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/iocontext.h>
  24 #include <linux/capability.h>
  25 #include <linux/ratelimit.h>
  26 #include <linux/kthread.h>
  27 #include <linux/raid/pq.h>
  28 #include <linux/semaphore.h>
  29 #include <linux/uuid.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);
 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_handle_fs_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_handle_fs_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         int i;
1735
1736         do {
1737                 seq = read_seqbegin(&fs_info->profiles_lock);
1738
1739                 all_avail = fs_info->avail_data_alloc_bits |
1740                             fs_info->avail_system_alloc_bits |
1741                             fs_info->avail_metadata_alloc_bits;
1742         } while (read_seqretry(&fs_info->profiles_lock, seq));
1743
1744         for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
1745                 if (!(all_avail & btrfs_raid_group[i]))
1746                         continue;
1747
1748                 if (num_devices < btrfs_raid_array[i].devs_min) {
1749                         int ret = btrfs_raid_mindev_error[i];
1750
1751                         if (ret)
1752                                 return ret;
1753                 }
1754         }
1755
1756         return 0;
1757 }
1758
1759 struct btrfs_device *btrfs_find_next_active_device(struct btrfs_fs_devices *fs_devs,
1760                                         struct btrfs_device *device)
1761 {
1762         struct btrfs_device *next_device;
1763
1764         list_for_each_entry(next_device, &fs_devs->devices, dev_list) {
1765                 if (next_device != device &&
1766                         !next_device->missing && next_device->bdev)
1767                         return next_device;
1768         }
1769
1770         return NULL;
1771 }
1772
1773 /*
1774  * Helper function to check if the given device is part of s_bdev / latest_bdev
1775  * and replace it with the provided or the next active device, in the context
1776  * where this function called, there should be always be another device (or
1777  * this_dev) which is active.
1778  */
1779 void btrfs_assign_next_active_device(struct btrfs_fs_info *fs_info,
1780                 struct btrfs_device *device, struct btrfs_device *this_dev)
1781 {
1782         struct btrfs_device *next_device;
1783
1784         if (this_dev)
1785                 next_device = this_dev;
1786         else
1787                 next_device = btrfs_find_next_active_device(fs_info->fs_devices,
1788                                                                 device);
1789         ASSERT(next_device);
1790
1791         if (fs_info->sb->s_bdev &&
1792                         (fs_info->sb->s_bdev == device->bdev))
1793                 fs_info->sb->s_bdev = next_device->bdev;
1794
1795         if (fs_info->fs_devices->latest_bdev == device->bdev)
1796                 fs_info->fs_devices->latest_bdev = next_device->bdev;
1797 }
1798
1799 int btrfs_rm_device(struct btrfs_root *root, char *device_path, u64 devid)
1800 {
1801         struct btrfs_device *device;
1802         struct btrfs_fs_devices *cur_devices;
1803         u64 num_devices;
1804         int ret = 0;
1805         bool clear_super = false;
1806         char *dev_name = NULL;
1807
1808         mutex_lock(&uuid_mutex);
1809
1810         num_devices = root->fs_info->fs_devices->num_devices;
1811         btrfs_dev_replace_lock(&root->fs_info->dev_replace, 0);
1812         if (btrfs_dev_replace_is_ongoing(&root->fs_info->dev_replace)) {
1813                 WARN_ON(num_devices < 1);
1814                 num_devices--;
1815         }
1816         btrfs_dev_replace_unlock(&root->fs_info->dev_replace, 0);
1817
1818         ret = btrfs_check_raid_min_devices(root->fs_info, num_devices - 1);
1819         if (ret)
1820                 goto out;
1821
1822         ret = btrfs_find_device_by_devspec(root, devid, device_path,
1823                                 &device);
1824         if (ret)
1825                 goto out;
1826
1827         if (device->is_tgtdev_for_dev_replace) {
1828                 ret = BTRFS_ERROR_DEV_TGT_REPLACE;
1829                 goto out;
1830         }
1831
1832         if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1833                 ret = BTRFS_ERROR_DEV_ONLY_WRITABLE;
1834                 goto out;
1835         }
1836
1837         if (device->writeable) {
1838                 lock_chunks(root);
1839                 list_del_init(&device->dev_alloc_list);
1840                 device->fs_devices->rw_devices--;
1841                 unlock_chunks(root);
1842                 dev_name = kstrdup(device->name->str, GFP_KERNEL);
1843                 if (!dev_name) {
1844                         ret = -ENOMEM;
1845                         goto error_undo;
1846                 }
1847                 clear_super = true;
1848         }
1849
1850         mutex_unlock(&uuid_mutex);
1851         ret = btrfs_shrink_device(device, 0);
1852         mutex_lock(&uuid_mutex);
1853         if (ret)
1854                 goto error_undo;
1855
1856         /*
1857          * TODO: the superblock still includes this device in its num_devices
1858          * counter although write_all_supers() is not locked out. This
1859          * could give a filesystem state which requires a degraded mount.
1860          */
1861         ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1862         if (ret)
1863                 goto error_undo;
1864
1865         device->in_fs_metadata = 0;
1866         btrfs_scrub_cancel_dev(root->fs_info, device);
1867
1868         /*
1869          * the device list mutex makes sure that we don't change
1870          * the device list while someone else is writing out all
1871          * the device supers. Whoever is writing all supers, should
1872          * lock the device list mutex before getting the number of
1873          * devices in the super block (super_copy). Conversely,
1874          * whoever updates the number of devices in the super block
1875          * (super_copy) should hold the device list mutex.
