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_fs_info *fs_info);
 138 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info);
 139 static void __btrfs_reset_dev_stats(struct btrfs_device *dev);
 140 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev);
 141 static void btrfs_dev_stat_print_on_load(struct btrfs_device *device);
 142 static int __btrfs_map_block(struct btrfs_fs_info *fs_info,
 143                              enum btrfs_map_op op,
 144                              u64 logical, u64 *length,
 145                              struct btrfs_bio **bbio_ret,
 146                              int mirror_num, int need_raid_map);
 147
 148 DEFINE_MUTEX(uuid_mutex);
 149 static LIST_HEAD(fs_uuids);
 150 struct list_head *btrfs_get_fs_uuids(void)
 151 {
 152         return &fs_uuids;
 153 }
 154
 155 static struct btrfs_fs_devices *__alloc_fs_devices(void)
 156 {
 157         struct btrfs_fs_devices *fs_devs;
 158
 159         fs_devs = kzalloc(sizeof(*fs_devs), GFP_KERNEL);
 160         if (!fs_devs)
 161                 return ERR_PTR(-ENOMEM);
 162
 163         mutex_init(&fs_devs->device_list_mutex);
 164
 165         INIT_LIST_HEAD(&fs_devs->devices);
 166         INIT_LIST_HEAD(&fs_devs->resized_devices);
 167         INIT_LIST_HEAD(&fs_devs->alloc_list);
 168         INIT_LIST_HEAD(&fs_devs->list);
 169
 170         return fs_devs;
 171 }
 172
 173 /**
 174  * alloc_fs_devices - allocate struct btrfs_fs_devices
 175  * @fsid:       a pointer to UUID for this FS.  If NULL a new UUID is
 176  *              generated.
 177  *
 178  * Return: a pointer to a new &struct btrfs_fs_devices on success;
 179  * ERR_PTR() on error.  Returned struct is not linked onto any lists and
 180  * can be destroyed with kfree() right away.
 181  */
 182 static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid)
 183 {
 184         struct btrfs_fs_devices *fs_devs;
 185
 186         fs_devs = __alloc_fs_devices();
 187         if (IS_ERR(fs_devs))
 188                 return fs_devs;
 189
 190         if (fsid)
 191                 memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE);
 192         else
 193                 generate_random_uuid(fs_devs->fsid);
 194
 195         return fs_devs;
 196 }
 197
 198 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
 199 {
 200         struct btrfs_device *device;
 201         WARN_ON(fs_devices->opened);
 202         while (!list_empty(&fs_devices->devices)) {
 203                 device = list_entry(fs_devices->devices.next,
 204                                     struct btrfs_device, dev_list);
 205                 list_del(&device->dev_list);
 206                 rcu_string_free(device->name);
 207                 kfree(device);
 208         }
 209         kfree(fs_devices);
 210 }
 211
 212 static void btrfs_kobject_uevent(struct block_device *bdev,
 213                                  enum kobject_action action)
 214 {
 215         int ret;
 216
 217         ret = kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, action);
 218         if (ret)
 219                 pr_warn("BTRFS: Sending event '%d' to kobject: '%s' (%p): failed\n",
 220                         action,
 221                         kobject_name(&disk_to_dev(bdev->bd_disk)->kobj),
 222                         &disk_to_dev(bdev->bd_disk)->kobj);
 223 }
 224
 225 void btrfs_cleanup_fs_uuids(void)
 226 {
 227         struct btrfs_fs_devices *fs_devices;
 228
 229         while (!list_empty(&fs_uuids)) {
 230                 fs_devices = list_entry(fs_uuids.next,
 231                                         struct btrfs_fs_devices, list);
 232                 list_del(&fs_devices->list);
 233                 free_fs_devices(fs_devices);
 234         }
 235 }
 236
 237 static struct btrfs_device *__alloc_device(void)
 238 {
 239         struct btrfs_device *dev;
 240
 241         dev = kzalloc(sizeof(*dev), GFP_KERNEL);
 242         if (!dev)
 243                 return ERR_PTR(-ENOMEM);
 244
 245         INIT_LIST_HEAD(&dev->dev_list);
 246         INIT_LIST_HEAD(&dev->dev_alloc_list);
 247         INIT_LIST_HEAD(&dev->resized_list);
 248
 249         spin_lock_init(&dev->io_lock);
 250
 251         spin_lock_init(&dev->reada_lock);
 252         atomic_set(&dev->reada_in_flight, 0);
 253         atomic_set(&dev->dev_stats_ccnt, 0);
 254         btrfs_device_data_ordered_init(dev);
 255         INIT_RADIX_TREE(&dev->reada_zones, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
 256         INIT_RADIX_TREE(&dev->reada_extents, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
 257
 258         return dev;
 259 }
 260
 261 static noinline struct btrfs_device *__find_device(struct list_head *head,
 262                                                    u64 devid, u8 *uuid)
 263 {
 264         struct btrfs_device *dev;
 265
 266         list_for_each_entry(dev, head, dev_list) {
 267                 if (dev->devid == devid &&
 268                     (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
 269                         return dev;
 270                 }
 271         }
 272         return NULL;
 273 }
 274
 275 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
 276 {
 277         struct btrfs_fs_devices *fs_devices;
 278
 279         list_for_each_entry(fs_devices, &fs_uuids, list) {
 280                 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
 281                         return fs_devices;
 282         }
 283         return NULL;
 284 }
 285
 286 static int
 287 btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder,
 288                       int flush, struct block_device **bdev,
 289                       struct buffer_head **bh)
 290 {
 291         int ret;
 292
 293         *bdev = blkdev_get_by_path(device_path, flags, holder);
 294
 295         if (IS_ERR(*bdev)) {
 296                 ret = PTR_ERR(*bdev);
 297                 goto error;
 298         }
 299
 300         if (flush)
 301                 filemap_write_and_wait((*bdev)->bd_inode->i_mapping);
 302         ret = set_blocksize(*bdev, 4096);
 303         if (ret) {
 304                 blkdev_put(*bdev, flags);
 305                 goto error;
 306         }
 307         invalidate_bdev(*bdev);
 308         *bh = btrfs_read_dev_super(*bdev);
 309         if (IS_ERR(*bh)) {
 310                 ret = PTR_ERR(*bh);
 311                 blkdev_put(*bdev, flags);
 312                 goto error;
 313         }
 314
 315         return 0;
 316
 317 error:
 318         *bdev = NULL;
 319         *bh = NULL;
 320         return ret;
 321 }
 322
 323 static void requeue_list(struct btrfs_pending_bios *pending_bios,
 324                         struct bio *head, struct bio *tail)
 325 {
 326
 327         struct bio *old_head;
 328
 329         old_head = pending_bios->head;
 330         pending_bios->head = head;
 331         if (pending_bios->tail)
 332                 tail->bi_next = old_head;
 333         else
 334                 pending_bios->tail = tail;
 335 }
 336
 337 /*
 338  * we try to collect pending bios for a device so we don't get a large
 339  * number of procs sending bios down to the same device.  This greatly
 340  * improves the schedulers ability to collect and merge the bios.
 341  *
 342  * But, it also turns into a long list of bios to process and that is sure
 343  * to eventually make the worker thread block.  The solution here is to
 344  * make some progress and then put this work struct back at the end of
 345  * the list if the block device is congested.  This way, multiple devices
 346  * can make progress from a single worker thread.
 347  */
 348 static noinline void run_scheduled_bios(struct btrfs_device *device)
 349 {
 350         struct btrfs_fs_info *fs_info = device->fs_info;
 351         struct bio *pending;
 352         struct backing_dev_info *bdi;
 353         struct btrfs_pending_bios *pending_bios;
 354         struct bio *tail;
 355         struct bio *cur;
 356         int again = 0;
 357         unsigned long num_run;
 358         unsigned long batch_run = 0;
 359         unsigned long limit;
 360         unsigned long last_waited = 0;
 361         int force_reg = 0;
 362         int sync_pending = 0;
 363         struct blk_plug plug;
 364
 365         /*
 366          * this function runs all the bios we've collected for
 367          * a particular device.  We don't want to wander off to
 368          * another device without first sending all of these down.
 369          * So, setup a plug here and finish it off before we return
 370          */
 371         blk_start_plug(&plug);
 372
 373         bdi = device->bdev->bd_bdi;
 374         limit = btrfs_async_submit_limit(fs_info);
 375         limit = limit * 2 / 3;
 376
 377 loop:
 378         spin_lock(&device->io_lock);
 379
 380 loop_lock:
 381         num_run = 0;
 382
 383         /* take all the bios off the list at once and process them
 384          * later on (without the lock held).  But, remember the
 385          * tail and other pointers so the bios can be properly reinserted
 386          * into the list if we hit congestion
 387          */
 388         if (!force_reg && device->pending_sync_bios.head) {
 389                 pending_bios = &device->pending_sync_bios;
 390                 force_reg = 1;
 391         } else {
 392                 pending_bios = &device->pending_bios;
 393                 force_reg = 0;
 394         }
 395
 396         pending = pending_bios->head;
 397         tail = pending_bios->tail;
 398         WARN_ON(pending && !tail);
 399
 400         /*
 401          * if pending was null this time around, no bios need processing
 402          * at all and we can stop.  Otherwise it'll loop back up again
 403          * and do an additional check so no bios are missed.
 404          *
 405          * device->running_pending is used to synchronize with the
 406          * schedule_bio code.
 407          */
 408         if (device->pending_sync_bios.head == NULL &&
 409             device->pending_bios.head == NULL) {
 410                 again = 0;
 411                 device->running_pending = 0;
 412         } else {
 413                 again = 1;
 414                 device->running_pending = 1;
 415         }
 416
 417         pending_bios->head = NULL;
 418         pending_bios->tail = NULL;
 419
 420         spin_unlock(&device->io_lock);
 421
 422         while (pending) {
 423
 424                 rmb();
 425                 /* we want to work on both lists, but do more bios on the
 426                  * sync list than the regular list
 427                  */
 428                 if ((num_run > 32 &&
 429                     pending_bios != &device->pending_sync_bios &&
 430                     device->pending_sync_bios.head) ||
 431                    (num_run > 64 && pending_bios == &device->pending_sync_bios &&
 432                     device->pending_bios.head)) {
 433                         spin_lock(&device->io_lock);
 434                         requeue_list(pending_bios, pending, tail);
 435                         goto loop_lock;
 436                 }
 437
 438                 cur = pending;
 439                 pending = pending->bi_next;
 440                 cur->bi_next = NULL;
 441
 442                 /*
 443                  * atomic_dec_return implies a barrier for waitqueue_active
 444                  */
 445                 if (atomic_dec_return(&fs_info->nr_async_bios) < limit &&
 446                     waitqueue_active(&fs_info->async_submit_wait))
 447                         wake_up(&fs_info->async_submit_wait);
 448
 449                 BUG_ON(atomic_read(&cur->__bi_cnt) == 0);
 450
 451                 /*
 452                  * if we're doing the sync list, record that our
 453                  * plug has some sync requests on it
 454                  *
 455                  * If we're doing the regular list and there are
 456                  * sync requests sitting around, unplug before
 457                  * we add more
 458                  */
 459                 if (pending_bios == &device->pending_sync_bios) {
 460                         sync_pending = 1;
 461                 } else if (sync_pending) {
 462                         blk_finish_plug(&plug);
 463                         blk_start_plug(&plug);
 464                         sync_pending = 0;
 465                 }
 466
 467                 btrfsic_submit_bio(cur);
 468                 num_run++;
 469                 batch_run++;
 470
 471                 cond_resched();
 472
 473                 /*
 474                  * we made progress, there is more work to do and the bdi
 475                  * is now congested.  Back off and let other work structs
 476                  * run instead
 477                  */
 478                 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
 479                     fs_info->fs_devices->open_devices > 1) {
 480                         struct io_context *ioc;
 481
 482                         ioc = current->io_context;
 483
 484                         /*
 485                          * the main goal here is that we don't want to
 486                          * block if we're going to be able to submit
 487                          * more requests without blocking.
 488                          *
 489                          * This code does two great things, it pokes into
 490                          * the elevator code from a filesystem _and_
 491                          * it makes assumptions about how batching works.
 492                          */
 493                         if (ioc && ioc->nr_batch_requests > 0 &&
 494                             time_before(jiffies, ioc->last_waited + HZ/50UL) &&
 495                             (last_waited == 0 ||
 496                              ioc->last_waited == last_waited)) {
 497                                 /*
 498                                  * we want to go through our batch of
 499                                  * requests and stop.  So, we copy out
 500                                  * the ioc->last_waited time and test
 501                                  * against it before looping
 502                                  */
 503                                 last_waited = ioc->last_waited;
 504                                 cond_resched();
 505                                 continue;
 506                         }
 507                         spin_lock(&device->io_lock);
 508                         requeue_list(pending_bios, pending, tail);
 509                         device->running_pending = 1;
 510
 511                         spin_unlock(&device->io_lock);
 512                         btrfs_queue_work(fs_info->submit_workers,
 513                                          &device->work);
 514                         goto done;
 515                 }
 516                 /* unplug every 64 requests just for good measure */
 517                 if (batch_run % 64 == 0) {
 518                         blk_finish_plug(&plug);
 519                         blk_start_plug(&plug);
 520                         sync_pending = 0;
 521                 }
 522         }
 523
 524         cond_resched();
 525         if (again)
 526                 goto loop;
 527
 528         spin_lock(&device->io_lock);
 529         if (device->pending_bios.head || device->pending_sync_bios.head)
 530                 goto loop_lock;
 531         spin_unlock(&device->io_lock);
 532
 533 done:
 534         blk_finish_plug(&plug);
 535 }
 536
 537 static void pending_bios_fn(struct btrfs_work *work)
 538 {
 539         struct btrfs_device *device;
 540
 541         device = container_of(work, struct btrfs_device, work);
 542         run_scheduled_bios(device);
 543 }
 544
 545
 546 void btrfs_free_stale_device(struct btrfs_device *cur_dev)
 547 {
 548         struct btrfs_fs_devices *fs_devs;
 549         struct btrfs_device *dev;
 550
 551         if (!cur_dev->name)
 552                 return;
 553
 554         list_for_each_entry(fs_devs, &fs_uuids, list) {
 555                 int del = 1;
 556
 557                 if (fs_devs->opened)
 558                         continue;
 559                 if (fs_devs->seeding)
 560                         continue;
 561
 562                 list_for_each_entry(dev, &fs_devs->devices, dev_list) {
 563
 564                         if (dev == cur_dev)
 565                                 continue;
 566                         if (!dev->name)
 567                                 continue;
 568
 569                         /*
 570                          * Todo: This won't be enough. What if the same device
 571                          * comes back (with new uuid and) with its mapper path?
 572                          * But for now, this does help as mostly an admin will
 573                          * either use mapper or non mapper path throughout.
 574                          */
 575                         rcu_read_lock();
 576                         del = strcmp(rcu_str_deref(dev->name),
 577                                                 rcu_str_deref(cur_dev->name));
 578                         rcu_read_unlock();
 579                         if (!del)
 580                                 break;
 581                 }
 582
 583                 if (!del) {
 584                         /* delete the stale device */
 585                         if (fs_devs->num_devices == 1) {
 586                                 btrfs_sysfs_remove_fsid(fs_devs);
 587                                 list_del(&fs_devs->list);
 588                                 free_fs_devices(fs_devs);
 589                         } else {
 590                                 fs_devs->num_devices--;
 591                                 list_del(&dev->dev_list);
 592                                 rcu_string_free(dev->name);
 593                                 kfree(dev);
 594                         }
 595                         break;
 596                 }
 597         }
 598 }
 599
 600 /*
 601  * Add new device to list of registered devices
 602  *
 603  * Returns:
 604  * 1   - first time device is seen
 605  * 0   - device already known
 606  * < 0 - error
 607  */
 608 static noinline int device_list_add(const char *path,
 609                            struct btrfs_super_block *disk_super,
 610                            u64 devid, struct btrfs_fs_devices **fs_devices_ret)
 611 {
 612         struct btrfs_device *device;
 613         struct btrfs_fs_devices *fs_devices;
 614         struct rcu_string *name;
 615         int ret = 0;
 616         u64 found_transid = btrfs_super_generation(disk_super);
 617
 618         fs_devices = find_fsid(disk_super->fsid);
 619         if (!fs_devices) {
 620                 fs_devices = alloc_fs_devices(disk_super->fsid);
 621                 if (IS_ERR(fs_devices))
 622                         return PTR_ERR(fs_devices);
 623
 624                 list_add(&fs_devices->list, &fs_uuids);
 625
 626                 device = NULL;
 627         } else {
 628                 device = __find_device(&fs_devices->devices, devid,
 629                                        disk_super->dev_item.uuid);
 630         }
 631
 632         if (!device) {
 633                 if (fs_devices->opened)
 634                         return -EBUSY;
 635
 636                 device = btrfs_alloc_device(NULL, &devid,
 637                                             disk_super->dev_item.uuid);
 638                 if (IS_ERR(device)) {
 639                         /* we can safely leave the fs_devices entry around */
 640                         return PTR_ERR(device);
 641                 }
 642
 643                 name = rcu_string_strdup(path, GFP_NOFS);
 644                 if (!name) {
 645                         kfree(device);
 646                         return -ENOMEM;
 647                 }
 648                 rcu_assign_pointer(device->name, name);
 649
 650                 mutex_lock(&fs_devices->device_list_mutex);
 651                 list_add_rcu(&device->dev_list, &fs_devices->devices);
 652                 fs_devices->num_devices++;
 653                 mutex_unlock(&fs_devices->device_list_mutex);
 654
 655                 ret = 1;
 656                 device->fs_devices = fs_devices;
 657         } else if (!device->name || strcmp(device->name->str, path)) {
 658                 /*
 659                  * When FS is already mounted.
 660                  * 1. If you are here and if the device->name is NULL that
 661                  *    means this device was missing at time of FS mount.
 662                  * 2. If you are here and if the device->name is different
 663                  *    from 'path' that means either
 664                  *      a. The same device disappeared and reappeared with
 665                  *         different name. or
 666                  *      b. The missing-disk-which-was-replaced, has
 667                  *         reappeared now.
 668                  *
 669                  * We must allow 1 and 2a above. But 2b would be a spurious
 670                  * and unintentional.
 671                  *
 672                  * Further in case of 1 and 2a above, the disk at 'path'
 673                  * would have missed some transaction when it was away and
 674                  * in case of 2a the stale bdev has to be updated as well.
 675                  * 2b must not be allowed at all time.
 676                  */
 677
 678                 /*
 679                  * For now, we do allow update to btrfs_fs_device through the
 680                  * btrfs dev scan cli after FS has been mounted.  We're still
 681                  * tracking a problem where systems fail mount by subvolume id
 682                  * when we reject replacement on a mounted FS.
 683                  */
 684                 if (!fs_devices->opened && found_transid < device->generation) {
 685                         /*
 686                          * That is if the FS is _not_ mounted and if you
 687                          * are here, that means there is more than one
 688                          * disk with same uuid and devid.We keep the one
 689                          * with larger generation number or the last-in if
 690                          * generation are equal.
 691                          */
 692                         return -EEXIST;
 693                 }
 694
 695                 name = rcu_string_strdup(path, GFP_NOFS);
 696                 if (!name)
 697                         return -ENOMEM;
 698                 rcu_string_free(device->name);
 699                 rcu_assign_pointer(device->name, name);
 700                 if (device->missing) {
 701                         fs_devices->missing_devices--;
 702                         device->missing = 0;
 703                 }
 704         }
 705
 706         /*
 707          * Unmount does not free the btrfs_device struct but would zero
 708          * generation along with most of the other members. So just update
 709          * it back. We need it to pick the disk with largest generation
 710          * (as above).
 711          */
 712         if (!fs_devices->opened)
 713                 device->generation = found_transid;
 714
 715         /*
 716          * if there is new btrfs on an already registered device,
 717          * then remove the stale device entry.
 718          */
 719         if (ret > 0)
 720                 btrfs_free_stale_device(device);
 721
 722         *fs_devices_ret = fs_devices;
 723
 724         return ret;
 725 }
 726
 727 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
 728 {
 729         struct btrfs_fs_devices *fs_devices;
 730         struct btrfs_device *device;
 731         struct btrfs_device *orig_dev;
 732
 733         fs_devices = alloc_fs_devices(orig->fsid);
 734         if (IS_ERR(fs_devices))
 735                 return fs_devices;
 736
 737         mutex_lock(&orig->device_list_mutex);
 738         fs_devices->total_devices = orig->total_devices;
 739
 740         /* We have held the volume lock, it is safe to get the devices. */
 741         list_for_each_entry(orig_dev, &orig->devices, dev_list) {
 742                 struct rcu_string *name;
 743
 744                 device = btrfs_alloc_device(NULL, &orig_dev->devid,
 745                                             orig_dev->uuid);
 746                 if (IS_ERR(device))
 747                         goto error;
 748
 749                 /*
 750                  * This is ok to do without rcu read locked because we hold the
 751                  * uuid mutex so nothing we touch in here is going to disappear.
 752                  */
 753                 if (orig_dev->name) {
 754                         name = rcu_string_strdup(orig_dev->name->str,
 755                                         GFP_KERNEL);
 756                         if (!name) {
 757                                 kfree(device);
 758                                 goto error;
 759                         }
 760                         rcu_assign_pointer(device->name, name);
 761                 }
 762
 763                 list_add(&device->dev_list, &fs_devices->devices);
 764                 device->fs_devices = fs_devices;
 765                 fs_devices->num_devices++;
 766         }
 767         mutex_unlock(&orig->device_list_mutex);
 768         return fs_devices;
 769 error:
 770         mutex_unlock(&orig->device_list_mutex);
 771         free_fs_devices(fs_devices);
 772         return ERR_PTR(-ENOMEM);
 773 }
 774
 775 void btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices, int step)
 776 {
 777         struct btrfs_device *device, *next;
 778         struct btrfs_device *latest_dev = NULL;
 779
 780         mutex_lock(&uuid_mutex);
 781 again:
 782         /* This is the initialized path, it is safe to release the devices. */
 783         list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
 784                 if (device->in_fs_metadata) {
 785                         if (!device->is_tgtdev_for_dev_replace &&
 786                             (!latest_dev ||
 787                              device->generation > latest_dev->generation)) {
 788                                 latest_dev = device;
 789                         }
 790                         continue;
 791                 }
 792
 793                 if (device->devid == BTRFS_DEV_REPLACE_DEVID) {
 794                         /*
 795                          * In the first step, keep the device which has
 796                          * the correct fsid and the devid that is used
 797                          * for the dev_replace procedure.
 798                          * In the second step, the dev_replace state is
 799                          * read from the device tree and it is known
 800                          * whether the procedure is really active or
 801                          * not, which means whether this device is
 802                          * used or whether it should be removed.
 803                          */
 804                         if (step == 0 || device->is_tgtdev_for_dev_replace) {
 805                                 continue;
 806                         }
 807                 }
 808                 if (device->bdev) {
 809                         blkdev_put(device->bdev, device->mode);
 810                         device->bdev = NULL;
 811                         fs_devices->open_devices--;
 812                 }
 813                 if (device->writeable) {
 814                         list_del_init(&device->dev_alloc_list);
 815                         device->writeable = 0;
 816                         if (!device->is_tgtdev_for_dev_replace)
 817                                 fs_devices->rw_devices--;
 818                 }
 819                 list_del_init(&device->dev_list);
 820                 fs_devices->num_devices--;
 821                 rcu_string_free(device->name);
 822                 kfree(device);
 823         }
 824
 825         if (fs_devices->seed) {
 826                 fs_devices = fs_devices->seed;
 827                 goto again;
 828         }
 829
 830         fs_devices->latest_bdev = latest_dev->bdev;
 831
 832         mutex_unlock(&uuid_mutex);
 833 }
 834
 835 static void __free_device(struct work_struct *work)
 836 {
 837         struct btrfs_device *device;
 838
 839         device = container_of(work, struct btrfs_device, rcu_work);
 840         rcu_string_free(device->name);
 841         kfree(device);
 842 }
 843
 844 static void free_device(struct rcu_head *head)
 845 {
 846         struct btrfs_device *device;
 847
 848         device = container_of(head, struct btrfs_device, rcu);
 849
 850         INIT_WORK(&device->rcu_work, __free_device);
 851         schedule_work(&device->rcu_work);
 852 }
 853
 854 static void btrfs_close_bdev(struct btrfs_device *device)
 855 {
 856         if (device->bdev && device->writeable) {
 857                 sync_blockdev(device->bdev);
 858                 invalidate_bdev(device->bdev);
 859         }
 860
 861         if (device->bdev)
 862                 blkdev_put(device->bdev, device->mode);
 863 }
 864
 865 static void btrfs_prepare_close_one_device(struct btrfs_device *device)
 866 {
 867         struct btrfs_fs_devices *fs_devices = device->fs_devices;
 868         struct btrfs_device *new_device;
 869         struct rcu_string *name;
 870
 871         if (device->bdev)
 872                 fs_devices->open_devices--;
 873
 874         if (device->writeable &&
 875             device->devid != BTRFS_DEV_REPLACE_DEVID) {
 876                 list_del_init(&device->dev_alloc_list);
 877                 fs_devices->rw_devices--;
 878         }
 879
 880         if (device->missing)
 881                 fs_devices->missing_devices--;
 882
 883         new_device = btrfs_alloc_device(NULL, &device->devid,
 884                                         device->uuid);
 885         BUG_ON(IS_ERR(new_device)); /* -ENOMEM */
 886
 887         /* Safe because we are under uuid_mutex */
 888         if (device->name) {
 889                 name = rcu_string_strdup(device->name->str, GFP_NOFS);
 890                 BUG_ON(!name); /* -ENOMEM */
 891                 rcu_assign_pointer(new_device->name, name);
 892         }
 893
 894         list_replace_rcu(&device->dev_list, &new_device->dev_list);
 895         new_device->fs_devices = device->fs_devices;
 896 }
 897
 898 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
 899 {
 900         struct btrfs_device *device, *tmp;
 901         struct list_head pending_put;
 902
 903         INIT_LIST_HEAD(&pending_put);
 904
 905         if (--fs_devices->opened > 0)
 906                 return 0;
 907
 908         mutex_lock(&fs_devices->device_list_mutex);
 909         list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list) {
 910                 btrfs_prepare_close_one_device(device);
 911                 list_add(&device->dev_list, &pending_put);
 912         }
 913         mutex_unlock(&fs_devices->device_list_mutex);
 914
 915         /*
 916          * btrfs_show_devname() is using the device_list_mutex,
 917          * sometimes call to blkdev_put() leads vfs calling
 918          * into this func. So do put outside of device_list_mutex,
 919          * as of now.
 920          */
 921         while (!list_empty(&pending_put)) {
 922                 device = list_first_entry(&pending_put,
 923                                 struct btrfs_device, dev_list);
 924                 list_del(&device->dev_list);
 925                 btrfs_close_bdev(device);
 926                 call_rcu(&device->rcu, free_device);
 927         }
 928
 929         WARN_ON(fs_devices->open_devices);
 930         WARN_ON(fs_devices->rw_devices);
 931         fs_devices->opened = 0;
 932         fs_devices->seeding = 0;
 933
 934         return 0;
 935 }
 936
 937 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
 938 {
 939         struct btrfs_fs_devices *seed_devices = NULL;
 940         int ret;
 941
 942         mutex_lock(&uuid_mutex);
 943         ret = __btrfs_close_devices(fs_devices);
 944         if (!fs_devices->opened) {
 945                 seed_devices = fs_devices->seed;
 946                 fs_devices->seed = NULL;
 947         }
 948         mutex_unlock(&uuid_mutex);
 949
 950         while (seed_devices) {
 951                 fs_devices = seed_devices;
 952                 seed_devices = fs_devices->seed;
 953                 __btrfs_close_devices(fs_devices);
 954                 free_fs_devices(fs_devices);
 955         }
 956         /*
 957          * Wait for rcu kworkers under __btrfs_close_devices
 958          * to finish all blkdev_puts so device is really
 959          * free when umount is done.
