5400                 mirror_num = index_srcdev + 1;
5401                 patch_the_first_stripe_for_dev_replace = 1;
5402                 physical_to_patch_in_first_stripe = physical_of_found;
5403         } else if (mirror_num > map->num_stripes) {
5404                 mirror_num = 0;
5405         }
5406
5407         num_stripes = 1;
5408         stripe_index = 0;
5409         stripe_nr_orig = stripe_nr;
5410         stripe_nr_end = ALIGN(offset + *length, map->stripe_len);
5411         stripe_nr_end = div_u64(stripe_nr_end, map->stripe_len);
5412         stripe_end_offset = stripe_nr_end * map->stripe_len -
5413                             (offset + *length);
5414
5415         if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5416                 if (rw & REQ_DISCARD)
5417                         num_stripes = min_t(u64, map->num_stripes,
5418                                             stripe_nr_end - stripe_nr_orig);
5419                 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5420                                 &stripe_index);
5421                 if (!(rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)))
5422                         mirror_num = 1;
5423         } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
5424                 if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS))
5425                         num_stripes = map->num_stripes;
5426                 else if (mirror_num)
5427                         stripe_index = mirror_num - 1;
5428                 else {
5429                         stripe_index = find_live_mirror(fs_info, map, 0,
5430                                             map->num_stripes,
5431                                             current->pid % map->num_stripes,
5432                                             dev_replace_is_ongoing);
5433                         mirror_num = stripe_index + 1;
5434                 }
5435
5436         } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
5437                 if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) {
5438                         num_stripes = map->num_stripes;
5439                 } else if (mirror_num) {
5440                         stripe_index = mirror_num - 1;
5441                 } else {
5442                         mirror_num = 1;
5443                 }
5444
5445         } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5446                 u32 factor = map->num_stripes / map->sub_stripes;
5447
5448                 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
5449                 stripe_index *= map->sub_stripes;
5450
5451                 if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS))
5452                         num_stripes = map->sub_stripes;
5453                 else if (rw & REQ_DISCARD)
5454                         num_stripes = min_t(u64, map->sub_stripes *
5455                                             (stripe_nr_end - stripe_nr_orig),
5456                                             map->num_stripes);
5457                 else if (mirror_num)
5458                         stripe_index += mirror_num - 1;
5459                 else {
5460                         int old_stripe_index = stripe_index;
5461                         stripe_index = find_live_mirror(fs_info, map,
5462                                               stripe_index,
5463                                               map->sub_stripes, stripe_index +
5464                                               current->pid % map->sub_stripes,
5465                                               dev_replace_is_ongoing);
5466                         mirror_num = stripe_index - old_stripe_index + 1;
5467                 }
5468
5469         } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5470                 if (need_raid_map &&
5471                     ((rw & (REQ_WRITE | REQ_GET_READ_MIRRORS)) ||
5472                      mirror_num > 1)) {
5473                         /* push stripe_nr back to the start of the full stripe */
5474                         stripe_nr = div_u64(raid56_full_stripe_start,
5475                                         stripe_len * nr_data_stripes(map));
5476
5477                         /* RAID[56] write or recovery. Return all stripes */
5478                         num_stripes = map->num_stripes;
5479                         max_errors = nr_parity_stripes(map);
5480
5481                         *length = map->stripe_len;
5482                         stripe_index = 0;
5483                         stripe_offset = 0;
5484                 } else {
5485                         /*
5486                          * Mirror #0 or #1 means the original data block.
5487                          * Mirror #2 is RAID5 parity block.
5488                          * Mirror #3 is RAID6 Q block.
5489                          */
5490                         stripe_nr = div_u64_rem(stripe_nr,
5491                                         nr_data_stripes(map), &stripe_index);
5492                         if (mirror_num > 1)
5493                                 stripe_index = nr_data_stripes(map) +
5494                                                 mirror_num - 2;
5495
5496                         /* We distribute the parity blocks across stripes */
5497                         div_u64_rem(stripe_nr + stripe_index, map->num_stripes,
5498                                         &stripe_index);
5499                         if (!(rw & (REQ_WRITE | REQ_DISCARD |
5500                                     REQ_GET_READ_MIRRORS)) && mirror_num <= 1)
5501                                 mirror_num = 1;
5502                 }
5503         } else {
5504                 /*
5505                  * after this, stripe_nr is the number of stripes on this
5506                  * device we have to walk to find the data, and stripe_index is
5507                  * the number of our device in the stripe array
5508                  */
5509                 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5510                                 &stripe_index);
5511                 mirror_num = stripe_index + 1;
5512         }
5513         BUG_ON(stripe_index >= map->num_stripes);
5514
5515         num_alloc_stripes = num_stripes;
5516         if (dev_replace_is_ongoing) {
5517                 if (rw & (REQ_WRITE | REQ_DISCARD))
5518                         num_alloc_stripes <<= 1;
5519                 if (rw & REQ_GET_READ_MIRRORS)
5520                         num_alloc_stripes++;
5521                 tgtdev_indexes = num_stripes;
5522         }
5523
5524         bbio = alloc_btrfs_bio(num_alloc_stripes, tgtdev_indexes);
5525         if (!bbio) {
5526                 ret = -ENOMEM;
5527                 goto out;
5528         }
5529         if (dev_replace_is_ongoing)
5530                 bbio->tgtdev_map = (int *)(bbio->stripes + num_alloc_stripes);
5531
5532         /* build raid_map */
5533         if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK &&
5534             need_raid_map && ((rw & (REQ_WRITE | REQ_GET_READ_MIRRORS)) ||
5535             mirror_num > 1)) {
5536                 u64 tmp;
5537                 unsigned rot;
5538
5539                 bbio->raid_map = (u64 *)((void *)bbio->stripes +
5540                                  sizeof(struct btrfs_bio_stripe) *
5541                                  num_alloc_stripes +
5542                                  sizeof(int) * tgtdev_indexes);
5543
5544                 /* Work out the disk rotation on this stripe-set */
5545                 div_u64_rem(stripe_nr, num_stripes, &rot);
5546
5547                 /* Fill in the logical address of each stripe */
5548                 tmp = stripe_nr * nr_data_stripes(map);
5549                 for (i = 0; i < nr_data_stripes(map); i++)
5550                         bbio->raid_map[(i+rot) % num_stripes] =
5551                                 em->start + (tmp + i) * map->stripe_len;
5552
5553                 bbio->raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE;
5554                 if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5555                         bbio->raid_map[(i+rot+1) % num_stripes] =
5556                                 RAID6_Q_STRIPE;
5557         }
5558
5559         if (rw & REQ_DISCARD) {
5560                 u32 factor = 0;
5561                 u32 sub_stripes = 0;
5562                 u64 stripes_per_dev = 0;
5563                 u32 remaining_stripes = 0;
5564                 u32 last_stripe = 0;
5565
5566                 if (map->type &
5567                     (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID10)) {
