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,
3242                                struct btrfs_balance_args *bargs)
3243 {
3244         struct btrfs_stripe *stripe;
3245         int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3246         u64 stripe_offset;
3247         u64 stripe_length;
3248         int factor;
3249         int i;
3250
3251         if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
3252                 return 0;
3253
3254         if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP |
3255              BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) {
3256                 factor = num_stripes / 2;
3257         } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) {
3258                 factor = num_stripes - 1;
3259         } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) {
3260                 factor = num_stripes - 2;
3261         } else {
3262                 factor = num_stripes;
3263         }
3264
3265         for (i = 0; i < num_stripes; i++) {
3266                 stripe = btrfs_stripe_nr(chunk, i);
3267                 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
3268                         continue;
3269
3270                 stripe_offset = btrfs_stripe_offset(leaf, stripe);
3271                 stripe_length = btrfs_chunk_length(leaf, chunk);
3272                 stripe_length = div_u64(stripe_length, factor);
3273
3274                 if (stripe_offset < bargs->pend &&
3275                     stripe_offset + stripe_length > bargs->pstart)
3276                         return 0;
3277         }
3278
3279         return 1;
3280 }
3281
3282 /* [vstart, vend) */
3283 static int chunk_vrange_filter(struct extent_buffer *leaf,
3284                                struct btrfs_chunk *chunk,
3285                                u64 chunk_offset,
3286                                struct btrfs_balance_args *bargs)
3287 {
3288         if (chunk_offset < bargs->vend &&
3289             chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
3290                 /* at least part of the chunk is inside this vrange */
3291                 return 0;
3292
3293         return 1;
3294 }
3295
3296 static int chunk_stripes_range_filter(struct extent_buffer *leaf,
3297                                struct btrfs_chunk *chunk,
3298                                struct btrfs_balance_args *bargs)
3299 {
3300         int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3301
3302         if (bargs->stripes_min <= num_stripes
3303                         && num_stripes <= bargs->stripes_max)
3304                 return 0;
3305
3306         return 1;
3307 }
3308
3309 static int chunk_soft_convert_filter(u64 chunk_type,
3310                                      struct btrfs_balance_args *bargs)
3311 {
3312         if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
3313                 return 0;
3314
3315         chunk_type = chunk_to_extended(chunk_type) &
3316                                 BTRFS_EXTENDED_PROFILE_MASK;
3317
3318         if (bargs->target == chunk_type)
3319                 return 1;
3320
3321         return 0;
3322 }
3323
3324 static int should_balance_chunk(struct btrfs_root *root,
3325                                 struct extent_buffer *leaf,
3326                                 struct btrfs_chunk *chunk, u64 chunk_offset)
3327 {
3328         struct btrfs_balance_control *bctl = root->fs_info->balance_ctl;
3329         struct btrfs_balance_args *bargs = NULL;
3330         u64 chunk_type = btrfs_chunk_type(leaf, chunk);
3331
3332         /* type filter */
3333         if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
3334               (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
3335                 return 0;
3336         }
3337
3338         if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3339                 bargs = &bctl->data;
3340         else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3341                 bargs = &bctl->sys;
3342         else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3343                 bargs = &bctl->meta;
3344
3345         /* profiles filter */
3346         if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
3347             chunk_profiles_filter(chunk_type, bargs)) {
3348                 return 0;
3349         }
3350
3351         /* usage filter */
3352         if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
3353             chunk_usage_filter(bctl->fs_info, chunk_offset, bargs)) {
3354                 return 0;
3355         } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3356             chunk_usage_range_filter(bctl->fs_info, chunk_offset, bargs)) {
3357                 return 0;
3358         }
3359
3360         /* devid filter */
3361         if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
3362             chunk_devid_filter(leaf, chunk, bargs)) {
3363                 return 0;
3364         }
3365
3366         /* drange filter, makes sense only with devid filter */
3367         if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
3368             chunk_drange_filter(leaf, chunk, chunk_offset, bargs)) {
3369                 return 0;
3370         }
3371
3372         /* vrange filter */
3373         if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
3374             chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
3375                 return 0;
3376         }
3377
3378         /* stripes filter */
3379         if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) &&
3380             chunk_stripes_range_filter(leaf, chunk, bargs)) {
3381                 return 0;
3382         }
3383
3384         /* soft profile changing mode */
3385         if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
3386             chunk_soft_convert_filter(chunk_type, bargs)) {
3387                 return 0;
3388         }
3389
3390         /*
3391          * limited by count, must be the last filter
3392          */
3393         if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) {
3394                 if (bargs->limit == 0)
3395                         return 0;
3396                 else
3397                         bargs->limit--;
3398         } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) {
3399                 /*
3400                  * Same logic as the 'limit' filter; the minimum cannot be
3401                  * determined here because we do not have the global information
3402                  * about the count of all chunks that satisfy the filters.
3403                  */
3404                 if (bargs->limit_max == 0)
3405                         return 0;
3406                 else
3407                         bargs->limit_max--;
3408         }
3409
3410         return 1;
3411 }
3412
3413 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
3414 {
3415         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3416         struct btrfs_root *chunk_root = fs_info->chunk_root;
3417         struct btrfs_root *dev_root = fs_info->dev_root;
3418         struct list_head *devices;
3419         struct btrfs_device *device;
3420         u64 old_size;
3421         u64 size_to_free;
3422         u64 chunk_type;
3423         struct btrfs_chunk *chunk;
3424         struct btrfs_path *path;
3425         struct btrfs_key key;
3426         struct btrfs_key found_key;
3427         struct btrfs_trans_handle *trans;
3428         struct extent_buffer *leaf;
3429         int slot;
3430         int ret;
3431         int enospc_errors = 0;
3432         bool counting = true;
3433         /* The single value limit and min/max limits use the same bytes in the */
3434         u64 limit_data = bctl->data.limit;
3435         u64 limit_meta = bctl->meta.limit;
3436         u64 limit_sys = bctl->sys.limit;
3437         u32 count_data = 0;
3438         u32 count_meta = 0;
3439         u32 count_sys = 0;
3440         int chunk_reserved = 0;
3441         u64 bytes_used = 0;
3442
3443         /* step one make some room on all the devices */
3444         devices = &fs_info->fs_devices->devices;
3445         list_for_each_entry(device, devices, dev_list) {
3446                 old_size = btrfs_device_get_total_bytes(device);
3447                 size_to_free = div_factor(old_size, 1);
3448                 size_to_free = min_t(u64, size_to_free, SZ_1M);
3449                 if (!device->writeable ||
3450                     btrfs_device_get_total_bytes(device) -
3451                     btrfs_device_get_bytes_used(device) > size_to_free ||
3452                     device->is_tgtdev_for_dev_replace)
3453                         continue;
3454
3455                 ret = btrfs_shrink_device(device, old_size - size_to_free);
3456                 if (ret == -ENOSPC)
3457                         break;
3458                 BUG_ON(ret);
3459
3460                 trans = btrfs_start_transaction(dev_root, 0);
3461                 BUG_ON(IS_ERR(trans));
3462
3463                 ret = btrfs_grow_device(trans, device, old_size);
3464                 BUG_ON(ret);
3465
3466                 btrfs_end_transaction(trans, dev_root);
3467         }
3468
3469         /* step two, relocate all the chunks */
3470         path = btrfs_alloc_path();
3471         if (!path) {
3472                 ret = -ENOMEM;
3473                 goto error;
3474         }
3475
3476         /* zero out stat counters */
3477         spin_lock(&fs_info->balance_lock);
3478         memset(&bctl->stat, 0, sizeof(bctl->stat));
3479         spin_unlock(&fs_info->balance_lock);
3480 again:
3481         if (!counting) {
3482                 /*
3483                  * The single value limit and min/max limits use the same bytes
3484                  * in the
3485                  */
3486                 bctl->data.limit = limit_data;
3487                 bctl->meta.limit = limit_meta;
3488                 bctl->sys.limit = limit_sys;
3489         }
3490         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3491         key.offset = (u64)-1;
3492         key.type = BTRFS_CHUNK_ITEM_KEY;
3493
3494         while (1) {
3495                 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
3496                     atomic_read(&fs_info->balance_cancel_req)) {
3497                         ret = -ECANCELED;
3498                         goto error;
3499                 }
3500
3501                 mutex_lock(&fs_info->delete_unused_bgs_mutex);
3502                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3503                 if (ret < 0) {
3504                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3505                         goto error;
3506                 }
3507
3508                 /*
3509                  * this shouldn't happen, it means the last relocate
3510                  * failed
3511                  */
3512                 if (ret == 0)
3513                         BUG(); /* FIXME break ? */
3514
3515                 ret = btrfs_previous_item(chunk_root, path, 0,
3516                                           BTRFS_CHUNK_ITEM_KEY);
3517                 if (ret) {
3518                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3519                         ret = 0;
3520                         break;
3521                 }
3522
3523                 leaf = path->nodes[0];
3524                 slot = path->slots[0];
3525                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3526
3527                 if (found_key.objectid != key.objectid) {
3528                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3529                         break;
3530                 }
3531
3532                 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3533                 chunk_type = btrfs_chunk_type(leaf, chunk);
3534
3535                 if (!counting) {
3536                         spin_lock(&fs_info->balance_lock);
3537                         bctl->stat.considered++;
3538                         spin_unlock(&fs_info->balance_lock);
3539                 }
3540
3541                 ret = should_balance_chunk(chunk_root, leaf, chunk,
3542                                            found_key.offset);
3543
3544                 btrfs_release_path(path);
3545                 if (!ret) {
3546                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3547                         goto loop;
3548                 }
3549
3550                 if (counting) {
3551                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3552                         spin_lock(&fs_info->balance_lock);
3553                         bctl->stat.expected++;
3554                         spin_unlock(&fs_info->balance_lock);
3555
3556                         if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3557                                 count_data++;
3558                         else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3559                                 count_sys++;
3560                         else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3561                                 count_meta++;
3562
3563                         goto loop;
3564                 }
3565
3566                 /*
3567                  * Apply limit_min filter, no need to check if the LIMITS
3568                  * filter is used, limit_min is 0 by default
3569                  */
3570                 if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
3571                                         count_data < bctl->data.limit_min)
3572                                 || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) &&
3573                                         count_meta < bctl->meta.limit_min)
3574                                 || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) &&
3575                                         count_sys < bctl->sys.limit_min)) {
3576                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3577                         goto loop;
3578                 }
3579
3580                 ASSERT(fs_info->data_sinfo);
3581                 spin_lock(&fs_info->data_sinfo->lock);
3582                 bytes_used = fs_info->data_sinfo->bytes_used;
3583                 spin_unlock(&fs_info->data_sinfo->lock);
3584
3585                 if ((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
3586                     !chunk_reserved && !bytes_used) {
3587                         trans = btrfs_start_transaction(chunk_root, 0);
3588                         if (IS_ERR(trans)) {
3589                                 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3590                                 ret = PTR_ERR(trans);
3591                                 goto error;
3592                         }
3593
3594                         ret = btrfs_force_chunk_alloc(trans, chunk_root,
3595                                                       BTRFS_BLOCK_GROUP_DATA);
3596                         btrfs_end_transaction(trans, chunk_root);
3597                         if (ret < 0) {
3598                                 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3599                                 goto error;
3600                         }
3601                         chunk_reserved = 1;
3602                 }
3603
3604                 ret = btrfs_relocate_chunk(chunk_root,
3605                                            found_key.offset);
3606                 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3607                 if (ret && ret != -ENOSPC)
3608                         goto error;
3609                 if (ret == -ENOSPC) {
3610                         enospc_errors++;
3611                 } else {
3612                         spin_lock(&fs_info->balance_lock);
3613                         bctl->stat.completed++;
3614                         spin_unlock(&fs_info->balance_lock);
3615                 }
3616 loop:
3617                 if (found_key.offset == 0)
3618                         break;
3619                 key.offset = found_key.offset - 1;
3620         }
3621
3622         if (counting) {
3623                 btrfs_release_path(path);
3624                 counting = false;
3625                 goto again;
3626         }
3627 error:
3628         btrfs_free_path(path);
3629         if (enospc_errors) {
3630                 btrfs_info(fs_info, "%d enospc errors during balance",
3631                        enospc_errors);
3632                 if (!ret)
3633                         ret = -ENOSPC;
3634         }
3635
3636         return ret;
3637 }
3638
3639 /**
3640  * alloc_profile_is_valid - see if a given profile is valid and reduced
3641  * @flags: profile to validate
3642  * @extended: if true @flags is treated as an extended profile
3643  */
3644 static int alloc_profile_is_valid(u64 flags, int extended)
3645 {
3646         u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
3647                                BTRFS_BLOCK_GROUP_PROFILE_MASK);
3648
3649         flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
3650
3651         /* 1) check that all other bits are zeroed */
3652         if (flags & ~mask)
3653                 return 0;
3654
3655         /* 2) see if profile is reduced */
3656         if (flags == 0)
3657                 return !extended; /* "0" is valid for usual profiles */
3658
3659         /* true if exactly one bit set */
3660         return (flags & (flags - 1)) == 0;
3661 }
3662
3663 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
3664 {
3665         /* cancel requested || normal exit path */
3666         return atomic_read(&fs_info->balance_cancel_req) ||
3667                 (atomic_read(&fs_info->balance_pause_req) == 0 &&
3668                  atomic_read(&fs_info->balance_cancel_req) == 0);
3669 }
3670
3671 static void __cancel_balance(struct btrfs_fs_info *fs_info)
3672 {
3673         int ret;
3674
3675         unset_balance_control(fs_info);
3676         ret = del_balance_item(fs_info->tree_root);
3677         if (ret)
3678                 btrfs_handle_fs_error(fs_info, ret, NULL);
3679
3680         atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3681 }
3682
3683 /* Non-zero return value signifies invalidity */
3684 static inline int validate_convert_profile(struct btrfs_balance_args *bctl_arg,
3685                 u64 allowed)
3686 {
3687         return ((bctl_arg->flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3688                 (!alloc_profile_is_valid(bctl_arg->target, 1) ||
3689                  (bctl_arg->target & ~allowed)));
3690 }
3691
3692 /*
3693  * Should be called with both balance and volume mutexes held
3694  */
3695 int btrfs_balance(struct btrfs_balance_control *bctl,
3696                   struct btrfs_ioctl_balance_args *bargs)
3697 {
3698         struct btrfs_fs_info *fs_info = bctl->fs_info;
3699         u64 allowed;
3700         int mixed = 0;
3701         int ret;
3702         u64 num_devices;
3703         unsigned seq;
3704
3705         if (btrfs_fs_closing(fs_info) ||
3706             atomic_read(&fs_info->balance_pause_req) ||
3707             atomic_read(&fs_info->balance_cancel_req)) {
3708                 ret = -EINVAL;
3709                 goto out;
3710         }
3711
3712         allowed = btrfs_super_incompat_flags(fs_info->super_copy);
3713         if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
3714                 mixed = 1;
3715
3716         /*
3717          * In case of mixed groups both data and meta should be picked,
3718          * and identical options should be given for both of them.