 960          */
 961         rcu_barrier();
 962         return ret;
 963 }
 964
 965 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
 966                                 fmode_t flags, void *holder)
 967 {
 968         struct request_queue *q;
 969         struct block_device *bdev;
 970         struct list_head *head = &fs_devices->devices;
 971         struct btrfs_device *device;
 972         struct btrfs_device *latest_dev = NULL;
 973         struct buffer_head *bh;
 974         struct btrfs_super_block *disk_super;
 975         u64 devid;
 976         int seeding = 1;
 977         int ret = 0;
 978
 979         flags |= FMODE_EXCL;
 980
 981         list_for_each_entry(device, head, dev_list) {
 982                 if (device->bdev)
 983                         continue;
 984                 if (!device->name)
 985                         continue;
 986
 987                 /* Just open everything we can; ignore failures here */
 988                 if (btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1,
 989                                             &bdev, &bh))
 990                         continue;
 991
 992                 disk_super = (struct btrfs_super_block *)bh->b_data;
 993                 devid = btrfs_stack_device_id(&disk_super->dev_item);
 994                 if (devid != device->devid)
 995                         goto error_brelse;
 996
 997                 if (memcmp(device->uuid, disk_super->dev_item.uuid,
 998                            BTRFS_UUID_SIZE))
 999                         goto error_brelse;
1000
1001                 device->generation = btrfs_super_generation(disk_super);
1002                 if (!latest_dev ||
1003                     device->generation > latest_dev->generation)
1004                         latest_dev = device;
1005
1006                 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
1007                         device->writeable = 0;
1008                 } else {
1009                         device->writeable = !bdev_read_only(bdev);
1010                         seeding = 0;
1011                 }
1012
1013                 q = bdev_get_queue(bdev);
1014                 if (blk_queue_discard(q))
1015                         device->can_discard = 1;
1016                 if (!blk_queue_nonrot(q))
1017                         fs_devices->rotating = 1;
1018
1019                 device->bdev = bdev;
1020                 device->in_fs_metadata = 0;
1021                 device->mode = flags;
1022
1023                 fs_devices->open_devices++;
1024                 if (device->writeable &&
1025                     device->devid != BTRFS_DEV_REPLACE_DEVID) {
1026                         fs_devices->rw_devices++;
1027                         list_add(&device->dev_alloc_list,
1028                                  &fs_devices->alloc_list);
1029                 }
1030                 brelse(bh);
1031                 continue;
1032
1033 error_brelse:
1034                 brelse(bh);
1035                 blkdev_put(bdev, flags);
1036                 continue;
1037         }
1038         if (fs_devices->open_devices == 0) {
1039                 ret = -EINVAL;
1040                 goto out;
1041         }
1042         fs_devices->seeding = seeding;
1043         fs_devices->opened = 1;
1044         fs_devices->latest_bdev = latest_dev->bdev;
1045         fs_devices->total_rw_bytes = 0;
1046 out:
1047         return ret;
1048 }
1049
1050 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
1051                        fmode_t flags, void *holder)
1052 {
1053         int ret;
1054
1055         mutex_lock(&uuid_mutex);
1056         if (fs_devices->opened) {
1057                 fs_devices->opened++;
1058                 ret = 0;
1059         } else {
1060                 ret = __btrfs_open_devices(fs_devices, flags, holder);
1061         }
1062         mutex_unlock(&uuid_mutex);
1063         return ret;
1064 }
1065
1066 void btrfs_release_disk_super(struct page *page)
1067 {
1068         kunmap(page);
1069         put_page(page);
1070 }
1071
1072 int btrfs_read_disk_super(struct block_device *bdev, u64 bytenr,
1073                 struct page **page, struct btrfs_super_block **disk_super)
1074 {
1075         void *p;
1076         pgoff_t index;
1077
1078         /* make sure our super fits in the device */
1079         if (bytenr + PAGE_SIZE >= i_size_read(bdev->bd_inode))
1080                 return 1;
1081
1082         /* make sure our super fits in the page */
1083         if (sizeof(**disk_super) > PAGE_SIZE)
1084                 return 1;
1085
1086         /* make sure our super doesn't straddle pages on disk */
1087         index = bytenr >> PAGE_SHIFT;
1088         if ((bytenr + sizeof(**disk_super) - 1) >> PAGE_SHIFT != index)
1089                 return 1;
1090
1091         /* pull in the page with our super */
1092         *page = read_cache_page_gfp(bdev->bd_inode->i_mapping,
1093                                    index, GFP_KERNEL);
1094
1095         if (IS_ERR_OR_NULL(*page))
1096                 return 1;
1097
1098         p = kmap(*page);
1099
1100         /* align our pointer to the offset of the super block */
1101         *disk_super = p + (bytenr & ~PAGE_MASK);
1102
1103         if (btrfs_super_bytenr(*disk_super) != bytenr ||
1104             btrfs_super_magic(*disk_super) != BTRFS_MAGIC) {
1105                 btrfs_release_disk_super(*page);
1106                 return 1;
1107         }
1108
1109         if ((*disk_super)->label[0] &&
1110                 (*disk_super)->label[BTRFS_LABEL_SIZE - 1])
1111                 (*disk_super)->label[BTRFS_LABEL_SIZE - 1] = '\0';
1112
1113         return 0;
1114 }
1115
1116 /*
1117  * Look for a btrfs signature on a device. This may be called out of the mount path
1118  * and we are not allowed to call set_blocksize during the scan. The superblock
1119  * is read via pagecache
1120  */
1121 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
1122                           struct btrfs_fs_devices **fs_devices_ret)
1123 {
1124         struct btrfs_super_block *disk_super;
1125         struct block_device *bdev;
1126         struct page *page;
1127         int ret = -EINVAL;
1128         u64 devid;
1129         u64 transid;
1130         u64 total_devices;
1131         u64 bytenr;
1132
1133         /*
1134          * we would like to check all the supers, but that would make
1135          * a btrfs mount succeed after a mkfs from a different FS.
1136          * So, we need to add a special mount option to scan for
1137          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
1138          */
1139         bytenr = btrfs_sb_offset(0);
1140         flags |= FMODE_EXCL;
1141         mutex_lock(&uuid_mutex);
1142
1143         bdev = blkdev_get_by_path(path, flags, holder);
1144         if (IS_ERR(bdev)) {
1145                 ret = PTR_ERR(bdev);
1146                 goto error;
1147         }
1148
1149         if (btrfs_read_disk_super(bdev, bytenr, &page, &disk_super))
1150                 goto error_bdev_put;
1151
1152         devid = btrfs_stack_device_id(&disk_super->dev_item);
1153         transid = btrfs_super_generation(disk_super);
1154         total_devices = btrfs_super_num_devices(disk_super);
1155
1156         ret = device_list_add(path, disk_super, devid, fs_devices_ret);
1157         if (ret > 0) {
1158                 if (disk_super->label[0]) {
1159                         pr_info("BTRFS: device label %s ", disk_super->label);
1160                 } else {
1161                         pr_info("BTRFS: device fsid %pU ", disk_super->fsid);
1162                 }
1163
1164                 pr_cont("devid %llu transid %llu %s\n", devid, transid, path);
1165                 ret = 0;
1166         }
1167         if (!ret && fs_devices_ret)
1168                 (*fs_devices_ret)->total_devices = total_devices;
1169
1170         btrfs_release_disk_super(page);
1171
1172 error_bdev_put:
1173         blkdev_put(bdev, flags);
1174 error:
1175         mutex_unlock(&uuid_mutex);
1176         return ret;
1177 }
1178
1179 /* helper to account the used device space in the range */
1180 int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
1181                                    u64 end, u64 *length)
1182 {
1183         struct btrfs_key key;
1184         struct btrfs_root *root = device->fs_info->dev_root;
1185         struct btrfs_dev_extent *dev_extent;
1186         struct btrfs_path *path;
1187         u64 extent_end;
1188         int ret;
1189         int slot;
1190         struct extent_buffer *l;
1191
1192         *length = 0;
1193
1194         if (start >= device->total_bytes || device->is_tgtdev_for_dev_replace)
1195                 return 0;
1196
1197         path = btrfs_alloc_path();
1198         if (!path)
1199                 return -ENOMEM;
1200         path->reada = READA_FORWARD;
1201
1202         key.objectid = device->devid;
1203         key.offset = start;
1204         key.type = BTRFS_DEV_EXTENT_KEY;
1205
1206         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1207         if (ret < 0)
1208                 goto out;
1209         if (ret > 0) {
1210                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1211                 if (ret < 0)
1212                         goto out;
1213         }
1214
1215         while (1) {
1216                 l = path->nodes[0];
1217                 slot = path->slots[0];
1218                 if (slot >= btrfs_header_nritems(l)) {
1219                         ret = btrfs_next_leaf(root, path);
1220                         if (ret == 0)
1221                                 continue;
1222                         if (ret < 0)
1223                                 goto out;
1224
1225                         break;
1226                 }
1227                 btrfs_item_key_to_cpu(l, &key, slot);
1228
1229                 if (key.objectid < device->devid)
1230                         goto next;
1231
1232                 if (key.objectid > device->devid)
1233                         break;
1234
1235                 if (key.type != BTRFS_DEV_EXTENT_KEY)
1236                         goto next;
1237
1238                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1239                 extent_end = key.offset + btrfs_dev_extent_length(l,
1240                                                                   dev_extent);
1241                 if (key.offset <= start && extent_end > end) {
1242                         *length = end - start + 1;
1243                         break;
1244                 } else if (key.offset <= start && extent_end > start)
1245                         *length += extent_end - start;
1246                 else if (key.offset > start && extent_end <= end)
1247                         *length += extent_end - key.offset;
1248                 else if (key.offset > start && key.offset <= end) {
1249                         *length += end - key.offset + 1;
1250                         break;
1251                 } else if (key.offset > end)
1252                         break;
1253
1254 next:
1255                 path->slots[0]++;
1256         }
1257         ret = 0;
1258 out:
1259         btrfs_free_path(path);
1260         return ret;
1261 }
1262
1263 static int contains_pending_extent(struct btrfs_transaction *transaction,
1264                                    struct btrfs_device *device,
1265                                    u64 *start, u64 len)
1266 {
1267         struct btrfs_fs_info *fs_info = device->fs_info;
1268         struct extent_map *em;
1269         struct list_head *search_list = &fs_info->pinned_chunks;
1270         int ret = 0;
1271         u64 physical_start = *start;
1272
1273         if (transaction)
1274                 search_list = &transaction->pending_chunks;
1275 again:
1276         list_for_each_entry(em, search_list, list) {
1277                 struct map_lookup *map;
1278                 int i;
1279
1280                 map = em->map_lookup;
1281                 for (i = 0; i < map->num_stripes; i++) {
1282                         u64 end;
1283
1284                         if (map->stripes[i].dev != device)
1285                                 continue;
1286                         if (map->stripes[i].physical >= physical_start + len ||
1287                             map->stripes[i].physical + em->orig_block_len <=
1288                             physical_start)
1289                                 continue;
1290                         /*
1291                          * Make sure that while processing the pinned list we do
1292                          * not override our *start with a lower value, because
1293                          * we can have pinned chunks that fall within this
1294                          * device hole and that have lower physical addresses
1295                          * than the pending chunks we processed before. If we
1296                          * do not take this special care we can end up getting
1297                          * 2 pending chunks that start at the same physical
1298                          * device offsets because the end offset of a pinned
1299                          * chunk can be equal to the start offset of some
1300                          * pending chunk.
1301                          */
1302                         end = map->stripes[i].physical + em->orig_block_len;
1303                         if (end > *start) {
1304                                 *start = end;
1305                                 ret = 1;
1306                         }
1307                 }
1308         }
1309         if (search_list != &fs_info->pinned_chunks) {
1310                 search_list = &fs_info->pinned_chunks;
1311                 goto again;
1312         }
1313
1314         return ret;
1315 }
1316
1317
1318 /*
1319  * find_free_dev_extent_start - find free space in the specified device
1320  * @device:       the device which we search the free space in
1321  * @num_bytes:    the size of the free space that we need
1322  * @search_start: the position from which to begin the search
1323  * @start:        store the start of the free space.
1324  * @len:          the size of the free space. that we find, or the size
1325  *                of the max free space if we don't find suitable free space
1326  *
1327  * this uses a pretty simple search, the expectation is that it is
1328  * called very infrequently and that a given device has a small number
1329  * of extents
1330  *
1331  * @start is used to store the start of the free space if we find. But if we
1332  * don't find suitable free space, it will be used to store the start position
1333  * of the max free space.
1334  *
1335  * @len is used to store the size of the free space that we find.
1336  * But if we don't find suitable free space, it is used to store the size of
1337  * the max free space.
1338  */
1339 int find_free_dev_extent_start(struct btrfs_transaction *transaction,
1340                                struct btrfs_device *device, u64 num_bytes,
1341                                u64 search_start, u64 *start, u64 *len)
1342 {
1343         struct btrfs_fs_info *fs_info = device->fs_info;
1344         struct btrfs_root *root = fs_info->dev_root;
1345         struct btrfs_key key;
1346         struct btrfs_dev_extent *dev_extent;
1347         struct btrfs_path *path;
1348         u64 hole_size;
1349         u64 max_hole_start;
1350         u64 max_hole_size;
1351         u64 extent_end;
1352         u64 search_end = device->total_bytes;
1353         int ret;
1354         int slot;
1355         struct extent_buffer *l;
1356         u64 min_search_start;
1357
1358         /*
1359          * We don't want to overwrite the superblock on the drive nor any area
1360          * used by the boot loader (grub for example), so we make sure to start
1361          * at an offset of at least 1MB.
1362          */
1363         min_search_start = max(fs_info->alloc_start, 1024ull * 1024);
1364         search_start = max(search_start, min_search_start);
1365
1366         path = btrfs_alloc_path();
1367         if (!path)
1368                 return -ENOMEM;
1369
1370         max_hole_start = search_start;
1371         max_hole_size = 0;
1372
1373 again:
1374         if (search_start >= search_end || device->is_tgtdev_for_dev_replace) {
1375                 ret = -ENOSPC;
1376                 goto out;
1377         }
1378
1379         path->reada = READA_FORWARD;
1380         path->search_commit_root = 1;
1381         path->skip_locking = 1;
1382
1383         key.objectid = device->devid;
1384         key.offset = search_start;
1385         key.type = BTRFS_DEV_EXTENT_KEY;
1386
1387         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1388         if (ret < 0)
1389                 goto out;
1390         if (ret > 0) {
1391                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1392                 if (ret < 0)
1393                         goto out;
1394         }
1395
1396         while (1) {
1397                 l = path->nodes[0];
1398                 slot = path->slots[0];
1399                 if (slot >= btrfs_header_nritems(l)) {
1400                         ret = btrfs_next_leaf(root, path);
1401                         if (ret == 0)
1402                                 continue;
1403                         if (ret < 0)
1404                                 goto out;
1405
1406                         break;
1407                 }
1408                 btrfs_item_key_to_cpu(l, &key, slot);
1409
1410                 if (key.objectid < device->devid)
1411                         goto next;
1412
1413                 if (key.objectid > device->devid)
1414                         break;
1415
1416                 if (key.type != BTRFS_DEV_EXTENT_KEY)
1417                         goto next;
1418
1419                 if (key.offset > search_start) {
1420                         hole_size = key.offset - search_start;
1421
1422                         /*
1423                          * Have to check before we set max_hole_start, otherwise
1424                          * we could end up sending back this offset anyway.
1425                          */
1426                         if (contains_pending_extent(transaction, device,
1427                                                     &search_start,
1428                                                     hole_size)) {
1429                                 if (key.offset >= search_start) {
1430                                         hole_size = key.offset - search_start;
1431                                 } else {
1432                                         WARN_ON_ONCE(1);
1433                                         hole_size = 0;
1434                                 }
1435                         }
1436
1437                         if (hole_size > max_hole_size) {
1438                                 max_hole_start = search_start;
1439                                 max_hole_size = hole_size;
1440                         }
1441
1442                         /*
1443                          * If this free space is greater than which we need,
1444                          * it must be the max free space that we have found
1445                          * until now, so max_hole_start must point to the start
1446                          * of this free space and the length of this free space
1447                          * is stored in max_hole_size. Thus, we return
1448                          * max_hole_start and max_hole_size and go back to the
1449                          * caller.
1450                          */
1451                         if (hole_size >= num_bytes) {
1452                                 ret = 0;
1453                                 goto out;
1454                         }
1455                 }
1456
1457                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1458                 extent_end = key.offset + btrfs_dev_extent_length(l,
1459                                                                   dev_extent);
1460                 if (extent_end > search_start)
1461                         search_start = extent_end;
1462 next:
1463                 path->slots[0]++;
1464                 cond_resched();
1465         }
1466
1467         /*
1468          * At this point, search_start should be the end of
1469          * allocated dev extents, and when shrinking the device,
1470          * search_end may be smaller than search_start.
1471          */
1472         if (search_end > search_start) {
1473                 hole_size = search_end - search_start;
1474
1475                 if (contains_pending_extent(transaction, device, &search_start,
1476                                             hole_size)) {
1477                         btrfs_release_path(path);
1478                         goto again;
1479                 }
1480
1481                 if (hole_size > max_hole_size) {
1482                         max_hole_start = search_start;
1483                         max_hole_size = hole_size;
1484                 }
1485         }
1486
1487         /* See above. */
1488         if (max_hole_size < num_bytes)
1489                 ret = -ENOSPC;
1490         else
1491                 ret = 0;
1492
1493 out:
1494         btrfs_free_path(path);
1495         *start = max_hole_start;
1496         if (len)
1497                 *len = max_hole_size;
1498         return ret;
1499 }
1500
1501 int find_free_dev_extent(struct btrfs_trans_handle *trans,
1502                          struct btrfs_device *device, u64 num_bytes,
1503                          u64 *start, u64 *len)
1504 {
1505         /* FIXME use last free of some kind */
1506         return find_free_dev_extent_start(trans->transaction, device,
1507                                           num_bytes, 0, start, len);
1508 }
1509
1510 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1511                           struct btrfs_device *device,
1512                           u64 start, u64 *dev_extent_len)
1513 {
1514         struct btrfs_fs_info *fs_info = device->fs_info;
1515         struct btrfs_root *root = fs_info->dev_root;
1516         int ret;
1517         struct btrfs_path *path;
1518         struct btrfs_key key;
1519         struct btrfs_key found_key;
1520         struct extent_buffer *leaf = NULL;
1521         struct btrfs_dev_extent *extent = NULL;
1522
1523         path = btrfs_alloc_path();
1524         if (!path)
1525                 return -ENOMEM;
1526
1527         key.objectid = device->devid;
1528         key.offset = start;
1529         key.type = BTRFS_DEV_EXTENT_KEY;
1530 again:
1531         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1532         if (ret > 0) {
1533                 ret = btrfs_previous_item(root, path, key.objectid,
1534                                           BTRFS_DEV_EXTENT_KEY);
1535                 if (ret)
1536                         goto out;
1537                 leaf = path->nodes[0];
1538                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1539                 extent = btrfs_item_ptr(leaf, path->slots[0],
1540                                         struct btrfs_dev_extent);
1541                 BUG_ON(found_key.offset > start || found_key.offset +
1542                        btrfs_dev_extent_length(leaf, extent) < start);
1543                 key = found_key;
1544                 btrfs_release_path(path);
1545                 goto again;
1546         } else if (ret == 0) {
1547                 leaf = path->nodes[0];
1548                 extent = btrfs_item_ptr(leaf, path->slots[0],
1549                                         struct btrfs_dev_extent);
1550         } else {
1551                 btrfs_handle_fs_error(fs_info, ret, "Slot search failed");
1552                 goto out;
1553         }
1554
1555         *dev_extent_len = btrfs_dev_extent_length(leaf, extent);
1556
1557         ret = btrfs_del_item(trans, root, path);
1558         if (ret) {
1559                 btrfs_handle_fs_error(fs_info, ret,
1560                                       "Failed to remove dev extent item");
1561         } else {
1562                 set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags);
1563         }
1564 out:
1565         btrfs_free_path(path);
1566         return ret;
1567 }
1568
1569 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1570                                   struct btrfs_device *device,
1571                                   u64 chunk_tree, u64 chunk_objectid,
1572                                   u64 chunk_offset, u64 start, u64 num_bytes)
1573 {
1574         int ret;
1575         struct btrfs_path *path;
1576         struct btrfs_fs_info *fs_info = device->fs_info;
1577         struct btrfs_root *root = fs_info->dev_root;
1578         struct btrfs_dev_extent *extent;
1579         struct extent_buffer *leaf;
1580         struct btrfs_key key;
1581
1582         WARN_ON(!device->in_fs_metadata);
1583         WARN_ON(device->is_tgtdev_for_dev_replace);
1584         path = btrfs_alloc_path();
1585         if (!path)
1586                 return -ENOMEM;
1587
1588         key.objectid = device->devid;
1589         key.offset = start;
1590         key.type = BTRFS_DEV_EXTENT_KEY;
1591         ret = btrfs_insert_empty_item(trans, root, path, &key,
1592                                       sizeof(*extent));
1593         if (ret)
1594                 goto out;
1595
1596         leaf = path->nodes[0];
1597         extent = btrfs_item_ptr(leaf, path->slots[0],
1598                                 struct btrfs_dev_extent);
1599         btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
1600         btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
1601         btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1602
1603         write_extent_buffer_chunk_tree_uuid(leaf, fs_info->chunk_tree_uuid);
1604
1605         btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1606         btrfs_mark_buffer_dirty(leaf);
1607 out:
1608         btrfs_free_path(path);
1609         return ret;
1610 }
1611
1612 static u64 find_next_chunk(struct btrfs_fs_info *fs_info)
1613 {
1614         struct extent_map_tree *em_tree;
1615         struct extent_map *em;
1616         struct rb_node *n;
1617         u64 ret = 0;
1618
1619         em_tree = &fs_info->mapping_tree.map_tree;
1620         read_lock(&em_tree->lock);
1621         n = rb_last(&em_tree->map);
1622         if (n) {
1623                 em = rb_entry(n, struct extent_map, rb_node);
1624                 ret = em->start + em->len;
1625         }
1626         read_unlock(&em_tree->lock);
1627
1628         return ret;
1629 }
1630
1631 static noinline int find_next_devid(struct btrfs_fs_info *fs_info,
1632                                     u64 *devid_ret)
1633 {
1634         int ret;
1635         struct btrfs_key key;
1636         struct btrfs_key found_key;
1637         struct btrfs_path *path;
1638
1639         path = btrfs_alloc_path();
1640         if (!path)
1641                 return -ENOMEM;
1642
1643         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1644         key.type = BTRFS_DEV_ITEM_KEY;
1645         key.offset = (u64)-1;
1646
1647         ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
1648         if (ret < 0)
1649                 goto error;
1650
1651         BUG_ON(ret == 0); /* Corruption */
1652
1653         ret = btrfs_previous_item(fs_info->chunk_root, path,
1654                                   BTRFS_DEV_ITEMS_OBJECTID,
1655                                   BTRFS_DEV_ITEM_KEY);
1656         if (ret) {
1657                 *devid_ret = 1;
1658         } else {
1659                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1660                                       path->slots[0]);
1661                 *devid_ret = found_key.offset + 1;
1662         }
1663         ret = 0;
1664 error:
1665         btrfs_free_path(path);
1666         return ret;
1667 }
1668
1669 /*
1670  * the device information is stored in the chunk root
1671  * the btrfs_device struct should be fully filled in
1672  */
1673 static int btrfs_add_device(struct btrfs_trans_handle *trans,
1674                             struct btrfs_fs_info *fs_info,
1675                             struct btrfs_device *device)
1676 {
1677         struct btrfs_root *root = fs_info->chunk_root;
1678         int ret;
1679         struct btrfs_path *path;
1680         struct btrfs_dev_item *dev_item;
1681         struct extent_buffer *leaf;
1682         struct btrfs_key key;
1683         unsigned long ptr;
1684
1685         path = btrfs_alloc_path();
1686         if (!path)
1687                 return -ENOMEM;
1688
1689         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1690         key.type = BTRFS_DEV_ITEM_KEY;
1691         key.offset = device->devid;
1692
1693         ret = btrfs_insert_empty_item(trans, root, path, &key,
1694                                       sizeof(*dev_item));
1695         if (ret)
1696                 goto out;
1697
1698         leaf = path->nodes[0];
1699         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1700
1701         btrfs_set_device_id(leaf, dev_item, device->devid);
1702         btrfs_set_device_generation(leaf, dev_item, 0);
1703         btrfs_set_device_type(leaf, dev_item, device->type);
1704         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1705         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1706         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1707         btrfs_set_device_total_bytes(leaf, dev_item,
1708                                      btrfs_device_get_disk_total_bytes(device));
1709         btrfs_set_device_bytes_used(leaf, dev_item,
1710                                     btrfs_device_get_bytes_used(device));
1711         btrfs_set_device_group(leaf, dev_item, 0);
1712         btrfs_set_device_seek_speed(leaf, dev_item, 0);
1713         btrfs_set_device_bandwidth(leaf, dev_item, 0);
1714         btrfs_set_device_start_offset(leaf, dev_item, 0);
1715
1716         ptr = btrfs_device_uuid(dev_item);
1717         write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1718         ptr = btrfs_device_fsid(dev_item);
1719         write_extent_buffer(leaf, fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1720         btrfs_mark_buffer_dirty(leaf);
1721
1722         ret = 0;
1723 out:
1724         btrfs_free_path(path);
1725         return ret;
1726 }
1727
1728 /*
1729  * Function to update ctime/mtime for a given device path.
1730  * Mainly used for ctime/mtime based probe like libblkid.
1731  */
1732 static void update_dev_time(const char *path_name)
1733 {
1734         struct file *filp;
1735
1736         filp = filp_open(path_name, O_RDWR, 0);
1737         if (IS_ERR(filp))
1738                 return;
1739         file_update_time(filp);
1740         filp_close(filp, NULL);
1741 }
1742
1743 static int btrfs_rm_dev_item(struct btrfs_fs_info *fs_info,
1744                              struct btrfs_device *device)
1745 {
1746         struct btrfs_root *root = fs_info->chunk_root;
1747         int ret;
1748         struct btrfs_path *path;
1749         struct btrfs_key key;
1750         struct btrfs_trans_handle *trans;
1751
1752         path = btrfs_alloc_path();
1753         if (!path)
1754                 return -ENOMEM;
1755
1756         trans = btrfs_start_transaction(root, 0);
1757         if (IS_ERR(trans)) {
1758                 btrfs_free_path(path);
1759                 return PTR_ERR(trans);
1760         }
1761         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1762         key.type = BTRFS_DEV_ITEM_KEY;
1763         key.offset = device->devid;
1764
1765         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1766         if (ret < 0)
1767                 goto out;
1768
1769         if (ret > 0) {
1770                 ret = -ENOENT;
1771                 goto out;
1772         }
1773
1774         ret = btrfs_del_item(trans, root, path);
1775         if (ret)
1776                 goto out;
1777 out:
1778         btrfs_free_path(path);
1779         btrfs_commit_transaction(trans);
1780         return ret;
1781 }
1782
1783 /*
1784  * Verify that @num_devices satisfies the RAID profile constraints in the whole
1785  * filesystem. It's up to the caller to adjust that number regarding eg. device
1786  * replace.
1787  */
1788 static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info,
1789                 u64 num_devices)
1790 {
1791         u64 all_avail;
1792         unsigned seq;
1793         int i;
1794
1795         do {
1796                 seq = read_seqbegin(&fs_info->profiles_lock);
1797
1798                 all_avail = fs_info->avail_data_alloc_bits |
1799                             fs_info->avail_system_alloc_bits |
1800                             fs_info->avail_metadata_alloc_bits;
1801         } while (read_seqretry(&fs_info->profiles_lock, seq));
1802
1803         for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
1804                 if (!(all_avail & btrfs_raid_group[i]))
1805                         continue;
1806
1807                 if (num_devices < btrfs_raid_array[i].devs_min) {
1808                         int ret = btrfs_raid_mindev_error[i];
1809
1810                         if (ret)
1811                                 return ret;
1812                 }
1813         }
1814
1815         return 0;
1816 }
1817
1818 struct btrfs_device *btrfs_find_next_active_device(struct btrfs_fs_devices *fs_devs,
1819                                         struct btrfs_device *device)
1820 {
1821         struct btrfs_device *next_device;
1822
1823         list_for_each_entry(next_device, &fs_devs->devices, dev_list) {
1824                 if (next_device != device &&
1825                         !next_device->missing && next_device->bdev)
1826                         return next_device;
1827         }
1828
1829         return NULL;
1830 }
1831
1832 /*
1833  * Helper function to check if the given device is part of s_bdev / latest_bdev
1834  * and replace it with the provided or the next active device, in the context
1835  * where this function called, there should be always be another device (or
1836  * this_dev) which is active.
1837  */
1838 void btrfs_assign_next_active_device(struct btrfs_fs_info *fs_info,
1839                 struct btrfs_device *device, struct btrfs_device *this_dev)
1840 {
1841         struct btrfs_device *next_device;
1842
1843         if (this_dev)
1844                 next_device = this_dev;
1845         else
1846                 next_device = btrfs_find_next_active_device(fs_info->fs_devices,
1847                                                                 device);
1848         ASSERT(next_device);
1849
1850         if (fs_info->sb->s_bdev &&
1851                         (fs_info->sb->s_bdev == device->bdev))
1852                 fs_info->sb->s_bdev = next_device->bdev;
1853
1854         if (fs_info->fs_devices->latest_bdev == device->bdev)
1855                 fs_info->fs_devices->latest_bdev = next_device->bdev;
1856 }
1857
1858 int btrfs_rm_device(struct btrfs_fs_info *fs_info, const char *device_path,
1859                 u64 devid)
1860 {
1861         struct btrfs_device *device;
1862         struct btrfs_fs_devices *cur_devices;
1863         u64 num_devices;
1864         int ret = 0;
1865         bool clear_super = false;
1866
1867         mutex_lock(&uuid_mutex);
1868
1869         num_devices = fs_info->fs_devices->num_devices;
1870         btrfs_dev_replace_lock(&fs_info->dev_replace, 0);
1871         if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
1872                 WARN_ON(num_devices < 1);
1873                 num_devices--;
1874         }
1875         btrfs_dev_replace_unlock(&fs_info->dev_replace, 0);
1876
1877         ret = btrfs_check_raid_min_devices(fs_info, num_devices - 1);
1878         if (ret)
1879                 goto out;
1880
1881         ret = btrfs_find_device_by_devspec(fs_info, devid, device_path,
1882                                            &device);
1883         if (ret)
1884                 goto out;
1885
1886         if (device->is_tgtdev_for_dev_replace) {
1887                 ret = BTRFS_ERROR_DEV_TGT_REPLACE;
1888                 goto out;
1889         }
1890
1891         if (device->writeable && fs_info->fs_devices->rw_devices == 1) {
1892                 ret = BTRFS_ERROR_DEV_ONLY_WRITABLE;
1893                 goto out;
1894         }
1895
1896         if (device->writeable) {
1897                 mutex_lock(&fs_info->chunk_mutex);
1898                 list_del_init(&device->dev_alloc_list);
1899                 device->fs_devices->rw_devices--;
1900                 mutex_unlock(&fs_info->chunk_mutex);
1901                 clear_super = true;
1902         }
1903
1904         mutex_unlock(&uuid_mutex);
1905         ret = btrfs_shrink_device(device, 0);
1906         mutex_lock(&uuid_mutex);
1907         if (ret)
1908                 goto error_undo;
1909
1910         /*
1911          * TODO: the superblock still includes this device in its num_devices
1912          * counter although write_all_supers() is not locked out. This
1913          * could give a filesystem state which requires a degraded mount.
1914          */
1915         ret = btrfs_rm_dev_item(fs_info, device);
1916         if (ret)
1917                 goto error_undo;
1918
1919         device->in_fs_metadata = 0;
1920         btrfs_scrub_cancel_dev(fs_info, device);
1921
1922         /*
1923          * the device list mutex makes sure that we don't change
1924          * the device list while someone else is writing out all
1925          * the device supers. Whoever is writing all supers, should
1926          * lock the device list mutex before getting the number of
1927          * devices in the super block (super_copy). Conversely,
1928          * whoever updates the number of devices in the super block
1929          * (super_copy) should hold the device list mutex.
1930          */
1931
1932         cur_devices = device->fs_devices;
1933         mutex_lock(&fs_info->fs_devices->device_list_mutex);
1934         list_del_rcu(&device->dev_list);
1935
1936         device->fs_devices->num_devices--;
1937         device->fs_devices->total_devices--;
1938
1939         if (device->missing)
1940                 device->fs_devices->missing_devices--;
1941
1942         btrfs_assign_next_active_device(fs_info, device, NULL);
1943
1944         if (device->bdev) {
1945                 device->fs_devices->open_devices--;
1946                 /* remove sysfs entry */
1947                 btrfs_sysfs_rm_device_link(fs_info->fs_devices, device);
1948         }
1949
1950         num_devices = btrfs_super_num_devices(fs_info->super_copy) - 1;
1951         btrfs_set_super_num_devices(fs_info->super_copy, num_devices);
1952         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
1953
1954         /*
1955          * at this point, the device is zero sized and detached from
1956          * the devices list.  All that's left is to zero out the old
1957          * supers and free the device.
