fs/btrfs/volumes.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18 #include <linux/sched.h>
  19 #include <linux/bio.h>
  20 #include <linux/slab.h>
  21 #include <linux/buffer_head.h>
  22 #include <linux/blkdev.h>
  23 #include <linux/iocontext.h>
  24 #include <linux/capability.h>
  25 #include <linux/ratelimit.h>
  26 #include <linux/kthread.h>
  27 #include <linux/raid/pq.h>
  28 #include <linux/semaphore.h>
  29 #include <linux/uuid.h>
  30 #include <asm/div64.h>
  31 #include "ctree.h"
  32 #include "extent_map.h"
  33 #include "disk-io.h"
  34 #include "transaction.h"
  35 #include "print-tree.h"
  36 #include "volumes.h"
  37 #include "raid56.h"
  38 #include "async-thread.h"
  39 #include "check-integrity.h"
  40 #include "rcu-string.h"
  41 #include "math.h"
  42 #include "dev-replace.h"
  43 #include "sysfs.h"
  44
  45 const struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = {
  46         [BTRFS_RAID_RAID10] = {
  47                 .sub_stripes    = 2,
  48                 .dev_stripes    = 1,
  49                 .devs_max       = 0,    /* 0 == as many as possible */
  50                 .devs_min       = 4,
  51                 .tolerated_failures = 1,
  52                 .devs_increment = 2,
  53                 .ncopies        = 2,
  54         },
  55         [BTRFS_RAID_RAID1] = {
  56                 .sub_stripes    = 1,
  57                 .dev_stripes    = 1,
  58                 .devs_max       = 2,
  59                 .devs_min       = 2,
  60                 .tolerated_failures = 1,
  61                 .devs_increment = 2,
  62                 .ncopies        = 2,
  63         },
  64         [BTRFS_RAID_DUP] = {
  65                 .sub_stripes    = 1,
  66                 .dev_stripes    = 2,
  67                 .devs_max       = 1,
  68                 .devs_min       = 1,
  69                 .tolerated_failures = 0,
  70                 .devs_increment = 1,
  71                 .ncopies        = 2,
  72         },
  73         [BTRFS_RAID_RAID0] = {
  74                 .sub_stripes    = 1,
  75                 .dev_stripes    = 1,
  76                 .devs_max       = 0,
  77                 .devs_min       = 2,
  78                 .tolerated_failures = 0,
  79                 .devs_increment = 1,
  80                 .ncopies        = 1,
  81         },
  82         [BTRFS_RAID_SINGLE] = {
  83                 .sub_stripes    = 1,
  84                 .dev_stripes    = 1,
  85                 .devs_max       = 1,
  86                 .devs_min       = 1,
  87                 .tolerated_failures = 0,
  88                 .devs_increment = 1,
  89                 .ncopies        = 1,
  90         },
  91         [BTRFS_RAID_RAID5] = {
  92                 .sub_stripes    = 1,
  93                 .dev_stripes    = 1,
  94                 .devs_max       = 0,
  95                 .devs_min       = 2,
  96                 .tolerated_failures = 1,
  97                 .devs_increment = 1,
  98                 .ncopies        = 2,
  99         },
 100         [BTRFS_RAID_RAID6] = {
 101                 .sub_stripes    = 1,
 102                 .dev_stripes    = 1,
 103                 .devs_max       = 0,
 104                 .devs_min       = 3,
 105                 .tolerated_failures = 2,
 106                 .devs_increment = 1,
 107                 .ncopies        = 3,
 108         },
 109 };
 110
 111 const u64 btrfs_raid_group[BTRFS_NR_RAID_TYPES] = {
 112         [BTRFS_RAID_RAID10] = BTRFS_BLOCK_GROUP_RAID10,
 113         [BTRFS_RAID_RAID1]  = BTRFS_BLOCK_GROUP_RAID1,
 114         [BTRFS_RAID_DUP]    = BTRFS_BLOCK_GROUP_DUP,
 115         [BTRFS_RAID_RAID0]  = BTRFS_BLOCK_GROUP_RAID0,
 116         [BTRFS_RAID_SINGLE] = 0,
 117         [BTRFS_RAID_RAID5]  = BTRFS_BLOCK_GROUP_RAID5,
 118         [BTRFS_RAID_RAID6]  = BTRFS_BLOCK_GROUP_RAID6,
 119 };
 120
 121 /*
 122  * Table to convert BTRFS_RAID_* to the error code if minimum number of devices
 123  * condition is not met. Zero means there's no corresponding
 124  * BTRFS_ERROR_DEV_*_NOT_MET value.
 125  */
 126 const int btrfs_raid_mindev_error[BTRFS_NR_RAID_TYPES] = {
 127         [BTRFS_RAID_RAID10] = BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET,
 128         [BTRFS_RAID_RAID1]  = BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET,
 129         [BTRFS_RAID_DUP]    = 0,
 130         [BTRFS_RAID_RAID0]  = 0,
 131         [BTRFS_RAID_SINGLE] = 0,
 132         [BTRFS_RAID_RAID5]  = BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET,
 133         [BTRFS_RAID_RAID6]  = BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET,
 134 };
 135
 136 static int init_first_rw_device(struct btrfs_trans_handle *trans,
 137                                 struct btrfs_root *root,
 138                                 struct btrfs_device *device);
 139 static int btrfs_relocate_sys_chunks(struct btrfs_root *root);
 140 static void __btrfs_reset_dev_stats(struct btrfs_device *dev);
 141 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev);
 142 static void btrfs_dev_stat_print_on_load(struct btrfs_device *device);
 143
 144 DEFINE_MUTEX(uuid_mutex);
 145 static LIST_HEAD(fs_uuids);
 146 struct list_head *btrfs_get_fs_uuids(void)
 147 {
 148         return &fs_uuids;
 149 }
 150
 151 static struct btrfs_fs_devices *__alloc_fs_devices(void)
 152 {
 153         struct btrfs_fs_devices *fs_devs;
 154
 155         fs_devs = kzalloc(sizeof(*fs_devs), GFP_KERNEL);
 156         if (!fs_devs)
 157                 return ERR_PTR(-ENOMEM);
 158
 159         mutex_init(&fs_devs->device_list_mutex);
 160
 161         INIT_LIST_HEAD(&fs_devs->devices);
 162         INIT_LIST_HEAD(&fs_devs->resized_devices);
 163         INIT_LIST_HEAD(&fs_devs->alloc_list);
 164         INIT_LIST_HEAD(&fs_devs->list);
 165
 166         return fs_devs;
 167 }
 168
 169 /**
 170  * alloc_fs_devices - allocate struct btrfs_fs_devices
 171  * @fsid:       a pointer to UUID for this FS.  If NULL a new UUID is
 172  *              generated.
 173  *
 174  * Return: a pointer to a new &struct btrfs_fs_devices on success;
 175  * ERR_PTR() on error.  Returned struct is not linked onto any lists and
 176  * can be destroyed with kfree() right away.
 177  */
 178 static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid)
 179 {
 180         struct btrfs_fs_devices *fs_devs;
 181
 182         fs_devs = __alloc_fs_devices();
 183         if (IS_ERR(fs_devs))
 184                 return fs_devs;
 185
 186         if (fsid)
 187                 memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE);
 188         else
 189                 generate_random_uuid(fs_devs->fsid);
 190
 191         return fs_devs;
 192 }
 193
 194 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
 195 {
 196         struct btrfs_device *device;
 197         WARN_ON(fs_devices->opened);
 198         while (!list_empty(&fs_devices->devices)) {
 199                 device = list_entry(fs_devices->devices.next,
 200                                     struct btrfs_device, dev_list);
 201                 list_del(&device->dev_list);
 202                 rcu_string_free(device->name);
 203                 kfree(device);
 204         }
 205         kfree(fs_devices);
 206 }
 207
 208 static void btrfs_kobject_uevent(struct block_device *bdev,
 209                                  enum kobject_action action)
 210 {
 211         int ret;
 212
 213         ret = kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, action);
 214         if (ret)
 215                 pr_warn("BTRFS: Sending event '%d' to kobject: '%s' (%p): failed\n",
 216                         action,
 217                         kobject_name(&disk_to_dev(bdev->bd_disk)->kobj),
 218                         &disk_to_dev(bdev->bd_disk)->kobj);
 219 }
 220
 221 void btrfs_cleanup_fs_uuids(void)
 222 {
 223         struct btrfs_fs_devices *fs_devices;
 224
 225         while (!list_empty(&fs_uuids)) {
 226                 fs_devices = list_entry(fs_uuids.next,
 227                                         struct btrfs_fs_devices, list);
 228                 list_del(&fs_devices->list);
 229                 free_fs_devices(fs_devices);
 230         }
 231 }
 232
 233 static struct btrfs_device *__alloc_device(void)
 234 {
 235         struct btrfs_device *dev;
 236
 237         dev = kzalloc(sizeof(*dev), GFP_KERNEL);
 238         if (!dev)
 239                 return ERR_PTR(-ENOMEM);
 240
 241         INIT_LIST_HEAD(&dev->dev_list);
 242         INIT_LIST_HEAD(&dev->dev_alloc_list);
 243         INIT_LIST_HEAD(&dev->resized_list);
 244
 245         spin_lock_init(&dev->io_lock);
 246
 247         spin_lock_init(&dev->reada_lock);
 248         atomic_set(&dev->reada_in_flight, 0);
 249         atomic_set(&dev->dev_stats_ccnt, 0);
 250         btrfs_device_data_ordered_init(dev);
 251         INIT_RADIX_TREE(&dev->reada_zones, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
 252         INIT_RADIX_TREE(&dev->reada_extents, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
 253
 254         return dev;
 255 }
 256
 257 static noinline struct btrfs_device *__find_device(struct list_head *head,
 258                                                    u64 devid, u8 *uuid)
 259 {
 260         struct btrfs_device *dev;
 261
 262         list_for_each_entry(dev, head, dev_list) {
 263                 if (dev->devid == devid &&
 264                     (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
 265                         return dev;
 266                 }
 267         }
 268         return NULL;
 269 }
 270
 271 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
 272 {
 273         struct btrfs_fs_devices *fs_devices;
 274
 275         list_for_each_entry(fs_devices, &fs_uuids, list) {
 276                 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
 277                         return fs_devices;
 278         }
 279         return NULL;
 280 }
 281
 282 static int
 283 btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder,
 284                       int flush, struct block_device **bdev,
 285                       struct buffer_head **bh)
 286 {
 287         int ret;
 288
 289         *bdev = blkdev_get_by_path(device_path, flags, holder);
 290
 291         if (IS_ERR(*bdev)) {
 292                 ret = PTR_ERR(*bdev);
 293                 goto error;
 294         }
 295
 296         if (flush)
 297                 filemap_write_and_wait((*bdev)->bd_inode->i_mapping);
 298         ret = set_blocksize(*bdev, 4096);
 299         if (ret) {
 300                 blkdev_put(*bdev, flags);
 301                 goto error;
 302         }
 303         invalidate_bdev(*bdev);
 304         *bh = btrfs_read_dev_super(*bdev);
 305         if (IS_ERR(*bh)) {
 306                 ret = PTR_ERR(*bh);
 307                 blkdev_put(*bdev, flags);
 308                 goto error;
 309         }
 310
 311         return 0;
 312
 313 error:
 314         *bdev = NULL;
 315         *bh = NULL;
 316         return ret;
 317 }
 318
 319 static void requeue_list(struct btrfs_pending_bios *pending_bios,
 320                         struct bio *head, struct bio *tail)
 321 {
 322
 323         struct bio *old_head;
 324
 325         old_head = pending_bios->head;
 326         pending_bios->head = head;
 327         if (pending_bios->tail)
 328                 tail->bi_next = old_head;
 329         else
 330                 pending_bios->tail = tail;
 331 }
 332
 333 /*
 334  * we try to collect pending bios for a device so we don't get a large
 335  * number of procs sending bios down to the same device.  This greatly
 336  * improves the schedulers ability to collect and merge the bios.
 337  *
 338  * But, it also turns into a long list of bios to process and that is sure
 339  * to eventually make the worker thread block.  The solution here is to
 340  * make some progress and then put this work struct back at the end of
 341  * the list if the block device is congested.  This way, multiple devices
 342  * can make progress from a single worker thread.
 343  */
 344 static noinline void run_scheduled_bios(struct btrfs_device *device)
 345 {
 346         struct bio *pending;
 347         struct backing_dev_info *bdi;
 348         struct btrfs_fs_info *fs_info;
 349         struct btrfs_pending_bios *pending_bios;
 350         struct bio *tail;
 351         struct bio *cur;
 352         int again = 0;
 353         unsigned long num_run;
 354         unsigned long batch_run = 0;
 355         unsigned long limit;
 356         unsigned long last_waited = 0;
 357         int force_reg = 0;
 358         int sync_pending = 0;
 359         struct blk_plug plug;
 360
 361         /*
 362          * this function runs all the bios we've collected for
 363          * a particular device.  We don't want to wander off to
 364          * another device without first sending all of these down.
 365          * So, setup a plug here and finish it off before we return
 366          */
 367         blk_start_plug(&plug);
 368
 369         bdi = blk_get_backing_dev_info(device->bdev);
 370         fs_info = device->dev_root->fs_info;
 371         limit = btrfs_async_submit_limit(fs_info);
 372         limit = limit * 2 / 3;
 373
 374 loop:
 375         spin_lock(&device->io_lock);
 376
 377 loop_lock:
 378         num_run = 0;
 379
 380         /* take all the bios off the list at once and process them
 381          * later on (without the lock held).  But, remember the
 382          * tail and other pointers so the bios can be properly reinserted
 383          * into the list if we hit congestion
 384          */
 385         if (!force_reg && device->pending_sync_bios.head) {
 386                 pending_bios = &device->pending_sync_bios;
 387                 force_reg = 1;
 388         } else {
 389                 pending_bios = &device->pending_bios;
 390                 force_reg = 0;
 391         }
 392
 393         pending = pending_bios->head;
 394         tail = pending_bios->tail;
 395         WARN_ON(pending && !tail);
 396
 397         /*
 398          * if pending was null this time around, no bios need processing
 399          * at all and we can stop.  Otherwise it'll loop back up again
 400          * and do an additional check so no bios are missed.
 401          *
 402          * device->running_pending is used to synchronize with the
 403          * schedule_bio code.
 404          */
 405         if (device->pending_sync_bios.head == NULL &&
 406             device->pending_bios.head == NULL) {
 407                 again = 0;
 408                 device->running_pending = 0;
 409         } else {
 410                 again = 1;
 411                 device->running_pending = 1;
 412         }
 413
 414         pending_bios->head = NULL;
 415         pending_bios->tail = NULL;
 416
 417         spin_unlock(&device->io_lock);
 418
 419         while (pending) {
 420
 421                 rmb();
 422                 /* we want to work on both lists, but do more bios on the
 423                  * sync list than the regular list
 424                  */
 425                 if ((num_run > 32 &&
 426                     pending_bios != &device->pending_sync_bios &&
 427                     device->pending_sync_bios.head) ||
 428                    (num_run > 64 && pending_bios == &device->pending_sync_bios &&
 429                     device->pending_bios.head)) {
 430                         spin_lock(&device->io_lock);
 431                         requeue_list(pending_bios, pending, tail);
 432                         goto loop_lock;
 433                 }
 434
 435                 cur = pending;
 436                 pending = pending->bi_next;
 437                 cur->bi_next = NULL;
 438
 439                 /*
 440                  * atomic_dec_return implies a barrier for waitqueue_active
 441                  */
 442                 if (atomic_dec_return(&fs_info->nr_async_bios) < limit &&
 443                     waitqueue_active(&fs_info->async_submit_wait))
 444                         wake_up(&fs_info->async_submit_wait);
 445
 446                 BUG_ON(atomic_read(&cur->__bi_cnt) == 0);
 447
 448                 /*
 449                  * if we're doing the sync list, record that our
 450                  * plug has some sync requests on it
 451                  *
 452                  * If we're doing the regular list and there are
 453                  * sync requests sitting around, unplug before
 454                  * we add more
 455                  */
 456                 if (pending_bios == &device->pending_sync_bios) {
 457                         sync_pending = 1;
 458                 } else if (sync_pending) {
 459                         blk_finish_plug(&plug);
 460                         blk_start_plug(&plug);
 461                         sync_pending = 0;
 462                 }
 463
 464                 btrfsic_submit_bio(cur);
 465                 num_run++;
 466                 batch_run++;
 467
 468                 cond_resched();
 469
 470                 /*
 471                  * we made progress, there is more work to do and the bdi
 472                  * is now congested.  Back off and let other work structs
 473                  * run instead
 474                  */
 475                 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
 476                     fs_info->fs_devices->open_devices > 1) {
 477                         struct io_context *ioc;
 478
 479                         ioc = current->io_context;
 480
 481                         /*
 482                          * the main goal here is that we don't want to
 483                          * block if we're going to be able to submit
 484                          * more requests without blocking.
 485                          *
 486                          * This code does two great things, it pokes into
 487                          * the elevator code from a filesystem _and_
 488                          * it makes assumptions about how batching works.
 489                          */
 490                         if (ioc && ioc->nr_batch_requests > 0 &&
 491                             time_before(jiffies, ioc->last_waited + HZ/50UL) &&
 492                             (last_waited == 0 ||
 493                              ioc->last_waited == last_waited)) {
 494                                 /*
 495                                  * we want to go through our batch of
 496                                  * requests and stop.  So, we copy out
 497                                  * the ioc->last_waited time and test
 498                                  * against it before looping
 499                                  */
 500                                 last_waited = ioc->last_waited;
 501                                 cond_resched();
 502                                 continue;
 503                         }
 504                         spin_lock(&device->io_lock);
 505                         requeue_list(pending_bios, pending, tail);
 506                         device->running_pending = 1;
 507
 508                         spin_unlock(&device->io_lock);
 509                         btrfs_queue_work(fs_info->submit_workers,
 510                                          &device->work);
 511                         goto done;
 512                 }
 513                 /* unplug every 64 requests just for good measure */
 514                 if (batch_run % 64 == 0) {
 515                         blk_finish_plug(&plug);
 516                         blk_start_plug(&plug);
 517                         sync_pending = 0;
 518                 }
 519         }
 520
 521         cond_resched();
 522         if (again)
 523                 goto loop;
 524
 525         spin_lock(&device->io_lock);
 526         if (device->pending_bios.head || device->pending_sync_bios.head)
 527                 goto loop_lock;
 528         spin_unlock(&device->io_lock);
 529
 530 done:
 531         blk_finish_plug(&plug);
 532 }
 533
 534 static void pending_bios_fn(struct btrfs_work *work)
 535 {
 536         struct btrfs_device *device;
 537
 538         device = container_of(work, struct btrfs_device, work);
 539         run_scheduled_bios(device);
 540 }
 541
 542
 543 void btrfs_free_stale_device(struct btrfs_device *cur_dev)
 544 {
 545         struct btrfs_fs_devices *fs_devs;
 546         struct btrfs_device *dev;
 547
 548         if (!cur_dev->name)
 549                 return;
 550
 551         list_for_each_entry(fs_devs, &fs_uuids, list) {
 552                 int del = 1;
 553
 554                 if (fs_devs->opened)
 555                         continue;
 556                 if (fs_devs->seeding)
 557                         continue;
 558
 559                 list_for_each_entry(dev, &fs_devs->devices, dev_list) {
 560
 561                         if (dev == cur_dev)
 562                                 continue;
 563                         if (!dev->name)
 564                                 continue;
 565
 566                         /*
 567                          * Todo: This won't be enough. What if the same device
 568                          * comes back (with new uuid and) with its mapper path?
 569                          * But for now, this does help as mostly an admin will
 570                          * either use mapper or non mapper path throughout.
 571                          */
 572                         rcu_read_lock();
 573                         del = strcmp(rcu_str_deref(dev->name),
 574                                                 rcu_str_deref(cur_dev->name));
 575                         rcu_read_unlock();
 576                         if (!del)
 577                                 break;
 578                 }
 579
 580                 if (!del) {
 581                         /* delete the stale device */
 582                         if (fs_devs->num_devices == 1) {
 583                                 btrfs_sysfs_remove_fsid(fs_devs);
 584                                 list_del(&fs_devs->list);
 585                                 free_fs_devices(fs_devs);
 586                         } else {
 587                                 fs_devs->num_devices--;
 588                                 list_del(&dev->dev_list);
 589                                 rcu_string_free(dev->name);
 590                                 kfree(dev);
 591                         }
 592                         break;
 593                 }
 594         }
 595 }
 596
 597 /*
 598  * Add new device to list of registered devices
 599  *
 600  * Returns:
 601  * 1   - first time device is seen
 602  * 0   - device already known
 603  * < 0 - error
 604  */
 605 static noinline int device_list_add(const char *path,
 606                            struct btrfs_super_block *disk_super,
 607                            u64 devid, struct btrfs_fs_devices **fs_devices_ret)
 608 {
 609         struct btrfs_device *device;
 610         struct btrfs_fs_devices *fs_devices;
 611         struct rcu_string *name;
 612         int ret = 0;
 613         u64 found_transid = btrfs_super_generation(disk_super);
 614
 615         fs_devices = find_fsid(disk_super->fsid);
 616         if (!fs_devices) {
 617                 fs_devices = alloc_fs_devices(disk_super->fsid);
 618                 if (IS_ERR(fs_devices))
 619                         return PTR_ERR(fs_devices);
 620
 621                 list_add(&fs_devices->list, &fs_uuids);
 622
 623                 device = NULL;
 624         } else {
 625                 device = __find_device(&fs_devices->devices, devid,
 626                                        disk_super->dev_item.uuid);
 627         }
 628
 629         if (!device) {
 630                 if (fs_devices->opened)
 631                         return -EBUSY;
 632
 633                 device = btrfs_alloc_device(NULL, &devid,
 634                                             disk_super->dev_item.uuid);
 635                 if (IS_ERR(device)) {
 636                         /* we can safely leave the fs_devices entry around */
 637                         return PTR_ERR(device);
 638                 }
 639
 640                 name = rcu_string_strdup(path, GFP_NOFS);
 641                 if (!name) {
 642                         kfree(device);
 643                         return -ENOMEM;
 644                 }
 645                 rcu_assign_pointer(device->name, name);
 646
 647                 mutex_lock(&fs_devices->device_list_mutex);
 648                 list_add_rcu(&device->dev_list, &fs_devices->devices);
 649                 fs_devices->num_devices++;
 650                 mutex_unlock(&fs_devices->device_list_mutex);
 651
 652                 ret = 1;
 653                 device->fs_devices = fs_devices;
 654         } else if (!device->name || strcmp(device->name->str, path)) {
 655                 /*
 656                  * When FS is already mounted.
 657                  * 1. If you are here and if the device->name is NULL that
 658                  *    means this device was missing at time of FS mount.
 659                  * 2. If you are here and if the device->name is different
 660                  *    from 'path' that means either
 661                  *      a. The same device disappeared and reappeared with
 662                  *         different name. or
 663                  *      b. The missing-disk-which-was-replaced, has
 664                  *         reappeared now.
 665                  *
 666                  * We must allow 1 and 2a above. But 2b would be a spurious
 667                  * and unintentional.
 668                  *
 669                  * Further in case of 1 and 2a above, the disk at 'path'
 670                  * would have missed some transaction when it was away and
 671                  * in case of 2a the stale bdev has to be updated as well.
 672                  * 2b must not be allowed at all time.
 673                  */
 674
 675                 /*
 676                  * For now, we do allow update to btrfs_fs_device through the
 677                  * btrfs dev scan cli after FS has been mounted.  We're still
 678                  * tracking a problem where systems fail mount by subvolume id
 679                  * when we reject replacement on a mounted FS.
 680                  */
 681                 if (!fs_devices->opened && found_transid < device->generation) {
 682                         /*
 683                          * That is if the FS is _not_ mounted and if you
 684                          * are here, that means there is more than one
 685                          * disk with same uuid and devid.We keep the one
 686                          * with larger generation number or the last-in if
 687                          * generation are equal.
 688                          */
 689                         return -EEXIST;
 690                 }
 691
 692                 name = rcu_string_strdup(path, GFP_NOFS);
 693                 if (!name)
 694                         return -ENOMEM;
 695                 rcu_string_free(device->name);
 696                 rcu_assign_pointer(device->name, name);
 697                 if (device->missing) {
 698                         fs_devices->missing_devices--;
 699                         device->missing = 0;
 700                 }
 701         }
 702
 703         /*
 704          * Unmount does not free the btrfs_device struct but would zero
 705          * generation along with most of the other members. So just update
 706          * it back. We need it to pick the disk with largest generation
 707          * (as above).
 708          */
 709         if (!fs_devices->opened)
 710                 device->generation = found_transid;
 711
 712         /*
 713          * if there is new btrfs on an already registered device,
 714          * then remove the stale device entry.
 715          */
 716         if (ret > 0)
 717                 btrfs_free_stale_device(device);
 718
 719         *fs_devices_ret = fs_devices;
 720
 721         return ret;
 722 }
 723
 724 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
 725 {
 726         struct btrfs_fs_devices *fs_devices;
 727         struct btrfs_device *device;
 728         struct btrfs_device *orig_dev;
 729
 730         fs_devices = alloc_fs_devices(orig->fsid);
 731         if (IS_ERR(fs_devices))
 732                 return fs_devices;
 733
 734         mutex_lock(&orig->device_list_mutex);
 735         fs_devices->total_devices = orig->total_devices;
 736
 737         /* We have held the volume lock, it is safe to get the devices. */
 738         list_for_each_entry(orig_dev, &orig->devices, dev_list) {
 739                 struct rcu_string *name;
 740
 741                 device = btrfs_alloc_device(NULL, &orig_dev->devid,
 742                                             orig_dev->uuid);
 743                 if (IS_ERR(device))
 744                         goto error;
 745
 746                 /*
 747                  * This is ok to do without rcu read locked because we hold the
 748                  * uuid mutex so nothing we touch in here is going to disappear.
 749                  */
 750                 if (orig_dev->name) {
 751                         name = rcu_string_strdup(orig_dev->name->str,
 752                                         GFP_KERNEL);
 753                         if (!name) {
 754                                 kfree(device);
 755                                 goto error;
 756                         }
 757                         rcu_assign_pointer(device->name, name);
 758                 }
 759
 760                 list_add(&device->dev_list, &fs_devices->devices);
 761                 device->fs_devices = fs_devices;
 762                 fs_devices->num_devices++;
 763         }
 764         mutex_unlock(&orig->device_list_mutex);
 765         return fs_devices;
 766 error:
 767         mutex_unlock(&orig->device_list_mutex);
 768         free_fs_devices(fs_devices);
 769         return ERR_PTR(-ENOMEM);
 770 }
 771
 772 void btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices, int step)
 773 {
 774         struct btrfs_device *device, *next;
 775         struct btrfs_device *latest_dev = NULL;
 776
 777         mutex_lock(&uuid_mutex);
 778 again:
 779         /* This is the initialized path, it is safe to release the devices. */
 780         list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
 781                 if (device->in_fs_metadata) {
 782                         if (!device->is_tgtdev_for_dev_replace &&
 783                             (!latest_dev ||
 784                              device->generation > latest_dev->generation)) {
 785                                 latest_dev = device;
 786                         }
 787                         continue;
 788                 }
 789
 790                 if (device->devid == BTRFS_DEV_REPLACE_DEVID) {
 791                         /*
 792                          * In the first step, keep the device which has
 793                          * the correct fsid and the devid that is used
 794                          * for the dev_replace procedure.
 795                          * In the second step, the dev_replace state is
 796                          * read from the device tree and it is known
 797                          * whether the procedure is really active or
 798                          * not, which means whether this device is
 799                          * used or whether it should be removed.
 800                          */
 801                         if (step == 0 || device->is_tgtdev_for_dev_replace) {
 802                                 continue;
 803                         }
 804                 }
 805                 if (device->bdev) {
 806                         blkdev_put(device->bdev, device->mode);
 807                         device->bdev = NULL;
 808                         fs_devices->open_devices--;
 809                 }
 810                 if (device->writeable) {
 811                         list_del_init(&device->dev_alloc_list);
 812                         device->writeable = 0;
 813                         if (!device->is_tgtdev_for_dev_replace)
 814                                 fs_devices->rw_devices--;
 815                 }
 816                 list_del_init(&device->dev_list);
 817                 fs_devices->num_devices--;
 818                 rcu_string_free(device->name);
 819                 kfree(device);
 820         }
 821
 822         if (fs_devices->seed) {
 823                 fs_devices = fs_devices->seed;
 824                 goto again;
 825         }
 826
 827         fs_devices->latest_bdev = latest_dev->bdev;
 828
 829         mutex_unlock(&uuid_mutex);
 830 }
 831
 832 static void __free_device(struct work_struct *work)
 833 {
 834         struct btrfs_device *device;
 835
 836         device = container_of(work, struct btrfs_device, rcu_work);
 837
 838         if (device->bdev)
 839                 blkdev_put(device->bdev, device->mode);
 840
 841         rcu_string_free(device->name);
 842         kfree(device);
 843 }
 844
 845 static void free_device(struct rcu_head *head)
 846 {
 847         struct btrfs_device *device;
 848
 849         device = container_of(head, struct btrfs_device, rcu);
 850
 851         INIT_WORK(&device->rcu_work, __free_device);
 852         schedule_work(&device->rcu_work);
 853 }
 854
 855 static void btrfs_close_one_device(struct btrfs_device *device)
 856 {
 857         struct btrfs_fs_devices *fs_devices = device->fs_devices;
 858         struct btrfs_device *new_device;
 859         struct rcu_string *name;
 860
 861         if (device->bdev)
 862                 fs_devices->open_devices--;
 863
 864         if (device->writeable &&
 865             device->devid != BTRFS_DEV_REPLACE_DEVID) {
 866                 list_del_init(&device->dev_alloc_list);
 867                 fs_devices->rw_devices--;
 868         }
 869
 870         if (device->missing)
 871                 fs_devices->missing_devices--;
 872
 873         if (device->bdev && device->writeable) {
 874                 sync_blockdev(device->bdev);
 875                 invalidate_bdev(device->bdev);
 876         }
 877
 878         new_device = btrfs_alloc_device(NULL, &device->devid,
 879                                         device->uuid);
 880         BUG_ON(IS_ERR(new_device)); /* -ENOMEM */
 881
 882         /* Safe because we are under uuid_mutex */
 883         if (device->name) {
 884                 name = rcu_string_strdup(device->name->str, GFP_NOFS);
 885                 BUG_ON(!name); /* -ENOMEM */
 886                 rcu_assign_pointer(new_device->name, name);
 887         }
 888
 889         list_replace_rcu(&device->dev_list, &new_device->dev_list);
 890         new_device->fs_devices = device->fs_devices;
 891
 892         call_rcu(&device->rcu, free_device);
 893 }
 894
 895 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
 896 {
 897         struct btrfs_device *device, *tmp;
 898
 899         if (--fs_devices->opened > 0)
 900                 return 0;
 901
 902         mutex_lock(&fs_devices->device_list_mutex);
 903         list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list) {
 904                 btrfs_close_one_device(device);
 905         }
 906         mutex_unlock(&fs_devices->device_list_mutex);
 907
 908         WARN_ON(fs_devices->open_devices);
 909         WARN_ON(fs_devices->rw_devices);
 910         fs_devices->opened = 0;
 911         fs_devices->seeding = 0;
 912
 913         return 0;
 914 }
 915
 916 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
 917 {
 918         struct btrfs_fs_devices *seed_devices = NULL;
 919         int ret;
 920
 921         mutex_lock(&uuid_mutex);
 922         ret = __btrfs_close_devices(fs_devices);
 923         if (!fs_devices->opened) {
 924                 seed_devices = fs_devices->seed;
 925                 fs_devices->seed = NULL;
 926         }
 927         mutex_unlock(&uuid_mutex);
 928
 929         while (seed_devices) {
 930                 fs_devices = seed_devices;
 931                 seed_devices = fs_devices->seed;
 932                 __btrfs_close_devices(fs_devices);
 933                 free_fs_devices(fs_devices);
 934         }
 935         /*
 936          * Wait for rcu kworkers under __btrfs_close_devices
 937          * to finish all blkdev_puts so device is really
 938          * free when umount is done.
