fs/btrfs/volumes.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18 #include <linux/sched.h>
  19 #include <linux/bio.h>
  20 #include <linux/slab.h>
  21 #include <linux/buffer_head.h>
  22 #include <linux/blkdev.h>
  23 #include <linux/iocontext.h>
  24 #include <linux/capability.h>
  25 #include <linux/ratelimit.h>
  26 #include <linux/kthread.h>
  27 #include <linux/raid/pq.h>
  28 #include <linux/semaphore.h>
  29 #include <linux/uuid.h>
  30 #include <asm/div64.h>
  31 #include "ctree.h"
  32 #include "extent_map.h"
  33 #include "disk-io.h"
  34 #include "transaction.h"
  35 #include "print-tree.h"
  36 #include "volumes.h"
  37 #include "raid56.h"
  38 #include "async-thread.h"
  39 #include "check-integrity.h"
  40 #include "rcu-string.h"
  41 #include "math.h"
  42 #include "dev-replace.h"
  43 #include "sysfs.h"
  44
  45 const struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = {
  46         [BTRFS_RAID_RAID10] = {
  47                 .sub_stripes    = 2,
  48                 .dev_stripes    = 1,
  49                 .devs_max       = 0,    /* 0 == as many as possible */
  50                 .devs_min       = 4,
  51                 .tolerated_failures = 1,
  52                 .devs_increment = 2,
  53                 .ncopies        = 2,
  54         },
  55         [BTRFS_RAID_RAID1] = {
  56                 .sub_stripes    = 1,
  57                 .dev_stripes    = 1,
  58                 .devs_max       = 2,
  59                 .devs_min       = 2,
  60                 .tolerated_failures = 1,
  61                 .devs_increment = 2,
  62                 .ncopies        = 2,
  63         },
  64         [BTRFS_RAID_DUP] = {
  65                 .sub_stripes    = 1,
  66                 .dev_stripes    = 2,
  67                 .devs_max       = 1,
  68                 .devs_min       = 1,
  69                 .tolerated_failures = 0,
  70                 .devs_increment = 1,
  71                 .ncopies        = 2,
  72         },
  73         [BTRFS_RAID_RAID0] = {
  74                 .sub_stripes    = 1,
  75                 .dev_stripes    = 1,
  76                 .devs_max       = 0,
  77                 .devs_min       = 2,
  78                 .tolerated_failures = 0,
  79                 .devs_increment = 1,
  80                 .ncopies        = 1,
  81         },
  82         [BTRFS_RAID_SINGLE] = {
  83                 .sub_stripes    = 1,
  84                 .dev_stripes    = 1,
  85                 .devs_max       = 1,
  86                 .devs_min       = 1,
  87                 .tolerated_failures = 0,
  88                 .devs_increment = 1,
  89                 .ncopies        = 1,
  90         },
  91         [BTRFS_RAID_RAID5] = {
  92                 .sub_stripes    = 1,
  93                 .dev_stripes    = 1,
  94                 .devs_max       = 0,
  95                 .devs_min       = 2,
  96                 .tolerated_failures = 1,
  97                 .devs_increment = 1,
  98                 .ncopies        = 2,
  99         },
 100         [BTRFS_RAID_RAID6] = {
 101                 .sub_stripes    = 1,
 102                 .dev_stripes    = 1,
 103                 .devs_max       = 0,
 104                 .devs_min       = 3,
 105                 .tolerated_failures = 2,
 106                 .devs_increment = 1,
 107                 .ncopies        = 3,
 108         },
 109 };
 110
 111 const u64 btrfs_raid_group[BTRFS_NR_RAID_TYPES] = {
 112         [BTRFS_RAID_RAID10] = BTRFS_BLOCK_GROUP_RAID10,
 113         [BTRFS_RAID_RAID1]  = BTRFS_BLOCK_GROUP_RAID1,
 114         [BTRFS_RAID_DUP]    = BTRFS_BLOCK_GROUP_DUP,
 115         [BTRFS_RAID_RAID0]  = BTRFS_BLOCK_GROUP_RAID0,
 116         [BTRFS_RAID_SINGLE] = 0,
 117         [BTRFS_RAID_RAID5]  = BTRFS_BLOCK_GROUP_RAID5,
 118         [BTRFS_RAID_RAID6]  = BTRFS_BLOCK_GROUP_RAID6,
 119 };
 120
 121 /*
 122  * Table to convert BTRFS_RAID_* to the error code if minimum number of devices
 123  * condition is not met. Zero means there's no corresponding
 124  * BTRFS_ERROR_DEV_*_NOT_MET value.