1958          */
1959         if (device->writeable)
1960                 btrfs_scratch_superblocks(device->bdev, device->name->str);
1961
1962         btrfs_close_bdev(device);
1963         call_rcu(&device->rcu, free_device);
1964
1965         if (cur_devices->open_devices == 0) {
1966                 struct btrfs_fs_devices *fs_devices;
1967                 fs_devices = fs_info->fs_devices;
1968                 while (fs_devices) {
1969                         if (fs_devices->seed == cur_devices) {
1970                                 fs_devices->seed = cur_devices->seed;
1971                                 break;
1972                         }
1973                         fs_devices = fs_devices->seed;
1974                 }
1975                 cur_devices->seed = NULL;
1976                 __btrfs_close_devices(cur_devices);
1977                 free_fs_devices(cur_devices);
1978         }
1979
1980         fs_info->num_tolerated_disk_barrier_failures =
1981                 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
1982
1983 out:
1984         mutex_unlock(&uuid_mutex);
1985         return ret;
1986
1987 error_undo:
1988         if (device->writeable) {
1989                 mutex_lock(&fs_info->chunk_mutex);
1990                 list_add(&device->dev_alloc_list,
1991                          &fs_info->fs_devices->alloc_list);
1992                 device->fs_devices->rw_devices++;
1993                 mutex_unlock(&fs_info->chunk_mutex);
1994         }
1995         goto out;
1996 }
1997
1998 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_fs_info *fs_info,
1999                                         struct btrfs_device *srcdev)
2000 {
2001         struct btrfs_fs_devices *fs_devices;
2002
2003         WARN_ON(!mutex_is_locked(&fs_info->fs_devices->device_list_mutex));
2004
2005         /*
2006          * in case of fs with no seed, srcdev->fs_devices will point
2007          * to fs_devices of fs_info. However when the dev being replaced is
2008          * a seed dev it will point to the seed's local fs_devices. In short
2009          * srcdev will have its correct fs_devices in both the cases.
2010          */
2011         fs_devices = srcdev->fs_devices;
2012
2013         list_del_rcu(&srcdev->dev_list);
2014         list_del_rcu(&srcdev->dev_alloc_list);
2015         fs_devices->num_devices--;
2016         if (srcdev->missing)
2017                 fs_devices->missing_devices--;
2018
2019         if (srcdev->writeable)
2020                 fs_devices->rw_devices--;
2021
2022         if (srcdev->bdev)
2023                 fs_devices->open_devices--;
2024 }
2025
2026 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_fs_info *fs_info,
2027                                       struct btrfs_device *srcdev)
2028 {
2029         struct btrfs_fs_devices *fs_devices = srcdev->fs_devices;
2030
2031         if (srcdev->writeable) {
2032                 /* zero out the old super if it is writable */
2033                 btrfs_scratch_superblocks(srcdev->bdev, srcdev->name->str);
2034         }
2035
2036         btrfs_close_bdev(srcdev);
2037
2038         call_rcu(&srcdev->rcu, free_device);
2039
2040         /*
2041          * unless fs_devices is seed fs, num_devices shouldn't go
2042          * zero
2043          */
2044         BUG_ON(!fs_devices->num_devices && !fs_devices->seeding);
2045
2046         /* if this is no devs we rather delete the fs_devices */
2047         if (!fs_devices->num_devices) {
2048                 struct btrfs_fs_devices *tmp_fs_devices;
2049
2050                 tmp_fs_devices = fs_info->fs_devices;
2051                 while (tmp_fs_devices) {
2052                         if (tmp_fs_devices->seed == fs_devices) {
2053                                 tmp_fs_devices->seed = fs_devices->seed;
2054                                 break;
2055                         }
2056                         tmp_fs_devices = tmp_fs_devices->seed;
2057                 }
2058                 fs_devices->seed = NULL;
2059                 __btrfs_close_devices(fs_devices);
2060                 free_fs_devices(fs_devices);
2061         }
2062 }
2063
2064 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
2065                                       struct btrfs_device *tgtdev)
2066 {
2067         mutex_lock(&uuid_mutex);
2068         WARN_ON(!tgtdev);
2069         mutex_lock(&fs_info->fs_devices->device_list_mutex);
2070
2071         btrfs_sysfs_rm_device_link(fs_info->fs_devices, tgtdev);
2072
2073         if (tgtdev->bdev)
2074                 fs_info->fs_devices->open_devices--;
2075
2076         fs_info->fs_devices->num_devices--;
2077
2078         btrfs_assign_next_active_device(fs_info, tgtdev, NULL);
2079
2080         list_del_rcu(&tgtdev->dev_list);
2081
2082         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2083         mutex_unlock(&uuid_mutex);
2084
2085         /*
2086          * The update_dev_time() with in btrfs_scratch_superblocks()
2087          * may lead to a call to btrfs_show_devname() which will try
2088          * to hold device_list_mutex. And here this device
2089          * is already out of device list, so we don't have to hold
2090          * the device_list_mutex lock.
2091          */
2092         btrfs_scratch_superblocks(tgtdev->bdev, tgtdev->name->str);
2093
2094         btrfs_close_bdev(tgtdev);
2095         call_rcu(&tgtdev->rcu, free_device);
2096 }
2097
2098 static int btrfs_find_device_by_path(struct btrfs_fs_info *fs_info,
2099                                      const char *device_path,
2100                                      struct btrfs_device **device)
2101 {
2102         int ret = 0;
2103         struct btrfs_super_block *disk_super;
2104         u64 devid;
2105         u8 *dev_uuid;
2106         struct block_device *bdev;
2107         struct buffer_head *bh;
2108
2109         *device = NULL;
2110         ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
2111                                     fs_info->bdev_holder, 0, &bdev, &bh);
2112         if (ret)
2113                 return ret;
2114         disk_super = (struct btrfs_super_block *)bh->b_data;
2115         devid = btrfs_stack_device_id(&disk_super->dev_item);
2116         dev_uuid = disk_super->dev_item.uuid;
2117         *device = btrfs_find_device(fs_info, devid, dev_uuid, disk_super->fsid);
2118         brelse(bh);
2119         if (!*device)
2120                 ret = -ENOENT;
2121         blkdev_put(bdev, FMODE_READ);
2122         return ret;
2123 }
2124
2125 int btrfs_find_device_missing_or_by_path(struct btrfs_fs_info *fs_info,
2126                                          const char *device_path,
2127                                          struct btrfs_device **device)
2128 {
2129         *device = NULL;
2130         if (strcmp(device_path, "missing") == 0) {
2131                 struct list_head *devices;
2132                 struct btrfs_device *tmp;
2133
2134                 devices = &fs_info->fs_devices->devices;
2135                 /*
2136                  * It is safe to read the devices since the volume_mutex
2137                  * is held by the caller.
2138                  */
2139                 list_for_each_entry(tmp, devices, dev_list) {
2140                         if (tmp->in_fs_metadata && !tmp->bdev) {
2141                                 *device = tmp;
2142                                 break;
2143                         }
2144                 }
2145
2146                 if (!*device)
2147                         return BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
2148
2149                 return 0;
2150         } else {
2151                 return btrfs_find_device_by_path(fs_info, device_path, device);
2152         }
2153 }
2154
2155 /*
2156  * Lookup a device given by device id, or the path if the id is 0.
2157  */
2158 int btrfs_find_device_by_devspec(struct btrfs_fs_info *fs_info, u64 devid,
2159                                  const char *devpath,
2160                                  struct btrfs_device **device)
2161 {
2162         int ret;
2163
2164         if (devid) {
2165                 ret = 0;
2166                 *device = btrfs_find_device(fs_info, devid, NULL, NULL);
2167                 if (!*device)
2168                         ret = -ENOENT;
2169         } else {
2170                 if (!devpath || !devpath[0])
2171                         return -EINVAL;
2172
2173                 ret = btrfs_find_device_missing_or_by_path(fs_info, devpath,
2174                                                            device);
2175         }
2176         return ret;
2177 }
2178
2179 /*
2180  * does all the dirty work required for changing file system's UUID.
2181  */
2182 static int btrfs_prepare_sprout(struct btrfs_fs_info *fs_info)
2183 {
2184         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2185         struct btrfs_fs_devices *old_devices;
2186         struct btrfs_fs_devices *seed_devices;
2187         struct btrfs_super_block *disk_super = fs_info->super_copy;
2188         struct btrfs_device *device;
2189         u64 super_flags;
2190
2191         BUG_ON(!mutex_is_locked(&uuid_mutex));
2192         if (!fs_devices->seeding)
2193                 return -EINVAL;
2194
2195         seed_devices = __alloc_fs_devices();
2196         if (IS_ERR(seed_devices))
2197                 return PTR_ERR(seed_devices);
2198
2199         old_devices = clone_fs_devices(fs_devices);
2200         if (IS_ERR(old_devices)) {
2201                 kfree(seed_devices);
2202                 return PTR_ERR(old_devices);
2203         }
2204
2205         list_add(&old_devices->list, &fs_uuids);
2206
2207         memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
2208         seed_devices->opened = 1;
2209         INIT_LIST_HEAD(&seed_devices->devices);
2210         INIT_LIST_HEAD(&seed_devices->alloc_list);
2211         mutex_init(&seed_devices->device_list_mutex);
2212
2213         mutex_lock(&fs_info->fs_devices->device_list_mutex);
2214         list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
2215                               synchronize_rcu);
2216         list_for_each_entry(device, &seed_devices->devices, dev_list)
2217                 device->fs_devices = seed_devices;
2218
2219         mutex_lock(&fs_info->chunk_mutex);
2220         list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
2221         mutex_unlock(&fs_info->chunk_mutex);
2222
2223         fs_devices->seeding = 0;
2224         fs_devices->num_devices = 0;
2225         fs_devices->open_devices = 0;
2226         fs_devices->missing_devices = 0;
2227         fs_devices->rotating = 0;
2228         fs_devices->seed = seed_devices;
2229
2230         generate_random_uuid(fs_devices->fsid);
2231         memcpy(fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2232         memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2233         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2234
2235         super_flags = btrfs_super_flags(disk_super) &
2236                       ~BTRFS_SUPER_FLAG_SEEDING;
2237         btrfs_set_super_flags(disk_super, super_flags);
2238
2239         return 0;
2240 }
2241
2242 /*
2243  * Store the expected generation for seed devices in device items.
2244  */
2245 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
2246                                struct btrfs_fs_info *fs_info)
2247 {
2248         struct btrfs_root *root = fs_info->chunk_root;
2249         struct btrfs_path *path;
2250         struct extent_buffer *leaf;
2251         struct btrfs_dev_item *dev_item;
2252         struct btrfs_device *device;
2253         struct btrfs_key key;
2254         u8 fs_uuid[BTRFS_UUID_SIZE];
2255         u8 dev_uuid[BTRFS_UUID_SIZE];
2256         u64 devid;
2257         int ret;
2258
2259         path = btrfs_alloc_path();
2260         if (!path)
2261                 return -ENOMEM;
2262
2263         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2264         key.offset = 0;
2265         key.type = BTRFS_DEV_ITEM_KEY;
2266
2267         while (1) {
2268                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2269                 if (ret < 0)
2270                         goto error;
2271
2272                 leaf = path->nodes[0];
2273 next_slot:
2274                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2275                         ret = btrfs_next_leaf(root, path);
2276                         if (ret > 0)
2277                                 break;
2278                         if (ret < 0)
2279                                 goto error;
2280                         leaf = path->nodes[0];
2281                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2282                         btrfs_release_path(path);
2283                         continue;
2284                 }
2285
2286                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2287                 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
2288                     key.type != BTRFS_DEV_ITEM_KEY)
2289                         break;
2290
2291                 dev_item = btrfs_item_ptr(leaf, path->slots[0],
2292                                           struct btrfs_dev_item);
2293                 devid = btrfs_device_id(leaf, dev_item);
2294                 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
2295                                    BTRFS_UUID_SIZE);
2296                 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
2297                                    BTRFS_UUID_SIZE);
2298                 device = btrfs_find_device(fs_info, devid, dev_uuid, fs_uuid);
2299                 BUG_ON(!device); /* Logic error */
2300
2301                 if (device->fs_devices->seeding) {
2302                         btrfs_set_device_generation(leaf, dev_item,
2303                                                     device->generation);
2304                         btrfs_mark_buffer_dirty(leaf);
2305                 }
2306
2307                 path->slots[0]++;
2308                 goto next_slot;
2309         }
2310         ret = 0;
2311 error:
2312         btrfs_free_path(path);
2313         return ret;
2314 }
2315
2316 int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path)
2317 {
2318         struct btrfs_root *root = fs_info->dev_root;
2319         struct request_queue *q;
2320         struct btrfs_trans_handle *trans;
2321         struct btrfs_device *device;
2322         struct block_device *bdev;
2323         struct list_head *devices;
2324         struct super_block *sb = fs_info->sb;
2325         struct rcu_string *name;
2326         u64 tmp;
2327         int seeding_dev = 0;
2328         int ret = 0;
2329
2330         if ((sb->s_flags & MS_RDONLY) && !fs_info->fs_devices->seeding)
2331                 return -EROFS;
2332
2333         bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2334                                   fs_info->bdev_holder);
2335         if (IS_ERR(bdev))
2336                 return PTR_ERR(bdev);
2337
2338         if (fs_info->fs_devices->seeding) {
2339                 seeding_dev = 1;
2340                 down_write(&sb->s_umount);
2341                 mutex_lock(&uuid_mutex);
2342         }
2343
2344         filemap_write_and_wait(bdev->bd_inode->i_mapping);
2345
2346         devices = &fs_info->fs_devices->devices;
2347
2348         mutex_lock(&fs_info->fs_devices->device_list_mutex);
2349         list_for_each_entry(device, devices, dev_list) {
2350                 if (device->bdev == bdev) {
2351                         ret = -EEXIST;
2352                         mutex_unlock(
2353                                 &fs_info->fs_devices->device_list_mutex);
2354                         goto error;
2355                 }
2356         }
2357         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2358
2359         device = btrfs_alloc_device(fs_info, NULL, NULL);
2360         if (IS_ERR(device)) {
2361                 /* we can safely leave the fs_devices entry around */
2362                 ret = PTR_ERR(device);
2363                 goto error;
2364         }
2365
2366         name = rcu_string_strdup(device_path, GFP_KERNEL);
2367         if (!name) {
2368                 kfree(device);
2369                 ret = -ENOMEM;
2370                 goto error;
2371         }
2372         rcu_assign_pointer(device->name, name);
2373
2374         trans = btrfs_start_transaction(root, 0);
2375         if (IS_ERR(trans)) {
2376                 rcu_string_free(device->name);
2377                 kfree(device);
2378                 ret = PTR_ERR(trans);
2379                 goto error;
2380         }
2381
2382         q = bdev_get_queue(bdev);
2383         if (blk_queue_discard(q))
2384                 device->can_discard = 1;
2385         device->writeable = 1;
2386         device->generation = trans->transid;
2387         device->io_width = fs_info->sectorsize;
2388         device->io_align = fs_info->sectorsize;
2389         device->sector_size = fs_info->sectorsize;
2390         device->total_bytes = i_size_read(bdev->bd_inode);
2391         device->disk_total_bytes = device->total_bytes;
2392         device->commit_total_bytes = device->total_bytes;
2393         device->fs_info = fs_info;
2394         device->bdev = bdev;
2395         device->in_fs_metadata = 1;
2396         device->is_tgtdev_for_dev_replace = 0;
2397         device->mode = FMODE_EXCL;
2398         device->dev_stats_valid = 1;
2399         set_blocksize(device->bdev, 4096);
2400
2401         if (seeding_dev) {
2402                 sb->s_flags &= ~MS_RDONLY;
2403                 ret = btrfs_prepare_sprout(fs_info);
2404                 BUG_ON(ret); /* -ENOMEM */
2405         }
2406
2407         device->fs_devices = fs_info->fs_devices;
2408
2409         mutex_lock(&fs_info->fs_devices->device_list_mutex);
2410         mutex_lock(&fs_info->chunk_mutex);
2411         list_add_rcu(&device->dev_list, &fs_info->fs_devices->devices);
2412         list_add(&device->dev_alloc_list,
2413                  &fs_info->fs_devices->alloc_list);
2414         fs_info->fs_devices->num_devices++;
2415         fs_info->fs_devices->open_devices++;
2416         fs_info->fs_devices->rw_devices++;
2417         fs_info->fs_devices->total_devices++;
2418         fs_info->fs_devices->total_rw_bytes += device->total_bytes;
2419
2420         atomic64_add(device->total_bytes, &fs_info->free_chunk_space);
2421
2422         if (!blk_queue_nonrot(q))
2423                 fs_info->fs_devices->rotating = 1;
2424
2425         tmp = btrfs_super_total_bytes(fs_info->super_copy);
2426         btrfs_set_super_total_bytes(fs_info->super_copy,
2427                                     tmp + device->total_bytes);
2428
2429         tmp = btrfs_super_num_devices(fs_info->super_copy);
2430         btrfs_set_super_num_devices(fs_info->super_copy, tmp + 1);
2431
2432         /* add sysfs device entry */
2433         btrfs_sysfs_add_device_link(fs_info->fs_devices, device);
2434
2435         /*
2436          * we've got more storage, clear any full flags on the space
2437          * infos
2438          */
2439         btrfs_clear_space_info_full(fs_info);
2440
2441         mutex_unlock(&fs_info->chunk_mutex);
2442         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2443
2444         if (seeding_dev) {
2445                 mutex_lock(&fs_info->chunk_mutex);
2446                 ret = init_first_rw_device(trans, fs_info);
2447                 mutex_unlock(&fs_info->chunk_mutex);
2448                 if (ret) {
2449                         btrfs_abort_transaction(trans, ret);
2450                         goto error_trans;
2451                 }
2452         }
2453
2454         ret = btrfs_add_device(trans, fs_info, device);
2455         if (ret) {
2456                 btrfs_abort_transaction(trans, ret);
2457                 goto error_trans;
2458         }
2459
2460         if (seeding_dev) {
2461                 char fsid_buf[BTRFS_UUID_UNPARSED_SIZE];
2462
2463                 ret = btrfs_finish_sprout(trans, fs_info);
2464                 if (ret) {
2465                         btrfs_abort_transaction(trans, ret);
2466                         goto error_trans;
2467                 }
2468
2469                 /* Sprouting would change fsid of the mounted root,
2470                  * so rename the fsid on the sysfs
2471                  */
2472                 snprintf(fsid_buf, BTRFS_UUID_UNPARSED_SIZE, "%pU",
2473                                                 fs_info->fsid);
2474                 if (kobject_rename(&fs_info->fs_devices->fsid_kobj, fsid_buf))
2475                         btrfs_warn(fs_info,
2476                                    "sysfs: failed to create fsid for sprout");
2477         }
2478
2479         fs_info->num_tolerated_disk_barrier_failures =
2480                 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
2481         ret = btrfs_commit_transaction(trans);
2482
2483         if (seeding_dev) {
2484                 mutex_unlock(&uuid_mutex);
2485                 up_write(&sb->s_umount);
2486
2487                 if (ret) /* transaction commit */
2488                         return ret;
2489
2490                 ret = btrfs_relocate_sys_chunks(fs_info);
2491                 if (ret < 0)
2492                         btrfs_handle_fs_error(fs_info, ret,
2493                                     "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command.");
2494                 trans = btrfs_attach_transaction(root);
2495                 if (IS_ERR(trans)) {
2496                         if (PTR_ERR(trans) == -ENOENT)
2497                                 return 0;
2498                         return PTR_ERR(trans);
2499                 }
2500                 ret = btrfs_commit_transaction(trans);
2501         }
2502
2503         /* Update ctime/mtime for libblkid */
2504         update_dev_time(device_path);
2505         return ret;
2506
2507 error_trans:
2508         btrfs_end_transaction(trans);
2509         rcu_string_free(device->name);
2510         btrfs_sysfs_rm_device_link(fs_info->fs_devices, device);
2511         kfree(device);
2512 error:
2513         blkdev_put(bdev, FMODE_EXCL);
2514         if (seeding_dev) {
2515                 mutex_unlock(&uuid_mutex);
2516                 up_write(&sb->s_umount);
2517         }
2518         return ret;
2519 }
2520
2521 int btrfs_init_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
2522                                   const char *device_path,
2523                                   struct btrfs_device *srcdev,
2524                                   struct btrfs_device **device_out)
2525 {
2526         struct request_queue *q;
2527         struct btrfs_device *device;
2528         struct block_device *bdev;
2529         struct list_head *devices;
2530         struct rcu_string *name;
2531         u64 devid = BTRFS_DEV_REPLACE_DEVID;
2532         int ret = 0;
2533
2534         *device_out = NULL;
2535         if (fs_info->fs_devices->seeding) {
2536                 btrfs_err(fs_info, "the filesystem is a seed filesystem!");
2537                 return -EINVAL;
2538         }
2539
2540         bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2541                                   fs_info->bdev_holder);
2542         if (IS_ERR(bdev)) {
2543                 btrfs_err(fs_info, "target device %s is invalid!", device_path);
2544                 return PTR_ERR(bdev);
2545         }
2546
2547         filemap_write_and_wait(bdev->bd_inode->i_mapping);
2548
2549         devices = &fs_info->fs_devices->devices;
2550         list_for_each_entry(device, devices, dev_list) {
2551                 if (device->bdev == bdev) {
2552                         btrfs_err(fs_info,
2553                                   "target device is in the filesystem!");
2554                         ret = -EEXIST;
2555                         goto error;
2556                 }
2557         }
2558
2559
2560         if (i_size_read(bdev->bd_inode) <
2561             btrfs_device_get_total_bytes(srcdev)) {
2562                 btrfs_err(fs_info,
2563                           "target device is smaller than source device!");
2564                 ret = -EINVAL;
2565                 goto error;
2566         }
2567
2568
2569         device = btrfs_alloc_device(NULL, &devid, NULL);
2570         if (IS_ERR(device)) {
2571                 ret = PTR_ERR(device);
2572                 goto error;
2573         }
2574
2575         name = rcu_string_strdup(device_path, GFP_NOFS);
2576         if (!name) {
2577                 kfree(device);
2578                 ret = -ENOMEM;
2579                 goto error;
2580         }
2581         rcu_assign_pointer(device->name, name);
2582
2583         q = bdev_get_queue(bdev);
2584         if (blk_queue_discard(q))
2585                 device->can_discard = 1;
2586         mutex_lock(&fs_info->fs_devices->device_list_mutex);
2587         device->writeable = 1;
2588         device->generation = 0;
2589         device->io_width = fs_info->sectorsize;
2590         device->io_align = fs_info->sectorsize;
2591         device->sector_size = fs_info->sectorsize;
2592         device->total_bytes = btrfs_device_get_total_bytes(srcdev);
2593         device->disk_total_bytes = btrfs_device_get_disk_total_bytes(srcdev);
2594         device->bytes_used = btrfs_device_get_bytes_used(srcdev);
2595         ASSERT(list_empty(&srcdev->resized_list));
2596         device->commit_total_bytes = srcdev->commit_total_bytes;
2597         device->commit_bytes_used = device->bytes_used;
2598         device->fs_info = fs_info;
2599         device->bdev = bdev;
2600         device->in_fs_metadata = 1;
2601         device->is_tgtdev_for_dev_replace = 1;
2602         device->mode = FMODE_EXCL;
2603         device->dev_stats_valid = 1;
2604         set_blocksize(device->bdev, 4096);
2605         device->fs_devices = fs_info->fs_devices;
2606         list_add(&device->dev_list, &fs_info->fs_devices->devices);
2607         fs_info->fs_devices->num_devices++;
2608         fs_info->fs_devices->open_devices++;
2609         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2610
2611         *device_out = device;
2612         return ret;
2613
2614 error:
2615         blkdev_put(bdev, FMODE_EXCL);
2616         return ret;
2617 }
2618
2619 void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info *fs_info,
2620                                               struct btrfs_device *tgtdev)
2621 {
2622         u32 sectorsize = fs_info->sectorsize;
2623
2624         WARN_ON(fs_info->fs_devices->rw_devices == 0);
2625         tgtdev->io_width = sectorsize;
2626         tgtdev->io_align = sectorsize;
2627         tgtdev->sector_size = sectorsize;
2628         tgtdev->fs_info = fs_info;
2629         tgtdev->in_fs_metadata = 1;
2630 }
2631
2632 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2633                                         struct btrfs_device *device)
2634 {
2635         int ret;
2636         struct btrfs_path *path;
2637         struct btrfs_root *root = device->fs_info->chunk_root;
2638         struct btrfs_dev_item *dev_item;
2639         struct extent_buffer *leaf;
2640         struct btrfs_key key;
2641
2642         path = btrfs_alloc_path();
2643         if (!path)
2644                 return -ENOMEM;
2645
2646         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2647         key.type = BTRFS_DEV_ITEM_KEY;
2648         key.offset = device->devid;
2649
2650         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2651         if (ret < 0)
2652                 goto out;
2653
2654         if (ret > 0) {
2655                 ret = -ENOENT;
2656                 goto out;
2657         }
2658
2659         leaf = path->nodes[0];
2660         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2661
2662         btrfs_set_device_id(leaf, dev_item, device->devid);
2663         btrfs_set_device_type(leaf, dev_item, device->type);
2664         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2665         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2666         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2667         btrfs_set_device_total_bytes(leaf, dev_item,
2668                                      btrfs_device_get_disk_total_bytes(device));
2669         btrfs_set_device_bytes_used(leaf, dev_item,
2670                                     btrfs_device_get_bytes_used(device));
2671         btrfs_mark_buffer_dirty(leaf);
2672
2673 out:
2674         btrfs_free_path(path);
2675         return ret;
2676 }
2677
2678 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2679                       struct btrfs_device *device, u64 new_size)
2680 {
2681         struct btrfs_fs_info *fs_info = device->fs_info;
2682         struct btrfs_super_block *super_copy = fs_info->super_copy;
2683         struct btrfs_fs_devices *fs_devices;
2684         u64 old_total;
2685         u64 diff;
2686
2687         if (!device->writeable)
2688                 return -EACCES;
2689
2690         mutex_lock(&fs_info->chunk_mutex);
2691         old_total = btrfs_super_total_bytes(super_copy);
2692         diff = new_size - device->total_bytes;
2693
2694         if (new_size <= device->total_bytes ||
2695             device->is_tgtdev_for_dev_replace) {
2696                 mutex_unlock(&fs_info->chunk_mutex);
2697                 return -EINVAL;
2698         }
2699
2700         fs_devices = fs_info->fs_devices;
2701
2702         btrfs_set_super_total_bytes(super_copy, old_total + diff);
2703         device->fs_devices->total_rw_bytes += diff;
2704
2705         btrfs_device_set_total_bytes(device, new_size);
2706         btrfs_device_set_disk_total_bytes(device, new_size);
2707         btrfs_clear_space_info_full(device->fs_info);
2708         if (list_empty(&device->resized_list))
2709                 list_add_tail(&device->resized_list,
2710                               &fs_devices->resized_devices);
2711         mutex_unlock(&fs_info->chunk_mutex);
2712
2713         return btrfs_update_device(trans, device);
2714 }
2715
2716 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
2717                             struct btrfs_fs_info *fs_info, u64 chunk_objectid,
2718                             u64 chunk_offset)
2719 {
2720         struct btrfs_root *root = fs_info->chunk_root;
2721         int ret;
2722         struct btrfs_path *path;
2723         struct btrfs_key key;
2724
2725         path = btrfs_alloc_path();
2726         if (!path)
2727                 return -ENOMEM;
2728
2729         key.objectid = chunk_objectid;
2730         key.offset = chunk_offset;
2731         key.type = BTRFS_CHUNK_ITEM_KEY;
2732
2733         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2734         if (ret < 0)
2735                 goto out;
2736         else if (ret > 0) { /* Logic error or corruption */
2737                 btrfs_handle_fs_error(fs_info, -ENOENT,
2738                                       "Failed lookup while freeing chunk.");
2739                 ret = -ENOENT;
2740                 goto out;
2741         }
2742
2743         ret = btrfs_del_item(trans, root, path);
2744         if (ret < 0)
2745                 btrfs_handle_fs_error(fs_info, ret,
2746                                       "Failed to delete chunk item.");
2747 out:
2748         btrfs_free_path(path);
2749         return ret;
2750 }
2751
2752 static int btrfs_del_sys_chunk(struct btrfs_fs_info *fs_info,
2753                                u64 chunk_objectid, u64 chunk_offset)
2754 {
2755         struct btrfs_super_block *super_copy = fs_info->super_copy;
2756         struct btrfs_disk_key *disk_key;
2757         struct btrfs_chunk *chunk;
2758         u8 *ptr;
2759         int ret = 0;
2760         u32 num_stripes;
2761         u32 array_size;
2762         u32 len = 0;
2763         u32 cur;
2764         struct btrfs_key key;
2765
2766         mutex_lock(&fs_info->chunk_mutex);
2767         array_size = btrfs_super_sys_array_size(super_copy);
2768
2769         ptr = super_copy->sys_chunk_array;
2770         cur = 0;
2771
2772         while (cur < array_size) {
2773                 disk_key = (struct btrfs_disk_key *)ptr;
2774                 btrfs_disk_key_to_cpu(&key, disk_key);
2775
2776                 len = sizeof(*disk_key);
2777
2778                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2779                         chunk = (struct btrfs_chunk *)(ptr + len);
2780                         num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2781                         len += btrfs_chunk_item_size(num_stripes);
2782                 } else {
2783                         ret = -EIO;
2784                         break;
2785                 }
2786                 if (key.objectid == chunk_objectid &&
2787                     key.offset == chunk_offset) {
2788                         memmove(ptr, ptr + len, array_size - (cur + len));
2789                         array_size -= len;
2790                         btrfs_set_super_sys_array_size(super_copy, array_size);
2791                 } else {
2792                         ptr += len;
2793                         cur += len;
2794                 }
2795         }
2796         mutex_unlock(&fs_info->chunk_mutex);
2797         return ret;
2798 }
2799
2800 static struct extent_map *get_chunk_map(struct btrfs_fs_info *fs_info,
2801                                         u64 logical, u64 length)
2802 {
2803         struct extent_map_tree *em_tree;
2804         struct extent_map *em;
2805
2806         em_tree = &fs_info->mapping_tree.map_tree;
2807         read_lock(&em_tree->lock);
2808         em = lookup_extent_mapping(em_tree, logical, length);
2809         read_unlock(&em_tree->lock);
2810
2811         if (!em) {
2812                 btrfs_crit(fs_info, "unable to find logical %llu length %llu",
2813                            logical, length);
2814                 return ERR_PTR(-EINVAL);
2815         }
2816
2817         if (em->start > logical || em->start + em->len < logical) {
2818                 btrfs_crit(fs_info,
2819                            "found a bad mapping, wanted %llu-%llu, found %llu-%llu",
2820                            logical, length, em->start, em->start + em->len);
2821                 free_extent_map(em);
2822                 return ERR_PTR(-EINVAL);
2823         }
2824
2825         /* callers are responsible for dropping em's ref. */
2826         return em;
2827 }
2828
2829 int btrfs_remove_chunk(struct btrfs_trans_handle *trans,
2830                        struct btrfs_fs_info *fs_info, u64 chunk_offset)
2831 {
2832         struct extent_map *em;
2833         struct map_lookup *map;
2834         u64 dev_extent_len = 0;
2835         u64 chunk_objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2836         int i, ret = 0;
2837         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2838
2839         em = get_chunk_map(fs_info, chunk_offset, 1);
2840         if (IS_ERR(em)) {
2841                 /*
2842                  * This is a logic error, but we don't want to just rely on the
2843                  * user having built with ASSERT enabled, so if ASSERT doesn't
2844                  * do anything we still error out.