 939          */
 940         rcu_barrier();
 941         return ret;
 942 }
 943
 944 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
 945                                 fmode_t flags, void *holder)
 946 {
 947         struct request_queue *q;
 948         struct block_device *bdev;
 949         struct list_head *head = &fs_devices->devices;
 950         struct btrfs_device *device;
 951         struct btrfs_device *latest_dev = NULL;
 952         struct buffer_head *bh;
 953         struct btrfs_super_block *disk_super;
 954         u64 devid;
 955         int seeding = 1;
 956         int ret = 0;
 957
 958         flags |= FMODE_EXCL;
 959
 960         list_for_each_entry(device, head, dev_list) {
 961                 if (device->bdev)
 962                         continue;
 963                 if (!device->name)
 964                         continue;
 965
 966                 /* Just open everything we can; ignore failures here */
 967                 if (btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1,
 968                                             &bdev, &bh))
 969                         continue;
 970
 971                 disk_super = (struct btrfs_super_block *)bh->b_data;
 972                 devid = btrfs_stack_device_id(&disk_super->dev_item);
 973                 if (devid != device->devid)
 974                         goto error_brelse;
 975
 976                 if (memcmp(device->uuid, disk_super->dev_item.uuid,
 977                            BTRFS_UUID_SIZE))
 978                         goto error_brelse;
 979
 980                 device->generation = btrfs_super_generation(disk_super);
 981                 if (!latest_dev ||
 982                     device->generation > latest_dev->generation)
 983                         latest_dev = device;
 984
 985                 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
 986                         device->writeable = 0;
 987                 } else {
 988                         device->writeable = !bdev_read_only(bdev);
 989                         seeding = 0;
 990                 }
 991
 992                 q = bdev_get_queue(bdev);
 993                 if (blk_queue_discard(q))
 994                         device->can_discard = 1;
 995
 996                 device->bdev = bdev;
 997                 device->in_fs_metadata = 0;
 998                 device->mode = flags;
 999
1000                 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
1001                         fs_devices->rotating = 1;
1002
1003                 fs_devices->open_devices++;
1004                 if (device->writeable &&
1005                     device->devid != BTRFS_DEV_REPLACE_DEVID) {
1006                         fs_devices->rw_devices++;
1007                         list_add(&device->dev_alloc_list,
1008                                  &fs_devices->alloc_list);
1009                 }
1010                 brelse(bh);
1011                 continue;
1012
1013 error_brelse:
1014                 brelse(bh);
1015                 blkdev_put(bdev, flags);
1016                 continue;
1017         }
1018         if (fs_devices->open_devices == 0) {
1019                 ret = -EINVAL;
1020                 goto out;
1021         }
1022         fs_devices->seeding = seeding;
1023         fs_devices->opened = 1;
1024         fs_devices->latest_bdev = latest_dev->bdev;
1025         fs_devices->total_rw_bytes = 0;
1026 out:
1027         return ret;
1028 }
1029
1030 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
1031                        fmode_t flags, void *holder)
1032 {
1033         int ret;
1034
1035         mutex_lock(&uuid_mutex);
1036         if (fs_devices->opened) {
1037                 fs_devices->opened++;
1038                 ret = 0;
1039         } else {
1040                 ret = __btrfs_open_devices(fs_devices, flags, holder);
1041         }
1042         mutex_unlock(&uuid_mutex);
1043         return ret;
1044 }
1045
1046 void btrfs_release_disk_super(struct page *page)
1047 {
1048         kunmap(page);
1049         put_page(page);
1050 }
1051
1052 int btrfs_read_disk_super(struct block_device *bdev, u64 bytenr,
1053                 struct page **page, struct btrfs_super_block **disk_super)
1054 {
1055         void *p;
1056         pgoff_t index;
1057
1058         /* make sure our super fits in the device */
1059         if (bytenr + PAGE_SIZE >= i_size_read(bdev->bd_inode))
1060                 return 1;
1061
1062         /* make sure our super fits in the page */
1063         if (sizeof(**disk_super) > PAGE_SIZE)
1064                 return 1;
1065
1066         /* make sure our super doesn't straddle pages on disk */
1067         index = bytenr >> PAGE_SHIFT;
1068         if ((bytenr + sizeof(**disk_super) - 1) >> PAGE_SHIFT != index)
1069                 return 1;
1070
1071         /* pull in the page with our super */
1072         *page = read_cache_page_gfp(bdev->bd_inode->i_mapping,
1073                                    index, GFP_KERNEL);
1074
1075         if (IS_ERR_OR_NULL(*page))
1076                 return 1;
1077
1078         p = kmap(*page);
1079
1080         /* align our pointer to the offset of the super block */
1081         *disk_super = p + (bytenr & ~PAGE_MASK);
1082
1083         if (btrfs_super_bytenr(*disk_super) != bytenr ||
1084             btrfs_super_magic(*disk_super) != BTRFS_MAGIC) {
1085                 btrfs_release_disk_super(*page);
1086                 return 1;
1087         }
1088
1089         if ((*disk_super)->label[0] &&
1090                 (*disk_super)->label[BTRFS_LABEL_SIZE - 1])
1091                 (*disk_super)->label[BTRFS_LABEL_SIZE - 1] = '\0';
1092
1093         return 0;
1094 }
1095
1096 /*
1097  * Look for a btrfs signature on a device. This may be called out of the mount path
1098  * and we are not allowed to call set_blocksize during the scan. The superblock
1099  * is read via pagecache
1100  */
1101 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
1102                           struct btrfs_fs_devices **fs_devices_ret)
1103 {
1104         struct btrfs_super_block *disk_super;
1105         struct block_device *bdev;
1106         struct page *page;
1107         int ret = -EINVAL;
1108         u64 devid;
1109         u64 transid;
1110         u64 total_devices;
1111         u64 bytenr;
1112
1113         /*
1114          * we would like to check all the supers, but that would make
1115          * a btrfs mount succeed after a mkfs from a different FS.
1116          * So, we need to add a special mount option to scan for
1117          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
1118          */
1119         bytenr = btrfs_sb_offset(0);
1120         flags |= FMODE_EXCL;
1121         mutex_lock(&uuid_mutex);
1122
1123         bdev = blkdev_get_by_path(path, flags, holder);
1124         if (IS_ERR(bdev)) {
1125                 ret = PTR_ERR(bdev);
1126                 goto error;
1127         }
1128
1129         if (btrfs_read_disk_super(bdev, bytenr, &page, &disk_super))
1130                 goto error_bdev_put;
1131
1132         devid = btrfs_stack_device_id(&disk_super->dev_item);
1133         transid = btrfs_super_generation(disk_super);
1134         total_devices = btrfs_super_num_devices(disk_super);
1135
1136         ret = device_list_add(path, disk_super, devid, fs_devices_ret);
1137         if (ret > 0) {
1138                 if (disk_super->label[0]) {
1139                         printk(KERN_INFO "BTRFS: device label %s ", disk_super->label);
1140                 } else {
1141                         printk(KERN_INFO "BTRFS: device fsid %pU ", disk_super->fsid);
1142                 }
1143
1144                 printk(KERN_CONT "devid %llu transid %llu %s\n", devid, transid, path);
1145                 ret = 0;
1146         }
1147         if (!ret && fs_devices_ret)
1148                 (*fs_devices_ret)->total_devices = total_devices;
1149
1150         btrfs_release_disk_super(page);
1151
1152 error_bdev_put:
1153         blkdev_put(bdev, flags);
1154 error:
1155         mutex_unlock(&uuid_mutex);
1156         return ret;
1157 }
1158
1159 /* helper to account the used device space in the range */
1160 int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
1161                                    u64 end, u64 *length)
1162 {
1163         struct btrfs_key key;
1164         struct btrfs_root *root = device->dev_root;
1165         struct btrfs_dev_extent *dev_extent;
1166         struct btrfs_path *path;
1167         u64 extent_end;
1168         int ret;
1169         int slot;
1170         struct extent_buffer *l;
1171
1172         *length = 0;
1173
1174         if (start >= device->total_bytes || device->is_tgtdev_for_dev_replace)
1175                 return 0;
1176
1177         path = btrfs_alloc_path();
1178         if (!path)
1179                 return -ENOMEM;
1180         path->reada = READA_FORWARD;
1181
1182         key.objectid = device->devid;
1183         key.offset = start;
1184         key.type = BTRFS_DEV_EXTENT_KEY;
1185
1186         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1187         if (ret < 0)
1188                 goto out;
1189         if (ret > 0) {
1190                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1191                 if (ret < 0)
1192                         goto out;
1193         }
1194
1195         while (1) {
1196                 l = path->nodes[0];
1197                 slot = path->slots[0];
1198                 if (slot >= btrfs_header_nritems(l)) {
1199                         ret = btrfs_next_leaf(root, path);
1200                         if (ret == 0)
1201                                 continue;
1202                         if (ret < 0)
1203                                 goto out;
1204
1205                         break;
1206                 }
1207                 btrfs_item_key_to_cpu(l, &key, slot);
1208
1209                 if (key.objectid < device->devid)
1210                         goto next;
1211
1212                 if (key.objectid > device->devid)
1213                         break;
1214
1215                 if (key.type != BTRFS_DEV_EXTENT_KEY)
1216                         goto next;
1217
1218                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1219                 extent_end = key.offset + btrfs_dev_extent_length(l,
1220                                                                   dev_extent);
1221                 if (key.offset <= start && extent_end > end) {
1222                         *length = end - start + 1;
1223                         break;
1224                 } else if (key.offset <= start && extent_end > start)
1225                         *length += extent_end - start;
1226                 else if (key.offset > start && extent_end <= end)
1227                         *length += extent_end - key.offset;
1228                 else if (key.offset > start && key.offset <= end) {
1229                         *length += end - key.offset + 1;
1230                         break;
1231                 } else if (key.offset > end)
1232                         break;
1233
1234 next:
1235                 path->slots[0]++;
1236         }
1237         ret = 0;
1238 out:
1239         btrfs_free_path(path);
1240         return ret;
1241 }
1242
1243 static int contains_pending_extent(struct btrfs_transaction *transaction,
1244                                    struct btrfs_device *device,
1245                                    u64 *start, u64 len)
1246 {
1247         struct btrfs_fs_info *fs_info = device->dev_root->fs_info;
1248         struct extent_map *em;
1249         struct list_head *search_list = &fs_info->pinned_chunks;
1250         int ret = 0;
1251         u64 physical_start = *start;
1252
1253         if (transaction)
1254                 search_list = &transaction->pending_chunks;
1255 again:
1256         list_for_each_entry(em, search_list, list) {
1257                 struct map_lookup *map;
1258                 int i;
1259
1260                 map = em->map_lookup;
1261                 for (i = 0; i < map->num_stripes; i++) {
1262                         u64 end;
1263
1264                         if (map->stripes[i].dev != device)
1265                                 continue;
1266                         if (map->stripes[i].physical >= physical_start + len ||
1267                             map->stripes[i].physical + em->orig_block_len <=
1268                             physical_start)
1269                                 continue;
1270                         /*
1271                          * Make sure that while processing the pinned list we do
1272                          * not override our *start with a lower value, because
1273                          * we can have pinned chunks that fall within this
1274                          * device hole and that have lower physical addresses
1275                          * than the pending chunks we processed before. If we
1276                          * do not take this special care we can end up getting
1277                          * 2 pending chunks that start at the same physical
1278                          * device offsets because the end offset of a pinned
1279                          * chunk can be equal to the start offset of some
1280                          * pending chunk.
1281                          */
1282                         end = map->stripes[i].physical + em->orig_block_len;
1283                         if (end > *start) {
1284                                 *start = end;
1285                                 ret = 1;
1286                         }
1287                 }
1288         }
1289         if (search_list != &fs_info->pinned_chunks) {
1290                 search_list = &fs_info->pinned_chunks;
1291                 goto again;
1292         }
1293
1294         return ret;
1295 }
1296
1297
1298 /*
1299  * find_free_dev_extent_start - find free space in the specified device
1300  * @device:       the device which we search the free space in
1301  * @num_bytes:    the size of the free space that we need
1302  * @search_start: the position from which to begin the search
1303  * @start:        store the start of the free space.
1304  * @len:          the size of the free space. that we find, or the size
1305  *                of the max free space if we don't find suitable free space
1306  *
1307  * this uses a pretty simple search, the expectation is that it is
1308  * called very infrequently and that a given device has a small number
1309  * of extents
1310  *
1311  * @start is used to store the start of the free space if we find. But if we
1312  * don't find suitable free space, it will be used to store the start position
1313  * of the max free space.
1314  *
1315  * @len is used to store the size of the free space that we find.
1316  * But if we don't find suitable free space, it is used to store the size of
1317  * the max free space.
1318  */
1319 int find_free_dev_extent_start(struct btrfs_transaction *transaction,
1320                                struct btrfs_device *device, u64 num_bytes,
1321                                u64 search_start, u64 *start, u64 *len)
1322 {
1323         struct btrfs_key key;
1324         struct btrfs_root *root = device->dev_root;
1325         struct btrfs_dev_extent *dev_extent;
1326         struct btrfs_path *path;
1327         u64 hole_size;
1328         u64 max_hole_start;
1329         u64 max_hole_size;
1330         u64 extent_end;
1331         u64 search_end = device->total_bytes;
1332         int ret;
1333         int slot;
1334         struct extent_buffer *l;
1335         u64 min_search_start;
1336
1337         /*
1338          * We don't want to overwrite the superblock on the drive nor any area
1339          * used by the boot loader (grub for example), so we make sure to start
1340          * at an offset of at least 1MB.
1341          */
1342         min_search_start = max(root->fs_info->alloc_start, 1024ull * 1024);
1343         search_start = max(search_start, min_search_start);
1344
1345         path = btrfs_alloc_path();
1346         if (!path)
1347                 return -ENOMEM;
1348
1349         max_hole_start = search_start;
1350         max_hole_size = 0;
1351
1352 again:
1353         if (search_start >= search_end || device->is_tgtdev_for_dev_replace) {
1354                 ret = -ENOSPC;
1355                 goto out;
1356         }
1357
1358         path->reada = READA_FORWARD;
1359         path->search_commit_root = 1;
1360         path->skip_locking = 1;
1361
1362         key.objectid = device->devid;
1363         key.offset = search_start;
1364         key.type = BTRFS_DEV_EXTENT_KEY;
1365
1366         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1367         if (ret < 0)
1368                 goto out;
1369         if (ret > 0) {
1370                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1371                 if (ret < 0)
1372                         goto out;
1373         }
1374
1375         while (1) {
1376                 l = path->nodes[0];
1377                 slot = path->slots[0];
1378                 if (slot >= btrfs_header_nritems(l)) {
1379                         ret = btrfs_next_leaf(root, path);
1380                         if (ret == 0)
1381                                 continue;
1382                         if (ret < 0)
1383                                 goto out;
1384
1385                         break;
1386                 }
1387                 btrfs_item_key_to_cpu(l, &key, slot);
1388
1389                 if (key.objectid < device->devid)
1390                         goto next;
1391
1392                 if (key.objectid > device->devid)
1393                         break;
1394
1395                 if (key.type != BTRFS_DEV_EXTENT_KEY)
1396                         goto next;
1397
1398                 if (key.offset > search_start) {
1399                         hole_size = key.offset - search_start;
1400
1401                         /*
1402                          * Have to check before we set max_hole_start, otherwise
1403                          * we could end up sending back this offset anyway.
1404                          */
1405                         if (contains_pending_extent(transaction, device,
1406                                                     &search_start,
1407                                                     hole_size)) {
1408                                 if (key.offset >= search_start) {
1409                                         hole_size = key.offset - search_start;
1410                                 } else {
1411                                         WARN_ON_ONCE(1);
1412                                         hole_size = 0;
1413                                 }
1414                         }
1415
1416                         if (hole_size > max_hole_size) {
1417                                 max_hole_start = search_start;
1418                                 max_hole_size = hole_size;
1419                         }
1420
1421                         /*
1422                          * If this free space is greater than which we need,
1423                          * it must be the max free space that we have found
1424                          * until now, so max_hole_start must point to the start
1425                          * of this free space and the length of this free space
1426                          * is stored in max_hole_size. Thus, we return
1427                          * max_hole_start and max_hole_size and go back to the
1428                          * caller.
1429                          */
1430                         if (hole_size >= num_bytes) {
1431                                 ret = 0;
1432                                 goto out;
1433                         }
1434                 }
1435
1436                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1437                 extent_end = key.offset + btrfs_dev_extent_length(l,
1438                                                                   dev_extent);
1439                 if (extent_end > search_start)
1440                         search_start = extent_end;
1441 next:
1442                 path->slots[0]++;
1443                 cond_resched();
1444         }
1445
1446         /*
1447          * At this point, search_start should be the end of
1448          * allocated dev extents, and when shrinking the device,
1449          * search_end may be smaller than search_start.
1450          */
1451         if (search_end > search_start) {
1452                 hole_size = search_end - search_start;
1453
1454                 if (contains_pending_extent(transaction, device, &search_start,
1455                                             hole_size)) {
1456                         btrfs_release_path(path);
1457                         goto again;
1458                 }
1459
1460                 if (hole_size > max_hole_size) {
1461                         max_hole_start = search_start;
1462                         max_hole_size = hole_size;
1463                 }
1464         }
1465
1466         /* See above. */
1467         if (max_hole_size < num_bytes)
1468                 ret = -ENOSPC;
1469         else
1470                 ret = 0;
1471
1472 out:
1473         btrfs_free_path(path);
1474         *start = max_hole_start;
1475         if (len)
1476                 *len = max_hole_size;
1477         return ret;
1478 }
1479
1480 int find_free_dev_extent(struct btrfs_trans_handle *trans,
1481                          struct btrfs_device *device, u64 num_bytes,
1482                          u64 *start, u64 *len)
1483 {
1484         /* FIXME use last free of some kind */
1485         return find_free_dev_extent_start(trans->transaction, device,
1486                                           num_bytes, 0, start, len);
1487 }
1488
1489 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1490                           struct btrfs_device *device,
1491                           u64 start, u64 *dev_extent_len)
1492 {
1493         int ret;
1494         struct btrfs_path *path;
1495         struct btrfs_root *root = device->dev_root;
1496         struct btrfs_key key;
1497         struct btrfs_key found_key;
1498         struct extent_buffer *leaf = NULL;
1499         struct btrfs_dev_extent *extent = NULL;
1500
1501         path = btrfs_alloc_path();
1502         if (!path)
1503                 return -ENOMEM;
1504
1505         key.objectid = device->devid;
1506         key.offset = start;
1507         key.type = BTRFS_DEV_EXTENT_KEY;
1508 again:
1509         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1510         if (ret > 0) {
1511                 ret = btrfs_previous_item(root, path, key.objectid,
1512                                           BTRFS_DEV_EXTENT_KEY);
1513                 if (ret)
1514                         goto out;
1515                 leaf = path->nodes[0];
1516                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1517                 extent = btrfs_item_ptr(leaf, path->slots[0],
1518                                         struct btrfs_dev_extent);
1519                 BUG_ON(found_key.offset > start || found_key.offset +
1520                        btrfs_dev_extent_length(leaf, extent) < start);
1521                 key = found_key;
1522                 btrfs_release_path(path);
1523                 goto again;
1524         } else if (ret == 0) {
1525                 leaf = path->nodes[0];
1526                 extent = btrfs_item_ptr(leaf, path->slots[0],
1527                                         struct btrfs_dev_extent);
1528         } else {
1529                 btrfs_handle_fs_error(root->fs_info, ret, "Slot search failed");
1530                 goto out;
1531         }
1532
1533         *dev_extent_len = btrfs_dev_extent_length(leaf, extent);
1534
1535         ret = btrfs_del_item(trans, root, path);
1536         if (ret) {
1537                 btrfs_handle_fs_error(root->fs_info, ret,
1538                             "Failed to remove dev extent item");
1539         } else {
1540                 set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags);
1541         }
1542 out:
1543         btrfs_free_path(path);
1544         return ret;
1545 }
1546
1547 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1548                                   struct btrfs_device *device,
1549                                   u64 chunk_tree, u64 chunk_objectid,
1550                                   u64 chunk_offset, u64 start, u64 num_bytes)
1551 {
1552         int ret;
1553         struct btrfs_path *path;
1554         struct btrfs_root *root = device->dev_root;
1555         struct btrfs_dev_extent *extent;
1556         struct extent_buffer *leaf;
1557         struct btrfs_key key;
1558
1559         WARN_ON(!device->in_fs_metadata);
1560         WARN_ON(device->is_tgtdev_for_dev_replace);
1561         path = btrfs_alloc_path();
1562         if (!path)
1563                 return -ENOMEM;
1564
1565         key.objectid = device->devid;
1566         key.offset = start;
1567         key.type = BTRFS_DEV_EXTENT_KEY;
1568         ret = btrfs_insert_empty_item(trans, root, path, &key,
1569                                       sizeof(*extent));
1570         if (ret)
1571                 goto out;
1572
1573         leaf = path->nodes[0];
1574         extent = btrfs_item_ptr(leaf, path->slots[0],
1575                                 struct btrfs_dev_extent);
1576         btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
1577         btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
1578         btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1579
1580         write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
1581                     btrfs_dev_extent_chunk_tree_uuid(extent), BTRFS_UUID_SIZE);
1582
1583         btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1584         btrfs_mark_buffer_dirty(leaf);
1585 out:
1586         btrfs_free_path(path);
1587         return ret;
1588 }
1589
1590 static u64 find_next_chunk(struct btrfs_fs_info *fs_info)
1591 {
1592         struct extent_map_tree *em_tree;
1593         struct extent_map *em;
1594         struct rb_node *n;
1595         u64 ret = 0;
1596
1597         em_tree = &fs_info->mapping_tree.map_tree;
1598         read_lock(&em_tree->lock);
1599         n = rb_last(&em_tree->map);
1600         if (n) {
1601                 em = rb_entry(n, struct extent_map, rb_node);
1602                 ret = em->start + em->len;
1603         }
1604         read_unlock(&em_tree->lock);
1605
1606         return ret;
1607 }
1608
1609 static noinline int find_next_devid(struct btrfs_fs_info *fs_info,
1610                                     u64 *devid_ret)
1611 {
1612         int ret;
1613         struct btrfs_key key;
1614         struct btrfs_key found_key;
1615         struct btrfs_path *path;
1616
1617         path = btrfs_alloc_path();
1618         if (!path)
1619                 return -ENOMEM;
1620
1621         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1622         key.type = BTRFS_DEV_ITEM_KEY;
1623         key.offset = (u64)-1;
1624
1625         ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
1626         if (ret < 0)
1627                 goto error;
1628
1629         BUG_ON(ret == 0); /* Corruption */
1630
1631         ret = btrfs_previous_item(fs_info->chunk_root, path,
1632                                   BTRFS_DEV_ITEMS_OBJECTID,
1633                                   BTRFS_DEV_ITEM_KEY);
1634         if (ret) {
1635                 *devid_ret = 1;
1636         } else {
1637                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1638                                       path->slots[0]);
1639                 *devid_ret = found_key.offset + 1;
1640         }
1641         ret = 0;
1642 error:
1643         btrfs_free_path(path);
1644         return ret;
1645 }
1646
1647 /*
1648  * the device information is stored in the chunk root
1649  * the btrfs_device struct should be fully filled in
1650  */
1651 static int btrfs_add_device(struct btrfs_trans_handle *trans,
1652                             struct btrfs_root *root,
1653                             struct btrfs_device *device)
1654 {
1655         int ret;
1656         struct btrfs_path *path;
1657         struct btrfs_dev_item *dev_item;
1658         struct extent_buffer *leaf;
1659         struct btrfs_key key;
1660         unsigned long ptr;
1661
1662         root = root->fs_info->chunk_root;
1663
1664         path = btrfs_alloc_path();
1665         if (!path)
1666                 return -ENOMEM;
1667
1668         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1669         key.type = BTRFS_DEV_ITEM_KEY;
1670         key.offset = device->devid;
1671
1672         ret = btrfs_insert_empty_item(trans, root, path, &key,
1673                                       sizeof(*dev_item));
1674         if (ret)
1675                 goto out;
1676
1677         leaf = path->nodes[0];
1678         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1679
1680         btrfs_set_device_id(leaf, dev_item, device->devid);
1681         btrfs_set_device_generation(leaf, dev_item, 0);
1682         btrfs_set_device_type(leaf, dev_item, device->type);
1683         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1684         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1685         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1686         btrfs_set_device_total_bytes(leaf, dev_item,
1687                                      btrfs_device_get_disk_total_bytes(device));
1688         btrfs_set_device_bytes_used(leaf, dev_item,
1689                                     btrfs_device_get_bytes_used(device));
1690         btrfs_set_device_group(leaf, dev_item, 0);
1691         btrfs_set_device_seek_speed(leaf, dev_item, 0);
1692         btrfs_set_device_bandwidth(leaf, dev_item, 0);
1693         btrfs_set_device_start_offset(leaf, dev_item, 0);
1694
1695         ptr = btrfs_device_uuid(dev_item);
1696         write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1697         ptr = btrfs_device_fsid(dev_item);
1698         write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1699         btrfs_mark_buffer_dirty(leaf);
1700
1701         ret = 0;
1702 out:
1703         btrfs_free_path(path);
1704         return ret;
1705 }
1706
1707 /*
1708  * Function to update ctime/mtime for a given device path.
1709  * Mainly used for ctime/mtime based probe like libblkid.
1710  */
1711 static void update_dev_time(char *path_name)
1712 {
1713         struct file *filp;
1714
1715         filp = filp_open(path_name, O_RDWR, 0);
1716         if (IS_ERR(filp))
1717                 return;
1718         file_update_time(filp);
1719         filp_close(filp, NULL);
1720 }
1721
1722 static int btrfs_rm_dev_item(struct btrfs_root *root,
1723                              struct btrfs_device *device)
1724 {
1725         int ret;
1726         struct btrfs_path *path;
1727         struct btrfs_key key;
1728         struct btrfs_trans_handle *trans;
1729
1730         root = root->fs_info->chunk_root;
1731
1732         path = btrfs_alloc_path();
1733         if (!path)
1734                 return -ENOMEM;
1735
1736         trans = btrfs_start_transaction(root, 0);
1737         if (IS_ERR(trans)) {
1738                 btrfs_free_path(path);
1739                 return PTR_ERR(trans);
1740         }
1741         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1742         key.type = BTRFS_DEV_ITEM_KEY;
1743         key.offset = device->devid;
1744
1745         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1746         if (ret < 0)
1747                 goto out;
1748
1749         if (ret > 0) {
1750                 ret = -ENOENT;
1751                 goto out;
1752         }
1753
1754         ret = btrfs_del_item(trans, root, path);
1755         if (ret)
1756                 goto out;
1757 out:
1758         btrfs_free_path(path);
1759         btrfs_commit_transaction(trans, root);
1760         return ret;
1761 }
1762
1763 /*
1764  * Verify that @num_devices satisfies the RAID profile constraints in the whole
1765  * filesystem. It's up to the caller to adjust that number regarding eg. device
1766  * replace.
1767  */
1768 static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info,
1769                 u64 num_devices)
1770 {
1771         u64 all_avail;
1772         unsigned seq;
1773         int i;
1774
1775         do {
1776                 seq = read_seqbegin(&fs_info->profiles_lock);
1777
1778                 all_avail = fs_info->avail_data_alloc_bits |
1779                             fs_info->avail_system_alloc_bits |
1780                             fs_info->avail_metadata_alloc_bits;
1781         } while (read_seqretry(&fs_info->profiles_lock, seq));
1782
1783         for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
1784                 if (!(all_avail & btrfs_raid_group[i]))
1785                         continue;
1786
1787                 if (num_devices < btrfs_raid_array[i].devs_min) {
1788                         int ret = btrfs_raid_mindev_error[i];
1789
1790                         if (ret)
1791                                 return ret;
1792                 }
1793         }
1794
1795         return 0;
1796 }
1797
1798 struct btrfs_device *btrfs_find_next_active_device(struct btrfs_fs_devices *fs_devs,
1799                                         struct btrfs_device *device)
1800 {
1801         struct btrfs_device *next_device;
1802
1803         list_for_each_entry(next_device, &fs_devs->devices, dev_list) {
1804                 if (next_device != device &&
1805                         !next_device->missing && next_device->bdev)
1806                         return next_device;
1807         }
1808
1809         return NULL;
1810 }
1811
1812 /*
1813  * Helper function to check if the given device is part of s_bdev / latest_bdev
1814  * and replace it with the provided or the next active device, in the context
1815  * where this function called, there should be always be another device (or
1816  * this_dev) which is active.