3719          */
3720         allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
3721         if (mixed && (bctl->flags & allowed)) {
3722                 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
3723                     !(bctl->flags & BTRFS_BALANCE_METADATA) ||
3724                     memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
3725                         btrfs_err(fs_info, "with mixed groups data and "
3726                                    "metadata balance options must be the same");
3727                         ret = -EINVAL;
3728                         goto out;
3729                 }
3730         }
3731
3732         num_devices = fs_info->fs_devices->num_devices;
3733         btrfs_dev_replace_lock(&fs_info->dev_replace, 0);
3734         if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
3735                 BUG_ON(num_devices < 1);
3736                 num_devices--;
3737         }
3738         btrfs_dev_replace_unlock(&fs_info->dev_replace, 0);
3739         allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE | BTRFS_BLOCK_GROUP_DUP;
3740         if (num_devices > 1)
3741                 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1);
3742         if (num_devices > 2)
3743                 allowed |= BTRFS_BLOCK_GROUP_RAID5;
3744         if (num_devices > 3)
3745                 allowed |= (BTRFS_BLOCK_GROUP_RAID10 |
3746                             BTRFS_BLOCK_GROUP_RAID6);
3747         if (validate_convert_profile(&bctl->data, allowed)) {
3748                 btrfs_err(fs_info, "unable to start balance with target "
3749                            "data profile %llu",
3750                        bctl->data.target);
3751                 ret = -EINVAL;
3752                 goto out;
3753         }
3754         if (validate_convert_profile(&bctl->meta, allowed)) {
3755                 btrfs_err(fs_info,
3756                            "unable to start balance with target metadata profile %llu",
3757                        bctl->meta.target);
3758                 ret = -EINVAL;
3759                 goto out;
3760         }
3761         if (validate_convert_profile(&bctl->sys, allowed)) {
3762                 btrfs_err(fs_info,
3763                            "unable to start balance with target system profile %llu",
3764                        bctl->sys.target);
3765                 ret = -EINVAL;
3766                 goto out;
3767         }
3768
3769         /* allow to reduce meta or sys integrity only if force set */
3770         allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3771                         BTRFS_BLOCK_GROUP_RAID10 |
3772                         BTRFS_BLOCK_GROUP_RAID5 |
3773                         BTRFS_BLOCK_GROUP_RAID6;
3774         do {
3775                 seq = read_seqbegin(&fs_info->profiles_lock);
3776
3777                 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3778                      (fs_info->avail_system_alloc_bits & allowed) &&
3779                      !(bctl->sys.target & allowed)) ||
3780                     ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3781                      (fs_info->avail_metadata_alloc_bits & allowed) &&
3782                      !(bctl->meta.target & allowed))) {
3783                         if (bctl->flags & BTRFS_BALANCE_FORCE) {
3784                                 btrfs_info(fs_info, "force reducing metadata integrity");
3785                         } else {
3786                                 btrfs_err(fs_info, "balance will reduce metadata "
3787                                            "integrity, use force if you want this");
3788                                 ret = -EINVAL;
3789                                 goto out;
3790                         }
3791                 }
3792         } while (read_seqretry(&fs_info->profiles_lock, seq));
3793
3794         if (btrfs_get_num_tolerated_disk_barrier_failures(bctl->meta.target) <
3795                 btrfs_get_num_tolerated_disk_barrier_failures(bctl->data.target)) {
3796                 btrfs_warn(fs_info,
3797         "metadata profile 0x%llx has lower redundancy than data profile 0x%llx",
3798                         bctl->meta.target, bctl->data.target);
3799         }
3800
3801         if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3802                 fs_info->num_tolerated_disk_barrier_failures = min(
3803                         btrfs_calc_num_tolerated_disk_barrier_failures(fs_info),
3804                         btrfs_get_num_tolerated_disk_barrier_failures(
3805                                 bctl->sys.target));
3806         }
3807
3808         ret = insert_balance_item(fs_info->tree_root, bctl);
3809         if (ret && ret != -EEXIST)
3810                 goto out;
3811
3812         if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
3813                 BUG_ON(ret == -EEXIST);
3814                 set_balance_control(bctl);
3815         } else {
3816                 BUG_ON(ret != -EEXIST);
3817                 spin_lock(&fs_info->balance_lock);
3818                 update_balance_args(bctl);
3819                 spin_unlock(&fs_info->balance_lock);
3820         }
3821
3822         atomic_inc(&fs_info->balance_running);
3823         mutex_unlock(&fs_info->balance_mutex);
3824
3825         ret = __btrfs_balance(fs_info);
3826
3827         mutex_lock(&fs_info->balance_mutex);
3828         atomic_dec(&fs_info->balance_running);
3829
3830         if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3831                 fs_info->num_tolerated_disk_barrier_failures =
3832                         btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3833         }
3834
3835         if (bargs) {
3836                 memset(bargs, 0, sizeof(*bargs));
3837                 update_ioctl_balance_args(fs_info, 0, bargs);
3838         }
3839
3840         if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
3841             balance_need_close(fs_info)) {
3842                 __cancel_balance(fs_info);
3843         }
3844
3845         wake_up(&fs_info->balance_wait_q);
3846
3847         return ret;
3848 out:
3849         if (bctl->flags & BTRFS_BALANCE_RESUME)
3850                 __cancel_balance(fs_info);
3851         else {
3852                 kfree(bctl);
3853                 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3854         }
3855         return ret;
3856 }
3857
3858 static int balance_kthread(void *data)
3859 {
3860         struct btrfs_fs_info *fs_info = data;
3861         int ret = 0;
3862
3863         mutex_lock(&fs_info->volume_mutex);
3864         mutex_lock(&fs_info->balance_mutex);
3865
3866         if (fs_info->balance_ctl) {
3867                 btrfs_info(fs_info, "continuing balance");
3868                 ret = btrfs_balance(fs_info->balance_ctl, NULL);
3869         }
3870
3871         mutex_unlock(&fs_info->balance_mutex);
3872         mutex_unlock(&fs_info->volume_mutex);
3873
3874         return ret;
3875 }
3876
3877 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
3878 {
3879         struct task_struct *tsk;
3880
3881         spin_lock(&fs_info->balance_lock);
3882         if (!fs_info->balance_ctl) {
3883                 spin_unlock(&fs_info->balance_lock);
3884                 return 0;
3885         }
3886         spin_unlock(&fs_info->balance_lock);
3887
3888         if (btrfs_test_opt(fs_info->tree_root, SKIP_BALANCE)) {
3889                 btrfs_info(fs_info, "force skipping balance");
3890                 return 0;
3891         }
3892
3893         tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
3894         return PTR_ERR_OR_ZERO(tsk);
3895 }
3896
3897 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
3898 {
3899         struct btrfs_balance_control *bctl;
3900         struct btrfs_balance_item *item;
3901         struct btrfs_disk_balance_args disk_bargs;
3902         struct btrfs_path *path;
3903         struct extent_buffer *leaf;
3904         struct btrfs_key key;
3905         int ret;
3906
3907         path = btrfs_alloc_path();
3908         if (!path)
3909                 return -ENOMEM;
3910
3911         key.objectid = BTRFS_BALANCE_OBJECTID;
3912         key.type = BTRFS_TEMPORARY_ITEM_KEY;
3913         key.offset = 0;
3914
3915         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
3916         if (ret < 0)
3917                 goto out;
3918         if (ret > 0) { /* ret = -ENOENT; */
3919                 ret = 0;
3920                 goto out;
3921         }
3922
3923         bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
3924         if (!bctl) {
3925                 ret = -ENOMEM;
3926                 goto out;
3927         }
3928
3929         leaf = path->nodes[0];
3930         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3931
3932         bctl->fs_info = fs_info;
3933         bctl->flags = btrfs_balance_flags(leaf, item);
3934         bctl->flags |= BTRFS_BALANCE_RESUME;
3935
3936         btrfs_balance_data(leaf, item, &disk_bargs);
3937         btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
3938         btrfs_balance_meta(leaf, item, &disk_bargs);
3939         btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
3940         btrfs_balance_sys(leaf, item, &disk_bargs);
3941         btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
3942
3943         WARN_ON(atomic_xchg(&fs_info->mutually_exclusive_operation_running, 1));
3944
3945         mutex_lock(&fs_info->volume_mutex);
3946         mutex_lock(&fs_info->balance_mutex);
3947
3948         set_balance_control(bctl);
3949
3950         mutex_unlock(&fs_info->balance_mutex);
3951         mutex_unlock(&fs_info->volume_mutex);
3952 out:
3953         btrfs_free_path(path);
3954         return ret;
3955 }
3956
3957 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
3958 {
3959         int ret = 0;
3960
3961         mutex_lock(&fs_info->balance_mutex);
3962         if (!fs_info->balance_ctl) {
3963                 mutex_unlock(&fs_info->balance_mutex);
3964                 return -ENOTCONN;
3965         }
3966
3967         if (atomic_read(&fs_info->balance_running)) {
3968                 atomic_inc(&fs_info->balance_pause_req);
3969                 mutex_unlock(&fs_info->balance_mutex);
3970
3971                 wait_event(fs_info->balance_wait_q,
3972                            atomic_read(&fs_info->balance_running) == 0);
3973
3974                 mutex_lock(&fs_info->balance_mutex);
3975                 /* we are good with balance_ctl ripped off from under us */
3976                 BUG_ON(atomic_read(&fs_info->balance_running));
3977                 atomic_dec(&fs_info->balance_pause_req);
3978         } else {
3979                 ret = -ENOTCONN;
3980         }
3981
3982         mutex_unlock(&fs_info->balance_mutex);
3983         return ret;
3984 }
3985
3986 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
3987 {
3988         if (fs_info->sb->s_flags & MS_RDONLY)
3989                 return -EROFS;
3990
3991         mutex_lock(&fs_info->balance_mutex);
3992         if (!fs_info->balance_ctl) {
3993                 mutex_unlock(&fs_info->balance_mutex);
3994                 return -ENOTCONN;
3995         }
3996
3997         atomic_inc(&fs_info->balance_cancel_req);
3998         /*
3999          * if we are running just wait and return, balance item is
4000          * deleted in btrfs_balance in this case
4001          */
4002         if (atomic_read(&fs_info->balance_running)) {
4003                 mutex_unlock(&fs_info->balance_mutex);
4004                 wait_event(fs_info->balance_wait_q,
4005                            atomic_read(&fs_info->balance_running) == 0);
4006                 mutex_lock(&fs_info->balance_mutex);
4007         } else {
4008                 /* __cancel_balance needs volume_mutex */
4009                 mutex_unlock(&fs_info->balance_mutex);
4010                 mutex_lock(&fs_info->volume_mutex);
4011                 mutex_lock(&fs_info->balance_mutex);
4012
4013                 if (fs_info->balance_ctl)
4014                         __cancel_balance(fs_info);
4015
4016                 mutex_unlock(&fs_info->volume_mutex);
4017         }
4018
4019         BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running));
4020         atomic_dec(&fs_info->balance_cancel_req);
4021         mutex_unlock(&fs_info->balance_mutex);
4022         return 0;
4023 }
4024
4025 static int btrfs_uuid_scan_kthread(void *data)
4026 {
4027         struct btrfs_fs_info *fs_info = data;
4028         struct btrfs_root *root = fs_info->tree_root;
4029         struct btrfs_key key;
4030         struct btrfs_key max_key;
4031         struct btrfs_path *path = NULL;
4032         int ret = 0;
4033         struct extent_buffer *eb;
4034         int slot;
4035         struct btrfs_root_item root_item;
4036         u32 item_size;
4037         struct btrfs_trans_handle *trans = NULL;
4038
4039         path = btrfs_alloc_path();
4040         if (!path) {
4041                 ret = -ENOMEM;
4042                 goto out;
4043         }
4044
4045         key.objectid = 0;
4046         key.type = BTRFS_ROOT_ITEM_KEY;
4047         key.offset = 0;
4048
4049         max_key.objectid = (u64)-1;
4050         max_key.type = BTRFS_ROOT_ITEM_KEY;
4051         max_key.offset = (u64)-1;
4052
4053         while (1) {
4054                 ret = btrfs_search_forward(root, &key, path, 0);
4055                 if (ret) {
4056                         if (ret > 0)
4057                                 ret = 0;
4058                         break;
4059                 }
4060
4061                 if (key.type != BTRFS_ROOT_ITEM_KEY ||
4062                     (key.objectid < BTRFS_FIRST_FREE_OBJECTID &&
4063                      key.objectid != BTRFS_FS_TREE_OBJECTID) ||
4064                     key.objectid > BTRFS_LAST_FREE_OBJECTID)
4065                         goto skip;
4066
4067                 eb = path->nodes[0];
4068                 slot = path->slots[0];
4069                 item_size = btrfs_item_size_nr(eb, slot);
4070                 if (item_size < sizeof(root_item))
4071                         goto skip;
4072
4073                 read_extent_buffer(eb, &root_item,
4074                                    btrfs_item_ptr_offset(eb, slot),
4075                                    (int)sizeof(root_item));
4076                 if (btrfs_root_refs(&root_item) == 0)
4077                         goto skip;
4078
4079                 if (!btrfs_is_empty_uuid(root_item.uuid) ||
4080                     !btrfs_is_empty_uuid(root_item.received_uuid)) {
4081                         if (trans)
4082                                 goto update_tree;
4083
4084                         btrfs_release_path(path);
4085                         /*
4086                          * 1 - subvol uuid item
4087                          * 1 - received_subvol uuid item
4088                          */
4089                         trans = btrfs_start_transaction(fs_info->uuid_root, 2);
4090                         if (IS_ERR(trans)) {
4091                                 ret = PTR_ERR(trans);
4092                                 break;
4093                         }
4094                         continue;
4095                 } else {
4096                         goto skip;
4097                 }
4098 update_tree:
4099                 if (!btrfs_is_empty_uuid(root_item.uuid)) {
4100                         ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
4101                                                   root_item.uuid,
4102                                                   BTRFS_UUID_KEY_SUBVOL,
4103                                                   key.objectid);
4104                         if (ret < 0) {
4105                                 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4106                                         ret);
4107                                 break;
4108                         }
4109                 }
4110
4111                 if (!btrfs_is_empty_uuid(root_item.received_uuid)) {
4112                         ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
4113                                                   root_item.received_uuid,
4114                                                  BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4115                                                   key.objectid);
4116                         if (ret < 0) {
4117                                 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4118                                         ret);
4119                                 break;
4120                         }
4121                 }
4122
4123 skip:
4124                 if (trans) {
4125                         ret = btrfs_end_transaction(trans, fs_info->uuid_root);
4126                         trans = NULL;
4127                         if (ret)
4128                                 break;
4129                 }
4130
4131                 btrfs_release_path(path);
4132                 if (key.offset < (u64)-1) {
4133                         key.offset++;
4134                 } else if (key.type < BTRFS_ROOT_ITEM_KEY) {
4135                         key.offset = 0;
4136                         key.type = BTRFS_ROOT_ITEM_KEY;
4137                 } else if (key.objectid < (u64)-1) {
4138                         key.offset = 0;
4139                         key.type = BTRFS_ROOT_ITEM_KEY;
4140                         key.objectid++;
4141                 } else {
4142                         break;
4143                 }
4144                 cond_resched();
4145         }
4146
4147 out:
4148         btrfs_free_path(path);
4149         if (trans && !IS_ERR(trans))
4150                 btrfs_end_transaction(trans, fs_info->uuid_root);
4151         if (ret)
4152                 btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret);
4153         else
4154                 fs_info->update_uuid_tree_gen = 1;
4155         up(&fs_info->uuid_tree_rescan_sem);
4156         return 0;
4157 }
4158
4159 /*
4160  * Callback for btrfs_uuid_tree_iterate().
4161  * returns:
4162  * 0    check succeeded, the entry is not outdated.
4163  * < 0  if an error occurred.
4164  * > 0  if the check failed, which means the caller shall remove the entry.
4165  */
4166 static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info *fs_info,
4167                                        u8 *uuid, u8 type, u64 subid)
4168 {
4169         struct btrfs_key key;
4170         int ret = 0;
4171         struct btrfs_root *subvol_root;
4172
4173         if (type != BTRFS_UUID_KEY_SUBVOL &&
4174             type != BTRFS_UUID_KEY_RECEIVED_SUBVOL)
4175                 goto out;
4176
4177         key.objectid = subid;
4178         key.type = BTRFS_ROOT_ITEM_KEY;
4179         key.offset = (u64)-1;
4180         subvol_root = btrfs_read_fs_root_no_name(fs_info, &key);
4181         if (IS_ERR(subvol_root)) {
4182                 ret = PTR_ERR(subvol_root);
4183                 if (ret == -ENOENT)
4184                         ret = 1;
4185                 goto out;
4186         }
4187
4188         switch (type) {
4189         case BTRFS_UUID_KEY_SUBVOL:
4190                 if (memcmp(uuid, subvol_root->root_item.uuid, BTRFS_UUID_SIZE))
4191                         ret = 1;
4192                 break;
4193         case BTRFS_UUID_KEY_RECEIVED_SUBVOL:
4194                 if (memcmp(uuid, subvol_root->root_item.received_uuid,
4195                            BTRFS_UUID_SIZE))
4196                         ret = 1;
4197                 break;
4198         }
4199
4200 out:
4201         return ret;
4202 }
4203
4204 static int btrfs_uuid_rescan_kthread(void *data)
4205 {
4206         struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data;
4207         int ret;
4208
4209         /*
4210          * 1st step is to iterate through the existing UUID tree and
4211          * to delete all entries that contain outdated data.
4212          * 2nd step is to add all missing entries to the UUID tree.
4213          */
4214         ret = btrfs_uuid_tree_iterate(fs_info, btrfs_check_uuid_tree_entry);
4215         if (ret < 0) {
4216                 btrfs_warn(fs_info, "iterating uuid_tree failed %d", ret);
4217                 up(&fs_info->uuid_tree_rescan_sem);
4218                 return ret;
4219         }
4220         return btrfs_uuid_scan_kthread(data);
4221 }
4222
4223 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info)
4224 {
4225         struct btrfs_trans_handle *trans;
4226         struct btrfs_root *tree_root = fs_info->tree_root;
4227         struct btrfs_root *uuid_root;
4228         struct task_struct *task;
4229         int ret;
4230
4231         /*
4232          * 1 - root node
4233          * 1 - root item
4234          */
4235         trans = btrfs_start_transaction(tree_root, 2);
4236         if (IS_ERR(trans))
4237                 return PTR_ERR(trans);
4238
4239         uuid_root = btrfs_create_tree(trans, fs_info,
4240                                       BTRFS_UUID_TREE_OBJECTID);
4241         if (IS_ERR(uuid_root)) {
4242                 ret = PTR_ERR(uuid_root);
4243                 btrfs_abort_transaction(trans, tree_root, ret);
4244                 btrfs_end_transaction(trans, tree_root);
4245                 return ret;
4246         }
4247
4248         fs_info->uuid_root = uuid_root;
4249
4250         ret = btrfs_commit_transaction(trans, tree_root);
4251         if (ret)
4252                 return ret;
4253
4254         down(&fs_info->uuid_tree_rescan_sem);
4255         task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid");
4256         if (IS_ERR(task)) {
4257                 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4258                 btrfs_warn(fs_info, "failed to start uuid_scan task");
4259                 up(&fs_info->uuid_tree_rescan_sem);
4260                 return PTR_ERR(task);
4261         }
4262
4263         return 0;
4264 }
4265
4266 int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
4267 {
4268         struct task_struct *task;
4269
4270         down(&fs_info->uuid_tree_rescan_sem);
4271         task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
4272         if (IS_ERR(task)) {
4273                 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4274                 btrfs_warn(fs_info, "failed to start uuid_rescan task");
4275                 up(&fs_info->uuid_tree_rescan_sem);
4276                 return PTR_ERR(task);
4277         }
4278
4279         return 0;
4280 }
4281
4282 /*
4283  * shrinking a device means finding all of the device extents past
4284  * the new size, and then following the back refs to the chunks.