 125  */
 126 const int btrfs_raid_mindev_error[BTRFS_NR_RAID_TYPES] = {
 127         [BTRFS_RAID_RAID10] = BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET,
 128         [BTRFS_RAID_RAID1]  = BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET,
 129         [BTRFS_RAID_DUP]    = 0,
 130         [BTRFS_RAID_RAID0]  = 0,
 131         [BTRFS_RAID_SINGLE] = 0,
 132         [BTRFS_RAID_RAID5]  = BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET,
 133         [BTRFS_RAID_RAID6]  = BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET,
 134 };
 135
 136 static int init_first_rw_device(struct btrfs_trans_handle *trans,
 137                                 struct btrfs_fs_info *fs_info);
 138 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info);
 139 static void __btrfs_reset_dev_stats(struct btrfs_device *dev);
 140 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev);
 141 static void btrfs_dev_stat_print_on_load(struct btrfs_device *device);
 142 static int __btrfs_map_block(struct btrfs_fs_info *fs_info,
 143                              enum btrfs_map_op op,
 144                              u64 logical, u64 *length,
 145                              struct btrfs_bio **bbio_ret,
 146                              int mirror_num, int need_raid_map);
 147
 148 DEFINE_MUTEX(uuid_mutex);
 149 static LIST_HEAD(fs_uuids);
 150 struct list_head *btrfs_get_fs_uuids(void)
 151 {
 152         return &fs_uuids;
 153 }
 154
 155 static struct btrfs_fs_devices *__alloc_fs_devices(void)
 156 {
 157         struct btrfs_fs_devices *fs_devs;
 158
 159         fs_devs = kzalloc(sizeof(*fs_devs), GFP_KERNEL);
 160         if (!fs_devs)
 161                 return ERR_PTR(-ENOMEM);
 162
 163         mutex_init(&fs_devs->device_list_mutex);
 164
 165         INIT_LIST_HEAD(&fs_devs->devices);
 166         INIT_LIST_HEAD(&fs_devs->resized_devices);
 167         INIT_LIST_HEAD(&fs_devs->alloc_list);
 168         INIT_LIST_HEAD(&fs_devs->list);
 169
 170         return fs_devs;
 171 }
 172
 173 /**
 174  * alloc_fs_devices - allocate struct btrfs_fs_devices
 175  * @fsid:       a pointer to UUID for this FS.  If NULL a new UUID is
 176  *              generated.
 177  *
 178  * Return: a pointer to a new &struct btrfs_fs_devices on success;
 179  * ERR_PTR() on error.  Returned struct is not linked onto any lists and
 180  * can be destroyed with kfree() right away.