5568                         if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5569                                 sub_stripes = 1;
5570                         else
5571                                 sub_stripes = map->sub_stripes;
5572
5573                         factor = map->num_stripes / sub_stripes;
5574                         stripes_per_dev = div_u64_rem(stripe_nr_end -
5575                                                       stripe_nr_orig,
5576                                                       factor,
5577                                                       &remaining_stripes);
5578                         div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
5579                         last_stripe *= sub_stripes;
5580                 }
5581
5582                 for (i = 0; i < num_stripes; i++) {
5583                         bbio->stripes[i].physical =
5584                                 map->stripes[stripe_index].physical +
5585                                 stripe_offset + stripe_nr * map->stripe_len;
5586                         bbio->stripes[i].dev = map->stripes[stripe_index].dev;
5587
5588                         if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5589                                          BTRFS_BLOCK_GROUP_RAID10)) {
5590                                 bbio->stripes[i].length = stripes_per_dev *
5591                                                           map->stripe_len;
5592
5593                                 if (i / sub_stripes < remaining_stripes)
5594                                         bbio->stripes[i].length +=
5595                                                 map->stripe_len;
5596
5597                                 /*
5598                                  * Special for the first stripe and
5599                                  * the last stripe:
5600                                  *
5601                                  * |-------|...|-------|
5602                                  *     |----------|
5603                                  *    off     end_off
5604                                  */
5605                                 if (i < sub_stripes)
5606                                         bbio->stripes[i].length -=
5607                                                 stripe_offset;
5608
5609                                 if (stripe_index >= last_stripe &&
5610                                     stripe_index <= (last_stripe +
5611                                                      sub_stripes - 1))
5612                                         bbio->stripes[i].length -=
5613                                                 stripe_end_offset;
5614
5615                                 if (i == sub_stripes - 1)
5616                                         stripe_offset = 0;
5617                         } else
5618                                 bbio->stripes[i].length = *length;
5619
5620                         stripe_index++;
5621                         if (stripe_index == map->num_stripes) {
5622                                 /* This could only happen for RAID0/10 */
5623                                 stripe_index = 0;
5624                                 stripe_nr++;
5625                         }
5626                 }
5627         } else {
5628                 for (i = 0; i < num_stripes; i++) {
5629                         bbio->stripes[i].physical =
5630                                 map->stripes[stripe_index].physical +
5631                                 stripe_offset +
5632                                 stripe_nr * map->stripe_len;
5633                         bbio->stripes[i].dev =
5634                                 map->stripes[stripe_index].dev;
5635                         stripe_index++;
5636                 }
5637         }
5638
5639         if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS))
5640                 max_errors = btrfs_chunk_max_errors(map);
5641
5642         if (bbio->raid_map)
5643                 sort_parity_stripes(bbio, num_stripes);
5644
5645         tgtdev_indexes = 0;
5646         if (dev_replace_is_ongoing && (rw & (REQ_WRITE | REQ_DISCARD)) &&
5647             dev_replace->tgtdev != NULL) {
5648                 int index_where_to_add;
5649                 u64 srcdev_devid = dev_replace->srcdev->devid;
5650
5651                 /*
5652                  * duplicate the write operations while the dev replace
5653                  * procedure is running. Since the copying of the old disk
5654                  * to the new disk takes place at run time while the
5655                  * filesystem is mounted writable, the regular write
5656                  * operations to the old disk have to be duplicated to go
5657                  * to the new disk as well.
5658                  * Note that device->missing is handled by the caller, and
5659                  * that the write to the old disk is already set up in the
5660                  * stripes array.
5661                  */
5662                 index_where_to_add = num_stripes;
5663                 for (i = 0; i < num_stripes; i++) {
5664                         if (bbio->stripes[i].dev->devid == srcdev_devid) {
5665                                 /* write to new disk, too */
5666                                 struct btrfs_bio_stripe *new =
5667                                         bbio->stripes + index_where_to_add;
5668                                 struct btrfs_bio_stripe *old =
5669                                         bbio->stripes + i;
5670
5671                                 new->physical = old->physical;
5672                                 new->length = old->length;
5673                                 new->dev = dev_replace->tgtdev;
5674                                 bbio->tgtdev_map[i] = index_where_to_add;
5675                                 index_where_to_add++;
5676                                 max_errors++;
5677                                 tgtdev_indexes++;
5678                         }
5679                 }
5680                 num_stripes = index_where_to_add;
5681         } else if (dev_replace_is_ongoing && (rw & REQ_GET_READ_MIRRORS) &&
5682                    dev_replace->tgtdev != NULL) {
5683                 u64 srcdev_devid = dev_replace->srcdev->devid;
5684                 int index_srcdev = 0;
5685                 int found = 0;
5686                 u64 physical_of_found = 0;
5687
5688                 /*
5689                  * During the dev-replace procedure, the target drive can
5690                  * also be used to read data in case it is needed to repair
5691                  * a corrupt block elsewhere. This is possible if the
5692                  * requested area is left of the left cursor. In this area,
5693                  * the target drive is a full copy of the source drive.
5694                  */
5695                 for (i = 0; i < num_stripes; i++) {
5696                         if (bbio->stripes[i].dev->devid == srcdev_devid) {
5697                                 /*
5698                                  * In case of DUP, in order to keep it
5699                                  * simple, only add the mirror with the
5700                                  * lowest physical address
5701                                  */
5702                                 if (found &&
5703                                     physical_of_found <=
5704                                      bbio->stripes[i].physical)
5705                                         continue;
5706                                 index_srcdev = i;
5707                                 found = 1;
5708                                 physical_of_found = bbio->stripes[i].physical;
5709                         }
5710                 }
5711                 if (found) {
5712                         if (physical_of_found + map->stripe_len <=
5713                             dev_replace->cursor_left) {
5714                                 struct btrfs_bio_stripe *tgtdev_stripe =
5715                                         bbio->stripes + num_stripes;
5716
5717                                 tgtdev_stripe->physical = physical_of_found;
5718                                 tgtdev_stripe->length =
5719                                         bbio->stripes[index_srcdev].length;
5720                                 tgtdev_stripe->dev = dev_replace->tgtdev;
5721                                 bbio->tgtdev_map[index_srcdev] = num_stripes;
5722
5723                                 tgtdev_indexes++;
5724                                 num_stripes++;
5725                         }