1817  */
1818 void btrfs_assign_next_active_device(struct btrfs_fs_info *fs_info,
1819                 struct btrfs_device *device, struct btrfs_device *this_dev)
1820 {
1821         struct btrfs_device *next_device;
1822
1823         if (this_dev)
1824                 next_device = this_dev;
1825         else
1826                 next_device = btrfs_find_next_active_device(fs_info->fs_devices,
1827                                                                 device);
1828         ASSERT(next_device);
1829
1830         if (fs_info->sb->s_bdev &&
1831                         (fs_info->sb->s_bdev == device->bdev))
1832                 fs_info->sb->s_bdev = next_device->bdev;
1833
1834         if (fs_info->fs_devices->latest_bdev == device->bdev)
1835                 fs_info->fs_devices->latest_bdev = next_device->bdev;
1836 }
1837
1838 int btrfs_rm_device(struct btrfs_root *root, char *device_path, u64 devid)
1839 {
1840         struct btrfs_device *device;
1841         struct btrfs_fs_devices *cur_devices;
1842         u64 num_devices;
1843         int ret = 0;
1844         bool clear_super = false;
1845         char *dev_name = NULL;
1846
1847         mutex_lock(&uuid_mutex);
1848
1849         num_devices = root->fs_info->fs_devices->num_devices;
1850         btrfs_dev_replace_lock(&root->fs_info->dev_replace, 0);
1851         if (btrfs_dev_replace_is_ongoing(&root->fs_info->dev_replace)) {
1852                 WARN_ON(num_devices < 1);
1853                 num_devices--;
1854         }
1855         btrfs_dev_replace_unlock(&root->fs_info->dev_replace, 0);
1856
1857         ret = btrfs_check_raid_min_devices(root->fs_info, num_devices - 1);
1858         if (ret)
1859                 goto out;
1860
1861         ret = btrfs_find_device_by_devspec(root, devid, device_path,
1862                                 &device);
1863         if (ret)
1864                 goto out;
1865
1866         if (device->is_tgtdev_for_dev_replace) {
1867                 ret = BTRFS_ERROR_DEV_TGT_REPLACE;
1868                 goto out;
1869         }
1870
1871         if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1872                 ret = BTRFS_ERROR_DEV_ONLY_WRITABLE;
1873                 goto out;
1874         }
1875
1876         if (device->writeable) {
1877                 lock_chunks(root);
1878                 list_del_init(&device->dev_alloc_list);
1879                 device->fs_devices->rw_devices--;
1880                 unlock_chunks(root);
1881                 dev_name = kstrdup(device->name->str, GFP_KERNEL);
1882                 if (!dev_name) {
1883                         ret = -ENOMEM;
1884                         goto error_undo;
1885                 }
1886                 clear_super = true;
1887         }
1888
1889         mutex_unlock(&uuid_mutex);
1890         ret = btrfs_shrink_device(device, 0);
1891         mutex_lock(&uuid_mutex);
1892         if (ret)
1893                 goto error_undo;
1894
1895         /*
1896          * TODO: the superblock still includes this device in its num_devices
1897          * counter although write_all_supers() is not locked out. This
1898          * could give a filesystem state which requires a degraded mount.
1899          */
1900         ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1901         if (ret)
1902                 goto error_undo;
1903
1904         device->in_fs_metadata = 0;
1905         btrfs_scrub_cancel_dev(root->fs_info, device);
1906
1907         /*
1908          * the device list mutex makes sure that we don't change
1909          * the device list while someone else is writing out all
1910          * the device supers. Whoever is writing all supers, should
1911          * lock the device list mutex before getting the number of
1912          * devices in the super block (super_copy). Conversely,
1913          * whoever updates the number of devices in the super block
1914          * (super_copy) should hold the device list mutex.
1915          */
1916
1917         cur_devices = device->fs_devices;
1918         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1919         list_del_rcu(&device->dev_list);
1920
1921         device->fs_devices->num_devices--;
1922         device->fs_devices->total_devices--;
1923
1924         if (device->missing)
1925                 device->fs_devices->missing_devices--;
1926
1927         btrfs_assign_next_active_device(root->fs_info, device, NULL);
1928
1929         if (device->bdev) {
1930                 device->fs_devices->open_devices--;
1931                 /* remove sysfs entry */
1932                 btrfs_sysfs_rm_device_link(root->fs_info->fs_devices, device);
1933         }
1934
1935         call_rcu(&device->rcu, free_device);
1936
1937         num_devices = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
1938         btrfs_set_super_num_devices(root->fs_info->super_copy, num_devices);
1939         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1940
1941         if (cur_devices->open_devices == 0) {
1942                 struct btrfs_fs_devices *fs_devices;
1943                 fs_devices = root->fs_info->fs_devices;
1944                 while (fs_devices) {
1945                         if (fs_devices->seed == cur_devices) {
1946                                 fs_devices->seed = cur_devices->seed;
1947                                 break;
1948                         }
1949                         fs_devices = fs_devices->seed;
1950                 }
1951                 cur_devices->seed = NULL;
1952                 __btrfs_close_devices(cur_devices);
1953                 free_fs_devices(cur_devices);
1954         }
1955
1956         root->fs_info->num_tolerated_disk_barrier_failures =
1957                 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
1958
1959         /*
1960          * at this point, the device is zero sized.  We want to
1961          * remove it from the devices list and zero out the old super
1962          */
1963         if (clear_super) {
1964                 struct block_device *bdev;
1965
1966                 bdev = blkdev_get_by_path(dev_name, FMODE_READ | FMODE_EXCL,
1967                                                 root->fs_info->bdev_holder);
1968                 if (!IS_ERR(bdev)) {
1969                         btrfs_scratch_superblocks(bdev, dev_name);
1970                         blkdev_put(bdev, FMODE_READ | FMODE_EXCL);
1971                 }
1972         }
1973
1974 out:
1975         kfree(dev_name);
1976
1977         mutex_unlock(&uuid_mutex);
1978         return ret;
1979
1980 error_undo:
1981         if (device->writeable) {
1982                 lock_chunks(root);
1983                 list_add(&device->dev_alloc_list,
1984                          &root->fs_info->fs_devices->alloc_list);
1985                 device->fs_devices->rw_devices++;
1986                 unlock_chunks(root);
1987         }
1988         goto out;
1989 }
1990
1991 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_fs_info *fs_info,
1992                                         struct btrfs_device *srcdev)
1993 {
1994         struct btrfs_fs_devices *fs_devices;
1995
1996         WARN_ON(!mutex_is_locked(&fs_info->fs_devices->device_list_mutex));
1997
1998         /*
1999          * in case of fs with no seed, srcdev->fs_devices will point
2000          * to fs_devices of fs_info. However when the dev being replaced is
2001          * a seed dev it will point to the seed's local fs_devices. In short
2002          * srcdev will have its correct fs_devices in both the cases.
2003          */
2004         fs_devices = srcdev->fs_devices;
2005
2006         list_del_rcu(&srcdev->dev_list);
2007         list_del_rcu(&srcdev->dev_alloc_list);
2008         fs_devices->num_devices--;
2009         if (srcdev->missing)
2010                 fs_devices->missing_devices--;
2011
2012         if (srcdev->writeable)
2013                 fs_devices->rw_devices--;
2014
2015         if (srcdev->bdev)
2016                 fs_devices->open_devices--;
2017 }
2018
2019 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_fs_info *fs_info,
2020                                       struct btrfs_device *srcdev)
2021 {
2022         struct btrfs_fs_devices *fs_devices = srcdev->fs_devices;
2023
2024         if (srcdev->writeable) {
2025                 /* zero out the old super if it is writable */
2026                 btrfs_scratch_superblocks(srcdev->bdev, srcdev->name->str);
2027         }
2028         call_rcu(&srcdev->rcu, free_device);
2029
2030         /*
2031          * unless fs_devices is seed fs, num_devices shouldn't go
2032          * zero
2033          */
2034         BUG_ON(!fs_devices->num_devices && !fs_devices->seeding);
2035
2036         /* if this is no devs we rather delete the fs_devices */
2037         if (!fs_devices->num_devices) {
2038                 struct btrfs_fs_devices *tmp_fs_devices;
2039
2040                 tmp_fs_devices = fs_info->fs_devices;
2041                 while (tmp_fs_devices) {
2042                         if (tmp_fs_devices->seed == fs_devices) {
2043                                 tmp_fs_devices->seed = fs_devices->seed;
2044                                 break;
2045                         }
2046                         tmp_fs_devices = tmp_fs_devices->seed;
2047                 }
2048                 fs_devices->seed = NULL;
2049                 __btrfs_close_devices(fs_devices);
2050                 free_fs_devices(fs_devices);
2051         }
2052 }
2053
2054 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
2055                                       struct btrfs_device *tgtdev)
2056 {
2057         mutex_lock(&uuid_mutex);
2058         WARN_ON(!tgtdev);
2059         mutex_lock(&fs_info->fs_devices->device_list_mutex);
2060
2061         btrfs_sysfs_rm_device_link(fs_info->fs_devices, tgtdev);
2062
2063         if (tgtdev->bdev)
2064                 fs_info->fs_devices->open_devices--;
2065
2066         fs_info->fs_devices->num_devices--;
2067
2068         btrfs_assign_next_active_device(fs_info, tgtdev, NULL);
2069
2070         list_del_rcu(&tgtdev->dev_list);
2071
2072         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2073         mutex_unlock(&uuid_mutex);
2074
2075         /*
2076          * The update_dev_time() with in btrfs_scratch_superblocks()
2077          * may lead to a call to btrfs_show_devname() which will try
2078          * to hold device_list_mutex. And here this device
2079          * is already out of device list, so we don't have to hold
2080          * the device_list_mutex lock.
2081          */
2082         btrfs_scratch_superblocks(tgtdev->bdev, tgtdev->name->str);
2083         call_rcu(&tgtdev->rcu, free_device);
2084 }
2085
2086 static int btrfs_find_device_by_path(struct btrfs_root *root, char *device_path,
2087                                      struct btrfs_device **device)
2088 {
2089         int ret = 0;
2090         struct btrfs_super_block *disk_super;
2091         u64 devid;
2092         u8 *dev_uuid;
2093         struct block_device *bdev;
2094         struct buffer_head *bh;
2095
2096         *device = NULL;
2097         ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
2098                                     root->fs_info->bdev_holder, 0, &bdev, &bh);
2099         if (ret)
2100                 return ret;
2101         disk_super = (struct btrfs_super_block *)bh->b_data;
2102         devid = btrfs_stack_device_id(&disk_super->dev_item);
2103         dev_uuid = disk_super->dev_item.uuid;
2104         *device = btrfs_find_device(root->fs_info, devid, dev_uuid,
2105                                     disk_super->fsid);
2106         brelse(bh);
2107         if (!*device)
2108                 ret = -ENOENT;
2109         blkdev_put(bdev, FMODE_READ);
2110         return ret;
2111 }
2112
2113 int btrfs_find_device_missing_or_by_path(struct btrfs_root *root,
2114                                          char *device_path,
2115                                          struct btrfs_device **device)
2116 {
2117         *device = NULL;
2118         if (strcmp(device_path, "missing") == 0) {
2119                 struct list_head *devices;
2120                 struct btrfs_device *tmp;
2121
2122                 devices = &root->fs_info->fs_devices->devices;
2123                 /*
2124                  * It is safe to read the devices since the volume_mutex
2125                  * is held by the caller.
2126                  */
2127                 list_for_each_entry(tmp, devices, dev_list) {
2128                         if (tmp->in_fs_metadata && !tmp->bdev) {
2129                                 *device = tmp;
2130                                 break;
2131                         }
2132                 }
2133
2134                 if (!*device)
2135                         return BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
2136
2137                 return 0;
2138         } else {
2139                 return btrfs_find_device_by_path(root, device_path, device);
2140         }
2141 }
2142
2143 /*
2144  * Lookup a device given by device id, or the path if the id is 0.
2145  */
2146 int btrfs_find_device_by_devspec(struct btrfs_root *root, u64 devid,
2147                                          char *devpath,
2148                                          struct btrfs_device **device)
2149 {
2150         int ret;
2151
2152         if (devid) {
2153                 ret = 0;
2154                 *device = btrfs_find_device(root->fs_info, devid, NULL,
2155                                             NULL);
2156                 if (!*device)
2157                         ret = -ENOENT;
2158         } else {
2159                 if (!devpath || !devpath[0])
2160                         return -EINVAL;
2161
2162                 ret = btrfs_find_device_missing_or_by_path(root, devpath,
2163                                                            device);
2164         }
2165         return ret;
2166 }
2167
2168 /*
2169  * does all the dirty work required for changing file system's UUID.
2170  */
2171 static int btrfs_prepare_sprout(struct btrfs_root *root)
2172 {
2173         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2174         struct btrfs_fs_devices *old_devices;
2175         struct btrfs_fs_devices *seed_devices;
2176         struct btrfs_super_block *disk_super = root->fs_info->super_copy;
2177         struct btrfs_device *device;
2178         u64 super_flags;
2179
2180         BUG_ON(!mutex_is_locked(&uuid_mutex));
2181         if (!fs_devices->seeding)
2182                 return -EINVAL;
2183
2184         seed_devices = __alloc_fs_devices();
2185         if (IS_ERR(seed_devices))
2186                 return PTR_ERR(seed_devices);
2187
2188         old_devices = clone_fs_devices(fs_devices);
2189         if (IS_ERR(old_devices)) {
2190                 kfree(seed_devices);
2191                 return PTR_ERR(old_devices);
2192         }
2193
2194         list_add(&old_devices->list, &fs_uuids);
2195
2196         memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
2197         seed_devices->opened = 1;
2198         INIT_LIST_HEAD(&seed_devices->devices);
2199         INIT_LIST_HEAD(&seed_devices->alloc_list);
2200         mutex_init(&seed_devices->device_list_mutex);
2201
2202         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2203         list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
2204                               synchronize_rcu);
2205         list_for_each_entry(device, &seed_devices->devices, dev_list)
2206                 device->fs_devices = seed_devices;
2207
2208         lock_chunks(root);
2209         list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
2210         unlock_chunks(root);
2211
2212         fs_devices->seeding = 0;
2213         fs_devices->num_devices = 0;
2214         fs_devices->open_devices = 0;
2215         fs_devices->missing_devices = 0;
2216         fs_devices->rotating = 0;
2217         fs_devices->seed = seed_devices;
2218
2219         generate_random_uuid(fs_devices->fsid);
2220         memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2221         memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2222         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2223
2224         super_flags = btrfs_super_flags(disk_super) &
2225                       ~BTRFS_SUPER_FLAG_SEEDING;
2226         btrfs_set_super_flags(disk_super, super_flags);
2227
2228         return 0;
2229 }
2230
2231 /*
2232  * Store the expected generation for seed devices in device items.
2233  */
2234 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
2235                                struct btrfs_root *root)
2236 {
2237         struct btrfs_path *path;
2238         struct extent_buffer *leaf;
2239         struct btrfs_dev_item *dev_item;
2240         struct btrfs_device *device;
2241         struct btrfs_key key;
2242         u8 fs_uuid[BTRFS_UUID_SIZE];
2243         u8 dev_uuid[BTRFS_UUID_SIZE];
2244         u64 devid;
2245         int ret;
2246
2247         path = btrfs_alloc_path();
2248         if (!path)
2249                 return -ENOMEM;
2250
2251         root = root->fs_info->chunk_root;
2252         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2253         key.offset = 0;
2254         key.type = BTRFS_DEV_ITEM_KEY;
2255
2256         while (1) {
2257                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2258                 if (ret < 0)
2259                         goto error;
2260
2261                 leaf = path->nodes[0];
2262 next_slot:
2263                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2264                         ret = btrfs_next_leaf(root, path);
2265                         if (ret > 0)
2266                                 break;
2267                         if (ret < 0)
2268                                 goto error;
2269                         leaf = path->nodes[0];
2270                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2271                         btrfs_release_path(path);
2272                         continue;
2273                 }
2274
2275                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2276                 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
2277                     key.type != BTRFS_DEV_ITEM_KEY)
2278                         break;
2279
2280                 dev_item = btrfs_item_ptr(leaf, path->slots[0],
2281                                           struct btrfs_dev_item);
2282                 devid = btrfs_device_id(leaf, dev_item);
2283                 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
2284                                    BTRFS_UUID_SIZE);
2285                 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
2286                                    BTRFS_UUID_SIZE);
2287                 device = btrfs_find_device(root->fs_info, devid, dev_uuid,
2288                                            fs_uuid);
2289                 BUG_ON(!device); /* Logic error */
2290
2291                 if (device->fs_devices->seeding) {
2292                         btrfs_set_device_generation(leaf, dev_item,
2293                                                     device->generation);
2294                         btrfs_mark_buffer_dirty(leaf);
2295                 }
2296
2297                 path->slots[0]++;
2298                 goto next_slot;
2299         }
2300         ret = 0;
2301 error:
2302         btrfs_free_path(path);
2303         return ret;
2304 }
2305
2306 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
2307 {
2308         struct request_queue *q;
2309         struct btrfs_trans_handle *trans;
2310         struct btrfs_device *device;
2311         struct block_device *bdev;
2312         struct list_head *devices;
2313         struct super_block *sb = root->fs_info->sb;
2314         struct rcu_string *name;
2315         u64 tmp;
2316         int seeding_dev = 0;
2317         int ret = 0;
2318
2319         if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
2320                 return -EROFS;
2321
2322         bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2323                                   root->fs_info->bdev_holder);
2324         if (IS_ERR(bdev))
2325                 return PTR_ERR(bdev);
2326
2327         if (root->fs_info->fs_devices->seeding) {
2328                 seeding_dev = 1;
2329                 down_write(&sb->s_umount);
2330                 mutex_lock(&uuid_mutex);
2331         }
2332
2333         filemap_write_and_wait(bdev->bd_inode->i_mapping);
2334
2335         devices = &root->fs_info->fs_devices->devices;
2336
2337         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2338         list_for_each_entry(device, devices, dev_list) {
2339                 if (device->bdev == bdev) {
2340                         ret = -EEXIST;
2341                         mutex_unlock(
2342                                 &root->fs_info->fs_devices->device_list_mutex);
2343                         goto error;
2344                 }
2345         }
2346         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2347
2348         device = btrfs_alloc_device(root->fs_info, NULL, NULL);
2349         if (IS_ERR(device)) {
2350                 /* we can safely leave the fs_devices entry around */
2351                 ret = PTR_ERR(device);
2352                 goto error;
2353         }
2354
2355         name = rcu_string_strdup(device_path, GFP_KERNEL);
2356         if (!name) {
2357                 kfree(device);
2358                 ret = -ENOMEM;
2359                 goto error;
2360         }
2361         rcu_assign_pointer(device->name, name);
2362
2363         trans = btrfs_start_transaction(root, 0);
2364         if (IS_ERR(trans)) {
2365                 rcu_string_free(device->name);
2366                 kfree(device);
2367                 ret = PTR_ERR(trans);
2368                 goto error;
2369         }
2370
2371         q = bdev_get_queue(bdev);
2372         if (blk_queue_discard(q))
2373                 device->can_discard = 1;
2374         device->writeable = 1;
2375         device->generation = trans->transid;
2376         device->io_width = root->sectorsize;
2377         device->io_align = root->sectorsize;
2378         device->sector_size = root->sectorsize;
2379         device->total_bytes = i_size_read(bdev->bd_inode);
2380         device->disk_total_bytes = device->total_bytes;
2381         device->commit_total_bytes = device->total_bytes;
2382         device->dev_root = root->fs_info->dev_root;
2383         device->bdev = bdev;
2384         device->in_fs_metadata = 1;
2385         device->is_tgtdev_for_dev_replace = 0;
2386         device->mode = FMODE_EXCL;
2387         device->dev_stats_valid = 1;
2388         set_blocksize(device->bdev, 4096);
2389
2390         if (seeding_dev) {
2391                 sb->s_flags &= ~MS_RDONLY;
2392                 ret = btrfs_prepare_sprout(root);
2393                 BUG_ON(ret); /* -ENOMEM */
2394         }
2395
2396         device->fs_devices = root->fs_info->fs_devices;
2397
2398         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2399         lock_chunks(root);
2400         list_add_rcu(&device->dev_list, &root->fs_info->fs_devices->devices);
2401         list_add(&device->dev_alloc_list,
2402                  &root->fs_info->fs_devices->alloc_list);
2403         root->fs_info->fs_devices->num_devices++;
2404         root->fs_info->fs_devices->open_devices++;
2405         root->fs_info->fs_devices->rw_devices++;
2406         root->fs_info->fs_devices->total_devices++;
2407         root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
2408
2409         spin_lock(&root->fs_info->free_chunk_lock);
2410         root->fs_info->free_chunk_space += device->total_bytes;
2411         spin_unlock(&root->fs_info->free_chunk_lock);
2412
2413         if (!blk_queue_nonrot(bdev_get_queue(bdev)))
2414                 root->fs_info->fs_devices->rotating = 1;
2415
2416         tmp = btrfs_super_total_bytes(root->fs_info->super_copy);
2417         btrfs_set_super_total_bytes(root->fs_info->super_copy,
2418                                     tmp + device->total_bytes);
2419
2420         tmp = btrfs_super_num_devices(root->fs_info->super_copy);
2421         btrfs_set_super_num_devices(root->fs_info->super_copy,
2422                                     tmp + 1);
2423
2424         /* add sysfs device entry */
2425         btrfs_sysfs_add_device_link(root->fs_info->fs_devices, device);
2426
2427         /*
2428          * we've got more storage, clear any full flags on the space
2429          * infos
2430          */
2431         btrfs_clear_space_info_full(root->fs_info);
2432
2433         unlock_chunks(root);
2434         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2435
2436         if (seeding_dev) {
2437                 lock_chunks(root);
2438                 ret = init_first_rw_device(trans, root, device);
2439                 unlock_chunks(root);
2440                 if (ret) {
2441                         btrfs_abort_transaction(trans, ret);
2442                         goto error_trans;
2443                 }
2444         }
2445
2446         ret = btrfs_add_device(trans, root, device);
2447         if (ret) {
2448                 btrfs_abort_transaction(trans, ret);
2449                 goto error_trans;
2450         }
2451
2452         if (seeding_dev) {
2453                 char fsid_buf[BTRFS_UUID_UNPARSED_SIZE];
2454
2455                 ret = btrfs_finish_sprout(trans, root);
2456                 if (ret) {
2457                         btrfs_abort_transaction(trans, ret);
2458                         goto error_trans;
2459                 }
2460
2461                 /* Sprouting would change fsid of the mounted root,
2462                  * so rename the fsid on the sysfs
2463                  */
2464                 snprintf(fsid_buf, BTRFS_UUID_UNPARSED_SIZE, "%pU",
2465                                                 root->fs_info->fsid);
2466                 if (kobject_rename(&root->fs_info->fs_devices->fsid_kobj,
2467                                                                 fsid_buf))
2468                         btrfs_warn(root->fs_info,
2469                                 "sysfs: failed to create fsid for sprout");
2470         }
2471
2472         root->fs_info->num_tolerated_disk_barrier_failures =
2473                 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
2474         ret = btrfs_commit_transaction(trans, root);
2475
2476         if (seeding_dev) {
2477                 mutex_unlock(&uuid_mutex);
2478                 up_write(&sb->s_umount);
2479
2480                 if (ret) /* transaction commit */
2481                         return ret;
2482
2483                 ret = btrfs_relocate_sys_chunks(root);
2484                 if (ret < 0)
2485                         btrfs_handle_fs_error(root->fs_info, ret,
2486                                     "Failed to relocate sys chunks after "
2487                                     "device initialization. This can be fixed "
2488                                     "using the \"btrfs balance\" command.");
2489                 trans = btrfs_attach_transaction(root);
2490                 if (IS_ERR(trans)) {
2491                         if (PTR_ERR(trans) == -ENOENT)
2492                                 return 0;
2493                         return PTR_ERR(trans);
2494                 }
2495                 ret = btrfs_commit_transaction(trans, root);
2496         }
2497
2498         /* Update ctime/mtime for libblkid */
2499         update_dev_time(device_path);
2500         return ret;
2501
2502 error_trans:
2503         btrfs_end_transaction(trans, root);
2504         rcu_string_free(device->name);
2505         btrfs_sysfs_rm_device_link(root->fs_info->fs_devices, device);
2506         kfree(device);
2507 error:
2508         blkdev_put(bdev, FMODE_EXCL);
2509         if (seeding_dev) {
2510                 mutex_unlock(&uuid_mutex);
2511                 up_write(&sb->s_umount);
2512         }
2513         return ret;
2514 }
2515
2516 int btrfs_init_dev_replace_tgtdev(struct btrfs_root *root, char *device_path,
2517                                   struct btrfs_device *srcdev,
2518                                   struct btrfs_device **device_out)
2519 {
2520         struct request_queue *q;
2521         struct btrfs_device *device;
2522         struct block_device *bdev;
2523         struct btrfs_fs_info *fs_info = root->fs_info;
2524         struct list_head *devices;
2525         struct rcu_string *name;
2526         u64 devid = BTRFS_DEV_REPLACE_DEVID;
2527         int ret = 0;
2528
2529         *device_out = NULL;
2530         if (fs_info->fs_devices->seeding) {
2531                 btrfs_err(fs_info, "the filesystem is a seed filesystem!");
2532                 return -EINVAL;
2533         }
2534
2535         bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2536                                   fs_info->bdev_holder);
2537         if (IS_ERR(bdev)) {
2538                 btrfs_err(fs_info, "target device %s is invalid!", device_path);
2539                 return PTR_ERR(bdev);
2540         }
2541
2542         filemap_write_and_wait(bdev->bd_inode->i_mapping);
2543
2544         devices = &fs_info->fs_devices->devices;
2545         list_for_each_entry(device, devices, dev_list) {
2546                 if (device->bdev == bdev) {
2547                         btrfs_err(fs_info, "target device is in the filesystem!");
2548                         ret = -EEXIST;
2549                         goto error;
2550                 }
2551         }
2552
2553
2554         if (i_size_read(bdev->bd_inode) <
2555             btrfs_device_get_total_bytes(srcdev)) {
2556                 btrfs_err(fs_info, "target device is smaller than source device!");
2557                 ret = -EINVAL;
2558                 goto error;
2559         }
2560
2561
2562         device = btrfs_alloc_device(NULL, &devid, NULL);
2563         if (IS_ERR(device)) {
2564                 ret = PTR_ERR(device);
2565                 goto error;
2566         }
2567
2568         name = rcu_string_strdup(device_path, GFP_NOFS);
2569         if (!name) {
2570                 kfree(device);
2571                 ret = -ENOMEM;
2572                 goto error;
2573         }
2574         rcu_assign_pointer(device->name, name);
2575
2576         q = bdev_get_queue(bdev);
2577         if (blk_queue_discard(q))
2578                 device->can_discard = 1;
2579         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2580         device->writeable = 1;
2581         device->generation = 0;
2582         device->io_width = root->sectorsize;
2583         device->io_align = root->sectorsize;
2584         device->sector_size = root->sectorsize;
2585         device->total_bytes = btrfs_device_get_total_bytes(srcdev);
2586         device->disk_total_bytes = btrfs_device_get_disk_total_bytes(srcdev);
2587         device->bytes_used = btrfs_device_get_bytes_used(srcdev);
2588         ASSERT(list_empty(&srcdev->resized_list));
2589         device->commit_total_bytes = srcdev->commit_total_bytes;
2590         device->commit_bytes_used = device->bytes_used;
2591         device->dev_root = fs_info->dev_root;
2592         device->bdev = bdev;
2593         device->in_fs_metadata = 1;
2594         device->is_tgtdev_for_dev_replace = 1;
2595         device->mode = FMODE_EXCL;
2596         device->dev_stats_valid = 1;
2597         set_blocksize(device->bdev, 4096);
2598         device->fs_devices = fs_info->fs_devices;
2599         list_add(&device->dev_list, &fs_info->fs_devices->devices);
2600         fs_info->fs_devices->num_devices++;
2601         fs_info->fs_devices->open_devices++;
2602         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2603
2604         *device_out = device;
2605         return ret;
2606
2607 error:
2608         blkdev_put(bdev, FMODE_EXCL);
2609         return ret;
2610 }
2611
2612 void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info *fs_info,
2613                                               struct btrfs_device *tgtdev)
2614 {
2615         WARN_ON(fs_info->fs_devices->rw_devices == 0);
2616         tgtdev->io_width = fs_info->dev_root->sectorsize;
2617         tgtdev->io_align = fs_info->dev_root->sectorsize;
2618         tgtdev->sector_size = fs_info->dev_root->sectorsize;
2619         tgtdev->dev_root = fs_info->dev_root;
2620         tgtdev->in_fs_metadata = 1;
2621 }
2622
2623 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2624                                         struct btrfs_device *device)
2625 {
2626         int ret;
2627         struct btrfs_path *path;
2628         struct btrfs_root *root;
2629         struct btrfs_dev_item *dev_item;
2630         struct extent_buffer *leaf;
2631         struct btrfs_key key;
2632
2633         root = device->dev_root->fs_info->chunk_root;
2634
2635         path = btrfs_alloc_path();
2636         if (!path)
2637                 return -ENOMEM;
2638
2639         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2640         key.type = BTRFS_DEV_ITEM_KEY;
2641         key.offset = device->devid;
2642
2643         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2644         if (ret < 0)
2645                 goto out;
2646
2647         if (ret > 0) {
2648                 ret = -ENOENT;
2649                 goto out;
2650         }
2651
2652         leaf = path->nodes[0];
2653         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2654
2655         btrfs_set_device_id(leaf, dev_item, device->devid);
2656         btrfs_set_device_type(leaf, dev_item, device->type);
2657         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2658         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2659         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2660         btrfs_set_device_total_bytes(leaf, dev_item,
2661                                      btrfs_device_get_disk_total_bytes(device));
2662         btrfs_set_device_bytes_used(leaf, dev_item,
2663                                     btrfs_device_get_bytes_used(device));
2664         btrfs_mark_buffer_dirty(leaf);
2665
2666 out:
2667         btrfs_free_path(path);
2668         return ret;
2669 }
2670
2671 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2672                       struct btrfs_device *device, u64 new_size)
2673 {
2674         struct btrfs_super_block *super_copy =
2675                 device->dev_root->fs_info->super_copy;
2676         struct btrfs_fs_devices *fs_devices;
2677         u64 old_total;
2678         u64 diff;
2679
2680         if (!device->writeable)
2681                 return -EACCES;
2682
2683         lock_chunks(device->dev_root);
2684         old_total = btrfs_super_total_bytes(super_copy);
2685         diff = new_size - device->total_bytes;
2686
2687         if (new_size <= device->total_bytes ||
2688             device->is_tgtdev_for_dev_replace) {
2689                 unlock_chunks(device->dev_root);
2690                 return -EINVAL;
2691         }
2692
2693         fs_devices = device->dev_root->fs_info->fs_devices;
2694
2695         btrfs_set_super_total_bytes(super_copy, old_total + diff);
2696         device->fs_devices->total_rw_bytes += diff;
2697
2698         btrfs_device_set_total_bytes(device, new_size);
2699         btrfs_device_set_disk_total_bytes(device, new_size);
2700         btrfs_clear_space_info_full(device->dev_root->fs_info);
2701         if (list_empty(&device->resized_list))
2702                 list_add_tail(&device->resized_list,
2703                               &fs_devices->resized_devices);
2704         unlock_chunks(device->dev_root);
2705
2706         return btrfs_update_device(trans, device);
2707 }
2708
2709 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
2710                             struct btrfs_root *root, u64 chunk_objectid,
2711                             u64 chunk_offset)
2712 {
2713         int ret;
2714         struct btrfs_path *path;
2715         struct btrfs_key key;
2716
2717         root = root->fs_info->chunk_root;
2718         path = btrfs_alloc_path();
2719         if (!path)
2720                 return -ENOMEM;
2721
2722         key.objectid = chunk_objectid;
2723         key.offset = chunk_offset;
2724         key.type = BTRFS_CHUNK_ITEM_KEY;
2725
2726         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2727         if (ret < 0)
2728                 goto out;
2729         else if (ret > 0) { /* Logic error or corruption */
2730                 btrfs_handle_fs_error(root->fs_info, -ENOENT,
2731                             "Failed lookup while freeing chunk.");
2732                 ret = -ENOENT;
2733                 goto out;
2734         }
2735
2736         ret = btrfs_del_item(trans, root, path);
2737         if (ret < 0)
2738                 btrfs_handle_fs_error(root->fs_info, ret,
2739                             "Failed to delete chunk item.");
2740 out:
2741         btrfs_free_path(path);
2742         return ret;
2743 }
2744
2745 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
2746                         chunk_offset)
2747 {
2748         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
2749         struct btrfs_disk_key *disk_key;
2750         struct btrfs_chunk *chunk;
2751         u8 *ptr;
2752         int ret = 0;
2753         u32 num_stripes;
2754         u32 array_size;
2755         u32 len = 0;
2756         u32 cur;
2757         struct btrfs_key key;
2758
2759         lock_chunks(root);
2760         array_size = btrfs_super_sys_array_size(super_copy);
2761
2762         ptr = super_copy->sys_chunk_array;
2763         cur = 0;
2764
2765         while (cur < array_size) {
2766                 disk_key = (struct btrfs_disk_key *)ptr;
2767                 btrfs_disk_key_to_cpu(&key, disk_key);
2768
2769                 len = sizeof(*disk_key);
2770
2771                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2772                         chunk = (struct btrfs_chunk *)(ptr + len);
2773                         num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2774                         len += btrfs_chunk_item_size(num_stripes);
2775                 } else {
2776                         ret = -EIO;
2777                         break;
2778                 }
2779                 if (key.objectid == chunk_objectid &&
2780                     key.offset == chunk_offset) {
2781                         memmove(ptr, ptr + len, array_size - (cur + len));
2782                         array_size -= len;
2783                         btrfs_set_super_sys_array_size(super_copy, array_size);
2784                 } else {
2785                         ptr += len;
2786                         cur += len;
2787                 }
2788         }
2789         unlock_chunks(root);
2790         return ret;
2791 }
2792
2793 int btrfs_remove_chunk(struct btrfs_trans_handle *trans,
2794                        struct btrfs_root *root, u64 chunk_offset)
2795 {
2796         struct extent_map_tree *em_tree;
2797         struct extent_map *em;
2798         struct btrfs_root *extent_root = root->fs_info->extent_root;
2799         struct map_lookup *map;
2800         u64 dev_extent_len = 0;
2801         u64 chunk_objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2802         int i, ret = 0;
2803         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2804
2805         /* Just in case */
2806         root = root->fs_info->chunk_root;
2807         em_tree = &root->fs_info->mapping_tree.map_tree;
2808
2809         read_lock(&em_tree->lock);
2810         em = lookup_extent_mapping(em_tree, chunk_offset, 1);
2811         read_unlock(&em_tree->lock);
2812
2813         if (!em || em->start > chunk_offset ||
2814             em->start + em->len < chunk_offset) {
2815                 /*
2816                  * This is a logic error, but we don't want to just rely on the
2817                  * user having built with ASSERT enabled, so if ASSERT doesn't
2818                  * do anything we still error out.