 181  */
 182 static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid)
 183 {
 184         struct btrfs_fs_devices *fs_devs;
 185
 186         fs_devs = __alloc_fs_devices();
 187         if (IS_ERR(fs_devs))
 188                 return fs_devs;
 189
 190         if (fsid)
 191                 memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE);
 192         else
 193                 generate_random_uuid(fs_devs->fsid);
 194
 195         return fs_devs;
 196 }
 197
 198 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
 199 {
 200         struct btrfs_device *device;
 201         WARN_ON(fs_devices->opened);
 202         while (!list_empty(&fs_devices->devices)) {
 203                 device = list_entry(fs_devices->devices.next,
 204                                     struct btrfs_device, dev_list);
 205                 list_del(&device->dev_list);
 206                 rcu_string_free(device->name);
 207                 kfree(device);
 208         }
 209         kfree(fs_devices);
 210 }
 211
 212 static void btrfs_kobject_uevent(struct block_device *bdev,
 213                                  enum kobject_action action)
 214 {
 215         int ret;
 216
 217         ret = kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, action);
 218         if (ret)
 219                 pr_warn("BTRFS: Sending event '%d' to kobject: '%s' (%p): failed\n",
 220                         action,
 221                         kobject_name(&disk_to_dev(bdev->bd_disk)->kobj),
 222                         &disk_to_dev(bdev->bd_disk)->kobj);
 223 }
 224
 225 void btrfs_cleanup_fs_uuids(void)
 226 {
 227         struct btrfs_fs_devices *fs_devices;
 228
 229         while (!list_empty(&fs_uuids)) {
 230                 fs_devices = list_entry(fs_uuids.next,
 231                                         struct btrfs_fs_devices, list);
 232                 list_del(&fs_devices->list);
 233                 free_fs_devices(fs_devices);
 234         }
 235 }
 236
 237 static struct btrfs_device *__alloc_device(void)
 238 {
 239         struct btrfs_device *dev;
 240
 241         dev = kzalloc(sizeof(*dev), GFP_KERNEL);
 242         if (!dev)
 243                 return ERR_PTR(-ENOMEM);
 244
 245         /*
 246          * Preallocate a bio that's always going to be used for flushing device
 247          * barriers and matches the device lifespan
 248          */
 249         dev->flush_bio = bio_alloc_bioset(GFP_KERNEL, 0, NULL);
 250         if (!dev->flush_bio) {
 251                 kfree(dev);
 252                 return ERR_PTR(-ENOMEM);
 253         }
 254         bio_get(dev->flush_bio);
 255
 256         INIT_LIST_HEAD(&dev->dev_list);
 257         INIT_LIST_HEAD(&dev->dev_alloc_list);
 258         INIT_LIST_HEAD(&dev->resized_list);
 259
 260         spin_lock_init(&dev->io_lock);
 261
 262         spin_lock_init(&dev->reada_lock);
 263         atomic_set(&dev->reada_in_flight, 0);
 264         atomic_set(&dev->dev_stats_ccnt, 0);
 265         btrfs_device_data_ordered_init(dev);
 266         INIT_RADIX_TREE(&dev->reada_zones, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
 267         INIT_RADIX_TREE(&dev->reada_extents, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
 268
 269         return dev;
 270 }
 271
 272 static noinline struct btrfs_device *__find_device(struct list_head *head,
 273                                                    u64 devid, u8 *uuid)
 274 {
 275         struct btrfs_device *dev;
 276
 277         list_for_each_entry(dev, head, dev_list) {
 278                 if (dev->devid == devid &&
 279                     (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
 280                         return dev;
 281                 }
 282         }
 283         return NULL;
 284 }
 285
 286 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
 287 {
 288         struct btrfs_fs_devices *fs_devices;
 289
 290         list_for_each_entry(fs_devices, &fs_uuids, list) {
 291                 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
 292                         return fs_devices;
 293         }
 294         return NULL;
 295 }
 296
 297 static int
 298 btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder,
 299                       int flush, struct block_device **bdev,
 300                       struct buffer_head **bh)
 301 {
 302         int ret;
 303
 304         *bdev = blkdev_get_by_path(device_path, flags, holder);
 305
 306         if (IS_ERR(*bdev)) {
 307                 ret = PTR_ERR(*bdev);
 308                 goto error;
 309         }
 310
 311         if (flush)
 312                 filemap_write_and_wait((*bdev)->bd_inode->i_mapping);
 313         ret = set_blocksize(*bdev, 4096);
 314         if (ret) {
 315                 blkdev_put(*bdev, flags);
 316                 goto error;
 317         }
 318         invalidate_bdev(*bdev);
 319         *bh = btrfs_read_dev_super(*bdev);
 320         if (IS_ERR(*bh)) {
 321                 ret = PTR_ERR(*bh);
 322                 blkdev_put(*bdev, flags);
 323                 goto error;
 324         }
 325
 326         return 0;
 327
 328 error:
 329         *bdev = NULL;
 330         *bh = NULL;
 331         return ret;
 332 }
 333
 334 static void requeue_list(struct btrfs_pending_bios *pending_bios,
 335                         struct bio *head, struct bio *tail)
 336 {
 337
 338         struct bio *old_head;
 339
 340         old_head = pending_bios->head;
 341         pending_bios->head = head;
 342         if (pending_bios->tail)
 343                 tail->bi_next = old_head;
 344         else
 345                 pending_bios->tail = tail;
 346 }
 347
 348 /*
 349  * we try to collect pending bios for a device so we don't get a large
 350  * number of procs sending bios down to the same device.  This greatly
 351  * improves the schedulers ability to collect and merge the bios.