1876          */
1877
1878         cur_devices = device->fs_devices;
1879         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1880         list_del_rcu(&device->dev_list);
1881
1882         device->fs_devices->num_devices--;
1883         device->fs_devices->total_devices--;
1884
1885         if (device->missing)
1886                 device->fs_devices->missing_devices--;
1887
1888         btrfs_assign_next_active_device(root->fs_info, device, NULL);
1889
1890         if (device->bdev) {
1891                 device->fs_devices->open_devices--;
1892                 /* remove sysfs entry */
1893                 btrfs_sysfs_rm_device_link(root->fs_info->fs_devices, device);
1894         }
1895
1896         call_rcu(&device->rcu, free_device);
1897
1898         num_devices = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
1899         btrfs_set_super_num_devices(root->fs_info->super_copy, num_devices);
1900         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1901
1902         if (cur_devices->open_devices == 0) {
1903                 struct btrfs_fs_devices *fs_devices;
1904                 fs_devices = root->fs_info->fs_devices;
1905                 while (fs_devices) {
1906                         if (fs_devices->seed == cur_devices) {
1907                                 fs_devices->seed = cur_devices->seed;
1908                                 break;
1909                         }
1910                         fs_devices = fs_devices->seed;
1911                 }
1912                 cur_devices->seed = NULL;
1913                 __btrfs_close_devices(cur_devices);
1914                 free_fs_devices(cur_devices);
1915         }
1916
1917         root->fs_info->num_tolerated_disk_barrier_failures =
1918                 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
1919
1920         /*
1921          * at this point, the device is zero sized.  We want to
1922          * remove it from the devices list and zero out the old super
1923          */
1924         if (clear_super) {
1925                 struct block_device *bdev;
1926
1927                 bdev = blkdev_get_by_path(dev_name, FMODE_READ | FMODE_EXCL,
1928                                                 root->fs_info->bdev_holder);
1929                 if (!IS_ERR(bdev)) {
1930                         btrfs_scratch_superblocks(bdev, dev_name);
1931                         blkdev_put(bdev, FMODE_READ | FMODE_EXCL);
1932                 }
1933         }
1934
1935 out:
1936         kfree(dev_name);
1937
1938         mutex_unlock(&uuid_mutex);
1939         return ret;
1940
1941 error_undo:
1942         if (device->writeable) {
1943                 lock_chunks(root);
1944                 list_add(&device->dev_alloc_list,
1945                          &root->fs_info->fs_devices->alloc_list);
1946                 device->fs_devices->rw_devices++;
1947                 unlock_chunks(root);
1948         }
1949         goto out;
1950 }
1951
1952 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_fs_info *fs_info,
1953                                         struct btrfs_device *srcdev)
1954 {
1955         struct btrfs_fs_devices *fs_devices;
1956
1957         WARN_ON(!mutex_is_locked(&fs_info->fs_devices->device_list_mutex));
1958
1959         /*
1960          * in case of fs with no seed, srcdev->fs_devices will point
1961          * to fs_devices of fs_info. However when the dev being replaced is
1962          * a seed dev it will point to the seed's local fs_devices. In short
1963          * srcdev will have its correct fs_devices in both the cases.
1964          */
1965         fs_devices = srcdev->fs_devices;
1966
1967         list_del_rcu(&srcdev->dev_list);
1968         list_del_rcu(&srcdev->dev_alloc_list);
1969         fs_devices->num_devices--;
1970         if (srcdev->missing)
1971                 fs_devices->missing_devices--;
1972
1973         if (srcdev->writeable)
1974                 fs_devices->rw_devices--;
1975
1976         if (srcdev->bdev)
1977                 fs_devices->open_devices--;
1978 }
1979
1980 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_fs_info *fs_info,
1981                                       struct btrfs_device *srcdev)
1982 {
1983         struct btrfs_fs_devices *fs_devices = srcdev->fs_devices;
1984
1985         if (srcdev->writeable) {
1986                 /* zero out the old super if it is writable */
1987                 btrfs_scratch_superblocks(srcdev->bdev, srcdev->name->str);
1988         }
1989         call_rcu(&srcdev->rcu, free_device);
1990
1991         /*
1992          * unless fs_devices is seed fs, num_devices shouldn't go
1993          * zero
1994          */
1995         BUG_ON(!fs_devices->num_devices && !fs_devices->seeding);
1996
1997         /* if this is no devs we rather delete the fs_devices */
1998         if (!fs_devices->num_devices) {
1999                 struct btrfs_fs_devices *tmp_fs_devices;
2000
2001                 tmp_fs_devices = fs_info->fs_devices;
2002                 while (tmp_fs_devices) {
2003                         if (tmp_fs_devices->seed == fs_devices) {
2004                                 tmp_fs_devices->seed = fs_devices->seed;
2005                                 break;
2006                         }
2007                         tmp_fs_devices = tmp_fs_devices->seed;
2008                 }
2009                 fs_devices->seed = NULL;
2010                 __btrfs_close_devices(fs_devices);
2011                 free_fs_devices(fs_devices);
2012         }
2013 }
2014
2015 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
2016                                       struct btrfs_device *tgtdev)
2017 {
2018         mutex_lock(&uuid_mutex);
2019         WARN_ON(!tgtdev);
2020         mutex_lock(&fs_info->fs_devices->device_list_mutex);
2021
2022         btrfs_sysfs_rm_device_link(fs_info->fs_devices, tgtdev);
2023
2024         if (tgtdev->bdev)
2025                 fs_info->fs_devices->open_devices--;
2026
2027         fs_info->fs_devices->num_devices--;
2028
2029         btrfs_assign_next_active_device(fs_info, tgtdev, NULL);
2030
2031         list_del_rcu(&tgtdev->dev_list);
2032
2033         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2034         mutex_unlock(&uuid_mutex);
2035
2036         /*
2037          * The update_dev_time() with in btrfs_scratch_superblocks()
2038          * may lead to a call to btrfs_show_devname() which will try
2039          * to hold device_list_mutex. And here this device
2040          * is already out of device list, so we don't have to hold
2041          * the device_list_mutex lock.
2042          */
2043         btrfs_scratch_superblocks(tgtdev->bdev, tgtdev->name->str);
2044         call_rcu(&tgtdev->rcu, free_device);
2045 }
2046
2047 static int btrfs_find_device_by_path(struct btrfs_root *root, char *device_path,
2048                                      struct btrfs_device **device)
2049 {
2050         int ret = 0;
2051         struct btrfs_super_block *disk_super;
2052         u64 devid;
2053         u8 *dev_uuid;
2054         struct block_device *bdev;
2055         struct buffer_head *bh;
2056
2057         *device = NULL;
2058         ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
2059                                     root->fs_info->bdev_holder, 0, &bdev, &bh);
2060         if (ret)
2061                 return ret;
2062         disk_super = (struct btrfs_super_block *)bh->b_data;
2063         devid = btrfs_stack_device_id(&disk_super->dev_item);
2064         dev_uuid = disk_super->dev_item.uuid;
2065         *device = btrfs_find_device(root->fs_info, devid, dev_uuid,
2066                                     disk_super->fsid);
2067         brelse(bh);
2068         if (!*device)
2069                 ret = -ENOENT;
2070         blkdev_put(bdev, FMODE_READ);
2071         return ret;
2072 }
2073
2074 int btrfs_find_device_missing_or_by_path(struct btrfs_root *root,
2075                                          char *device_path,
2076                                          struct btrfs_device **device)
2077 {
2078         *device = NULL;
2079         if (strcmp(device_path, "missing") == 0) {
2080                 struct list_head *devices;
2081                 struct btrfs_device *tmp;
2082
2083                 devices = &root->fs_info->fs_devices->devices;
2084                 /*
2085                  * It is safe to read the devices since the volume_mutex
2086                  * is held by the caller.