2819                  */
2820                 ASSERT(0);
2821                 if (em)
2822                         free_extent_map(em);
2823                 return -EINVAL;
2824         }
2825         map = em->map_lookup;
2826         lock_chunks(root->fs_info->chunk_root);
2827         check_system_chunk(trans, extent_root, map->type);
2828         unlock_chunks(root->fs_info->chunk_root);
2829
2830         /*
2831          * Take the device list mutex to prevent races with the final phase of
2832          * a device replace operation that replaces the device object associated
2833          * with map stripes (dev-replace.c:btrfs_dev_replace_finishing()).
2834          */
2835         mutex_lock(&fs_devices->device_list_mutex);
2836         for (i = 0; i < map->num_stripes; i++) {
2837                 struct btrfs_device *device = map->stripes[i].dev;
2838                 ret = btrfs_free_dev_extent(trans, device,
2839                                             map->stripes[i].physical,
2840                                             &dev_extent_len);
2841                 if (ret) {
2842                         mutex_unlock(&fs_devices->device_list_mutex);
2843                         btrfs_abort_transaction(trans, ret);
2844                         goto out;
2845                 }
2846
2847                 if (device->bytes_used > 0) {
2848                         lock_chunks(root);
2849                         btrfs_device_set_bytes_used(device,
2850                                         device->bytes_used - dev_extent_len);
2851                         spin_lock(&root->fs_info->free_chunk_lock);
2852                         root->fs_info->free_chunk_space += dev_extent_len;
2853                         spin_unlock(&root->fs_info->free_chunk_lock);
2854                         btrfs_clear_space_info_full(root->fs_info);
2855                         unlock_chunks(root);
2856                 }
2857
2858                 if (map->stripes[i].dev) {
2859                         ret = btrfs_update_device(trans, map->stripes[i].dev);
2860                         if (ret) {
2861                                 mutex_unlock(&fs_devices->device_list_mutex);
2862                                 btrfs_abort_transaction(trans, ret);
2863                                 goto out;
2864                         }
2865                 }
2866         }
2867         mutex_unlock(&fs_devices->device_list_mutex);
2868
2869         ret = btrfs_free_chunk(trans, root, chunk_objectid, chunk_offset);
2870         if (ret) {
2871                 btrfs_abort_transaction(trans, ret);
2872                 goto out;
2873         }
2874
2875         trace_btrfs_chunk_free(root, map, chunk_offset, em->len);
2876
2877         if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2878                 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
2879                 if (ret) {
2880                         btrfs_abort_transaction(trans, ret);
2881                         goto out;
2882                 }
2883         }
2884
2885         ret = btrfs_remove_block_group(trans, extent_root, chunk_offset, em);
2886         if (ret) {
2887                 btrfs_abort_transaction(trans, ret);
2888                 goto out;
2889         }
2890
2891 out:
2892         /* once for us */
2893         free_extent_map(em);
2894         return ret;
2895 }
2896
2897 static int btrfs_relocate_chunk(struct btrfs_root *root, u64 chunk_offset)
2898 {
2899         struct btrfs_root *extent_root;
2900         struct btrfs_trans_handle *trans;
2901         int ret;
2902
2903         root = root->fs_info->chunk_root;
2904         extent_root = root->fs_info->extent_root;
2905
2906         /*
2907          * Prevent races with automatic removal of unused block groups.
2908          * After we relocate and before we remove the chunk with offset
2909          * chunk_offset, automatic removal of the block group can kick in,
2910          * resulting in a failure when calling btrfs_remove_chunk() below.
2911          *
2912          * Make sure to acquire this mutex before doing a tree search (dev
2913          * or chunk trees) to find chunks. Otherwise the cleaner kthread might
2914          * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after
2915          * we release the path used to search the chunk/dev tree and before
2916          * the current task acquires this mutex and calls us.
2917          */
2918         ASSERT(mutex_is_locked(&root->fs_info->delete_unused_bgs_mutex));
2919
2920         ret = btrfs_can_relocate(extent_root, chunk_offset);
2921         if (ret)
2922                 return -ENOSPC;
2923
2924         /* step one, relocate all the extents inside this chunk */
2925         btrfs_scrub_pause(root);
2926         ret = btrfs_relocate_block_group(extent_root, chunk_offset);
2927         btrfs_scrub_continue(root);
2928         if (ret)
2929                 return ret;
2930
2931         trans = btrfs_start_trans_remove_block_group(root->fs_info,
2932                                                      chunk_offset);
2933         if (IS_ERR(trans)) {
2934                 ret = PTR_ERR(trans);
2935                 btrfs_handle_fs_error(root->fs_info, ret, NULL);
2936                 return ret;
2937         }
2938
2939         /*
2940          * step two, delete the device extents and the
2941          * chunk tree entries
2942          */
2943         ret = btrfs_remove_chunk(trans, root, chunk_offset);
2944         btrfs_end_transaction(trans, extent_root);
2945         return ret;
2946 }
2947
2948 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
2949 {
2950         struct btrfs_root *chunk_root = root->fs_info->chunk_root;
2951         struct btrfs_path *path;
2952         struct extent_buffer *leaf;
2953         struct btrfs_chunk *chunk;
2954         struct btrfs_key key;
2955         struct btrfs_key found_key;
2956         u64 chunk_type;
2957         bool retried = false;
2958         int failed = 0;
2959         int ret;
2960
2961         path = btrfs_alloc_path();
2962         if (!path)
2963                 return -ENOMEM;
2964
2965 again:
2966         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2967         key.offset = (u64)-1;
2968         key.type = BTRFS_CHUNK_ITEM_KEY;
2969
2970         while (1) {
2971                 mutex_lock(&root->fs_info->delete_unused_bgs_mutex);
2972                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2973                 if (ret < 0) {
2974                         mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
2975                         goto error;
2976                 }
2977                 BUG_ON(ret == 0); /* Corruption */
2978
2979                 ret = btrfs_previous_item(chunk_root, path, key.objectid,
2980                                           key.type);
2981                 if (ret)
2982                         mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
2983                 if (ret < 0)
2984                         goto error;
2985                 if (ret > 0)
2986                         break;
2987
2988                 leaf = path->nodes[0];
2989                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2990
2991                 chunk = btrfs_item_ptr(leaf, path->slots[0],
2992                                        struct btrfs_chunk);
2993                 chunk_type = btrfs_chunk_type(leaf, chunk);
2994                 btrfs_release_path(path);
2995
2996                 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
2997                         ret = btrfs_relocate_chunk(chunk_root,
2998                                                    found_key.offset);
2999                         if (ret == -ENOSPC)
3000                                 failed++;
3001                         else
3002                                 BUG_ON(ret);
3003                 }
3004                 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
3005
3006                 if (found_key.offset == 0)
3007                         break;
3008                 key.offset = found_key.offset - 1;
3009         }
3010         ret = 0;
3011         if (failed && !retried) {
3012                 failed = 0;
3013                 retried = true;
3014                 goto again;
3015         } else if (WARN_ON(failed && retried)) {
3016                 ret = -ENOSPC;
3017         }
3018 error:
3019         btrfs_free_path(path);
3020         return ret;
3021 }
3022
3023 static int insert_balance_item(struct btrfs_root *root,
3024                                struct btrfs_balance_control *bctl)
3025 {
3026         struct btrfs_trans_handle *trans;
3027         struct btrfs_balance_item *item;
3028         struct btrfs_disk_balance_args disk_bargs;
3029         struct btrfs_path *path;
3030         struct extent_buffer *leaf;
3031         struct btrfs_key key;
3032         int ret, err;
3033
3034         path = btrfs_alloc_path();
3035         if (!path)
3036                 return -ENOMEM;
3037
3038         trans = btrfs_start_transaction(root, 0);
3039         if (IS_ERR(trans)) {
3040                 btrfs_free_path(path);
3041                 return PTR_ERR(trans);
3042         }
3043
3044         key.objectid = BTRFS_BALANCE_OBJECTID;
3045         key.type = BTRFS_TEMPORARY_ITEM_KEY;
3046         key.offset = 0;
3047
3048         ret = btrfs_insert_empty_item(trans, root, path, &key,
3049                                       sizeof(*item));
3050         if (ret)
3051                 goto out;
3052
3053         leaf = path->nodes[0];
3054         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3055
3056         memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
3057
3058         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
3059         btrfs_set_balance_data(leaf, item, &disk_bargs);
3060         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
3061         btrfs_set_balance_meta(leaf, item, &disk_bargs);
3062         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
3063         btrfs_set_balance_sys(leaf, item, &disk_bargs);
3064
3065         btrfs_set_balance_flags(leaf, item, bctl->flags);
3066
3067         btrfs_mark_buffer_dirty(leaf);
3068 out:
3069         btrfs_free_path(path);
3070         err = btrfs_commit_transaction(trans, root);
3071         if (err && !ret)
3072                 ret = err;
3073         return ret;
3074 }
3075
3076 static int del_balance_item(struct btrfs_root *root)
3077 {
3078         struct btrfs_trans_handle *trans;
3079         struct btrfs_path *path;
3080         struct btrfs_key key;
3081         int ret, err;
3082
3083         path = btrfs_alloc_path();
3084         if (!path)
3085                 return -ENOMEM;
3086
3087         trans = btrfs_start_transaction(root, 0);
3088         if (IS_ERR(trans)) {
3089                 btrfs_free_path(path);
3090                 return PTR_ERR(trans);
3091         }
3092
3093         key.objectid = BTRFS_BALANCE_OBJECTID;
3094         key.type = BTRFS_TEMPORARY_ITEM_KEY;
3095         key.offset = 0;
3096
3097         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3098         if (ret < 0)
3099                 goto out;
3100         if (ret > 0) {
3101                 ret = -ENOENT;
3102                 goto out;
3103         }
3104
3105         ret = btrfs_del_item(trans, root, path);
3106 out:
3107         btrfs_free_path(path);
3108         err = btrfs_commit_transaction(trans, root);
3109         if (err && !ret)
3110                 ret = err;
3111         return ret;
3112 }
3113
3114 /*
3115  * This is a heuristic used to reduce the number of chunks balanced on
3116  * resume after balance was interrupted.
3117  */
3118 static void update_balance_args(struct btrfs_balance_control *bctl)
3119 {
3120         /*
3121          * Turn on soft mode for chunk types that were being converted.
3122          */
3123         if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
3124                 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
3125         if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
3126                 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
3127         if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
3128                 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
3129
3130         /*
3131          * Turn on usage filter if is not already used.  The idea is
3132          * that chunks that we have already balanced should be
3133          * reasonably full.  Don't do it for chunks that are being
3134          * converted - that will keep us from relocating unconverted
3135          * (albeit full) chunks.
3136          */
3137         if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3138             !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3139             !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3140                 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
3141                 bctl->data.usage = 90;
3142         }
3143         if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3144             !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3145             !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3146                 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
3147                 bctl->sys.usage = 90;
3148         }
3149         if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3150             !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3151             !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3152                 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
3153                 bctl->meta.usage = 90;
3154         }
3155 }
3156
3157 /*
3158  * Should be called with both balance and volume mutexes held to
3159  * serialize other volume operations (add_dev/rm_dev/resize) with
3160  * restriper.  Same goes for unset_balance_control.
3161  */
3162 static void set_balance_control(struct btrfs_balance_control *bctl)
3163 {
3164         struct btrfs_fs_info *fs_info = bctl->fs_info;
3165
3166         BUG_ON(fs_info->balance_ctl);
3167
3168         spin_lock(&fs_info->balance_lock);
3169         fs_info->balance_ctl = bctl;
3170         spin_unlock(&fs_info->balance_lock);
3171 }
3172
3173 static void unset_balance_control(struct btrfs_fs_info *fs_info)
3174 {
3175         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3176
3177         BUG_ON(!fs_info->balance_ctl);
3178
3179         spin_lock(&fs_info->balance_lock);
3180         fs_info->balance_ctl = NULL;
3181         spin_unlock(&fs_info->balance_lock);
3182
3183         kfree(bctl);
3184 }
3185
3186 /*
3187  * Balance filters.  Return 1 if chunk should be filtered out
3188  * (should not be balanced).
3189  */
3190 static int chunk_profiles_filter(u64 chunk_type,
3191                                  struct btrfs_balance_args *bargs)
3192 {
3193         chunk_type = chunk_to_extended(chunk_type) &
3194                                 BTRFS_EXTENDED_PROFILE_MASK;
3195
3196         if (bargs->profiles & chunk_type)
3197                 return 0;
3198
3199         return 1;
3200 }
3201
3202 static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
3203                               struct btrfs_balance_args *bargs)
3204 {
3205         struct btrfs_block_group_cache *cache;
3206         u64 chunk_used;
3207         u64 user_thresh_min;
3208         u64 user_thresh_max;
3209         int ret = 1;
3210
3211         cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3212         chunk_used = btrfs_block_group_used(&cache->item);
3213
3214         if (bargs->usage_min == 0)
3215                 user_thresh_min = 0;
3216         else
3217                 user_thresh_min = div_factor_fine(cache->key.offset,
3218                                         bargs->usage_min);
3219
3220         if (bargs->usage_max == 0)
3221                 user_thresh_max = 1;
3222         else if (bargs->usage_max > 100)
3223                 user_thresh_max = cache->key.offset;
3224         else
3225                 user_thresh_max = div_factor_fine(cache->key.offset,
3226                                         bargs->usage_max);
3227
3228         if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max)
3229                 ret = 0;
3230
3231         btrfs_put_block_group(cache);
3232         return ret;
3233 }
3234
3235 static int chunk_usage_filter(struct btrfs_fs_info *fs_info,
3236                 u64 chunk_offset, struct btrfs_balance_args *bargs)
3237 {
3238         struct btrfs_block_group_cache *cache;
3239         u64 chunk_used, user_thresh;
3240         int ret = 1;
3241
3242         cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3243         chunk_used = btrfs_block_group_used(&cache->item);
3244
3245         if (bargs->usage_min == 0)
3246                 user_thresh = 1;
3247         else if (bargs->usage > 100)
3248                 user_thresh = cache->key.offset;
3249         else
3250                 user_thresh = div_factor_fine(cache->key.offset,
3251                                               bargs->usage);
3252
3253         if (chunk_used < user_thresh)
3254                 ret = 0;
3255
3256         btrfs_put_block_group(cache);
3257         return ret;
3258 }
3259
3260 static int chunk_devid_filter(struct extent_buffer *leaf,
3261                               struct btrfs_chunk *chunk,
3262                               struct btrfs_balance_args *bargs)
3263 {
3264         struct btrfs_stripe *stripe;
3265         int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3266         int i;
3267
3268         for (i = 0; i < num_stripes; i++) {
3269                 stripe = btrfs_stripe_nr(chunk, i);
3270                 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
3271                         return 0;
3272         }
3273
3274         return 1;
3275 }
3276
3277 /* [pstart, pend) */
3278 static int chunk_drange_filter(struct extent_buffer *leaf,
3279                                struct btrfs_chunk *chunk,
3280                                u64 chunk_offset,
3281                                struct btrfs_balance_args *bargs)
3282 {
3283         struct btrfs_stripe *stripe;
3284         int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3285         u64 stripe_offset;
3286         u64 stripe_length;
3287         int factor;
3288         int i;
3289
3290         if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
3291                 return 0;
3292
3293         if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP |
3294              BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) {
3295                 factor = num_stripes / 2;
3296         } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) {
3297                 factor = num_stripes - 1;
3298         } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) {
3299                 factor = num_stripes - 2;
3300         } else {
3301                 factor = num_stripes;
3302         }
3303
3304         for (i = 0; i < num_stripes; i++) {
3305                 stripe = btrfs_stripe_nr(chunk, i);
3306                 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
3307                         continue;
3308
3309                 stripe_offset = btrfs_stripe_offset(leaf, stripe);
3310                 stripe_length = btrfs_chunk_length(leaf, chunk);
3311                 stripe_length = div_u64(stripe_length, factor);
3312
3313                 if (stripe_offset < bargs->pend &&
3314                     stripe_offset + stripe_length > bargs->pstart)
3315                         return 0;
3316         }
3317
3318         return 1;
3319 }
3320
3321 /* [vstart, vend) */
3322 static int chunk_vrange_filter(struct extent_buffer *leaf,
3323                                struct btrfs_chunk *chunk,
3324                                u64 chunk_offset,
3325                                struct btrfs_balance_args *bargs)
3326 {
3327         if (chunk_offset < bargs->vend &&
3328             chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
3329                 /* at least part of the chunk is inside this vrange */
3330                 return 0;
3331
3332         return 1;
3333 }
3334
3335 static int chunk_stripes_range_filter(struct extent_buffer *leaf,
3336                                struct btrfs_chunk *chunk,
3337                                struct btrfs_balance_args *bargs)
3338 {
3339         int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3340
3341         if (bargs->stripes_min <= num_stripes
3342                         && num_stripes <= bargs->stripes_max)
3343                 return 0;
3344
3345         return 1;
3346 }
3347
3348 static int chunk_soft_convert_filter(u64 chunk_type,
3349                                      struct btrfs_balance_args *bargs)
3350 {
3351         if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
3352                 return 0;
3353
3354         chunk_type = chunk_to_extended(chunk_type) &
3355                                 BTRFS_EXTENDED_PROFILE_MASK;
3356
3357         if (bargs->target == chunk_type)
3358                 return 1;
3359
3360         return 0;
3361 }
3362
3363 static int should_balance_chunk(struct btrfs_root *root,
3364                                 struct extent_buffer *leaf,
3365                                 struct btrfs_chunk *chunk, u64 chunk_offset)
3366 {
3367         struct btrfs_balance_control *bctl = root->fs_info->balance_ctl;
3368         struct btrfs_balance_args *bargs = NULL;
3369         u64 chunk_type = btrfs_chunk_type(leaf, chunk);
3370
3371         /* type filter */
3372         if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
3373               (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
3374                 return 0;
3375         }
3376
3377         if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3378                 bargs = &bctl->data;
3379         else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3380                 bargs = &bctl->sys;
3381         else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3382                 bargs = &bctl->meta;
3383
3384         /* profiles filter */
3385         if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
3386             chunk_profiles_filter(chunk_type, bargs)) {
3387                 return 0;
3388         }
3389
3390         /* usage filter */
3391         if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
3392             chunk_usage_filter(bctl->fs_info, chunk_offset, bargs)) {
3393                 return 0;
3394         } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3395             chunk_usage_range_filter(bctl->fs_info, chunk_offset, bargs)) {
3396                 return 0;
3397         }
3398
3399         /* devid filter */
3400         if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
3401             chunk_devid_filter(leaf, chunk, bargs)) {
3402                 return 0;
3403         }
3404
3405         /* drange filter, makes sense only with devid filter */
3406         if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
3407             chunk_drange_filter(leaf, chunk, chunk_offset, bargs)) {
3408                 return 0;
3409         }
3410
3411         /* vrange filter */
3412         if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
3413             chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
3414                 return 0;
3415         }
3416
3417         /* stripes filter */
3418         if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) &&
3419             chunk_stripes_range_filter(leaf, chunk, bargs)) {
3420                 return 0;
3421         }
3422
3423         /* soft profile changing mode */
3424         if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
3425             chunk_soft_convert_filter(chunk_type, bargs)) {
3426                 return 0;
3427         }
3428
3429         /*
3430          * limited by count, must be the last filter
3431          */
3432         if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) {
3433                 if (bargs->limit == 0)
3434                         return 0;
3435                 else
3436                         bargs->limit--;
3437         } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) {
3438                 /*
3439                  * Same logic as the 'limit' filter; the minimum cannot be
3440                  * determined here because we do not have the global information
3441                  * about the count of all chunks that satisfy the filters.
3442                  */
3443                 if (bargs->limit_max == 0)
3444                         return 0;
3445                 else
3446                         bargs->limit_max--;
3447         }
3448
3449         return 1;
3450 }
3451
3452 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
3453 {
3454         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3455         struct btrfs_root *chunk_root = fs_info->chunk_root;
3456         struct btrfs_root *dev_root = fs_info->dev_root;
3457         struct list_head *devices;
3458         struct btrfs_device *device;
3459         u64 old_size;
3460         u64 size_to_free;
3461         u64 chunk_type;
3462         struct btrfs_chunk *chunk;
3463         struct btrfs_path *path = NULL;
3464         struct btrfs_key key;
3465         struct btrfs_key found_key;
3466         struct btrfs_trans_handle *trans;
3467         struct extent_buffer *leaf;
3468         int slot;
3469         int ret;
3470         int enospc_errors = 0;
3471         bool counting = true;
3472         /* The single value limit and min/max limits use the same bytes in the */
3473         u64 limit_data = bctl->data.limit;
3474         u64 limit_meta = bctl->meta.limit;
3475         u64 limit_sys = bctl->sys.limit;
3476         u32 count_data = 0;
3477         u32 count_meta = 0;
3478         u32 count_sys = 0;
3479         int chunk_reserved = 0;
3480         u64 bytes_used = 0;
3481
3482         /* step one make some room on all the devices */
3483         devices = &fs_info->fs_devices->devices;
3484         list_for_each_entry(device, devices, dev_list) {
3485                 old_size = btrfs_device_get_total_bytes(device);
3486                 size_to_free = div_factor(old_size, 1);
3487                 size_to_free = min_t(u64, size_to_free, SZ_1M);
3488                 if (!device->writeable ||
3489                     btrfs_device_get_total_bytes(device) -
3490                     btrfs_device_get_bytes_used(device) > size_to_free ||
3491                     device->is_tgtdev_for_dev_replace)
3492                         continue;
3493
3494                 ret = btrfs_shrink_device(device, old_size - size_to_free);
3495                 if (ret == -ENOSPC)
3496                         break;
3497                 if (ret) {
3498                         /* btrfs_shrink_device never returns ret > 0 */
3499                         WARN_ON(ret > 0);
3500                         goto error;
3501                 }
3502
3503                 trans = btrfs_start_transaction(dev_root, 0);
3504                 if (IS_ERR(trans)) {
3505                         ret = PTR_ERR(trans);
3506                         btrfs_info_in_rcu(fs_info,
3507                  "resize: unable to start transaction after shrinking device %s (error %d), old size %llu, new size %llu",
3508                                           rcu_str_deref(device->name), ret,
3509                                           old_size, old_size - size_to_free);
3510                         goto error;
3511                 }
3512
3513                 ret = btrfs_grow_device(trans, device, old_size);
3514                 if (ret) {
3515                         btrfs_end_transaction(trans, dev_root);
3516                         /* btrfs_grow_device never returns ret > 0 */
3517                         WARN_ON(ret > 0);
3518                         btrfs_info_in_rcu(fs_info,
3519                  "resize: unable to grow device after shrinking device %s (error %d), old size %llu, new size %llu",
3520                                           rcu_str_deref(device->name), ret,
3521                                           old_size, old_size - size_to_free);
3522                         goto error;
3523                 }
3524
3525                 btrfs_end_transaction(trans, dev_root);
3526         }
3527
3528         /* step two, relocate all the chunks */
3529         path = btrfs_alloc_path();
3530         if (!path) {
3531                 ret = -ENOMEM;
3532                 goto error;
3533         }
3534
3535         /* zero out stat counters */
3536         spin_lock(&fs_info->balance_lock);
3537         memset(&bctl->stat, 0, sizeof(bctl->stat));
3538         spin_unlock(&fs_info->balance_lock);
3539 again:
3540         if (!counting) {
3541                 /*
3542                  * The single value limit and min/max limits use the same bytes
3543                  * in the
3544                  */
3545                 bctl->data.limit = limit_data;
3546                 bctl->meta.limit = limit_meta;
3547                 bctl->sys.limit = limit_sys;
3548         }
3549         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3550         key.offset = (u64)-1;
3551         key.type = BTRFS_CHUNK_ITEM_KEY;
3552
3553         while (1) {
3554                 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
3555                     atomic_read(&fs_info->balance_cancel_req)) {
3556                         ret = -ECANCELED;
3557                         goto error;
3558                 }
3559
3560                 mutex_lock(&fs_info->delete_unused_bgs_mutex);
3561                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3562                 if (ret < 0) {
3563                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3564                         goto error;
3565                 }
3566
3567                 /*
3568                  * this shouldn't happen, it means the last relocate
3569                  * failed
3570                  */
3571                 if (ret == 0)
3572                         BUG(); /* FIXME break ? */
3573
3574                 ret = btrfs_previous_item(chunk_root, path, 0,
3575                                           BTRFS_CHUNK_ITEM_KEY);
3576                 if (ret) {
3577                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3578                         ret = 0;
3579                         break;
3580                 }
3581
3582                 leaf = path->nodes[0];
3583                 slot = path->slots[0];
3584                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3585
3586                 if (found_key.objectid != key.objectid) {
3587                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3588                         break;
3589                 }
3590
3591                 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3592                 chunk_type = btrfs_chunk_type(leaf, chunk);
3593
3594                 if (!counting) {
3595                         spin_lock(&fs_info->balance_lock);
3596                         bctl->stat.considered++;
3597                         spin_unlock(&fs_info->balance_lock);
3598                 }
3599
3600                 ret = should_balance_chunk(chunk_root, leaf, chunk,
3601                                            found_key.offset);
3602
3603                 btrfs_release_path(path);
3604                 if (!ret) {
3605                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3606                         goto loop;
3607                 }
3608
3609                 if (counting) {
3610                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3611                         spin_lock(&fs_info->balance_lock);
3612                         bctl->stat.expected++;
3613                         spin_unlock(&fs_info->balance_lock);
3614
3615                         if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3616                                 count_data++;
3617                         else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3618                                 count_sys++;
3619                         else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3620                                 count_meta++;
3621
3622                         goto loop;
3623                 }
3624
3625                 /*
3626                  * Apply limit_min filter, no need to check if the LIMITS
3627                  * filter is used, limit_min is 0 by default
3628                  */
3629                 if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
3630                                         count_data < bctl->data.limit_min)
3631                                 || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) &&
3632                                         count_meta < bctl->meta.limit_min)
3633                                 || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) &&
3634                                         count_sys < bctl->sys.limit_min)) {
3635                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3636                         goto loop;
3637                 }
3638
3639                 ASSERT(fs_info->data_sinfo);
3640                 spin_lock(&fs_info->data_sinfo->lock);
3641                 bytes_used = fs_info->data_sinfo->bytes_used;
3642                 spin_unlock(&fs_info->data_sinfo->lock);
3643
3644                 if ((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
3645                     !chunk_reserved && !bytes_used) {
3646                         trans = btrfs_start_transaction(chunk_root, 0);
3647                         if (IS_ERR(trans)) {
3648                                 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3649                                 ret = PTR_ERR(trans);
3650                                 goto error;
3651                         }
3652
3653                         ret = btrfs_force_chunk_alloc(trans, chunk_root,
3654                                                       BTRFS_BLOCK_GROUP_DATA);
3655                         btrfs_end_transaction(trans, chunk_root);
3656                         if (ret < 0) {
3657                                 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3658                                 goto error;
3659                         }
3660                         chunk_reserved = 1;
3661                 }
3662
3663                 ret = btrfs_relocate_chunk(chunk_root,
3664                                            found_key.offset);
3665                 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3666                 if (ret && ret != -ENOSPC)
3667                         goto error;
3668                 if (ret == -ENOSPC) {
3669                         enospc_errors++;
3670                 } else {
3671                         spin_lock(&fs_info->balance_lock);
3672                         bctl->stat.completed++;
3673                         spin_unlock(&fs_info->balance_lock);
3674                 }
3675 loop:
3676                 if (found_key.offset == 0)
3677                         break;
3678                 key.offset = found_key.offset - 1;
3679         }
3680
3681         if (counting) {
3682                 btrfs_release_path(path);
3683                 counting = false;
3684                 goto again;
3685         }
3686 error:
3687         btrfs_free_path(path);
3688         if (enospc_errors) {
3689                 btrfs_info(fs_info, "%d enospc errors during balance",
3690                        enospc_errors);
3691                 if (!ret)
3692                         ret = -ENOSPC;
3693         }
3694
3695         return ret;
3696 }
3697
3698 /**
3699  * alloc_profile_is_valid - see if a given profile is valid and reduced
3700  * @flags: profile to validate
3701  * @extended: if true @flags is treated as an extended profile
3702  */
3703 static int alloc_profile_is_valid(u64 flags, int extended)
3704 {
3705         u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
3706                                BTRFS_BLOCK_GROUP_PROFILE_MASK);
3707
3708         flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
3709
3710         /* 1) check that all other bits are zeroed */
3711         if (flags & ~mask)
3712                 return 0;
3713
3714         /* 2) see if profile is reduced */
3715         if (flags == 0)
3716                 return !extended; /* "0" is valid for usual profiles */
3717
3718         /* true if exactly one bit set */
3719         return (flags & (flags - 1)) == 0;
3720 }
3721
3722 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
3723 {
3724         /* cancel requested || normal exit path */
3725         return atomic_read(&fs_info->balance_cancel_req) ||
3726                 (atomic_read(&fs_info->balance_pause_req) == 0 &&
3727                  atomic_read(&fs_info->balance_cancel_req) == 0);
3728 }
3729
3730 static void __cancel_balance(struct btrfs_fs_info *fs_info)
3731 {
3732         int ret;
3733
3734         unset_balance_control(fs_info);
3735         ret = del_balance_item(fs_info->tree_root);
3736         if (ret)
3737                 btrfs_handle_fs_error(fs_info, ret, NULL);
3738
3739         atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3740 }
3741
3742 /* Non-zero return value signifies invalidity */
3743 static inline int validate_convert_profile(struct btrfs_balance_args *bctl_arg,
3744                 u64 allowed)
3745 {
3746         return ((bctl_arg->flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3747                 (!alloc_profile_is_valid(bctl_arg->target, 1) ||
3748                  (bctl_arg->target & ~allowed)));
3749 }
3750
3751 /*
3752  * Should be called with both balance and volume mutexes held
3753  */
3754 int btrfs_balance(struct btrfs_balance_control *bctl,
3755                   struct btrfs_ioctl_balance_args *bargs)
3756 {
3757         struct btrfs_fs_info *fs_info = bctl->fs_info;
3758         u64 allowed;
3759         int mixed = 0;
3760         int ret;
3761         u64 num_devices;
3762         unsigned seq;
3763
3764         if (btrfs_fs_closing(fs_info) ||
3765             atomic_read(&fs_info->balance_pause_req) ||
3766             atomic_read(&fs_info->balance_cancel_req)) {
3767                 ret = -EINVAL;
3768                 goto out;
3769         }
3770
3771         allowed = btrfs_super_incompat_flags(fs_info->super_copy);
3772         if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
3773                 mixed = 1;
3774
3775         /*
3776          * In case of mixed groups both data and meta should be picked,
3777          * and identical options should be given for both of them.