4285  * The chunk relocation code actually frees the device extent
4286  */
4287 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
4288 {
4289         struct btrfs_trans_handle *trans;
4290         struct btrfs_root *root = device->dev_root;
4291         struct btrfs_dev_extent *dev_extent = NULL;
4292         struct btrfs_path *path;
4293         u64 length;
4294         u64 chunk_offset;
4295         int ret;
4296         int slot;
4297         int failed = 0;
4298         bool retried = false;
4299         bool checked_pending_chunks = false;
4300         struct extent_buffer *l;
4301         struct btrfs_key key;
4302         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
4303         u64 old_total = btrfs_super_total_bytes(super_copy);
4304         u64 old_size = btrfs_device_get_total_bytes(device);
4305         u64 diff = old_size - new_size;
4306
4307         if (device->is_tgtdev_for_dev_replace)
4308                 return -EINVAL;
4309
4310         path = btrfs_alloc_path();
4311         if (!path)
4312                 return -ENOMEM;
4313
4314         path->reada = READA_FORWARD;
4315
4316         lock_chunks(root);
4317
4318         btrfs_device_set_total_bytes(device, new_size);
4319         if (device->writeable) {
4320                 device->fs_devices->total_rw_bytes -= diff;
4321                 spin_lock(&root->fs_info->free_chunk_lock);
4322                 root->fs_info->free_chunk_space -= diff;
4323                 spin_unlock(&root->fs_info->free_chunk_lock);
4324         }
4325         unlock_chunks(root);
4326
4327 again:
4328         key.objectid = device->devid;
4329         key.offset = (u64)-1;
4330         key.type = BTRFS_DEV_EXTENT_KEY;
4331
4332         do {
4333                 mutex_lock(&root->fs_info->delete_unused_bgs_mutex);
4334                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4335                 if (ret < 0) {
4336                         mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4337                         goto done;
4338                 }
4339
4340                 ret = btrfs_previous_item(root, path, 0, key.type);
4341                 if (ret)
4342                         mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4343                 if (ret < 0)
4344                         goto done;
4345                 if (ret) {
4346                         ret = 0;
4347                         btrfs_release_path(path);
4348                         break;
4349                 }
4350
4351                 l = path->nodes[0];
4352                 slot = path->slots[0];
4353                 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
4354
4355                 if (key.objectid != device->devid) {
4356                         mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4357                         btrfs_release_path(path);
4358                         break;
4359                 }
4360
4361                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
4362                 length = btrfs_dev_extent_length(l, dev_extent);
4363
4364                 if (key.offset + length <= new_size) {
4365                         mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4366                         btrfs_release_path(path);
4367                         break;
4368                 }
4369
4370                 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
4371                 btrfs_release_path(path);
4372
4373                 ret = btrfs_relocate_chunk(root, chunk_offset);
4374                 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4375                 if (ret && ret != -ENOSPC)
4376                         goto done;
4377                 if (ret == -ENOSPC)
4378                         failed++;
4379         } while (key.offset-- > 0);
4380
4381         if (failed && !retried) {
4382                 failed = 0;
4383                 retried = true;
4384                 goto again;
4385         } else if (failed && retried) {
4386                 ret = -ENOSPC;
4387                 goto done;
4388         }
4389
4390         /* Shrinking succeeded, else we would be at "done". */
4391         trans = btrfs_start_transaction(root, 0);
4392         if (IS_ERR(trans)) {
4393                 ret = PTR_ERR(trans);
4394                 goto done;
4395         }
4396
4397         lock_chunks(root);
4398
4399         /*
4400          * We checked in the above loop all device extents that were already in
4401          * the device tree. However before we have updated the device's
4402          * total_bytes to the new size, we might have had chunk allocations that
4403          * have not complete yet (new block groups attached to transaction
4404          * handles), and therefore their device extents were not yet in the
4405          * device tree and we missed them in the loop above. So if we have any
4406          * pending chunk using a device extent that overlaps the device range
4407          * that we can not use anymore, commit the current transaction and
4408          * repeat the search on the device tree - this way we guarantee we will
4409          * not have chunks using device extents that end beyond 'new_size'.
4410          */
4411         if (!checked_pending_chunks) {
4412                 u64 start = new_size;
4413                 u64 len = old_size - new_size;
4414
4415                 if (contains_pending_extent(trans->transaction, device,
4416                                             &start, len)) {
4417                         unlock_chunks(root);
4418                         checked_pending_chunks = true;
4419                         failed = 0;
4420                         retried = false;
4421                         ret = btrfs_commit_transaction(trans, root);
4422                         if (ret)
4423                                 goto done;
4424                         goto again;
4425                 }
4426         }
4427
4428         btrfs_device_set_disk_total_bytes(device, new_size);
4429         if (list_empty(&device->resized_list))
4430                 list_add_tail(&device->resized_list,
4431                               &root->fs_info->fs_devices->resized_devices);
4432
4433         WARN_ON(diff > old_total);
4434         btrfs_set_super_total_bytes(super_copy, old_total - diff);
4435         unlock_chunks(root);
4436
4437         /* Now btrfs_update_device() will change the on-disk size. */
4438         ret = btrfs_update_device(trans, device);
4439         btrfs_end_transaction(trans, root);
4440 done:
4441         btrfs_free_path(path);
4442         if (ret) {
4443                 lock_chunks(root);
4444                 btrfs_device_set_total_bytes(device, old_size);
4445                 if (device->writeable)
4446                         device->fs_devices->total_rw_bytes += diff;
4447                 spin_lock(&root->fs_info->free_chunk_lock);
4448                 root->fs_info->free_chunk_space += diff;
4449                 spin_unlock(&root->fs_info->free_chunk_lock);
4450                 unlock_chunks(root);
4451         }
4452         return ret;
4453 }
4454
4455 static int btrfs_add_system_chunk(struct btrfs_root *root,
4456                            struct btrfs_key *key,
4457                            struct btrfs_chunk *chunk, int item_size)
4458 {
4459         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
4460         struct btrfs_disk_key disk_key;
4461         u32 array_size;
4462         u8 *ptr;
4463
4464         lock_chunks(root);
4465         array_size = btrfs_super_sys_array_size(super_copy);
4466         if (array_size + item_size + sizeof(disk_key)
4467                         > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
4468                 unlock_chunks(root);
4469                 return -EFBIG;
4470         }
4471
4472         ptr = super_copy->sys_chunk_array + array_size;
4473         btrfs_cpu_key_to_disk(&disk_key, key);
4474         memcpy(ptr, &disk_key, sizeof(disk_key));
4475         ptr += sizeof(disk_key);
4476         memcpy(ptr, chunk, item_size);
4477         item_size += sizeof(disk_key);
4478         btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
4479         unlock_chunks(root);
4480
4481         return 0;
4482 }
4483
4484 /*
4485  * sort the devices in descending order by max_avail, total_avail
4486  */
4487 static int btrfs_cmp_device_info(const void *a, const void *b)
4488 {
4489         const struct btrfs_device_info *di_a = a;
4490         const struct btrfs_device_info *di_b = b;
4491
4492         if (di_a->max_avail > di_b->max_avail)
4493                 return -1;
4494         if (di_a->max_avail < di_b->max_avail)
4495                 return 1;
4496         if (di_a->total_avail > di_b->total_avail)
4497                 return -1;
4498         if (di_a->total_avail < di_b->total_avail)
4499                 return 1;
4500         return 0;
4501 }
4502
4503 static u32 find_raid56_stripe_len(u32 data_devices, u32 dev_stripe_target)
4504 {
4505         /* TODO allow them to set a preferred stripe size */
4506         return SZ_64K;
4507 }
4508
4509 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
4510 {
4511         if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK))
4512                 return;
4513
4514         btrfs_set_fs_incompat(info, RAID56);
4515 }
4516
4517 #define BTRFS_MAX_DEVS(r) ((BTRFS_LEAF_DATA_SIZE(r)             \
4518                         - sizeof(struct btrfs_item)             \
4519                         - sizeof(struct btrfs_chunk))           \
4520                         / sizeof(struct btrfs_stripe) + 1)
4521
4522 #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE        \
4523                                 - 2 * sizeof(struct btrfs_disk_key)     \
4524                                 - 2 * sizeof(struct btrfs_chunk))       \
4525                                 / sizeof(struct btrfs_stripe) + 1)
4526
4527 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4528                                struct btrfs_root *extent_root, u64 start,
4529                                u64 type)
4530 {
4531         struct btrfs_fs_info *info = extent_root->fs_info;
4532         struct btrfs_fs_devices *fs_devices = info->fs_devices;
4533         struct list_head *cur;
4534         struct map_lookup *map = NULL;
4535         struct extent_map_tree *em_tree;
4536         struct extent_map *em;
4537         struct btrfs_device_info *devices_info = NULL;
4538         u64 total_avail;
4539         int num_stripes;        /* total number of stripes to allocate */
4540         int data_stripes;       /* number of stripes that count for
4541                                    block group size */
4542         int sub_stripes;        /* sub_stripes info for map */
4543         int dev_stripes;        /* stripes per dev */
4544         int devs_max;           /* max devs to use */
4545         int devs_min;           /* min devs needed */
4546         int devs_increment;     /* ndevs has to be a multiple of this */
4547         int ncopies;            /* how many copies to data has */
4548         int ret;
4549         u64 max_stripe_size;
4550         u64 max_chunk_size;
4551         u64 stripe_size;
4552         u64 num_bytes;
4553         u64 raid_stripe_len = BTRFS_STRIPE_LEN;
4554         int ndevs;
4555         int i;
4556         int j;
4557         int index;
4558
4559         BUG_ON(!alloc_profile_is_valid(type, 0));
4560
4561         if (list_empty(&fs_devices->alloc_list))
4562                 return -ENOSPC;
4563
4564         index = __get_raid_index(type);
4565
4566         sub_stripes = btrfs_raid_array[index].sub_stripes;
4567         dev_stripes = btrfs_raid_array[index].dev_stripes;
4568         devs_max = btrfs_raid_array[index].devs_max;
4569         devs_min = btrfs_raid_array[index].devs_min;
4570         devs_increment = btrfs_raid_array[index].devs_increment;
4571         ncopies = btrfs_raid_array[index].ncopies;
4572
4573         if (type & BTRFS_BLOCK_GROUP_DATA) {
4574                 max_stripe_size = SZ_1G;
4575                 max_chunk_size = 10 * max_stripe_size;
4576                 if (!devs_max)
4577                         devs_max = BTRFS_MAX_DEVS(info->chunk_root);
4578         } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
4579                 /* for larger filesystems, use larger metadata chunks */
4580                 if (fs_devices->total_rw_bytes > 50ULL * SZ_1G)
4581                         max_stripe_size = SZ_1G;
4582                 else
4583                         max_stripe_size = SZ_256M;
4584                 max_chunk_size = max_stripe_size;
4585                 if (!devs_max)
4586                         devs_max = BTRFS_MAX_DEVS(info->chunk_root);
4587         } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
4588                 max_stripe_size = SZ_32M;
4589                 max_chunk_size = 2 * max_stripe_size;
4590                 if (!devs_max)
4591                         devs_max = BTRFS_MAX_DEVS_SYS_CHUNK;
4592         } else {
4593                 btrfs_err(info, "invalid chunk type 0x%llx requested",
4594                        type);
4595                 BUG_ON(1);
4596         }
4597
4598         /* we don't want a chunk larger than 10% of writeable space */
4599         max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
4600                              max_chunk_size);
4601
4602         devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info),
4603                                GFP_NOFS);
4604         if (!devices_info)
4605                 return -ENOMEM;
4606
4607         cur = fs_devices->alloc_list.next;
4608
4609         /*
4610          * in the first pass through the devices list, we gather information
4611          * about the available holes on each device.
4612          */
4613         ndevs = 0;
4614         while (cur != &fs_devices->alloc_list) {
4615                 struct btrfs_device *device;
4616                 u64 max_avail;
4617                 u64 dev_offset;
4618
4619                 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
4620
4621                 cur = cur->next;
4622
4623                 if (!device->writeable) {
4624                         WARN(1, KERN_ERR
4625                                "BTRFS: read-only device in alloc_list\n");
4626                         continue;
4627                 }
4628
4629                 if (!device->in_fs_metadata ||
4630                     device->is_tgtdev_for_dev_replace)
4631                         continue;
4632
4633                 if (device->total_bytes > device->bytes_used)
4634                         total_avail = device->total_bytes - device->bytes_used;
4635                 else
4636                         total_avail = 0;
4637
4638                 /* If there is no space on this device, skip it. */
4639                 if (total_avail == 0)
4640                         continue;
4641
4642                 ret = find_free_dev_extent(trans, device,
4643                                            max_stripe_size * dev_stripes,
4644                                            &dev_offset, &max_avail);
4645                 if (ret && ret != -ENOSPC)
4646                         goto error;
4647
4648                 if (ret == 0)
4649                         max_avail = max_stripe_size * dev_stripes;
4650
4651                 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes)
4652                         continue;
4653
4654                 if (ndevs == fs_devices->rw_devices) {
4655                         WARN(1, "%s: found more than %llu devices\n",
4656                              __func__, fs_devices->rw_devices);
4657                         break;
4658                 }
4659                 devices_info[ndevs].dev_offset = dev_offset;
4660                 devices_info[ndevs].max_avail = max_avail;
4661                 devices_info[ndevs].total_avail = total_avail;
4662                 devices_info[ndevs].dev = device;
4663                 ++ndevs;
4664         }
4665
4666         /*
4667          * now sort the devices by hole size / available space
4668          */
4669         sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
4670              btrfs_cmp_device_info, NULL);
4671
4672         /* round down to number of usable stripes */
4673         ndevs -= ndevs % devs_increment;
4674
4675         if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) {
4676                 ret = -ENOSPC;
4677                 goto error;
4678         }
4679
4680         if (devs_max && ndevs > devs_max)
4681                 ndevs = devs_max;
4682         /*
4683          * the primary goal is to maximize the number of stripes, so use as many
4684          * devices as possible, even if the stripes are not maximum sized.