5726                 }
5727         }
5728
5729         *bbio_ret = bbio;
5730         bbio->map_type = map->type;
5731         bbio->num_stripes = num_stripes;
5732         bbio->max_errors = max_errors;
5733         bbio->mirror_num = mirror_num;
5734         bbio->num_tgtdevs = tgtdev_indexes;
5735
5736         /*
5737          * this is the case that REQ_READ && dev_replace_is_ongoing &&
5738          * mirror_num == num_stripes + 1 && dev_replace target drive is
5739          * available as a mirror
5740          */
5741         if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) {
5742                 WARN_ON(num_stripes > 1);
5743                 bbio->stripes[0].dev = dev_replace->tgtdev;
5744                 bbio->stripes[0].physical = physical_to_patch_in_first_stripe;
5745                 bbio->mirror_num = map->num_stripes + 1;
5746         }
5747 out:
5748         if (dev_replace_is_ongoing) {
5749                 btrfs_dev_replace_clear_lock_blocking(dev_replace);
5750                 btrfs_dev_replace_unlock(dev_replace, 0);
5751         }
5752         free_extent_map(em);
5753         return ret;
5754 }
5755
5756 int btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
5757                       u64 logical, u64 *length,
5758                       struct btrfs_bio **bbio_ret, int mirror_num)
5759 {
5760         return __btrfs_map_block(fs_info, rw, logical, length, bbio_ret,
5761                                  mirror_num, 0);
5762 }
5763
5764 /* For Scrub/replace */
5765 int btrfs_map_sblock(struct btrfs_fs_info *fs_info, int rw,
5766                      u64 logical, u64 *length,
5767                      struct btrfs_bio **bbio_ret, int mirror_num,
5768                      int need_raid_map)
5769 {
5770         return __btrfs_map_block(fs_info, rw, logical, length, bbio_ret,
5771                                  mirror_num, need_raid_map);
5772 }
5773
5774 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
5775                      u64 chunk_start, u64 physical, u64 devid,
5776                      u64 **logical, int *naddrs, int *stripe_len)
5777 {
5778         struct extent_map_tree *em_tree = &map_tree->map_tree;
5779         struct extent_map *em;
5780         struct map_lookup *map;
5781         u64 *buf;
5782         u64 bytenr;
5783         u64 length;
5784         u64 stripe_nr;
5785         u64 rmap_len;
5786         int i, j, nr = 0;
5787
5788         read_lock(&em_tree->lock);
5789         em = lookup_extent_mapping(em_tree, chunk_start, 1);
5790         read_unlock(&em_tree->lock);
5791
5792         if (!em) {
5793                 printk(KERN_ERR "BTRFS: couldn't find em for chunk %Lu\n",
5794                        chunk_start);
5795                 return -EIO;
5796         }
5797
5798         if (em->start != chunk_start) {
5799                 printk(KERN_ERR "BTRFS: bad chunk start, em=%Lu, wanted=%Lu\n",
5800                        em->start, chunk_start);
5801                 free_extent_map(em);
5802                 return -EIO;
5803         }
5804         map = em->map_lookup;
5805
5806         length = em->len;
5807         rmap_len = map->stripe_len;
5808
5809         if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5810                 length = div_u64(length, map->num_stripes / map->sub_stripes);
5811         else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5812                 length = div_u64(length, map->num_stripes);
5813         else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5814                 length = div_u64(length, nr_data_stripes(map));
5815                 rmap_len = map->stripe_len * nr_data_stripes(map);
5816         }
5817
5818         buf = kcalloc(map->num_stripes, sizeof(u64), GFP_NOFS);
5819         BUG_ON(!buf); /* -ENOMEM */
5820
5821         for (i = 0; i < map->num_stripes; i++) {
5822                 if (devid && map->stripes[i].dev->devid != devid)
5823                         continue;
5824                 if (map->stripes[i].physical > physical ||
5825                     map->stripes[i].physical + length <= physical)
5826                         continue;
5827
5828                 stripe_nr = physical - map->stripes[i].physical;
5829                 stripe_nr = div_u64(stripe_nr, map->stripe_len);
5830
5831                 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5832                         stripe_nr = stripe_nr * map->num_stripes + i;
5833                         stripe_nr = div_u64(stripe_nr, map->sub_stripes);
5834                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5835                         stripe_nr = stripe_nr * map->num_stripes + i;
5836                 } /* else if RAID[56], multiply by nr_data_stripes().
5837                    * Alternatively, just use rmap_len below instead of
5838                    * map->stripe_len */
5839
5840                 bytenr = chunk_start + stripe_nr * rmap_len;
5841                 WARN_ON(nr >= map->num_stripes);
5842                 for (j = 0; j < nr; j++) {
5843                         if (buf[j] == bytenr)
5844                                 break;
5845                 }
5846                 if (j == nr) {
5847                         WARN_ON(nr >= map->num_stripes);
5848                         buf[nr++] = bytenr;
5849                 }
5850         }
5851
5852         *logical = buf;
5853         *naddrs = nr;
5854         *stripe_len = rmap_len;
5855
5856         free_extent_map(em);
5857         return 0;
5858 }
5859
5860 static inline void btrfs_end_bbio(struct btrfs_bio *bbio, struct bio *bio)
5861 {
5862         bio->bi_private = bbio->private;
5863         bio->bi_end_io = bbio->end_io;
5864         bio_endio(bio);
5865
5866         btrfs_put_bbio(bbio);
5867 }
5868
5869 static void btrfs_end_bio(struct bio *bio)
5870 {
5871         struct btrfs_bio *bbio = bio->bi_private;
5872         int is_orig_bio = 0;
5873
5874         if (bio->bi_error) {
5875                 atomic_inc(&bbio->error);
5876                 if (bio->bi_error == -EIO || bio->bi_error == -EREMOTEIO) {
5877                         unsigned int stripe_index =
5878                                 btrfs_io_bio(bio)->stripe_index;
5879                         struct btrfs_device *dev;
5880
5881                         BUG_ON(stripe_index >= bbio->num_stripes);
5882                         dev = bbio->stripes[stripe_index].dev;
5883                         if (dev->bdev) {
5884                                 if (bio->bi_rw & WRITE)
5885                                         btrfs_dev_stat_inc(dev,
5886                                                 BTRFS_DEV_STAT_WRITE_ERRS);
5887                                 else
5888                                         btrfs_dev_stat_inc(dev,
5889                                                 BTRFS_DEV_STAT_READ_ERRS);
5890                                 if ((bio->bi_rw & WRITE_FLUSH) == WRITE_FLUSH)
5891                                         btrfs_dev_stat_inc(dev,
5892                                                 BTRFS_DEV_STAT_FLUSH_ERRS);
5893                                 btrfs_dev_stat_print_on_error(dev);
5894                         }
5895                 }
5896         }
5897
5898         if (bio == bbio->orig_bio)
5899                 is_orig_bio = 1;
5900
5901         btrfs_bio_counter_dec(bbio->fs_info);
5902
5903         if (atomic_dec_and_test(&bbio->stripes_pending)) {
5904                 if (!is_orig_bio) {
5905                         bio_put(bio);
5906                         bio = bbio->orig_bio;
5907                 }
5908
5909                 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
5910                 /* only send an error to the higher layers if it is
5911                  * beyond the tolerance of the btrfs bio
5912                  */
5913                 if (atomic_read(&bbio->error) > bbio->max_errors) {
5914                         bio->bi_error = -EIO;
5915                 } else {
5916                         /*
5917                          * this bio is actually up to date, we didn't
5918                          * go over the max number of errors
5919                          */
5920                         bio->bi_error = 0;
5921                 }
5922
5923                 btrfs_end_bbio(bbio, bio);
5924         } else if (!is_orig_bio) {
5925                 bio_put(bio);
5926         }
5927 }
5928
5929 /*
5930  * see run_scheduled_bios for a description of why bios are collected for
5931  * async submit.
5932  *
5933  * This will add one bio to the pending list for a device and make sure
5934  * the work struct is scheduled.