3778          */
3779         allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
3780         if (mixed && (bctl->flags & allowed)) {
3781                 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
3782                     !(bctl->flags & BTRFS_BALANCE_METADATA) ||
3783                     memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
3784                         btrfs_err(fs_info, "with mixed groups data and "
3785                                    "metadata balance options must be the same");
3786                         ret = -EINVAL;
3787                         goto out;
3788                 }
3789         }
3790
3791         num_devices = fs_info->fs_devices->num_devices;
3792         btrfs_dev_replace_lock(&fs_info->dev_replace, 0);
3793         if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
3794                 BUG_ON(num_devices < 1);
3795                 num_devices--;
3796         }
3797         btrfs_dev_replace_unlock(&fs_info->dev_replace, 0);
3798         allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE | BTRFS_BLOCK_GROUP_DUP;
3799         if (num_devices > 1)
3800                 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1);
3801         if (num_devices > 2)
3802                 allowed |= BTRFS_BLOCK_GROUP_RAID5;
3803         if (num_devices > 3)
3804                 allowed |= (BTRFS_BLOCK_GROUP_RAID10 |
3805                             BTRFS_BLOCK_GROUP_RAID6);
3806         if (validate_convert_profile(&bctl->data, allowed)) {
3807                 btrfs_err(fs_info, "unable to start balance with target "
3808                            "data profile %llu",
3809                        bctl->data.target);
3810                 ret = -EINVAL;
3811                 goto out;
3812         }
3813         if (validate_convert_profile(&bctl->meta, allowed)) {
3814                 btrfs_err(fs_info,
3815                            "unable to start balance with target metadata profile %llu",
3816                        bctl->meta.target);
3817                 ret = -EINVAL;
3818                 goto out;
3819         }
3820         if (validate_convert_profile(&bctl->sys, allowed)) {
3821                 btrfs_err(fs_info,
3822                            "unable to start balance with target system profile %llu",
3823                        bctl->sys.target);
3824                 ret = -EINVAL;
3825                 goto out;
3826         }
3827
3828         /* allow to reduce meta or sys integrity only if force set */
3829         allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3830                         BTRFS_BLOCK_GROUP_RAID10 |
3831                         BTRFS_BLOCK_GROUP_RAID5 |
3832                         BTRFS_BLOCK_GROUP_RAID6;
3833         do {
3834                 seq = read_seqbegin(&fs_info->profiles_lock);
3835
3836                 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3837                      (fs_info->avail_system_alloc_bits & allowed) &&
3838                      !(bctl->sys.target & allowed)) ||
3839                     ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3840                      (fs_info->avail_metadata_alloc_bits & allowed) &&
3841                      !(bctl->meta.target & allowed))) {
3842                         if (bctl->flags & BTRFS_BALANCE_FORCE) {
3843                                 btrfs_info(fs_info, "force reducing metadata integrity");
3844                         } else {
3845                                 btrfs_err(fs_info, "balance will reduce metadata "
3846                                            "integrity, use force if you want this");
3847                                 ret = -EINVAL;
3848                                 goto out;
3849                         }
3850                 }
3851         } while (read_seqretry(&fs_info->profiles_lock, seq));
3852
3853         if (btrfs_get_num_tolerated_disk_barrier_failures(bctl->meta.target) <
3854                 btrfs_get_num_tolerated_disk_barrier_failures(bctl->data.target)) {
3855                 btrfs_warn(fs_info,
3856         "metadata profile 0x%llx has lower redundancy than data profile 0x%llx",
3857                         bctl->meta.target, bctl->data.target);
3858         }
3859
3860         if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3861                 fs_info->num_tolerated_disk_barrier_failures = min(
3862                         btrfs_calc_num_tolerated_disk_barrier_failures(fs_info),
3863                         btrfs_get_num_tolerated_disk_barrier_failures(
3864                                 bctl->sys.target));
3865         }
3866
3867         ret = insert_balance_item(fs_info->tree_root, bctl);
3868         if (ret && ret != -EEXIST)
3869                 goto out;
3870
3871         if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
3872                 BUG_ON(ret == -EEXIST);
3873                 set_balance_control(bctl);
3874         } else {
3875                 BUG_ON(ret != -EEXIST);
3876                 spin_lock(&fs_info->balance_lock);
3877                 update_balance_args(bctl);
3878                 spin_unlock(&fs_info->balance_lock);
3879         }
3880
3881         atomic_inc(&fs_info->balance_running);
3882         mutex_unlock(&fs_info->balance_mutex);
3883
3884         ret = __btrfs_balance(fs_info);
3885
3886         mutex_lock(&fs_info->balance_mutex);
3887         atomic_dec(&fs_info->balance_running);
3888
3889         if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3890                 fs_info->num_tolerated_disk_barrier_failures =
3891                         btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3892         }
3893
3894         if (bargs) {
3895                 memset(bargs, 0, sizeof(*bargs));
3896                 update_ioctl_balance_args(fs_info, 0, bargs);
3897         }
3898
3899         if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
3900             balance_need_close(fs_info)) {
3901                 __cancel_balance(fs_info);
3902         }
3903
3904         wake_up(&fs_info->balance_wait_q);
3905
3906         return ret;
3907 out:
3908         if (bctl->flags & BTRFS_BALANCE_RESUME)
3909                 __cancel_balance(fs_info);
3910         else {
3911                 kfree(bctl);
3912                 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3913         }
3914         return ret;
3915 }
3916
3917 static int balance_kthread(void *data)
3918 {
3919         struct btrfs_fs_info *fs_info = data;
3920         int ret = 0;
3921
3922         mutex_lock(&fs_info->volume_mutex);
3923         mutex_lock(&fs_info->balance_mutex);
3924
3925         if (fs_info->balance_ctl) {
3926                 btrfs_info(fs_info, "continuing balance");
3927                 ret = btrfs_balance(fs_info->balance_ctl, NULL);
3928         }
3929
3930         mutex_unlock(&fs_info->balance_mutex);
3931         mutex_unlock(&fs_info->volume_mutex);
3932
3933         return ret;
3934 }
3935
3936 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
3937 {
3938         struct task_struct *tsk;
3939
3940         spin_lock(&fs_info->balance_lock);
3941         if (!fs_info->balance_ctl) {
3942                 spin_unlock(&fs_info->balance_lock);
3943                 return 0;
3944         }
3945         spin_unlock(&fs_info->balance_lock);
3946
3947         if (btrfs_test_opt(fs_info, SKIP_BALANCE)) {
3948                 btrfs_info(fs_info, "force skipping balance");
3949                 return 0;
3950         }
3951
3952         tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
3953         return PTR_ERR_OR_ZERO(tsk);
3954 }
3955
3956 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
3957 {
3958         struct btrfs_balance_control *bctl;
3959         struct btrfs_balance_item *item;
3960         struct btrfs_disk_balance_args disk_bargs;
3961         struct btrfs_path *path;
3962         struct extent_buffer *leaf;
3963         struct btrfs_key key;
3964         int ret;
3965
3966         path = btrfs_alloc_path();
3967         if (!path)
3968                 return -ENOMEM;
3969
3970         key.objectid = BTRFS_BALANCE_OBJECTID;
3971         key.type = BTRFS_TEMPORARY_ITEM_KEY;
3972         key.offset = 0;
3973
3974         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
3975         if (ret < 0)
3976                 goto out;
3977         if (ret > 0) { /* ret = -ENOENT; */
3978                 ret = 0;
3979                 goto out;
3980         }
3981
3982         bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
3983         if (!bctl) {
3984                 ret = -ENOMEM;
3985                 goto out;
3986         }
3987
3988         leaf = path->nodes[0];
3989         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3990
3991         bctl->fs_info = fs_info;
3992         bctl->flags = btrfs_balance_flags(leaf, item);
3993         bctl->flags |= BTRFS_BALANCE_RESUME;
3994
3995         btrfs_balance_data(leaf, item, &disk_bargs);
3996         btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
3997         btrfs_balance_meta(leaf, item, &disk_bargs);
3998         btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
3999         btrfs_balance_sys(leaf, item, &disk_bargs);
4000         btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
4001
4002         WARN_ON(atomic_xchg(&fs_info->mutually_exclusive_operation_running, 1));
4003
4004         mutex_lock(&fs_info->volume_mutex);
4005         mutex_lock(&fs_info->balance_mutex);
4006
4007         set_balance_control(bctl);
4008
4009         mutex_unlock(&fs_info->balance_mutex);
4010         mutex_unlock(&fs_info->volume_mutex);
4011 out:
4012         btrfs_free_path(path);
4013         return ret;
4014 }
4015
4016 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
4017 {
4018         int ret = 0;
4019
4020         mutex_lock(&fs_info->balance_mutex);
4021         if (!fs_info->balance_ctl) {
4022                 mutex_unlock(&fs_info->balance_mutex);
4023                 return -ENOTCONN;
4024         }
4025
4026         if (atomic_read(&fs_info->balance_running)) {
4027                 atomic_inc(&fs_info->balance_pause_req);
4028                 mutex_unlock(&fs_info->balance_mutex);
4029
4030                 wait_event(fs_info->balance_wait_q,
4031                            atomic_read(&fs_info->balance_running) == 0);
4032
4033                 mutex_lock(&fs_info->balance_mutex);
4034                 /* we are good with balance_ctl ripped off from under us */
4035                 BUG_ON(atomic_read(&fs_info->balance_running));
4036                 atomic_dec(&fs_info->balance_pause_req);
4037         } else {
4038                 ret = -ENOTCONN;
4039         }
4040
4041         mutex_unlock(&fs_info->balance_mutex);
4042         return ret;
4043 }
4044
4045 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
4046 {
4047         if (fs_info->sb->s_flags & MS_RDONLY)
4048                 return -EROFS;
4049
4050         mutex_lock(&fs_info->balance_mutex);
4051         if (!fs_info->balance_ctl) {
4052                 mutex_unlock(&fs_info->balance_mutex);
4053                 return -ENOTCONN;
4054         }
4055
4056         atomic_inc(&fs_info->balance_cancel_req);
4057         /*
4058          * if we are running just wait and return, balance item is
4059          * deleted in btrfs_balance in this case
4060          */
4061         if (atomic_read(&fs_info->balance_running)) {
4062                 mutex_unlock(&fs_info->balance_mutex);
4063                 wait_event(fs_info->balance_wait_q,
4064                            atomic_read(&fs_info->balance_running) == 0);
4065                 mutex_lock(&fs_info->balance_mutex);
4066         } else {
4067                 /* __cancel_balance needs volume_mutex */
4068                 mutex_unlock(&fs_info->balance_mutex);
4069                 mutex_lock(&fs_info->volume_mutex);
4070                 mutex_lock(&fs_info->balance_mutex);
4071
4072                 if (fs_info->balance_ctl)
4073                         __cancel_balance(fs_info);
4074
4075                 mutex_unlock(&fs_info->volume_mutex);
4076         }
4077
4078         BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running));
4079         atomic_dec(&fs_info->balance_cancel_req);
4080         mutex_unlock(&fs_info->balance_mutex);
4081         return 0;
4082 }
4083
4084 static int btrfs_uuid_scan_kthread(void *data)
4085 {
4086         struct btrfs_fs_info *fs_info = data;
4087         struct btrfs_root *root = fs_info->tree_root;
4088         struct btrfs_key key;
4089         struct btrfs_key max_key;
4090         struct btrfs_path *path = NULL;
4091         int ret = 0;
4092         struct extent_buffer *eb;
4093         int slot;
4094         struct btrfs_root_item root_item;
4095         u32 item_size;
4096         struct btrfs_trans_handle *trans = NULL;
4097
4098         path = btrfs_alloc_path();
4099         if (!path) {
4100                 ret = -ENOMEM;
4101                 goto out;
4102         }
4103
4104         key.objectid = 0;
4105         key.type = BTRFS_ROOT_ITEM_KEY;
4106         key.offset = 0;
4107
4108         max_key.objectid = (u64)-1;
4109         max_key.type = BTRFS_ROOT_ITEM_KEY;
4110         max_key.offset = (u64)-1;
4111
4112         while (1) {
4113                 ret = btrfs_search_forward(root, &key, path, 0);
4114                 if (ret) {
4115                         if (ret > 0)
4116                                 ret = 0;
4117                         break;
4118                 }
4119
4120                 if (key.type != BTRFS_ROOT_ITEM_KEY ||
4121                     (key.objectid < BTRFS_FIRST_FREE_OBJECTID &&
4122                      key.objectid != BTRFS_FS_TREE_OBJECTID) ||
4123                     key.objectid > BTRFS_LAST_FREE_OBJECTID)
4124                         goto skip;
4125
4126                 eb = path->nodes[0];
4127                 slot = path->slots[0];
4128                 item_size = btrfs_item_size_nr(eb, slot);
4129                 if (item_size < sizeof(root_item))
4130                         goto skip;
4131
4132                 read_extent_buffer(eb, &root_item,
4133                                    btrfs_item_ptr_offset(eb, slot),
4134                                    (int)sizeof(root_item));
4135                 if (btrfs_root_refs(&root_item) == 0)
4136                         goto skip;
4137
4138                 if (!btrfs_is_empty_uuid(root_item.uuid) ||
4139                     !btrfs_is_empty_uuid(root_item.received_uuid)) {
4140                         if (trans)
4141                                 goto update_tree;
4142
4143                         btrfs_release_path(path);
4144                         /*
4145                          * 1 - subvol uuid item
4146                          * 1 - received_subvol uuid item
4147                          */
4148                         trans = btrfs_start_transaction(fs_info->uuid_root, 2);
4149                         if (IS_ERR(trans)) {
4150                                 ret = PTR_ERR(trans);
4151                                 break;
4152                         }
4153                         continue;
4154                 } else {
4155                         goto skip;
4156                 }
4157 update_tree:
4158                 if (!btrfs_is_empty_uuid(root_item.uuid)) {
4159                         ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
4160                                                   root_item.uuid,
4161                                                   BTRFS_UUID_KEY_SUBVOL,
4162                                                   key.objectid);
4163                         if (ret < 0) {
4164                                 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4165                                         ret);
4166                                 break;
4167                         }
4168                 }
4169
4170                 if (!btrfs_is_empty_uuid(root_item.received_uuid)) {
4171                         ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
4172                                                   root_item.received_uuid,
4173                                                  BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4174                                                   key.objectid);
4175                         if (ret < 0) {
4176                                 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4177                                         ret);
4178                                 break;
4179                         }
4180                 }
4181
4182 skip:
4183                 if (trans) {
4184                         ret = btrfs_end_transaction(trans, fs_info->uuid_root);
4185                         trans = NULL;
4186                         if (ret)
4187                                 break;
4188                 }
4189
4190                 btrfs_release_path(path);
4191                 if (key.offset < (u64)-1) {
4192                         key.offset++;
4193                 } else if (key.type < BTRFS_ROOT_ITEM_KEY) {
4194                         key.offset = 0;
4195                         key.type = BTRFS_ROOT_ITEM_KEY;
4196                 } else if (key.objectid < (u64)-1) {
4197                         key.offset = 0;
4198                         key.type = BTRFS_ROOT_ITEM_KEY;
4199                         key.objectid++;
4200                 } else {
4201                         break;
4202                 }
4203                 cond_resched();
4204         }
4205
4206 out:
4207         btrfs_free_path(path);
4208         if (trans && !IS_ERR(trans))
4209                 btrfs_end_transaction(trans, fs_info->uuid_root);
4210         if (ret)
4211                 btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret);
4212         else
4213                 fs_info->update_uuid_tree_gen = 1;
4214         up(&fs_info->uuid_tree_rescan_sem);
4215         return 0;
4216 }
4217
4218 /*
4219  * Callback for btrfs_uuid_tree_iterate().
4220  * returns:
4221  * 0    check succeeded, the entry is not outdated.
4222  * < 0  if an error occurred.
4223  * > 0  if the check failed, which means the caller shall remove the entry.
4224  */
4225 static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info *fs_info,
4226                                        u8 *uuid, u8 type, u64 subid)
4227 {
4228         struct btrfs_key key;
4229         int ret = 0;
4230         struct btrfs_root *subvol_root;
4231
4232         if (type != BTRFS_UUID_KEY_SUBVOL &&
4233             type != BTRFS_UUID_KEY_RECEIVED_SUBVOL)
4234                 goto out;
4235
4236         key.objectid = subid;
4237         key.type = BTRFS_ROOT_ITEM_KEY;
4238         key.offset = (u64)-1;
4239         subvol_root = btrfs_read_fs_root_no_name(fs_info, &key);
4240         if (IS_ERR(subvol_root)) {
4241                 ret = PTR_ERR(subvol_root);
4242                 if (ret == -ENOENT)
4243                         ret = 1;
4244                 goto out;
4245         }
4246
4247         switch (type) {
4248         case BTRFS_UUID_KEY_SUBVOL:
4249                 if (memcmp(uuid, subvol_root->root_item.uuid, BTRFS_UUID_SIZE))
4250                         ret = 1;
4251                 break;
4252         case BTRFS_UUID_KEY_RECEIVED_SUBVOL:
4253                 if (memcmp(uuid, subvol_root->root_item.received_uuid,
4254                            BTRFS_UUID_SIZE))
4255                         ret = 1;
4256                 break;
4257         }
4258
4259 out:
4260         return ret;
4261 }
4262
4263 static int btrfs_uuid_rescan_kthread(void *data)
4264 {
4265         struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data;
4266         int ret;
4267
4268         /*
4269          * 1st step is to iterate through the existing UUID tree and
4270          * to delete all entries that contain outdated data.
4271          * 2nd step is to add all missing entries to the UUID tree.
4272          */
4273         ret = btrfs_uuid_tree_iterate(fs_info, btrfs_check_uuid_tree_entry);
4274         if (ret < 0) {
4275                 btrfs_warn(fs_info, "iterating uuid_tree failed %d", ret);
4276                 up(&fs_info->uuid_tree_rescan_sem);
4277                 return ret;
4278         }
4279         return btrfs_uuid_scan_kthread(data);
4280 }
4281
4282 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info)
4283 {
4284         struct btrfs_trans_handle *trans;
4285         struct btrfs_root *tree_root = fs_info->tree_root;
4286         struct btrfs_root *uuid_root;
4287         struct task_struct *task;
4288         int ret;
4289
4290         /*
4291          * 1 - root node
4292          * 1 - root item
4293          */
4294         trans = btrfs_start_transaction(tree_root, 2);
4295         if (IS_ERR(trans))
4296                 return PTR_ERR(trans);
4297
4298         uuid_root = btrfs_create_tree(trans, fs_info,
4299                                       BTRFS_UUID_TREE_OBJECTID);
4300         if (IS_ERR(uuid_root)) {
4301                 ret = PTR_ERR(uuid_root);
4302                 btrfs_abort_transaction(trans, ret);
4303                 btrfs_end_transaction(trans, tree_root);
4304                 return ret;
4305         }
4306
4307         fs_info->uuid_root = uuid_root;
4308
4309         ret = btrfs_commit_transaction(trans, tree_root);
4310         if (ret)
4311                 return ret;
4312
4313         down(&fs_info->uuid_tree_rescan_sem);
4314         task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid");
4315         if (IS_ERR(task)) {
4316                 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4317                 btrfs_warn(fs_info, "failed to start uuid_scan task");
4318                 up(&fs_info->uuid_tree_rescan_sem);
4319                 return PTR_ERR(task);
4320         }
4321
4322         return 0;
4323 }
4324
4325 int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
4326 {
4327         struct task_struct *task;
4328
4329         down(&fs_info->uuid_tree_rescan_sem);
4330         task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
4331         if (IS_ERR(task)) {
4332                 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4333                 btrfs_warn(fs_info, "failed to start uuid_rescan task");
4334                 up(&fs_info->uuid_tree_rescan_sem);
4335                 return PTR_ERR(task);
4336         }
4337
4338         return 0;
4339 }
4340
4341 /*
4342  * shrinking a device means finding all of the device extents past
4343  * the new size, and then following the back refs to the chunks.
4344  * The chunk relocation code actually frees the device extent
4345  */
4346 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
4347 {
4348         struct btrfs_trans_handle *trans;
4349         struct btrfs_root *root = device->dev_root;
4350         struct btrfs_dev_extent *dev_extent = NULL;
4351         struct btrfs_path *path;
4352         u64 length;
4353         u64 chunk_offset;
4354         int ret;
4355         int slot;
4356         int failed = 0;
4357         bool retried = false;
4358         bool checked_pending_chunks = false;
4359         struct extent_buffer *l;
4360         struct btrfs_key key;
4361         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
4362         u64 old_total = btrfs_super_total_bytes(super_copy);
4363         u64 old_size = btrfs_device_get_total_bytes(device);
4364         u64 diff = old_size - new_size;
4365
4366         if (device->is_tgtdev_for_dev_replace)
4367                 return -EINVAL;
4368
4369         path = btrfs_alloc_path();
4370         if (!path)
4371                 return -ENOMEM;
4372
4373         path->reada = READA_FORWARD;
4374
4375         lock_chunks(root);
4376
4377         btrfs_device_set_total_bytes(device, new_size);
4378         if (device->writeable) {
4379                 device->fs_devices->total_rw_bytes -= diff;
4380                 spin_lock(&root->fs_info->free_chunk_lock);
4381                 root->fs_info->free_chunk_space -= diff;
4382                 spin_unlock(&root->fs_info->free_chunk_lock);
4383         }
4384         unlock_chunks(root);
4385
4386 again:
4387         key.objectid = device->devid;
4388         key.offset = (u64)-1;
4389         key.type = BTRFS_DEV_EXTENT_KEY;
4390
4391         do {
4392                 mutex_lock(&root->fs_info->delete_unused_bgs_mutex);
4393                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4394                 if (ret < 0) {
4395                         mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4396                         goto done;
4397                 }
4398
4399                 ret = btrfs_previous_item(root, path, 0, key.type);
4400                 if (ret)
4401                         mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4402                 if (ret < 0)
4403                         goto done;
4404                 if (ret) {
4405                         ret = 0;
4406                         btrfs_release_path(path);
4407                         break;
4408                 }
4409
4410                 l = path->nodes[0];
4411                 slot = path->slots[0];
4412                 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
4413
4414                 if (key.objectid != device->devid) {
4415                         mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4416                         btrfs_release_path(path);
4417                         break;
4418                 }
4419
4420                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
4421                 length = btrfs_dev_extent_length(l, dev_extent);
4422
4423                 if (key.offset + length <= new_size) {
4424                         mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4425                         btrfs_release_path(path);
4426                         break;
4427                 }
4428
4429                 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
4430                 btrfs_release_path(path);
4431
4432                 ret = btrfs_relocate_chunk(root, chunk_offset);
4433                 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4434                 if (ret && ret != -ENOSPC)
4435                         goto done;
4436                 if (ret == -ENOSPC)
4437                         failed++;
4438         } while (key.offset-- > 0);
4439
4440         if (failed && !retried) {
4441                 failed = 0;
4442                 retried = true;
4443                 goto again;
4444         } else if (failed && retried) {
4445                 ret = -ENOSPC;
4446                 goto done;
4447         }
4448
4449         /* Shrinking succeeded, else we would be at "done". */
4450         trans = btrfs_start_transaction(root, 0);
4451         if (IS_ERR(trans)) {
4452                 ret = PTR_ERR(trans);
4453                 goto done;
4454         }
4455
4456         lock_chunks(root);
4457
4458         /*
4459          * We checked in the above loop all device extents that were already in
4460          * the device tree. However before we have updated the device's
4461          * total_bytes to the new size, we might have had chunk allocations that
4462          * have not complete yet (new block groups attached to transaction
4463          * handles), and therefore their device extents were not yet in the
4464          * device tree and we missed them in the loop above. So if we have any
4465          * pending chunk using a device extent that overlaps the device range
4466          * that we can not use anymore, commit the current transaction and
4467          * repeat the search on the device tree - this way we guarantee we will
4468          * not have chunks using device extents that end beyond 'new_size'.
4469          */
4470         if (!checked_pending_chunks) {
4471                 u64 start = new_size;
4472                 u64 len = old_size - new_size;
4473
4474                 if (contains_pending_extent(trans->transaction, device,
4475                                             &start, len)) {
4476                         unlock_chunks(root);
4477                         checked_pending_chunks = true;
4478                         failed = 0;
4479                         retried = false;
4480                         ret = btrfs_commit_transaction(trans, root);
4481                         if (ret)
4482                                 goto done;
4483                         goto again;
4484                 }
4485         }
4486
4487         btrfs_device_set_disk_total_bytes(device, new_size);
4488         if (list_empty(&device->resized_list))
4489                 list_add_tail(&device->resized_list,
4490                               &root->fs_info->fs_devices->resized_devices);
4491
4492         WARN_ON(diff > old_total);
4493         btrfs_set_super_total_bytes(super_copy, old_total - diff);
4494         unlock_chunks(root);
4495
4496         /* Now btrfs_update_device() will change the on-disk size. */
4497         ret = btrfs_update_device(trans, device);
4498         btrfs_end_transaction(trans, root);
4499 done:
4500         btrfs_free_path(path);
4501         if (ret) {
4502                 lock_chunks(root);
4503                 btrfs_device_set_total_bytes(device, old_size);
4504                 if (device->writeable)
4505                         device->fs_devices->total_rw_bytes += diff;
4506                 spin_lock(&root->fs_info->free_chunk_lock);
4507                 root->fs_info->free_chunk_space += diff;
4508                 spin_unlock(&root->fs_info->free_chunk_lock);
4509                 unlock_chunks(root);
4510         }
4511         return ret;
4512 }
4513
4514 static int btrfs_add_system_chunk(struct btrfs_root *root,
4515                            struct btrfs_key *key,
4516                            struct btrfs_chunk *chunk, int item_size)
4517 {
4518         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
4519         struct btrfs_disk_key disk_key;
4520         u32 array_size;
4521         u8 *ptr;
4522
4523         lock_chunks(root);
4524         array_size = btrfs_super_sys_array_size(super_copy);
4525         if (array_size + item_size + sizeof(disk_key)
4526                         > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
4527                 unlock_chunks(root);
4528                 return -EFBIG;
4529         }
4530
4531         ptr = super_copy->sys_chunk_array + array_size;
4532         btrfs_cpu_key_to_disk(&disk_key, key);
4533         memcpy(ptr, &disk_key, sizeof(disk_key));
4534         ptr += sizeof(disk_key);
4535         memcpy(ptr, chunk, item_size);
4536         item_size += sizeof(disk_key);
4537         btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
4538         unlock_chunks(root);
4539
4540         return 0;
4541 }
4542
4543 /*
4544  * sort the devices in descending order by max_avail, total_avail
4545  */
4546 static int btrfs_cmp_device_info(const void *a, const void *b)
4547 {
4548         const struct btrfs_device_info *di_a = a;
4549         const struct btrfs_device_info *di_b = b;
4550
4551         if (di_a->max_avail > di_b->max_avail)
4552                 return -1;
4553         if (di_a->max_avail < di_b->max_avail)
4554                 return 1;
4555         if (di_a->total_avail > di_b->total_avail)
4556                 return -1;
4557         if (di_a->total_avail < di_b->total_avail)
4558                 return 1;
4559         return 0;
4560 }
4561
4562 static u32 find_raid56_stripe_len(u32 data_devices, u32 dev_stripe_target)
4563 {
4564         /* TODO allow them to set a preferred stripe size */
4565         return SZ_64K;
4566 }
4567
4568 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
4569 {
4570         if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK))
4571                 return;
4572
4573         btrfs_set_fs_incompat(info, RAID56);
4574 }
4575
4576 #define BTRFS_MAX_DEVS(r) ((BTRFS_MAX_ITEM_SIZE(r)              \
4577                         - sizeof(struct btrfs_chunk))           \
4578                         / sizeof(struct btrfs_stripe) + 1)
4579
4580 #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE        \
4581                                 - 2 * sizeof(struct btrfs_disk_key)     \
4582                                 - 2 * sizeof(struct btrfs_chunk))       \
4583                                 / sizeof(struct btrfs_stripe) + 1)
4584
4585 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4586                                struct btrfs_root *extent_root, u64 start,
4587                                u64 type)
4588 {
4589         struct btrfs_fs_info *info = extent_root->fs_info;
4590         struct btrfs_fs_devices *fs_devices = info->fs_devices;
4591         struct list_head *cur;
4592         struct map_lookup *map = NULL;
4593         struct extent_map_tree *em_tree;
4594         struct extent_map *em;
4595         struct btrfs_device_info *devices_info = NULL;
4596         u64 total_avail;
4597         int num_stripes;        /* total number of stripes to allocate */
4598         int data_stripes;       /* number of stripes that count for
4599                                    block group size */
4600         int sub_stripes;        /* sub_stripes info for map */
4601         int dev_stripes;        /* stripes per dev */
4602         int devs_max;           /* max devs to use */
4603         int devs_min;           /* min devs needed */
4604         int devs_increment;     /* ndevs has to be a multiple of this */
4605         int ncopies;            /* how many copies to data has */
4606         int ret;
4607         u64 max_stripe_size;
4608         u64 max_chunk_size;
4609         u64 stripe_size;
4610         u64 num_bytes;
4611         u64 raid_stripe_len = BTRFS_STRIPE_LEN;
4612         int ndevs;
4613         int i;
4614         int j;
4615         int index;
4616
4617         BUG_ON(!alloc_profile_is_valid(type, 0));
4618
4619         if (list_empty(&fs_devices->alloc_list))
4620                 return -ENOSPC;
4621
4622         index = __get_raid_index(type);
4623
4624         sub_stripes = btrfs_raid_array[index].sub_stripes;
4625         dev_stripes = btrfs_raid_array[index].dev_stripes;
4626         devs_max = btrfs_raid_array[index].devs_max;
4627         devs_min = btrfs_raid_array[index].devs_min;
4628         devs_increment = btrfs_raid_array[index].devs_increment;
4629         ncopies = btrfs_raid_array[index].ncopies;
4630
4631         if (type & BTRFS_BLOCK_GROUP_DATA) {
4632                 max_stripe_size = SZ_1G;
4633                 max_chunk_size = 10 * max_stripe_size;
4634                 if (!devs_max)
4635                         devs_max = BTRFS_MAX_DEVS(info->chunk_root);
4636         } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
4637                 /* for larger filesystems, use larger metadata chunks */
4638                 if (fs_devices->total_rw_bytes > 50ULL * SZ_1G)
4639                         max_stripe_size = SZ_1G;
4640                 else
4641                         max_stripe_size = SZ_256M;
4642                 max_chunk_size = max_stripe_size;
4643                 if (!devs_max)
4644                         devs_max = BTRFS_MAX_DEVS(info->chunk_root);
4645         } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
4646                 max_stripe_size = SZ_32M;
4647                 max_chunk_size = 2 * max_stripe_size;
4648                 if (!devs_max)
4649                         devs_max = BTRFS_MAX_DEVS_SYS_CHUNK;
4650         } else {
4651                 btrfs_err(info, "invalid chunk type 0x%llx requested",
4652                        type);
4653                 BUG_ON(1);
4654         }
4655
4656         /* we don't want a chunk larger than 10% of writeable space */
4657         max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
4658                              max_chunk_size);
4659
4660         devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info),
4661                                GFP_NOFS);
4662         if (!devices_info)
4663                 return -ENOMEM;
4664
4665         cur = fs_devices->alloc_list.next;
4666
4667         /*
4668          * in the first pass through the devices list, we gather information
4669          * about the available holes on each device.