4685          */
4686         stripe_size = devices_info[ndevs-1].max_avail;
4687         num_stripes = ndevs * dev_stripes;
4688
4689         /*
4690          * this will have to be fixed for RAID1 and RAID10 over
4691          * more drives
4692          */
4693         data_stripes = num_stripes / ncopies;
4694
4695         if (type & BTRFS_BLOCK_GROUP_RAID5) {
4696                 raid_stripe_len = find_raid56_stripe_len(ndevs - 1,
4697                                                 extent_root->stripesize);
4698                 data_stripes = num_stripes - 1;
4699         }
4700         if (type & BTRFS_BLOCK_GROUP_RAID6) {
4701                 raid_stripe_len = find_raid56_stripe_len(ndevs - 2,
4702                                                 extent_root->stripesize);
4703                 data_stripes = num_stripes - 2;
4704         }
4705
4706         /*
4707          * Use the number of data stripes to figure out how big this chunk
4708          * is really going to be in terms of logical address space,
4709          * and compare that answer with the max chunk size
4710          */
4711         if (stripe_size * data_stripes > max_chunk_size) {
4712                 u64 mask = (1ULL << 24) - 1;
4713
4714                 stripe_size = div_u64(max_chunk_size, data_stripes);
4715
4716                 /* bump the answer up to a 16MB boundary */
4717                 stripe_size = (stripe_size + mask) & ~mask;
4718
4719                 /* but don't go higher than the limits we found
4720                  * while searching for free extents
4721                  */
4722                 if (stripe_size > devices_info[ndevs-1].max_avail)
4723                         stripe_size = devices_info[ndevs-1].max_avail;
4724         }
4725
4726         stripe_size = div_u64(stripe_size, dev_stripes);
4727
4728         /* align to BTRFS_STRIPE_LEN */
4729         stripe_size = div_u64(stripe_size, raid_stripe_len);
4730         stripe_size *= raid_stripe_len;
4731
4732         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
4733         if (!map) {
4734                 ret = -ENOMEM;
4735                 goto error;
4736         }
4737         map->num_stripes = num_stripes;
4738
4739         for (i = 0; i < ndevs; ++i) {
4740                 for (j = 0; j < dev_stripes; ++j) {
4741                         int s = i * dev_stripes + j;
4742                         map->stripes[s].dev = devices_info[i].dev;
4743                         map->stripes[s].physical = devices_info[i].dev_offset +
4744                                                    j * stripe_size;
4745                 }
4746         }
4747         map->sector_size = extent_root->sectorsize;
4748         map->stripe_len = raid_stripe_len;
4749         map->io_align = raid_stripe_len;
4750         map->io_width = raid_stripe_len;
4751         map->type = type;
4752         map->sub_stripes = sub_stripes;
4753
4754         num_bytes = stripe_size * data_stripes;
4755
4756         trace_btrfs_chunk_alloc(info->chunk_root, map, start, num_bytes);
4757
4758         em = alloc_extent_map();
4759         if (!em) {
4760                 kfree(map);
4761                 ret = -ENOMEM;
4762                 goto error;
4763         }
4764         set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
4765         em->map_lookup = map;
4766         em->start = start;
4767         em->len = num_bytes;
4768         em->block_start = 0;
4769         em->block_len = em->len;
4770         em->orig_block_len = stripe_size;
4771
4772         em_tree = &extent_root->fs_info->mapping_tree.map_tree;
4773         write_lock(&em_tree->lock);
4774         ret = add_extent_mapping(em_tree, em, 0);
4775         if (!ret) {
4776                 list_add_tail(&em->list, &trans->transaction->pending_chunks);
4777                 atomic_inc(&em->refs);
4778         }
4779         write_unlock(&em_tree->lock);
4780         if (ret) {
4781                 free_extent_map(em);
4782                 goto error;
4783         }
4784
4785         ret = btrfs_make_block_group(trans, extent_root, 0, type,
4786                                      BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4787                                      start, num_bytes);
4788         if (ret)
4789                 goto error_del_extent;
4790
4791         for (i = 0; i < map->num_stripes; i++) {
4792                 num_bytes = map->stripes[i].dev->bytes_used + stripe_size;
4793                 btrfs_device_set_bytes_used(map->stripes[i].dev, num_bytes);
4794         }
4795
4796         spin_lock(&extent_root->fs_info->free_chunk_lock);
4797         extent_root->fs_info->free_chunk_space -= (stripe_size *
4798                                                    map->num_stripes);
4799         spin_unlock(&extent_root->fs_info->free_chunk_lock);
4800
4801         free_extent_map(em);
4802         check_raid56_incompat_flag(extent_root->fs_info, type);
4803
4804         kfree(devices_info);
4805         return 0;
4806
4807 error_del_extent:
4808         write_lock(&em_tree->lock);
4809         remove_extent_mapping(em_tree, em);
4810         write_unlock(&em_tree->lock);
4811
4812         /* One for our allocation */
4813         free_extent_map(em);
4814         /* One for the tree reference */
4815         free_extent_map(em);
4816         /* One for the pending_chunks list reference */
4817         free_extent_map(em);
4818 error:
4819         kfree(devices_info);
4820         return ret;
4821 }
4822
4823 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans,
4824                                 struct btrfs_root *extent_root,
4825                                 u64 chunk_offset, u64 chunk_size)
4826 {
4827         struct btrfs_key key;
4828         struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
4829         struct btrfs_device *device;
4830         struct btrfs_chunk *chunk;
4831         struct btrfs_stripe *stripe;
4832         struct extent_map_tree *em_tree;
4833         struct extent_map *em;
4834         struct map_lookup *map;
4835         size_t item_size;
4836         u64 dev_offset;
4837         u64 stripe_size;
4838         int i = 0;
4839         int ret = 0;
4840
4841         em_tree = &extent_root->fs_info->mapping_tree.map_tree;
4842         read_lock(&em_tree->lock);
4843         em = lookup_extent_mapping(em_tree, chunk_offset, chunk_size);
4844         read_unlock(&em_tree->lock);
4845
4846         if (!em) {
4847                 btrfs_crit(extent_root->fs_info, "unable to find logical "
4848                            "%Lu len %Lu", chunk_offset, chunk_size);
4849                 return -EINVAL;
4850         }
4851
4852         if (em->start != chunk_offset || em->len != chunk_size) {
4853                 btrfs_crit(extent_root->fs_info, "found a bad mapping, wanted"
4854                           " %Lu-%Lu, found %Lu-%Lu", chunk_offset,
4855                           chunk_size, em->start, em->len);
4856                 free_extent_map(em);
4857                 return -EINVAL;
4858         }
4859
4860         map = em->map_lookup;
4861         item_size = btrfs_chunk_item_size(map->num_stripes);
4862         stripe_size = em->orig_block_len;
4863
4864         chunk = kzalloc(item_size, GFP_NOFS);
4865         if (!chunk) {
4866                 ret = -ENOMEM;
4867                 goto out;
4868         }
4869
4870         /*
4871          * Take the device list mutex to prevent races with the final phase of
4872          * a device replace operation that replaces the device object associated
4873          * with the map's stripes, because the device object's id can change
4874          * at any time during that final phase of the device replace operation
4875          * (dev-replace.c:btrfs_dev_replace_finishing()).
4876          */
4877         mutex_lock(&chunk_root->fs_info->fs_devices->device_list_mutex);
4878         for (i = 0; i < map->num_stripes; i++) {
4879                 device = map->stripes[i].dev;
4880                 dev_offset = map->stripes[i].physical;
4881
4882                 ret = btrfs_update_device(trans, device);
4883                 if (ret)
4884                         break;
4885                 ret = btrfs_alloc_dev_extent(trans, device,
4886                                              chunk_root->root_key.objectid,
4887                                              BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4888                                              chunk_offset, dev_offset,
4889                                              stripe_size);
4890                 if (ret)
4891                         break;
4892         }
4893         if (ret) {
4894                 mutex_unlock(&chunk_root->fs_info->fs_devices->device_list_mutex);
4895                 goto out;
4896         }
4897
4898         stripe = &chunk->stripe;
4899         for (i = 0; i < map->num_stripes; i++) {
4900                 device = map->stripes[i].dev;
4901                 dev_offset = map->stripes[i].physical;
4902
4903                 btrfs_set_stack_stripe_devid(stripe, device->devid);
4904                 btrfs_set_stack_stripe_offset(stripe, dev_offset);
4905                 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
4906                 stripe++;
4907         }
4908         mutex_unlock(&chunk_root->fs_info->fs_devices->device_list_mutex);
4909
4910         btrfs_set_stack_chunk_length(chunk, chunk_size);
4911         btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
4912         btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
4913         btrfs_set_stack_chunk_type(chunk, map->type);
4914         btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
4915         btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
4916         btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
4917         btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
4918         btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
4919
4920         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
4921         key.type = BTRFS_CHUNK_ITEM_KEY;
4922         key.offset = chunk_offset;
4923
4924         ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
4925         if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
4926                 /*
4927                  * TODO: Cleanup of inserted chunk root in case of
4928                  * failure.
4929                  */
4930                 ret = btrfs_add_system_chunk(chunk_root, &key, chunk,
4931                                              item_size);
4932         }
4933
4934 out:
4935         kfree(chunk);
4936         free_extent_map(em);
4937         return ret;
4938 }
4939
4940 /*
4941  * Chunk allocation falls into two parts. The first part does works
4942  * that make the new allocated chunk useable, but not do any operation
4943  * that modifies the chunk tree. The second part does the works that
4944  * require modifying the chunk tree. This division is important for the
4945  * bootstrap process of adding storage to a seed btrfs.
4946  */
4947 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4948                       struct btrfs_root *extent_root, u64 type)
4949 {
4950         u64 chunk_offset;
4951
4952         ASSERT(mutex_is_locked(&extent_root->fs_info->chunk_mutex));
4953         chunk_offset = find_next_chunk(extent_root->fs_info);
4954         return __btrfs_alloc_chunk(trans, extent_root, chunk_offset, type);
4955 }
4956
4957 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
4958                                          struct btrfs_root *root,
4959                                          struct btrfs_device *device)
4960 {
4961         u64 chunk_offset;
4962         u64 sys_chunk_offset;
4963         u64 alloc_profile;
4964         struct btrfs_fs_info *fs_info = root->fs_info;
4965         struct btrfs_root *extent_root = fs_info->extent_root;
4966         int ret;
4967
4968         chunk_offset = find_next_chunk(fs_info);
4969         alloc_profile = btrfs_get_alloc_profile(extent_root, 0);
4970         ret = __btrfs_alloc_chunk(trans, extent_root, chunk_offset,
4971                                   alloc_profile);
4972         if (ret)
4973                 return ret;
4974
4975         sys_chunk_offset = find_next_chunk(root->fs_info);
4976         alloc_profile = btrfs_get_alloc_profile(fs_info->chunk_root, 0);
4977         ret = __btrfs_alloc_chunk(trans, extent_root, sys_chunk_offset,
4978                                   alloc_profile);
4979         return ret;
4980 }
4981
4982 static inline int btrfs_chunk_max_errors(struct map_lookup *map)
4983 {
4984         int max_errors;
4985
4986         if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
4987                          BTRFS_BLOCK_GROUP_RAID10 |
4988                          BTRFS_BLOCK_GROUP_RAID5 |
4989                          BTRFS_BLOCK_GROUP_DUP)) {
4990                 max_errors = 1;
4991         } else if (map->type & BTRFS_BLOCK_GROUP_RAID6) {
4992                 max_errors = 2;
4993         } else {
4994                 max_errors = 0;
4995         }
4996
4997         return max_errors;
4998 }
4999
5000 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
5001 {
5002         struct extent_map *em;
5003         struct map_lookup *map;
5004         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
5005         int readonly = 0;
5006         int miss_ndevs = 0;
5007         int i;
5008
5009         read_lock(&map_tree->map_tree.lock);
5010         em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
5011         read_unlock(&map_tree->map_tree.lock);
5012         if (!em)
5013                 return 1;
5014
5015         map = em->map_lookup;
5016         for (i = 0; i < map->num_stripes; i++) {
5017                 if (map->stripes[i].dev->missing) {
5018                         miss_ndevs++;
5019                         continue;
5020                 }
5021
5022                 if (!map->stripes[i].dev->writeable) {
5023                         readonly = 1;
5024                         goto end;
5025                 }
5026         }
5027
5028         /*
5029          * If the number of missing devices is larger than max errors,
5030          * we can not write the data into that chunk successfully, so
5031          * set it readonly.
5032          */
5033         if (miss_ndevs > btrfs_chunk_max_errors(map))
5034                 readonly = 1;
5035 end:
5036         free_extent_map(em);
5037         return readonly;
5038 }
5039
5040 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
5041 {
5042         extent_map_tree_init(&tree->map_tree);
5043 }
5044
5045 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
5046 {
5047         struct extent_map *em;
5048
5049         while (1) {
5050                 write_lock(&tree->map_tree.lock);
5051                 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
5052                 if (em)
5053                         remove_extent_mapping(&tree->map_tree, em);
5054                 write_unlock(&tree->map_tree.lock);
5055                 if (!em)
5056                         break;
5057                 /* once for us */
5058                 free_extent_map(em);
5059                 /* once for the tree */
5060                 free_extent_map(em);
5061         }
5062 }
5063
5064 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5065 {
5066         struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
5067         struct extent_map *em;
5068         struct map_lookup *map;
5069         struct extent_map_tree *em_tree = &map_tree->map_tree;
5070         int ret;
5071
5072         read_lock(&em_tree->lock);
5073         em = lookup_extent_mapping(em_tree, logical, len);
5074         read_unlock(&em_tree->lock);
5075
5076         /*
5077          * We could return errors for these cases, but that could get ugly and
5078          * we'd probably do the same thing which is just not do anything else
5079          * and exit, so return 1 so the callers don't try to use other copies.
5080          */
5081         if (!em) {
5082                 btrfs_crit(fs_info, "No mapping for %Lu-%Lu", logical,
5083                             logical+len);
5084                 return 1;
5085         }
5086
5087         if (em->start > logical || em->start + em->len < logical) {
5088                 btrfs_crit(fs_info, "Invalid mapping for %Lu-%Lu, got "
5089                             "%Lu-%Lu", logical, logical+len, em->start,
5090                             em->start + em->len);
5091                 free_extent_map(em);
5092                 return 1;
5093         }
5094
5095         map = em->map_lookup;
5096         if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
5097                 ret = map->num_stripes;
5098         else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5099                 ret = map->sub_stripes;
5100         else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
5101                 ret = 2;
5102         else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5103                 ret = 3;
5104         else
5105                 ret = 1;
5106         free_extent_map(em);
5107
5108         btrfs_dev_replace_lock(&fs_info->dev_replace, 0);
5109         if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace))
5110                 ret++;
5111         btrfs_dev_replace_unlock(&fs_info->dev_replace, 0);
5112
5113         return ret;
5114 }
5115
5116 unsigned long btrfs_full_stripe_len(struct btrfs_root *root,
5117                                     struct btrfs_mapping_tree *map_tree,
5118                                     u64 logical)
5119 {
5120         struct extent_map *em;
5121         struct map_lookup *map;
5122         struct extent_map_tree *em_tree = &map_tree->map_tree;
5123         unsigned long len = root->sectorsize;
5124
5125         read_lock(&em_tree->lock);
5126         em = lookup_extent_mapping(em_tree, logical, len);
5127         read_unlock(&em_tree->lock);
5128         BUG_ON(!em);
5129
5130         BUG_ON(em->start > logical || em->start + em->len < logical);
5131         map = em->map_lookup;
5132         if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5133                 len = map->stripe_len * nr_data_stripes(map);
5134         free_extent_map(em);
5135         return len;
5136 }
5137
5138 int btrfs_is_parity_mirror(struct btrfs_mapping_tree *map_tree,
5139                            u64 logical, u64 len, int mirror_num)
5140 {
5141         struct extent_map *em;
5142         struct map_lookup *map;
5143         struct extent_map_tree *em_tree = &map_tree->map_tree;
5144         int ret = 0;
5145
5146         read_lock(&em_tree->lock);
5147         em = lookup_extent_mapping(em_tree, logical, len);
5148         read_unlock(&em_tree->lock);
5149         BUG_ON(!em);
5150
5151         BUG_ON(em->start > logical || em->start + em->len < logical);
5152         map = em->map_lookup;
5153         if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5154                 ret = 1;
5155         free_extent_map(em);
5156         return ret;
5157 }
5158
5159 static int find_live_mirror(struct btrfs_fs_info *fs_info,
5160                             struct map_lookup *map, int first, int num,
5161                             int optimal, int dev_replace_is_ongoing)
5162 {
5163         int i;
5164         int tolerance;
5165         struct btrfs_device *srcdev;
5166
5167         if (dev_replace_is_ongoing &&
5168             fs_info->dev_replace.cont_reading_from_srcdev_mode ==
5169              BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
5170                 srcdev = fs_info->dev_replace.srcdev;
5171         else
5172                 srcdev = NULL;
5173
5174         /*
5175          * try to avoid the drive that is the source drive for a
5176          * dev-replace procedure, only choose it if no other non-missing
5177          * mirror is available
5178          */
5179         for (tolerance = 0; tolerance < 2; tolerance++) {
5180                 if (map->stripes[optimal].dev->bdev &&
5181                     (tolerance || map->stripes[optimal].dev != srcdev))
5182                         return optimal;
5183                 for (i = first; i < first + num; i++) {
5184                         if (map->stripes[i].dev->bdev &&
5185                             (tolerance || map->stripes[i].dev != srcdev))
5186                                 return i;
5187                 }
5188         }
5189
5190         /* we couldn't find one that doesn't fail.  Just return something
5191          * and the io error handling code will clean up eventually
5192          */
5193         return optimal;
5194 }
5195
5196 static inline int parity_smaller(u64 a, u64 b)
5197 {
5198         return a > b;
5199 }
5200
5201 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
5202 static void sort_parity_stripes(struct btrfs_bio *bbio, int num_stripes)
5203 {
5204         struct btrfs_bio_stripe s;
5205         int i;
5206         u64 l;
5207         int again = 1;
5208
5209         while (again) {
5210                 again = 0;
5211                 for (i = 0; i < num_stripes - 1; i++) {
5212                         if (parity_smaller(bbio->raid_map[i],
5213                                            bbio->raid_map[i+1])) {
5214                                 s = bbio->stripes[i];
5215                                 l = bbio->raid_map[i];
5216                                 bbio->stripes[i] = bbio->stripes[i+1];
5217                                 bbio->raid_map[i] = bbio->raid_map[i+1];
5218                                 bbio->stripes[i+1] = s;
5219                                 bbio->raid_map[i+1] = l;
5220
5221                                 again = 1;
5222                         }
5223                 }
5224         }
5225 }
5226
5227 static struct btrfs_bio *alloc_btrfs_bio(int total_stripes, int real_stripes)
5228 {
5229         struct btrfs_bio *bbio = kzalloc(
5230                  /* the size of the btrfs_bio */
5231                 sizeof(struct btrfs_bio) +
5232                 /* plus the variable array for the stripes */
5233                 sizeof(struct btrfs_bio_stripe) * (total_stripes) +
5234                 /* plus the variable array for the tgt dev */
5235                 sizeof(int) * (real_stripes) +
5236                 /*
5237                  * plus the raid_map, which includes both the tgt dev
5238                  * and the stripes
5239                  */
5240                 sizeof(u64) * (total_stripes),
5241                 GFP_NOFS|__GFP_NOFAIL);
5242
5243         atomic_set(&bbio->error, 0);
5244         atomic_set(&bbio->refs, 1);
5245
5246         return bbio;
5247 }
5248
5249 void btrfs_get_bbio(struct btrfs_bio *bbio)
5250 {
5251         WARN_ON(!atomic_read(&bbio->refs));
5252         atomic_inc(&bbio->refs);
5253 }
5254
5255 void btrfs_put_bbio(struct btrfs_bio *bbio)
5256 {
5257         if (!bbio)
5258                 return;
5259         if (atomic_dec_and_test(&bbio->refs))
5260                 kfree(bbio);
5261 }
5262
5263 static int __btrfs_map_block(struct btrfs_fs_info *fs_info, int op,
5264                              u64 logical, u64 *length,
5265                              struct btrfs_bio **bbio_ret,
5266                              int mirror_num, int need_raid_map)
5267 {
5268         struct extent_map *em;
5269         struct map_lookup *map;
5270         struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
5271         struct extent_map_tree *em_tree = &map_tree->map_tree;
5272         u64 offset;
5273         u64 stripe_offset;
5274         u64 stripe_end_offset;
5275         u64 stripe_nr;
5276         u64 stripe_nr_orig;
5277         u64 stripe_nr_end;
5278         u64 stripe_len;
5279         u32 stripe_index;
5280         int i;
5281         int ret = 0;
5282         int num_stripes;
5283         int max_errors = 0;
5284         int tgtdev_indexes = 0;
5285         struct btrfs_bio *bbio = NULL;
5286         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
5287         int dev_replace_is_ongoing = 0;
5288         int num_alloc_stripes;
5289         int patch_the_first_stripe_for_dev_replace = 0;
5290         u64 physical_to_patch_in_first_stripe = 0;
5291         u64 raid56_full_stripe_start = (u64)-1;
5292
5293         read_lock(&em_tree->lock);
5294         em = lookup_extent_mapping(em_tree, logical, *length);
5295         read_unlock(&em_tree->lock);
5296
5297         if (!em) {
5298                 btrfs_crit(fs_info, "unable to find logical %llu len %llu",
5299                         logical, *length);
5300                 return -EINVAL;
5301         }
5302
5303         if (em->start > logical || em->start + em->len < logical) {
5304                 btrfs_crit(fs_info, "found a bad mapping, wanted %Lu, "
5305                            "found %Lu-%Lu", logical, em->start,
5306                            em->start + em->len);
5307                 free_extent_map(em);
5308                 return -EINVAL;
5309         }
5310
5311         map = em->map_lookup;
5312         offset = logical - em->start;
5313
5314         stripe_len = map->stripe_len;
5315         stripe_nr = offset;
5316         /*
5317          * stripe_nr counts the total number of stripes we have to stride
5318          * to get to this block
5319          */
5320         stripe_nr = div64_u64(stripe_nr, stripe_len);
5321
5322         stripe_offset = stripe_nr * stripe_len;
5323         if (offset < stripe_offset) {
5324                 btrfs_crit(fs_info, "stripe math has gone wrong, "
5325                            "stripe_offset=%llu, offset=%llu, start=%llu, "
5326                            "logical=%llu, stripe_len=%llu",
5327                            stripe_offset, offset, em->start, logical,
5328                            stripe_len);
5329                 free_extent_map(em);
5330                 return -EINVAL;
5331         }
5332
5333         /* stripe_offset is the offset of this block in its stripe*/
5334         stripe_offset = offset - stripe_offset;
5335
5336         /* if we're here for raid56, we need to know the stripe aligned start */
5337         if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5338                 unsigned long full_stripe_len = stripe_len * nr_data_stripes(map);
5339                 raid56_full_stripe_start = offset;
5340
5341                 /* allow a write of a full stripe, but make sure we don't
5342                  * allow straddling of stripes
5343                  */
5344                 raid56_full_stripe_start = div64_u64(raid56_full_stripe_start,
5345                                 full_stripe_len);
5346                 raid56_full_stripe_start *= full_stripe_len;
5347         }
5348
5349         if (op == REQ_OP_DISCARD) {
5350                 /* we don't discard raid56 yet */
5351                 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5352                         ret = -EOPNOTSUPP;
5353                         goto out;
5354                 }
5355                 *length = min_t(u64, em->len - offset, *length);
5356         } else if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
5357                 u64 max_len;
5358                 /* For writes to RAID[56], allow a full stripeset across all disks.