2087                  */
2088                 list_for_each_entry(tmp, devices, dev_list) {
2089                         if (tmp->in_fs_metadata && !tmp->bdev) {
2090                                 *device = tmp;
2091                                 break;
2092                         }
2093                 }
2094
2095                 if (!*device)
2096                         return BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
2097
2098                 return 0;
2099         } else {
2100                 return btrfs_find_device_by_path(root, device_path, device);
2101         }
2102 }
2103
2104 /*
2105  * Lookup a device given by device id, or the path if the id is 0.
2106  */
2107 int btrfs_find_device_by_devspec(struct btrfs_root *root, u64 devid,
2108                                          char *devpath,
2109                                          struct btrfs_device **device)
2110 {
2111         int ret;
2112
2113         if (devid) {
2114                 ret = 0;
2115                 *device = btrfs_find_device(root->fs_info, devid, NULL,
2116                                             NULL);
2117                 if (!*device)
2118                         ret = -ENOENT;
2119         } else {
2120                 if (!devpath || !devpath[0])
2121                         return -EINVAL;
2122
2123                 ret = btrfs_find_device_missing_or_by_path(root, devpath,
2124                                                            device);
2125         }
2126         return ret;
2127 }
2128
2129 /*
2130  * does all the dirty work required for changing file system's UUID.
2131  */
2132 static int btrfs_prepare_sprout(struct btrfs_root *root)
2133 {
2134         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2135         struct btrfs_fs_devices *old_devices;
2136         struct btrfs_fs_devices *seed_devices;
2137         struct btrfs_super_block *disk_super = root->fs_info->super_copy;
2138         struct btrfs_device *device;
2139         u64 super_flags;
2140
2141         BUG_ON(!mutex_is_locked(&uuid_mutex));
2142         if (!fs_devices->seeding)
2143                 return -EINVAL;
2144
2145         seed_devices = __alloc_fs_devices();
2146         if (IS_ERR(seed_devices))
2147                 return PTR_ERR(seed_devices);
2148
2149         old_devices = clone_fs_devices(fs_devices);
2150         if (IS_ERR(old_devices)) {
2151                 kfree(seed_devices);
2152                 return PTR_ERR(old_devices);
2153         }
2154
2155         list_add(&old_devices->list, &fs_uuids);
2156
2157         memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
2158         seed_devices->opened = 1;
2159         INIT_LIST_HEAD(&seed_devices->devices);
2160         INIT_LIST_HEAD(&seed_devices->alloc_list);
2161         mutex_init(&seed_devices->device_list_mutex);
2162
2163         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2164         list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
2165                               synchronize_rcu);
2166         list_for_each_entry(device, &seed_devices->devices, dev_list)
2167                 device->fs_devices = seed_devices;
2168
2169         lock_chunks(root);
2170         list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
2171         unlock_chunks(root);
2172
2173         fs_devices->seeding = 0;
2174         fs_devices->num_devices = 0;
2175         fs_devices->open_devices = 0;
2176         fs_devices->missing_devices = 0;
2177         fs_devices->rotating = 0;
2178         fs_devices->seed = seed_devices;
2179
2180         generate_random_uuid(fs_devices->fsid);
2181         memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2182         memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2183         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2184
2185         super_flags = btrfs_super_flags(disk_super) &
2186                       ~BTRFS_SUPER_FLAG_SEEDING;
2187         btrfs_set_super_flags(disk_super, super_flags);
2188
2189         return 0;
2190 }
2191
2192 /*
2193  * Store the expected generation for seed devices in device items.
2194  */
2195 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
2196                                struct btrfs_root *root)
2197 {
2198         struct btrfs_path *path;
2199         struct extent_buffer *leaf;
2200         struct btrfs_dev_item *dev_item;
2201         struct btrfs_device *device;
2202         struct btrfs_key key;
2203         u8 fs_uuid[BTRFS_UUID_SIZE];
2204         u8 dev_uuid[BTRFS_UUID_SIZE];
2205         u64 devid;
2206         int ret;
2207
2208         path = btrfs_alloc_path();
2209         if (!path)
2210                 return -ENOMEM;
2211
2212         root = root->fs_info->chunk_root;
2213         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2214         key.offset = 0;
2215         key.type = BTRFS_DEV_ITEM_KEY;
2216
2217         while (1) {
2218                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2219                 if (ret < 0)
2220                         goto error;
2221
2222                 leaf = path->nodes[0];
2223 next_slot:
2224                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2225                         ret = btrfs_next_leaf(root, path);
2226                         if (ret > 0)
2227                                 break;
2228                         if (ret < 0)
2229                                 goto error;
2230                         leaf = path->nodes[0];
2231                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2232                         btrfs_release_path(path);
2233                         continue;
2234                 }
2235
2236                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2237                 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
2238                     key.type != BTRFS_DEV_ITEM_KEY)
2239                         break;
2240
2241                 dev_item = btrfs_item_ptr(leaf, path->slots[0],
2242                                           struct btrfs_dev_item);
2243                 devid = btrfs_device_id(leaf, dev_item);
2244                 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
2245                                    BTRFS_UUID_SIZE);
2246                 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
2247                                    BTRFS_UUID_SIZE);
2248                 device = btrfs_find_device(root->fs_info, devid, dev_uuid,
2249                                            fs_uuid);
2250                 BUG_ON(!device); /* Logic error */
2251
2252                 if (device->fs_devices->seeding) {
2253                         btrfs_set_device_generation(leaf, dev_item,
2254                                                     device->generation);
2255                         btrfs_mark_buffer_dirty(leaf);
2256                 }
2257
2258                 path->slots[0]++;
2259                 goto next_slot;
2260         }
2261         ret = 0;
2262 error:
2263         btrfs_free_path(path);
2264         return ret;
2265 }
2266
2267 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
2268 {
2269         struct request_queue *q;
2270         struct btrfs_trans_handle *trans;
2271         struct btrfs_device *device;
2272         struct block_device *bdev;
2273         struct list_head *devices;
2274         struct super_block *sb = root->fs_info->sb;
2275         struct rcu_string *name;
2276         u64 tmp;
2277         int seeding_dev = 0;
2278         int ret = 0;
2279
2280         if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
2281                 return -EROFS;
2282
2283         bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2284                                   root->fs_info->bdev_holder);
2285         if (IS_ERR(bdev))
2286                 return PTR_ERR(bdev);
2287