4670          */
4671         ndevs = 0;
4672         while (cur != &fs_devices->alloc_list) {
4673                 struct btrfs_device *device;
4674                 u64 max_avail;
4675                 u64 dev_offset;
4676
4677                 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
4678
4679                 cur = cur->next;
4680
4681                 if (!device->writeable) {
4682                         WARN(1, KERN_ERR
4683                                "BTRFS: read-only device in alloc_list\n");
4684                         continue;
4685                 }
4686
4687                 if (!device->in_fs_metadata ||
4688                     device->is_tgtdev_for_dev_replace)
4689                         continue;
4690
4691                 if (device->total_bytes > device->bytes_used)
4692                         total_avail = device->total_bytes - device->bytes_used;
4693                 else
4694                         total_avail = 0;
4695
4696                 /* If there is no space on this device, skip it. */
4697                 if (total_avail == 0)
4698                         continue;
4699
4700                 ret = find_free_dev_extent(trans, device,
4701                                            max_stripe_size * dev_stripes,
4702                                            &dev_offset, &max_avail);
4703                 if (ret && ret != -ENOSPC)
4704                         goto error;
4705
4706                 if (ret == 0)
4707                         max_avail = max_stripe_size * dev_stripes;
4708
4709                 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes)
4710                         continue;
4711
4712                 if (ndevs == fs_devices->rw_devices) {
4713                         WARN(1, "%s: found more than %llu devices\n",
4714                              __func__, fs_devices->rw_devices);
4715                         break;
4716                 }
4717                 devices_info[ndevs].dev_offset = dev_offset;
4718                 devices_info[ndevs].max_avail = max_avail;
4719                 devices_info[ndevs].total_avail = total_avail;
4720                 devices_info[ndevs].dev = device;
4721                 ++ndevs;
4722         }
4723
4724         /*
4725          * now sort the devices by hole size / available space
4726          */
4727         sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
4728              btrfs_cmp_device_info, NULL);
4729
4730         /* round down to number of usable stripes */
4731         ndevs -= ndevs % devs_increment;
4732
4733         if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) {
4734                 ret = -ENOSPC;
4735                 goto error;
4736         }
4737
4738         if (devs_max && ndevs > devs_max)
4739                 ndevs = devs_max;
4740         /*
4741          * the primary goal is to maximize the number of stripes, so use as many
4742          * devices as possible, even if the stripes are not maximum sized.
4743          */
4744         stripe_size = devices_info[ndevs-1].max_avail;
4745         num_stripes = ndevs * dev_stripes;
4746
4747         /*
4748          * this will have to be fixed for RAID1 and RAID10 over
4749          * more drives
4750          */
4751         data_stripes = num_stripes / ncopies;
4752
4753         if (type & BTRFS_BLOCK_GROUP_RAID5) {
4754                 raid_stripe_len = find_raid56_stripe_len(ndevs - 1,
4755                                                 extent_root->stripesize);
4756                 data_stripes = num_stripes - 1;
4757         }
4758         if (type & BTRFS_BLOCK_GROUP_RAID6) {
4759                 raid_stripe_len = find_raid56_stripe_len(ndevs - 2,
4760                                                 extent_root->stripesize);
4761                 data_stripes = num_stripes - 2;
4762         }
4763
4764         /*
4765          * Use the number of data stripes to figure out how big this chunk
4766          * is really going to be in terms of logical address space,
4767          * and compare that answer with the max chunk size
4768          */
4769         if (stripe_size * data_stripes > max_chunk_size) {
4770                 u64 mask = (1ULL << 24) - 1;
4771
4772                 stripe_size = div_u64(max_chunk_size, data_stripes);
4773
4774                 /* bump the answer up to a 16MB boundary */
4775                 stripe_size = (stripe_size + mask) & ~mask;
4776
4777                 /* but don't go higher than the limits we found
4778                  * while searching for free extents
4779                  */
4780                 if (stripe_size > devices_info[ndevs-1].max_avail)
4781                         stripe_size = devices_info[ndevs-1].max_avail;
4782         }
4783
4784         stripe_size = div_u64(stripe_size, dev_stripes);
4785
4786         /* align to BTRFS_STRIPE_LEN */
4787         stripe_size = div_u64(stripe_size, raid_stripe_len);
4788         stripe_size *= raid_stripe_len;
4789
4790         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
4791         if (!map) {
4792                 ret = -ENOMEM;
4793                 goto error;
4794         }
4795         map->num_stripes = num_stripes;
4796
4797         for (i = 0; i < ndevs; ++i) {
4798                 for (j = 0; j < dev_stripes; ++j) {
4799                         int s = i * dev_stripes + j;
4800                         map->stripes[s].dev = devices_info[i].dev;
4801                         map->stripes[s].physical = devices_info[i].dev_offset +
4802                                                    j * stripe_size;
4803                 }
4804         }
4805         map->sector_size = extent_root->sectorsize;
4806         map->stripe_len = raid_stripe_len;
4807         map->io_align = raid_stripe_len;
4808         map->io_width = raid_stripe_len;
4809         map->type = type;
4810         map->sub_stripes = sub_stripes;
4811
4812         num_bytes = stripe_size * data_stripes;
4813
4814         trace_btrfs_chunk_alloc(info->chunk_root, map, start, num_bytes);
4815
4816         em = alloc_extent_map();
4817         if (!em) {
4818                 kfree(map);
4819                 ret = -ENOMEM;
4820                 goto error;
4821         }
4822         set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
4823         em->map_lookup = map;
4824         em->start = start;
4825         em->len = num_bytes;
4826         em->block_start = 0;
4827         em->block_len = em->len;
4828         em->orig_block_len = stripe_size;
4829
4830         em_tree = &extent_root->fs_info->mapping_tree.map_tree;
4831         write_lock(&em_tree->lock);
4832         ret = add_extent_mapping(em_tree, em, 0);
4833         if (!ret) {
4834                 list_add_tail(&em->list, &trans->transaction->pending_chunks);
4835                 atomic_inc(&em->refs);
4836         }
4837         write_unlock(&em_tree->lock);
4838         if (ret) {
4839                 free_extent_map(em);
4840                 goto error;
4841         }
4842
4843         ret = btrfs_make_block_group(trans, extent_root, 0, type,
4844                                      BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4845                                      start, num_bytes);
4846         if (ret)
4847                 goto error_del_extent;
4848
4849         for (i = 0; i < map->num_stripes; i++) {
4850                 num_bytes = map->stripes[i].dev->bytes_used + stripe_size;
4851                 btrfs_device_set_bytes_used(map->stripes[i].dev, num_bytes);
4852         }
4853
4854         spin_lock(&extent_root->fs_info->free_chunk_lock);
4855         extent_root->fs_info->free_chunk_space -= (stripe_size *
4856                                                    map->num_stripes);
4857         spin_unlock(&extent_root->fs_info->free_chunk_lock);
4858
4859         free_extent_map(em);
4860         check_raid56_incompat_flag(extent_root->fs_info, type);
4861
4862         kfree(devices_info);
4863         return 0;
4864
4865 error_del_extent:
4866         write_lock(&em_tree->lock);
4867         remove_extent_mapping(em_tree, em);
4868         write_unlock(&em_tree->lock);
4869
4870         /* One for our allocation */
4871         free_extent_map(em);
4872         /* One for the tree reference */
4873         free_extent_map(em);
4874         /* One for the pending_chunks list reference */
4875         free_extent_map(em);
4876 error:
4877         kfree(devices_info);
4878         return ret;
4879 }
4880
4881 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans,
4882                                 struct btrfs_root *extent_root,
4883                                 u64 chunk_offset, u64 chunk_size)
4884 {
4885         struct btrfs_key key;
4886         struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
4887         struct btrfs_device *device;
4888         struct btrfs_chunk *chunk;
4889         struct btrfs_stripe *stripe;
4890         struct extent_map_tree *em_tree;
4891         struct extent_map *em;
4892         struct map_lookup *map;
4893         size_t item_size;
4894         u64 dev_offset;
4895         u64 stripe_size;
4896         int i = 0;
4897         int ret = 0;
4898
4899         em_tree = &extent_root->fs_info->mapping_tree.map_tree;
4900         read_lock(&em_tree->lock);
4901         em = lookup_extent_mapping(em_tree, chunk_offset, chunk_size);
4902         read_unlock(&em_tree->lock);
4903
4904         if (!em) {
4905                 btrfs_crit(extent_root->fs_info, "unable to find logical "
4906                            "%Lu len %Lu", chunk_offset, chunk_size);
4907                 return -EINVAL;
4908         }
4909
4910         if (em->start != chunk_offset || em->len != chunk_size) {
4911                 btrfs_crit(extent_root->fs_info, "found a bad mapping, wanted"
4912                           " %Lu-%Lu, found %Lu-%Lu", chunk_offset,
4913                           chunk_size, em->start, em->len);
4914                 free_extent_map(em);
4915                 return -EINVAL;
4916         }
4917
4918         map = em->map_lookup;
4919         item_size = btrfs_chunk_item_size(map->num_stripes);
4920         stripe_size = em->orig_block_len;
4921
4922         chunk = kzalloc(item_size, GFP_NOFS);
4923         if (!chunk) {
4924                 ret = -ENOMEM;
4925                 goto out;
4926         }
4927
4928         /*
4929          * Take the device list mutex to prevent races with the final phase of
4930          * a device replace operation that replaces the device object associated
4931          * with the map's stripes, because the device object's id can change
4932          * at any time during that final phase of the device replace operation
4933          * (dev-replace.c:btrfs_dev_replace_finishing()).
4934          */
4935         mutex_lock(&chunk_root->fs_info->fs_devices->device_list_mutex);
4936         for (i = 0; i < map->num_stripes; i++) {
4937                 device = map->stripes[i].dev;
4938                 dev_offset = map->stripes[i].physical;
4939
4940                 ret = btrfs_update_device(trans, device);
4941                 if (ret)
4942                         break;
4943                 ret = btrfs_alloc_dev_extent(trans, device,
4944                                              chunk_root->root_key.objectid,
4945                                              BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4946                                              chunk_offset, dev_offset,
4947                                              stripe_size);
4948                 if (ret)
4949                         break;
4950         }
4951         if (ret) {
4952                 mutex_unlock(&chunk_root->fs_info->fs_devices->device_list_mutex);
4953                 goto out;
4954         }
4955
4956         stripe = &chunk->stripe;
4957         for (i = 0; i < map->num_stripes; i++) {
4958                 device = map->stripes[i].dev;
4959                 dev_offset = map->stripes[i].physical;
4960
4961                 btrfs_set_stack_stripe_devid(stripe, device->devid);
4962                 btrfs_set_stack_stripe_offset(stripe, dev_offset);
4963                 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
4964                 stripe++;
4965         }
4966         mutex_unlock(&chunk_root->fs_info->fs_devices->device_list_mutex);
4967
4968         btrfs_set_stack_chunk_length(chunk, chunk_size);
4969         btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
4970         btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
4971         btrfs_set_stack_chunk_type(chunk, map->type);
4972         btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
4973         btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
4974         btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
4975         btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
4976         btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
4977
4978         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
4979         key.type = BTRFS_CHUNK_ITEM_KEY;
4980         key.offset = chunk_offset;
4981
4982         ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
4983         if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
4984                 /*
4985                  * TODO: Cleanup of inserted chunk root in case of
4986                  * failure.
4987                  */
4988                 ret = btrfs_add_system_chunk(chunk_root, &key, chunk,
4989                                              item_size);
4990         }
4991
4992 out:
4993         kfree(chunk);
4994         free_extent_map(em);
4995         return ret;
4996 }
4997
4998 /*
4999  * Chunk allocation falls into two parts. The first part does works
5000  * that make the new allocated chunk useable, but not do any operation
5001  * that modifies the chunk tree. The second part does the works that
5002  * require modifying the chunk tree. This division is important for the
5003  * bootstrap process of adding storage to a seed btrfs.
5004  */
5005 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
5006                       struct btrfs_root *extent_root, u64 type)
5007 {
5008         u64 chunk_offset;
5009
5010         ASSERT(mutex_is_locked(&extent_root->fs_info->chunk_mutex));
5011         chunk_offset = find_next_chunk(extent_root->fs_info);
5012         return __btrfs_alloc_chunk(trans, extent_root, chunk_offset, type);
5013 }
5014
5015 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
5016                                          struct btrfs_root *root,
5017                                          struct btrfs_device *device)
5018 {
5019         u64 chunk_offset;
5020         u64 sys_chunk_offset;
5021         u64 alloc_profile;
5022         struct btrfs_fs_info *fs_info = root->fs_info;
5023         struct btrfs_root *extent_root = fs_info->extent_root;
5024         int ret;
5025
5026         chunk_offset = find_next_chunk(fs_info);
5027         alloc_profile = btrfs_get_alloc_profile(extent_root, 0);
5028         ret = __btrfs_alloc_chunk(trans, extent_root, chunk_offset,
5029                                   alloc_profile);
5030         if (ret)
5031                 return ret;
5032
5033         sys_chunk_offset = find_next_chunk(root->fs_info);
5034         alloc_profile = btrfs_get_alloc_profile(fs_info->chunk_root, 0);
5035         ret = __btrfs_alloc_chunk(trans, extent_root, sys_chunk_offset,
5036                                   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_root *root, u64 chunk_offset)
5059 {
5060         struct extent_map *em;
5061         struct map_lookup *map;
5062         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
5063         int readonly = 0;
5064         int miss_ndevs = 0;
5065         int i;
5066
5067         read_lock(&map_tree->map_tree.lock);
5068         em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
5069         read_unlock(&map_tree->map_tree.lock);
5070         if (!em)
5071                 return 1;
5072
5073         map = em->map_lookup;
5074         for (i = 0; i < map->num_stripes; i++) {
5075                 if (map->stripes[i].dev->missing) {
5076                         miss_ndevs++;
5077                         continue;
5078                 }
5079
5080                 if (!map->stripes[i].dev->writeable) {
5081                         readonly = 1;
5082                         goto end;
5083                 }
5084         }
5085
5086         /*
5087          * If the number of missing devices is larger than max errors,
5088          * we can not write the data into that chunk successfully, so
5089          * set it readonly.
5090          */
5091         if (miss_ndevs > btrfs_chunk_max_errors(map))
5092                 readonly = 1;
5093 end:
5094         free_extent_map(em);
5095         return readonly;
5096 }
5097
5098 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
5099 {
5100         extent_map_tree_init(&tree->map_tree);
5101 }
5102
5103 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
5104 {
5105         struct extent_map *em;
5106
5107         while (1) {
5108                 write_lock(&tree->map_tree.lock);
5109                 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
5110                 if (em)
5111                         remove_extent_mapping(&tree->map_tree, em);
5112                 write_unlock(&tree->map_tree.lock);
5113                 if (!em)
5114                         break;
5115                 /* once for us */
5116                 free_extent_map(em);
5117                 /* once for the tree */
5118                 free_extent_map(em);
5119         }
5120 }
5121
5122 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5123 {
5124         struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
5125         struct extent_map *em;
5126         struct map_lookup *map;
5127         struct extent_map_tree *em_tree = &map_tree->map_tree;
5128         int ret;
5129
5130         read_lock(&em_tree->lock);
5131         em = lookup_extent_mapping(em_tree, logical, len);
5132         read_unlock(&em_tree->lock);
5133
5134         /*
5135          * We could return errors for these cases, but that could get ugly and
5136          * we'd probably do the same thing which is just not do anything else
5137          * and exit, so return 1 so the callers don't try to use other copies.
5138          */
5139         if (!em) {
5140                 btrfs_crit(fs_info, "No mapping for %Lu-%Lu", logical,
5141                             logical+len);
5142                 return 1;
5143         }
5144
5145         if (em->start > logical || em->start + em->len < logical) {
5146                 btrfs_crit(fs_info, "Invalid mapping for %Lu-%Lu, got "
5147                             "%Lu-%Lu", logical, logical+len, em->start,
5148                             em->start + em->len);
5149                 free_extent_map(em);
5150                 return 1;
5151         }
5152
5153         map = em->map_lookup;
5154         if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
5155                 ret = map->num_stripes;
5156         else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5157                 ret = map->sub_stripes;
5158         else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
5159                 ret = 2;
5160         else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5161                 ret = 3;
5162         else
5163                 ret = 1;
5164         free_extent_map(em);
5165
5166         btrfs_dev_replace_lock(&fs_info->dev_replace, 0);
5167         if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace))
5168                 ret++;
5169         btrfs_dev_replace_unlock(&fs_info->dev_replace, 0);
5170
5171         return ret;
5172 }
5173
5174 unsigned long btrfs_full_stripe_len(struct btrfs_root *root,
5175                                     struct btrfs_mapping_tree *map_tree,
5176                                     u64 logical)
5177 {
5178         struct extent_map *em;
5179         struct map_lookup *map;
5180         struct extent_map_tree *em_tree = &map_tree->map_tree;
5181         unsigned long len = root->sectorsize;
5182
5183         read_lock(&em_tree->lock);
5184         em = lookup_extent_mapping(em_tree, logical, len);
5185         read_unlock(&em_tree->lock);
5186         BUG_ON(!em);
5187
5188         BUG_ON(em->start > logical || em->start + em->len < logical);
5189         map = em->map_lookup;
5190         if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5191                 len = map->stripe_len * nr_data_stripes(map);
5192         free_extent_map(em);
5193         return len;
5194 }
5195
5196 int btrfs_is_parity_mirror(struct btrfs_mapping_tree *map_tree,
5197                            u64 logical, u64 len, int mirror_num)
5198 {
5199         struct extent_map *em;
5200         struct map_lookup *map;
5201         struct extent_map_tree *em_tree = &map_tree->map_tree;
5202         int ret = 0;
5203
5204         read_lock(&em_tree->lock);
5205         em = lookup_extent_mapping(em_tree, logical, len);
5206         read_unlock(&em_tree->lock);
5207         BUG_ON(!em);
5208
5209         BUG_ON(em->start > logical || em->start + em->len < logical);
5210         map = em->map_lookup;
5211         if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5212                 ret = 1;
5213         free_extent_map(em);
5214         return ret;
5215 }
5216
5217 static int find_live_mirror(struct btrfs_fs_info *fs_info,
5218                             struct map_lookup *map, int first, int num,
5219                             int optimal, int dev_replace_is_ongoing)
5220 {
5221         int i;
5222         int tolerance;
5223         struct btrfs_device *srcdev;
5224
5225         if (dev_replace_is_ongoing &&
5226             fs_info->dev_replace.cont_reading_from_srcdev_mode ==
5227              BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
5228                 srcdev = fs_info->dev_replace.srcdev;
5229         else
5230                 srcdev = NULL;
5231
5232         /*
5233          * try to avoid the drive that is the source drive for a
5234          * dev-replace procedure, only choose it if no other non-missing
5235          * mirror is available
5236          */
5237         for (tolerance = 0; tolerance < 2; tolerance++) {
5238                 if (map->stripes[optimal].dev->bdev &&
5239                     (tolerance || map->stripes[optimal].dev != srcdev))
5240                         return optimal;
5241                 for (i = first; i < first + num; i++) {
5242                         if (map->stripes[i].dev->bdev &&
5243                             (tolerance || map->stripes[i].dev != srcdev))
5244                                 return i;
5245                 }
5246         }
5247
5248         /* we couldn't find one that doesn't fail.  Just return something
5249          * and the io error handling code will clean up eventually
5250          */
5251         return optimal;
5252 }
5253
5254 static inline int parity_smaller(u64 a, u64 b)
5255 {
5256         return a > b;
5257 }
5258
5259 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
5260 static void sort_parity_stripes(struct btrfs_bio *bbio, int num_stripes)
5261 {
5262         struct btrfs_bio_stripe s;
5263         int i;
5264         u64 l;
5265         int again = 1;
5266
5267         while (again) {
5268                 again = 0;
5269                 for (i = 0; i < num_stripes - 1; i++) {
5270                         if (parity_smaller(bbio->raid_map[i],
5271                                            bbio->raid_map[i+1])) {
5272                                 s = bbio->stripes[i];
5273                                 l = bbio->raid_map[i];
5274                                 bbio->stripes[i] = bbio->stripes[i+1];
5275                                 bbio->raid_map[i] = bbio->raid_map[i+1];
5276                                 bbio->stripes[i+1] = s;
5277                                 bbio->raid_map[i+1] = l;
5278
5279                                 again = 1;
5280                         }
5281                 }
5282         }
5283 }
5284
5285 static struct btrfs_bio *alloc_btrfs_bio(int total_stripes, int real_stripes)
5286 {
5287         struct btrfs_bio *bbio = kzalloc(
5288                  /* the size of the btrfs_bio */
5289                 sizeof(struct btrfs_bio) +
5290                 /* plus the variable array for the stripes */
5291                 sizeof(struct btrfs_bio_stripe) * (total_stripes) +
5292                 /* plus the variable array for the tgt dev */
5293                 sizeof(int) * (real_stripes) +
5294                 /*
5295                  * plus the raid_map, which includes both the tgt dev
5296                  * and the stripes
5297                  */
5298                 sizeof(u64) * (total_stripes),
5299                 GFP_NOFS|__GFP_NOFAIL);
5300
5301         atomic_set(&bbio->error, 0);
5302         atomic_set(&bbio->refs, 1);
5303
5304         return bbio;
5305 }
5306
5307 void btrfs_get_bbio(struct btrfs_bio *bbio)
5308 {
5309         WARN_ON(!atomic_read(&bbio->refs));
5310         atomic_inc(&bbio->refs);
5311 }
5312
5313 void btrfs_put_bbio(struct btrfs_bio *bbio)
5314 {
5315         if (!bbio)
5316                 return;
5317         if (atomic_dec_and_test(&bbio->refs))
5318                 kfree(bbio);
5319 }
5320
5321 static int __btrfs_map_block(struct btrfs_fs_info *fs_info, int op,
5322                              u64 logical, u64 *length,
5323                              struct btrfs_bio **bbio_ret,
5324                              int mirror_num, int need_raid_map)
5325 {
5326         struct extent_map *em;
5327         struct map_lookup *map;
5328         struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
5329         struct extent_map_tree *em_tree = &map_tree->map_tree;
5330         u64 offset;
5331         u64 stripe_offset;
5332         u64 stripe_end_offset;
5333         u64 stripe_nr;
5334         u64 stripe_nr_orig;
5335         u64 stripe_nr_end;
5336         u64 stripe_len;
5337         u32 stripe_index;
5338         int i;
5339         int ret = 0;
5340         int num_stripes;
5341         int max_errors = 0;
5342         int tgtdev_indexes = 0;
5343         struct btrfs_bio *bbio = NULL;
5344         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
5345         int dev_replace_is_ongoing = 0;
5346         int num_alloc_stripes;
5347         int patch_the_first_stripe_for_dev_replace = 0;
5348         u64 physical_to_patch_in_first_stripe = 0;
5349         u64 raid56_full_stripe_start = (u64)-1;
5350
5351         read_lock(&em_tree->lock);
5352         em = lookup_extent_mapping(em_tree, logical, *length);
5353         read_unlock(&em_tree->lock);
5354
5355         if (!em) {
5356                 btrfs_crit(fs_info, "unable to find logical %llu len %llu",
5357                         logical, *length);
5358                 return -EINVAL;
5359         }
5360
5361         if (em->start > logical || em->start + em->len < logical) {
5362                 btrfs_crit(fs_info, "found a bad mapping, wanted %Lu, "
5363                            "found %Lu-%Lu", logical, em->start,
5364                            em->start + em->len);
5365                 free_extent_map(em);
5366                 return -EINVAL;
5367         }
5368
5369         map = em->map_lookup;
5370         offset = logical - em->start;
5371
5372         stripe_len = map->stripe_len;
5373         stripe_nr = offset;
5374         /*
5375          * stripe_nr counts the total number of stripes we have to stride
5376          * to get to this block
5377          */
5378         stripe_nr = div64_u64(stripe_nr, stripe_len);
5379
5380         stripe_offset = stripe_nr * stripe_len;
5381         if (offset < stripe_offset) {
5382                 btrfs_crit(fs_info, "stripe math has gone wrong, "
5383                            "stripe_offset=%llu, offset=%llu, start=%llu, "
5384                            "logical=%llu, stripe_len=%llu",
5385                            stripe_offset, offset, em->start, logical,
5386                            stripe_len);
5387                 free_extent_map(em);
5388                 return -EINVAL;
5389         }
5390
5391         /* stripe_offset is the offset of this block in its stripe*/
5392         stripe_offset = offset - stripe_offset;
5393
5394         /* if we're here for raid56, we need to know the stripe aligned start */
5395         if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5396                 unsigned long full_stripe_len = stripe_len * nr_data_stripes(map);
5397                 raid56_full_stripe_start = offset;
5398
5399                 /* allow a write of a full stripe, but make sure we don't
5400                  * allow straddling of stripes
5401                  */
5402                 raid56_full_stripe_start = div64_u64(raid56_full_stripe_start,
5403                                 full_stripe_len);
5404                 raid56_full_stripe_start *= full_stripe_len;
5405         }
5406
5407         if (op == REQ_OP_DISCARD) {
5408                 /* we don't discard raid56 yet */
5409                 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5410                         ret = -EOPNOTSUPP;
5411                         goto out;
5412                 }
5413                 *length = min_t(u64, em->len - offset, *length);
5414         } else if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
5415                 u64 max_len;
5416                 /* For writes to RAID[56], allow a full stripeset across all disks.
5417                    For other RAID types and for RAID[56] reads, just allow a single
5418                    stripe (on a single disk). */
5419                 if ((map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
5420                     (op == REQ_OP_WRITE)) {
5421                         max_len = stripe_len * nr_data_stripes(map) -
5422                                 (offset - raid56_full_stripe_start);
5423                 } else {
5424                         /* we limit the length of each bio to what fits in a stripe */
5425                         max_len = stripe_len - stripe_offset;
5426                 }
5427                 *length = min_t(u64, em->len - offset, max_len);
5428         } else {
5429                 *length = em->len - offset;
5430         }
5431
5432         /* This is for when we're called from btrfs_merge_bio_hook() and all
5433            it cares about is the length */
5434         if (!bbio_ret)
5435                 goto out;
5436
5437         btrfs_dev_replace_lock(dev_replace, 0);
5438         dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
5439         if (!dev_replace_is_ongoing)
5440                 btrfs_dev_replace_unlock(dev_replace, 0);
5441         else
5442                 btrfs_dev_replace_set_lock_blocking(dev_replace);
5443
5444         if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
5445             op != REQ_OP_WRITE && op != REQ_OP_DISCARD &&
5446             op != REQ_GET_READ_MIRRORS && dev_replace->tgtdev != NULL) {
5447                 /*
5448                  * in dev-replace case, for repair case (that's the only
5449                  * case where the mirror is selected explicitly when
5450                  * calling btrfs_map_block), blocks left of the left cursor
5451                  * can also be read from the target drive.
5452                  * For REQ_GET_READ_MIRRORS, the target drive is added as
5453                  * the last one to the array of stripes. For READ, it also
5454                  * needs to be supported using the same mirror number.
5455                  * If the requested block is not left of the left cursor,
5456                  * EIO is returned. This can happen because btrfs_num_copies()
5457                  * returns one more in the dev-replace case.
5458                  */
5459                 u64 tmp_length = *length;
5460                 struct btrfs_bio *tmp_bbio = NULL;
5461                 int tmp_num_stripes;
5462                 u64 srcdev_devid = dev_replace->srcdev->devid;
5463                 int index_srcdev = 0;
5464                 int found = 0;
5465                 u64 physical_of_found = 0;
5466
5467                 ret = __btrfs_map_block(fs_info, REQ_GET_READ_MIRRORS,
5468                              logical, &tmp_length, &tmp_bbio, 0, 0);
5469                 if (ret) {
5470                         WARN_ON(tmp_bbio != NULL);
5471                         goto out;
5472                 }
5473
5474                 tmp_num_stripes = tmp_bbio->num_stripes;
5475                 if (mirror_num > tmp_num_stripes) {
5476                         /*
5477                          * REQ_GET_READ_MIRRORS does not contain this
5478                          * mirror, that means that the requested area
5479                          * is not left of the left cursor
5480                          */
5481                         ret = -EIO;
5482                         btrfs_put_bbio(tmp_bbio);
5483                         goto out;
5484                 }
5485
5486                 /*
5487                  * process the rest of the function using the mirror_num
5488                  * of the source drive. Therefore look it up first.
5489                  * At the end, patch the device pointer to the one of the
5490                  * target drive.