5359                    For other RAID types and for RAID[56] reads, just allow a single
5360                    stripe (on a single disk). */
5361                 if ((map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
5362                     (op == REQ_OP_WRITE)) {
5363                         max_len = stripe_len * nr_data_stripes(map) -
5364                                 (offset - raid56_full_stripe_start);
5365                 } else {
5366                         /* we limit the length of each bio to what fits in a stripe */
5367                         max_len = stripe_len - stripe_offset;
5368                 }
5369                 *length = min_t(u64, em->len - offset, max_len);
5370         } else {
5371                 *length = em->len - offset;
5372         }
5373
5374         /* This is for when we're called from btrfs_merge_bio_hook() and all
5375            it cares about is the length */
5376         if (!bbio_ret)
5377                 goto out;
5378
5379         btrfs_dev_replace_lock(dev_replace, 0);
5380         dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
5381         if (!dev_replace_is_ongoing)
5382                 btrfs_dev_replace_unlock(dev_replace, 0);
5383         else
5384                 btrfs_dev_replace_set_lock_blocking(dev_replace);
5385
5386         if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
5387             op != REQ_OP_WRITE && op != REQ_OP_DISCARD &&
5388             op != REQ_GET_READ_MIRRORS && dev_replace->tgtdev != NULL) {
5389                 /*
5390                  * in dev-replace case, for repair case (that's the only
5391                  * case where the mirror is selected explicitly when
5392                  * calling btrfs_map_block), blocks left of the left cursor
5393                  * can also be read from the target drive.
5394                  * For REQ_GET_READ_MIRRORS, the target drive is added as
5395                  * the last one to the array of stripes. For READ, it also
5396                  * needs to be supported using the same mirror number.
5397                  * If the requested block is not left of the left cursor,
5398                  * EIO is returned. This can happen because btrfs_num_copies()
5399                  * returns one more in the dev-replace case.
5400                  */
5401                 u64 tmp_length = *length;
5402                 struct btrfs_bio *tmp_bbio = NULL;
5403                 int tmp_num_stripes;
5404                 u64 srcdev_devid = dev_replace->srcdev->devid;
5405                 int index_srcdev = 0;
5406                 int found = 0;
5407                 u64 physical_of_found = 0;
5408
5409                 ret = __btrfs_map_block(fs_info, REQ_GET_READ_MIRRORS,
5410                              logical, &tmp_length, &tmp_bbio, 0, 0);
5411                 if (ret) {
5412                         WARN_ON(tmp_bbio != NULL);
5413                         goto out;
5414                 }
5415
5416                 tmp_num_stripes = tmp_bbio->num_stripes;
5417                 if (mirror_num > tmp_num_stripes) {
5418                         /*
5419                          * REQ_GET_READ_MIRRORS does not contain this
5420                          * mirror, that means that the requested area
5421                          * is not left of the left cursor
5422                          */
5423                         ret = -EIO;
5424                         btrfs_put_bbio(tmp_bbio);
5425                         goto out;
5426                 }
5427
5428                 /*
5429                  * process the rest of the function using the mirror_num
5430                  * of the source drive. Therefore look it up first.
5431                  * At the end, patch the device pointer to the one of the
5432                  * target drive.
5433                  */
5434                 for (i = 0; i < tmp_num_stripes; i++) {
5435                         if (tmp_bbio->stripes[i].dev->devid != srcdev_devid)
5436                                 continue;
5437
5438                         /*
5439                          * In case of DUP, in order to keep it simple, only add
5440                          * the mirror with the lowest physical address
5441                          */
5442                         if (found &&
5443                             physical_of_found <= tmp_bbio->stripes[i].physical)
5444                                 continue;
5445
5446                         index_srcdev = i;
5447                         found = 1;
5448                         physical_of_found = tmp_bbio->stripes[i].physical;
5449                 }
5450
5451                 btrfs_put_bbio(tmp_bbio);
5452
5453                 if (!found) {
5454                         WARN_ON(1);
5455                         ret = -EIO;
5456                         goto out;
5457                 }
5458
5459                 mirror_num = index_srcdev + 1;
5460                 patch_the_first_stripe_for_dev_replace = 1;
5461                 physical_to_patch_in_first_stripe = physical_of_found;
5462         } else if (mirror_num > map->num_stripes) {
5463                 mirror_num = 0;
5464         }
5465
5466         num_stripes = 1;
5467         stripe_index = 0;
5468         stripe_nr_orig = stripe_nr;
5469         stripe_nr_end = ALIGN(offset + *length, map->stripe_len);
5470         stripe_nr_end = div_u64(stripe_nr_end, map->stripe_len);
5471         stripe_end_offset = stripe_nr_end * map->stripe_len -
5472                             (offset + *length);
5473
5474         if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5475                 if (op == REQ_OP_DISCARD)
5476                         num_stripes = min_t(u64, map->num_stripes,
5477                                             stripe_nr_end - stripe_nr_orig);
5478                 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5479                                 &stripe_index);
5480                 if (op != REQ_OP_WRITE && op != REQ_OP_DISCARD &&
5481                     op != REQ_GET_READ_MIRRORS)
5482                         mirror_num = 1;
5483         } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
5484                 if (op == REQ_OP_WRITE || op == REQ_OP_DISCARD ||
5485                     op == REQ_GET_READ_MIRRORS)
5486                         num_stripes = map->num_stripes;
5487                 else if (mirror_num)
5488                         stripe_index = mirror_num - 1;
5489                 else {
5490                         stripe_index = find_live_mirror(fs_info, map, 0,
5491                                             map->num_stripes,
5492                                             current->pid % map->num_stripes,
5493                                             dev_replace_is_ongoing);
5494                         mirror_num = stripe_index + 1;
5495                 }
5496
5497         } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
5498                 if (op == REQ_OP_WRITE || op == REQ_OP_DISCARD ||
5499                     op == REQ_GET_READ_MIRRORS) {
5500                         num_stripes = map->num_stripes;
5501                 } else if (mirror_num) {
5502                         stripe_index = mirror_num - 1;
5503                 } else {
5504                         mirror_num = 1;
5505                 }
5506
5507         } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5508                 u32 factor = map->num_stripes / map->sub_stripes;
5509
5510                 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
5511                 stripe_index *= map->sub_stripes;
5512
5513                 if (op == REQ_OP_WRITE || op == REQ_GET_READ_MIRRORS)
5514                         num_stripes = map->sub_stripes;
5515                 else if (op == REQ_OP_DISCARD)
5516                         num_stripes = min_t(u64, map->sub_stripes *
5517                                             (stripe_nr_end - stripe_nr_orig),
5518                                             map->num_stripes);
5519                 else if (mirror_num)
5520                         stripe_index += mirror_num - 1;
5521                 else {
5522                         int old_stripe_index = stripe_index;
5523                         stripe_index = find_live_mirror(fs_info, map,
5524                                               stripe_index,
5525                                               map->sub_stripes, stripe_index +
5526                                               current->pid % map->sub_stripes,
5527                                               dev_replace_is_ongoing);
5528                         mirror_num = stripe_index - old_stripe_index + 1;
5529                 }
5530
5531         } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5532                 if (need_raid_map &&
5533                     (op == REQ_OP_WRITE || op == REQ_GET_READ_MIRRORS ||
5534                      mirror_num > 1)) {
5535                         /* push stripe_nr back to the start of the full stripe */
5536                         stripe_nr = div_u64(raid56_full_stripe_start,
5537                                         stripe_len * nr_data_stripes(map));
5538
5539                         /* RAID[56] write or recovery. Return all stripes */
5540                         num_stripes = map->num_stripes;
5541                         max_errors = nr_parity_stripes(map);
5542
5543                         *length = map->stripe_len;
5544                         stripe_index = 0;
5545                         stripe_offset = 0;
5546                 } else {
5547                         /*
5548                          * Mirror #0 or #1 means the original data block.
5549                          * Mirror #2 is RAID5 parity block.
5550                          * Mirror #3 is RAID6 Q block.
5551                          */
5552                         stripe_nr = div_u64_rem(stripe_nr,
5553                                         nr_data_stripes(map), &stripe_index);
5554                         if (mirror_num > 1)
5555                                 stripe_index = nr_data_stripes(map) +
5556                                                 mirror_num - 2;
5557
5558                         /* We distribute the parity blocks across stripes */
5559                         div_u64_rem(stripe_nr + stripe_index, map->num_stripes,
5560                                         &stripe_index);
5561                         if ((op != REQ_OP_WRITE && op != REQ_OP_DISCARD &&
5562                             op != REQ_GET_READ_MIRRORS) && mirror_num <= 1)
5563                                 mirror_num = 1;
5564                 }
5565         } else {
5566                 /*
5567                  * after this, stripe_nr is the number of stripes on this
5568                  * device we have to walk to find the data, and stripe_index is
5569                  * the number of our device in the stripe array
5570                  */
5571                 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5572                                 &stripe_index);
5573                 mirror_num = stripe_index + 1;
5574         }
5575         if (stripe_index >= map->num_stripes) {
5576                 btrfs_crit(fs_info, "stripe index math went horribly wrong, "
5577                            "got stripe_index=%u, num_stripes=%u",
5578                            stripe_index, map->num_stripes);
5579                 ret = -EINVAL;
5580                 goto out;
5581         }
5582
5583         num_alloc_stripes = num_stripes;
5584         if (dev_replace_is_ongoing) {
5585                 if (op == REQ_OP_WRITE || op == REQ_OP_DISCARD)
5586                         num_alloc_stripes <<= 1;
5587                 if (op == REQ_GET_READ_MIRRORS)
5588                         num_alloc_stripes++;
5589                 tgtdev_indexes = num_stripes;
5590         }
5591
5592         bbio = alloc_btrfs_bio(num_alloc_stripes, tgtdev_indexes);
5593         if (!bbio) {
5594                 ret = -ENOMEM;
5595                 goto out;
5596         }
5597         if (dev_replace_is_ongoing)
5598                 bbio->tgtdev_map = (int *)(bbio->stripes + num_alloc_stripes);
5599
5600         /* build raid_map */
5601         if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK &&
5602             need_raid_map &&
5603             ((op == REQ_OP_WRITE || op == REQ_GET_READ_MIRRORS) ||
5604             mirror_num > 1)) {
5605                 u64 tmp;
5606                 unsigned rot;
5607
5608                 bbio->raid_map = (u64 *)((void *)bbio->stripes +
5609                                  sizeof(struct btrfs_bio_stripe) *
5610                                  num_alloc_stripes +
5611                                  sizeof(int) * tgtdev_indexes);
5612
5613                 /* Work out the disk rotation on this stripe-set */
5614                 div_u64_rem(stripe_nr, num_stripes, &rot);
5615
5616                 /* Fill in the logical address of each stripe */
5617                 tmp = stripe_nr * nr_data_stripes(map);
5618                 for (i = 0; i < nr_data_stripes(map); i++)
5619                         bbio->raid_map[(i+rot) % num_stripes] =
5620                                 em->start + (tmp + i) * map->stripe_len;
5621
5622                 bbio->raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE;
5623                 if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5624                         bbio->raid_map[(i+rot+1) % num_stripes] =
5625                                 RAID6_Q_STRIPE;
5626         }
5627
5628         if (op == REQ_OP_DISCARD) {
5629                 u32 factor = 0;
5630                 u32 sub_stripes = 0;
5631                 u64 stripes_per_dev = 0;
5632                 u32 remaining_stripes = 0;
5633                 u32 last_stripe = 0;
5634
5635                 if (map->type &
5636                     (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID10)) {
5637                         if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5638                                 sub_stripes = 1;
5639                         else
5640                                 sub_stripes = map->sub_stripes;
5641
5642                         factor = map->num_stripes / sub_stripes;
5643                         stripes_per_dev = div_u64_rem(stripe_nr_end -
5644                                                       stripe_nr_orig,
5645                                                       factor,
5646                                                       &remaining_stripes);
5647                         div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
5648                         last_stripe *= sub_stripes;
5649                 }
5650
5651                 for (i = 0; i < num_stripes; i++) {
5652                         bbio->stripes[i].physical =
5653                                 map->stripes[stripe_index].physical +
5654                                 stripe_offset + stripe_nr * map->stripe_len;
5655                         bbio->stripes[i].dev = map->stripes[stripe_index].dev;
5656
5657                         if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5658                                          BTRFS_BLOCK_GROUP_RAID10)) {
5659                                 bbio->stripes[i].length = stripes_per_dev *
5660                                                           map->stripe_len;
5661
5662                                 if (i / sub_stripes < remaining_stripes)
5663                                         bbio->stripes[i].length +=
5664                                                 map->stripe_len;
5665
5666                                 /*
5667                                  * Special for the first stripe and
5668                                  * the last stripe:
5669                                  *
5670                                  * |-------|...|-------|
5671                                  *     |----------|
5672                                  *    off     end_off
5673                                  */
5674                                 if (i < sub_stripes)
5675                                         bbio->stripes[i].length -=
5676                                                 stripe_offset;
5677
5678                                 if (stripe_index >= last_stripe &&
5679                                     stripe_index <= (last_stripe +
5680                                                      sub_stripes - 1))
5681                                         bbio->stripes[i].length -=
5682                                                 stripe_end_offset;
5683
5684                                 if (i == sub_stripes - 1)
5685                                         stripe_offset = 0;
5686                         } else
5687                                 bbio->stripes[i].length = *length;
5688
5689                         stripe_index++;
5690                         if (stripe_index == map->num_stripes) {
5691                                 /* This could only happen for RAID0/10 */
5692                                 stripe_index = 0;
5693                                 stripe_nr++;
5694                         }
5695                 }
5696         } else {
5697                 for (i = 0; i < num_stripes; i++) {
5698                         bbio->stripes[i].physical =
5699                                 map->stripes[stripe_index].physical +
5700                                 stripe_offset +
5701                                 stripe_nr * map->stripe_len;
5702                         bbio->stripes[i].dev =
5703                                 map->stripes[stripe_index].dev;
5704                         stripe_index++;
5705                 }
5706         }
5707
5708         if (op == REQ_OP_WRITE || op == REQ_GET_READ_MIRRORS)
5709                 max_errors = btrfs_chunk_max_errors(map);
5710
5711         if (bbio->raid_map)
5712                 sort_parity_stripes(bbio, num_stripes);
5713
5714         tgtdev_indexes = 0;
5715         if (dev_replace_is_ongoing &&
5716            (op == REQ_OP_WRITE || op == REQ_OP_DISCARD) &&
5717             dev_replace->tgtdev != NULL) {
5718                 int index_where_to_add;
5719                 u64 srcdev_devid = dev_replace->srcdev->devid;
5720
5721                 /*
5722                  * duplicate the write operations while the dev replace
5723                  * procedure is running. Since the copying of the old disk
5724                  * to the new disk takes place at run time while the
5725                  * filesystem is mounted writable, the regular write
5726                  * operations to the old disk have to be duplicated to go
5727                  * to the new disk as well.