2288         if (root->fs_info->fs_devices->seeding) {
2289                 seeding_dev = 1;
2290                 down_write(&sb->s_umount);
2291                 mutex_lock(&uuid_mutex);
2292         }
2293
2294         filemap_write_and_wait(bdev->bd_inode->i_mapping);
2295
2296         devices = &root->fs_info->fs_devices->devices;
2297
2298         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2299         list_for_each_entry(device, devices, dev_list) {
2300                 if (device->bdev == bdev) {
2301                         ret = -EEXIST;
2302                         mutex_unlock(
2303                                 &root->fs_info->fs_devices->device_list_mutex);
2304                         goto error;
2305                 }
2306         }
2307         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2308
2309         device = btrfs_alloc_device(root->fs_info, NULL, NULL);
2310         if (IS_ERR(device)) {
2311                 /* we can safely leave the fs_devices entry around */
2312                 ret = PTR_ERR(device);
2313                 goto error;
2314         }
2315
2316         name = rcu_string_strdup(device_path, GFP_KERNEL);
2317         if (!name) {
2318                 kfree(device);
2319                 ret = -ENOMEM;
2320                 goto error;
2321         }
2322         rcu_assign_pointer(device->name, name);
2323
2324         trans = btrfs_start_transaction(root, 0);
2325         if (IS_ERR(trans)) {
2326                 rcu_string_free(device->name);
2327                 kfree(device);
2328                 ret = PTR_ERR(trans);
2329                 goto error;
2330         }
2331
2332         q = bdev_get_queue(bdev);
2333         if (blk_queue_discard(q))
2334                 device->can_discard = 1;
2335         device->writeable = 1;
2336         device->generation = trans->transid;
2337         device->io_width = root->sectorsize;
2338         device->io_align = root->sectorsize;
2339         device->sector_size = root->sectorsize;
2340         device->total_bytes = i_size_read(bdev->bd_inode);
2341         device->disk_total_bytes = device->total_bytes;
2342         device->commit_total_bytes = device->total_bytes;
2343         device->dev_root = root->fs_info->dev_root;
2344         device->bdev = bdev;
2345         device->in_fs_metadata = 1;
2346         device->is_tgtdev_for_dev_replace = 0;
2347         device->mode = FMODE_EXCL;
2348         device->dev_stats_valid = 1;
2349         set_blocksize(device->bdev, 4096);
2350
2351         if (seeding_dev) {
2352                 sb->s_flags &= ~MS_RDONLY;
2353                 ret = btrfs_prepare_sprout(root);
2354                 BUG_ON(ret); /* -ENOMEM */
2355         }
2356
2357         device->fs_devices = root->fs_info->fs_devices;
2358
2359         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2360         lock_chunks(root);
2361         list_add_rcu(&device->dev_list, &root->fs_info->fs_devices->devices);
2362         list_add(&device->dev_alloc_list,
2363                  &root->fs_info->fs_devices->alloc_list);
2364         root->fs_info->fs_devices->num_devices++;
2365         root->fs_info->fs_devices->open_devices++;
2366         root->fs_info->fs_devices->rw_devices++;
2367         root->fs_info->fs_devices->total_devices++;
2368         root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
2369
2370         spin_lock(&root->fs_info->free_chunk_lock);
2371         root->fs_info->free_chunk_space += device->total_bytes;
2372         spin_unlock(&root->fs_info->free_chunk_lock);
2373
2374         if (!blk_queue_nonrot(bdev_get_queue(bdev)))
2375                 root->fs_info->fs_devices->rotating = 1;
2376
2377         tmp = btrfs_super_total_bytes(root->fs_info->super_copy);
2378         btrfs_set_super_total_bytes(root->fs_info->super_copy,
2379                                     tmp + device->total_bytes);
2380
2381         tmp = btrfs_super_num_devices(root->fs_info->super_copy);
2382         btrfs_set_super_num_devices(root->fs_info->super_copy,
2383                                     tmp + 1);
2384
2385         /* add sysfs device entry */
2386         btrfs_sysfs_add_device_link(root->fs_info->fs_devices, device);
2387
2388         /*
2389          * we've got more storage, clear any full flags on the space
2390          * infos
2391          */
2392         btrfs_clear_space_info_full(root->fs_info);
2393
2394         unlock_chunks(root);
2395         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2396
2397         if (seeding_dev) {
2398                 lock_chunks(root);
2399                 ret = init_first_rw_device(trans, root, device);
2400                 unlock_chunks(root);
2401                 if (ret) {
2402                         btrfs_abort_transaction(trans, root, ret);
2403                         goto error_trans;
2404                 }
2405         }
2406
2407         ret = btrfs_add_device(trans, root, device);
2408         if (ret) {
2409                 btrfs_abort_transaction(trans, root, ret);
2410                 goto error_trans;
2411         }
2412
2413         if (seeding_dev) {
2414                 char fsid_buf[BTRFS_UUID_UNPARSED_SIZE];
2415
2416                 ret = btrfs_finish_sprout(trans, root);
2417                 if (ret) {
2418                         btrfs_abort_transaction(trans, root, ret);
2419                         goto error_trans;
2420                 }
2421
2422                 /* Sprouting would change fsid of the mounted root,
2423                  * so rename the fsid on the sysfs
2424                  */
2425                 snprintf(fsid_buf, BTRFS_UUID_UNPARSED_SIZE, "%pU",
2426                                                 root->fs_info->fsid);
2427                 if (kobject_rename(&root->fs_info->fs_devices->fsid_kobj,
2428                                                                 fsid_buf))
2429                         btrfs_warn(root->fs_info,
2430                                 "sysfs: failed to create fsid for sprout");
2431         }
2432
2433         root->fs_info->num_tolerated_disk_barrier_failures =
2434                 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
2435         ret = btrfs_commit_transaction(trans, root);
2436
2437         if (seeding_dev) {
2438                 mutex_unlock(&uuid_mutex);
2439                 up_write(&sb->s_umount);
2440
2441                 if (ret) /* transaction commit */
2442                         return ret;
2443
2444                 ret = btrfs_relocate_sys_chunks(root);
2445                 if (ret < 0)
2446                         btrfs_handle_fs_error(root->fs_info, ret,
2447                                     "Failed to relocate sys chunks after "
2448                                     "device initialization. This can be fixed "
2449                                     "using the \"btrfs balance\" command.");
2450                 trans = btrfs_attach_transaction(root);
2451                 if (IS_ERR(trans)) {
2452                         if (PTR_ERR(trans) == -ENOENT)
2453                                 return 0;
2454                         return PTR_ERR(trans);
2455                 }
2456                 ret = btrfs_commit_transaction(trans, root);
2457         }
2458
2459         /* Update ctime/mtime for libblkid */
2460         update_dev_time(device_path);
2461         return ret;
2462
2463 error_trans:
2464         btrfs_end_transaction(trans, root);
2465         rcu_string_free(device->name);
2466         btrfs_sysfs_rm_device_link(root->fs_info->fs_devices, device);
2467         kfree(device);
2468 error:
2469         blkdev_put(bdev, FMODE_EXCL);
2470         if (seeding_dev) {
2471                 mutex_unlock(&uuid_mutex);
2472                 up_write(&sb->s_umount);
2473         }
2474         return ret;
2475 }
2476
2477 int btrfs_init_dev_replace_tgtdev(struct btrfs_root *root, char *device_path,