 352  *
 353  * But, it also turns into a long list of bios to process and that is sure
 354  * to eventually make the worker thread block.  The solution here is to
 355  * make some progress and then put this work struct back at the end of
 356  * the list if the block device is congested.  This way, multiple devices
 357  * can make progress from a single worker thread.
 358  */
 359 static noinline void run_scheduled_bios(struct btrfs_device *device)
 360 {
 361         struct btrfs_fs_info *fs_info = device->fs_info;
 362         struct bio *pending;
 363         struct backing_dev_info *bdi;
 364         struct btrfs_pending_bios *pending_bios;
 365         struct bio *tail;
 366         struct bio *cur;
 367         int again = 0;
 368         unsigned long num_run;
 369         unsigned long batch_run = 0;
 370         unsigned long limit;
 371         unsigned long last_waited = 0;
 372         int force_reg = 0;
 373         int sync_pending = 0;
 374         struct blk_plug plug;
 375
 376         /*
 377          * this function runs all the bios we've collected for
 378          * a particular device.  We don't want to wander off to
 379          * another device without first sending all of these down.
 380          * So, setup a plug here and finish it off before we return
 381          */
 382         blk_start_plug(&plug);
 383
 384         bdi = device->bdev->bd_bdi;
 385         limit = btrfs_async_submit_limit(fs_info);
 386         limit = limit * 2 / 3;
 387
 388 loop:
 389         spin_lock(&device->io_lock);
 390
 391 loop_lock:
 392         num_run = 0;
 393
 394         /* take all the bios off the list at once and process them
 395          * later on (without the lock held).  But, remember the
 396          * tail and other pointers so the bios can be properly reinserted
 397          * into the list if we hit congestion
 398          */
 399         if (!force_reg && device->pending_sync_bios.head) {
 400                 pending_bios = &device->pending_sync_bios;
 401                 force_reg = 1;
 402         } else {
 403                 pending_bios = &device->pending_bios;
 404                 force_reg = 0;
 405         }
 406
 407         pending = pending_bios->head;
 408         tail = pending_bios->tail;
 409         WARN_ON(pending && !tail);
 410
 411         /*
 412          * if pending was null this time around, no bios need processing
 413          * at all and we can stop.  Otherwise it'll loop back up again
 414          * and do an additional check so no bios are missed.
 415          *
 416          * device->running_pending is used to synchronize with the
 417          * schedule_bio code.