2845                  */
2846                 ASSERT(0);
2847                 return PTR_ERR(em);
2848         }
2849         map = em->map_lookup;
2850         mutex_lock(&fs_info->chunk_mutex);
2851         check_system_chunk(trans, fs_info, map->type);
2852         mutex_unlock(&fs_info->chunk_mutex);
2853
2854         /*
2855          * Take the device list mutex to prevent races with the final phase of
2856          * a device replace operation that replaces the device object associated
2857          * with map stripes (dev-replace.c:btrfs_dev_replace_finishing()).
2858          */
2859         mutex_lock(&fs_devices->device_list_mutex);
2860         for (i = 0; i < map->num_stripes; i++) {
2861                 struct btrfs_device *device = map->stripes[i].dev;
2862                 ret = btrfs_free_dev_extent(trans, device,
2863                                             map->stripes[i].physical,
2864                                             &dev_extent_len);
2865                 if (ret) {
2866                         mutex_unlock(&fs_devices->device_list_mutex);
2867                         btrfs_abort_transaction(trans, ret);
2868                         goto out;
2869                 }
2870
2871                 if (device->bytes_used > 0) {
2872                         mutex_lock(&fs_info->chunk_mutex);
2873                         btrfs_device_set_bytes_used(device,
2874                                         device->bytes_used - dev_extent_len);
2875                         atomic64_add(dev_extent_len, &fs_info->free_chunk_space);
2876                         btrfs_clear_space_info_full(fs_info);
2877                         mutex_unlock(&fs_info->chunk_mutex);
2878                 }
2879
2880                 if (map->stripes[i].dev) {
2881                         ret = btrfs_update_device(trans, map->stripes[i].dev);
2882                         if (ret) {
2883                                 mutex_unlock(&fs_devices->device_list_mutex);
2884                                 btrfs_abort_transaction(trans, ret);
2885                                 goto out;
2886                         }
2887                 }
2888         }
2889         mutex_unlock(&fs_devices->device_list_mutex);
2890
2891         ret = btrfs_free_chunk(trans, fs_info, chunk_objectid, chunk_offset);
2892         if (ret) {
2893                 btrfs_abort_transaction(trans, ret);
2894                 goto out;
2895         }
2896
2897         trace_btrfs_chunk_free(fs_info, map, chunk_offset, em->len);
2898
2899         if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2900                 ret = btrfs_del_sys_chunk(fs_info, chunk_objectid,
2901                                           chunk_offset);
2902                 if (ret) {
2903                         btrfs_abort_transaction(trans, ret);
2904                         goto out;
2905                 }
2906         }
2907
2908         ret = btrfs_remove_block_group(trans, fs_info, chunk_offset, em);
2909         if (ret) {
2910                 btrfs_abort_transaction(trans, ret);
2911                 goto out;
2912         }
2913
2914 out:
2915         /* once for us */
2916         free_extent_map(em);
2917         return ret;
2918 }
2919
2920 static int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
2921 {
2922         struct btrfs_root *root = fs_info->chunk_root;
2923         struct btrfs_trans_handle *trans;
2924         int ret;
2925
2926         /*
2927          * Prevent races with automatic removal of unused block groups.
2928          * After we relocate and before we remove the chunk with offset
2929          * chunk_offset, automatic removal of the block group can kick in,
2930          * resulting in a failure when calling btrfs_remove_chunk() below.
2931          *
2932          * Make sure to acquire this mutex before doing a tree search (dev
2933          * or chunk trees) to find chunks. Otherwise the cleaner kthread might
2934          * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after
2935          * we release the path used to search the chunk/dev tree and before
2936          * the current task acquires this mutex and calls us.
2937          */
2938         ASSERT(mutex_is_locked(&fs_info->delete_unused_bgs_mutex));
2939
2940         ret = btrfs_can_relocate(fs_info, chunk_offset);
2941         if (ret)
2942                 return -ENOSPC;
2943
2944         /* step one, relocate all the extents inside this chunk */
2945         btrfs_scrub_pause(fs_info);
2946         ret = btrfs_relocate_block_group(fs_info, chunk_offset);
2947         btrfs_scrub_continue(fs_info);
2948         if (ret)
2949                 return ret;
2950
2951         trans = btrfs_start_trans_remove_block_group(root->fs_info,
2952                                                      chunk_offset);
2953         if (IS_ERR(trans)) {
2954                 ret = PTR_ERR(trans);
2955                 btrfs_handle_fs_error(root->fs_info, ret, NULL);
2956                 return ret;
2957         }
2958
2959         /*
2960          * step two, delete the device extents and the
2961          * chunk tree entries
2962          */
2963         ret = btrfs_remove_chunk(trans, fs_info, chunk_offset);
2964         btrfs_end_transaction(trans);
2965         return ret;
2966 }
2967
2968 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info)
2969 {
2970         struct btrfs_root *chunk_root = fs_info->chunk_root;
2971         struct btrfs_path *path;
2972         struct extent_buffer *leaf;
2973         struct btrfs_chunk *chunk;
2974         struct btrfs_key key;
2975         struct btrfs_key found_key;
2976         u64 chunk_type;
2977         bool retried = false;
2978         int failed = 0;
2979         int ret;
2980
2981         path = btrfs_alloc_path();
2982         if (!path)
2983                 return -ENOMEM;
2984
2985 again:
2986         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2987         key.offset = (u64)-1;
2988         key.type = BTRFS_CHUNK_ITEM_KEY;
2989
2990         while (1) {
2991                 mutex_lock(&fs_info->delete_unused_bgs_mutex);
2992                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2993                 if (ret < 0) {
2994                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
2995                         goto error;
2996                 }
2997                 BUG_ON(ret == 0); /* Corruption */
2998
2999                 ret = btrfs_previous_item(chunk_root, path, key.objectid,
3000                                           key.type);
3001                 if (ret)
3002                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3003                 if (ret < 0)
3004                         goto error;
3005                 if (ret > 0)
3006                         break;
3007
3008                 leaf = path->nodes[0];
3009                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3010
3011                 chunk = btrfs_item_ptr(leaf, path->slots[0],
3012                                        struct btrfs_chunk);
3013                 chunk_type = btrfs_chunk_type(leaf, chunk);
3014                 btrfs_release_path(path);
3015
3016                 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
3017                         ret = btrfs_relocate_chunk(fs_info, found_key.offset);
3018                         if (ret == -ENOSPC)
3019                                 failed++;
3020                         else
3021                                 BUG_ON(ret);
3022                 }
3023                 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3024
3025                 if (found_key.offset == 0)
3026                         break;
3027                 key.offset = found_key.offset - 1;
3028         }
3029         ret = 0;
3030         if (failed && !retried) {
3031                 failed = 0;
3032                 retried = true;
3033                 goto again;
3034         } else if (WARN_ON(failed && retried)) {
3035                 ret = -ENOSPC;
3036         }
3037 error:
3038         btrfs_free_path(path);
3039         return ret;
3040 }
3041
3042 static int insert_balance_item(struct btrfs_fs_info *fs_info,
3043                                struct btrfs_balance_control *bctl)
3044 {
3045         struct btrfs_root *root = fs_info->tree_root;
3046         struct btrfs_trans_handle *trans;
3047         struct btrfs_balance_item *item;
3048         struct btrfs_disk_balance_args disk_bargs;
3049         struct btrfs_path *path;
3050         struct extent_buffer *leaf;
3051         struct btrfs_key key;
3052         int ret, err;
3053
3054         path = btrfs_alloc_path();
3055         if (!path)
3056                 return -ENOMEM;
3057
3058         trans = btrfs_start_transaction(root, 0);
3059         if (IS_ERR(trans)) {
3060                 btrfs_free_path(path);
3061                 return PTR_ERR(trans);
3062         }
3063
3064         key.objectid = BTRFS_BALANCE_OBJECTID;
3065         key.type = BTRFS_TEMPORARY_ITEM_KEY;
3066         key.offset = 0;
3067
3068         ret = btrfs_insert_empty_item(trans, root, path, &key,
3069                                       sizeof(*item));
3070         if (ret)
3071                 goto out;
3072
3073         leaf = path->nodes[0];
3074         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3075
3076         memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item));
3077
3078         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
3079         btrfs_set_balance_data(leaf, item, &disk_bargs);
3080         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
3081         btrfs_set_balance_meta(leaf, item, &disk_bargs);
3082         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
3083         btrfs_set_balance_sys(leaf, item, &disk_bargs);
3084
3085         btrfs_set_balance_flags(leaf, item, bctl->flags);
3086
3087         btrfs_mark_buffer_dirty(leaf);
3088 out:
3089         btrfs_free_path(path);
3090         err = btrfs_commit_transaction(trans);
3091         if (err && !ret)
3092                 ret = err;
3093         return ret;
3094 }
3095
3096 static int del_balance_item(struct btrfs_fs_info *fs_info)
3097 {
3098         struct btrfs_root *root = fs_info->tree_root;
3099         struct btrfs_trans_handle *trans;
3100         struct btrfs_path *path;
3101         struct btrfs_key key;
3102         int ret, err;
3103
3104         path = btrfs_alloc_path();
3105         if (!path)
3106                 return -ENOMEM;
3107
3108         trans = btrfs_start_transaction(root, 0);
3109         if (IS_ERR(trans)) {
3110                 btrfs_free_path(path);
3111                 return PTR_ERR(trans);
3112         }
3113
3114         key.objectid = BTRFS_BALANCE_OBJECTID;
3115         key.type = BTRFS_TEMPORARY_ITEM_KEY;
3116         key.offset = 0;
3117
3118         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3119         if (ret < 0)
3120                 goto out;
3121         if (ret > 0) {
3122                 ret = -ENOENT;
3123                 goto out;
3124         }
3125
3126         ret = btrfs_del_item(trans, root, path);
3127 out:
3128         btrfs_free_path(path);
3129         err = btrfs_commit_transaction(trans);
3130         if (err && !ret)
3131                 ret = err;
3132         return ret;
3133 }
3134
3135 /*
3136  * This is a heuristic used to reduce the number of chunks balanced on
3137  * resume after balance was interrupted.
3138  */
3139 static void update_balance_args(struct btrfs_balance_control *bctl)
3140 {
3141         /*
3142          * Turn on soft mode for chunk types that were being converted.
3143          */
3144         if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
3145                 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
3146         if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
3147                 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
3148         if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
3149                 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
3150
3151         /*
3152          * Turn on usage filter if is not already used.  The idea is
3153          * that chunks that we have already balanced should be
3154          * reasonably full.  Don't do it for chunks that are being
3155          * converted - that will keep us from relocating unconverted
3156          * (albeit full) chunks.
3157          */
3158         if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3159             !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3160             !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3161                 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
3162                 bctl->data.usage = 90;
3163         }
3164         if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3165             !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3166             !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3167                 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
3168                 bctl->sys.usage = 90;
3169         }
3170         if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3171             !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3172             !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3173                 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
3174                 bctl->meta.usage = 90;
3175         }
3176 }
3177
3178 /*
3179  * Should be called with both balance and volume mutexes held to
3180  * serialize other volume operations (add_dev/rm_dev/resize) with
3181  * restriper.  Same goes for unset_balance_control.
3182  */
3183 static void set_balance_control(struct btrfs_balance_control *bctl)
3184 {
3185         struct btrfs_fs_info *fs_info = bctl->fs_info;
3186
3187         BUG_ON(fs_info->balance_ctl);
3188
3189         spin_lock(&fs_info->balance_lock);
3190         fs_info->balance_ctl = bctl;
3191         spin_unlock(&fs_info->balance_lock);
3192 }
3193
3194 static void unset_balance_control(struct btrfs_fs_info *fs_info)
3195 {
3196         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3197
3198         BUG_ON(!fs_info->balance_ctl);
3199
3200         spin_lock(&fs_info->balance_lock);
3201         fs_info->balance_ctl = NULL;
3202         spin_unlock(&fs_info->balance_lock);
3203
3204         kfree(bctl);
3205 }
3206
3207 /*
3208  * Balance filters.  Return 1 if chunk should be filtered out
3209  * (should not be balanced).
3210  */
3211 static int chunk_profiles_filter(u64 chunk_type,
3212                                  struct btrfs_balance_args *bargs)
3213 {
3214         chunk_type = chunk_to_extended(chunk_type) &
3215                                 BTRFS_EXTENDED_PROFILE_MASK;
3216
3217         if (bargs->profiles & chunk_type)
3218                 return 0;
3219
3220         return 1;
3221 }
3222
3223 static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
3224                               struct btrfs_balance_args *bargs)
3225 {
3226         struct btrfs_block_group_cache *cache;
3227         u64 chunk_used;
3228         u64 user_thresh_min;
3229         u64 user_thresh_max;
3230         int ret = 1;
3231
3232         cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3233         chunk_used = btrfs_block_group_used(&cache->item);
3234
3235         if (bargs->usage_min == 0)
3236                 user_thresh_min = 0;
3237         else
3238                 user_thresh_min = div_factor_fine(cache->key.offset,
3239                                         bargs->usage_min);
3240
3241         if (bargs->usage_max == 0)
3242                 user_thresh_max = 1;
3243         else if (bargs->usage_max > 100)
3244                 user_thresh_max = cache->key.offset;
3245         else
3246                 user_thresh_max = div_factor_fine(cache->key.offset,
3247                                         bargs->usage_max);
3248
3249         if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max)
3250                 ret = 0;
3251
3252         btrfs_put_block_group(cache);
3253         return ret;
3254 }
3255
3256 static int chunk_usage_filter(struct btrfs_fs_info *fs_info,
3257                 u64 chunk_offset, struct btrfs_balance_args *bargs)
3258 {
3259         struct btrfs_block_group_cache *cache;
3260         u64 chunk_used, user_thresh;
3261         int ret = 1;
3262
3263         cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3264         chunk_used = btrfs_block_group_used(&cache->item);
3265
3266         if (bargs->usage_min == 0)
3267                 user_thresh = 1;
3268         else if (bargs->usage > 100)
3269                 user_thresh = cache->key.offset;
3270         else
3271                 user_thresh = div_factor_fine(cache->key.offset,
3272                                               bargs->usage);
3273
3274         if (chunk_used < user_thresh)
3275                 ret = 0;
3276
3277         btrfs_put_block_group(cache);
3278         return ret;
3279 }
3280
3281 static int chunk_devid_filter(struct extent_buffer *leaf,
3282                               struct btrfs_chunk *chunk,
3283                               struct btrfs_balance_args *bargs)
3284 {
3285         struct btrfs_stripe *stripe;
3286         int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3287         int i;
3288
3289         for (i = 0; i < num_stripes; i++) {
3290                 stripe = btrfs_stripe_nr(chunk, i);
3291                 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
3292                         return 0;
3293         }
3294
3295         return 1;
3296 }
3297
3298 /* [pstart, pend) */
3299 static int chunk_drange_filter(struct extent_buffer *leaf,
3300                                struct btrfs_chunk *chunk,
3301                                u64 chunk_offset,
3302                                struct btrfs_balance_args *bargs)
3303 {
3304         struct btrfs_stripe *stripe;
3305         int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3306         u64 stripe_offset;
3307         u64 stripe_length;
3308         int factor;
3309         int i;
3310
3311         if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
3312                 return 0;
3313
3314         if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP |
3315              BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) {
3316                 factor = num_stripes / 2;
3317         } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) {
3318                 factor = num_stripes - 1;
3319         } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) {
3320                 factor = num_stripes - 2;
3321         } else {
3322                 factor = num_stripes;
3323         }
3324
3325         for (i = 0; i < num_stripes; i++) {
3326                 stripe = btrfs_stripe_nr(chunk, i);
3327                 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
3328                         continue;
3329
3330                 stripe_offset = btrfs_stripe_offset(leaf, stripe);
3331                 stripe_length = btrfs_chunk_length(leaf, chunk);
3332                 stripe_length = div_u64(stripe_length, factor);
3333
3334                 if (stripe_offset < bargs->pend &&
3335                     stripe_offset + stripe_length > bargs->pstart)
3336                         return 0;
3337         }
3338
3339         return 1;
3340 }
3341
3342 /* [vstart, vend) */
3343 static int chunk_vrange_filter(struct extent_buffer *leaf,
3344                                struct btrfs_chunk *chunk,
3345                                u64 chunk_offset,
3346                                struct btrfs_balance_args *bargs)
3347 {
3348         if (chunk_offset < bargs->vend &&
3349             chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
3350                 /* at least part of the chunk is inside this vrange */
3351                 return 0;
3352
3353         return 1;
3354 }
3355
3356 static int chunk_stripes_range_filter(struct extent_buffer *leaf,
3357                                struct btrfs_chunk *chunk,
3358                                struct btrfs_balance_args *bargs)
3359 {
3360         int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3361
3362         if (bargs->stripes_min <= num_stripes
3363                         && num_stripes <= bargs->stripes_max)
3364                 return 0;
3365
3366         return 1;
3367 }
3368
3369 static int chunk_soft_convert_filter(u64 chunk_type,
3370                                      struct btrfs_balance_args *bargs)
3371 {
3372         if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
3373                 return 0;
3374
3375         chunk_type = chunk_to_extended(chunk_type) &
3376                                 BTRFS_EXTENDED_PROFILE_MASK;
3377
3378         if (bargs->target == chunk_type)
3379                 return 1;
3380
3381         return 0;
3382 }
3383
3384 static int should_balance_chunk(struct btrfs_fs_info *fs_info,
3385                                 struct extent_buffer *leaf,
3386                                 struct btrfs_chunk *chunk, u64 chunk_offset)
3387 {
3388         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3389         struct btrfs_balance_args *bargs = NULL;
3390         u64 chunk_type = btrfs_chunk_type(leaf, chunk);
3391
3392         /* type filter */
3393         if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
3394               (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
3395                 return 0;
3396         }
3397
3398         if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3399                 bargs = &bctl->data;
3400         else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3401                 bargs = &bctl->sys;
3402         else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3403                 bargs = &bctl->meta;
3404
3405         /* profiles filter */
3406         if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
3407             chunk_profiles_filter(chunk_type, bargs)) {
3408                 return 0;
3409         }
3410
3411         /* usage filter */
3412         if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
3413             chunk_usage_filter(fs_info, chunk_offset, bargs)) {
3414                 return 0;
3415         } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3416             chunk_usage_range_filter(fs_info, chunk_offset, bargs)) {
3417                 return 0;
3418         }
3419
3420         /* devid filter */
3421         if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
3422             chunk_devid_filter(leaf, chunk, bargs)) {
3423                 return 0;
3424         }
3425
3426         /* drange filter, makes sense only with devid filter */
3427         if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
3428             chunk_drange_filter(leaf, chunk, chunk_offset, bargs)) {
3429                 return 0;
3430         }
3431
3432         /* vrange filter */
3433         if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
3434             chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
3435                 return 0;
3436         }
3437
3438         /* stripes filter */
3439         if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) &&
3440             chunk_stripes_range_filter(leaf, chunk, bargs)) {
3441                 return 0;
3442         }
3443
3444         /* soft profile changing mode */
3445         if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
3446             chunk_soft_convert_filter(chunk_type, bargs)) {
3447                 return 0;
3448         }
3449
3450         /*
3451          * limited by count, must be the last filter
3452          */
3453         if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) {
3454                 if (bargs->limit == 0)
3455                         return 0;
3456                 else
3457                         bargs->limit--;
3458         } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) {
3459                 /*
3460                  * Same logic as the 'limit' filter; the minimum cannot be
3461                  * determined here because we do not have the global information
3462                  * about the count of all chunks that satisfy the filters.
3463                  */
3464                 if (bargs->limit_max == 0)
3465                         return 0;
3466                 else
3467                         bargs->limit_max--;
3468         }
3469
3470         return 1;
3471 }
3472
3473 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
3474 {
3475         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3476         struct btrfs_root *chunk_root = fs_info->chunk_root;
3477         struct btrfs_root *dev_root = fs_info->dev_root;
3478         struct list_head *devices;
3479         struct btrfs_device *device;
3480         u64 old_size;
3481         u64 size_to_free;
3482         u64 chunk_type;
3483         struct btrfs_chunk *chunk;
3484         struct btrfs_path *path = NULL;
3485         struct btrfs_key key;
3486         struct btrfs_key found_key;
3487         struct btrfs_trans_handle *trans;
3488         struct extent_buffer *leaf;
3489         int slot;
3490         int ret;
3491         int enospc_errors = 0;
3492         bool counting = true;
3493         /* The single value limit and min/max limits use the same bytes in the */
3494         u64 limit_data = bctl->data.limit;
3495         u64 limit_meta = bctl->meta.limit;
3496         u64 limit_sys = bctl->sys.limit;
3497         u32 count_data = 0;
3498         u32 count_meta = 0;
3499         u32 count_sys = 0;
3500         int chunk_reserved = 0;
3501         u64 bytes_used = 0;
3502
3503         /* step one make some room on all the devices */
3504         devices = &fs_info->fs_devices->devices;
3505         list_for_each_entry(device, devices, dev_list) {
3506                 old_size = btrfs_device_get_total_bytes(device);
3507                 size_to_free = div_factor(old_size, 1);
3508                 size_to_free = min_t(u64, size_to_free, SZ_1M);
3509                 if (!device->writeable ||
3510                     btrfs_device_get_total_bytes(device) -
3511                     btrfs_device_get_bytes_used(device) > size_to_free ||
3512                     device->is_tgtdev_for_dev_replace)
3513                         continue;
3514
3515                 ret = btrfs_shrink_device(device, old_size - size_to_free);
3516                 if (ret == -ENOSPC)
3517                         break;
3518                 if (ret) {
3519                         /* btrfs_shrink_device never returns ret > 0 */
3520                         WARN_ON(ret > 0);
3521                         goto error;
3522                 }
3523
3524                 trans = btrfs_start_transaction(dev_root, 0);
3525                 if (IS_ERR(trans)) {
3526                         ret = PTR_ERR(trans);
3527                         btrfs_info_in_rcu(fs_info,
3528                  "resize: unable to start transaction after shrinking device %s (error %d), old size %llu, new size %llu",
3529                                           rcu_str_deref(device->name), ret,
3530                                           old_size, old_size - size_to_free);
3531                         goto error;
3532                 }
3533
3534                 ret = btrfs_grow_device(trans, device, old_size);
3535                 if (ret) {
3536                         btrfs_end_transaction(trans);
3537                         /* btrfs_grow_device never returns ret > 0 */
3538                         WARN_ON(ret > 0);
3539                         btrfs_info_in_rcu(fs_info,
3540                  "resize: unable to grow device after shrinking device %s (error %d), old size %llu, new size %llu",
3541                                           rcu_str_deref(device->name), ret,
3542                                           old_size, old_size - size_to_free);
3543                         goto error;
3544                 }
3545
3546                 btrfs_end_transaction(trans);
3547         }
3548
3549         /* step two, relocate all the chunks */
3550         path = btrfs_alloc_path();
3551         if (!path) {
3552                 ret = -ENOMEM;
3553                 goto error;
3554         }
3555
3556         /* zero out stat counters */
3557         spin_lock(&fs_info->balance_lock);
3558         memset(&bctl->stat, 0, sizeof(bctl->stat));
3559         spin_unlock(&fs_info->balance_lock);
3560 again:
3561         if (!counting) {
3562                 /*
3563                  * The single value limit and min/max limits use the same bytes
3564                  * in the
3565                  */
3566                 bctl->data.limit = limit_data;
3567                 bctl->meta.limit = limit_meta;
3568                 bctl->sys.limit = limit_sys;
3569         }
3570         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3571         key.offset = (u64)-1;
3572         key.type = BTRFS_CHUNK_ITEM_KEY;
3573
3574         while (1) {
3575                 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
3576                     atomic_read(&fs_info->balance_cancel_req)) {
3577                         ret = -ECANCELED;
3578                         goto error;
3579                 }
3580
3581                 mutex_lock(&fs_info->delete_unused_bgs_mutex);
3582                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3583                 if (ret < 0) {
3584                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3585                         goto error;
3586                 }
3587
3588                 /*
3589                  * this shouldn't happen, it means the last relocate
3590                  * failed
3591                  */
3592                 if (ret == 0)
3593                         BUG(); /* FIXME break ? */
3594
3595                 ret = btrfs_previous_item(chunk_root, path, 0,
3596                                           BTRFS_CHUNK_ITEM_KEY);
3597                 if (ret) {
3598                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3599                         ret = 0;
3600                         break;
3601                 }
3602
3603                 leaf = path->nodes[0];
3604                 slot = path->slots[0];
3605                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3606
3607                 if (found_key.objectid != key.objectid) {
3608                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3609                         break;
3610                 }
3611
3612                 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3613                 chunk_type = btrfs_chunk_type(leaf, chunk);
3614
3615                 if (!counting) {
3616                         spin_lock(&fs_info->balance_lock);
3617                         bctl->stat.considered++;
3618                         spin_unlock(&fs_info->balance_lock);
3619                 }
3620
3621                 ret = should_balance_chunk(fs_info, leaf, chunk,
3622                                            found_key.offset);
3623
3624                 btrfs_release_path(path);
3625                 if (!ret) {
3626                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3627                         goto loop;
3628                 }
3629
3630                 if (counting) {
3631                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3632                         spin_lock(&fs_info->balance_lock);
3633                         bctl->stat.expected++;
3634                         spin_unlock(&fs_info->balance_lock);
3635
3636                         if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3637                                 count_data++;
3638                         else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3639                                 count_sys++;
3640                         else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3641                                 count_meta++;
3642
3643                         goto loop;
3644                 }
3645
3646                 /*
3647                  * Apply limit_min filter, no need to check if the LIMITS
3648                  * filter is used, limit_min is 0 by default
3649                  */
3650                 if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
3651                                         count_data < bctl->data.limit_min)
3652                                 || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) &&
3653                                         count_meta < bctl->meta.limit_min)
3654                                 || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) &&
3655                                         count_sys < bctl->sys.limit_min)) {
3656                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3657                         goto loop;
3658                 }
3659
3660                 ASSERT(fs_info->data_sinfo);
3661                 spin_lock(&fs_info->data_sinfo->lock);
3662                 bytes_used = fs_info->data_sinfo->bytes_used;
3663                 spin_unlock(&fs_info->data_sinfo->lock);
3664
3665                 if ((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
3666                     !chunk_reserved && !bytes_used) {
3667                         trans = btrfs_start_transaction(chunk_root, 0);
3668                         if (IS_ERR(trans)) {
3669                                 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3670                                 ret = PTR_ERR(trans);
3671                                 goto error;
3672                         }
3673
3674                         ret = btrfs_force_chunk_alloc(trans, fs_info,
3675                                                       BTRFS_BLOCK_GROUP_DATA);
3676                         btrfs_end_transaction(trans);
3677                         if (ret < 0) {
3678                                 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3679                                 goto error;
3680                         }
3681                         chunk_reserved = 1;
3682                 }
3683
3684                 ret = btrfs_relocate_chunk(fs_info, found_key.offset);
3685                 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3686                 if (ret && ret != -ENOSPC)
3687                         goto error;
3688                 if (ret == -ENOSPC) {
3689                         enospc_errors++;
3690                 } else {
3691                         spin_lock(&fs_info->balance_lock);
3692                         bctl->stat.completed++;
3693                         spin_unlock(&fs_info->balance_lock);
3694                 }
3695 loop:
3696                 if (found_key.offset == 0)
3697                         break;
3698                 key.offset = found_key.offset - 1;
3699         }
3700
3701         if (counting) {
3702                 btrfs_release_path(path);
3703                 counting = false;
3704                 goto again;
3705         }
3706 error:
3707         btrfs_free_path(path);
3708         if (enospc_errors) {
3709                 btrfs_info(fs_info, "%d enospc errors during balance",
3710                            enospc_errors);
3711                 if (!ret)
3712                         ret = -ENOSPC;
3713         }
3714
3715         return ret;
3716 }
3717
3718 /**
3719  * alloc_profile_is_valid - see if a given profile is valid and reduced
3720  * @flags: profile to validate
3721  * @extended: if true @flags is treated as an extended profile
3722  */
3723 static int alloc_profile_is_valid(u64 flags, int extended)
3724 {
3725         u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
3726                                BTRFS_BLOCK_GROUP_PROFILE_MASK);
3727
3728         flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
3729
3730         /* 1) check that all other bits are zeroed */
3731         if (flags & ~mask)
3732                 return 0;
3733
3734         /* 2) see if profile is reduced */
3735         if (flags == 0)
3736                 return !extended; /* "0" is valid for usual profiles */
3737
3738         /* true if exactly one bit set */
3739         return (flags & (flags - 1)) == 0;
3740 }
3741
3742 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
3743 {
3744         /* cancel requested || normal exit path */
3745         return atomic_read(&fs_info->balance_cancel_req) ||
3746                 (atomic_read(&fs_info->balance_pause_req) == 0 &&
3747                  atomic_read(&fs_info->balance_cancel_req) == 0);
3748 }
3749
3750 static void __cancel_balance(struct btrfs_fs_info *fs_info)
3751 {
3752         int ret;
3753
3754         unset_balance_control(fs_info);
3755         ret = del_balance_item(fs_info);
3756         if (ret)
3757                 btrfs_handle_fs_error(fs_info, ret, NULL);
3758
3759         clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
3760 }
3761
3762 /* Non-zero return value signifies invalidity */
3763 static inline int validate_convert_profile(struct btrfs_balance_args *bctl_arg,
3764                 u64 allowed)
3765 {
3766         return ((bctl_arg->flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3767                 (!alloc_profile_is_valid(bctl_arg->target, 1) ||
3768                  (bctl_arg->target & ~allowed)));
3769 }
3770
3771 /*
3772  * Should be called with both balance and volume mutexes held
3773  */
3774 int btrfs_balance(struct btrfs_balance_control *bctl,
3775                   struct btrfs_ioctl_balance_args *bargs)
3776 {
3777         struct btrfs_fs_info *fs_info = bctl->fs_info;
3778         u64 meta_target, data_target;
3779         u64 allowed;
3780         int mixed = 0;
3781         int ret;
3782         u64 num_devices;
3783         unsigned seq;
3784
3785         if (btrfs_fs_closing(fs_info) ||
3786             atomic_read(&fs_info->balance_pause_req) ||
3787             atomic_read(&fs_info->balance_cancel_req)) {
3788                 ret = -EINVAL;
3789                 goto out;
3790         }
3791
3792         allowed = btrfs_super_incompat_flags(fs_info->super_copy);
3793         if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
3794                 mixed = 1;
3795
3796         /*
3797          * In case of mixed groups both data and meta should be picked,
3798          * and identical options should be given for both of them.