5491                  */
5492                 for (i = 0; i < tmp_num_stripes; i++) {
5493                         if (tmp_bbio->stripes[i].dev->devid != srcdev_devid)
5494                                 continue;
5495
5496                         /*
5497                          * In case of DUP, in order to keep it simple, only add
5498                          * the mirror with the lowest physical address
5499                          */
5500                         if (found &&
5501                             physical_of_found <= tmp_bbio->stripes[i].physical)
5502                                 continue;
5503
5504                         index_srcdev = i;
5505                         found = 1;
5506                         physical_of_found = tmp_bbio->stripes[i].physical;
5507                 }
5508
5509                 btrfs_put_bbio(tmp_bbio);
5510
5511                 if (!found) {
5512                         WARN_ON(1);
5513                         ret = -EIO;
5514                         goto out;
5515                 }
5516
5517                 mirror_num = index_srcdev + 1;
5518                 patch_the_first_stripe_for_dev_replace = 1;
5519                 physical_to_patch_in_first_stripe = physical_of_found;
5520         } else if (mirror_num > map->num_stripes) {
5521                 mirror_num = 0;
5522         }
5523
5524         num_stripes = 1;
5525         stripe_index = 0;
5526         stripe_nr_orig = stripe_nr;
5527         stripe_nr_end = ALIGN(offset + *length, map->stripe_len);
5528         stripe_nr_end = div_u64(stripe_nr_end, map->stripe_len);
5529         stripe_end_offset = stripe_nr_end * map->stripe_len -
5530                             (offset + *length);
5531
5532         if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5533                 if (op == REQ_OP_DISCARD)
5534                         num_stripes = min_t(u64, map->num_stripes,
5535                                             stripe_nr_end - stripe_nr_orig);
5536                 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5537                                 &stripe_index);
5538                 if (op != REQ_OP_WRITE && op != REQ_OP_DISCARD &&
5539                     op != REQ_GET_READ_MIRRORS)
5540                         mirror_num = 1;
5541         } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
5542                 if (op == REQ_OP_WRITE || op == REQ_OP_DISCARD ||
5543                     op == REQ_GET_READ_MIRRORS)
5544                         num_stripes = map->num_stripes;
5545                 else if (mirror_num)
5546                         stripe_index = mirror_num - 1;
5547                 else {
5548                         stripe_index = find_live_mirror(fs_info, map, 0,
5549                                             map->num_stripes,
5550                                             current->pid % map->num_stripes,
5551                                             dev_replace_is_ongoing);
5552                         mirror_num = stripe_index + 1;
5553                 }
5554
5555         } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
5556                 if (op == REQ_OP_WRITE || op == REQ_OP_DISCARD ||
5557                     op == REQ_GET_READ_MIRRORS) {
5558                         num_stripes = map->num_stripes;
5559                 } else if (mirror_num) {
5560                         stripe_index = mirror_num - 1;
5561                 } else {
5562                         mirror_num = 1;
5563                 }
5564
5565         } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5566                 u32 factor = map->num_stripes / map->sub_stripes;
5567
5568                 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
5569                 stripe_index *= map->sub_stripes;
5570
5571                 if (op == REQ_OP_WRITE || op == REQ_GET_READ_MIRRORS)
5572                         num_stripes = map->sub_stripes;
5573                 else if (op == REQ_OP_DISCARD)
5574                         num_stripes = min_t(u64, map->sub_stripes *
5575                                             (stripe_nr_end - stripe_nr_orig),
5576                                             map->num_stripes);
5577                 else if (mirror_num)
5578                         stripe_index += mirror_num - 1;
5579                 else {
5580                         int old_stripe_index = stripe_index;
5581                         stripe_index = find_live_mirror(fs_info, map,
5582                                               stripe_index,
5583                                               map->sub_stripes, stripe_index +
5584                                               current->pid % map->sub_stripes,
5585                                               dev_replace_is_ongoing);
5586                         mirror_num = stripe_index - old_stripe_index + 1;
5587                 }
5588
5589         } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5590                 if (need_raid_map &&
5591                     (op == REQ_OP_WRITE || op == REQ_GET_READ_MIRRORS ||
5592                      mirror_num > 1)) {
5593                         /* push stripe_nr back to the start of the full stripe */
5594                         stripe_nr = div_u64(raid56_full_stripe_start,
5595                                         stripe_len * nr_data_stripes(map));
5596
5597                         /* RAID[56] write or recovery. Return all stripes */
5598                         num_stripes = map->num_stripes;
5599                         max_errors = nr_parity_stripes(map);
5600
5601                         *length = map->stripe_len;
5602                         stripe_index = 0;
5603                         stripe_offset = 0;
5604                 } else {
5605                         /*
5606                          * Mirror #0 or #1 means the original data block.
5607                          * Mirror #2 is RAID5 parity block.
5608                          * Mirror #3 is RAID6 Q block.
5609                          */
5610                         stripe_nr = div_u64_rem(stripe_nr,
5611                                         nr_data_stripes(map), &stripe_index);
5612                         if (mirror_num > 1)
5613                                 stripe_index = nr_data_stripes(map) +
5614                                                 mirror_num - 2;
5615
5616                         /* We distribute the parity blocks across stripes */
5617                         div_u64_rem(stripe_nr + stripe_index, map->num_stripes,
5618                                         &stripe_index);
5619                         if ((op != REQ_OP_WRITE && op != REQ_OP_DISCARD &&
5620                             op != REQ_GET_READ_MIRRORS) && mirror_num <= 1)
5621                                 mirror_num = 1;
5622                 }
5623         } else {
5624                 /*
5625                  * after this, stripe_nr is the number of stripes on this
5626                  * device we have to walk to find the data, and stripe_index is
5627                  * the number of our device in the stripe array
5628                  */
5629                 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5630                                 &stripe_index);
5631                 mirror_num = stripe_index + 1;
5632         }
5633         if (stripe_index >= map->num_stripes) {
5634                 btrfs_crit(fs_info, "stripe index math went horribly wrong, "
5635                            "got stripe_index=%u, num_stripes=%u",
5636                            stripe_index, map->num_stripes);
5637                 ret = -EINVAL;
5638                 goto out;
5639         }
5640
5641         num_alloc_stripes = num_stripes;
5642         if (dev_replace_is_ongoing) {
5643                 if (op == REQ_OP_WRITE || op == REQ_OP_DISCARD)
5644                         num_alloc_stripes <<= 1;
5645                 if (op == REQ_GET_READ_MIRRORS)
5646                         num_alloc_stripes++;
5647                 tgtdev_indexes = num_stripes;
5648         }
5649
5650         bbio = alloc_btrfs_bio(num_alloc_stripes, tgtdev_indexes);
5651         if (!bbio) {
5652                 ret = -ENOMEM;
5653                 goto out;
5654         }
5655         if (dev_replace_is_ongoing)
5656                 bbio->tgtdev_map = (int *)(bbio->stripes + num_alloc_stripes);
5657
5658         /* build raid_map */
5659         if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK &&
5660             need_raid_map &&
5661             ((op == REQ_OP_WRITE || op == REQ_GET_READ_MIRRORS) ||
5662             mirror_num > 1)) {
5663                 u64 tmp;
5664                 unsigned rot;
5665
5666                 bbio->raid_map = (u64 *)((void *)bbio->stripes +
5667                                  sizeof(struct btrfs_bio_stripe) *
5668                                  num_alloc_stripes +
5669                                  sizeof(int) * tgtdev_indexes);
5670
5671                 /* Work out the disk rotation on this stripe-set */
5672                 div_u64_rem(stripe_nr, num_stripes, &rot);
5673
5674                 /* Fill in the logical address of each stripe */
5675                 tmp = stripe_nr * nr_data_stripes(map);
5676                 for (i = 0; i < nr_data_stripes(map); i++)
5677                         bbio->raid_map[(i+rot) % num_stripes] =
5678                                 em->start + (tmp + i) * map->stripe_len;
5679
5680                 bbio->raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE;
5681                 if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5682                         bbio->raid_map[(i+rot+1) % num_stripes] =
5683                                 RAID6_Q_STRIPE;
5684         }
5685
5686         if (op == REQ_OP_DISCARD) {
5687                 u32 factor = 0;
5688                 u32 sub_stripes = 0;
5689                 u64 stripes_per_dev = 0;
5690                 u32 remaining_stripes = 0;
5691                 u32 last_stripe = 0;
5692
5693                 if (map->type &
5694                     (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID10)) {
5695                         if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5696                                 sub_stripes = 1;
5697                         else
5698                                 sub_stripes = map->sub_stripes;
5699
5700                         factor = map->num_stripes / sub_stripes;
5701                         stripes_per_dev = div_u64_rem(stripe_nr_end -
5702                                                       stripe_nr_orig,
5703                                                       factor,
5704                                                       &remaining_stripes);
5705                         div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
5706                         last_stripe *= sub_stripes;
5707                 }
5708
5709                 for (i = 0; i < num_stripes; i++) {
5710                         bbio->stripes[i].physical =
5711                                 map->stripes[stripe_index].physical +
5712                                 stripe_offset + stripe_nr * map->stripe_len;
5713                         bbio->stripes[i].dev = map->stripes[stripe_index].dev;
5714
5715                         if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5716                                          BTRFS_BLOCK_GROUP_RAID10)) {
5717                                 bbio->stripes[i].length = stripes_per_dev *
5718                                                           map->stripe_len;
5719
5720                                 if (i / sub_stripes < remaining_stripes)
5721                                         bbio->stripes[i].length +=
5722                                                 map->stripe_len;
5723
5724                                 /*
5725                                  * Special for the first stripe and
5726                                  * the last stripe:
5727                                  *
5728                                  * |-------|...|-------|
5729                                  *     |----------|
5730                                  *    off     end_off
5731                                  */
5732                                 if (i < sub_stripes)
5733                                         bbio->stripes[i].length -=
5734                                                 stripe_offset;
5735
5736                                 if (stripe_index >= last_stripe &&
5737                                     stripe_index <= (last_stripe +
5738                                                      sub_stripes - 1))
5739                                         bbio->stripes[i].length -=
5740                                                 stripe_end_offset;
5741
5742                                 if (i == sub_stripes - 1)
5743                                         stripe_offset = 0;
5744                         } else
5745                                 bbio->stripes[i].length = *length;
5746
5747                         stripe_index++;
5748                         if (stripe_index == map->num_stripes) {
5749                                 /* This could only happen for RAID0/10 */
5750                                 stripe_index = 0;
5751                                 stripe_nr++;
5752                         }
5753                 }
5754         } else {
5755                 for (i = 0; i < num_stripes; i++) {
5756                         bbio->stripes[i].physical =
5757                                 map->stripes[stripe_index].physical +
5758                                 stripe_offset +
5759                                 stripe_nr * map->stripe_len;
5760                         bbio->stripes[i].dev =
5761                                 map->stripes[stripe_index].dev;
5762                         stripe_index++;
5763                 }
5764         }
5765
5766         if (op == REQ_OP_WRITE || op == REQ_GET_READ_MIRRORS)
5767                 max_errors = btrfs_chunk_max_errors(map);
5768
5769         if (bbio->raid_map)
5770                 sort_parity_stripes(bbio, num_stripes);
5771
5772         tgtdev_indexes = 0;
5773         if (dev_replace_is_ongoing &&
5774            (op == REQ_OP_WRITE || op == REQ_OP_DISCARD) &&
5775             dev_replace->tgtdev != NULL) {
5776                 int index_where_to_add;
5777                 u64 srcdev_devid = dev_replace->srcdev->devid;
5778
5779                 /*
5780                  * duplicate the write operations while the dev replace
5781                  * procedure is running. Since the copying of the old disk
5782                  * to the new disk takes place at run time while the
5783                  * filesystem is mounted writable, the regular write
5784                  * operations to the old disk have to be duplicated to go
5785                  * to the new disk as well.
5786                  * Note that device->missing is handled by the caller, and
5787                  * that the write to the old disk is already set up in the
5788                  * stripes array.
5789                  */
5790                 index_where_to_add = num_stripes;
5791                 for (i = 0; i < num_stripes; i++) {
5792                         if (bbio->stripes[i].dev->devid == srcdev_devid) {
5793                                 /* write to new disk, too */
5794                                 struct btrfs_bio_stripe *new =
5795                                         bbio->stripes + index_where_to_add;
5796                                 struct btrfs_bio_stripe *old =
5797                                         bbio->stripes + i;
5798
5799                                 new->physical = old->physical;
5800                                 new->length = old->length;
5801                                 new->dev = dev_replace->tgtdev;
5802                                 bbio->tgtdev_map[i] = index_where_to_add;
5803                                 index_where_to_add++;
5804                                 max_errors++;
5805                                 tgtdev_indexes++;
5806                         }
5807                 }
5808                 num_stripes = index_where_to_add;
5809         } else if (dev_replace_is_ongoing && (op == REQ_GET_READ_MIRRORS) &&
5810                    dev_replace->tgtdev != NULL) {
5811                 u64 srcdev_devid = dev_replace->srcdev->devid;
5812                 int index_srcdev = 0;
5813                 int found = 0;
5814                 u64 physical_of_found = 0;
5815
5816                 /*
5817                  * During the dev-replace procedure, the target drive can
5818                  * also be used to read data in case it is needed to repair
5819                  * a corrupt block elsewhere. This is possible if the
5820                  * requested area is left of the left cursor. In this area,
5821                  * the target drive is a full copy of the source drive.
5822                  */
5823                 for (i = 0; i < num_stripes; i++) {
5824                         if (bbio->stripes[i].dev->devid == srcdev_devid) {
5825                                 /*
5826                                  * In case of DUP, in order to keep it
5827                                  * simple, only add the mirror with the
5828                                  * lowest physical address
5829                                  */
5830                                 if (found &&
5831                                     physical_of_found <=
5832                                      bbio->stripes[i].physical)
5833                                         continue;
5834                                 index_srcdev = i;
5835                                 found = 1;
5836                                 physical_of_found = bbio->stripes[i].physical;
5837                         }
5838                 }
5839                 if (found) {
5840                         struct btrfs_bio_stripe *tgtdev_stripe =
5841                                 bbio->stripes + num_stripes;
5842
5843                         tgtdev_stripe->physical = physical_of_found;
5844                         tgtdev_stripe->length =
5845                                 bbio->stripes[index_srcdev].length;
5846                         tgtdev_stripe->dev = dev_replace->tgtdev;
5847                         bbio->tgtdev_map[index_srcdev] = num_stripes;
5848
5849                         tgtdev_indexes++;
5850                         num_stripes++;
5851                 }
5852         }
5853
5854         *bbio_ret = bbio;
5855         bbio->map_type = map->type;
5856         bbio->num_stripes = num_stripes;
5857         bbio->max_errors = max_errors;
5858         bbio->mirror_num = mirror_num;
5859         bbio->num_tgtdevs = tgtdev_indexes;
5860
5861         /*
5862          * this is the case that REQ_READ && dev_replace_is_ongoing &&
5863          * mirror_num == num_stripes + 1 && dev_replace target drive is
5864          * available as a mirror
5865          */
5866         if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) {
5867                 WARN_ON(num_stripes > 1);
5868                 bbio->stripes[0].dev = dev_replace->tgtdev;
5869                 bbio->stripes[0].physical = physical_to_patch_in_first_stripe;
5870                 bbio->mirror_num = map->num_stripes + 1;
5871         }
5872 out:
5873         if (dev_replace_is_ongoing) {
5874                 btrfs_dev_replace_clear_lock_blocking(dev_replace);
5875                 btrfs_dev_replace_unlock(dev_replace, 0);
5876         }
5877         free_extent_map(em);
5878         return ret;
5879 }
5880
5881 int btrfs_map_block(struct btrfs_fs_info *fs_info, int op,
5882                       u64 logical, u64 *length,
5883                       struct btrfs_bio **bbio_ret, int mirror_num)
5884 {
5885         return __btrfs_map_block(fs_info, op, logical, length, bbio_ret,
5886                                  mirror_num, 0);
5887 }
5888
5889 /* For Scrub/replace */
5890 int btrfs_map_sblock(struct btrfs_fs_info *fs_info, int op,
5891                      u64 logical, u64 *length,
5892                      struct btrfs_bio **bbio_ret, int mirror_num,
5893                      int need_raid_map)
5894 {
5895         return __btrfs_map_block(fs_info, op, logical, length, bbio_ret,
5896                                  mirror_num, need_raid_map);
5897 }
5898
5899 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
5900                      u64 chunk_start, u64 physical, u64 devid,
5901                      u64 **logical, int *naddrs, int *stripe_len)
5902 {
5903         struct extent_map_tree *em_tree = &map_tree->map_tree;
5904         struct extent_map *em;
5905         struct map_lookup *map;
5906         u64 *buf;
5907         u64 bytenr;
5908         u64 length;
5909         u64 stripe_nr;
5910         u64 rmap_len;
5911         int i, j, nr = 0;
5912
5913         read_lock(&em_tree->lock);
5914         em = lookup_extent_mapping(em_tree, chunk_start, 1);
5915         read_unlock(&em_tree->lock);
5916
5917         if (!em) {
5918                 printk(KERN_ERR "BTRFS: couldn't find em for chunk %Lu\n",
5919                        chunk_start);
5920                 return -EIO;
5921         }
5922
5923         if (em->start != chunk_start) {
5924                 printk(KERN_ERR "BTRFS: bad chunk start, em=%Lu, wanted=%Lu\n",
5925                        em->start, chunk_start);
5926                 free_extent_map(em);
5927                 return -EIO;
5928         }
5929         map = em->map_lookup;
5930
5931         length = em->len;
5932         rmap_len = map->stripe_len;
5933
5934         if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5935                 length = div_u64(length, map->num_stripes / map->sub_stripes);
5936         else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5937                 length = div_u64(length, map->num_stripes);
5938         else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5939                 length = div_u64(length, nr_data_stripes(map));
5940                 rmap_len = map->stripe_len * nr_data_stripes(map);
5941         }
5942
5943         buf = kcalloc(map->num_stripes, sizeof(u64), GFP_NOFS);
5944         BUG_ON(!buf); /* -ENOMEM */
5945
5946         for (i = 0; i < map->num_stripes; i++) {
5947                 if (devid && map->stripes[i].dev->devid != devid)
5948                         continue;
5949                 if (map->stripes[i].physical > physical ||
5950                     map->stripes[i].physical + length <= physical)
5951                         continue;
5952
5953                 stripe_nr = physical - map->stripes[i].physical;
5954                 stripe_nr = div_u64(stripe_nr, map->stripe_len);
5955
5956                 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5957                         stripe_nr = stripe_nr * map->num_stripes + i;
5958                         stripe_nr = div_u64(stripe_nr, map->sub_stripes);
5959                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5960                         stripe_nr = stripe_nr * map->num_stripes + i;
5961                 } /* else if RAID[56], multiply by nr_data_stripes().
5962                    * Alternatively, just use rmap_len below instead of
5963                    * map->stripe_len */
5964
5965                 bytenr = chunk_start + stripe_nr * rmap_len;
5966                 WARN_ON(nr >= map->num_stripes);
5967                 for (j = 0; j < nr; j++) {
5968                         if (buf[j] == bytenr)
5969                                 break;
5970                 }
5971                 if (j == nr) {
5972                         WARN_ON(nr >= map->num_stripes);
5973                         buf[nr++] = bytenr;
5974                 }
5975         }
5976
5977         *logical = buf;
5978         *naddrs = nr;
5979         *stripe_len = rmap_len;
5980
5981         free_extent_map(em);
5982         return 0;
5983 }
5984
5985 static inline void btrfs_end_bbio(struct btrfs_bio *bbio, struct bio *bio)
5986 {
5987         bio->bi_private = bbio->private;
5988         bio->bi_end_io = bbio->end_io;
5989         bio_endio(bio);
5990
5991         btrfs_put_bbio(bbio);
5992 }
5993
5994 static void btrfs_end_bio(struct bio *bio)
5995 {
5996         struct btrfs_bio *bbio = bio->bi_private;
5997         int is_orig_bio = 0;
5998
5999         if (bio->bi_error) {
6000                 atomic_inc(&bbio->error);
6001                 if (bio->bi_error == -EIO || bio->bi_error == -EREMOTEIO) {
6002                         unsigned int stripe_index =
6003                                 btrfs_io_bio(bio)->stripe_index;
6004                         struct btrfs_device *dev;
6005
6006                         BUG_ON(stripe_index >= bbio->num_stripes);
6007                         dev = bbio->stripes[stripe_index].dev;
6008                         if (dev->bdev) {
6009                                 if (bio_op(bio) == REQ_OP_WRITE)
6010                                         btrfs_dev_stat_inc(dev,
6011                                                 BTRFS_DEV_STAT_WRITE_ERRS);
6012                                 else
6013                                         btrfs_dev_stat_inc(dev,
6014                                                 BTRFS_DEV_STAT_READ_ERRS);
6015                                 if ((bio->bi_rw & WRITE_FLUSH) == WRITE_FLUSH)
6016                                         btrfs_dev_stat_inc(dev,
6017                                                 BTRFS_DEV_STAT_FLUSH_ERRS);
6018                                 btrfs_dev_stat_print_on_error(dev);
6019                         }
6020                 }
6021         }
6022
6023         if (bio == bbio->orig_bio)
6024                 is_orig_bio = 1;
6025
6026         btrfs_bio_counter_dec(bbio->fs_info);
6027
6028         if (atomic_dec_and_test(&bbio->stripes_pending)) {
6029                 if (!is_orig_bio) {
6030                         bio_put(bio);
6031                         bio = bbio->orig_bio;
6032                 }
6033
6034                 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
6035                 /* only send an error to the higher layers if it is
6036                  * beyond the tolerance of the btrfs bio
6037                  */
6038                 if (atomic_read(&bbio->error) > bbio->max_errors) {
6039                         bio->bi_error = -EIO;
6040                 } else {
6041                         /*
6042                          * this bio is actually up to date, we didn't
6043                          * go over the max number of errors
6044                          */
6045                         bio->bi_error = 0;
6046                 }
6047
6048                 btrfs_end_bbio(bbio, bio);
6049         } else if (!is_orig_bio) {
6050                 bio_put(bio);
6051         }
6052 }
6053
6054 /*
6055  * see run_scheduled_bios for a description of why bios are collected for
6056  * async submit.
6057  *
6058  * This will add one bio to the pending list for a device and make sure
6059  * the work struct is scheduled.
6060  */
6061 static noinline void btrfs_schedule_bio(struct btrfs_root *root,
6062                                         struct btrfs_device *device,
6063                                         struct bio *bio)
6064 {
6065         int should_queue = 1;
6066         struct btrfs_pending_bios *pending_bios;
6067
6068         if (device->missing || !device->bdev) {
6069                 bio_io_error(bio);
6070                 return;
6071         }
6072
6073         /* don't bother with additional async steps for reads, right now */
6074         if (bio_op(bio) == REQ_OP_READ) {
6075                 bio_get(bio);
6076                 btrfsic_submit_bio(bio);
6077                 bio_put(bio);
6078                 return;
6079         }
6080
6081         /*
6082          * nr_async_bios allows us to reliably return congestion to the
6083          * higher layers.  Otherwise, the async bio makes it appear we have
6084          * made progress against dirty pages when we've really just put it
6085          * on a queue for later
6086          */
6087         atomic_inc(&root->fs_info->nr_async_bios);
6088         WARN_ON(bio->bi_next);
6089         bio->bi_next = NULL;
6090
6091         spin_lock(&device->io_lock);
6092         if (bio->bi_rw & REQ_SYNC)
6093                 pending_bios = &device->pending_sync_bios;
6094         else
6095                 pending_bios = &device->pending_bios;
6096
6097         if (pending_bios->tail)
6098                 pending_bios->tail->bi_next = bio;
6099
6100         pending_bios->tail = bio;
6101         if (!pending_bios->head)
6102                 pending_bios->head = bio;
6103         if (device->running_pending)
6104                 should_queue = 0;
6105
6106         spin_unlock(&device->io_lock);
6107
6108         if (should_queue)
6109                 btrfs_queue_work(root->fs_info->submit_workers,
6110                                  &device->work);
6111 }
6112
6113 static void submit_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio,
6114                               struct bio *bio, u64 physical, int dev_nr,
6115                               int async)
6116 {
6117         struct btrfs_device *dev = bbio->stripes[dev_nr].dev;
6118
6119         bio->bi_private = bbio;
6120         btrfs_io_bio(bio)->stripe_index = dev_nr;
6121         bio->bi_end_io = btrfs_end_bio;
6122         bio->bi_iter.bi_sector = physical >> 9;
6123 #ifdef DEBUG
6124         {
6125                 struct rcu_string *name;
6126
6127                 rcu_read_lock();
6128                 name = rcu_dereference(dev->name);
6129                 pr_debug("btrfs_map_bio: rw %d 0x%x, sector=%llu, dev=%lu "
6130                          "(%s id %llu), size=%u\n", bio_op(bio), bio->bi_rw,
6131                          (u64)bio->bi_iter.bi_sector, (u_long)dev->bdev->bd_dev,
6132                          name->str, dev->devid, bio->bi_iter.bi_size);
6133                 rcu_read_unlock();
6134         }
6135 #endif
6136         bio->bi_bdev = dev->bdev;
6137
6138         btrfs_bio_counter_inc_noblocked(root->fs_info);
6139
6140         if (async)
6141                 btrfs_schedule_bio(root, dev, bio);
6142         else
6143                 btrfsic_submit_bio(bio);
6144 }
6145
6146 static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical)
6147 {
6148         atomic_inc(&bbio->error);
6149         if (atomic_dec_and_test(&bbio->stripes_pending)) {
6150                 /* Should be the original bio. */
6151                 WARN_ON(bio != bbio->orig_bio);
6152
6153                 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
6154                 bio->bi_iter.bi_sector = logical >> 9;
6155                 bio->bi_error = -EIO;
6156                 btrfs_end_bbio(bbio, bio);
6157         }
6158 }
6159
6160 int btrfs_map_bio(struct btrfs_root *root, struct bio *bio,
6161                   int mirror_num, int async_submit)
6162 {
6163         struct btrfs_device *dev;
6164         struct bio *first_bio = bio;
6165         u64 logical = (u64)bio->bi_iter.bi_sector << 9;
6166         u64 length = 0;
6167         u64 map_length;
6168         int ret;
6169         int dev_nr;
6170         int total_devs;
6171         struct btrfs_bio *bbio = NULL;
6172
6173         length = bio->bi_iter.bi_size;
6174         map_length = length;
6175
6176         btrfs_bio_counter_inc_blocked(root->fs_info);
6177         ret = __btrfs_map_block(root->fs_info, bio_op(bio), logical,
6178                                 &map_length, &bbio, mirror_num, 1);
6179         if (ret) {
6180                 btrfs_bio_counter_dec(root->fs_info);
6181                 return ret;
6182         }
6183
6184         total_devs = bbio->num_stripes;
6185         bbio->orig_bio = first_bio;
6186         bbio->private = first_bio->bi_private;
6187         bbio->end_io = first_bio->bi_end_io;
6188         bbio->fs_info = root->fs_info;
6189         atomic_set(&bbio->stripes_pending, bbio->num_stripes);
6190
6191         if ((bbio->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
6192             ((bio_op(bio) == REQ_OP_WRITE) || (mirror_num > 1))) {
6193                 /* In this case, map_length has been set to the length of
6194                    a single stripe; not the whole write */
6195                 if (bio_op(bio) == REQ_OP_WRITE) {
6196                         ret = raid56_parity_write(root, bio, bbio, map_length);
6197                 } else {
6198                         ret = raid56_parity_recover(root, bio, bbio, map_length,
6199                                                     mirror_num, 1);
6200                 }
6201
6202                 btrfs_bio_counter_dec(root->fs_info);
6203                 return ret;
6204         }
6205
6206         if (map_length < length) {
6207                 btrfs_crit(root->fs_info, "mapping failed logical %llu bio len %llu len %llu",
6208                         logical, length, map_length);
6209                 BUG();
6210         }
6211
6212         for (dev_nr = 0; dev_nr < total_devs; dev_nr++) {
6213                 dev = bbio->stripes[dev_nr].dev;
6214                 if (!dev || !dev->bdev ||
6215                     (bio_op(bio) == REQ_OP_WRITE && !dev->writeable)) {
6216                         bbio_error(bbio, first_bio, logical);
6217                         continue;
6218                 }
6219
6220                 if (dev_nr < total_devs - 1) {
6221                         bio = btrfs_bio_clone(first_bio, GFP_NOFS);
6222                         BUG_ON(!bio); /* -ENOMEM */
6223                 } else
6224                         bio = first_bio;
6225
6226                 submit_stripe_bio(root, bbio, bio,
6227                                   bbio->stripes[dev_nr].physical, dev_nr,
6228                                   async_submit);
6229         }
6230         btrfs_bio_counter_dec(root->fs_info);
6231         return 0;
6232 }
6233
6234 struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid,
6235                                        u8 *uuid, u8 *fsid)
6236 {
6237         struct btrfs_device *device;
6238         struct btrfs_fs_devices *cur_devices;
6239
6240         cur_devices = fs_info->fs_devices;
6241         while (cur_devices) {
6242                 if (!fsid ||
6243                     !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
6244                         device = __find_device(&cur_devices->devices,
6245                                                devid, uuid);
6246                         if (device)
6247                                 return device;
6248                 }
6249                 cur_devices = cur_devices->seed;
6250         }
6251         return NULL;
6252 }
6253
6254 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
6255                                             struct btrfs_fs_devices *fs_devices,
6256                                             u64 devid, u8 *dev_uuid)
6257 {
6258         struct btrfs_device *device;
6259
6260         device = btrfs_alloc_device(NULL, &devid, dev_uuid);
6261         if (IS_ERR(device))
6262                 return NULL;
6263
6264         list_add(&device->dev_list, &fs_devices->devices);
6265         device->fs_devices = fs_devices;
6266         fs_devices->num_devices++;
6267
6268         device->missing = 1;
6269         fs_devices->missing_devices++;
6270
6271         return device;
6272 }
6273
6274 /**
6275  * btrfs_alloc_device - allocate struct btrfs_device
6276  * @fs_info:    used only for generating a new devid, can be NULL if
6277  *              devid is provided (i.e. @devid != NULL).
6278  * @devid:      a pointer to devid for this device.  If NULL a new devid
6279  *              is generated.
6280  * @uuid:       a pointer to UUID for this device.  If NULL a new UUID
6281  *              is generated.
6282  *
6283  * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
6284  * on error.  Returned struct is not linked onto any lists and can be
6285  * destroyed with kfree() right away.