 418          */
 419         if (device->pending_sync_bios.head == NULL &&
 420             device->pending_bios.head == NULL) {
 421                 again = 0;
 422                 device->running_pending = 0;
 423         } else {
 424                 again = 1;
 425                 device->running_pending = 1;
 426         }
 427
 428         pending_bios->head = NULL;
 429         pending_bios->tail = NULL;
 430
 431         spin_unlock(&device->io_lock);
 432
 433         while (pending) {
 434
 435                 rmb();
 436                 /* we want to work on both lists, but do more bios on the
 437                  * sync list than the regular list
 438                  */
 439                 if ((num_run > 32 &&
 440                     pending_bios != &device->pending_sync_bios &&
 441                     device->pending_sync_bios.head) ||
 442                    (num_run > 64 && pending_bios == &device->pending_sync_bios &&
 443                     device->pending_bios.head)) {
 444                         spin_lock(&device->io_lock);
 445                         requeue_list(pending_bios, pending, tail);
 446                         goto loop_lock;
 447                 }
 448
 449                 cur = pending;
 450                 pending = pending->bi_next;
 451                 cur->bi_next = NULL;
 452
 453                 /*
 454                  * atomic_dec_return implies a barrier for waitqueue_active
 455                  */
 456                 if (atomic_dec_return(&fs_info->nr_async_bios) < limit &&
 457                     waitqueue_active(&fs_info->async_submit_wait))
 458                         wake_up(&fs_info->async_submit_wait);
 459
 460                 BUG_ON(atomic_read(&cur->__bi_cnt) == 0);
 461
 462                 /*
 463                  * if we're doing the sync list, record that our
 464                  * plug has some sync requests on it
 465                  *
 466                  * If we're doing the regular list and there are
 467                  * sync requests sitting around, unplug before
 468                  * we add more
 469                  */
 470                 if (pending_bios == &device->pending_sync_bios) {
 471                         sync_pending = 1;
 472                 } else if (sync_pending) {
 473                         blk_finish_plug(&plug);
 474                         blk_start_plug(&plug);
 475                         sync_pending = 0;
 476                 }
 477
 478                 btrfsic_submit_bio(cur);
 479                 num_run++;
 480                 batch_run++;
 481
 482                 cond_resched();
 483
 484                 /*
 485                  * we made progress, there is more work to do and the bdi
 486                  * is now congested.  Back off and let other work structs
 487                  * run instead
 488                  */
 489                 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
 490                     fs_info->fs_devices->open_devices > 1) {
 491                         struct io_context *ioc;
 492
 493                         ioc = current->io_context;
 494
 495                         /*
 496                          * the main goal here is that we don't want to
 497                          * block if we're going to be able to submit
 498                          * more requests without blocking.
 499                          *
 500                          * This code does two great things, it pokes into
 501                          * the elevator code from a filesystem _and_
 502                          * it makes assumptions about how batching works.
 503                          */
 504                         if (ioc && ioc->nr_batch_requests > 0 &&
 505                             time_before(jiffies, ioc->last_waited + HZ/50UL) &&
 506                             (last_waited == 0 ||
 507                              ioc->last_waited == last_waited)) {
 508                                 /*
 509                                  * we want to go through our batch of
 510                                  * requests and stop.  So, we copy out
 511                                  * the ioc->last_waited time and test
 512                                  * against it before looping
 513                                  */
 514                                 last_waited = ioc->last_waited;
 515                                 cond_resched();
 516                                 continue;
 517                         }
 518                         spin_lock(&device->io_lock);
 519                         requeue_list(pending_bios, pending, tail);
 520                         device->running_pending = 1;
 521
 522                         spin_unlock(&device->io_lock);
 523                         btrfs_queue_work(fs_info->submit_workers,
 524                                          &device->work);
 525                         goto done;
 526                 }
 527                 /* unplug every 64 requests just for good measure */