5728                  * Note that device->missing is handled by the caller, and
5729                  * that the write to the old disk is already set up in the
5730                  * stripes array.
5731                  */
5732                 index_where_to_add = num_stripes;
5733                 for (i = 0; i < num_stripes; i++) {
5734                         if (bbio->stripes[i].dev->devid == srcdev_devid) {
5735                                 /* write to new disk, too */
5736                                 struct btrfs_bio_stripe *new =
5737                                         bbio->stripes + index_where_to_add;
5738                                 struct btrfs_bio_stripe *old =
5739                                         bbio->stripes + i;
5740
5741                                 new->physical = old->physical;
5742                                 new->length = old->length;
5743                                 new->dev = dev_replace->tgtdev;
5744                                 bbio->tgtdev_map[i] = index_where_to_add;
5745                                 index_where_to_add++;
5746                                 max_errors++;
5747                                 tgtdev_indexes++;
5748                         }
5749                 }
5750                 num_stripes = index_where_to_add;
5751         } else if (dev_replace_is_ongoing && (op == REQ_GET_READ_MIRRORS) &&
5752                    dev_replace->tgtdev != NULL) {
5753                 u64 srcdev_devid = dev_replace->srcdev->devid;
5754                 int index_srcdev = 0;
5755                 int found = 0;
5756                 u64 physical_of_found = 0;
5757
5758                 /*
5759                  * During the dev-replace procedure, the target drive can
5760                  * also be used to read data in case it is needed to repair
5761                  * a corrupt block elsewhere. This is possible if the
5762                  * requested area is left of the left cursor. In this area,
5763                  * the target drive is a full copy of the source drive.
5764                  */
5765                 for (i = 0; i < num_stripes; i++) {
5766                         if (bbio->stripes[i].dev->devid == srcdev_devid) {
5767                                 /*
5768                                  * In case of DUP, in order to keep it
5769                                  * simple, only add the mirror with the
5770                                  * lowest physical address
5771                                  */
5772                                 if (found &&
5773                                     physical_of_found <=
5774                                      bbio->stripes[i].physical)
5775                                         continue;
5776                                 index_srcdev = i;
5777                                 found = 1;
5778                                 physical_of_found = bbio->stripes[i].physical;
5779                         }
5780                 }
5781                 if (found) {
5782                         struct btrfs_bio_stripe *tgtdev_stripe =
5783                                 bbio->stripes + num_stripes;
5784
5785                         tgtdev_stripe->physical = physical_of_found;
5786                         tgtdev_stripe->length =
5787                                 bbio->stripes[index_srcdev].length;
5788                         tgtdev_stripe->dev = dev_replace->tgtdev;
5789                         bbio->tgtdev_map[index_srcdev] = num_stripes;
5790
5791                         tgtdev_indexes++;
5792                         num_stripes++;
5793                 }
5794         }
5795
5796         *bbio_ret = bbio;
5797         bbio->map_type = map->type;
5798         bbio->num_stripes = num_stripes;
5799         bbio->max_errors = max_errors;
5800         bbio->mirror_num = mirror_num;
5801         bbio->num_tgtdevs = tgtdev_indexes;
5802
5803         /*
5804          * this is the case that REQ_READ && dev_replace_is_ongoing &&
5805          * mirror_num == num_stripes + 1 && dev_replace target drive is
5806          * available as a mirror
5807          */
5808         if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) {
5809                 WARN_ON(num_stripes > 1);
5810                 bbio->stripes[0].dev = dev_replace->tgtdev;
5811                 bbio->stripes[0].physical = physical_to_patch_in_first_stripe;
5812                 bbio->mirror_num = map->num_stripes + 1;
5813         }
5814 out:
5815         if (dev_replace_is_ongoing) {
5816                 btrfs_dev_replace_clear_lock_blocking(dev_replace);
5817                 btrfs_dev_replace_unlock(dev_replace, 0);
5818         }
5819         free_extent_map(em);
5820         return ret;
5821 }
5822
5823 int btrfs_map_block(struct btrfs_fs_info *fs_info, int op,
5824                       u64 logical, u64 *length,
5825                       struct btrfs_bio **bbio_ret, int mirror_num)
5826 {
5827         return __btrfs_map_block(fs_info, op, logical, length, bbio_ret,
5828                                  mirror_num, 0);
5829 }
5830
5831 /* For Scrub/replace */
5832 int btrfs_map_sblock(struct btrfs_fs_info *fs_info, int op,
5833                      u64 logical, u64 *length,
5834                      struct btrfs_bio **bbio_ret, int mirror_num,
5835                      int need_raid_map)
5836 {
5837         return __btrfs_map_block(fs_info, op, logical, length, bbio_ret,
5838                                  mirror_num, need_raid_map);
5839 }
5840
5841 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
5842                      u64 chunk_start, u64 physical, u64 devid,
5843                      u64 **logical, int *naddrs, int *stripe_len)
5844 {
5845         struct extent_map_tree *em_tree = &map_tree->map_tree;
5846         struct extent_map *em;
5847         struct map_lookup *map;
5848         u64 *buf;
5849         u64 bytenr;
5850         u64 length;
5851         u64 stripe_nr;
5852         u64 rmap_len;
5853         int i, j, nr = 0;
5854
5855         read_lock(&em_tree->lock);
5856         em = lookup_extent_mapping(em_tree, chunk_start, 1);
5857         read_unlock(&em_tree->lock);
5858
5859         if (!em) {
5860                 printk(KERN_ERR "BTRFS: couldn't find em for chunk %Lu\n",
5861                        chunk_start);
5862                 return -EIO;
5863         }
5864
5865         if (em->start != chunk_start) {
5866                 printk(KERN_ERR "BTRFS: bad chunk start, em=%Lu, wanted=%Lu\n",
5867                        em->start, chunk_start);
5868                 free_extent_map(em);
5869                 return -EIO;
5870         }
5871         map = em->map_lookup;
5872
5873         length = em->len;
5874         rmap_len = map->stripe_len;
5875
5876         if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5877                 length = div_u64(length, map->num_stripes / map->sub_stripes);
5878         else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5879                 length = div_u64(length, map->num_stripes);
5880         else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5881                 length = div_u64(length, nr_data_stripes(map));
5882                 rmap_len = map->stripe_len * nr_data_stripes(map);
5883         }
5884
5885         buf = kcalloc(map->num_stripes, sizeof(u64), GFP_NOFS);
5886         BUG_ON(!buf); /* -ENOMEM */
5887
5888         for (i = 0; i < map->num_stripes; i++) {
5889                 if (devid && map->stripes[i].dev->devid != devid)
5890                         continue;
5891                 if (map->stripes[i].physical > physical ||
5892                     map->stripes[i].physical + length <= physical)
5893                         continue;
5894
5895                 stripe_nr = physical - map->stripes[i].physical;
5896                 stripe_nr = div_u64(stripe_nr, map->stripe_len);
5897
5898                 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5899                         stripe_nr = stripe_nr * map->num_stripes + i;
5900                         stripe_nr = div_u64(stripe_nr, map->sub_stripes);
5901                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5902                         stripe_nr = stripe_nr * map->num_stripes + i;
5903                 } /* else if RAID[56], multiply by nr_data_stripes().
5904                    * Alternatively, just use rmap_len below instead of
5905                    * map->stripe_len */
5906
5907                 bytenr = chunk_start + stripe_nr * rmap_len;
5908                 WARN_ON(nr >= map->num_stripes);
5909                 for (j = 0; j < nr; j++) {
5910                         if (buf[j] == bytenr)
5911                                 break;
5912                 }
5913                 if (j == nr) {
5914                         WARN_ON(nr >= map->num_stripes);
5915                         buf[nr++] = bytenr;
5916                 }
5917         }
5918
5919         *logical = buf;
5920         *naddrs = nr;
5921         *stripe_len = rmap_len;
5922
5923         free_extent_map(em);
5924         return 0;
5925 }
5926
5927 static inline void btrfs_end_bbio(struct btrfs_bio *bbio, struct bio *bio)
5928 {
5929         bio->bi_private = bbio->private;
5930         bio->bi_end_io = bbio->end_io;
5931         bio_endio(bio);
5932
5933         btrfs_put_bbio(bbio);
5934 }
5935
5936 static void btrfs_end_bio(struct bio *bio)
5937 {
5938         struct btrfs_bio *bbio = bio->bi_private;
5939         int is_orig_bio = 0;
5940
5941         if (bio->bi_error) {
5942                 atomic_inc(&bbio->error);
5943                 if (bio->bi_error == -EIO || bio->bi_error == -EREMOTEIO) {
5944                         unsigned int stripe_index =
5945                                 btrfs_io_bio(bio)->stripe_index;
5946                         struct btrfs_device *dev;
5947
5948                         BUG_ON(stripe_index >= bbio->num_stripes);
5949                         dev = bbio->stripes[stripe_index].dev;
5950                         if (dev->bdev) {
5951                                 if (bio_op(bio) == REQ_OP_WRITE)
5952                                         btrfs_dev_stat_inc(dev,
5953                                                 BTRFS_DEV_STAT_WRITE_ERRS);
5954                                 else
5955                                         btrfs_dev_stat_inc(dev,
5956                                                 BTRFS_DEV_STAT_READ_ERRS);
5957                                 if ((bio->bi_rw & WRITE_FLUSH) == WRITE_FLUSH)
5958                                         btrfs_dev_stat_inc(dev,
5959                                                 BTRFS_DEV_STAT_FLUSH_ERRS);
5960                                 btrfs_dev_stat_print_on_error(dev);
5961                         }
5962                 }
5963         }
5964
5965         if (bio == bbio->orig_bio)
5966                 is_orig_bio = 1;
5967
5968         btrfs_bio_counter_dec(bbio->fs_info);
5969
5970         if (atomic_dec_and_test(&bbio->stripes_pending)) {
5971                 if (!is_orig_bio) {
5972                         bio_put(bio);
5973                         bio = bbio->orig_bio;
5974                 }
5975
5976                 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
5977                 /* only send an error to the higher layers if it is
5978                  * beyond the tolerance of the btrfs bio
5979                  */
5980                 if (atomic_read(&bbio->error) > bbio->max_errors) {
5981                         bio->bi_error = -EIO;
5982                 } else {
5983                         /*
5984                          * this bio is actually up to date, we didn't
5985                          * go over the max number of errors
5986                          */
5987                         bio->bi_error = 0;
5988                 }
5989
5990                 btrfs_end_bbio(bbio, bio);
5991         } else if (!is_orig_bio) {
5992                 bio_put(bio);
5993         }
5994 }
5995
5996 /*
5997  * see run_scheduled_bios for a description of why bios are collected for
5998  * async submit.
5999  *
6000  * This will add one bio to the pending list for a device and make sure
6001  * the work struct is scheduled.
6002  */
6003 static noinline void btrfs_schedule_bio(struct btrfs_root *root,
6004                                         struct btrfs_device *device,
6005                                         struct bio *bio)
6006 {
6007         int should_queue = 1;
6008         struct btrfs_pending_bios *pending_bios;
6009
6010         if (device->missing || !device->bdev) {
6011                 bio_io_error(bio);
6012                 return;
6013         }
6014
6015         /* don't bother with additional async steps for reads, right now */
6016         if (bio_op(bio) == REQ_OP_READ) {
6017                 bio_get(bio);
6018                 btrfsic_submit_bio(bio);
6019                 bio_put(bio);
6020                 return;
6021         }
6022
6023         /*
6024          * nr_async_bios allows us to reliably return congestion to the
6025          * higher layers.  Otherwise, the async bio makes it appear we have
6026          * made progress against dirty pages when we've really just put it
6027          * on a queue for later
6028          */
6029         atomic_inc(&root->fs_info->nr_async_bios);
6030         WARN_ON(bio->bi_next);
6031         bio->bi_next = NULL;
6032
6033         spin_lock(&device->io_lock);
6034         if (bio->bi_rw & REQ_SYNC)
6035                 pending_bios = &device->pending_sync_bios;
6036         else
6037                 pending_bios = &device->pending_bios;
6038
6039         if (pending_bios->tail)
6040                 pending_bios->tail->bi_next = bio;
6041
6042         pending_bios->tail = bio;
6043         if (!pending_bios->head)
6044                 pending_bios->head = bio;
6045         if (device->running_pending)
6046                 should_queue = 0;
6047
6048         spin_unlock(&device->io_lock);
6049
6050         if (should_queue)
6051                 btrfs_queue_work(root->fs_info->submit_workers,
6052                                  &device->work);
6053 }
6054
6055 static void submit_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio,
6056                               struct bio *bio, u64 physical, int dev_nr,
6057                               int async)
6058 {
6059         struct btrfs_device *dev = bbio->stripes[dev_nr].dev;
6060
6061         bio->bi_private = bbio;
6062         btrfs_io_bio(bio)->stripe_index = dev_nr;
6063         bio->bi_end_io = btrfs_end_bio;
6064         bio->bi_iter.bi_sector = physical >> 9;
6065 #ifdef DEBUG
6066         {
6067                 struct rcu_string *name;
6068
6069                 rcu_read_lock();
6070                 name = rcu_dereference(dev->name);
6071                 pr_debug("btrfs_map_bio: rw %d 0x%x, sector=%llu, dev=%lu "
6072                          "(%s id %llu), size=%u\n", bio_op(bio), bio->bi_rw,
6073                          (u64)bio->bi_iter.bi_sector, (u_long)dev->bdev->bd_dev,
6074                          name->str, dev->devid, bio->bi_iter.bi_size);
6075                 rcu_read_unlock();
6076         }
6077 #endif
6078         bio->bi_bdev = dev->bdev;
6079
6080         btrfs_bio_counter_inc_noblocked(root->fs_info);
6081
6082         if (async)
6083                 btrfs_schedule_bio(root, dev, bio);
6084         else
6085                 btrfsic_submit_bio(bio);
6086 }
6087
6088 static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical)
6089 {
6090         atomic_inc(&bbio->error);
6091         if (atomic_dec_and_test(&bbio->stripes_pending)) {
6092                 /* Should be the original bio. */
6093                 WARN_ON(bio != bbio->orig_bio);
6094
6095                 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
6096                 bio->bi_iter.bi_sector = logical >> 9;
6097                 bio->bi_error = -EIO;
6098                 btrfs_end_bbio(bbio, bio);
6099         }
6100 }
6101
6102 int btrfs_map_bio(struct btrfs_root *root, struct bio *bio,
6103                   int mirror_num, int async_submit)
6104 {
6105         struct btrfs_device *dev;
6106         struct bio *first_bio = bio;
6107         u64 logical = (u64)bio->bi_iter.bi_sector << 9;
6108         u64 length = 0;
6109         u64 map_length;
6110         int ret;
6111         int dev_nr;
6112         int total_devs;
6113         struct btrfs_bio *bbio = NULL;
6114
6115         length = bio->bi_iter.bi_size;
6116         map_length = length;
6117
6118         btrfs_bio_counter_inc_blocked(root->fs_info);
6119         ret = __btrfs_map_block(root->fs_info, bio_op(bio), logical,
6120                                 &map_length, &bbio, mirror_num, 1);
6121         if (ret) {
6122                 btrfs_bio_counter_dec(root->fs_info);
6123                 return ret;
6124         }
6125
6126         total_devs = bbio->num_stripes;
6127         bbio->orig_bio = first_bio;
6128         bbio->private = first_bio->bi_private;
6129         bbio->end_io = first_bio->bi_end_io;
6130         bbio->fs_info = root->fs_info;
6131         atomic_set(&bbio->stripes_pending, bbio->num_stripes);
6132
6133         if ((bbio->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
6134             ((bio_op(bio) == REQ_OP_WRITE) || (mirror_num > 1))) {
6135                 /* In this case, map_length has been set to the length of
6136                    a single stripe; not the whole write */
6137                 if (bio_op(bio) == REQ_OP_WRITE) {
6138                         ret = raid56_parity_write(root, bio, bbio, map_length);
6139                 } else {
6140                         ret = raid56_parity_recover(root, bio, bbio, map_length,
6141                                                     mirror_num, 1);
6142                 }
6143
6144                 btrfs_bio_counter_dec(root->fs_info);
6145                 return ret;
6146         }
6147
6148         if (map_length < length) {
6149                 btrfs_crit(root->fs_info, "mapping failed logical %llu bio len %llu len %llu",
6150                         logical, length, map_length);
6151                 BUG();
6152         }
6153
6154         for (dev_nr = 0; dev_nr < total_devs; dev_nr++) {
6155                 dev = bbio->stripes[dev_nr].dev;
6156                 if (!dev || !dev->bdev ||
6157                     (bio_op(bio) == REQ_OP_WRITE && !dev->writeable)) {
6158                         bbio_error(bbio, first_bio, logical);
6159                         continue;
6160                 }
6161
6162                 if (dev_nr < total_devs - 1) {
6163                         bio = btrfs_bio_clone(first_bio, GFP_NOFS);
6164                         BUG_ON(!bio); /* -ENOMEM */
6165                 } else
6166                         bio = first_bio;
6167
6168                 submit_stripe_bio(root, bbio, bio,
6169                                   bbio->stripes[dev_nr].physical, dev_nr,
6170                                   async_submit);
6171         }
6172         btrfs_bio_counter_dec(root->fs_info);
6173         return 0;
6174 }
6175
6176 struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid,
6177                                        u8 *uuid, u8 *fsid)
6178 {
6179         struct btrfs_device *device;
6180         struct btrfs_fs_devices *cur_devices;
6181
6182         cur_devices = fs_info->fs_devices;
6183         while (cur_devices) {
6184                 if (!fsid ||
6185                     !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
6186                         device = __find_device(&cur_devices->devices,
6187                                                devid, uuid);
6188                         if (device)
6189                                 return device;
6190                 }
6191                 cur_devices = cur_devices->seed;
6192         }
6193         return NULL;
6194 }
6195
6196 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
6197                                             struct btrfs_fs_devices *fs_devices,
6198                                             u64 devid, u8 *dev_uuid)
6199 {
6200         struct btrfs_device *device;
6201
6202         device = btrfs_alloc_device(NULL, &devid, dev_uuid);
6203         if (IS_ERR(device))
6204                 return NULL;
6205
6206         list_add(&device->dev_list, &fs_devices->devices);
6207         device->fs_devices = fs_devices;
6208         fs_devices->num_devices++;
6209
6210         device->missing = 1;
6211         fs_devices->missing_devices++;
6212
6213         return device;
6214 }
6215
6216 /**
6217  * btrfs_alloc_device - allocate struct btrfs_device
6218  * @fs_info:    used only for generating a new devid, can be NULL if
6219  *              devid is provided (i.e. @devid != NULL).