5935  */
5936 static noinline void btrfs_schedule_bio(struct btrfs_root *root,
5937                                         struct btrfs_device *device,
5938                                         int rw, struct bio *bio)
5939 {
5940         int should_queue = 1;
5941         struct btrfs_pending_bios *pending_bios;
5942
5943         if (device->missing || !device->bdev) {
5944                 bio_io_error(bio);
5945                 return;
5946         }
5947
5948         /* don't bother with additional async steps for reads, right now */
5949         if (!(rw & REQ_WRITE)) {
5950                 bio_get(bio);
5951                 btrfsic_submit_bio(rw, bio);
5952                 bio_put(bio);
5953                 return;
5954         }
5955
5956         /*
5957          * nr_async_bios allows us to reliably return congestion to the
5958          * higher layers.  Otherwise, the async bio makes it appear we have
5959          * made progress against dirty pages when we've really just put it
5960          * on a queue for later
5961          */
5962         atomic_inc(&root->fs_info->nr_async_bios);
5963         WARN_ON(bio->bi_next);
5964         bio->bi_next = NULL;
5965         bio->bi_rw |= rw;
5966
5967         spin_lock(&device->io_lock);
5968         if (bio->bi_rw & REQ_SYNC)
5969                 pending_bios = &device->pending_sync_bios;
5970         else
5971                 pending_bios = &device->pending_bios;
5972
5973         if (pending_bios->tail)
5974                 pending_bios->tail->bi_next = bio;
5975
5976         pending_bios->tail = bio;
5977         if (!pending_bios->head)
5978                 pending_bios->head = bio;
5979         if (device->running_pending)
5980                 should_queue = 0;
5981
5982         spin_unlock(&device->io_lock);
5983
5984         if (should_queue)
5985                 btrfs_queue_work(root->fs_info->submit_workers,
5986                                  &device->work);
5987 }
5988
5989 static void submit_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio,
5990                               struct bio *bio, u64 physical, int dev_nr,
5991                               int rw, int async)
5992 {
5993         struct btrfs_device *dev = bbio->stripes[dev_nr].dev;
5994
5995         bio->bi_private = bbio;
5996         btrfs_io_bio(bio)->stripe_index = dev_nr;
5997         bio->bi_end_io = btrfs_end_bio;
5998         bio->bi_iter.bi_sector = physical >> 9;
5999 #ifdef DEBUG
6000         {
6001                 struct rcu_string *name;
6002
6003                 rcu_read_lock();
6004                 name = rcu_dereference(dev->name);
6005                 pr_debug("btrfs_map_bio: rw %d, sector=%llu, dev=%lu "
6006                          "(%s id %llu), size=%u\n", rw,
6007                          (u64)bio->bi_iter.bi_sector, (u_long)dev->bdev->bd_dev,
6008                          name->str, dev->devid, bio->bi_iter.bi_size);
6009                 rcu_read_unlock();
6010         }
6011 #endif
6012         bio->bi_bdev = dev->bdev;
6013
6014         btrfs_bio_counter_inc_noblocked(root->fs_info);
6015
6016         if (async)
6017                 btrfs_schedule_bio(root, dev, rw, bio);
6018         else
6019                 btrfsic_submit_bio(rw, bio);
6020 }
6021
6022 static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical)
6023 {
6024         atomic_inc(&bbio->error);
6025         if (atomic_dec_and_test(&bbio->stripes_pending)) {
6026                 /* Shoud be the original bio. */
6027                 WARN_ON(bio != bbio->orig_bio);
6028
6029                 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
6030                 bio->bi_iter.bi_sector = logical >> 9;
6031                 bio->bi_error = -EIO;
6032                 btrfs_end_bbio(bbio, bio);
6033         }
6034 }
6035
6036 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
6037                   int mirror_num, int async_submit)
6038 {
6039         struct btrfs_device *dev;
6040         struct bio *first_bio = bio;
6041         u64 logical = (u64)bio->bi_iter.bi_sector << 9;
6042         u64 length = 0;
6043         u64 map_length;
6044         int ret;
6045         int dev_nr;
6046         int total_devs;
6047         struct btrfs_bio *bbio = NULL;
6048
6049         length = bio->bi_iter.bi_size;
6050         map_length = length;
6051
6052         btrfs_bio_counter_inc_blocked(root->fs_info);
6053         ret = __btrfs_map_block(root->fs_info, rw, logical, &map_length, &bbio,
6054                               mirror_num, 1);
6055         if (ret) {
6056                 btrfs_bio_counter_dec(root->fs_info);
6057                 return ret;
6058         }
6059
6060         total_devs = bbio->num_stripes;
6061         bbio->orig_bio = first_bio;
6062         bbio->private = first_bio->bi_private;
6063         bbio->end_io = first_bio->bi_end_io;
6064         bbio->fs_info = root->fs_info;
6065         atomic_set(&bbio->stripes_pending, bbio->num_stripes);
6066
6067         if ((bbio->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
6068             ((rw & WRITE) || (mirror_num > 1))) {
6069                 /* In this case, map_length has been set to the length of
6070                    a single stripe; not the whole write */
6071                 if (rw & WRITE) {
6072                         ret = raid56_parity_write(root, bio, bbio, map_length);
6073                 } else {
6074                         ret = raid56_parity_recover(root, bio, bbio, map_length,
6075                                                     mirror_num, 1);
6076                 }
6077
6078                 btrfs_bio_counter_dec(root->fs_info);
6079                 return ret;
6080         }
6081
6082         if (map_length < length) {
6083                 btrfs_crit(root->fs_info, "mapping failed logical %llu bio len %llu len %llu",
6084                         logical, length, map_length);
6085                 BUG();
6086         }
6087
6088         for (dev_nr = 0; dev_nr < total_devs; dev_nr++) {
6089                 dev = bbio->stripes[dev_nr].dev;
6090                 if (!dev || !dev->bdev || (rw & WRITE && !dev->writeable)) {
6091                         bbio_error(bbio, first_bio, logical);
6092                         continue;
6093                 }
6094
6095                 if (dev_nr < total_devs - 1) {
6096                         bio = btrfs_bio_clone(first_bio, GFP_NOFS);
6097                         BUG_ON(!bio); /* -ENOMEM */
6098                 } else
6099                         bio = first_bio;
6100
6101                 submit_stripe_bio(root, bbio, bio,
6102                                   bbio->stripes[dev_nr].physical, dev_nr, rw,
6103                                   async_submit);
6104         }
6105         btrfs_bio_counter_dec(root->fs_info);
6106         return 0;
6107 }
6108
6109 struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid,
6110                                        u8 *uuid, u8 *fsid)
6111 {
6112         struct btrfs_device *device;
6113         struct btrfs_fs_devices *cur_devices;
6114
6115         cur_devices = fs_info->fs_devices;
6116         while (cur_devices) {
6117                 if (!fsid ||
6118                     !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
6119                         device = __find_device(&cur_devices->devices,
6120                                                devid, uuid);
6121                         if (device)
6122                                 return device;
6123                 }
6124                 cur_devices = cur_devices->seed;
6125         }
6126         return NULL;
6127 }
6128
6129 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
6130                                             struct btrfs_fs_devices *fs_devices,
6131                                             u64 devid, u8 *dev_uuid)
6132 {
6133         struct btrfs_device *device;
6134
6135         device = btrfs_alloc_device(NULL, &devid, dev_uuid);
6136         if (IS_ERR(device))
6137                 return NULL;
6138
6139         list_add(&device->dev_list, &fs_devices->devices);
6140         device->fs_devices = fs_devices;
6141         fs_devices->num_devices++;
6142
6143         device->missing = 1;
6144         fs_devices->missing_devices++;
6145
6146         return device;
6147 }
6148
6149 /**
6150  * btrfs_alloc_device - allocate struct btrfs_device
6151  * @fs_info:    used only for generating a new devid, can be NULL if
6152  *              devid is provided (i.e. @devid != NULL).
6153  * @devid:      a pointer to devid for this device.  If NULL a new devid
6154  *              is generated.
6155  * @uuid:       a pointer to UUID for this device.  If NULL a new UUID
6156  *              is generated.
6157  *
6158  * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
6159  * on error.  Returned struct is not linked onto any lists and can be
6160  * destroyed with kfree() right away.