2478                                   struct btrfs_device *srcdev,
2479                                   struct btrfs_device **device_out)
2480 {
2481         struct request_queue *q;
2482         struct btrfs_device *device;
2483         struct block_device *bdev;
2484         struct btrfs_fs_info *fs_info = root->fs_info;
2485         struct list_head *devices;
2486         struct rcu_string *name;
2487         u64 devid = BTRFS_DEV_REPLACE_DEVID;
2488         int ret = 0;
2489
2490         *device_out = NULL;
2491         if (fs_info->fs_devices->seeding) {
2492                 btrfs_err(fs_info, "the filesystem is a seed filesystem!");
2493                 return -EINVAL;
2494         }
2495
2496         bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2497                                   fs_info->bdev_holder);
2498         if (IS_ERR(bdev)) {
2499                 btrfs_err(fs_info, "target device %s is invalid!", device_path);
2500                 return PTR_ERR(bdev);
2501         }
2502
2503         filemap_write_and_wait(bdev->bd_inode->i_mapping);
2504
2505         devices = &fs_info->fs_devices->devices;
2506         list_for_each_entry(device, devices, dev_list) {
2507                 if (device->bdev == bdev) {
2508                         btrfs_err(fs_info, "target device is in the filesystem!");
2509                         ret = -EEXIST;
2510                         goto error;
2511                 }
2512         }
2513
2514
2515         if (i_size_read(bdev->bd_inode) <
2516             btrfs_device_get_total_bytes(srcdev)) {
2517                 btrfs_err(fs_info, "target device is smaller than source device!");
2518                 ret = -EINVAL;
2519                 goto error;
2520         }
2521
2522
2523         device = btrfs_alloc_device(NULL, &devid, NULL);
2524         if (IS_ERR(device)) {
2525                 ret = PTR_ERR(device);
2526                 goto error;
2527         }
2528
2529         name = rcu_string_strdup(device_path, GFP_NOFS);
2530         if (!name) {
2531                 kfree(device);
2532                 ret = -ENOMEM;
2533                 goto error;
2534         }
2535         rcu_assign_pointer(device->name, name);
2536
2537         q = bdev_get_queue(bdev);
2538         if (blk_queue_discard(q))
2539                 device->can_discard = 1;
2540         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2541         device->writeable = 1;
2542         device->generation = 0;
2543         device->io_width = root->sectorsize;
2544         device->io_align = root->sectorsize;
2545         device->sector_size = root->sectorsize;
2546         device->total_bytes = btrfs_device_get_total_bytes(srcdev);
2547         device->disk_total_bytes = btrfs_device_get_disk_total_bytes(srcdev);
2548         device->bytes_used = btrfs_device_get_bytes_used(srcdev);
2549         ASSERT(list_empty(&srcdev->resized_list));
2550         device->commit_total_bytes = srcdev->commit_total_bytes;
2551         device->commit_bytes_used = device->bytes_used;
2552         device->dev_root = fs_info->dev_root;
2553         device->bdev = bdev;
2554         device->in_fs_metadata = 1;
2555         device->is_tgtdev_for_dev_replace = 1;
2556         device->mode = FMODE_EXCL;
2557         device->dev_stats_valid = 1;
2558         set_blocksize(device->bdev, 4096);
2559         device->fs_devices = fs_info->fs_devices;
2560         list_add(&device->dev_list, &fs_info->fs_devices->devices);
2561         fs_info->fs_devices->num_devices++;
2562         fs_info->fs_devices->open_devices++;
2563         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2564
2565         *device_out = device;
2566         return ret;
2567
2568 error:
2569         blkdev_put(bdev, FMODE_EXCL);
2570         return ret;
2571 }
2572
2573 void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info *fs_info,
2574                                               struct btrfs_device *tgtdev)
2575 {
2576         WARN_ON(fs_info->fs_devices->rw_devices == 0);
2577         tgtdev->io_width = fs_info->dev_root->sectorsize;
2578         tgtdev->io_align = fs_info->dev_root->sectorsize;
2579         tgtdev->sector_size = fs_info->dev_root->sectorsize;
2580         tgtdev->dev_root = fs_info->dev_root;
2581         tgtdev->in_fs_metadata = 1;
2582 }
2583
2584 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2585                                         struct btrfs_device *device)
2586 {
2587         int ret;
2588         struct btrfs_path *path;
2589         struct btrfs_root *root;
2590         struct btrfs_dev_item *dev_item;
2591         struct extent_buffer *leaf;
2592         struct btrfs_key key;
2593
2594         root = device->dev_root->fs_info->chunk_root;
2595
2596         path = btrfs_alloc_path();
2597         if (!path)
2598                 return -ENOMEM;
2599
2600         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2601         key.type = BTRFS_DEV_ITEM_KEY;
2602         key.offset = device->devid;
2603
2604         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2605         if (ret < 0)
2606                 goto out;
2607
2608         if (ret > 0) {
2609                 ret = -ENOENT;
2610                 goto out;
2611         }
2612
2613         leaf = path->nodes[0];
2614         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2615
2616         btrfs_set_device_id(leaf, dev_item, device->devid);
2617         btrfs_set_device_type(leaf, dev_item, device->type);
2618         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2619         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2620         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2621         btrfs_set_device_total_bytes(leaf, dev_item,
2622                                      btrfs_device_get_disk_total_bytes(device));
2623         btrfs_set_device_bytes_used(leaf, dev_item,
2624                                     btrfs_device_get_bytes_used(device));
2625         btrfs_mark_buffer_dirty(leaf);
2626
2627 out:
2628         btrfs_free_path(path);
2629         return ret;
2630 }
2631
2632 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2633                       struct btrfs_device *device, u64 new_size)
2634 {
2635         struct btrfs_super_block *super_copy =
2636                 device->dev_root->fs_info->super_copy;
2637         struct btrfs_fs_devices *fs_devices;
2638         u64 old_total;
2639         u64 diff;
2640
2641         if (!device->writeable)
2642                 return -EACCES;
2643
2644         lock_chunks(device->dev_root);
2645         old_total = btrfs_super_total_bytes(super_copy);
2646         diff = new_size - device->total_bytes;
2647
2648         if (new_size <= device->total_bytes ||
2649             device->is_tgtdev_for_dev_replace) {
2650                 unlock_chunks(device->dev_root);
2651                 return -EINVAL;
2652         }
2653
2654         fs_devices = device->dev_root->fs_info->fs_devices;
2655
2656         btrfs_set_super_total_bytes(super_copy, old_total + diff);
2657         device->fs_devices->total_rw_bytes += diff;
2658
2659         btrfs_device_set_total_bytes(device, new_size);
2660         btrfs_device_set_disk_total_bytes(device, new_size);
2661         btrfs_clear_space_info_full(device->dev_root->fs_info);
2662         if (list_empty(&device->resized_list))
2663                 list_add_tail(&device->resized_list,
2664                               &fs_devices->resized_devices);
2665         unlock_chunks(device->dev_root);
2666
2667         return btrfs_update_device(trans, device);
2668 }
2669
2670 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
2671                             struct btrfs_root *root, u64 chunk_objectid,
2672                             u64 chunk_offset)
2673 {
2674         int ret;
2675         struct btrfs_path *path;