3799          */
3800         allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
3801         if (mixed && (bctl->flags & allowed)) {
3802                 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
3803                     !(bctl->flags & BTRFS_BALANCE_METADATA) ||
3804                     memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
3805                         btrfs_err(fs_info,
3806                                   "with mixed groups data and metadata balance options must be the same");
3807                         ret = -EINVAL;
3808                         goto out;
3809                 }
3810         }
3811
3812         num_devices = fs_info->fs_devices->num_devices;
3813         btrfs_dev_replace_lock(&fs_info->dev_replace, 0);
3814         if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
3815                 BUG_ON(num_devices < 1);
3816                 num_devices--;
3817         }
3818         btrfs_dev_replace_unlock(&fs_info->dev_replace, 0);
3819         allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE | BTRFS_BLOCK_GROUP_DUP;
3820         if (num_devices > 1)
3821                 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1);
3822         if (num_devices > 2)
3823                 allowed |= BTRFS_BLOCK_GROUP_RAID5;
3824         if (num_devices > 3)
3825                 allowed |= (BTRFS_BLOCK_GROUP_RAID10 |
3826                             BTRFS_BLOCK_GROUP_RAID6);
3827         if (validate_convert_profile(&bctl->data, allowed)) {
3828                 btrfs_err(fs_info,
3829                           "unable to start balance with target data profile %llu",
3830                           bctl->data.target);
3831                 ret = -EINVAL;
3832                 goto out;
3833         }
3834         if (validate_convert_profile(&bctl->meta, allowed)) {
3835                 btrfs_err(fs_info,
3836                           "unable to start balance with target metadata profile %llu",
3837                           bctl->meta.target);
3838                 ret = -EINVAL;
3839                 goto out;
3840         }
3841         if (validate_convert_profile(&bctl->sys, allowed)) {
3842                 btrfs_err(fs_info,
3843                           "unable to start balance with target system profile %llu",
3844                           bctl->sys.target);
3845                 ret = -EINVAL;
3846                 goto out;
3847         }
3848
3849         /* allow to reduce meta or sys integrity only if force set */
3850         allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3851                         BTRFS_BLOCK_GROUP_RAID10 |
3852                         BTRFS_BLOCK_GROUP_RAID5 |
3853                         BTRFS_BLOCK_GROUP_RAID6;
3854         do {
3855                 seq = read_seqbegin(&fs_info->profiles_lock);
3856
3857                 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3858                      (fs_info->avail_system_alloc_bits & allowed) &&
3859                      !(bctl->sys.target & allowed)) ||
3860                     ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3861                      (fs_info->avail_metadata_alloc_bits & allowed) &&
3862                      !(bctl->meta.target & allowed))) {
3863                         if (bctl->flags & BTRFS_BALANCE_FORCE) {
3864                                 btrfs_info(fs_info,
3865                                            "force reducing metadata integrity");
3866                         } else {
3867                                 btrfs_err(fs_info,
3868                                           "balance will reduce metadata integrity, use force if you want this");
3869                                 ret = -EINVAL;
3870                                 goto out;
3871                         }
3872                 }
3873         } while (read_seqretry(&fs_info->profiles_lock, seq));
3874
3875         /* if we're not converting, the target field is uninitialized */
3876         meta_target = (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
3877                 bctl->meta.target : fs_info->avail_metadata_alloc_bits;
3878         data_target = (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
3879                 bctl->data.target : fs_info->avail_data_alloc_bits;
3880         if (btrfs_get_num_tolerated_disk_barrier_failures(meta_target) <
3881                 btrfs_get_num_tolerated_disk_barrier_failures(data_target)) {
3882                 btrfs_warn(fs_info,
3883                            "metadata profile 0x%llx has lower redundancy than data profile 0x%llx",
3884                            meta_target, data_target);
3885         }
3886
3887         if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3888                 fs_info->num_tolerated_disk_barrier_failures = min(
3889                         btrfs_calc_num_tolerated_disk_barrier_failures(fs_info),
3890                         btrfs_get_num_tolerated_disk_barrier_failures(
3891                                 bctl->sys.target));
3892         }
3893
3894         ret = insert_balance_item(fs_info, bctl);
3895         if (ret && ret != -EEXIST)
3896                 goto out;
3897
3898         if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
3899                 BUG_ON(ret == -EEXIST);
3900                 set_balance_control(bctl);
3901         } else {
3902                 BUG_ON(ret != -EEXIST);
3903                 spin_lock(&fs_info->balance_lock);
3904                 update_balance_args(bctl);
3905                 spin_unlock(&fs_info->balance_lock);
3906         }
3907
3908         atomic_inc(&fs_info->balance_running);
3909         mutex_unlock(&fs_info->balance_mutex);
3910
3911         ret = __btrfs_balance(fs_info);
3912
3913         mutex_lock(&fs_info->balance_mutex);
3914         atomic_dec(&fs_info->balance_running);
3915
3916         if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3917                 fs_info->num_tolerated_disk_barrier_failures =
3918                         btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3919         }
3920
3921         if (bargs) {
3922                 memset(bargs, 0, sizeof(*bargs));
3923                 update_ioctl_balance_args(fs_info, 0, bargs);
3924         }
3925
3926         if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
3927             balance_need_close(fs_info)) {
3928                 __cancel_balance(fs_info);
3929         }
3930
3931         wake_up(&fs_info->balance_wait_q);
3932
3933         return ret;
3934 out:
3935         if (bctl->flags & BTRFS_BALANCE_RESUME)
3936                 __cancel_balance(fs_info);
3937         else {
3938                 kfree(bctl);
3939                 clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
3940         }
3941         return ret;
3942 }
3943
3944 static int balance_kthread(void *data)
3945 {
3946         struct btrfs_fs_info *fs_info = data;
3947         int ret = 0;
3948
3949         mutex_lock(&fs_info->volume_mutex);
3950         mutex_lock(&fs_info->balance_mutex);
3951
3952         if (fs_info->balance_ctl) {
3953                 btrfs_info(fs_info, "continuing balance");
3954                 ret = btrfs_balance(fs_info->balance_ctl, NULL);
3955         }
3956
3957         mutex_unlock(&fs_info->balance_mutex);
3958         mutex_unlock(&fs_info->volume_mutex);
3959
3960         return ret;
3961 }
3962
3963 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
3964 {
3965         struct task_struct *tsk;
3966
3967         spin_lock(&fs_info->balance_lock);
3968         if (!fs_info->balance_ctl) {
3969                 spin_unlock(&fs_info->balance_lock);
3970                 return 0;
3971         }
3972         spin_unlock(&fs_info->balance_lock);
3973
3974         if (btrfs_test_opt(fs_info, SKIP_BALANCE)) {
3975                 btrfs_info(fs_info, "force skipping balance");
3976                 return 0;
3977         }
3978
3979         tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
3980         return PTR_ERR_OR_ZERO(tsk);
3981 }
3982
3983 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
3984 {
3985         struct btrfs_balance_control *bctl;
3986         struct btrfs_balance_item *item;
3987         struct btrfs_disk_balance_args disk_bargs;
3988         struct btrfs_path *path;
3989         struct extent_buffer *leaf;
3990         struct btrfs_key key;
3991         int ret;
3992
3993         path = btrfs_alloc_path();
3994         if (!path)
3995                 return -ENOMEM;
3996
3997         key.objectid = BTRFS_BALANCE_OBJECTID;
3998         key.type = BTRFS_TEMPORARY_ITEM_KEY;
3999         key.offset = 0;
4000
4001         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
4002         if (ret < 0)
4003                 goto out;
4004         if (ret > 0) { /* ret = -ENOENT; */
4005                 ret = 0;
4006                 goto out;
4007         }
4008
4009         bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
4010         if (!bctl) {
4011                 ret = -ENOMEM;
4012                 goto out;
4013         }
4014
4015         leaf = path->nodes[0];
4016         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
4017
4018         bctl->fs_info = fs_info;
4019         bctl->flags = btrfs_balance_flags(leaf, item);
4020         bctl->flags |= BTRFS_BALANCE_RESUME;
4021
4022         btrfs_balance_data(leaf, item, &disk_bargs);
4023         btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
4024         btrfs_balance_meta(leaf, item, &disk_bargs);
4025         btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
4026         btrfs_balance_sys(leaf, item, &disk_bargs);
4027         btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
4028
4029         WARN_ON(test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags));
4030
4031         mutex_lock(&fs_info->volume_mutex);
4032         mutex_lock(&fs_info->balance_mutex);
4033
4034         set_balance_control(bctl);
4035
4036         mutex_unlock(&fs_info->balance_mutex);
4037         mutex_unlock(&fs_info->volume_mutex);
4038 out:
4039         btrfs_free_path(path);
4040         return ret;
4041 }
4042
4043 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
4044 {
4045         int ret = 0;
4046
4047         mutex_lock(&fs_info->balance_mutex);
4048         if (!fs_info->balance_ctl) {
4049                 mutex_unlock(&fs_info->balance_mutex);
4050                 return -ENOTCONN;
4051         }
4052
4053         if (atomic_read(&fs_info->balance_running)) {
4054                 atomic_inc(&fs_info->balance_pause_req);
4055                 mutex_unlock(&fs_info->balance_mutex);
4056
4057                 wait_event(fs_info->balance_wait_q,
4058                            atomic_read(&fs_info->balance_running) == 0);
4059
4060                 mutex_lock(&fs_info->balance_mutex);
4061                 /* we are good with balance_ctl ripped off from under us */
4062                 BUG_ON(atomic_read(&fs_info->balance_running));
4063                 atomic_dec(&fs_info->balance_pause_req);
4064         } else {
4065                 ret = -ENOTCONN;
4066         }
4067
4068         mutex_unlock(&fs_info->balance_mutex);
4069         return ret;
4070 }
4071
4072 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
4073 {
4074         if (fs_info->sb->s_flags & MS_RDONLY)
4075                 return -EROFS;
4076
4077         mutex_lock(&fs_info->balance_mutex);
4078         if (!fs_info->balance_ctl) {
4079                 mutex_unlock(&fs_info->balance_mutex);
4080                 return -ENOTCONN;
4081         }
4082
4083         atomic_inc(&fs_info->balance_cancel_req);
4084         /*
4085          * if we are running just wait and return, balance item is
4086          * deleted in btrfs_balance in this case
4087          */
4088         if (atomic_read(&fs_info->balance_running)) {
4089                 mutex_unlock(&fs_info->balance_mutex);
4090                 wait_event(fs_info->balance_wait_q,
4091                            atomic_read(&fs_info->balance_running) == 0);
4092                 mutex_lock(&fs_info->balance_mutex);
4093         } else {
4094                 /* __cancel_balance needs volume_mutex */
4095                 mutex_unlock(&fs_info->balance_mutex);
4096                 mutex_lock(&fs_info->volume_mutex);
4097                 mutex_lock(&fs_info->balance_mutex);
4098
4099                 if (fs_info->balance_ctl)
4100                         __cancel_balance(fs_info);
4101
4102                 mutex_unlock(&fs_info->volume_mutex);
4103         }
4104
4105         BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running));
4106         atomic_dec(&fs_info->balance_cancel_req);
4107         mutex_unlock(&fs_info->balance_mutex);
4108         return 0;
4109 }
4110
4111 static int btrfs_uuid_scan_kthread(void *data)
4112 {
4113         struct btrfs_fs_info *fs_info = data;
4114         struct btrfs_root *root = fs_info->tree_root;
4115         struct btrfs_key key;
4116         struct btrfs_key max_key;
4117         struct btrfs_path *path = NULL;
4118         int ret = 0;
4119         struct extent_buffer *eb;
4120         int slot;
4121         struct btrfs_root_item root_item;
4122         u32 item_size;
4123         struct btrfs_trans_handle *trans = NULL;
4124
4125         path = btrfs_alloc_path();
4126         if (!path) {
4127                 ret = -ENOMEM;
4128                 goto out;
4129         }
4130
4131         key.objectid = 0;
4132         key.type = BTRFS_ROOT_ITEM_KEY;
4133         key.offset = 0;
4134
4135         max_key.objectid = (u64)-1;
4136         max_key.type = BTRFS_ROOT_ITEM_KEY;
4137         max_key.offset = (u64)-1;
4138
4139         while (1) {
4140                 ret = btrfs_search_forward(root, &key, path, 0);
4141                 if (ret) {
4142                         if (ret > 0)
4143                                 ret = 0;
4144                         break;
4145                 }
4146
4147                 if (key.type != BTRFS_ROOT_ITEM_KEY ||
4148                     (key.objectid < BTRFS_FIRST_FREE_OBJECTID &&
4149                      key.objectid != BTRFS_FS_TREE_OBJECTID) ||
4150                     key.objectid > BTRFS_LAST_FREE_OBJECTID)
4151                         goto skip;
4152
4153                 eb = path->nodes[0];
4154                 slot = path->slots[0];
4155                 item_size = btrfs_item_size_nr(eb, slot);
4156                 if (item_size < sizeof(root_item))
4157                         goto skip;
4158
4159                 read_extent_buffer(eb, &root_item,
4160                                    btrfs_item_ptr_offset(eb, slot),
4161                                    (int)sizeof(root_item));
4162                 if (btrfs_root_refs(&root_item) == 0)
4163                         goto skip;
4164
4165                 if (!btrfs_is_empty_uuid(root_item.uuid) ||
4166                     !btrfs_is_empty_uuid(root_item.received_uuid)) {
4167                         if (trans)
4168                                 goto update_tree;
4169
4170                         btrfs_release_path(path);
4171                         /*
4172                          * 1 - subvol uuid item
4173                          * 1 - received_subvol uuid item
4174                          */
4175                         trans = btrfs_start_transaction(fs_info->uuid_root, 2);
4176                         if (IS_ERR(trans)) {
4177                                 ret = PTR_ERR(trans);
4178                                 break;
4179                         }
4180                         continue;
4181                 } else {
4182                         goto skip;
4183                 }
4184 update_tree:
4185                 if (!btrfs_is_empty_uuid(root_item.uuid)) {
4186                         ret = btrfs_uuid_tree_add(trans, fs_info,
4187                                                   root_item.uuid,
4188                                                   BTRFS_UUID_KEY_SUBVOL,
4189                                                   key.objectid);
4190                         if (ret < 0) {
4191                                 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4192                                         ret);
4193                                 break;
4194                         }
4195                 }
4196
4197                 if (!btrfs_is_empty_uuid(root_item.received_uuid)) {
4198                         ret = btrfs_uuid_tree_add(trans, fs_info,
4199                                                   root_item.received_uuid,
4200                                                  BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4201                                                   key.objectid);
4202                         if (ret < 0) {
4203                                 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4204                                         ret);
4205                                 break;
4206                         }
4207                 }
4208
4209 skip:
4210                 if (trans) {
4211                         ret = btrfs_end_transaction(trans);
4212                         trans = NULL;
4213                         if (ret)
4214                                 break;
4215                 }
4216
4217                 btrfs_release_path(path);
4218                 if (key.offset < (u64)-1) {
4219                         key.offset++;
4220                 } else if (key.type < BTRFS_ROOT_ITEM_KEY) {
4221                         key.offset = 0;
4222                         key.type = BTRFS_ROOT_ITEM_KEY;
4223                 } else if (key.objectid < (u64)-1) {
4224                         key.offset = 0;
4225                         key.type = BTRFS_ROOT_ITEM_KEY;
4226                         key.objectid++;
4227                 } else {
4228                         break;
4229                 }
4230                 cond_resched();
4231         }
4232
4233 out:
4234         btrfs_free_path(path);
4235         if (trans && !IS_ERR(trans))
4236                 btrfs_end_transaction(trans);
4237         if (ret)
4238                 btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret);
4239         else
4240                 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
4241         up(&fs_info->uuid_tree_rescan_sem);
4242         return 0;
4243 }
4244
4245 /*
4246  * Callback for btrfs_uuid_tree_iterate().
4247  * returns:
4248  * 0    check succeeded, the entry is not outdated.
4249  * < 0  if an error occurred.
4250  * > 0  if the check failed, which means the caller shall remove the entry.
4251  */
4252 static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info *fs_info,
4253                                        u8 *uuid, u8 type, u64 subid)
4254 {
4255         struct btrfs_key key;
4256         int ret = 0;
4257         struct btrfs_root *subvol_root;
4258
4259         if (type != BTRFS_UUID_KEY_SUBVOL &&
4260             type != BTRFS_UUID_KEY_RECEIVED_SUBVOL)
4261                 goto out;
4262
4263         key.objectid = subid;
4264         key.type = BTRFS_ROOT_ITEM_KEY;
4265         key.offset = (u64)-1;
4266         subvol_root = btrfs_read_fs_root_no_name(fs_info, &key);
4267         if (IS_ERR(subvol_root)) {
4268                 ret = PTR_ERR(subvol_root);
4269                 if (ret == -ENOENT)
4270                         ret = 1;
4271                 goto out;
4272         }
4273
4274         switch (type) {
4275         case BTRFS_UUID_KEY_SUBVOL:
4276                 if (memcmp(uuid, subvol_root->root_item.uuid, BTRFS_UUID_SIZE))
4277                         ret = 1;
4278                 break;
4279         case BTRFS_UUID_KEY_RECEIVED_SUBVOL:
4280                 if (memcmp(uuid, subvol_root->root_item.received_uuid,
4281                            BTRFS_UUID_SIZE))
4282                         ret = 1;
4283                 break;
4284         }
4285
4286 out:
4287         return ret;
4288 }
4289
4290 static int btrfs_uuid_rescan_kthread(void *data)
4291 {
4292         struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data;
4293         int ret;
4294
4295         /*
4296          * 1st step is to iterate through the existing UUID tree and
4297          * to delete all entries that contain outdated data.
4298          * 2nd step is to add all missing entries to the UUID tree.
4299          */
4300         ret = btrfs_uuid_tree_iterate(fs_info, btrfs_check_uuid_tree_entry);
4301         if (ret < 0) {
4302                 btrfs_warn(fs_info, "iterating uuid_tree failed %d", ret);
4303                 up(&fs_info->uuid_tree_rescan_sem);
4304                 return ret;
4305         }
4306         return btrfs_uuid_scan_kthread(data);
4307 }
4308
4309 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info)
4310 {
4311         struct btrfs_trans_handle *trans;
4312         struct btrfs_root *tree_root = fs_info->tree_root;
4313         struct btrfs_root *uuid_root;
4314         struct task_struct *task;
4315         int ret;
4316
4317         /*
4318          * 1 - root node
4319          * 1 - root item
4320          */
4321         trans = btrfs_start_transaction(tree_root, 2);
4322         if (IS_ERR(trans))
4323                 return PTR_ERR(trans);
4324
4325         uuid_root = btrfs_create_tree(trans, fs_info,
4326                                       BTRFS_UUID_TREE_OBJECTID);
4327         if (IS_ERR(uuid_root)) {
4328                 ret = PTR_ERR(uuid_root);
4329                 btrfs_abort_transaction(trans, ret);
4330                 btrfs_end_transaction(trans);
4331                 return ret;
4332         }
4333
4334         fs_info->uuid_root = uuid_root;
4335
4336         ret = btrfs_commit_transaction(trans);
4337         if (ret)
4338                 return ret;
4339
4340         down(&fs_info->uuid_tree_rescan_sem);
4341         task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid");
4342         if (IS_ERR(task)) {
4343                 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4344                 btrfs_warn(fs_info, "failed to start uuid_scan task");
4345                 up(&fs_info->uuid_tree_rescan_sem);
4346                 return PTR_ERR(task);
4347         }
4348
4349         return 0;
4350 }
4351
4352 int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
4353 {
4354         struct task_struct *task;
4355
4356         down(&fs_info->uuid_tree_rescan_sem);
4357         task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
4358         if (IS_ERR(task)) {
4359                 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4360                 btrfs_warn(fs_info, "failed to start uuid_rescan task");
4361                 up(&fs_info->uuid_tree_rescan_sem);
4362                 return PTR_ERR(task);
4363         }
4364
4365         return 0;
4366 }
4367
4368 /*
4369  * shrinking a device means finding all of the device extents past
4370  * the new size, and then following the back refs to the chunks.
4371  * The chunk relocation code actually frees the device extent
4372  */
4373 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
4374 {
4375         struct btrfs_fs_info *fs_info = device->fs_info;
4376         struct btrfs_root *root = fs_info->dev_root;
4377         struct btrfs_trans_handle *trans;
4378         struct btrfs_dev_extent *dev_extent = NULL;
4379         struct btrfs_path *path;
4380         u64 length;
4381         u64 chunk_offset;
4382         int ret;
4383         int slot;
4384         int failed = 0;
4385         bool retried = false;
4386         bool checked_pending_chunks = false;
4387         struct extent_buffer *l;
4388         struct btrfs_key key;
4389         struct btrfs_super_block *super_copy = fs_info->super_copy;
4390         u64 old_total = btrfs_super_total_bytes(super_copy);
4391         u64 old_size = btrfs_device_get_total_bytes(device);
4392         u64 diff = old_size - new_size;
4393
4394         if (device->is_tgtdev_for_dev_replace)
4395                 return -EINVAL;
4396
4397         path = btrfs_alloc_path();
4398         if (!path)
4399                 return -ENOMEM;
4400
4401         path->reada = READA_FORWARD;
4402
4403         mutex_lock(&fs_info->chunk_mutex);
4404
4405         btrfs_device_set_total_bytes(device, new_size);
4406         if (device->writeable) {
4407                 device->fs_devices->total_rw_bytes -= diff;
4408                 atomic64_sub(diff, &fs_info->free_chunk_space);
4409         }
4410         mutex_unlock(&fs_info->chunk_mutex);
4411
4412 again:
4413         key.objectid = device->devid;
4414         key.offset = (u64)-1;
4415         key.type = BTRFS_DEV_EXTENT_KEY;
4416
4417         do {
4418                 mutex_lock(&fs_info->delete_unused_bgs_mutex);
4419                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4420                 if (ret < 0) {
4421                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4422                         goto done;
4423                 }
4424
4425                 ret = btrfs_previous_item(root, path, 0, key.type);
4426                 if (ret)
4427                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4428                 if (ret < 0)
4429                         goto done;
4430                 if (ret) {
4431                         ret = 0;
4432                         btrfs_release_path(path);
4433                         break;
4434                 }
4435
4436                 l = path->nodes[0];
4437                 slot = path->slots[0];
4438                 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
4439
4440                 if (key.objectid != device->devid) {
4441                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4442                         btrfs_release_path(path);
4443                         break;
4444                 }
4445
4446                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
4447                 length = btrfs_dev_extent_length(l, dev_extent);
4448
4449                 if (key.offset + length <= new_size) {
4450                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4451                         btrfs_release_path(path);
4452                         break;
4453                 }
4454
4455                 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
4456                 btrfs_release_path(path);
4457
4458                 ret = btrfs_relocate_chunk(fs_info, chunk_offset);
4459                 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4460                 if (ret && ret != -ENOSPC)
4461                         goto done;
4462                 if (ret == -ENOSPC)
4463                         failed++;
4464         } while (key.offset-- > 0);
4465
4466         if (failed && !retried) {
4467                 failed = 0;
4468                 retried = true;
4469                 goto again;
4470         } else if (failed && retried) {
4471                 ret = -ENOSPC;
4472                 goto done;
4473         }
4474
4475         /* Shrinking succeeded, else we would be at "done". */
4476         trans = btrfs_start_transaction(root, 0);
4477         if (IS_ERR(trans)) {
4478                 ret = PTR_ERR(trans);
4479                 goto done;
4480         }
4481
4482         mutex_lock(&fs_info->chunk_mutex);
4483
4484         /*
4485          * We checked in the above loop all device extents that were already in
4486          * the device tree. However before we have updated the device's
4487          * total_bytes to the new size, we might have had chunk allocations that
4488          * have not complete yet (new block groups attached to transaction
4489          * handles), and therefore their device extents were not yet in the
4490          * device tree and we missed them in the loop above. So if we have any
4491          * pending chunk using a device extent that overlaps the device range
4492          * that we can not use anymore, commit the current transaction and
4493          * repeat the search on the device tree - this way we guarantee we will
4494          * not have chunks using device extents that end beyond 'new_size'.
4495          */
4496         if (!checked_pending_chunks) {
4497                 u64 start = new_size;
4498                 u64 len = old_size - new_size;
4499
4500                 if (contains_pending_extent(trans->transaction, device,
4501                                             &start, len)) {
4502                         mutex_unlock(&fs_info->chunk_mutex);
4503                         checked_pending_chunks = true;
4504                         failed = 0;
4505                         retried = false;
4506                         ret = btrfs_commit_transaction(trans);
4507                         if (ret)
4508                                 goto done;
4509                         goto again;
4510                 }
4511         }
4512
4513         btrfs_device_set_disk_total_bytes(device, new_size);
4514         if (list_empty(&device->resized_list))
4515                 list_add_tail(&device->resized_list,
4516                               &fs_info->fs_devices->resized_devices);
4517
4518         WARN_ON(diff > old_total);
4519         btrfs_set_super_total_bytes(super_copy, old_total - diff);
4520         mutex_unlock(&fs_info->chunk_mutex);
4521
4522         /* Now btrfs_update_device() will change the on-disk size. */
4523         ret = btrfs_update_device(trans, device);
4524         btrfs_end_transaction(trans);
4525 done:
4526         btrfs_free_path(path);
4527         if (ret) {
4528                 mutex_lock(&fs_info->chunk_mutex);
4529                 btrfs_device_set_total_bytes(device, old_size);
4530                 if (device->writeable)
4531                         device->fs_devices->total_rw_bytes += diff;
4532                 atomic64_add(diff, &fs_info->free_chunk_space);
4533                 mutex_unlock(&fs_info->chunk_mutex);
4534         }
4535         return ret;
4536 }
4537
4538 static int btrfs_add_system_chunk(struct btrfs_fs_info *fs_info,
4539                            struct btrfs_key *key,
4540                            struct btrfs_chunk *chunk, int item_size)
4541 {
4542         struct btrfs_super_block *super_copy = fs_info->super_copy;
4543         struct btrfs_disk_key disk_key;
4544         u32 array_size;
4545         u8 *ptr;
4546
4547         mutex_lock(&fs_info->chunk_mutex);
4548         array_size = btrfs_super_sys_array_size(super_copy);
4549         if (array_size + item_size + sizeof(disk_key)
4550                         > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
4551                 mutex_unlock(&fs_info->chunk_mutex);
4552                 return -EFBIG;
4553         }
4554
4555         ptr = super_copy->sys_chunk_array + array_size;
4556         btrfs_cpu_key_to_disk(&disk_key, key);
4557         memcpy(ptr, &disk_key, sizeof(disk_key));
4558         ptr += sizeof(disk_key);
4559         memcpy(ptr, chunk, item_size);
4560         item_size += sizeof(disk_key);
4561         btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
4562         mutex_unlock(&fs_info->chunk_mutex);
4563
4564         return 0;
4565 }
4566
4567 /*
4568  * sort the devices in descending order by max_avail, total_avail
4569  */
4570 static int btrfs_cmp_device_info(const void *a, const void *b)
4571 {
4572         const struct btrfs_device_info *di_a = a;
4573         const struct btrfs_device_info *di_b = b;
4574
4575         if (di_a->max_avail > di_b->max_avail)
4576                 return -1;
4577         if (di_a->max_avail < di_b->max_avail)
4578                 return 1;
4579         if (di_a->total_avail > di_b->total_avail)
4580                 return -1;
4581         if (di_a->total_avail < di_b->total_avail)
4582                 return 1;
4583         return 0;
4584 }
4585
4586 static u32 find_raid56_stripe_len(u32 data_devices, u32 dev_stripe_target)
4587 {
4588         /* TODO allow them to set a preferred stripe size */
4589         return SZ_64K;
4590 }
4591
4592 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
4593 {
4594         if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK))
4595                 return;
4596
4597         btrfs_set_fs_incompat(info, RAID56);
4598 }
4599
4600 #define BTRFS_MAX_DEVS(r) ((BTRFS_MAX_ITEM_SIZE(r->fs_info)             \
4601                         - sizeof(struct btrfs_chunk))           \
4602                         / sizeof(struct btrfs_stripe) + 1)
4603
4604 #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE        \
4605                                 - 2 * sizeof(struct btrfs_disk_key)     \
4606                                 - 2 * sizeof(struct btrfs_chunk))       \
4607                                 / sizeof(struct btrfs_stripe) + 1)
4608
4609 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4610                                u64 start, u64 type)
4611 {
4612         struct btrfs_fs_info *info = trans->fs_info;
4613         struct btrfs_fs_devices *fs_devices = info->fs_devices;
4614         struct list_head *cur;
4615         struct map_lookup *map = NULL;
4616         struct extent_map_tree *em_tree;
4617         struct extent_map *em;
4618         struct btrfs_device_info *devices_info = NULL;
4619         u64 total_avail;
4620         int num_stripes;        /* total number of stripes to allocate */
4621         int data_stripes;       /* number of stripes that count for
4622                                    block group size */
4623         int sub_stripes;        /* sub_stripes info for map */
4624         int dev_stripes;        /* stripes per dev */
4625         int devs_max;           /* max devs to use */
4626         int devs_min;           /* min devs needed */
4627         int devs_increment;     /* ndevs has to be a multiple of this */
4628         int ncopies;            /* how many copies to data has */
4629         int ret;
4630         u64 max_stripe_size;
4631         u64 max_chunk_size;
4632         u64 stripe_size;
4633         u64 num_bytes;
4634         u64 raid_stripe_len = BTRFS_STRIPE_LEN;
4635         int ndevs;
4636         int i;
4637         int j;
4638         int index;
4639
4640         BUG_ON(!alloc_profile_is_valid(type, 0));
4641
4642         if (list_empty(&fs_devices->alloc_list))
4643                 return -ENOSPC;
4644
4645         index = __get_raid_index(type);
4646
4647         sub_stripes = btrfs_raid_array[index].sub_stripes;
4648         dev_stripes = btrfs_raid_array[index].dev_stripes;
4649         devs_max = btrfs_raid_array[index].devs_max;
4650         devs_min = btrfs_raid_array[index].devs_min;
4651         devs_increment = btrfs_raid_array[index].devs_increment;
4652         ncopies = btrfs_raid_array[index].ncopies;
4653
4654         if (type & BTRFS_BLOCK_GROUP_DATA) {
4655                 max_stripe_size = SZ_1G;
4656                 max_chunk_size = 10 * max_stripe_size;
4657                 if (!devs_max)
4658                         devs_max = BTRFS_MAX_DEVS(info->chunk_root);
4659         } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
4660                 /* for larger filesystems, use larger metadata chunks */
4661                 if (fs_devices->total_rw_bytes > 50ULL * SZ_1G)
4662                         max_stripe_size = SZ_1G;
4663                 else
4664                         max_stripe_size = SZ_256M;
4665                 max_chunk_size = max_stripe_size;
4666                 if (!devs_max)
4667                         devs_max = BTRFS_MAX_DEVS(info->chunk_root);
4668         } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
4669                 max_stripe_size = SZ_32M;
4670                 max_chunk_size = 2 * max_stripe_size;
4671                 if (!devs_max)
4672                         devs_max = BTRFS_MAX_DEVS_SYS_CHUNK;
4673         } else {
4674                 btrfs_err(info, "invalid chunk type 0x%llx requested",
4675                        type);
4676                 BUG_ON(1);
4677         }
4678
4679         /* we don't want a chunk larger than 10% of writeable space */
4680         max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
4681                              max_chunk_size);
4682
4683         devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info),
4684                                GFP_NOFS);
4685         if (!devices_info)
4686                 return -ENOMEM;
4687
4688         cur = fs_devices->alloc_list.next;
4689
4690         /*
4691          * in the first pass through the devices list, we gather information
4692          * about the available holes on each device.