6161  */
6162 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
6163                                         const u64 *devid,
6164                                         const u8 *uuid)
6165 {
6166         struct btrfs_device *dev;
6167         u64 tmp;
6168
6169         if (WARN_ON(!devid && !fs_info))
6170                 return ERR_PTR(-EINVAL);
6171
6172         dev = __alloc_device();
6173         if (IS_ERR(dev))
6174                 return dev;
6175
6176         if (devid)
6177                 tmp = *devid;
6178         else {
6179                 int ret;
6180
6181                 ret = find_next_devid(fs_info, &tmp);
6182                 if (ret) {
6183                         kfree(dev);
6184                         return ERR_PTR(ret);
6185                 }
6186         }
6187         dev->devid = tmp;
6188
6189         if (uuid)
6190                 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE);
6191         else
6192                 generate_random_uuid(dev->uuid);
6193
6194         btrfs_init_work(&dev->work, btrfs_submit_helper,
6195                         pending_bios_fn, NULL, NULL);
6196
6197         return dev;
6198 }
6199
6200 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
6201                           struct extent_buffer *leaf,
6202                           struct btrfs_chunk *chunk)
6203 {
6204         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
6205         struct map_lookup *map;
6206         struct extent_map *em;
6207         u64 logical;
6208         u64 length;
6209         u64 stripe_len;
6210         u64 devid;
6211         u8 uuid[BTRFS_UUID_SIZE];
6212         int num_stripes;
6213         int ret;
6214         int i;
6215
6216         logical = key->offset;
6217         length = btrfs_chunk_length(leaf, chunk);
6218         stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6219         num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6220         /* Validation check */
6221         if (!num_stripes) {
6222                 btrfs_err(root->fs_info, "invalid chunk num_stripes: %u",
6223                           num_stripes);
6224                 return -EIO;
6225         }
6226         if (!IS_ALIGNED(logical, root->sectorsize)) {
6227                 btrfs_err(root->fs_info,
6228                           "invalid chunk logical %llu", logical);
6229                 return -EIO;
6230         }
6231         if (!length || !IS_ALIGNED(length, root->sectorsize)) {
6232                 btrfs_err(root->fs_info,
6233                         "invalid chunk length %llu", length);
6234                 return -EIO;
6235         }
6236         if (!is_power_of_2(stripe_len)) {
6237                 btrfs_err(root->fs_info, "invalid chunk stripe length: %llu",
6238                           stripe_len);
6239                 return -EIO;
6240         }
6241         if (~(BTRFS_BLOCK_GROUP_TYPE_MASK | BTRFS_BLOCK_GROUP_PROFILE_MASK) &
6242             btrfs_chunk_type(leaf, chunk)) {
6243                 btrfs_err(root->fs_info, "unrecognized chunk type: %llu",
6244                           ~(BTRFS_BLOCK_GROUP_TYPE_MASK |
6245                             BTRFS_BLOCK_GROUP_PROFILE_MASK) &
6246                           btrfs_chunk_type(leaf, chunk));
6247                 return -EIO;
6248         }
6249
6250         read_lock(&map_tree->map_tree.lock);
6251         em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
6252         read_unlock(&map_tree->map_tree.lock);
6253
6254         /* already mapped? */
6255         if (em && em->start <= logical && em->start + em->len > logical) {
6256                 free_extent_map(em);
6257                 return 0;
6258         } else if (em) {
6259                 free_extent_map(em);
6260         }
6261
6262         em = alloc_extent_map();
6263         if (!em)
6264                 return -ENOMEM;
6265         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
6266         if (!map) {
6267                 free_extent_map(em);
6268                 return -ENOMEM;
6269         }
6270
6271         set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
6272         em->map_lookup = map;
6273         em->start = logical;
6274         em->len = length;
6275         em->orig_start = 0;
6276         em->block_start = 0;
6277         em->block_len = em->len;
6278
6279         map->num_stripes = num_stripes;
6280         map->io_width = btrfs_chunk_io_width(leaf, chunk);
6281         map->io_align = btrfs_chunk_io_align(leaf, chunk);
6282         map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
6283         map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6284         map->type = btrfs_chunk_type(leaf, chunk);
6285         map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
6286         for (i = 0; i < num_stripes; i++) {
6287                 map->stripes[i].physical =
6288                         btrfs_stripe_offset_nr(leaf, chunk, i);
6289                 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
6290                 read_extent_buffer(leaf, uuid, (unsigned long)
6291                                    btrfs_stripe_dev_uuid_nr(chunk, i),
6292                                    BTRFS_UUID_SIZE);
6293                 map->stripes[i].dev = btrfs_find_device(root->fs_info, devid,
6294                                                         uuid, NULL);
6295                 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
6296                         free_extent_map(em);
6297                         return -EIO;
6298                 }
6299                 if (!map->stripes[i].dev) {
6300                         map->stripes[i].dev =
6301                                 add_missing_dev(root, root->fs_info->fs_devices,
6302                                                 devid, uuid);
6303                         if (!map->stripes[i].dev) {
6304                                 free_extent_map(em);
6305                                 return -EIO;
6306                         }
6307                         btrfs_warn(root->fs_info, "devid %llu uuid %pU is missing",
6308                                                 devid, uuid);
6309                 }
6310                 map->stripes[i].dev->in_fs_metadata = 1;
6311         }
6312
6313         write_lock(&map_tree->map_tree.lock);
6314         ret = add_extent_mapping(&map_tree->map_tree, em, 0);
6315         write_unlock(&map_tree->map_tree.lock);
6316         BUG_ON(ret); /* Tree corruption */
6317         free_extent_map(em);
6318
6319         return 0;
6320 }
6321
6322 static void fill_device_from_item(struct extent_buffer *leaf,
6323                                  struct btrfs_dev_item *dev_item,
6324                                  struct btrfs_device *device)
6325 {
6326         unsigned long ptr;
6327
6328         device->devid = btrfs_device_id(leaf, dev_item);
6329         device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
6330         device->total_bytes = device->disk_total_bytes;
6331         device->commit_total_bytes = device->disk_total_bytes;
6332         device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
6333         device->commit_bytes_used = device->bytes_used;
6334         device->type = btrfs_device_type(leaf, dev_item);
6335         device->io_align = btrfs_device_io_align(leaf, dev_item);
6336         device->io_width = btrfs_device_io_width(leaf, dev_item);
6337         device->sector_size = btrfs_device_sector_size(leaf, dev_item);
6338         WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
6339         device->is_tgtdev_for_dev_replace = 0;
6340
6341         ptr = btrfs_device_uuid(dev_item);
6342         read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
6343 }
6344
6345 static struct btrfs_fs_devices *open_seed_devices(struct btrfs_root *root,
6346                                                   u8 *fsid)
6347 {
6348         struct btrfs_fs_devices *fs_devices;
6349         int ret;
6350
6351         BUG_ON(!mutex_is_locked(&uuid_mutex));
6352
6353         fs_devices = root->fs_info->fs_devices->seed;
6354         while (fs_devices) {
6355                 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE))
6356                         return fs_devices;
6357
6358                 fs_devices = fs_devices->seed;
6359         }
6360
6361         fs_devices = find_fsid(fsid);
6362         if (!fs_devices) {
6363                 if (!btrfs_test_opt(root, DEGRADED))
6364                         return ERR_PTR(-ENOENT);
6365
6366                 fs_devices = alloc_fs_devices(fsid);
6367                 if (IS_ERR(fs_devices))
6368                         return fs_devices;
6369
6370                 fs_devices->seeding = 1;
6371                 fs_devices->opened = 1;
6372                 return fs_devices;
6373         }
6374
6375         fs_devices = clone_fs_devices(fs_devices);
6376         if (IS_ERR(fs_devices))
6377                 return fs_devices;
6378
6379         ret = __btrfs_open_devices(fs_devices, FMODE_READ,
6380                                    root->fs_info->bdev_holder);
6381         if (ret) {
6382                 free_fs_devices(fs_devices);
6383                 fs_devices = ERR_PTR(ret);
6384                 goto out;
6385         }
6386
6387         if (!fs_devices->seeding) {
6388                 __btrfs_close_devices(fs_devices);
6389                 free_fs_devices(fs_devices);
6390                 fs_devices = ERR_PTR(-EINVAL);
6391                 goto out;
6392         }
6393
6394         fs_devices->seed = root->fs_info->fs_devices->seed;
6395         root->fs_info->fs_devices->seed = fs_devices;
6396 out:
6397         return fs_devices;
6398 }
6399
6400 static int read_one_dev(struct btrfs_root *root,
6401                         struct extent_buffer *leaf,
6402                         struct btrfs_dev_item *dev_item)
6403 {
6404         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
6405         struct btrfs_device *device;
6406         u64 devid;
6407         int ret;
6408         u8 fs_uuid[BTRFS_UUID_SIZE];
6409         u8 dev_uuid[BTRFS_UUID_SIZE];
6410
6411         devid = btrfs_device_id(leaf, dev_item);
6412         read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
6413                            BTRFS_UUID_SIZE);
6414         read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
6415                            BTRFS_UUID_SIZE);
6416
6417         if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
6418                 fs_devices = open_seed_devices(root, fs_uuid);
6419                 if (IS_ERR(fs_devices))
6420                         return PTR_ERR(fs_devices);
6421         }
6422
6423         device = btrfs_find_device(root->fs_info, devid, dev_uuid, fs_uuid);
6424         if (!device) {
6425                 if (!btrfs_test_opt(root, DEGRADED))
6426                         return -EIO;
6427
6428                 device = add_missing_dev(root, fs_devices, devid, dev_uuid);
6429                 if (!device)
6430                         return -ENOMEM;
6431                 btrfs_warn(root->fs_info, "devid %llu uuid %pU missing",
6432                                 devid, dev_uuid);
6433         } else {
6434                 if (!device->bdev && !btrfs_test_opt(root, DEGRADED))
6435                         return -EIO;
6436
6437                 if(!device->bdev && !device->missing) {
6438                         /*
6439                          * this happens when a device that was properly setup
6440                          * in the device info lists suddenly goes bad.