2676         struct btrfs_key key;
2677
2678         root = root->fs_info->chunk_root;
2679         path = btrfs_alloc_path();
2680         if (!path)
2681                 return -ENOMEM;
2682
2683         key.objectid = chunk_objectid;
2684         key.offset = chunk_offset;
2685         key.type = BTRFS_CHUNK_ITEM_KEY;
2686
2687         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2688         if (ret < 0)
2689                 goto out;
2690         else if (ret > 0) { /* Logic error or corruption */
2691                 btrfs_handle_fs_error(root->fs_info, -ENOENT,
2692                             "Failed lookup while freeing chunk.");
2693                 ret = -ENOENT;
2694                 goto out;
2695         }
2696
2697         ret = btrfs_del_item(trans, root, path);
2698         if (ret < 0)
2699                 btrfs_handle_fs_error(root->fs_info, ret,
2700                             "Failed to delete chunk item.");
2701 out:
2702         btrfs_free_path(path);
2703         return ret;
2704 }
2705
2706 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
2707                         chunk_offset)
2708 {
2709         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
2710         struct btrfs_disk_key *disk_key;
2711         struct btrfs_chunk *chunk;
2712         u8 *ptr;
2713         int ret = 0;
2714         u32 num_stripes;
2715         u32 array_size;
2716         u32 len = 0;
2717         u32 cur;
2718         struct btrfs_key key;
2719
2720         lock_chunks(root);
2721         array_size = btrfs_super_sys_array_size(super_copy);
2722
2723         ptr = super_copy->sys_chunk_array;
2724         cur = 0;
2725
2726         while (cur < array_size) {
2727                 disk_key = (struct btrfs_disk_key *)ptr;
2728                 btrfs_disk_key_to_cpu(&key, disk_key);
2729
2730                 len = sizeof(*disk_key);
2731
2732                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2733                         chunk = (struct btrfs_chunk *)(ptr + len);
2734                         num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2735                         len += btrfs_chunk_item_size(num_stripes);
2736                 } else {
2737                         ret = -EIO;
2738                         break;
2739                 }
2740                 if (key.objectid == chunk_objectid &&
2741                     key.offset == chunk_offset) {
2742                         memmove(ptr, ptr + len, array_size - (cur + len));
2743                         array_size -= len;
2744                         btrfs_set_super_sys_array_size(super_copy, array_size);
2745                 } else {
2746                         ptr += len;
2747                         cur += len;
2748                 }
2749         }
2750         unlock_chunks(root);
2751         return ret;
2752 }
2753
2754 int btrfs_remove_chunk(struct btrfs_trans_handle *trans,
2755                        struct btrfs_root *root, u64 chunk_offset)
2756 {
2757         struct extent_map_tree *em_tree;
2758         struct extent_map *em;
2759         struct btrfs_root *extent_root = root->fs_info->extent_root;
2760         struct map_lookup *map;
2761         u64 dev_extent_len = 0;
2762         u64 chunk_objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2763         int i, ret = 0;
2764         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2765
2766         /* Just in case */
2767         root = root->fs_info->chunk_root;
2768         em_tree = &root->fs_info->mapping_tree.map_tree;
2769
2770         read_lock(&em_tree->lock);
2771         em = lookup_extent_mapping(em_tree, chunk_offset, 1);
2772         read_unlock(&em_tree->lock);
2773
2774         if (!em || em->start > chunk_offset ||
2775             em->start + em->len < chunk_offset) {
2776                 /*
2777                  * This is a logic error, but we don't want to just rely on the
2778                  * user having built with ASSERT enabled, so if ASSERT doesn't
2779                  * do anything we still error out.
2780                  */
2781                 ASSERT(0);
2782                 if (em)
2783                         free_extent_map(em);
2784                 return -EINVAL;
2785         }
2786         map = em->map_lookup;
2787         lock_chunks(root->fs_info->chunk_root);
2788         check_system_chunk(trans, extent_root, map->type);
2789         unlock_chunks(root->fs_info->chunk_root);
2790
2791         /*
2792          * Take the device list mutex to prevent races with the final phase of
2793          * a device replace operation that replaces the device object associated
2794          * with map stripes (dev-replace.c:btrfs_dev_replace_finishing()).
2795          */
2796         mutex_lock(&fs_devices->device_list_mutex);
2797         for (i = 0; i < map->num_stripes; i++) {
2798                 struct btrfs_device *device = map->stripes[i].dev;
2799                 ret = btrfs_free_dev_extent(trans, device,
2800                                             map->stripes[i].physical,
2801                                             &dev_extent_len);
2802                 if (ret) {
2803                         mutex_unlock(&fs_devices->device_list_mutex);
2804                         btrfs_abort_transaction(trans, root, ret);
2805                         goto out;
2806                 }
2807
2808                 if (device->bytes_used > 0) {
2809                         lock_chunks(root);
2810                         btrfs_device_set_bytes_used(device,
2811                                         device->bytes_used - dev_extent_len);
2812                         spin_lock(&root->fs_info->free_chunk_lock);
2813                         root->fs_info->free_chunk_space += dev_extent_len;
2814                         spin_unlock(&root->fs_info->free_chunk_lock);
2815                         btrfs_clear_space_info_full(root->fs_info);
2816                         unlock_chunks(root);
2817                 }
2818
2819                 if (map->stripes[i].dev) {
2820                         ret = btrfs_update_device(trans, map->stripes[i].dev);
2821                         if (ret) {
2822                                 mutex_unlock(&fs_devices->device_list_mutex);
2823                                 btrfs_abort_transaction(trans, root, ret);
2824                                 goto out;
2825                         }
2826                 }
2827         }
2828         mutex_unlock(&fs_devices->device_list_mutex);
2829
2830         ret = btrfs_free_chunk(trans, root, chunk_objectid, chunk_offset);
2831         if (ret) {
2832                 btrfs_abort_transaction(trans, root, ret);
2833                 goto out;
2834         }
2835
2836         trace_btrfs_chunk_free(root, map, chunk_offset, em->len);
2837
2838         if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2839                 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
2840                 if (ret) {
2841                         btrfs_abort_transaction(trans, root, ret);
2842                         goto out;
2843                 }
2844         }
2845
2846         ret = btrfs_remove_block_group(trans, extent_root, chunk_offset, em);
2847         if (ret) {
2848                 btrfs_abort_transaction(trans, extent_root, ret);
2849                 goto out;
2850         }
2851
2852 out:
2853         /* once for us */
2854         free_extent_map(em);
2855         return ret;
2856 }
2857
2858 static int btrfs_relocate_chunk(struct btrfs_root *root, u64 chunk_offset)
2859 {
2860         struct btrfs_root *extent_root;
2861         struct btrfs_trans_handle *trans;
2862         int ret;
2863
2864         root = root->fs_info->chunk_root;
2865         extent_root = root->fs_info->extent_root;
2866
2867         /*
2868          * Prevent races with automatic removal of unused block groups.