4693          */
4694         ndevs = 0;
4695         while (cur != &fs_devices->alloc_list) {
4696                 struct btrfs_device *device;
4697                 u64 max_avail;
4698                 u64 dev_offset;
4699
4700                 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
4701
4702                 cur = cur->next;
4703
4704                 if (!device->writeable) {
4705                         WARN(1, KERN_ERR
4706                                "BTRFS: read-only device in alloc_list\n");
4707                         continue;
4708                 }
4709
4710                 if (!device->in_fs_metadata ||
4711                     device->is_tgtdev_for_dev_replace)
4712                         continue;
4713
4714                 if (device->total_bytes > device->bytes_used)
4715                         total_avail = device->total_bytes - device->bytes_used;
4716                 else
4717                         total_avail = 0;
4718
4719                 /* If there is no space on this device, skip it. */
4720                 if (total_avail == 0)
4721                         continue;
4722
4723                 ret = find_free_dev_extent(trans, device,
4724                                            max_stripe_size * dev_stripes,
4725                                            &dev_offset, &max_avail);
4726                 if (ret && ret != -ENOSPC)
4727                         goto error;
4728
4729                 if (ret == 0)
4730                         max_avail = max_stripe_size * dev_stripes;
4731
4732                 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes)
4733                         continue;
4734
4735                 if (ndevs == fs_devices->rw_devices) {
4736                         WARN(1, "%s: found more than %llu devices\n",
4737                              __func__, fs_devices->rw_devices);
4738                         break;
4739                 }
4740                 devices_info[ndevs].dev_offset = dev_offset;
4741                 devices_info[ndevs].max_avail = max_avail;
4742                 devices_info[ndevs].total_avail = total_avail;
4743                 devices_info[ndevs].dev = device;
4744                 ++ndevs;
4745         }
4746
4747         /*
4748          * now sort the devices by hole size / available space
4749          */
4750         sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
4751              btrfs_cmp_device_info, NULL);
4752
4753         /* round down to number of usable stripes */
4754         ndevs -= ndevs % devs_increment;
4755
4756         if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) {
4757                 ret = -ENOSPC;
4758                 goto error;
4759         }
4760
4761         if (devs_max && ndevs > devs_max)
4762                 ndevs = devs_max;
4763         /*
4764          * the primary goal is to maximize the number of stripes, so use as many
4765          * devices as possible, even if the stripes are not maximum sized.
4766          */
4767         stripe_size = devices_info[ndevs-1].max_avail;
4768         num_stripes = ndevs * dev_stripes;
4769
4770         /*
4771          * this will have to be fixed for RAID1 and RAID10 over
4772          * more drives
4773          */
4774         data_stripes = num_stripes / ncopies;
4775
4776         if (type & BTRFS_BLOCK_GROUP_RAID5) {
4777                 raid_stripe_len = find_raid56_stripe_len(ndevs - 1,
4778                                                          info->stripesize);
4779                 data_stripes = num_stripes - 1;
4780         }
4781         if (type & BTRFS_BLOCK_GROUP_RAID6) {
4782                 raid_stripe_len = find_raid56_stripe_len(ndevs - 2,
4783                                                          info->stripesize);
4784                 data_stripes = num_stripes - 2;
4785         }
4786
4787         /*
4788          * Use the number of data stripes to figure out how big this chunk
4789          * is really going to be in terms of logical address space,
4790          * and compare that answer with the max chunk size
4791          */
4792         if (stripe_size * data_stripes > max_chunk_size) {
4793                 u64 mask = (1ULL << 24) - 1;
4794
4795                 stripe_size = div_u64(max_chunk_size, data_stripes);
4796
4797                 /* bump the answer up to a 16MB boundary */
4798                 stripe_size = (stripe_size + mask) & ~mask;
4799
4800                 /* but don't go higher than the limits we found
4801                  * while searching for free extents
4802                  */
4803                 if (stripe_size > devices_info[ndevs-1].max_avail)
4804                         stripe_size = devices_info[ndevs-1].max_avail;
4805         }
4806
4807         stripe_size = div_u64(stripe_size, dev_stripes);
4808
4809         /* align to BTRFS_STRIPE_LEN */
4810         stripe_size = div64_u64(stripe_size, raid_stripe_len);
4811         stripe_size *= raid_stripe_len;
4812
4813         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
4814         if (!map) {
4815                 ret = -ENOMEM;
4816                 goto error;
4817         }
4818         map->num_stripes = num_stripes;
4819
4820         for (i = 0; i < ndevs; ++i) {
4821                 for (j = 0; j < dev_stripes; ++j) {
4822                         int s = i * dev_stripes + j;
4823                         map->stripes[s].dev = devices_info[i].dev;
4824                         map->stripes[s].physical = devices_info[i].dev_offset +
4825                                                    j * stripe_size;
4826                 }
4827         }
4828         map->sector_size = info->sectorsize;
4829         map->stripe_len = raid_stripe_len;
4830         map->io_align = raid_stripe_len;
4831         map->io_width = raid_stripe_len;
4832         map->type = type;
4833         map->sub_stripes = sub_stripes;
4834
4835         num_bytes = stripe_size * data_stripes;
4836
4837         trace_btrfs_chunk_alloc(info, map, start, num_bytes);
4838
4839         em = alloc_extent_map();
4840         if (!em) {
4841                 kfree(map);
4842                 ret = -ENOMEM;
4843                 goto error;
4844         }
4845         set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
4846         em->map_lookup = map;
4847         em->start = start;
4848         em->len = num_bytes;
4849         em->block_start = 0;
4850         em->block_len = em->len;
4851         em->orig_block_len = stripe_size;
4852
4853         em_tree = &info->mapping_tree.map_tree;
4854         write_lock(&em_tree->lock);
4855         ret = add_extent_mapping(em_tree, em, 0);
4856         if (!ret) {
4857                 list_add_tail(&em->list, &trans->transaction->pending_chunks);
4858                 refcount_inc(&em->refs);
4859         }
4860         write_unlock(&em_tree->lock);
4861         if (ret) {
4862                 free_extent_map(em);
4863                 goto error;
4864         }
4865
4866         ret = btrfs_make_block_group(trans, info, 0, type,
4867                                      BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4868                                      start, num_bytes);
4869         if (ret)
4870                 goto error_del_extent;
4871
4872         for (i = 0; i < map->num_stripes; i++) {
4873                 num_bytes = map->stripes[i].dev->bytes_used + stripe_size;
4874                 btrfs_device_set_bytes_used(map->stripes[i].dev, num_bytes);
4875         }
4876
4877         atomic64_sub(stripe_size * map->num_stripes, &info->free_chunk_space);
4878
4879         free_extent_map(em);
4880         check_raid56_incompat_flag(info, type);
4881
4882         kfree(devices_info);
4883         return 0;
4884
4885 error_del_extent:
4886         write_lock(&em_tree->lock);
4887         remove_extent_mapping(em_tree, em);
4888         write_unlock(&em_tree->lock);
4889
4890         /* One for our allocation */
4891         free_extent_map(em);
4892         /* One for the tree reference */
4893         free_extent_map(em);
4894         /* One for the pending_chunks list reference */
4895         free_extent_map(em);
4896 error:
4897         kfree(devices_info);
4898         return ret;
4899 }
4900
4901 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans,
4902                                 struct btrfs_fs_info *fs_info,
4903                                 u64 chunk_offset, u64 chunk_size)
4904 {
4905         struct btrfs_root *extent_root = fs_info->extent_root;
4906         struct btrfs_root *chunk_root = fs_info->chunk_root;
4907         struct btrfs_key key;
4908         struct btrfs_device *device;
4909         struct btrfs_chunk *chunk;
4910         struct btrfs_stripe *stripe;
4911         struct extent_map *em;
4912         struct map_lookup *map;
4913         size_t item_size;
4914         u64 dev_offset;
4915         u64 stripe_size;
4916         int i = 0;
4917         int ret = 0;
4918
4919         em = get_chunk_map(fs_info, chunk_offset, chunk_size);
4920         if (IS_ERR(em))
4921                 return PTR_ERR(em);
4922
4923         map = em->map_lookup;
4924         item_size = btrfs_chunk_item_size(map->num_stripes);
4925         stripe_size = em->orig_block_len;
4926
4927         chunk = kzalloc(item_size, GFP_NOFS);
4928         if (!chunk) {
4929                 ret = -ENOMEM;
4930                 goto out;
4931         }
4932
4933         /*
4934          * Take the device list mutex to prevent races with the final phase of
4935          * a device replace operation that replaces the device object associated
4936          * with the map's stripes, because the device object's id can change
4937          * at any time during that final phase of the device replace operation
4938          * (dev-replace.c:btrfs_dev_replace_finishing()).
4939          */
4940         mutex_lock(&fs_info->fs_devices->device_list_mutex);
4941         for (i = 0; i < map->num_stripes; i++) {
4942                 device = map->stripes[i].dev;
4943                 dev_offset = map->stripes[i].physical;
4944
4945                 ret = btrfs_update_device(trans, device);
4946                 if (ret)
4947                         break;
4948                 ret = btrfs_alloc_dev_extent(trans, device,
4949                                              chunk_root->root_key.objectid,
4950                                              BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4951                                              chunk_offset, dev_offset,
4952                                              stripe_size);
4953                 if (ret)
4954                         break;
4955         }
4956         if (ret) {
4957                 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4958                 goto out;
4959         }
4960
4961         stripe = &chunk->stripe;
4962         for (i = 0; i < map->num_stripes; i++) {
4963                 device = map->stripes[i].dev;
4964                 dev_offset = map->stripes[i].physical;
4965
4966                 btrfs_set_stack_stripe_devid(stripe, device->devid);
4967                 btrfs_set_stack_stripe_offset(stripe, dev_offset);
4968                 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
4969                 stripe++;
4970         }
4971         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4972
4973         btrfs_set_stack_chunk_length(chunk, chunk_size);
4974         btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
4975         btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
4976         btrfs_set_stack_chunk_type(chunk, map->type);
4977         btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
4978         btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
4979         btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
4980         btrfs_set_stack_chunk_sector_size(chunk, fs_info->sectorsize);
4981         btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
4982
4983         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
4984         key.type = BTRFS_CHUNK_ITEM_KEY;
4985         key.offset = chunk_offset;
4986
4987         ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
4988         if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
4989                 /*
4990                  * TODO: Cleanup of inserted chunk root in case of
4991                  * failure.
4992                  */
4993                 ret = btrfs_add_system_chunk(fs_info, &key, chunk, item_size);
4994         }
4995
4996 out:
4997         kfree(chunk);
4998         free_extent_map(em);
4999         return ret;
5000 }
5001
5002 /*
5003  * Chunk allocation falls into two parts. The first part does works
5004  * that make the new allocated chunk useable, but not do any operation
5005  * that modifies the chunk tree. The second part does the works that
5006  * require modifying the chunk tree. This division is important for the
5007  * bootstrap process of adding storage to a seed btrfs.
5008  */
5009 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
5010                       struct btrfs_fs_info *fs_info, u64 type)
5011 {
5012         u64 chunk_offset;
5013
5014         ASSERT(mutex_is_locked(&fs_info->chunk_mutex));
5015         chunk_offset = find_next_chunk(fs_info);
5016         return __btrfs_alloc_chunk(trans, chunk_offset, type);
5017 }
5018
5019 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
5020                                          struct btrfs_fs_info *fs_info)
5021 {
5022         struct btrfs_root *extent_root = fs_info->extent_root;
5023         u64 chunk_offset;
5024         u64 sys_chunk_offset;
5025         u64 alloc_profile;
5026         int ret;
5027
5028         chunk_offset = find_next_chunk(fs_info);
5029         alloc_profile = btrfs_get_alloc_profile(extent_root, 0);
5030         ret = __btrfs_alloc_chunk(trans, chunk_offset, alloc_profile);
5031         if (ret)
5032                 return ret;
5033
5034         sys_chunk_offset = find_next_chunk(fs_info);
5035         alloc_profile = btrfs_get_alloc_profile(fs_info->chunk_root, 0);
5036         ret = __btrfs_alloc_chunk(trans, sys_chunk_offset, alloc_profile);
5037         return ret;
5038 }
5039
5040 static inline int btrfs_chunk_max_errors(struct map_lookup *map)
5041 {
5042         int max_errors;
5043
5044         if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
5045                          BTRFS_BLOCK_GROUP_RAID10 |
5046                          BTRFS_BLOCK_GROUP_RAID5 |
5047                          BTRFS_BLOCK_GROUP_DUP)) {
5048                 max_errors = 1;
5049         } else if (map->type & BTRFS_BLOCK_GROUP_RAID6) {
5050                 max_errors = 2;
5051         } else {
5052                 max_errors = 0;
5053         }
5054
5055         return max_errors;
5056 }
5057
5058 int btrfs_chunk_readonly(struct btrfs_fs_info *fs_info, u64 chunk_offset)
5059 {
5060         struct extent_map *em;
5061         struct map_lookup *map;
5062         int readonly = 0;
5063         int miss_ndevs = 0;
5064         int i;
5065
5066         em = get_chunk_map(fs_info, chunk_offset, 1);
5067         if (IS_ERR(em))
5068                 return 1;
5069
5070         map = em->map_lookup;
5071         for (i = 0; i < map->num_stripes; i++) {
5072                 if (map->stripes[i].dev->missing) {
5073                         miss_ndevs++;
5074                         continue;
5075                 }
5076
5077                 if (!map->stripes[i].dev->writeable) {
5078                         readonly = 1;
5079                         goto end;
5080                 }
5081         }
5082
5083         /*
5084          * If the number of missing devices is larger than max errors,
5085          * we can not write the data into that chunk successfully, so
5086          * set it readonly.
5087          */
5088         if (miss_ndevs > btrfs_chunk_max_errors(map))
5089                 readonly = 1;
5090 end:
5091         free_extent_map(em);
5092         return readonly;
5093 }
5094
5095 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
5096 {
5097         extent_map_tree_init(&tree->map_tree);
5098 }
5099
5100 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
5101 {
5102         struct extent_map *em;
5103
5104         while (1) {
5105                 write_lock(&tree->map_tree.lock);
5106                 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
5107                 if (em)
5108                         remove_extent_mapping(&tree->map_tree, em);
5109                 write_unlock(&tree->map_tree.lock);
5110                 if (!em)
5111                         break;
5112                 /* once for us */
5113                 free_extent_map(em);
5114                 /* once for the tree */
5115                 free_extent_map(em);
5116         }
5117 }
5118
5119 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5120 {
5121         struct extent_map *em;
5122         struct map_lookup *map;
5123         int ret;
5124
5125         em = get_chunk_map(fs_info, logical, len);
5126         if (IS_ERR(em))
5127                 /*
5128                  * We could return errors for these cases, but that could get
5129                  * ugly and we'd probably do the same thing which is just not do
5130                  * anything else and exit, so return 1 so the callers don't try
5131                  * to use other copies.
5132                  */
5133                 return 1;
5134
5135         map = em->map_lookup;
5136         if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
5137                 ret = map->num_stripes;
5138         else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5139                 ret = map->sub_stripes;
5140         else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
5141                 ret = 2;
5142         else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5143                 ret = 3;
5144         else
5145                 ret = 1;
5146         free_extent_map(em);
5147
5148         btrfs_dev_replace_lock(&fs_info->dev_replace, 0);
5149         if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace) &&
5150             fs_info->dev_replace.tgtdev)
5151                 ret++;
5152         btrfs_dev_replace_unlock(&fs_info->dev_replace, 0);
5153
5154         return ret;
5155 }
5156
5157 unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info,
5158                                     struct btrfs_mapping_tree *map_tree,
5159                                     u64 logical)
5160 {
5161         struct extent_map *em;
5162         struct map_lookup *map;
5163         unsigned long len = fs_info->sectorsize;
5164
5165         em = get_chunk_map(fs_info, logical, len);
5166         WARN_ON(IS_ERR(em));
5167
5168         map = em->map_lookup;
5169         if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5170                 len = map->stripe_len * nr_data_stripes(map);
5171         free_extent_map(em);
5172         return len;
5173 }
5174
5175 int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info,
5176                            u64 logical, u64 len, int mirror_num)
5177 {
5178         struct extent_map *em;
5179         struct map_lookup *map;
5180         int ret = 0;
5181
5182         em = get_chunk_map(fs_info, logical, len);
5183         WARN_ON(IS_ERR(em));
5184
5185         map = em->map_lookup;
5186         if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5187                 ret = 1;
5188         free_extent_map(em);
5189         return ret;
5190 }
5191
5192 static int find_live_mirror(struct btrfs_fs_info *fs_info,
5193                             struct map_lookup *map, int first, int num,
5194                             int optimal, int dev_replace_is_ongoing)
5195 {
5196         int i;
5197         int tolerance;
5198         struct btrfs_device *srcdev;
5199
5200         if (dev_replace_is_ongoing &&
5201             fs_info->dev_replace.cont_reading_from_srcdev_mode ==
5202              BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
5203                 srcdev = fs_info->dev_replace.srcdev;
5204         else
5205                 srcdev = NULL;
5206
5207         /*
5208          * try to avoid the drive that is the source drive for a
5209          * dev-replace procedure, only choose it if no other non-missing
5210          * mirror is available
5211          */
5212         for (tolerance = 0; tolerance < 2; tolerance++) {
5213                 if (map->stripes[optimal].dev->bdev &&
5214                     (tolerance || map->stripes[optimal].dev != srcdev))
5215                         return optimal;
5216                 for (i = first; i < first + num; i++) {
5217                         if (map->stripes[i].dev->bdev &&
5218                             (tolerance || map->stripes[i].dev != srcdev))
5219                                 return i;
5220                 }
5221         }
5222
5223         /* we couldn't find one that doesn't fail.  Just return something
5224          * and the io error handling code will clean up eventually
5225          */
5226         return optimal;
5227 }
5228
5229 static inline int parity_smaller(u64 a, u64 b)
5230 {
5231         return a > b;
5232 }
5233
5234 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
5235 static void sort_parity_stripes(struct btrfs_bio *bbio, int num_stripes)
5236 {
5237         struct btrfs_bio_stripe s;
5238         int i;
5239         u64 l;
5240         int again = 1;
5241
5242         while (again) {
5243                 again = 0;
5244                 for (i = 0; i < num_stripes - 1; i++) {
5245                         if (parity_smaller(bbio->raid_map[i],
5246                                            bbio->raid_map[i+1])) {
5247                                 s = bbio->stripes[i];
5248                                 l = bbio->raid_map[i];
5249                                 bbio->stripes[i] = bbio->stripes[i+1];
5250                                 bbio->raid_map[i] = bbio->raid_map[i+1];
5251                                 bbio->stripes[i+1] = s;
5252                                 bbio->raid_map[i+1] = l;
5253
5254                                 again = 1;
5255                         }
5256                 }
5257         }
5258 }
5259
5260 static struct btrfs_bio *alloc_btrfs_bio(int total_stripes, int real_stripes)
5261 {
5262         struct btrfs_bio *bbio = kzalloc(
5263                  /* the size of the btrfs_bio */
5264                 sizeof(struct btrfs_bio) +
5265                 /* plus the variable array for the stripes */
5266                 sizeof(struct btrfs_bio_stripe) * (total_stripes) +
5267                 /* plus the variable array for the tgt dev */
5268                 sizeof(int) * (real_stripes) +
5269                 /*
5270                  * plus the raid_map, which includes both the tgt dev
5271                  * and the stripes
5272                  */
5273                 sizeof(u64) * (total_stripes),
5274                 GFP_NOFS|__GFP_NOFAIL);
5275
5276         atomic_set(&bbio->error, 0);
5277         refcount_set(&bbio->refs, 1);
5278
5279         return bbio;
5280 }
5281
5282 void btrfs_get_bbio(struct btrfs_bio *bbio)
5283 {
5284         WARN_ON(!refcount_read(&bbio->refs));
5285         refcount_inc(&bbio->refs);
5286 }
5287
5288 void btrfs_put_bbio(struct btrfs_bio *bbio)
5289 {
5290         if (!bbio)
5291                 return;
5292         if (refcount_dec_and_test(&bbio->refs))
5293                 kfree(bbio);
5294 }
5295
5296 /* can REQ_OP_DISCARD be sent with other REQ like REQ_OP_WRITE? */
5297 /*
5298  * Please note that, discard won't be sent to target device of device
5299  * replace.
5300  */
5301 static int __btrfs_map_block_for_discard(struct btrfs_fs_info *fs_info,
5302                                          u64 logical, u64 length,
5303                                          struct btrfs_bio **bbio_ret)
5304 {
5305         struct extent_map *em;
5306         struct map_lookup *map;
5307         struct btrfs_bio *bbio;
5308         u64 offset;
5309         u64 stripe_nr;
5310         u64 stripe_nr_end;
5311         u64 stripe_end_offset;
5312         u64 stripe_cnt;
5313         u64 stripe_len;
5314         u64 stripe_offset;
5315         u64 num_stripes;
5316         u32 stripe_index;
5317         u32 factor = 0;
5318         u32 sub_stripes = 0;
5319         u64 stripes_per_dev = 0;
5320         u32 remaining_stripes = 0;
5321         u32 last_stripe = 0;
5322         int ret = 0;
5323         int i;
5324
5325         /* discard always return a bbio */
5326         ASSERT(bbio_ret);
5327
5328         em = get_chunk_map(fs_info, logical, length);
5329         if (IS_ERR(em))
5330                 return PTR_ERR(em);
5331
5332         map = em->map_lookup;
5333         /* we don't discard raid56 yet */
5334         if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5335                 ret = -EOPNOTSUPP;
5336                 goto out;
5337         }
5338
5339         offset = logical - em->start;
5340         length = min_t(u64, em->len - offset, length);
5341
5342         stripe_len = map->stripe_len;
5343         /*
5344          * stripe_nr counts the total number of stripes we have to stride
5345          * to get to this block
5346          */
5347         stripe_nr = div64_u64(offset, stripe_len);
5348
5349         /* stripe_offset is the offset of this block in its stripe */
5350         stripe_offset = offset - stripe_nr * stripe_len;
5351
5352         stripe_nr_end = round_up(offset + length, map->stripe_len);
5353         stripe_nr_end = div64_u64(stripe_nr_end, map->stripe_len);
5354         stripe_cnt = stripe_nr_end - stripe_nr;
5355         stripe_end_offset = stripe_nr_end * map->stripe_len -
5356                             (offset + length);
5357         /*
5358          * after this, stripe_nr is the number of stripes on this
5359          * device we have to walk to find the data, and stripe_index is
5360          * the number of our device in the stripe array
5361          */
5362         num_stripes = 1;
5363         stripe_index = 0;
5364         if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5365                          BTRFS_BLOCK_GROUP_RAID10)) {
5366                 if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5367                         sub_stripes = 1;
5368                 else
5369                         sub_stripes = map->sub_stripes;
5370
5371                 factor = map->num_stripes / sub_stripes;
5372                 num_stripes = min_t(u64, map->num_stripes,
5373                                     sub_stripes * stripe_cnt);
5374                 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
5375                 stripe_index *= sub_stripes;
5376                 stripes_per_dev = div_u64_rem(stripe_cnt, factor,
5377                                               &remaining_stripes);
5378                 div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
5379                 last_stripe *= sub_stripes;
5380         } else if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
5381                                 BTRFS_BLOCK_GROUP_DUP)) {
5382                 num_stripes = map->num_stripes;
5383         } else {
5384                 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5385                                         &stripe_index);
5386         }
5387
5388         bbio = alloc_btrfs_bio(num_stripes, 0);
5389         if (!bbio) {
5390                 ret = -ENOMEM;
5391                 goto out;
5392         }
5393
5394         for (i = 0; i < num_stripes; i++) {
5395                 bbio->stripes[i].physical =
5396                         map->stripes[stripe_index].physical +
5397                         stripe_offset + stripe_nr * map->stripe_len;
5398                 bbio->stripes[i].dev = map->stripes[stripe_index].dev;
5399
5400                 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5401                                  BTRFS_BLOCK_GROUP_RAID10)) {
5402                         bbio->stripes[i].length = stripes_per_dev *
5403                                 map->stripe_len;
5404
5405                         if (i / sub_stripes < remaining_stripes)
5406                                 bbio->stripes[i].length +=
5407                                         map->stripe_len;
5408
5409                         /*
5410                          * Special for the first stripe and
5411                          * the last stripe:
5412                          *
5413                          * |-------|...|-------|
5414                          *     |----------|
5415                          *    off     end_off
5416                          */
5417                         if (i < sub_stripes)
5418                                 bbio->stripes[i].length -=
5419                                         stripe_offset;
5420
5421                         if (stripe_index >= last_stripe &&
5422                             stripe_index <= (last_stripe +
5423                                              sub_stripes - 1))
5424                                 bbio->stripes[i].length -=
5425                                         stripe_end_offset;
5426
5427                         if (i == sub_stripes - 1)
5428                                 stripe_offset = 0;
5429                 } else {
5430                         bbio->stripes[i].length = length;
5431                 }
5432
5433                 stripe_index++;
5434                 if (stripe_index == map->num_stripes) {
5435                         stripe_index = 0;
5436                         stripe_nr++;
5437                 }
5438         }
5439
5440         *bbio_ret = bbio;
5441         bbio->map_type = map->type;
5442         bbio->num_stripes = num_stripes;
5443 out:
5444         free_extent_map(em);
5445         return ret;
5446 }
5447
5448 /*
5449  * In dev-replace case, for repair case (that's the only case where the mirror
5450  * is selected explicitly when calling btrfs_map_block), blocks left of the
5451  * left cursor can also be read from the target drive.
5452  *
5453  * For REQ_GET_READ_MIRRORS, the target drive is added as the last one to the
5454  * array of stripes.
5455  * For READ, it also needs to be supported using the same mirror number.
5456  *
5457  * If the requested block is not left of the left cursor, EIO is returned. This
5458  * can happen because btrfs_num_copies() returns one more in the dev-replace
5459  * case.
5460  */
5461 static int get_extra_mirror_from_replace(struct btrfs_fs_info *fs_info,
5462                                          u64 logical, u64 length,
5463                                          u64 srcdev_devid, int *mirror_num,
5464                                          u64 *physical)
5465 {
5466         struct btrfs_bio *bbio = NULL;
5467         int num_stripes;
5468         int index_srcdev = 0;
5469         int found = 0;
5470         u64 physical_of_found = 0;
5471         int i;
5472         int ret = 0;
5473
5474         ret = __btrfs_map_block(fs_info, BTRFS_MAP_GET_READ_MIRRORS,
5475                                 logical, &length, &bbio, 0, 0);
5476         if (ret) {
5477                 ASSERT(bbio == NULL);
5478                 return ret;
5479         }
5480
5481         num_stripes = bbio->num_stripes;
5482         if (*mirror_num > num_stripes) {
5483                 /*
5484                  * BTRFS_MAP_GET_READ_MIRRORS does not contain this mirror,
5485                  * that means that the requested area is not left of the left
5486                  * cursor
5487                  */
5488                 btrfs_put_bbio(bbio);
5489                 return -EIO;
5490         }
5491
5492         /*
5493          * process the rest of the function using the mirror_num of the source
5494          * drive. Therefore look it up first.  At the end, patch the device
5495          * pointer to the one of the target drive.
5496          */
5497         for (i = 0; i < num_stripes; i++) {
5498                 if (bbio->stripes[i].dev->devid != srcdev_devid)
5499                         continue;
5500
5501                 /*
5502                  * In case of DUP, in order to keep it simple, only add the
5503                  * mirror with the lowest physical address
5504                  */
5505                 if (found &&
5506                     physical_of_found <= bbio->stripes[i].physical)
5507                         continue;
5508
5509                 index_srcdev = i;
5510                 found = 1;
5511                 physical_of_found = bbio->stripes[i].physical;
5512         }
5513
5514         btrfs_put_bbio(bbio);
5515
5516         ASSERT(found);
5517         if (!found)
5518                 return -EIO;
5519
5520         *mirror_num = index_srcdev + 1;
5521         *physical = physical_of_found;
5522         return ret;
5523 }
5524
5525 static void handle_ops_on_dev_replace(enum btrfs_map_op op,
5526                                       struct btrfs_bio **bbio_ret,
5527                                       struct btrfs_dev_replace *dev_replace,
5528                                       int *num_stripes_ret, int *max_errors_ret)
5529 {
5530         struct btrfs_bio *bbio = *bbio_ret;
5531         u64 srcdev_devid = dev_replace->srcdev->devid;
5532         int tgtdev_indexes = 0;
5533         int num_stripes = *num_stripes_ret;
5534         int max_errors = *max_errors_ret;
5535         int i;
5536
5537         if (op == BTRFS_MAP_WRITE) {
5538                 int index_where_to_add;
5539
5540                 /*
5541                  * duplicate the write operations while the dev replace
5542                  * procedure is running. Since the copying of the old disk to
5543                  * the new disk takes place at run time while the filesystem is
5544                  * mounted writable, the regular write operations to the old
5545                  * disk have to be duplicated to go to the new disk as well.