6441                          * device->bdev is NULL, and so we have to set
6442                          * device->missing to one here
6443                          */
6444                         device->fs_devices->missing_devices++;
6445                         device->missing = 1;
6446                 }
6447
6448                 /* Move the device to its own fs_devices */
6449                 if (device->fs_devices != fs_devices) {
6450                         ASSERT(device->missing);
6451
6452                         list_move(&device->dev_list, &fs_devices->devices);
6453                         device->fs_devices->num_devices--;
6454                         fs_devices->num_devices++;
6455
6456                         device->fs_devices->missing_devices--;
6457                         fs_devices->missing_devices++;
6458
6459                         device->fs_devices = fs_devices;
6460                 }
6461         }
6462
6463         if (device->fs_devices != root->fs_info->fs_devices) {
6464                 BUG_ON(device->writeable);
6465                 if (device->generation !=
6466                     btrfs_device_generation(leaf, dev_item))
6467                         return -EINVAL;
6468         }
6469
6470         fill_device_from_item(leaf, dev_item, device);
6471         device->in_fs_metadata = 1;
6472         if (device->writeable && !device->is_tgtdev_for_dev_replace) {
6473                 device->fs_devices->total_rw_bytes += device->total_bytes;
6474                 spin_lock(&root->fs_info->free_chunk_lock);
6475                 root->fs_info->free_chunk_space += device->total_bytes -
6476                         device->bytes_used;
6477                 spin_unlock(&root->fs_info->free_chunk_lock);
6478         }
6479         ret = 0;
6480         return ret;
6481 }
6482
6483 int btrfs_read_sys_array(struct btrfs_root *root)
6484 {
6485         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
6486         struct extent_buffer *sb;
6487         struct btrfs_disk_key *disk_key;
6488         struct btrfs_chunk *chunk;
6489         u8 *array_ptr;
6490         unsigned long sb_array_offset;
6491         int ret = 0;
6492         u32 num_stripes;
6493         u32 array_size;
6494         u32 len = 0;
6495         u32 cur_offset;
6496         struct btrfs_key key;
6497
6498         ASSERT(BTRFS_SUPER_INFO_SIZE <= root->nodesize);
6499         /*
6500          * This will create extent buffer of nodesize, superblock size is
6501          * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will
6502          * overallocate but we can keep it as-is, only the first page is used.
6503          */
6504         sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET);
6505         if (!sb)
6506                 return -ENOMEM;
6507         set_extent_buffer_uptodate(sb);
6508         btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
6509         /*
6510          * The sb extent buffer is artifical and just used to read the system array.
6511          * set_extent_buffer_uptodate() call does not properly mark all it's
6512          * pages up-to-date when the page is larger: extent does not cover the
6513          * whole page and consequently check_page_uptodate does not find all
6514          * the page's extents up-to-date (the hole beyond sb),
6515          * write_extent_buffer then triggers a WARN_ON.
6516          *
6517          * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
6518          * but sb spans only this function. Add an explicit SetPageUptodate call
6519          * to silence the warning eg. on PowerPC 64.
6520          */
6521         if (PAGE_SIZE > BTRFS_SUPER_INFO_SIZE)
6522                 SetPageUptodate(sb->pages[0]);
6523
6524         write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
6525         array_size = btrfs_super_sys_array_size(super_copy);
6526
6527         array_ptr = super_copy->sys_chunk_array;
6528         sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array);
6529         cur_offset = 0;
6530
6531         while (cur_offset < array_size) {
6532                 disk_key = (struct btrfs_disk_key *)array_ptr;
6533                 len = sizeof(*disk_key);
6534                 if (cur_offset + len > array_size)
6535                         goto out_short_read;
6536
6537                 btrfs_disk_key_to_cpu(&key, disk_key);
6538
6539                 array_ptr += len;
6540                 sb_array_offset += len;
6541                 cur_offset += len;
6542
6543                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
6544                         chunk = (struct btrfs_chunk *)sb_array_offset;
6545                         /*
6546                          * At least one btrfs_chunk with one stripe must be
6547                          * present, exact stripe count check comes afterwards
6548                          */
6549                         len = btrfs_chunk_item_size(1);
6550                         if (cur_offset + len > array_size)
6551                                 goto out_short_read;
6552
6553                         num_stripes = btrfs_chunk_num_stripes(sb, chunk);
6554                         if (!num_stripes) {
6555                                 printk(KERN_ERR
6556             "BTRFS: invalid number of stripes %u in sys_array at offset %u\n",
6557                                         num_stripes, cur_offset);
6558                                 ret = -EIO;
6559                                 break;
6560                         }
6561
6562                         len = btrfs_chunk_item_size(num_stripes);
6563                         if (cur_offset + len > array_size)
6564                                 goto out_short_read;
6565
6566                         ret = read_one_chunk(root, &key, sb, chunk);
6567                         if (ret)
6568                                 break;
6569                 } else {
6570                         printk(KERN_ERR
6571                 "BTRFS: unexpected item type %u in sys_array at offset %u\n",
6572                                 (u32)key.type, cur_offset);
6573                         ret = -EIO;
6574                         break;
6575                 }
6576                 array_ptr += len;
6577                 sb_array_offset += len;
6578                 cur_offset += len;
6579         }
6580         free_extent_buffer(sb);
6581         return ret;
6582
6583 out_short_read:
6584         printk(KERN_ERR "BTRFS: sys_array too short to read %u bytes at offset %u\n",
6585                         len, cur_offset);
6586         free_extent_buffer(sb);
6587         return -EIO;
6588 }
6589
6590 int btrfs_read_chunk_tree(struct btrfs_root *root)
6591 {
6592         struct btrfs_path *path;
6593         struct extent_buffer *leaf;
6594         struct btrfs_key key;
6595         struct btrfs_key found_key;
6596         int ret;
6597         int slot;
6598
6599         root = root->fs_info->chunk_root;
6600
6601         path = btrfs_alloc_path();
6602         if (!path)
6603                 return -ENOMEM;
6604
6605         mutex_lock(&uuid_mutex);
6606         lock_chunks(root);
6607
6608         /*
6609          * Read all device items, and then all the chunk items. All
6610          * device items are found before any chunk item (their object id
6611          * is smaller than the lowest possible object id for a chunk
6612          * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
6613          */
6614         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
6615         key.offset = 0;
6616         key.type = 0;
6617         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6618         if (ret < 0)