2869          * After we relocate and before we remove the chunk with offset
2870          * chunk_offset, automatic removal of the block group can kick in,
2871          * resulting in a failure when calling btrfs_remove_chunk() below.
2872          *
2873          * Make sure to acquire this mutex before doing a tree search (dev
2874          * or chunk trees) to find chunks. Otherwise the cleaner kthread might
2875          * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after
2876          * we release the path used to search the chunk/dev tree and before
2877          * the current task acquires this mutex and calls us.
2878          */
2879         ASSERT(mutex_is_locked(&root->fs_info->delete_unused_bgs_mutex));
2880
2881         ret = btrfs_can_relocate(extent_root, chunk_offset);
2882         if (ret)
2883                 return -ENOSPC;
2884
2885         /* step one, relocate all the extents inside this chunk */
2886         btrfs_scrub_pause(root);
2887         ret = btrfs_relocate_block_group(extent_root, chunk_offset);
2888         btrfs_scrub_continue(root);
2889         if (ret)
2890                 return ret;
2891
2892         trans = btrfs_start_trans_remove_block_group(root->fs_info,
2893                                                      chunk_offset);
2894         if (IS_ERR(trans)) {
2895                 ret = PTR_ERR(trans);
2896                 btrfs_handle_fs_error(root->fs_info, ret, NULL);
2897                 return ret;
2898         }
2899
2900         /*
2901          * step two, delete the device extents and the
2902          * chunk tree entries
2903          */
2904         ret = btrfs_remove_chunk(trans, root, chunk_offset);
2905         btrfs_end_transaction(trans, root);
2906         return ret;
2907 }
2908
2909 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
2910 {
2911         struct btrfs_root *chunk_root = root->fs_info->chunk_root;
2912         struct btrfs_path *path;
2913         struct extent_buffer *leaf;
2914         struct btrfs_chunk *chunk;
2915         struct btrfs_key key;
2916         struct btrfs_key found_key;
2917         u64 chunk_type;
2918         bool retried = false;
2919         int failed = 0;
2920         int ret;
2921
2922         path = btrfs_alloc_path();
2923         if (!path)
2924                 return -ENOMEM;
2925
2926 again:
2927         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2928         key.offset = (u64)-1;
2929         key.type = BTRFS_CHUNK_ITEM_KEY;
2930
2931         while (1) {
2932                 mutex_lock(&root->fs_info->delete_unused_bgs_mutex);
2933                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2934                 if (ret < 0) {
2935                         mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
2936                         goto error;
2937                 }
2938                 BUG_ON(ret == 0); /* Corruption */
2939
2940                 ret = btrfs_previous_item(chunk_root, path, key.objectid,
2941                                           key.type);
2942                 if (ret)
2943                         mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
2944                 if (ret < 0)
2945                         goto error;
2946                 if (ret > 0)
2947                         break;
2948
2949                 leaf = path->nodes[0];
2950                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2951
2952                 chunk = btrfs_item_ptr(leaf, path->slots[0],
2953                                        struct btrfs_chunk);
2954                 chunk_type = btrfs_chunk_type(leaf, chunk);
2955                 btrfs_release_path(path);
2956
2957                 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
2958                         ret = btrfs_relocate_chunk(chunk_root,
2959                                                    found_key.offset);
2960                         if (ret == -ENOSPC)
2961                                 failed++;
2962                         else
2963                                 BUG_ON(ret);
2964                 }
2965                 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
2966
2967                 if (found_key.offset == 0)
2968                         break;
2969                 key.offset = found_key.offset - 1;
2970         }
2971         ret = 0;
2972         if (failed && !retried) {
2973                 failed = 0;
2974                 retried = true;
2975                 goto again;
2976         } else if (WARN_ON(failed && retried)) {
2977                 ret = -ENOSPC;
2978         }
2979 error:
2980         btrfs_free_path(path);
2981         return ret;
2982 }
2983
2984 static int insert_balance_item(struct btrfs_root *root,
2985                                struct btrfs_balance_control *bctl)
2986 {
2987         struct btrfs_trans_handle *trans;
2988         struct btrfs_balance_item *item;
2989         struct btrfs_disk_balance_args disk_bargs;
2990         struct btrfs_path *path;
2991         struct extent_buffer *leaf;
2992         struct btrfs_key key;
2993         int ret, err;
2994
2995         path = btrfs_alloc_path();
2996         if (!path)
2997                 return -ENOMEM;
2998
2999         trans = btrfs_start_transaction(root, 0);
3000         if (IS_ERR(trans)) {
3001                 btrfs_free_path(path);
3002                 return PTR_ERR(trans);
3003         }
3004
3005         key.objectid = BTRFS_BALANCE_OBJECTID;
3006         key.type = BTRFS_TEMPORARY_ITEM_KEY;
3007         key.offset = 0;
3008
3009         ret = btrfs_insert_empty_item(trans, root, path, &key,
3010                                       sizeof(*item));
3011         if (ret)
3012                 goto out;
3013
3014         leaf = path->nodes[0];
3015         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3016
3017         memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
3018
3019         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
3020         btrfs_set_balance_data(leaf, item, &disk_bargs);
3021         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
3022         btrfs_set_balance_meta(leaf, item, &disk_bargs);
3023         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
3024         btrfs_set_balance_sys(leaf, item, &disk_bargs);
3025
3026         btrfs_set_balance_flags(leaf, item, bctl->flags);
3027
3028         btrfs_mark_buffer_dirty(leaf);
3029 out:
3030         btrfs_free_path(path);
3031         err = btrfs_commit_transaction(trans, root);
3032         if (err && !ret)
3033                 ret = err;
3034         return ret;
3035 }
3036
3037 static int del_balance_item(struct btrfs_root *root)
3038 {
3039         struct btrfs_trans_handle *trans;
3040         struct btrfs_path *path;
3041         struct btrfs_key key;
3042         int ret, err;
3043
3044         path = btrfs_alloc_path();
3045         if (!path)
3046                 return -ENOMEM;
3047
3048         trans = btrfs_start_transaction(root, 0);
3049         if (IS_ERR(trans)) {
3050                 btrfs_free_path(path);
3051                 return PTR_ERR(trans);
3052         }
3053
3054         key.objectid = BTRFS_BALANCE_OBJECTID;
3055         key.type = BTRFS_TEMPORARY_ITEM_KEY;
3056         key.offset = 0;
3057
3058         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3059         if (ret < 0)
3060                 goto out;
3061         if (ret > 0) {
3062                 ret = -ENOENT;
3063                 goto out;
3064         }
3065
3066         ret = btrfs_del_item(trans, root, path);
3067 out:
3068         btrfs_free_path(path);
3069         err = btrfs_commit_transaction(trans, root);
3070         if (err && !ret)
3071                 ret = err;
3072         return ret;
3073 }
3074
3075 /*
3076  * This is a heuristic used to reduce the number of chunks balanced on
3077  * resume after balance was interrupted.