5546                  *
5547                  * Note that device->missing is handled by the caller, and that
5548                  * the write to the old disk is already set up in the stripes
5549                  * array.
5550                  */
5551                 index_where_to_add = num_stripes;
5552                 for (i = 0; i < num_stripes; i++) {
5553                         if (bbio->stripes[i].dev->devid == srcdev_devid) {
5554                                 /* write to new disk, too */
5555                                 struct btrfs_bio_stripe *new =
5556                                         bbio->stripes + index_where_to_add;
5557                                 struct btrfs_bio_stripe *old =
5558                                         bbio->stripes + i;
5559
5560                                 new->physical = old->physical;
5561                                 new->length = old->length;
5562                                 new->dev = dev_replace->tgtdev;
5563                                 bbio->tgtdev_map[i] = index_where_to_add;
5564                                 index_where_to_add++;
5565                                 max_errors++;
5566                                 tgtdev_indexes++;
5567                         }
5568                 }
5569                 num_stripes = index_where_to_add;
5570         } else if (op == BTRFS_MAP_GET_READ_MIRRORS) {
5571                 int index_srcdev = 0;
5572                 int found = 0;
5573                 u64 physical_of_found = 0;
5574
5575                 /*
5576                  * During the dev-replace procedure, the target drive can also
5577                  * be used to read data in case it is needed to repair a corrupt
5578                  * block elsewhere. This is possible if the requested area is
5579                  * left of the left cursor. In this area, the target drive is a
5580                  * full copy of the source drive.
5581                  */
5582                 for (i = 0; i < num_stripes; i++) {
5583                         if (bbio->stripes[i].dev->devid == srcdev_devid) {
5584                                 /*
5585                                  * In case of DUP, in order to keep it simple,
5586                                  * only add the mirror with the lowest physical
5587                                  * address
5588                                  */
5589                                 if (found &&
5590                                     physical_of_found <=
5591                                      bbio->stripes[i].physical)
5592                                         continue;
5593                                 index_srcdev = i;
5594                                 found = 1;
5595                                 physical_of_found = bbio->stripes[i].physical;
5596                         }
5597                 }
5598                 if (found) {
5599                         struct btrfs_bio_stripe *tgtdev_stripe =
5600                                 bbio->stripes + num_stripes;
5601
5602                         tgtdev_stripe->physical = physical_of_found;
5603                         tgtdev_stripe->length =
5604                                 bbio->stripes[index_srcdev].length;
5605                         tgtdev_stripe->dev = dev_replace->tgtdev;
5606                         bbio->tgtdev_map[index_srcdev] = num_stripes;
5607
5608                         tgtdev_indexes++;
5609                         num_stripes++;
5610                 }
5611         }
5612
5613         *num_stripes_ret = num_stripes;
5614         *max_errors_ret = max_errors;
5615         bbio->num_tgtdevs = tgtdev_indexes;
5616         *bbio_ret = bbio;
5617 }
5618
5619 static bool need_full_stripe(enum btrfs_map_op op)
5620 {
5621         return (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS);
5622 }
5623
5624 static int __btrfs_map_block(struct btrfs_fs_info *fs_info,
5625                              enum btrfs_map_op op,
5626                              u64 logical, u64 *length,
5627                              struct btrfs_bio **bbio_ret,
5628                              int mirror_num, int need_raid_map)
5629 {
5630         struct extent_map *em;
5631         struct map_lookup *map;
5632         u64 offset;
5633         u64 stripe_offset;
5634         u64 stripe_nr;
5635         u64 stripe_len;
5636         u32 stripe_index;
5637         int i;
5638         int ret = 0;
5639         int num_stripes;
5640         int max_errors = 0;
5641         int tgtdev_indexes = 0;
5642         struct btrfs_bio *bbio = NULL;
5643         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
5644         int dev_replace_is_ongoing = 0;
5645         int num_alloc_stripes;
5646         int patch_the_first_stripe_for_dev_replace = 0;
5647         u64 physical_to_patch_in_first_stripe = 0;
5648         u64 raid56_full_stripe_start = (u64)-1;
5649
5650         if (op == BTRFS_MAP_DISCARD)
5651                 return __btrfs_map_block_for_discard(fs_info, logical,
5652                                                      *length, bbio_ret);
5653
5654         em = get_chunk_map(fs_info, logical, *length);
5655         if (IS_ERR(em))
5656                 return PTR_ERR(em);
5657
5658         map = em->map_lookup;
5659         offset = logical - em->start;
5660
5661         stripe_len = map->stripe_len;
5662         stripe_nr = offset;
5663         /*
5664          * stripe_nr counts the total number of stripes we have to stride
5665          * to get to this block
5666          */
5667         stripe_nr = div64_u64(stripe_nr, stripe_len);
5668
5669         stripe_offset = stripe_nr * stripe_len;
5670         if (offset < stripe_offset) {
5671                 btrfs_crit(fs_info,
5672                            "stripe math has gone wrong, stripe_offset=%llu, offset=%llu, start=%llu, logical=%llu, stripe_len=%llu",
5673                            stripe_offset, offset, em->start, logical,
5674                            stripe_len);
5675                 free_extent_map(em);
5676                 return -EINVAL;
5677         }
5678
5679         /* stripe_offset is the offset of this block in its stripe*/
5680         stripe_offset = offset - stripe_offset;
5681
5682         /* if we're here for raid56, we need to know the stripe aligned start */
5683         if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5684                 unsigned long full_stripe_len = stripe_len * nr_data_stripes(map);
5685                 raid56_full_stripe_start = offset;
5686
5687                 /* allow a write of a full stripe, but make sure we don't
5688                  * allow straddling of stripes
5689                  */
5690                 raid56_full_stripe_start = div64_u64(raid56_full_stripe_start,
5691                                 full_stripe_len);
5692                 raid56_full_stripe_start *= full_stripe_len;
5693         }
5694
5695         if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
5696                 u64 max_len;
5697                 /* For writes to RAID[56], allow a full stripeset across all disks.
5698                    For other RAID types and for RAID[56] reads, just allow a single
5699                    stripe (on a single disk). */
5700                 if ((map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
5701                     (op == BTRFS_MAP_WRITE)) {
5702                         max_len = stripe_len * nr_data_stripes(map) -
5703                                 (offset - raid56_full_stripe_start);
5704                 } else {
5705                         /* we limit the length of each bio to what fits in a stripe */
5706                         max_len = stripe_len - stripe_offset;
5707                 }
5708                 *length = min_t(u64, em->len - offset, max_len);
5709         } else {
5710                 *length = em->len - offset;
5711         }
5712
5713         /* This is for when we're called from btrfs_merge_bio_hook() and all
5714            it cares about is the length */
5715         if (!bbio_ret)
5716                 goto out;
5717
5718         btrfs_dev_replace_lock(dev_replace, 0);
5719         dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
5720         if (!dev_replace_is_ongoing)
5721                 btrfs_dev_replace_unlock(dev_replace, 0);
5722         else
5723                 btrfs_dev_replace_set_lock_blocking(dev_replace);
5724
5725         if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
5726             !need_full_stripe(op) && dev_replace->tgtdev != NULL) {
5727                 ret = get_extra_mirror_from_replace(fs_info, logical, *length,
5728                                                     dev_replace->srcdev->devid,
5729                                                     &mirror_num,
5730                                             &physical_to_patch_in_first_stripe);
5731                 if (ret)
5732                         goto out;
5733                 else
5734                         patch_the_first_stripe_for_dev_replace = 1;
5735         } else if (mirror_num > map->num_stripes) {
5736                 mirror_num = 0;
5737         }
5738
5739         num_stripes = 1;
5740         stripe_index = 0;
5741         if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5742                 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5743                                 &stripe_index);
5744                 if (op != BTRFS_MAP_WRITE && op != BTRFS_MAP_GET_READ_MIRRORS)
5745                         mirror_num = 1;
5746         } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
5747                 if (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS)
5748                         num_stripes = map->num_stripes;
5749                 else if (mirror_num)
5750                         stripe_index = mirror_num - 1;
5751                 else {
5752                         stripe_index = find_live_mirror(fs_info, map, 0,
5753                                             map->num_stripes,
5754                                             current->pid % map->num_stripes,
5755                                             dev_replace_is_ongoing);
5756                         mirror_num = stripe_index + 1;
5757                 }
5758
5759         } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
5760                 if (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS) {
5761                         num_stripes = map->num_stripes;
5762                 } else if (mirror_num) {
5763                         stripe_index = mirror_num - 1;
5764                 } else {
5765                         mirror_num = 1;
5766                 }
5767
5768         } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5769                 u32 factor = map->num_stripes / map->sub_stripes;
5770
5771                 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
5772                 stripe_index *= map->sub_stripes;
5773
5774                 if (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS)
5775                         num_stripes = map->sub_stripes;
5776                 else if (mirror_num)
5777                         stripe_index += mirror_num - 1;
5778                 else {
5779                         int old_stripe_index = stripe_index;
5780                         stripe_index = find_live_mirror(fs_info, map,
5781                                               stripe_index,
5782                                               map->sub_stripes, stripe_index +
5783                                               current->pid % map->sub_stripes,
5784                                               dev_replace_is_ongoing);
5785                         mirror_num = stripe_index - old_stripe_index + 1;
5786                 }
5787
5788         } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5789                 if (need_raid_map &&
5790                     (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS ||
5791                      mirror_num > 1)) {
5792                         /* push stripe_nr back to the start of the full stripe */
5793                         stripe_nr = div64_u64(raid56_full_stripe_start,
5794                                         stripe_len * nr_data_stripes(map));
5795
5796                         /* RAID[56] write or recovery. Return all stripes */
5797                         num_stripes = map->num_stripes;
5798                         max_errors = nr_parity_stripes(map);
5799
5800                         *length = map->stripe_len;
5801                         stripe_index = 0;
5802                         stripe_offset = 0;
5803                 } else {
5804                         /*
5805                          * Mirror #0 or #1 means the original data block.
5806                          * Mirror #2 is RAID5 parity block.
5807                          * Mirror #3 is RAID6 Q block.
5808                          */
5809                         stripe_nr = div_u64_rem(stripe_nr,
5810                                         nr_data_stripes(map), &stripe_index);
5811                         if (mirror_num > 1)
5812                                 stripe_index = nr_data_stripes(map) +
5813                                                 mirror_num - 2;
5814
5815                         /* We distribute the parity blocks across stripes */
5816                         div_u64_rem(stripe_nr + stripe_index, map->num_stripes,
5817                                         &stripe_index);
5818                         if ((op != BTRFS_MAP_WRITE &&
5819                              op != BTRFS_MAP_GET_READ_MIRRORS) &&
5820                             mirror_num <= 1)
5821                                 mirror_num = 1;
5822                 }
5823         } else {
5824                 /*
5825                  * after this, stripe_nr is the number of stripes on this
5826                  * device we have to walk to find the data, and stripe_index is
5827                  * the number of our device in the stripe array
5828                  */
5829                 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5830                                 &stripe_index);
5831                 mirror_num = stripe_index + 1;
5832         }
5833         if (stripe_index >= map->num_stripes) {
5834                 btrfs_crit(fs_info,
5835                            "stripe index math went horribly wrong, got stripe_index=%u, num_stripes=%u",
5836                            stripe_index, map->num_stripes);
5837                 ret = -EINVAL;
5838                 goto out;
5839         }
5840
5841         num_alloc_stripes = num_stripes;
5842         if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL) {
5843                 if (op == BTRFS_MAP_WRITE)
5844                         num_alloc_stripes <<= 1;
5845                 if (op == BTRFS_MAP_GET_READ_MIRRORS)
5846                         num_alloc_stripes++;
5847                 tgtdev_indexes = num_stripes;
5848         }
5849
5850         bbio = alloc_btrfs_bio(num_alloc_stripes, tgtdev_indexes);
5851         if (!bbio) {
5852                 ret = -ENOMEM;
5853                 goto out;
5854         }
5855         if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL)
5856                 bbio->tgtdev_map = (int *)(bbio->stripes + num_alloc_stripes);
5857
5858         /* build raid_map */
5859         if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK && need_raid_map &&
5860             (need_full_stripe(op) || mirror_num > 1)) {
5861                 u64 tmp;
5862                 unsigned rot;
5863
5864                 bbio->raid_map = (u64 *)((void *)bbio->stripes +
5865                                  sizeof(struct btrfs_bio_stripe) *
5866                                  num_alloc_stripes +
5867                                  sizeof(int) * tgtdev_indexes);
5868
5869                 /* Work out the disk rotation on this stripe-set */
5870                 div_u64_rem(stripe_nr, num_stripes, &rot);
5871
5872                 /* Fill in the logical address of each stripe */
5873                 tmp = stripe_nr * nr_data_stripes(map);
5874                 for (i = 0; i < nr_data_stripes(map); i++)
5875                         bbio->raid_map[(i+rot) % num_stripes] =
5876                                 em->start + (tmp + i) * map->stripe_len;
5877
5878                 bbio->raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE;
5879                 if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5880                         bbio->raid_map[(i+rot+1) % num_stripes] =
5881                                 RAID6_Q_STRIPE;
5882         }
5883
5884
5885         for (i = 0; i < num_stripes; i++) {
5886                 bbio->stripes[i].physical =
5887                         map->stripes[stripe_index].physical +
5888                         stripe_offset +
5889                         stripe_nr * map->stripe_len;
5890                 bbio->stripes[i].dev =
5891                         map->stripes[stripe_index].dev;
5892                 stripe_index++;
5893         }
5894
5895         if (need_full_stripe(op))
5896                 max_errors = btrfs_chunk_max_errors(map);
5897
5898         if (bbio->raid_map)
5899                 sort_parity_stripes(bbio, num_stripes);
5900
5901         if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL &&
5902             need_full_stripe(op)) {
5903                 handle_ops_on_dev_replace(op, &bbio, dev_replace, &num_stripes,
5904                                           &max_errors);
5905         }
5906
5907         *bbio_ret = bbio;
5908         bbio->map_type = map->type;
5909         bbio->num_stripes = num_stripes;
5910         bbio->max_errors = max_errors;
5911         bbio->mirror_num = mirror_num;
5912
5913         /*
5914          * this is the case that REQ_READ && dev_replace_is_ongoing &&
5915          * mirror_num == num_stripes + 1 && dev_replace target drive is
5916          * available as a mirror
5917          */
5918         if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) {
5919                 WARN_ON(num_stripes > 1);
5920                 bbio->stripes[0].dev = dev_replace->tgtdev;
5921                 bbio->stripes[0].physical = physical_to_patch_in_first_stripe;
5922                 bbio->mirror_num = map->num_stripes + 1;
5923         }
5924 out:
5925         if (dev_replace_is_ongoing) {
5926                 btrfs_dev_replace_clear_lock_blocking(dev_replace);
5927                 btrfs_dev_replace_unlock(dev_replace, 0);
5928         }
5929         free_extent_map(em);
5930         return ret;
5931 }
5932
5933 int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
5934                       u64 logical, u64 *length,
5935                       struct btrfs_bio **bbio_ret, int mirror_num)
5936 {
5937         return __btrfs_map_block(fs_info, op, logical, length, bbio_ret,
5938                                  mirror_num, 0);
5939 }
5940
5941 /* For Scrub/replace */
5942 int btrfs_map_sblock(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
5943                      u64 logical, u64 *length,
5944                      struct btrfs_bio **bbio_ret)
5945 {
5946         return __btrfs_map_block(fs_info, op, logical, length, bbio_ret, 0, 1);
5947 }
5948
5949 int btrfs_rmap_block(struct btrfs_fs_info *fs_info,
5950                      u64 chunk_start, u64 physical, u64 devid,
5951                      u64 **logical, int *naddrs, int *stripe_len)
5952 {
5953         struct extent_map *em;
5954         struct map_lookup *map;
5955         u64 *buf;
5956         u64 bytenr;
5957         u64 length;
5958         u64 stripe_nr;
5959         u64 rmap_len;
5960         int i, j, nr = 0;
5961
5962         em = get_chunk_map(fs_info, chunk_start, 1);
5963         if (IS_ERR(em))
5964                 return -EIO;
5965
5966         map = em->map_lookup;
5967         length = em->len;
5968         rmap_len = map->stripe_len;
5969
5970         if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5971                 length = div_u64(length, map->num_stripes / map->sub_stripes);
5972         else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5973                 length = div_u64(length, map->num_stripes);
5974         else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5975                 length = div_u64(length, nr_data_stripes(map));
5976                 rmap_len = map->stripe_len * nr_data_stripes(map);
5977         }
5978
5979         buf = kcalloc(map->num_stripes, sizeof(u64), GFP_NOFS);
5980         BUG_ON(!buf); /* -ENOMEM */
5981
5982         for (i = 0; i < map->num_stripes; i++) {
5983                 if (devid && map->stripes[i].dev->devid != devid)
5984                         continue;
5985                 if (map->stripes[i].physical > physical ||
5986                     map->stripes[i].physical + length <= physical)
5987                         continue;
5988
5989                 stripe_nr = physical - map->stripes[i].physical;
5990                 stripe_nr = div64_u64(stripe_nr, map->stripe_len);
5991
5992                 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5993                         stripe_nr = stripe_nr * map->num_stripes + i;
5994                         stripe_nr = div_u64(stripe_nr, map->sub_stripes);
5995                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5996                         stripe_nr = stripe_nr * map->num_stripes + i;
5997                 } /* else if RAID[56], multiply by nr_data_stripes().
5998                    * Alternatively, just use rmap_len below instead of
5999                    * map->stripe_len */
6000
6001                 bytenr = chunk_start + stripe_nr * rmap_len;
6002                 WARN_ON(nr >= map->num_stripes);
6003                 for (j = 0; j < nr; j++) {
6004                         if (buf[j] == bytenr)
6005                                 break;
6006                 }
6007                 if (j == nr) {
6008                         WARN_ON(nr >= map->num_stripes);
6009                         buf[nr++] = bytenr;
6010                 }
6011         }
6012
6013         *logical = buf;
6014         *naddrs = nr;
6015         *stripe_len = rmap_len;
6016
6017         free_extent_map(em);
6018         return 0;
6019 }
6020
6021 static inline void btrfs_end_bbio(struct btrfs_bio *bbio, struct bio *bio)
6022 {
6023         bio->bi_private = bbio->private;
6024         bio->bi_end_io = bbio->end_io;
6025         bio_endio(bio);
6026
6027         btrfs_put_bbio(bbio);
6028 }
6029
6030 static void btrfs_end_bio(struct bio *bio)
6031 {
6032         struct btrfs_bio *bbio = bio->bi_private;
6033         int is_orig_bio = 0;
6034
6035         if (bio->bi_error) {
6036                 atomic_inc(&bbio->error);
6037                 if (bio->bi_error == -EIO || bio->bi_error == -EREMOTEIO) {
6038                         unsigned int stripe_index =
6039                                 btrfs_io_bio(bio)->stripe_index;
6040                         struct btrfs_device *dev;
6041
6042                         BUG_ON(stripe_index >= bbio->num_stripes);
6043                         dev = bbio->stripes[stripe_index].dev;
6044                         if (dev->bdev) {
6045                                 if (bio_op(bio) == REQ_OP_WRITE)
6046                                         btrfs_dev_stat_inc(dev,
6047                                                 BTRFS_DEV_STAT_WRITE_ERRS);
6048                                 else
6049                                         btrfs_dev_stat_inc(dev,
6050                                                 BTRFS_DEV_STAT_READ_ERRS);
6051                                 if (bio->bi_opf & REQ_PREFLUSH)
6052                                         btrfs_dev_stat_inc(dev,
6053                                                 BTRFS_DEV_STAT_FLUSH_ERRS);
6054                                 btrfs_dev_stat_print_on_error(dev);
6055                         }
6056                 }
6057         }
6058
6059         if (bio == bbio->orig_bio)
6060                 is_orig_bio = 1;
6061
6062         btrfs_bio_counter_dec(bbio->fs_info);
6063
6064         if (atomic_dec_and_test(&bbio->stripes_pending)) {
6065                 if (!is_orig_bio) {
6066                         bio_put(bio);
6067                         bio = bbio->orig_bio;
6068                 }
6069
6070                 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
6071                 /* only send an error to the higher layers if it is
6072                  * beyond the tolerance of the btrfs bio
6073                  */
6074                 if (atomic_read(&bbio->error) > bbio->max_errors) {
6075                         bio->bi_error = -EIO;
6076                 } else {
6077                         /*
6078                          * this bio is actually up to date, we didn't
6079                          * go over the max number of errors
6080                          */
6081                         bio->bi_error = 0;
6082                 }
6083
6084                 btrfs_end_bbio(bbio, bio);
6085         } else if (!is_orig_bio) {
6086                 bio_put(bio);
6087         }
6088 }
6089
6090 /*
6091  * see run_scheduled_bios for a description of why bios are collected for
6092  * async submit.
6093  *
6094  * This will add one bio to the pending list for a device and make sure
6095  * the work struct is scheduled.
6096  */
6097 static noinline void btrfs_schedule_bio(struct btrfs_device *device,
6098                                         struct bio *bio)
6099 {
6100         struct btrfs_fs_info *fs_info = device->fs_info;
6101         int should_queue = 1;
6102         struct btrfs_pending_bios *pending_bios;
6103
6104         if (device->missing || !device->bdev) {
6105                 bio_io_error(bio);
6106                 return;
6107         }
6108
6109         /* don't bother with additional async steps for reads, right now */
6110         if (bio_op(bio) == REQ_OP_READ) {
6111                 bio_get(bio);
6112                 btrfsic_submit_bio(bio);
6113                 bio_put(bio);
6114                 return;
6115         }
6116
6117         /*
6118          * nr_async_bios allows us to reliably return congestion to the
6119          * higher layers.  Otherwise, the async bio makes it appear we have
6120          * made progress against dirty pages when we've really just put it
6121          * on a queue for later
6122          */
6123         atomic_inc(&fs_info->nr_async_bios);
6124         WARN_ON(bio->bi_next);
6125         bio->bi_next = NULL;
6126
6127         spin_lock(&device->io_lock);
6128         if (op_is_sync(bio->bi_opf))
6129                 pending_bios = &device->pending_sync_bios;
6130         else
6131                 pending_bios = &device->pending_bios;
6132
6133         if (pending_bios->tail)
6134                 pending_bios->tail->bi_next = bio;
6135
6136         pending_bios->tail = bio;
6137         if (!pending_bios->head)
6138                 pending_bios->head = bio;
6139         if (device->running_pending)
6140                 should_queue = 0;
6141
6142         spin_unlock(&device->io_lock);
6143
6144         if (should_queue)
6145                 btrfs_queue_work(fs_info->submit_workers, &device->work);
6146 }
6147
6148 static void submit_stripe_bio(struct btrfs_bio *bbio, struct bio *bio,
6149                               u64 physical, int dev_nr, int async)
6150 {
6151         struct btrfs_device *dev = bbio->stripes[dev_nr].dev;
6152         struct btrfs_fs_info *fs_info = bbio->fs_info;
6153
6154         bio->bi_private = bbio;
6155         btrfs_io_bio(bio)->stripe_index = dev_nr;
6156         bio->bi_end_io = btrfs_end_bio;
6157         bio->bi_iter.bi_sector = physical >> 9;
6158 #ifdef DEBUG
6159         {
6160                 struct rcu_string *name;
6161
6162                 rcu_read_lock();
6163                 name = rcu_dereference(dev->name);
6164                 btrfs_debug(fs_info,
6165                         "btrfs_map_bio: rw %d 0x%x, sector=%llu, dev=%lu (%s id %llu), size=%u",
6166                         bio_op(bio), bio->bi_opf,
6167                         (u64)bio->bi_iter.bi_sector,
6168                         (u_long)dev->bdev->bd_dev, name->str, dev->devid,
6169                         bio->bi_iter.bi_size);
6170                 rcu_read_unlock();
6171         }
6172 #endif
6173         bio->bi_bdev = dev->bdev;
6174
6175         btrfs_bio_counter_inc_noblocked(fs_info);
6176
6177         if (async)
6178                 btrfs_schedule_bio(dev, bio);
6179         else
6180                 btrfsic_submit_bio(bio);
6181 }
6182
6183 static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical)
6184 {
6185         atomic_inc(&bbio->error);
6186         if (atomic_dec_and_test(&bbio->stripes_pending)) {
6187                 /* Should be the original bio. */
6188                 WARN_ON(bio != bbio->orig_bio);
6189
6190                 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
6191                 bio->bi_iter.bi_sector = logical >> 9;
6192                 bio->bi_error = -EIO;
6193                 btrfs_end_bbio(bbio, bio);
6194         }
6195 }
6196
6197 int btrfs_map_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
6198                   int mirror_num, int async_submit)
6199 {
6200         struct btrfs_device *dev;
6201         struct bio *first_bio = bio;
6202         u64 logical = (u64)bio->bi_iter.bi_sector << 9;
6203         u64 length = 0;
6204         u64 map_length;
6205         int ret;
6206         int dev_nr;
6207         int total_devs;
6208         struct btrfs_bio *bbio = NULL;
6209
6210         length = bio->bi_iter.bi_size;
6211         map_length = length;
6212
6213         btrfs_bio_counter_inc_blocked(fs_info);
6214         ret = __btrfs_map_block(fs_info, bio_op(bio), logical,
6215                                 &map_length, &bbio, mirror_num, 1);
6216         if (ret) {
6217                 btrfs_bio_counter_dec(fs_info);
6218                 return ret;
6219         }
6220
6221         total_devs = bbio->num_stripes;
6222         bbio->orig_bio = first_bio;
6223         bbio->private = first_bio->bi_private;
6224         bbio->end_io = first_bio->bi_end_io;
6225         bbio->fs_info = fs_info;
6226         atomic_set(&bbio->stripes_pending, bbio->num_stripes);
6227
6228         if ((bbio->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
6229             ((bio_op(bio) == REQ_OP_WRITE) || (mirror_num > 1))) {
6230                 /* In this case, map_length has been set to the length of
6231                    a single stripe; not the whole write */
6232                 if (bio_op(bio) == REQ_OP_WRITE) {
6233                         ret = raid56_parity_write(fs_info, bio, bbio,
6234                                                   map_length);
6235                 } else {
6236                         ret = raid56_parity_recover(fs_info, bio, bbio,
6237                                                     map_length, mirror_num, 1);
6238                 }
6239
6240                 btrfs_bio_counter_dec(fs_info);
6241                 return ret;
6242         }
6243
6244         if (map_length < length) {
6245                 btrfs_crit(fs_info,
6246                            "mapping failed logical %llu bio len %llu len %llu",
6247                            logical, length, map_length);
6248                 BUG();
6249         }
6250
6251         for (dev_nr = 0; dev_nr < total_devs; dev_nr++) {
6252                 dev = bbio->stripes[dev_nr].dev;
6253                 if (!dev || !dev->bdev ||
6254                     (bio_op(first_bio) == REQ_OP_WRITE && !dev->writeable)) {
6255                         bbio_error(bbio, first_bio, logical);
6256                         continue;
6257                 }
6258
6259                 if (dev_nr < total_devs - 1) {
6260                         bio = btrfs_bio_clone(first_bio, GFP_NOFS);
6261                         BUG_ON(!bio); /* -ENOMEM */
6262                 } else
6263                         bio = first_bio;
6264
6265                 submit_stripe_bio(bbio, bio, bbio->stripes[dev_nr].physical,
6266                                   dev_nr, async_submit);
6267         }
6268         btrfs_bio_counter_dec(fs_info);
6269         return 0;
6270 }
6271
6272 struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid,
6273                                        u8 *uuid, u8 *fsid)
6274 {
6275         struct btrfs_device *device;
6276         struct btrfs_fs_devices *cur_devices;
6277
6278         cur_devices = fs_info->fs_devices;
6279         while (cur_devices) {
6280                 if (!fsid ||
6281                     !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
6282                         device = __find_device(&cur_devices->devices,
6283                                                devid, uuid);
6284                         if (device)
6285                                 return device;
6286                 }
6287                 cur_devices = cur_devices->seed;
6288         }
6289         return NULL;
6290 }
6291
6292 static struct btrfs_device *add_missing_dev(struct btrfs_fs_devices *fs_devices,
6293                                             u64 devid, u8 *dev_uuid)
6294 {
6295         struct btrfs_device *device;
6296
6297         device = btrfs_alloc_device(NULL, &devid, dev_uuid);
6298         if (IS_ERR(device))
6299                 return NULL;
6300
6301         list_add(&device->dev_list, &fs_devices->devices);
6302         device->fs_devices = fs_devices;
6303         fs_devices->num_devices++;
6304
6305         device->missing = 1;
6306         fs_devices->missing_devices++;
6307
6308         return device;
6309 }
6310
6311 /**
6312  * btrfs_alloc_device - allocate struct btrfs_device
6313  * @fs_info:    used only for generating a new devid, can be NULL if
6314  *              devid is provided (i.e. @devid != NULL).
6315  * @devid:      a pointer to devid for this device.  If NULL a new devid
6316  *              is generated.
6317  * @uuid:       a pointer to UUID for this device.  If NULL a new UUID
6318  *              is generated.
6319  *
6320  * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
6321  * on error.  Returned struct is not linked onto any lists and can be
6322  * destroyed with kfree() right away.