6619                 goto error;
6620         while (1) {
6621                 leaf = path->nodes[0];
6622                 slot = path->slots[0];
6623                 if (slot >= btrfs_header_nritems(leaf)) {
6624                         ret = btrfs_next_leaf(root, path);
6625                         if (ret == 0)
6626                                 continue;
6627                         if (ret < 0)
6628                                 goto error;
6629                         break;
6630                 }
6631                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6632                 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
6633                         struct btrfs_dev_item *dev_item;
6634                         dev_item = btrfs_item_ptr(leaf, slot,
6635                                                   struct btrfs_dev_item);
6636                         ret = read_one_dev(root, leaf, dev_item);
6637                         if (ret)
6638                                 goto error;
6639                 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
6640                         struct btrfs_chunk *chunk;
6641                         chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
6642                         ret = read_one_chunk(root, &found_key, leaf, chunk);
6643                         if (ret)
6644                                 goto error;
6645                 }
6646                 path->slots[0]++;
6647         }
6648         ret = 0;
6649 error:
6650         unlock_chunks(root);
6651         mutex_unlock(&uuid_mutex);
6652
6653         btrfs_free_path(path);
6654         return ret;
6655 }
6656
6657 void btrfs_init_devices_late(struct btrfs_fs_info *fs_info)
6658 {
6659         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6660         struct btrfs_device *device;
6661
6662         while (fs_devices) {
6663                 mutex_lock(&fs_devices->device_list_mutex);
6664                 list_for_each_entry(device, &fs_devices->devices, dev_list)
6665                         device->dev_root = fs_info->dev_root;
6666                 mutex_unlock(&fs_devices->device_list_mutex);
6667
6668                 fs_devices = fs_devices->seed;
6669         }
6670 }
6671
6672 static void __btrfs_reset_dev_stats(struct btrfs_device *dev)
6673 {
6674         int i;
6675
6676         for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6677                 btrfs_dev_stat_reset(dev, i);
6678 }
6679
6680 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
6681 {
6682         struct btrfs_key key;
6683         struct btrfs_key found_key;
6684         struct btrfs_root *dev_root = fs_info->dev_root;
6685         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6686         struct extent_buffer *eb;
6687         int slot;
6688         int ret = 0;
6689         struct btrfs_device *device;
6690         struct btrfs_path *path = NULL;
6691         int i;
6692
6693         path = btrfs_alloc_path();
6694         if (!path) {
6695                 ret = -ENOMEM;
6696                 goto out;
6697         }
6698
6699         mutex_lock(&fs_devices->device_list_mutex);
6700         list_for_each_entry(device, &fs_devices->devices, dev_list) {
6701                 int item_size;
6702                 struct btrfs_dev_stats_item *ptr;
6703
6704                 key.objectid = BTRFS_DEV_STATS_OBJECTID;
6705                 key.type = BTRFS_PERSISTENT_ITEM_KEY;
6706                 key.offset = device->devid;
6707                 ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
6708                 if (ret) {
6709                         __btrfs_reset_dev_stats(device);
6710                         device->dev_stats_valid = 1;
6711                         btrfs_release_path(path);
6712                         continue;
6713                 }
6714                 slot = path->slots[0];
6715                 eb = path->nodes[0];
6716                 btrfs_item_key_to_cpu(eb, &found_key, slot);
6717                 item_size = btrfs_item_size_nr(eb, slot);
6718
6719                 ptr = btrfs_item_ptr(eb, slot,
6720                                      struct btrfs_dev_stats_item);
6721
6722                 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
6723                         if (item_size >= (1 + i) * sizeof(__le64))
6724                                 btrfs_dev_stat_set(device, i,
6725                                         btrfs_dev_stats_value(eb, ptr, i));
6726                         else
6727                                 btrfs_dev_stat_reset(device, i);
6728                 }
6729
6730                 device->dev_stats_valid = 1;
6731                 btrfs_dev_stat_print_on_load(device);
6732                 btrfs_release_path(path);
6733         }
6734         mutex_unlock(&fs_devices->device_list_mutex);
6735
6736 out:
6737         btrfs_free_path(path);
6738         return ret < 0 ? ret : 0;
6739 }
6740
6741 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
6742                                 struct btrfs_root *dev_root,
6743                                 struct btrfs_device *device)
6744 {
6745         struct btrfs_path *path;
6746         struct btrfs_key key;
6747         struct extent_buffer *eb;
6748         struct btrfs_dev_stats_item *ptr;
6749         int ret;
6750         int i;
6751
6752         key.objectid = BTRFS_DEV_STATS_OBJECTID;
6753         key.type = BTRFS_PERSISTENT_ITEM_KEY;
6754         key.offset = device->devid;
6755
6756         path = btrfs_alloc_path();
6757         BUG_ON(!path);
6758         ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
6759         if (ret < 0) {
6760                 btrfs_warn_in_rcu(dev_root->fs_info,
6761                         "error %d while searching for dev_stats item for device %s",
6762                               ret, rcu_str_deref(device->name));
6763                 goto out;
6764         }
6765
6766         if (ret == 0 &&
6767             btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
6768                 /* need to delete old one and insert a new one */
6769                 ret = btrfs_del_item(trans, dev_root, path);
6770                 if (ret != 0) {
6771                         btrfs_warn_in_rcu(dev_root->fs_info,
6772                                 "delete too small dev_stats item for device %s failed %d",
6773                                       rcu_str_deref(device->name), ret);
6774                         goto out;
6775                 }
6776                 ret = 1;
6777         }
6778
6779         if (ret == 1) {
6780                 /* need to insert a new item */
6781                 btrfs_release_path(path);
6782                 ret = btrfs_insert_empty_item(trans, dev_root, path,
6783                                               &key, sizeof(*ptr));
6784                 if (ret < 0) {
6785                         btrfs_warn_in_rcu(dev_root->fs_info,
6786                                 "insert dev_stats item for device %s failed %d",
6787                                 rcu_str_deref(device->name), ret);
6788                         goto out;
6789                 }
6790         }
6791
6792         eb = path->nodes[0];
6793         ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
6794         for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6795                 btrfs_set_dev_stats_value(eb, ptr, i,
6796                                           btrfs_dev_stat_read(device, i));
6797         btrfs_mark_buffer_dirty(eb);
6798
6799 out:
6800         btrfs_free_path(path);
6801         return ret;
6802 }
6803
6804 /*
6805  * called from commit_transaction. Writes all changed device stats to disk.