6220  * @devid:      a pointer to devid for this device.  If NULL a new devid
6221  *              is generated.
6222  * @uuid:       a pointer to UUID for this device.  If NULL a new UUID
6223  *              is generated.
6224  *
6225  * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
6226  * on error.  Returned struct is not linked onto any lists and can be
6227  * destroyed with kfree() right away.
6228  */
6229 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
6230                                         const u64 *devid,
6231                                         const u8 *uuid)
6232 {
6233         struct btrfs_device *dev;
6234         u64 tmp;
6235
6236         if (WARN_ON(!devid && !fs_info))
6237                 return ERR_PTR(-EINVAL);
6238
6239         dev = __alloc_device();
6240         if (IS_ERR(dev))
6241                 return dev;
6242
6243         if (devid)
6244                 tmp = *devid;
6245         else {
6246                 int ret;
6247
6248                 ret = find_next_devid(fs_info, &tmp);
6249                 if (ret) {
6250                         kfree(dev);
6251                         return ERR_PTR(ret);
6252                 }
6253         }
6254         dev->devid = tmp;
6255
6256         if (uuid)
6257                 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE);
6258         else
6259                 generate_random_uuid(dev->uuid);
6260
6261         btrfs_init_work(&dev->work, btrfs_submit_helper,
6262                         pending_bios_fn, NULL, NULL);
6263
6264         return dev;
6265 }
6266
6267 /* Return -EIO if any error, otherwise return 0. */
6268 static int btrfs_check_chunk_valid(struct btrfs_root *root,
6269                                    struct extent_buffer *leaf,
6270                                    struct btrfs_chunk *chunk, u64 logical)
6271 {
6272         u64 length;
6273         u64 stripe_len;
6274         u16 num_stripes;
6275         u16 sub_stripes;
6276         u64 type;
6277
6278         length = btrfs_chunk_length(leaf, chunk);
6279         stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6280         num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6281         sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
6282         type = btrfs_chunk_type(leaf, chunk);
6283
6284         if (!num_stripes) {
6285                 btrfs_err(root->fs_info, "invalid chunk num_stripes: %u",
6286                           num_stripes);
6287                 return -EIO;
6288         }
6289         if (!IS_ALIGNED(logical, root->sectorsize)) {
6290                 btrfs_err(root->fs_info,
6291                           "invalid chunk logical %llu", logical);
6292                 return -EIO;
6293         }
6294         if (btrfs_chunk_sector_size(leaf, chunk) != root->sectorsize) {
6295                 btrfs_err(root->fs_info, "invalid chunk sectorsize %u",
6296                           btrfs_chunk_sector_size(leaf, chunk));
6297                 return -EIO;
6298         }
6299         if (!length || !IS_ALIGNED(length, root->sectorsize)) {
6300                 btrfs_err(root->fs_info,
6301                         "invalid chunk length %llu", length);
6302                 return -EIO;
6303         }
6304         if (!is_power_of_2(stripe_len) || stripe_len != BTRFS_STRIPE_LEN) {
6305                 btrfs_err(root->fs_info, "invalid chunk stripe length: %llu",
6306                           stripe_len);
6307                 return -EIO;
6308         }
6309         if (~(BTRFS_BLOCK_GROUP_TYPE_MASK | BTRFS_BLOCK_GROUP_PROFILE_MASK) &
6310             type) {
6311                 btrfs_err(root->fs_info, "unrecognized chunk type: %llu",
6312                           ~(BTRFS_BLOCK_GROUP_TYPE_MASK |
6313                             BTRFS_BLOCK_GROUP_PROFILE_MASK) &
6314                           btrfs_chunk_type(leaf, chunk));
6315                 return -EIO;
6316         }
6317         if ((type & BTRFS_BLOCK_GROUP_RAID10 && sub_stripes != 2) ||
6318             (type & BTRFS_BLOCK_GROUP_RAID1 && num_stripes < 1) ||
6319             (type & BTRFS_BLOCK_GROUP_RAID5 && num_stripes < 2) ||
6320             (type & BTRFS_BLOCK_GROUP_RAID6 && num_stripes < 3) ||
6321             (type & BTRFS_BLOCK_GROUP_DUP && num_stripes > 2) ||
6322             ((type & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 &&
6323              num_stripes != 1)) {
6324                 btrfs_err(root->fs_info,
6325                         "invalid num_stripes:sub_stripes %u:%u for profile %llu",
6326                         num_stripes, sub_stripes,
6327                         type & BTRFS_BLOCK_GROUP_PROFILE_MASK);
6328                 return -EIO;
6329         }
6330
6331         return 0;
6332 }
6333
6334 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
6335                           struct extent_buffer *leaf,
6336                           struct btrfs_chunk *chunk)
6337 {
6338         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
6339         struct map_lookup *map;
6340         struct extent_map *em;
6341         u64 logical;
6342         u64 length;
6343         u64 stripe_len;
6344         u64 devid;
6345         u8 uuid[BTRFS_UUID_SIZE];
6346         int num_stripes;
6347         int ret;
6348         int i;
6349
6350         logical = key->offset;
6351         length = btrfs_chunk_length(leaf, chunk);
6352         stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6353         num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6354
6355         ret = btrfs_check_chunk_valid(root, leaf, chunk, logical);
6356         if (ret)
6357                 return ret;
6358
6359         read_lock(&map_tree->map_tree.lock);
6360         em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
6361         read_unlock(&map_tree->map_tree.lock);
6362
6363         /* already mapped? */
6364         if (em && em->start <= logical && em->start + em->len > logical) {
6365                 free_extent_map(em);
6366                 return 0;
6367         } else if (em) {
6368                 free_extent_map(em);
6369         }
6370
6371         em = alloc_extent_map();
6372         if (!em)
6373                 return -ENOMEM;
6374         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
6375         if (!map) {
6376                 free_extent_map(em);
6377                 return -ENOMEM;
6378         }
6379
6380         set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
6381         em->map_lookup = map;
6382         em->start = logical;
6383         em->len = length;
6384         em->orig_start = 0;
6385         em->block_start = 0;
6386         em->block_len = em->len;
6387
6388         map->num_stripes = num_stripes;
6389         map->io_width = btrfs_chunk_io_width(leaf, chunk);
6390         map->io_align = btrfs_chunk_io_align(leaf, chunk);
6391         map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
6392         map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6393         map->type = btrfs_chunk_type(leaf, chunk);
6394         map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
6395         for (i = 0; i < num_stripes; i++) {
6396                 map->stripes[i].physical =
6397                         btrfs_stripe_offset_nr(leaf, chunk, i);
6398                 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
6399                 read_extent_buffer(leaf, uuid, (unsigned long)
6400                                    btrfs_stripe_dev_uuid_nr(chunk, i),
6401                                    BTRFS_UUID_SIZE);
6402                 map->stripes[i].dev = btrfs_find_device(root->fs_info, devid,
6403                                                         uuid, NULL);
6404                 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
6405                         free_extent_map(em);
6406                         return -EIO;
6407                 }
6408                 if (!map->stripes[i].dev) {
6409                         map->stripes[i].dev =
6410                                 add_missing_dev(root, root->fs_info->fs_devices,
6411                                                 devid, uuid);
6412                         if (!map->stripes[i].dev) {
6413                                 free_extent_map(em);
6414                                 return -EIO;
6415                         }
6416                         btrfs_warn(root->fs_info, "devid %llu uuid %pU is missing",
6417                                                 devid, uuid);
6418                 }
6419                 map->stripes[i].dev->in_fs_metadata = 1;
6420         }
6421
6422         write_lock(&map_tree->map_tree.lock);
6423         ret = add_extent_mapping(&map_tree->map_tree, em, 0);
6424         write_unlock(&map_tree->map_tree.lock);
6425         BUG_ON(ret); /* Tree corruption */
6426         free_extent_map(em);
6427
6428         return 0;
6429 }
6430
6431 static void fill_device_from_item(struct extent_buffer *leaf,
6432                                  struct btrfs_dev_item *dev_item,
6433                                  struct btrfs_device *device)
6434 {
6435         unsigned long ptr;
6436
6437         device->devid = btrfs_device_id(leaf, dev_item);
6438         device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
6439         device->total_bytes = device->disk_total_bytes;
6440         device->commit_total_bytes = device->disk_total_bytes;
6441         device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
6442         device->commit_bytes_used = device->bytes_used;
6443         device->type = btrfs_device_type(leaf, dev_item);
6444         device->io_align = btrfs_device_io_align(leaf, dev_item);
6445         device->io_width = btrfs_device_io_width(leaf, dev_item);
6446         device->sector_size = btrfs_device_sector_size(leaf, dev_item);
6447         WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
6448         device->is_tgtdev_for_dev_replace = 0;
6449
6450         ptr = btrfs_device_uuid(dev_item);
6451         read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
6452 }
6453
6454 static struct btrfs_fs_devices *open_seed_devices(struct btrfs_root *root,
6455                                                   u8 *fsid)
6456 {
6457         struct btrfs_fs_devices *fs_devices;
6458         int ret;
6459
6460         BUG_ON(!mutex_is_locked(&uuid_mutex));
6461
6462         fs_devices = root->fs_info->fs_devices->seed;
6463         while (fs_devices) {
6464                 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE))
6465                         return fs_devices;
6466
6467                 fs_devices = fs_devices->seed;
6468         }
6469
6470         fs_devices = find_fsid(fsid);
6471         if (!fs_devices) {
6472                 if (!btrfs_test_opt(root, DEGRADED))
6473                         return ERR_PTR(-ENOENT);
6474
6475                 fs_devices = alloc_fs_devices(fsid);
6476                 if (IS_ERR(fs_devices))
6477                         return fs_devices;
6478
6479                 fs_devices->seeding = 1;
6480                 fs_devices->opened = 1;
6481                 return fs_devices;
6482         }
6483
6484         fs_devices = clone_fs_devices(fs_devices);
6485         if (IS_ERR(fs_devices))
6486                 return fs_devices;
6487
6488         ret = __btrfs_open_devices(fs_devices, FMODE_READ,
6489                                    root->fs_info->bdev_holder);
6490         if (ret) {
6491                 free_fs_devices(fs_devices);
6492                 fs_devices = ERR_PTR(ret);
6493                 goto out;
6494         }
6495
6496         if (!fs_devices->seeding) {
6497                 __btrfs_close_devices(fs_devices);
6498                 free_fs_devices(fs_devices);
6499                 fs_devices = ERR_PTR(-EINVAL);
6500                 goto out;
6501         }
6502
6503         fs_devices->seed = root->fs_info->fs_devices->seed;
6504         root->fs_info->fs_devices->seed = fs_devices;
6505 out:
6506         return fs_devices;
6507 }
6508
6509 static int read_one_dev(struct btrfs_root *root,
6510                         struct extent_buffer *leaf,
6511                         struct btrfs_dev_item *dev_item)
6512 {
6513         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
6514         struct btrfs_device *device;
6515         u64 devid;
6516         int ret;
6517         u8 fs_uuid[BTRFS_UUID_SIZE];
6518         u8 dev_uuid[BTRFS_UUID_SIZE];
6519
6520         devid = btrfs_device_id(leaf, dev_item);
6521         read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
6522                            BTRFS_UUID_SIZE);
6523         read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
6524                            BTRFS_UUID_SIZE);
6525
6526         if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
6527                 fs_devices = open_seed_devices(root, fs_uuid);
6528                 if (IS_ERR(fs_devices))
6529                         return PTR_ERR(fs_devices);
6530         }
6531
6532         device = btrfs_find_device(root->fs_info, devid, dev_uuid, fs_uuid);
6533         if (!device) {
6534                 if (!btrfs_test_opt(root, DEGRADED))
6535                         return -EIO;
6536
6537                 device = add_missing_dev(root, fs_devices, devid, dev_uuid);
6538                 if (!device)
6539                         return -ENOMEM;
6540                 btrfs_warn(root->fs_info, "devid %llu uuid %pU missing",
6541                                 devid, dev_uuid);
6542         } else {
6543                 if (!device->bdev && !btrfs_test_opt(root, DEGRADED))
6544                         return -EIO;
6545
6546                 if(!device->bdev && !device->missing) {
6547                         /*
6548                          * this happens when a device that was properly setup
6549                          * in the device info lists suddenly goes bad.
6550                          * device->bdev is NULL, and so we have to set
6551                          * device->missing to one here
6552                          */
6553                         device->fs_devices->missing_devices++;
6554                         device->missing = 1;
6555                 }
6556
6557                 /* Move the device to its own fs_devices */
6558                 if (device->fs_devices != fs_devices) {
6559                         ASSERT(device->missing);
6560
6561                         list_move(&device->dev_list, &fs_devices->devices);
6562                         device->fs_devices->num_devices--;
6563                         fs_devices->num_devices++;
6564
6565                         device->fs_devices->missing_devices--;
6566                         fs_devices->missing_devices++;
6567
6568                         device->fs_devices = fs_devices;
6569                 }
6570         }
6571
6572         if (device->fs_devices != root->fs_info->fs_devices) {
6573                 BUG_ON(device->writeable);
6574                 if (device->generation !=
6575                     btrfs_device_generation(leaf, dev_item))
6576                         return -EINVAL;
6577         }
6578
6579         fill_device_from_item(leaf, dev_item, device);
6580         device->in_fs_metadata = 1;
6581         if (device->writeable && !device->is_tgtdev_for_dev_replace) {
6582                 device->fs_devices->total_rw_bytes += device->total_bytes;
6583                 spin_lock(&root->fs_info->free_chunk_lock);
6584                 root->fs_info->free_chunk_space += device->total_bytes -
6585                         device->bytes_used;
6586                 spin_unlock(&root->fs_info->free_chunk_lock);
6587         }
6588         ret = 0;
6589         return ret;
6590 }
6591
6592 int btrfs_read_sys_array(struct btrfs_root *root)
6593 {
6594         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
6595         struct extent_buffer *sb;
6596         struct btrfs_disk_key *disk_key;
6597         struct btrfs_chunk *chunk;
6598         u8 *array_ptr;
6599         unsigned long sb_array_offset;
6600         int ret = 0;
6601         u32 num_stripes;
6602         u32 array_size;
6603         u32 len = 0;
6604         u32 cur_offset;
6605         u64 type;
6606         struct btrfs_key key;
6607
6608         ASSERT(BTRFS_SUPER_INFO_SIZE <= root->nodesize);
6609         /*
6610          * This will create extent buffer of nodesize, superblock size is
6611          * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will
6612          * overallocate but we can keep it as-is, only the first page is used.
6613          */
6614         sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET);
6615         if (IS_ERR(sb))
6616                 return PTR_ERR(sb);
6617         set_extent_buffer_uptodate(sb);
6618         btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
6619         /*
6620          * The sb extent buffer is artificial and just used to read the system array.
6621          * set_extent_buffer_uptodate() call does not properly mark all it's
6622          * pages up-to-date when the page is larger: extent does not cover the
6623          * whole page and consequently check_page_uptodate does not find all
6624          * the page's extents up-to-date (the hole beyond sb),
6625          * write_extent_buffer then triggers a WARN_ON.
6626          *
6627          * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
6628          * but sb spans only this function. Add an explicit SetPageUptodate call
6629          * to silence the warning eg. on PowerPC 64.