6286  */
6287 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
6288                                         const u64 *devid,
6289                                         const u8 *uuid)
6290 {
6291         struct btrfs_device *dev;
6292         u64 tmp;
6293
6294         if (WARN_ON(!devid && !fs_info))
6295                 return ERR_PTR(-EINVAL);
6296
6297         dev = __alloc_device();
6298         if (IS_ERR(dev))
6299                 return dev;
6300
6301         if (devid)
6302                 tmp = *devid;
6303         else {
6304                 int ret;
6305
6306                 ret = find_next_devid(fs_info, &tmp);
6307                 if (ret) {
6308                         kfree(dev);
6309                         return ERR_PTR(ret);
6310                 }
6311         }
6312         dev->devid = tmp;
6313
6314         if (uuid)
6315                 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE);
6316         else
6317                 generate_random_uuid(dev->uuid);
6318
6319         btrfs_init_work(&dev->work, btrfs_submit_helper,
6320                         pending_bios_fn, NULL, NULL);
6321
6322         return dev;
6323 }
6324
6325 /* Return -EIO if any error, otherwise return 0. */
6326 static int btrfs_check_chunk_valid(struct btrfs_root *root,
6327                                    struct extent_buffer *leaf,
6328                                    struct btrfs_chunk *chunk, u64 logical)
6329 {
6330         u64 length;
6331         u64 stripe_len;
6332         u16 num_stripes;
6333         u16 sub_stripes;
6334         u64 type;
6335
6336         length = btrfs_chunk_length(leaf, chunk);
6337         stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6338         num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6339         sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
6340         type = btrfs_chunk_type(leaf, chunk);
6341
6342         if (!num_stripes) {
6343                 btrfs_err(root->fs_info, "invalid chunk num_stripes: %u",
6344                           num_stripes);
6345                 return -EIO;
6346         }
6347         if (!IS_ALIGNED(logical, root->sectorsize)) {
6348                 btrfs_err(root->fs_info,
6349                           "invalid chunk logical %llu", logical);
6350                 return -EIO;
6351         }
6352         if (btrfs_chunk_sector_size(leaf, chunk) != root->sectorsize) {
6353                 btrfs_err(root->fs_info, "invalid chunk sectorsize %u",
6354                           btrfs_chunk_sector_size(leaf, chunk));
6355                 return -EIO;
6356         }
6357         if (!length || !IS_ALIGNED(length, root->sectorsize)) {
6358                 btrfs_err(root->fs_info,
6359                         "invalid chunk length %llu", length);
6360                 return -EIO;
6361         }
6362         if (!is_power_of_2(stripe_len) || stripe_len != BTRFS_STRIPE_LEN) {
6363                 btrfs_err(root->fs_info, "invalid chunk stripe length: %llu",
6364                           stripe_len);
6365                 return -EIO;
6366         }
6367         if (~(BTRFS_BLOCK_GROUP_TYPE_MASK | BTRFS_BLOCK_GROUP_PROFILE_MASK) &
6368             type) {
6369                 btrfs_err(root->fs_info, "unrecognized chunk type: %llu",
6370                           ~(BTRFS_BLOCK_GROUP_TYPE_MASK |
6371                             BTRFS_BLOCK_GROUP_PROFILE_MASK) &
6372                           btrfs_chunk_type(leaf, chunk));
6373                 return -EIO;
6374         }
6375         if ((type & BTRFS_BLOCK_GROUP_RAID10 && sub_stripes != 2) ||
6376             (type & BTRFS_BLOCK_GROUP_RAID1 && num_stripes < 1) ||
6377             (type & BTRFS_BLOCK_GROUP_RAID5 && num_stripes < 2) ||
6378             (type & BTRFS_BLOCK_GROUP_RAID6 && num_stripes < 3) ||
6379             (type & BTRFS_BLOCK_GROUP_DUP && num_stripes > 2) ||
6380             ((type & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 &&
6381              num_stripes != 1)) {
6382                 btrfs_err(root->fs_info,
6383                         "invalid num_stripes:sub_stripes %u:%u for profile %llu",
6384                         num_stripes, sub_stripes,
6385                         type & BTRFS_BLOCK_GROUP_PROFILE_MASK);
6386                 return -EIO;
6387         }
6388
6389         return 0;
6390 }
6391
6392 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
6393                           struct extent_buffer *leaf,
6394                           struct btrfs_chunk *chunk)
6395 {
6396         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
6397         struct map_lookup *map;
6398         struct extent_map *em;
6399         u64 logical;
6400         u64 length;
6401         u64 stripe_len;
6402         u64 devid;
6403         u8 uuid[BTRFS_UUID_SIZE];
6404         int num_stripes;
6405         int ret;
6406         int i;
6407
6408         logical = key->offset;
6409         length = btrfs_chunk_length(leaf, chunk);
6410         stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6411         num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6412
6413         ret = btrfs_check_chunk_valid(root, leaf, chunk, logical);
6414         if (ret)
6415                 return ret;
6416
6417         read_lock(&map_tree->map_tree.lock);
6418         em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
6419         read_unlock(&map_tree->map_tree.lock);
6420
6421         /* already mapped? */
6422         if (em && em->start <= logical && em->start + em->len > logical) {
6423                 free_extent_map(em);
6424                 return 0;
6425         } else if (em) {
6426                 free_extent_map(em);
6427         }
6428
6429         em = alloc_extent_map();
6430         if (!em)
6431                 return -ENOMEM;
6432         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
6433         if (!map) {
6434                 free_extent_map(em);
6435                 return -ENOMEM;
6436         }
6437
6438         set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
6439         em->map_lookup = map;
6440         em->start = logical;
6441         em->len = length;
6442         em->orig_start = 0;
6443         em->block_start = 0;
6444         em->block_len = em->len;
6445
6446         map->num_stripes = num_stripes;
6447         map->io_width = btrfs_chunk_io_width(leaf, chunk);
6448         map->io_align = btrfs_chunk_io_align(leaf, chunk);
6449         map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
6450         map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6451         map->type = btrfs_chunk_type(leaf, chunk);
6452         map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
6453         for (i = 0; i < num_stripes; i++) {
6454                 map->stripes[i].physical =
6455                         btrfs_stripe_offset_nr(leaf, chunk, i);
6456                 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
6457                 read_extent_buffer(leaf, uuid, (unsigned long)
6458                                    btrfs_stripe_dev_uuid_nr(chunk, i),
6459                                    BTRFS_UUID_SIZE);
6460                 map->stripes[i].dev = btrfs_find_device(root->fs_info, devid,
6461                                                         uuid, NULL);
6462                 if (!map->stripes[i].dev &&
6463                     !btrfs_test_opt(root->fs_info, DEGRADED)) {
6464                         free_extent_map(em);
6465                         return -EIO;
6466                 }
6467                 if (!map->stripes[i].dev) {
6468                         map->stripes[i].dev =
6469                                 add_missing_dev(root, root->fs_info->fs_devices,
6470                                                 devid, uuid);
6471                         if (!map->stripes[i].dev) {
6472                                 free_extent_map(em);
6473                                 return -EIO;
6474                         }
6475                         btrfs_warn(root->fs_info, "devid %llu uuid %pU is missing",
6476                                                 devid, uuid);
6477                 }
6478                 map->stripes[i].dev->in_fs_metadata = 1;
6479         }
6480
6481         write_lock(&map_tree->map_tree.lock);
6482         ret = add_extent_mapping(&map_tree->map_tree, em, 0);
6483         write_unlock(&map_tree->map_tree.lock);
6484         BUG_ON(ret); /* Tree corruption */
6485         free_extent_map(em);
6486
6487         return 0;
6488 }
6489
6490 static void fill_device_from_item(struct extent_buffer *leaf,
6491                                  struct btrfs_dev_item *dev_item,
6492                                  struct btrfs_device *device)
6493 {
6494         unsigned long ptr;
6495
6496         device->devid = btrfs_device_id(leaf, dev_item);
6497         device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
6498         device->total_bytes = device->disk_total_bytes;
6499         device->commit_total_bytes = device->disk_total_bytes;
6500         device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
6501         device->commit_bytes_used = device->bytes_used;
6502         device->type = btrfs_device_type(leaf, dev_item);
6503         device->io_align = btrfs_device_io_align(leaf, dev_item);
6504         device->io_width = btrfs_device_io_width(leaf, dev_item);
6505         device->sector_size = btrfs_device_sector_size(leaf, dev_item);
6506         WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
6507         device->is_tgtdev_for_dev_replace = 0;
6508
6509         ptr = btrfs_device_uuid(dev_item);
6510         read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
6511 }
6512
6513 static struct btrfs_fs_devices *open_seed_devices(struct btrfs_root *root,
6514                                                   u8 *fsid)
6515 {
6516         struct btrfs_fs_devices *fs_devices;
6517         int ret;
6518
6519         BUG_ON(!mutex_is_locked(&uuid_mutex));
6520
6521         fs_devices = root->fs_info->fs_devices->seed;
6522         while (fs_devices) {
6523                 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE))
6524                         return fs_devices;
6525
6526                 fs_devices = fs_devices->seed;
6527         }
6528
6529         fs_devices = find_fsid(fsid);
6530         if (!fs_devices) {
6531                 if (!btrfs_test_opt(root->fs_info, DEGRADED))
6532                         return ERR_PTR(-ENOENT);
6533
6534                 fs_devices = alloc_fs_devices(fsid);
6535                 if (IS_ERR(fs_devices))
6536                         return fs_devices;
6537
6538                 fs_devices->seeding = 1;
6539                 fs_devices->opened = 1;
6540                 return fs_devices;
6541         }
6542
6543         fs_devices = clone_fs_devices(fs_devices);
6544         if (IS_ERR(fs_devices))
6545                 return fs_devices;
6546
6547         ret = __btrfs_open_devices(fs_devices, FMODE_READ,
6548                                    root->fs_info->bdev_holder);
6549         if (ret) {
6550                 free_fs_devices(fs_devices);
6551                 fs_devices = ERR_PTR(ret);
6552                 goto out;
6553         }
6554
6555         if (!fs_devices->seeding) {
6556                 __btrfs_close_devices(fs_devices);
6557                 free_fs_devices(fs_devices);
6558                 fs_devices = ERR_PTR(-EINVAL);
6559                 goto out;
6560         }
6561
6562         fs_devices->seed = root->fs_info->fs_devices->seed;
6563         root->fs_info->fs_devices->seed = fs_devices;
6564 out:
6565         return fs_devices;
6566 }
6567
6568 static int read_one_dev(struct btrfs_root *root,
6569                         struct extent_buffer *leaf,
6570                         struct btrfs_dev_item *dev_item)
6571 {
6572         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
6573         struct btrfs_device *device;
6574         u64 devid;
6575         int ret;
6576         u8 fs_uuid[BTRFS_UUID_SIZE];
6577         u8 dev_uuid[BTRFS_UUID_SIZE];
6578
6579         devid = btrfs_device_id(leaf, dev_item);
6580         read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
6581                            BTRFS_UUID_SIZE);
6582         read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
6583                            BTRFS_UUID_SIZE);
6584
6585         if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
6586                 fs_devices = open_seed_devices(root, fs_uuid);
6587                 if (IS_ERR(fs_devices))
6588                         return PTR_ERR(fs_devices);
6589         }
6590
6591         device = btrfs_find_device(root->fs_info, devid, dev_uuid, fs_uuid);
6592         if (!device) {
6593                 if (!btrfs_test_opt(root->fs_info, DEGRADED))
6594                         return -EIO;
6595
6596                 device = add_missing_dev(root, fs_devices, devid, dev_uuid);
6597                 if (!device)
6598                         return -ENOMEM;
6599                 btrfs_warn(root->fs_info, "devid %llu uuid %pU missing",
6600                                 devid, dev_uuid);
6601         } else {
6602                 if (!device->bdev && !btrfs_test_opt(root->fs_info, DEGRADED))
6603                         return -EIO;
6604
6605                 if(!device->bdev && !device->missing) {
6606                         /*
6607                          * this happens when a device that was properly setup
6608                          * in the device info lists suddenly goes bad.
6609                          * device->bdev is NULL, and so we have to set
6610                          * device->missing to one here
6611                          */
6612                         device->fs_devices->missing_devices++;
6613                         device->missing = 1;
6614                 }
6615
6616                 /* Move the device to its own fs_devices */
6617                 if (device->fs_devices != fs_devices) {
6618                         ASSERT(device->missing);
6619
6620                         list_move(&device->dev_list, &fs_devices->devices);
6621                         device->fs_devices->num_devices--;
6622                         fs_devices->num_devices++;
6623
6624                         device->fs_devices->missing_devices--;
6625                         fs_devices->missing_devices++;
6626
6627                         device->fs_devices = fs_devices;
6628                 }
6629         }
6630
6631         if (device->fs_devices != root->fs_info->fs_devices) {
6632                 BUG_ON(device->writeable);
6633                 if (device->generation !=
6634                     btrfs_device_generation(leaf, dev_item))
6635                         return -EINVAL;
6636         }
6637
6638         fill_device_from_item(leaf, dev_item, device);
6639         device->in_fs_metadata = 1;
6640         if (device->writeable && !device->is_tgtdev_for_dev_replace) {
6641                 device->fs_devices->total_rw_bytes += device->total_bytes;
6642                 spin_lock(&root->fs_info->free_chunk_lock);
6643                 root->fs_info->free_chunk_space += device->total_bytes -
6644                         device->bytes_used;
6645                 spin_unlock(&root->fs_info->free_chunk_lock);
6646         }
6647         ret = 0;
6648         return ret;
6649 }
6650
6651 int btrfs_read_sys_array(struct btrfs_root *root)
6652 {
6653         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
6654         struct extent_buffer *sb;
6655         struct btrfs_disk_key *disk_key;
6656         struct btrfs_chunk *chunk;
6657         u8 *array_ptr;
6658         unsigned long sb_array_offset;
6659         int ret = 0;
6660         u32 num_stripes;
6661         u32 array_size;
6662         u32 len = 0;
6663         u32 cur_offset;
6664         u64 type;
6665         struct btrfs_key key;
6666
6667         ASSERT(BTRFS_SUPER_INFO_SIZE <= root->nodesize);
6668         /*
6669          * This will create extent buffer of nodesize, superblock size is
6670          * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will
6671          * overallocate but we can keep it as-is, only the first page is used.
6672          */
6673         sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET);
6674         if (IS_ERR(sb))
6675                 return PTR_ERR(sb);
6676         set_extent_buffer_uptodate(sb);
6677         btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
6678         /*
6679          * The sb extent buffer is artificial and just used to read the system array.
6680          * set_extent_buffer_uptodate() call does not properly mark all it's
6681          * pages up-to-date when the page is larger: extent does not cover the
6682          * whole page and consequently check_page_uptodate does not find all
6683          * the page's extents up-to-date (the hole beyond sb),
6684          * write_extent_buffer then triggers a WARN_ON.
6685          *
6686          * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
6687          * but sb spans only this function. Add an explicit SetPageUptodate call
6688          * to silence the warning eg. on PowerPC 64.
6689          */
6690         if (PAGE_SIZE > BTRFS_SUPER_INFO_SIZE)
6691                 SetPageUptodate(sb->pages[0]);
6692
6693         write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
6694         array_size = btrfs_super_sys_array_size(super_copy);
6695
6696         array_ptr = super_copy->sys_chunk_array;
6697         sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array);
6698         cur_offset = 0;
6699
6700         while (cur_offset < array_size) {
6701                 disk_key = (struct btrfs_disk_key *)array_ptr;
6702                 len = sizeof(*disk_key);
6703                 if (cur_offset + len > array_size)
6704                         goto out_short_read;
6705
6706                 btrfs_disk_key_to_cpu(&key, disk_key);
6707
6708                 array_ptr += len;
6709                 sb_array_offset += len;
6710                 cur_offset += len;
6711
6712                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
6713                         chunk = (struct btrfs_chunk *)sb_array_offset;
6714                         /*
6715                          * At least one btrfs_chunk with one stripe must be
6716                          * present, exact stripe count check comes afterwards
6717                          */
6718                         len = btrfs_chunk_item_size(1);
6719                         if (cur_offset + len > array_size)
6720                                 goto out_short_read;
6721
6722                         num_stripes = btrfs_chunk_num_stripes(sb, chunk);
6723                         if (!num_stripes) {
6724                                 printk(KERN_ERR
6725             "BTRFS: invalid number of stripes %u in sys_array at offset %u\n",
6726                                         num_stripes, cur_offset);
6727                                 ret = -EIO;
6728                                 break;
6729                         }
6730
6731                         type = btrfs_chunk_type(sb, chunk);
6732                         if ((type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) {
6733                                 btrfs_err(root->fs_info,
6734                             "invalid chunk type %llu in sys_array at offset %u",
6735                                         type, cur_offset);
6736                                 ret = -EIO;
6737                                 break;
6738                         }
6739
6740                         len = btrfs_chunk_item_size(num_stripes);
6741                         if (cur_offset + len > array_size)
6742                                 goto out_short_read;
6743
6744                         ret = read_one_chunk(root, &key, sb, chunk);
6745                         if (ret)
6746                                 break;
6747                 } else {
6748                         printk(KERN_ERR
6749                 "BTRFS: unexpected item type %u in sys_array at offset %u\n",
6750                                 (u32)key.type, cur_offset);
6751                         ret = -EIO;
6752                         break;
6753                 }
6754                 array_ptr += len;
6755                 sb_array_offset += len;
6756                 cur_offset += len;
6757         }
6758         clear_extent_buffer_uptodate(sb);
6759         free_extent_buffer_stale(sb);
6760         return ret;
6761
6762 out_short_read:
6763         printk(KERN_ERR "BTRFS: sys_array too short to read %u bytes at offset %u\n",
6764                         len, cur_offset);
6765         clear_extent_buffer_uptodate(sb);
6766         free_extent_buffer_stale(sb);
6767         return -EIO;
6768 }
6769
6770 int btrfs_read_chunk_tree(struct btrfs_root *root)
6771 {
6772         struct btrfs_path *path;
6773         struct extent_buffer *leaf;
6774         struct btrfs_key key;
6775         struct btrfs_key found_key;
6776         int ret;
6777         int slot;
6778         u64 total_dev = 0;
6779
6780         root = root->fs_info->chunk_root;
6781
6782         path = btrfs_alloc_path();
6783         if (!path)
6784                 return -ENOMEM;
6785
6786         mutex_lock(&uuid_mutex);
6787         lock_chunks(root);
6788
6789         /*
6790          * Read all device items, and then all the chunk items. All
6791          * device items are found before any chunk item (their object id
6792          * is smaller than the lowest possible object id for a chunk
6793          * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
6794          */
6795         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
6796         key.offset = 0;
6797         key.type = 0;
6798         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6799         if (ret < 0)
6800                 goto error;
6801         while (1) {
6802                 leaf = path->nodes[0];
6803                 slot = path->slots[0];
6804                 if (slot >= btrfs_header_nritems(leaf)) {
6805                         ret = btrfs_next_leaf(root, path);
6806                         if (ret == 0)
6807                                 continue;
6808                         if (ret < 0)
6809                                 goto error;
6810                         break;
6811                 }
6812                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6813                 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
6814                         struct btrfs_dev_item *dev_item;
6815                         dev_item = btrfs_item_ptr(leaf, slot,
6816                                                   struct btrfs_dev_item);
6817                         ret = read_one_dev(root, leaf, dev_item);
6818                         if (ret)
6819                                 goto error;
6820                         total_dev++;
6821                 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
6822                         struct btrfs_chunk *chunk;
6823                         chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
6824                         ret = read_one_chunk(root, &found_key, leaf, chunk);
6825                         if (ret)
6826                                 goto error;
6827                 }
6828                 path->slots[0]++;
6829         }
6830
6831         /*
6832          * After loading chunk tree, we've got all device information,
6833          * do another round of validation checks.
6834          */
6835         if (total_dev != root->fs_info->fs_devices->total_devices) {
6836                 btrfs_err(root->fs_info,
6837            "super_num_devices %llu mismatch with num_devices %llu found here",
6838                           btrfs_super_num_devices(root->fs_info->super_copy),
6839                           total_dev);
6840                 ret = -EINVAL;
6841                 goto error;
6842         }
6843         if (btrfs_super_total_bytes(root->fs_info->super_copy) <
6844             root->fs_info->fs_devices->total_rw_bytes) {
6845                 btrfs_err(root->fs_info,
6846         "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu",
6847                           btrfs_super_total_bytes(root->fs_info->super_copy),
6848                           root->fs_info->fs_devices->total_rw_bytes);
6849                 ret = -EINVAL;
6850                 goto error;
6851         }
6852         ret = 0;
6853 error:
6854         unlock_chunks(root);
6855         mutex_unlock(&uuid_mutex);
6856
6857         btrfs_free_path(path);
6858         return ret;
6859 }
6860
6861 void btrfs_init_devices_late(struct btrfs_fs_info *fs_info)
6862 {
6863         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6864         struct btrfs_device *device;
6865
6866         while (fs_devices) {
6867                 mutex_lock(&fs_devices->device_list_mutex);
6868                 list_for_each_entry(device, &fs_devices->devices, dev_list)
6869                         device->dev_root = fs_info->dev_root;
6870                 mutex_unlock(&fs_devices->device_list_mutex);
6871
6872                 fs_devices = fs_devices->seed;
6873         }
6874 }
6875
6876 static void __btrfs_reset_dev_stats(struct btrfs_device *dev)
6877 {
6878         int i;
6879
6880         for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6881                 btrfs_dev_stat_reset(dev, i);
6882 }
6883
6884 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
6885 {
6886         struct btrfs_key key;
6887         struct btrfs_key found_key;
6888         struct btrfs_root *dev_root = fs_info->dev_root;
6889         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6890         struct extent_buffer *eb;
6891         int slot;
6892         int ret = 0;
6893         struct btrfs_device *device;
6894         struct btrfs_path *path = NULL;
6895         int i;
6896
6897         path = btrfs_alloc_path();
6898         if (!path) {
6899                 ret = -ENOMEM;
6900                 goto out;
6901         }
6902
6903         mutex_lock(&fs_devices->device_list_mutex);
6904         list_for_each_entry(device, &fs_devices->devices, dev_list) {
6905                 int item_size;
6906                 struct btrfs_dev_stats_item *ptr;
6907
6908                 key.objectid = BTRFS_DEV_STATS_OBJECTID;
6909                 key.type = BTRFS_PERSISTENT_ITEM_KEY;
6910                 key.offset = device->devid;
6911                 ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
6912                 if (ret) {
6913                         __btrfs_reset_dev_stats(device);
6914                         device->dev_stats_valid = 1;
6915                         btrfs_release_path(path);
6916                         continue;
6917                 }
6918                 slot = path->slots[0];
6919                 eb = path->nodes[0];
6920                 btrfs_item_key_to_cpu(eb, &found_key, slot);
6921                 item_size = btrfs_item_size_nr(eb, slot);
6922
6923                 ptr = btrfs_item_ptr(eb, slot,
6924                                      struct btrfs_dev_stats_item);
6925
6926                 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
6927                         if (item_size >= (1 + i) * sizeof(__le64))
6928                                 btrfs_dev_stat_set(device, i,
6929                                         btrfs_dev_stats_value(eb, ptr, i));
6930                         else
6931                                 btrfs_dev_stat_reset(device, i);
6932                 }
6933
6934                 device->dev_stats_valid = 1;
6935                 btrfs_dev_stat_print_on_load(device);
6936                 btrfs_release_path(path);
6937         }
6938         mutex_unlock(&fs_devices->device_list_mutex);
6939
6940 out:
6941         btrfs_free_path(path);
6942         return ret < 0 ? ret : 0;
6943 }
6944
6945 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
6946                                 struct btrfs_root *dev_root,
6947                                 struct btrfs_device *device)
6948 {
6949         struct btrfs_path *path;
6950         struct btrfs_key key;
6951         struct extent_buffer *eb;
6952         struct btrfs_dev_stats_item *ptr;
6953         int ret;
6954         int i;
6955
6956         key.objectid = BTRFS_DEV_STATS_OBJECTID;
6957         key.type = BTRFS_PERSISTENT_ITEM_KEY;
6958         key.offset = device->devid;
6959
6960         path = btrfs_alloc_path();
6961         BUG_ON(!path);
6962         ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
6963         if (ret < 0) {
6964                 btrfs_warn_in_rcu(dev_root->fs_info,
6965                         "error %d while searching for dev_stats item for device %s",
6966                               ret, rcu_str_deref(device->name));
6967                 goto out;
6968         }
6969
6970         if (ret == 0 &&
6971             btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
6972                 /* need to delete old one and insert a new one */
6973                 ret = btrfs_del_item(trans, dev_root, path);
6974                 if (ret != 0) {
6975                         btrfs_warn_in_rcu(dev_root->fs_info,
6976                                 "delete too small dev_stats item for device %s failed %d",
6977                                       rcu_str_deref(device->name), ret);
6978                         goto out;
6979                 }
6980                 ret = 1;
6981         }
6982
6983         if (ret == 1) {
6984                 /* need to insert a new item */
6985                 btrfs_release_path(path);
6986                 ret = btrfs_insert_empty_item(trans, dev_root, path,
6987                                               &key, sizeof(*ptr));
6988                 if (ret < 0) {
6989                         btrfs_warn_in_rcu(dev_root->fs_info,
6990                                 "insert dev_stats item for device %s failed %d",
6991                                 rcu_str_deref(device->name), ret);
6992                         goto out;
6993                 }
6994         }
6995
6996         eb = path->nodes[0];
6997         ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
6998         for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6999                 btrfs_set_dev_stats_value(eb, ptr, i,
7000                                           btrfs_dev_stat_read(device, i));
7001         btrfs_mark_buffer_dirty(eb);
7002
7003 out:
7004         btrfs_free_path(path);
7005         return ret;
7006 }
7007
7008 /*
7009  * called from commit_transaction. Writes all changed device stats to disk.
7010  */
7011 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans,
7012                         struct btrfs_fs_info *fs_info)
7013 {
7014         struct btrfs_root *dev_root = fs_info->dev_root;
7015         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7016         struct btrfs_device *device;
7017         int stats_cnt;
7018         int ret = 0;
7019
7020         mutex_lock(&fs_devices->device_list_mutex);
7021         list_for_each_entry(device, &fs_devices->devices, dev_list) {
7022                 if (!device->dev_stats_valid || !btrfs_dev_stats_dirty(device))
7023                         continue;
7024
7025                 stats_cnt = atomic_read(&device->dev_stats_ccnt);
7026                 ret = update_dev_stat_item(trans, dev_root, device);
7027                 if (!ret)
7028                         atomic_sub(stats_cnt, &device->dev_stats_ccnt);
7029         }
7030         mutex_unlock(&fs_devices->device_list_mutex);
7031
7032         return ret;
7033 }
7034
7035 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
7036 {
7037         btrfs_dev_stat_inc(dev, index);
7038         btrfs_dev_stat_print_on_error(dev);
7039 }
7040
7041 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
7042 {
7043         if (!dev->dev_stats_valid)
7044                 return;
7045         btrfs_err_rl_in_rcu(dev->dev_root->fs_info,
7046                 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7047                            rcu_str_deref(dev->name),
7048                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7049                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7050                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7051                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7052                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7053 }
7054
7055 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
7056 {
7057         int i;
7058
7059         for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7060                 if (btrfs_dev_stat_read(dev, i) != 0)
7061                         break;
7062         if (i == BTRFS_DEV_STAT_VALUES_MAX)
7063                 return; /* all values == 0, suppress message */
7064
7065         btrfs_info_in_rcu(dev->dev_root->fs_info,
7066                 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7067                rcu_str_deref(dev->name),
7068                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7069                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7070                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7071                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7072                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7073 }
7074
7075 int btrfs_get_dev_stats(struct btrfs_root *root,
7076                         struct btrfs_ioctl_get_dev_stats *stats)
7077 {
7078         struct btrfs_device *dev;
7079         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
7080         int i;
7081
7082         mutex_lock(&fs_devices->device_list_mutex);
7083         dev = btrfs_find_device(root->fs_info, stats->devid, NULL, NULL);
7084         mutex_unlock(&fs_devices->device_list_mutex);
7085
7086         if (!dev) {
7087                 btrfs_warn(root->fs_info, "get dev_stats failed, device not found");
7088                 return -ENODEV;
7089         } else if (!dev->dev_stats_valid) {
7090                 btrfs_warn(root->fs_info, "get dev_stats failed, not yet valid");
7091                 return -ENODEV;
7092         } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
7093                 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7094                         if (stats->nr_items > i)
7095                                 stats->values[i] =
7096                                         btrfs_dev_stat_read_and_reset(dev, i);
7097                         else
7098                                 btrfs_dev_stat_reset(dev, i);
7099                 }
7100         } else {
7101                 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7102                         if (stats->nr_items > i)
7103                                 stats->values[i] = btrfs_dev_stat_read(dev, i);
7104         }
7105         if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
7106                 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
7107         return 0;
7108 }
7109
7110 void btrfs_scratch_superblocks(struct block_device *bdev, char *device_path)
7111 {
7112         struct buffer_head *bh;
7113         struct btrfs_super_block *disk_super;
7114         int copy_num;
7115
7116         if (!bdev)
7117                 return;
7118
7119         for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX;
7120                 copy_num++) {
7121
7122                 if (btrfs_read_dev_one_super(bdev, copy_num, &bh))
7123                         continue;
7124
7125                 disk_super = (struct btrfs_super_block *)bh->b_data;
7126
7127                 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
7128                 set_buffer_dirty(bh);
7129                 sync_dirty_buffer(bh);
7130                 brelse(bh);
7131         }
7132
7133         /* Notify udev that device has changed */
7134         btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
7135
7136         /* Update ctime/mtime for device path for libblkid */
7137         update_dev_time(device_path);
7138 }
7139
7140 /*
7141  * Update the size of all devices, which is used for writing out the
7142  * super blocks.
7143  */
7144 void btrfs_update_commit_device_size(struct btrfs_fs_info *fs_info)
7145 {
7146         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7147         struct btrfs_device *curr, *next;
7148
7149         if (list_empty(&fs_devices->resized_devices))
7150                 return;
7151
7152         mutex_lock(&fs_devices->device_list_mutex);
7153         lock_chunks(fs_info->dev_root);
7154         list_for_each_entry_safe(curr, next, &fs_devices->resized_devices,
7155                                  resized_list) {
7156                 list_del_init(&curr->resized_list);
7157                 curr->commit_total_bytes = curr->disk_total_bytes;
7158         }
7159         unlock_chunks(fs_info->dev_root);
7160         mutex_unlock(&fs_devices->device_list_mutex);
7161 }
7162
7163 /* Must be invoked during the transaction commit */
7164 void btrfs_update_commit_device_bytes_used(struct btrfs_root *root,
7165                                         struct btrfs_transaction *transaction)
7166 {
7167         struct extent_map *em;
7168         struct map_lookup *map;
7169         struct btrfs_device *dev;
7170         int i;
7171
7172         if (list_empty(&transaction->pending_chunks))
7173                 return;
7174
7175         /* In order to kick the device replace finish process */
7176         lock_chunks(root);
7177         list_for_each_entry(em, &transaction->pending_chunks, list) {
7178                 map = em->map_lookup;
7179
7180                 for (i = 0; i < map->num_stripes; i++) {
7181                         dev = map->stripes[i].dev;
7182                         dev->commit_bytes_used = dev->bytes_used;
7183                 }
7184         }
7185         unlock_chunks(root);
7186 }
7187
7188 void btrfs_set_fs_info_ptr(struct btrfs_fs_info *fs_info)
7189 {
7190         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7191         while (fs_devices) {
7192                 fs_devices->fs_info = fs_info;
7193                 fs_devices = fs_devices->seed;
7194         }
7195 }
7196
7197 void btrfs_reset_fs_info_ptr(struct btrfs_fs_info *fs_info)
7198 {
7199         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7200         while (fs_devices) {
7201                 fs_devices->fs_info = NULL;
7202                 fs_devices = fs_devices->seed;
7203         }
7204 }