3078  */
3079 static void update_balance_args(struct btrfs_balance_control *bctl)
3080 {
3081         /*
3082          * Turn on soft mode for chunk types that were being converted.
3083          */
3084         if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
3085                 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
3086         if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
3087                 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
3088         if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
3089                 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
3090
3091         /*
3092          * Turn on usage filter if is not already used.  The idea is
3093          * that chunks that we have already balanced should be
3094          * reasonably full.  Don't do it for chunks that are being
3095          * converted - that will keep us from relocating unconverted
3096          * (albeit full) chunks.
3097          */
3098         if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3099             !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3100             !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3101                 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
3102                 bctl->data.usage = 90;
3103         }
3104         if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3105             !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3106             !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3107                 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
3108                 bctl->sys.usage = 90;
3109         }
3110         if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3111             !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3112             !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3113                 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
3114                 bctl->meta.usage = 90;
3115         }
3116 }
3117
3118 /*
3119  * Should be called with both balance and volume mutexes held to
3120  * serialize other volume operations (add_dev/rm_dev/resize) with
3121  * restriper.  Same goes for unset_balance_control.
3122  */
3123 static void set_balance_control(struct btrfs_balance_control *bctl)
3124 {
3125         struct btrfs_fs_info *fs_info = bctl->fs_info;
3126
3127         BUG_ON(fs_info->balance_ctl);
3128
3129         spin_lock(&fs_info->balance_lock);
3130         fs_info->balance_ctl = bctl;
3131         spin_unlock(&fs_info->balance_lock);
3132 }
3133
3134 static void unset_balance_control(struct btrfs_fs_info *fs_info)
3135 {
3136         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3137
3138         BUG_ON(!fs_info->balance_ctl);
3139
3140         spin_lock(&fs_info->balance_lock);
3141         fs_info->balance_ctl = NULL;
3142         spin_unlock(&fs_info->balance_lock);
3143
3144         kfree(bctl);
3145 }
3146
3147 /*
3148  * Balance filters.  Return 1 if chunk should be filtered out
3149  * (should not be balanced).
3150  */
3151 static int chunk_profiles_filter(u64 chunk_type,
3152                                  struct btrfs_balance_args *bargs)
3153 {
3154         chunk_type = chunk_to_extended(chunk_type) &
3155                                 BTRFS_EXTENDED_PROFILE_MASK;
3156
3157         if (bargs->profiles & chunk_type)
3158                 return 0;
3159
3160         return 1;
3161 }
3162
3163 static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
3164                               struct btrfs_balance_args *bargs)
3165 {
3166         struct btrfs_block_group_cache *cache;
3167         u64 chunk_used;
3168         u64 user_thresh_min;
3169         u64 user_thresh_max;
3170         int ret = 1;
3171
3172         cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3173         chunk_used = btrfs_block_group_used(&cache->item);
3174
3175         if (bargs->usage_min == 0)
3176                 user_thresh_min = 0;
3177         else
3178                 user_thresh_min = div_factor_fine(cache->key.offset,
3179                                         bargs->usage_min);
3180
3181         if (bargs->usage_max == 0)
3182                 user_thresh_max = 1;
3183         else if (bargs->usage_max > 100)
3184                 user_thresh_max = cache->key.offset;
3185         else
3186                 user_thresh_max = div_factor_fine(cache->key.offset,
3187                                         bargs->usage_max);
3188
3189         if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max)
3190                 ret = 0;
3191
3192         btrfs_put_block_group(cache);
3193         return ret;
3194 }
3195
3196 static int chunk_usage_filter(struct btrfs_fs_info *fs_info,
3197                 u64 chunk_offset, struct btrfs_balance_args *bargs)
3198 {
3199         struct btrfs_block_group_cache *cache;
3200         u64 chunk_used, user_thresh;
3201         int ret = 1;
3202
3203         cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3204         chunk_used = btrfs_block_group_used(&cache->item);
3205
3206         if (bargs->usage_min == 0)
3207                 user_thresh = 1;
3208         else if (bargs->usage > 100)
3209                 user_thresh = cache->key.offset;
3210         else
3211                 user_thresh = div_factor_fine(cache->key.offset,
3212                                               bargs->usage);
3213
3214         if (chunk_used < user_thresh)
3215                 ret = 0;
3216
3217         btrfs_put_block_group(cache);
3218         return ret;
3219 }
3220
3221 static int chunk_devid_filter(struct extent_buffer *leaf,
3222                               struct btrfs_chunk *chunk,
3223                               struct btrfs_balance_args *bargs)
3224 {
3225         struct btrfs_stripe *stripe;
3226         int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3227         int i;
3228
3229         for (i = 0; i < num_stripes; i++) {
3230                 stripe = btrfs_stripe_nr(chunk, i);
3231                 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
3232                         return 0;
3233         }
3234
3235         return 1;
3236 }
3237
3238 /* [pstart, pend) */
3239 static int chunk_drange_filter(struct extent_buffer *leaf,
3240                                struct btrfs_chunk *chunk,
3241                                u64 chunk_offset,