6323  */
6324 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
6325                                         const u64 *devid,
6326                                         const u8 *uuid)
6327 {
6328         struct btrfs_device *dev;
6329         u64 tmp;
6330
6331         if (WARN_ON(!devid && !fs_info))
6332                 return ERR_PTR(-EINVAL);
6333
6334         dev = __alloc_device();
6335         if (IS_ERR(dev))
6336                 return dev;
6337
6338         if (devid)
6339                 tmp = *devid;
6340         else {
6341                 int ret;
6342
6343                 ret = find_next_devid(fs_info, &tmp);
6344                 if (ret) {
6345                         kfree(dev);
6346                         return ERR_PTR(ret);
6347                 }
6348         }
6349         dev->devid = tmp;
6350
6351         if (uuid)
6352                 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE);
6353         else
6354                 generate_random_uuid(dev->uuid);
6355
6356         btrfs_init_work(&dev->work, btrfs_submit_helper,
6357                         pending_bios_fn, NULL, NULL);
6358
6359         return dev;
6360 }
6361
6362 /* Return -EIO if any error, otherwise return 0. */
6363 static int btrfs_check_chunk_valid(struct btrfs_fs_info *fs_info,
6364                                    struct extent_buffer *leaf,
6365                                    struct btrfs_chunk *chunk, u64 logical)
6366 {
6367         u64 length;
6368         u64 stripe_len;
6369         u16 num_stripes;
6370         u16 sub_stripes;
6371         u64 type;
6372
6373         length = btrfs_chunk_length(leaf, chunk);
6374         stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6375         num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6376         sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
6377         type = btrfs_chunk_type(leaf, chunk);
6378
6379         if (!num_stripes) {
6380                 btrfs_err(fs_info, "invalid chunk num_stripes: %u",
6381                           num_stripes);
6382                 return -EIO;
6383         }
6384         if (!IS_ALIGNED(logical, fs_info->sectorsize)) {
6385                 btrfs_err(fs_info, "invalid chunk logical %llu", logical);
6386                 return -EIO;
6387         }
6388         if (btrfs_chunk_sector_size(leaf, chunk) != fs_info->sectorsize) {
6389                 btrfs_err(fs_info, "invalid chunk sectorsize %u",
6390                           btrfs_chunk_sector_size(leaf, chunk));
6391                 return -EIO;
6392         }
6393         if (!length || !IS_ALIGNED(length, fs_info->sectorsize)) {
6394                 btrfs_err(fs_info, "invalid chunk length %llu", length);
6395                 return -EIO;
6396         }
6397         if (!is_power_of_2(stripe_len) || stripe_len != BTRFS_STRIPE_LEN) {
6398                 btrfs_err(fs_info, "invalid chunk stripe length: %llu",
6399                           stripe_len);
6400                 return -EIO;
6401         }
6402         if (~(BTRFS_BLOCK_GROUP_TYPE_MASK | BTRFS_BLOCK_GROUP_PROFILE_MASK) &
6403             type) {
6404                 btrfs_err(fs_info, "unrecognized chunk type: %llu",
6405                           ~(BTRFS_BLOCK_GROUP_TYPE_MASK |
6406                             BTRFS_BLOCK_GROUP_PROFILE_MASK) &
6407                           btrfs_chunk_type(leaf, chunk));
6408                 return -EIO;
6409         }
6410         if ((type & BTRFS_BLOCK_GROUP_RAID10 && sub_stripes != 2) ||
6411             (type & BTRFS_BLOCK_GROUP_RAID1 && num_stripes < 1) ||
6412             (type & BTRFS_BLOCK_GROUP_RAID5 && num_stripes < 2) ||
6413             (type & BTRFS_BLOCK_GROUP_RAID6 && num_stripes < 3) ||
6414             (type & BTRFS_BLOCK_GROUP_DUP && num_stripes > 2) ||
6415             ((type & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 &&
6416              num_stripes != 1)) {
6417                 btrfs_err(fs_info,
6418                         "invalid num_stripes:sub_stripes %u:%u for profile %llu",
6419                         num_stripes, sub_stripes,
6420                         type & BTRFS_BLOCK_GROUP_PROFILE_MASK);
6421                 return -EIO;
6422         }
6423
6424         return 0;
6425 }
6426
6427 static int read_one_chunk(struct btrfs_fs_info *fs_info, struct btrfs_key *key,
6428                           struct extent_buffer *leaf,
6429                           struct btrfs_chunk *chunk)
6430 {
6431         struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
6432         struct map_lookup *map;
6433         struct extent_map *em;
6434         u64 logical;
6435         u64 length;
6436         u64 stripe_len;
6437         u64 devid;
6438         u8 uuid[BTRFS_UUID_SIZE];
6439         int num_stripes;
6440         int ret;
6441         int i;
6442
6443         logical = key->offset;
6444         length = btrfs_chunk_length(leaf, chunk);
6445         stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6446         num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6447
6448         ret = btrfs_check_chunk_valid(fs_info, leaf, chunk, logical);
6449         if (ret)
6450                 return ret;
6451
6452         read_lock(&map_tree->map_tree.lock);
6453         em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
6454         read_unlock(&map_tree->map_tree.lock);
6455
6456         /* already mapped? */
6457         if (em && em->start <= logical && em->start + em->len > logical) {
6458                 free_extent_map(em);
6459                 return 0;
6460         } else if (em) {
6461                 free_extent_map(em);
6462         }
6463
6464         em = alloc_extent_map();
6465         if (!em)
6466                 return -ENOMEM;
6467         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
6468         if (!map) {
6469                 free_extent_map(em);
6470                 return -ENOMEM;
6471         }
6472
6473         set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
6474         em->map_lookup = map;
6475         em->start = logical;
6476         em->len = length;
6477         em->orig_start = 0;
6478         em->block_start = 0;
6479         em->block_len = em->len;
6480
6481         map->num_stripes = num_stripes;
6482         map->io_width = btrfs_chunk_io_width(leaf, chunk);
6483         map->io_align = btrfs_chunk_io_align(leaf, chunk);
6484         map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
6485         map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6486         map->type = btrfs_chunk_type(leaf, chunk);
6487         map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
6488         for (i = 0; i < num_stripes; i++) {
6489                 map->stripes[i].physical =
6490                         btrfs_stripe_offset_nr(leaf, chunk, i);
6491                 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
6492                 read_extent_buffer(leaf, uuid, (unsigned long)
6493                                    btrfs_stripe_dev_uuid_nr(chunk, i),
6494                                    BTRFS_UUID_SIZE);
6495                 map->stripes[i].dev = btrfs_find_device(fs_info, devid,
6496                                                         uuid, NULL);
6497                 if (!map->stripes[i].dev &&
6498                     !btrfs_test_opt(fs_info, DEGRADED)) {
6499                         free_extent_map(em);
6500                         return -EIO;
6501                 }
6502                 if (!map->stripes[i].dev) {
6503                         map->stripes[i].dev =
6504                                 add_missing_dev(fs_info->fs_devices, devid,
6505                                                 uuid);
6506                         if (!map->stripes[i].dev) {
6507                                 free_extent_map(em);
6508                                 return -EIO;
6509                         }
6510                         btrfs_warn(fs_info, "devid %llu uuid %pU is missing",
6511                                    devid, uuid);
6512                 }
6513                 map->stripes[i].dev->in_fs_metadata = 1;
6514         }
6515
6516         write_lock(&map_tree->map_tree.lock);
6517         ret = add_extent_mapping(&map_tree->map_tree, em, 0);
6518         write_unlock(&map_tree->map_tree.lock);
6519         BUG_ON(ret); /* Tree corruption */
6520         free_extent_map(em);
6521
6522         return 0;
6523 }
6524
6525 static void fill_device_from_item(struct extent_buffer *leaf,
6526                                  struct btrfs_dev_item *dev_item,
6527                                  struct btrfs_device *device)
6528 {
6529         unsigned long ptr;
6530
6531         device->devid = btrfs_device_id(leaf, dev_item);
6532         device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
6533         device->total_bytes = device->disk_total_bytes;
6534         device->commit_total_bytes = device->disk_total_bytes;
6535         device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
6536         device->commit_bytes_used = device->bytes_used;
6537         device->type = btrfs_device_type(leaf, dev_item);
6538         device->io_align = btrfs_device_io_align(leaf, dev_item);
6539         device->io_width = btrfs_device_io_width(leaf, dev_item);
6540         device->sector_size = btrfs_device_sector_size(leaf, dev_item);
6541         WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
6542         device->is_tgtdev_for_dev_replace = 0;
6543
6544         ptr = btrfs_device_uuid(dev_item);
6545         read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
6546 }
6547
6548 static struct btrfs_fs_devices *open_seed_devices(struct btrfs_fs_info *fs_info,
6549                                                   u8 *fsid)
6550 {
6551         struct btrfs_fs_devices *fs_devices;
6552         int ret;
6553
6554         BUG_ON(!mutex_is_locked(&uuid_mutex));
6555
6556         fs_devices = fs_info->fs_devices->seed;
6557         while (fs_devices) {
6558                 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE))
6559                         return fs_devices;
6560
6561                 fs_devices = fs_devices->seed;
6562         }
6563
6564         fs_devices = find_fsid(fsid);
6565         if (!fs_devices) {
6566                 if (!btrfs_test_opt(fs_info, DEGRADED))
6567                         return ERR_PTR(-ENOENT);
6568
6569                 fs_devices = alloc_fs_devices(fsid);
6570                 if (IS_ERR(fs_devices))
6571                         return fs_devices;
6572
6573                 fs_devices->seeding = 1;
6574                 fs_devices->opened = 1;
6575                 return fs_devices;
6576         }
6577
6578         fs_devices = clone_fs_devices(fs_devices);
6579         if (IS_ERR(fs_devices))
6580                 return fs_devices;
6581
6582         ret = __btrfs_open_devices(fs_devices, FMODE_READ,
6583                                    fs_info->bdev_holder);
6584         if (ret) {
6585                 free_fs_devices(fs_devices);
6586                 fs_devices = ERR_PTR(ret);
6587                 goto out;
6588         }
6589
6590         if (!fs_devices->seeding) {
6591                 __btrfs_close_devices(fs_devices);
6592                 free_fs_devices(fs_devices);
6593                 fs_devices = ERR_PTR(-EINVAL);
6594                 goto out;
6595         }
6596
6597         fs_devices->seed = fs_info->fs_devices->seed;
6598         fs_info->fs_devices->seed = fs_devices;
6599 out:
6600         return fs_devices;
6601 }
6602
6603 static int read_one_dev(struct btrfs_fs_info *fs_info,
6604                         struct extent_buffer *leaf,
6605                         struct btrfs_dev_item *dev_item)
6606 {
6607         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6608         struct btrfs_device *device;
6609         u64 devid;
6610         int ret;
6611         u8 fs_uuid[BTRFS_UUID_SIZE];
6612         u8 dev_uuid[BTRFS_UUID_SIZE];
6613
6614         devid = btrfs_device_id(leaf, dev_item);
6615         read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
6616                            BTRFS_UUID_SIZE);
6617         read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
6618                            BTRFS_UUID_SIZE);
6619
6620         if (memcmp(fs_uuid, fs_info->fsid, BTRFS_UUID_SIZE)) {
6621                 fs_devices = open_seed_devices(fs_info, fs_uuid);
6622                 if (IS_ERR(fs_devices))
6623                         return PTR_ERR(fs_devices);
6624         }
6625
6626         device = btrfs_find_device(fs_info, devid, dev_uuid, fs_uuid);
6627         if (!device) {
6628                 if (!btrfs_test_opt(fs_info, DEGRADED))
6629                         return -EIO;
6630
6631                 device = add_missing_dev(fs_devices, devid, dev_uuid);
6632                 if (!device)
6633                         return -ENOMEM;
6634                 btrfs_warn(fs_info, "devid %llu uuid %pU missing",
6635                                 devid, dev_uuid);
6636         } else {
6637                 if (!device->bdev && !btrfs_test_opt(fs_info, DEGRADED))
6638                         return -EIO;
6639
6640                 if(!device->bdev && !device->missing) {
6641                         /*
6642                          * this happens when a device that was properly setup
6643                          * in the device info lists suddenly goes bad.
6644                          * device->bdev is NULL, and so we have to set
6645                          * device->missing to one here
6646                          */
6647                         device->fs_devices->missing_devices++;
6648                         device->missing = 1;
6649                 }
6650
6651                 /* Move the device to its own fs_devices */
6652                 if (device->fs_devices != fs_devices) {
6653                         ASSERT(device->missing);
6654
6655                         list_move(&device->dev_list, &fs_devices->devices);
6656                         device->fs_devices->num_devices--;
6657                         fs_devices->num_devices++;
6658
6659                         device->fs_devices->missing_devices--;
6660                         fs_devices->missing_devices++;
6661
6662                         device->fs_devices = fs_devices;
6663                 }
6664         }
6665
6666         if (device->fs_devices != fs_info->fs_devices) {
6667                 BUG_ON(device->writeable);
6668                 if (device->generation !=
6669                     btrfs_device_generation(leaf, dev_item))
6670                         return -EINVAL;
6671         }
6672
6673         fill_device_from_item(leaf, dev_item, device);
6674         device->in_fs_metadata = 1;
6675         if (device->writeable && !device->is_tgtdev_for_dev_replace) {
6676                 device->fs_devices->total_rw_bytes += device->total_bytes;
6677                 atomic64_add(device->total_bytes - device->bytes_used,
6678                                 &fs_info->free_chunk_space);
6679         }
6680         ret = 0;
6681         return ret;
6682 }
6683
6684 int btrfs_read_sys_array(struct btrfs_fs_info *fs_info)
6685 {
6686         struct btrfs_root *root = fs_info->tree_root;
6687         struct btrfs_super_block *super_copy = fs_info->super_copy;
6688         struct extent_buffer *sb;
6689         struct btrfs_disk_key *disk_key;
6690         struct btrfs_chunk *chunk;
6691         u8 *array_ptr;
6692         unsigned long sb_array_offset;
6693         int ret = 0;
6694         u32 num_stripes;
6695         u32 array_size;
6696         u32 len = 0;
6697         u32 cur_offset;
6698         u64 type;
6699         struct btrfs_key key;
6700
6701         ASSERT(BTRFS_SUPER_INFO_SIZE <= fs_info->nodesize);
6702         /*
6703          * This will create extent buffer of nodesize, superblock size is
6704          * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will
6705          * overallocate but we can keep it as-is, only the first page is used.
6706          */
6707         sb = btrfs_find_create_tree_block(fs_info, BTRFS_SUPER_INFO_OFFSET);
6708         if (IS_ERR(sb))
6709                 return PTR_ERR(sb);
6710         set_extent_buffer_uptodate(sb);
6711         btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
6712         /*
6713          * The sb extent buffer is artificial and just used to read the system array.
6714          * set_extent_buffer_uptodate() call does not properly mark all it's
6715          * pages up-to-date when the page is larger: extent does not cover the
6716          * whole page and consequently check_page_uptodate does not find all
6717          * the page's extents up-to-date (the hole beyond sb),
6718          * write_extent_buffer then triggers a WARN_ON.
6719          *
6720          * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
6721          * but sb spans only this function. Add an explicit SetPageUptodate call
6722          * to silence the warning eg. on PowerPC 64.
6723          */
6724         if (PAGE_SIZE > BTRFS_SUPER_INFO_SIZE)
6725                 SetPageUptodate(sb->pages[0]);
6726
6727         write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
6728         array_size = btrfs_super_sys_array_size(super_copy);
6729
6730         array_ptr = super_copy->sys_chunk_array;
6731         sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array);
6732         cur_offset = 0;
6733
6734         while (cur_offset < array_size) {
6735                 disk_key = (struct btrfs_disk_key *)array_ptr;
6736                 len = sizeof(*disk_key);
6737                 if (cur_offset + len > array_size)
6738                         goto out_short_read;
6739
6740                 btrfs_disk_key_to_cpu(&key, disk_key);
6741
6742                 array_ptr += len;
6743                 sb_array_offset += len;
6744                 cur_offset += len;
6745
6746                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
6747                         chunk = (struct btrfs_chunk *)sb_array_offset;
6748                         /*
6749                          * At least one btrfs_chunk with one stripe must be
6750                          * present, exact stripe count check comes afterwards
6751                          */
6752                         len = btrfs_chunk_item_size(1);
6753                         if (cur_offset + len > array_size)
6754                                 goto out_short_read;
6755
6756                         num_stripes = btrfs_chunk_num_stripes(sb, chunk);
6757                         if (!num_stripes) {
6758                                 btrfs_err(fs_info,
6759                                         "invalid number of stripes %u in sys_array at offset %u",
6760                                         num_stripes, cur_offset);
6761                                 ret = -EIO;
6762                                 break;
6763                         }
6764
6765                         type = btrfs_chunk_type(sb, chunk);
6766                         if ((type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) {
6767                                 btrfs_err(fs_info,
6768                             "invalid chunk type %llu in sys_array at offset %u",
6769                                         type, cur_offset);
6770                                 ret = -EIO;
6771                                 break;
6772                         }
6773
6774                         len = btrfs_chunk_item_size(num_stripes);
6775                         if (cur_offset + len > array_size)
6776                                 goto out_short_read;
6777
6778                         ret = read_one_chunk(fs_info, &key, sb, chunk);
6779                         if (ret)
6780                                 break;
6781                 } else {
6782                         btrfs_err(fs_info,
6783                             "unexpected item type %u in sys_array at offset %u",
6784                                   (u32)key.type, cur_offset);
6785                         ret = -EIO;
6786                         break;
6787                 }
6788                 array_ptr += len;
6789                 sb_array_offset += len;
6790                 cur_offset += len;
6791         }
6792         clear_extent_buffer_uptodate(sb);
6793         free_extent_buffer_stale(sb);
6794         return ret;
6795
6796 out_short_read:
6797         btrfs_err(fs_info, "sys_array too short to read %u bytes at offset %u",
6798                         len, cur_offset);
6799         clear_extent_buffer_uptodate(sb);
6800         free_extent_buffer_stale(sb);
6801         return -EIO;
6802 }
6803
6804 int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info)
6805 {
6806         struct btrfs_root *root = fs_info->chunk_root;
6807         struct btrfs_path *path;
6808         struct extent_buffer *leaf;
6809         struct btrfs_key key;
6810         struct btrfs_key found_key;
6811         int ret;
6812         int slot;
6813         u64 total_dev = 0;
6814
6815         path = btrfs_alloc_path();
6816         if (!path)
6817                 return -ENOMEM;
6818
6819         mutex_lock(&uuid_mutex);
6820         mutex_lock(&fs_info->chunk_mutex);
6821
6822         /*
6823          * Read all device items, and then all the chunk items. All
6824          * device items are found before any chunk item (their object id
6825          * is smaller than the lowest possible object id for a chunk
6826          * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
6827          */
6828         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
6829         key.offset = 0;
6830         key.type = 0;
6831         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6832         if (ret < 0)
6833                 goto error;
6834         while (1) {
6835                 leaf = path->nodes[0];
6836                 slot = path->slots[0];
6837                 if (slot >= btrfs_header_nritems(leaf)) {
6838                         ret = btrfs_next_leaf(root, path);
6839                         if (ret == 0)
6840                                 continue;
6841                         if (ret < 0)
6842                                 goto error;
6843                         break;
6844                 }
6845                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6846                 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
6847                         struct btrfs_dev_item *dev_item;
6848                         dev_item = btrfs_item_ptr(leaf, slot,
6849                                                   struct btrfs_dev_item);
6850                         ret = read_one_dev(fs_info, leaf, dev_item);
6851                         if (ret)
6852                                 goto error;
6853                         total_dev++;
6854                 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
6855                         struct btrfs_chunk *chunk;
6856                         chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
6857                         ret = read_one_chunk(fs_info, &found_key, leaf, chunk);
6858                         if (ret)
6859                                 goto error;
6860                 }
6861                 path->slots[0]++;
6862         }
6863
6864         /*
6865          * After loading chunk tree, we've got all device information,
6866          * do another round of validation checks.
6867          */
6868         if (total_dev != fs_info->fs_devices->total_devices) {
6869                 btrfs_err(fs_info,
6870            "super_num_devices %llu mismatch with num_devices %llu found here",
6871                           btrfs_super_num_devices(fs_info->super_copy),
6872                           total_dev);
6873                 ret = -EINVAL;
6874                 goto error;
6875         }
6876         if (btrfs_super_total_bytes(fs_info->super_copy) <
6877             fs_info->fs_devices->total_rw_bytes) {
6878                 btrfs_err(fs_info,
6879         "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu",
6880                           btrfs_super_total_bytes(fs_info->super_copy),
6881                           fs_info->fs_devices->total_rw_bytes);
6882                 ret = -EINVAL;
6883                 goto error;
6884         }
6885         ret = 0;
6886 error:
6887         mutex_unlock(&fs_info->chunk_mutex);
6888         mutex_unlock(&uuid_mutex);
6889
6890         btrfs_free_path(path);
6891         return ret;
6892 }
6893
6894 void btrfs_init_devices_late(struct btrfs_fs_info *fs_info)
6895 {
6896         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6897         struct btrfs_device *device;
6898
6899         while (fs_devices) {
6900                 mutex_lock(&fs_devices->device_list_mutex);
6901                 list_for_each_entry(device, &fs_devices->devices, dev_list)
6902                         device->fs_info = fs_info;
6903                 mutex_unlock(&fs_devices->device_list_mutex);
6904
6905                 fs_devices = fs_devices->seed;
6906         }
6907 }
6908
6909 static void __btrfs_reset_dev_stats(struct btrfs_device *dev)
6910 {
6911         int i;
6912
6913         for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6914                 btrfs_dev_stat_reset(dev, i);
6915 }
6916
6917 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
6918 {
6919         struct btrfs_key key;
6920         struct btrfs_key found_key;
6921         struct btrfs_root *dev_root = fs_info->dev_root;
6922         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6923         struct extent_buffer *eb;
6924         int slot;
6925         int ret = 0;
6926         struct btrfs_device *device;
6927         struct btrfs_path *path = NULL;
6928         int i;
6929
6930         path = btrfs_alloc_path();
6931         if (!path) {
6932                 ret = -ENOMEM;
6933                 goto out;
6934         }
6935
6936         mutex_lock(&fs_devices->device_list_mutex);
6937         list_for_each_entry(device, &fs_devices->devices, dev_list) {
6938                 int item_size;
6939                 struct btrfs_dev_stats_item *ptr;
6940
6941                 key.objectid = BTRFS_DEV_STATS_OBJECTID;
6942                 key.type = BTRFS_PERSISTENT_ITEM_KEY;
6943                 key.offset = device->devid;
6944                 ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
6945                 if (ret) {
6946                         __btrfs_reset_dev_stats(device);
6947                         device->dev_stats_valid = 1;
6948                         btrfs_release_path(path);
6949                         continue;
6950                 }
6951                 slot = path->slots[0];
6952                 eb = path->nodes[0];
6953                 btrfs_item_key_to_cpu(eb, &found_key, slot);
6954                 item_size = btrfs_item_size_nr(eb, slot);
6955
6956                 ptr = btrfs_item_ptr(eb, slot,
6957                                      struct btrfs_dev_stats_item);
6958
6959                 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
6960                         if (item_size >= (1 + i) * sizeof(__le64))
6961                                 btrfs_dev_stat_set(device, i,
6962                                         btrfs_dev_stats_value(eb, ptr, i));
6963                         else
6964                                 btrfs_dev_stat_reset(device, i);
6965                 }
6966
6967                 device->dev_stats_valid = 1;
6968                 btrfs_dev_stat_print_on_load(device);
6969                 btrfs_release_path(path);
6970         }
6971         mutex_unlock(&fs_devices->device_list_mutex);
6972
6973 out:
6974         btrfs_free_path(path);
6975         return ret < 0 ? ret : 0;
6976 }
6977
6978 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
6979                                 struct btrfs_fs_info *fs_info,
6980                                 struct btrfs_device *device)
6981 {
6982         struct btrfs_root *dev_root = fs_info->dev_root;
6983         struct btrfs_path *path;
6984         struct btrfs_key key;
6985         struct extent_buffer *eb;
6986         struct btrfs_dev_stats_item *ptr;
6987         int ret;
6988         int i;
6989
6990         key.objectid = BTRFS_DEV_STATS_OBJECTID;
6991         key.type = BTRFS_PERSISTENT_ITEM_KEY;
6992         key.offset = device->devid;
6993
6994         path = btrfs_alloc_path();
6995         if (!path)
6996                 return -ENOMEM;
6997         ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
6998         if (ret < 0) {
6999                 btrfs_warn_in_rcu(fs_info,
7000                         "error %d while searching for dev_stats item for device %s",
7001                               ret, rcu_str_deref(device->name));
7002                 goto out;
7003         }
7004
7005         if (ret == 0 &&
7006             btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
7007                 /* need to delete old one and insert a new one */
7008                 ret = btrfs_del_item(trans, dev_root, path);
7009                 if (ret != 0) {
7010                         btrfs_warn_in_rcu(fs_info,
7011                                 "delete too small dev_stats item for device %s failed %d",
7012                                       rcu_str_deref(device->name), ret);
7013                         goto out;
7014                 }
7015                 ret = 1;
7016         }
7017
7018         if (ret == 1) {
7019                 /* need to insert a new item */
7020                 btrfs_release_path(path);
7021                 ret = btrfs_insert_empty_item(trans, dev_root, path,
7022                                               &key, sizeof(*ptr));
7023                 if (ret < 0) {
7024                         btrfs_warn_in_rcu(fs_info,
7025                                 "insert dev_stats item for device %s failed %d",
7026                                 rcu_str_deref(device->name), ret);
7027                         goto out;
7028                 }
7029         }
7030
7031         eb = path->nodes[0];
7032         ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
7033         for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7034                 btrfs_set_dev_stats_value(eb, ptr, i,
7035                                           btrfs_dev_stat_read(device, i));
7036         btrfs_mark_buffer_dirty(eb);
7037
7038 out:
7039         btrfs_free_path(path);
7040         return ret;
7041 }
7042
7043 /*
7044  * called from commit_transaction. Writes all changed device stats to disk.
7045  */
7046 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans,
7047                         struct btrfs_fs_info *fs_info)
7048 {
7049         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7050         struct btrfs_device *device;
7051         int stats_cnt;
7052         int ret = 0;
7053
7054         mutex_lock(&fs_devices->device_list_mutex);
7055         list_for_each_entry(device, &fs_devices->devices, dev_list) {
7056                 if (!device->dev_stats_valid || !btrfs_dev_stats_dirty(device))
7057                         continue;
7058
7059                 stats_cnt = atomic_read(&device->dev_stats_ccnt);
7060                 ret = update_dev_stat_item(trans, fs_info, device);
7061                 if (!ret)
7062                         atomic_sub(stats_cnt, &device->dev_stats_ccnt);
7063         }
7064         mutex_unlock(&fs_devices->device_list_mutex);
7065
7066         return ret;
7067 }
7068
7069 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
7070 {
7071         btrfs_dev_stat_inc(dev, index);
7072         btrfs_dev_stat_print_on_error(dev);
7073 }
7074
7075 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
7076 {
7077         if (!dev->dev_stats_valid)
7078                 return;
7079         btrfs_err_rl_in_rcu(dev->fs_info,
7080                 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7081                            rcu_str_deref(dev->name),
7082                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7083                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7084                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7085                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7086                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7087 }
7088
7089 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
7090 {
7091         int i;
7092
7093         for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7094                 if (btrfs_dev_stat_read(dev, i) != 0)
7095                         break;
7096         if (i == BTRFS_DEV_STAT_VALUES_MAX)
7097                 return; /* all values == 0, suppress message */
7098
7099         btrfs_info_in_rcu(dev->fs_info,
7100                 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7101                rcu_str_deref(dev->name),
7102                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7103                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7104                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7105                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7106                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7107 }
7108
7109 int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info,
7110                         struct btrfs_ioctl_get_dev_stats *stats)
7111 {
7112         struct btrfs_device *dev;
7113         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7114         int i;
7115
7116         mutex_lock(&fs_devices->device_list_mutex);
7117         dev = btrfs_find_device(fs_info, stats->devid, NULL, NULL);
7118         mutex_unlock(&fs_devices->device_list_mutex);
7119
7120         if (!dev) {
7121                 btrfs_warn(fs_info, "get dev_stats failed, device not found");
7122                 return -ENODEV;
7123         } else if (!dev->dev_stats_valid) {
7124                 btrfs_warn(fs_info, "get dev_stats failed, not yet valid");
7125                 return -ENODEV;
7126         } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
7127                 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7128                         if (stats->nr_items > i)
7129                                 stats->values[i] =
7130                                         btrfs_dev_stat_read_and_reset(dev, i);
7131                         else
7132                                 btrfs_dev_stat_reset(dev, i);
7133                 }
7134         } else {
7135                 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7136                         if (stats->nr_items > i)
7137                                 stats->values[i] = btrfs_dev_stat_read(dev, i);
7138         }
7139         if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
7140                 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
7141         return 0;
7142 }
7143
7144 void btrfs_scratch_superblocks(struct block_device *bdev, const char *device_path)
7145 {
7146         struct buffer_head *bh;
7147         struct btrfs_super_block *disk_super;
7148         int copy_num;
7149
7150         if (!bdev)
7151                 return;
7152
7153         for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX;
7154                 copy_num++) {
7155
7156                 if (btrfs_read_dev_one_super(bdev, copy_num, &bh))
7157                         continue;
7158
7159                 disk_super = (struct btrfs_super_block *)bh->b_data;
7160
7161                 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
7162                 set_buffer_dirty(bh);
7163                 sync_dirty_buffer(bh);
7164                 brelse(bh);
7165         }
7166
7167         /* Notify udev that device has changed */
7168         btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
7169
7170         /* Update ctime/mtime for device path for libblkid */
7171         update_dev_time(device_path);
7172 }
7173
7174 /*
7175  * Update the size of all devices, which is used for writing out the
7176  * super blocks.
7177  */
7178 void btrfs_update_commit_device_size(struct btrfs_fs_info *fs_info)
7179 {
7180         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7181         struct btrfs_device *curr, *next;
7182
7183         if (list_empty(&fs_devices->resized_devices))
7184                 return;
7185
7186         mutex_lock(&fs_devices->device_list_mutex);
7187         mutex_lock(&fs_info->chunk_mutex);
7188         list_for_each_entry_safe(curr, next, &fs_devices->resized_devices,
7189                                  resized_list) {
7190                 list_del_init(&curr->resized_list);
7191                 curr->commit_total_bytes = curr->disk_total_bytes;
7192         }
7193         mutex_unlock(&fs_info->chunk_mutex);
7194         mutex_unlock(&fs_devices->device_list_mutex);
7195 }
7196
7197 /* Must be invoked during the transaction commit */
7198 void btrfs_update_commit_device_bytes_used(struct btrfs_fs_info *fs_info,
7199                                         struct btrfs_transaction *transaction)
7200 {
7201         struct extent_map *em;
7202         struct map_lookup *map;
7203         struct btrfs_device *dev;
7204         int i;
7205
7206         if (list_empty(&transaction->pending_chunks))
7207                 return;
7208
7209         /* In order to kick the device replace finish process */
7210         mutex_lock(&fs_info->chunk_mutex);
7211         list_for_each_entry(em, &transaction->pending_chunks, list) {
7212                 map = em->map_lookup;
7213
7214                 for (i = 0; i < map->num_stripes; i++) {
7215                         dev = map->stripes[i].dev;
7216                         dev->commit_bytes_used = dev->bytes_used;
7217                 }
7218         }
7219         mutex_unlock(&fs_info->chunk_mutex);
7220 }
7221
7222 void btrfs_set_fs_info_ptr(struct btrfs_fs_info *fs_info)
7223 {
7224         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7225         while (fs_devices) {
7226                 fs_devices->fs_info = fs_info;
7227                 fs_devices = fs_devices->seed;
7228         }
7229 }
7230
7231 void btrfs_reset_fs_info_ptr(struct btrfs_fs_info *fs_info)
7232 {
7233         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7234         while (fs_devices) {
7235                 fs_devices->fs_info = NULL;
7236                 fs_devices = fs_devices->seed;
7237         }
7238 }