6630          */
6631         if (PAGE_SIZE > BTRFS_SUPER_INFO_SIZE)
6632                 SetPageUptodate(sb->pages[0]);
6633
6634         write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
6635         array_size = btrfs_super_sys_array_size(super_copy);
6636
6637         array_ptr = super_copy->sys_chunk_array;
6638         sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array);
6639         cur_offset = 0;
6640
6641         while (cur_offset < array_size) {
6642                 disk_key = (struct btrfs_disk_key *)array_ptr;
6643                 len = sizeof(*disk_key);
6644                 if (cur_offset + len > array_size)
6645                         goto out_short_read;
6646
6647                 btrfs_disk_key_to_cpu(&key, disk_key);
6648
6649                 array_ptr += len;
6650                 sb_array_offset += len;
6651                 cur_offset += len;
6652
6653                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
6654                         chunk = (struct btrfs_chunk *)sb_array_offset;
6655                         /*
6656                          * At least one btrfs_chunk with one stripe must be
6657                          * present, exact stripe count check comes afterwards
6658                          */
6659                         len = btrfs_chunk_item_size(1);
6660                         if (cur_offset + len > array_size)
6661                                 goto out_short_read;
6662
6663                         num_stripes = btrfs_chunk_num_stripes(sb, chunk);
6664                         if (!num_stripes) {
6665                                 printk(KERN_ERR
6666             "BTRFS: invalid number of stripes %u in sys_array at offset %u\n",
6667                                         num_stripes, cur_offset);
6668                                 ret = -EIO;
6669                                 break;
6670                         }
6671
6672                         type = btrfs_chunk_type(sb, chunk);
6673                         if ((type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) {
6674                                 btrfs_err(root->fs_info,
6675                             "invalid chunk type %llu in sys_array at offset %u",
6676                                         type, cur_offset);
6677                                 ret = -EIO;
6678                                 break;
6679                         }
6680
6681                         len = btrfs_chunk_item_size(num_stripes);
6682                         if (cur_offset + len > array_size)
6683                                 goto out_short_read;
6684
6685                         ret = read_one_chunk(root, &key, sb, chunk);
6686                         if (ret)
6687                                 break;
6688                 } else {
6689                         printk(KERN_ERR
6690                 "BTRFS: unexpected item type %u in sys_array at offset %u\n",
6691                                 (u32)key.type, cur_offset);
6692                         ret = -EIO;
6693                         break;
6694                 }
6695                 array_ptr += len;
6696                 sb_array_offset += len;
6697                 cur_offset += len;
6698         }
6699         clear_extent_buffer_uptodate(sb);
6700         free_extent_buffer_stale(sb);
6701         return ret;
6702
6703 out_short_read:
6704         printk(KERN_ERR "BTRFS: sys_array too short to read %u bytes at offset %u\n",
6705                         len, cur_offset);
6706         clear_extent_buffer_uptodate(sb);
6707         free_extent_buffer_stale(sb);
6708         return -EIO;
6709 }
6710
6711 int btrfs_read_chunk_tree(struct btrfs_root *root)
6712 {
6713         struct btrfs_path *path;
6714         struct extent_buffer *leaf;
6715         struct btrfs_key key;
6716         struct btrfs_key found_key;
6717         int ret;
6718         int slot;
6719         u64 total_dev = 0;
6720
6721         root = root->fs_info->chunk_root;
6722
6723         path = btrfs_alloc_path();
6724         if (!path)
6725                 return -ENOMEM;
6726
6727         mutex_lock(&uuid_mutex);
6728         lock_chunks(root);
6729
6730         /*
6731          * Read all device items, and then all the chunk items. All
6732          * device items are found before any chunk item (their object id
6733          * is smaller than the lowest possible object id for a chunk
6734          * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
6735          */
6736         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
6737         key.offset = 0;
6738         key.type = 0;
6739         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6740         if (ret < 0)
6741                 goto error;
6742         while (1) {
6743                 leaf = path->nodes[0];
6744                 slot = path->slots[0];
6745                 if (slot >= btrfs_header_nritems(leaf)) {
6746                         ret = btrfs_next_leaf(root, path);
6747                         if (ret == 0)
6748                                 continue;
6749                         if (ret < 0)
6750                                 goto error;
6751                         break;
6752                 }
6753                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6754                 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
6755                         struct btrfs_dev_item *dev_item;
6756                         dev_item = btrfs_item_ptr(leaf, slot,
6757                                                   struct btrfs_dev_item);
6758                         ret = read_one_dev(root, leaf, dev_item);
6759                         if (ret)
6760                                 goto error;
6761                         total_dev++;
6762                 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
6763                         struct btrfs_chunk *chunk;
6764                         chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
6765                         ret = read_one_chunk(root, &found_key, leaf, chunk);
6766                         if (ret)
6767                                 goto error;
6768                 }
6769                 path->slots[0]++;
6770         }
6771
6772         /*
6773          * After loading chunk tree, we've got all device information,
6774          * do another round of validation checks.
6775          */
6776         if (total_dev != root->fs_info->fs_devices->total_devices) {
6777                 btrfs_err(root->fs_info,
6778            "super_num_devices %llu mismatch with num_devices %llu found here",
6779                           btrfs_super_num_devices(root->fs_info->super_copy),
6780                           total_dev);
6781                 ret = -EINVAL;
6782                 goto error;
6783         }
6784         if (btrfs_super_total_bytes(root->fs_info->super_copy) <
6785             root->fs_info->fs_devices->total_rw_bytes) {
6786                 btrfs_err(root->fs_info,
6787         "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu",
6788                           btrfs_super_total_bytes(root->fs_info->super_copy),
6789                           root->fs_info->fs_devices->total_rw_bytes);
6790                 ret = -EINVAL;
6791                 goto error;
6792         }
6793         ret = 0;
6794 error:
6795         unlock_chunks(root);
6796         mutex_unlock(&uuid_mutex);
6797
6798         btrfs_free_path(path);
6799         return ret;
6800 }
6801
6802 void btrfs_init_devices_late(struct btrfs_fs_info *fs_info)
6803 {
6804         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6805         struct btrfs_device *device;
6806
6807         while (fs_devices) {
6808                 mutex_lock(&fs_devices->device_list_mutex);
6809                 list_for_each_entry(device, &fs_devices->devices, dev_list)
6810                         device->dev_root = fs_info->dev_root;
6811                 mutex_unlock(&fs_devices->device_list_mutex);
6812
6813                 fs_devices = fs_devices->seed;
6814         }
6815 }
6816
6817 static void __btrfs_reset_dev_stats(struct btrfs_device *dev)
6818 {
6819         int i;
6820
6821         for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6822                 btrfs_dev_stat_reset(dev, i);
6823 }
6824
6825 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
6826 {
6827         struct btrfs_key key;
6828         struct btrfs_key found_key;
6829         struct btrfs_root *dev_root = fs_info->dev_root;
6830         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6831         struct extent_buffer *eb;
6832         int slot;
6833         int ret = 0;
6834         struct btrfs_device *device;
6835         struct btrfs_path *path = NULL;
6836         int i;
6837
6838         path = btrfs_alloc_path();
6839         if (!path) {
6840                 ret = -ENOMEM;
6841                 goto out;
6842         }
6843
6844         mutex_lock(&fs_devices->device_list_mutex);
6845         list_for_each_entry(device, &fs_devices->devices, dev_list) {
6846                 int item_size;
6847                 struct btrfs_dev_stats_item *ptr;
6848
6849                 key.objectid = BTRFS_DEV_STATS_OBJECTID;
6850                 key.type = BTRFS_PERSISTENT_ITEM_KEY;
6851                 key.offset = device->devid;
6852                 ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
6853                 if (ret) {
6854                         __btrfs_reset_dev_stats(device);
6855                         device->dev_stats_valid = 1;
6856                         btrfs_release_path(path);
6857                         continue;
6858                 }
6859                 slot = path->slots[0];
6860                 eb = path->nodes[0];
6861                 btrfs_item_key_to_cpu(eb, &found_key, slot);
6862                 item_size = btrfs_item_size_nr(eb, slot);
6863
6864                 ptr = btrfs_item_ptr(eb, slot,
6865                                      struct btrfs_dev_stats_item);
6866
6867                 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
6868                         if (item_size >= (1 + i) * sizeof(__le64))
6869                                 btrfs_dev_stat_set(device, i,
6870                                         btrfs_dev_stats_value(eb, ptr, i));
6871                         else
6872                                 btrfs_dev_stat_reset(device, i);
6873                 }
6874
6875                 device->dev_stats_valid = 1;
6876                 btrfs_dev_stat_print_on_load(device);
6877                 btrfs_release_path(path);
6878         }
6879         mutex_unlock(&fs_devices->device_list_mutex);
6880
6881 out:
6882         btrfs_free_path(path);
6883         return ret < 0 ? ret : 0;
6884 }
6885
6886 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
6887                                 struct btrfs_root *dev_root,
6888                                 struct btrfs_device *device)
6889 {
6890         struct btrfs_path *path;
6891         struct btrfs_key key;
6892         struct extent_buffer *eb;
6893         struct btrfs_dev_stats_item *ptr;
6894         int ret;
6895         int i;
6896
6897         key.objectid = BTRFS_DEV_STATS_OBJECTID;
6898         key.type = BTRFS_PERSISTENT_ITEM_KEY;
6899         key.offset = device->devid;
6900
6901         path = btrfs_alloc_path();
6902         BUG_ON(!path);
6903         ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
6904         if (ret < 0) {
6905                 btrfs_warn_in_rcu(dev_root->fs_info,
6906                         "error %d while searching for dev_stats item for device %s",
6907                               ret, rcu_str_deref(device->name));
6908                 goto out;
6909         }
6910
6911         if (ret == 0 &&
6912             btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
6913                 /* need to delete old one and insert a new one */
6914                 ret = btrfs_del_item(trans, dev_root, path);
6915                 if (ret != 0) {
6916                         btrfs_warn_in_rcu(dev_root->fs_info,
6917                                 "delete too small dev_stats item for device %s failed %d",
6918                                       rcu_str_deref(device->name), ret);
6919                         goto out;
6920                 }
6921                 ret = 1;
6922         }
6923
6924         if (ret == 1) {
6925                 /* need to insert a new item */
6926                 btrfs_release_path(path);
6927                 ret = btrfs_insert_empty_item(trans, dev_root, path,
6928                                               &key, sizeof(*ptr));
6929                 if (ret < 0) {
6930                         btrfs_warn_in_rcu(dev_root->fs_info,
6931                                 "insert dev_stats item for device %s failed %d",
6932                                 rcu_str_deref(device->name), ret);
6933                         goto out;
6934                 }
6935         }
6936
6937         eb = path->nodes[0];
6938         ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
6939         for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6940                 btrfs_set_dev_stats_value(eb, ptr, i,
6941                                           btrfs_dev_stat_read(device, i));
6942         btrfs_mark_buffer_dirty(eb);
6943
6944 out:
6945         btrfs_free_path(path);
6946         return ret;
6947 }
6948
6949 /*
6950  * called from commit_transaction. Writes all changed device stats to disk.
6951  */
6952 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans,
6953                         struct btrfs_fs_info *fs_info)
6954 {
6955         struct btrfs_root *dev_root = fs_info->dev_root;
6956         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6957         struct btrfs_device *device;
6958         int stats_cnt;
6959         int ret = 0;
6960
6961         mutex_lock(&fs_devices->device_list_mutex);
6962         list_for_each_entry(device, &fs_devices->devices, dev_list) {
6963                 if (!device->dev_stats_valid || !btrfs_dev_stats_dirty(device))
6964                         continue;
6965
6966                 stats_cnt = atomic_read(&device->dev_stats_ccnt);
6967                 ret = update_dev_stat_item(trans, dev_root, device);
6968                 if (!ret)
6969                         atomic_sub(stats_cnt, &device->dev_stats_ccnt);
6970         }
6971         mutex_unlock(&fs_devices->device_list_mutex);
6972
6973         return ret;
6974 }
6975
6976 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
6977 {
6978         btrfs_dev_stat_inc(dev, index);
6979         btrfs_dev_stat_print_on_error(dev);
6980 }
6981
6982 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
6983 {
6984         if (!dev->dev_stats_valid)
6985                 return;
6986         btrfs_err_rl_in_rcu(dev->dev_root->fs_info,
6987                 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
6988                            rcu_str_deref(dev->name),
6989                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
6990                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
6991                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
6992                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
6993                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
6994 }
6995
6996 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
6997 {
6998         int i;
6999
7000         for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7001                 if (btrfs_dev_stat_read(dev, i) != 0)
7002                         break;
7003         if (i == BTRFS_DEV_STAT_VALUES_MAX)
7004                 return; /* all values == 0, suppress message */
7005
7006         btrfs_info_in_rcu(dev->dev_root->fs_info,
7007                 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7008                rcu_str_deref(dev->name),
7009                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7010                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7011                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7012                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7013                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7014 }
7015
7016 int btrfs_get_dev_stats(struct btrfs_root *root,
7017                         struct btrfs_ioctl_get_dev_stats *stats)
7018 {
7019         struct btrfs_device *dev;
7020         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
7021         int i;
7022
7023         mutex_lock(&fs_devices->device_list_mutex);
7024         dev = btrfs_find_device(root->fs_info, stats->devid, NULL, NULL);
7025         mutex_unlock(&fs_devices->device_list_mutex);
7026
7027         if (!dev) {
7028                 btrfs_warn(root->fs_info, "get dev_stats failed, device not found");
7029                 return -ENODEV;
7030         } else if (!dev->dev_stats_valid) {
7031                 btrfs_warn(root->fs_info, "get dev_stats failed, not yet valid");
7032                 return -ENODEV;
7033         } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
7034                 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7035                         if (stats->nr_items > i)
7036                                 stats->values[i] =
7037                                         btrfs_dev_stat_read_and_reset(dev, i);
7038                         else
7039                                 btrfs_dev_stat_reset(dev, i);
7040                 }
7041         } else {
7042                 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7043                         if (stats->nr_items > i)
7044                                 stats->values[i] = btrfs_dev_stat_read(dev, i);
7045         }
7046         if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
7047                 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
7048         return 0;
7049 }
7050
7051 void btrfs_scratch_superblocks(struct block_device *bdev, char *device_path)
7052 {
7053         struct buffer_head *bh;
7054         struct btrfs_super_block *disk_super;
7055         int copy_num;
7056
7057         if (!bdev)
7058                 return;
7059
7060         for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX;
7061                 copy_num++) {
7062
7063                 if (btrfs_read_dev_one_super(bdev, copy_num, &bh))
7064                         continue;
7065
7066                 disk_super = (struct btrfs_super_block *)bh->b_data;
7067
7068                 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
7069                 set_buffer_dirty(bh);
7070                 sync_dirty_buffer(bh);
7071                 brelse(bh);
7072         }
7073
7074         /* Notify udev that device has changed */
7075         btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
7076
7077         /* Update ctime/mtime for device path for libblkid */
7078         update_dev_time(device_path);
7079 }
7080
7081 /*
7082  * Update the size of all devices, which is used for writing out the
7083  * super blocks.
7084  */
7085 void btrfs_update_commit_device_size(struct btrfs_fs_info *fs_info)
7086 {
7087         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7088         struct btrfs_device *curr, *next;
7089
7090         if (list_empty(&fs_devices->resized_devices))
7091                 return;
7092
7093         mutex_lock(&fs_devices->device_list_mutex);
7094         lock_chunks(fs_info->dev_root);
7095         list_for_each_entry_safe(curr, next, &fs_devices->resized_devices,
7096                                  resized_list) {
7097                 list_del_init(&curr->resized_list);
7098                 curr->commit_total_bytes = curr->disk_total_bytes;
7099         }
7100         unlock_chunks(fs_info->dev_root);
7101         mutex_unlock(&fs_devices->device_list_mutex);
7102 }
7103
7104 /* Must be invoked during the transaction commit */
7105 void btrfs_update_commit_device_bytes_used(struct btrfs_root *root,
7106                                         struct btrfs_transaction *transaction)
7107 {
7108         struct extent_map *em;
7109         struct map_lookup *map;
7110         struct btrfs_device *dev;
7111         int i;
7112
7113         if (list_empty(&transaction->pending_chunks))
7114                 return;
7115
7116         /* In order to kick the device replace finish process */
7117         lock_chunks(root);
7118         list_for_each_entry(em, &transaction->pending_chunks, list) {
7119                 map = em->map_lookup;
7120
7121                 for (i = 0; i < map->num_stripes; i++) {
7122                         dev = map->stripes[i].dev;
7123                         dev->commit_bytes_used = dev->bytes_used;
7124                 }
7125         }
7126         unlock_chunks(root);
7127 }
7128
7129 void btrfs_set_fs_info_ptr(struct btrfs_fs_info *fs_info)
7130 {
7131         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7132         while (fs_devices) {
7133                 fs_devices->fs_info = fs_info;
7134                 fs_devices = fs_devices->seed;
7135         }
7136 }
7137
7138 void btrfs_reset_fs_info_ptr(struct btrfs_fs_info *fs_info)
7139 {
7140         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7141         while (fs_devices) {
7142                 fs_devices->fs_info = NULL;
7143                 fs_devices = fs_devices->seed;
7144         }
7145 }
7146
7147 static void btrfs_close_one_device(struct btrfs_device *device)
7148 {
7149         struct btrfs_fs_devices *fs_devices = device->fs_devices;
7150         struct btrfs_device *new_device;
7151         struct rcu_string *name;
7152
7153         if (device->bdev)
7154                 fs_devices->open_devices--;
7155
7156         if (device->writeable &&
7157             device->devid != BTRFS_DEV_REPLACE_DEVID) {
7158                 list_del_init(&device->dev_alloc_list);
7159                 fs_devices->rw_devices--;
7160         }
7161
7162         if (device->missing)
7163                 fs_devices->missing_devices--;
7164
7165         new_device = btrfs_alloc_device(NULL, &device->devid,
7166                                         device->uuid);
7167         BUG_ON(IS_ERR(new_device)); /* -ENOMEM */
7168
7169         /* Safe because we are under uuid_mutex */
7170         if (device->name) {
7171                 name = rcu_string_strdup(device->name->str, GFP_NOFS);
7172                 BUG_ON(!name); /* -ENOMEM */
7173                 rcu_assign_pointer(new_device->name, name);
7174         }
7175
7176         list_replace_rcu(&device->dev_list, &new_device->dev_list);
7177         new_device->fs_devices = device->fs_devices;
7178
7179         call_rcu(&device->rcu, free_device);
7180 }