6806  */
6807 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans,
6808                         struct btrfs_fs_info *fs_info)
6809 {
6810         struct btrfs_root *dev_root = fs_info->dev_root;
6811         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6812         struct btrfs_device *device;
6813         int stats_cnt;
6814         int ret = 0;
6815
6816         mutex_lock(&fs_devices->device_list_mutex);
6817         list_for_each_entry(device, &fs_devices->devices, dev_list) {
6818                 if (!device->dev_stats_valid || !btrfs_dev_stats_dirty(device))
6819                         continue;
6820
6821                 stats_cnt = atomic_read(&device->dev_stats_ccnt);
6822                 ret = update_dev_stat_item(trans, dev_root, device);
6823                 if (!ret)
6824                         atomic_sub(stats_cnt, &device->dev_stats_ccnt);
6825         }
6826         mutex_unlock(&fs_devices->device_list_mutex);
6827
6828         return ret;
6829 }
6830
6831 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
6832 {
6833         btrfs_dev_stat_inc(dev, index);
6834         btrfs_dev_stat_print_on_error(dev);
6835 }
6836
6837 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
6838 {
6839         if (!dev->dev_stats_valid)
6840                 return;
6841         btrfs_err_rl_in_rcu(dev->dev_root->fs_info,
6842                 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
6843                            rcu_str_deref(dev->name),
6844                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
6845                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
6846                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
6847                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
6848                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
6849 }
6850
6851 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
6852 {
6853         int i;
6854
6855         for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6856                 if (btrfs_dev_stat_read(dev, i) != 0)
6857                         break;
6858         if (i == BTRFS_DEV_STAT_VALUES_MAX)
6859                 return; /* all values == 0, suppress message */
6860
6861         btrfs_info_in_rcu(dev->dev_root->fs_info,
6862                 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
6863                rcu_str_deref(dev->name),
6864                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
6865                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
6866                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
6867                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
6868                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
6869 }
6870
6871 int btrfs_get_dev_stats(struct btrfs_root *root,
6872                         struct btrfs_ioctl_get_dev_stats *stats)
6873 {
6874         struct btrfs_device *dev;
6875         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
6876         int i;
6877
6878         mutex_lock(&fs_devices->device_list_mutex);
6879         dev = btrfs_find_device(root->fs_info, stats->devid, NULL, NULL);
6880         mutex_unlock(&fs_devices->device_list_mutex);
6881
6882         if (!dev) {
6883                 btrfs_warn(root->fs_info, "get dev_stats failed, device not found");
6884                 return -ENODEV;
6885         } else if (!dev->dev_stats_valid) {
6886                 btrfs_warn(root->fs_info, "get dev_stats failed, not yet valid");
6887                 return -ENODEV;
6888         } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
6889                 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
6890                         if (stats->nr_items > i)
6891                                 stats->values[i] =
6892                                         btrfs_dev_stat_read_and_reset(dev, i);
6893                         else
6894                                 btrfs_dev_stat_reset(dev, i);
6895                 }
6896         } else {
6897                 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6898                         if (stats->nr_items > i)
6899                                 stats->values[i] = btrfs_dev_stat_read(dev, i);
6900         }
6901         if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
6902                 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
6903         return 0;
6904 }
6905
6906 void btrfs_scratch_superblocks(struct block_device *bdev, char *device_path)
6907 {
6908         struct buffer_head *bh;
6909         struct btrfs_super_block *disk_super;
6910         int copy_num;
6911
6912         if (!bdev)
6913                 return;
6914
6915         for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX;
6916                 copy_num++) {
6917
6918                 if (btrfs_read_dev_one_super(bdev, copy_num, &bh))
6919                         continue;
6920
6921                 disk_super = (struct btrfs_super_block *)bh->b_data;
6922
6923                 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
6924                 set_buffer_dirty(bh);
6925                 sync_dirty_buffer(bh);
6926                 brelse(bh);
6927         }
6928
6929         /* Notify udev that device has changed */
6930         btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
6931
6932         /* Update ctime/mtime for device path for libblkid */
6933         update_dev_time(device_path);
6934 }
6935
6936 /*
6937  * Update the size of all devices, which is used for writing out the
6938  * super blocks.
6939  */
6940 void btrfs_update_commit_device_size(struct btrfs_fs_info *fs_info)
6941 {
6942         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6943         struct btrfs_device *curr, *next;
6944
6945         if (list_empty(&fs_devices->resized_devices))
6946                 return;
6947
6948         mutex_lock(&fs_devices->device_list_mutex);
6949         lock_chunks(fs_info->dev_root);
6950         list_for_each_entry_safe(curr, next, &fs_devices->resized_devices,
6951                                  resized_list) {
6952                 list_del_init(&curr->resized_list);
6953                 curr->commit_total_bytes = curr->disk_total_bytes;
6954         }
6955         unlock_chunks(fs_info->dev_root);
6956         mutex_unlock(&fs_devices->device_list_mutex);
6957 }
6958
6959 /* Must be invoked during the transaction commit */
6960 void btrfs_update_commit_device_bytes_used(struct btrfs_root *root,
6961                                         struct btrfs_transaction *transaction)
6962 {
6963         struct extent_map *em;
6964         struct map_lookup *map;
6965         struct btrfs_device *dev;
6966         int i;
6967
6968         if (list_empty(&transaction->pending_chunks))
6969                 return;
6970
6971         /* In order to kick the device replace finish process */
6972         lock_chunks(root);
6973         list_for_each_entry(em, &transaction->pending_chunks, list) {
6974                 map = em->map_lookup;
6975
6976                 for (i = 0; i < map->num_stripes; i++) {
6977                         dev = map->stripes[i].dev;
6978                         dev->commit_bytes_used = dev->bytes_used;
6979                 }
6980         }
6981         unlock_chunks(root);
6982 }
6983
6984 void btrfs_set_fs_info_ptr(struct btrfs_fs_info *fs_info)
6985 {
6986         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6987         while (fs_devices) {
6988                 fs_devices->fs_info = fs_info;
6989                 fs_devices = fs_devices->seed;
6990         }
6991 }
6992
6993 void btrfs_reset_fs_info_ptr(struct btrfs_fs_info *fs_info)
6994 {
6995         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6996         while (fs_devices) {
6997                 fs_devices->fs_info = NULL;
6998                 fs_devices = fs_devices->seed;
6999         }
7000 }
7001
7002 static void btrfs_close_one_device(struct btrfs_device *device)
7003 {
7004         struct btrfs_fs_devices *fs_devices = device->fs_devices;
7005         struct btrfs_device *new_device;
7006         struct rcu_string *name;
7007
7008         if (device->bdev)
7009                 fs_devices->open_devices--;
7010
7011         if (device->writeable &&
7012             device->devid != BTRFS_DEV_REPLACE_DEVID) {
7013                 list_del_init(&device->dev_alloc_list);
7014                 fs_devices->rw_devices--;
7015         }
7016
7017         if (device->missing)
7018                 fs_devices->missing_devices--;
7019
7020         new_device = btrfs_alloc_device(NULL, &device->devid,
7021                                         device->uuid);
7022         BUG_ON(IS_ERR(new_device)); /* -ENOMEM */
7023
7024         /* Safe because we are under uuid_mutex */
7025         if (device->name) {
7026                 name = rcu_string_strdup(device->name->str, GFP_NOFS);
7027                 BUG_ON(!name); /* -ENOMEM */
7028                 rcu_assign_pointer(new_device->name, name);
7029         }
7030
7031         list_replace_rcu(&device->dev_list, &new_device->dev_list);
7032         new_device->fs_devices = device->fs_devices;
7033
7034         call_rcu(&device->rcu, free_device);
7035 }