 528                 if (batch_run % 64 == 0) {
 529                         blk_finish_plug(&plug);
 530                         blk_start_plug(&plug);
 531                         sync_pending = 0;
 532                 }
 533         }
 534
 535         cond_resched();
 536         if (again)
 537                 goto loop;
 538
 539         spin_lock(&device->io_lock);
 540         if (device->pending_bios.head || device->pending_sync_bios.head)
 541                 goto loop_lock;
 542         spin_unlock(&device->io_lock);
 543
 544 done:
 545         blk_finish_plug(&plug);
 546 }
 547
 548 static void pending_bios_fn(struct btrfs_work *work)
 549 {
 550         struct btrfs_device *device;
 551
 552         device = container_of(work, struct btrfs_device, work);
 553         run_scheduled_bios(device);
 554 }
 555
 556
 557 void btrfs_free_stale_device(struct btrfs_device *cur_dev)
 558 {
 559         struct btrfs_fs_devices *fs_devs;
 560         struct btrfs_device *dev;
 561
 562         if (!cur_dev->name)
 563                 return;
 564
 565         list_for_each_entry(fs_devs, &fs_uuids, list) {
 566                 int del = 1;
 567
 568                 if (fs_devs->opened)
 569                         continue;
 570                 if (fs_devs->seeding)
 571                         continue;
 572
 573                 list_for_each_entry(dev, &fs_devs->devices, dev_list) {
 574
 575                         if (dev == cur_dev)
 576                                 continue;
 577                         if (!dev->name)
 578                                 continue;
 579
 580                         /*
 581                          * Todo: This won't be enough. What if the same device
 582                          * comes back (with new uuid and) with its mapper path?
 583                          * But for now, this does help as mostly an admin will
 584                          * either use mapper or non mapper path throughout.
 585                          */
 586                         rcu_read_lock();
 587                         del = strcmp(rcu_str_deref(dev->name),
 588                                                 rcu_str_deref(cur_dev->name));
 589                         rcu_read_unlock();
 590                         if (!del)
 591                                 break;
 592                 }
 593
 594                 if (!del) {
 595                         /* delete the stale device */
 596                         if (fs_devs->num_devices == 1) {
 597                                 btrfs_sysfs_remove_fsid(fs_devs);
 598                                 list_del(&fs_devs->list);
 599                                 free_fs_devices(fs_devs);
 600                         } else {
 601                                 fs_devs->num_devices--;
 602                                 list_del(&dev->dev_list);
 603                                 rcu_string_free(dev->name);
 604                                 kfree(dev);
 605                         }
 606                         break;
 607                 }
 608         }
 609 }
 610
 611 /*
 612  * Add new device to list of registered devices
 613  *
 614  * Returns:
 615  * 1   - first time device is seen
 616  * 0   - device already known
 617  * < 0 - error
 618  */
 619 static noinline int device_list_add(const char *path,
 620                            struct btrfs_super_block *disk_super,
 621                            u64 devid, struct btrfs_fs_devices **fs_devices_ret)
 622 {
 623         struct btrfs_device *device;
 624         struct btrfs_fs_devices *fs_devices;
 625         struct rcu_string *name;
 626         int ret = 0;
 627         u64 found_transid = btrfs_super_generation(disk_super);
 628
 629         fs_devices = find_fsid(disk_super->fsid);
 630         if (!fs_devices) {
 631                 fs_devices = alloc_fs_devices(disk_super->fsid);
 632                 if (IS_ERR(fs_devices))
 633                         return PTR_ERR(fs_devices);
 634
 635                 list_add(&fs_devices->list, &fs_uuids);
 636
 637                 device = NULL;
 638         } else {
 639                 device = __find_device(&fs_devices->devices, devid,
 640                                        disk_super->dev_item.uuid);
 641         }
 642
 643         if (!device) {
 644                 if (fs_devices->opened)
 645                         return -EBUSY;
 646
 647                 device = btrfs_alloc_device(NULL, &devid,
 648                                             disk_super->dev_item.uuid);
 649                 if (IS_ERR(device)) {
 650                         /* we can safely leave the fs_devices entry around */
 651                         return PTR_ERR(device);
 652                 }
 653
 654                 name = rcu_string_strdup(path, GFP_NOFS);
 655                 if (!name) {
 656                         kfree(device);
 657                         return -ENOMEM;
 658                 }
 659                 rcu_assign_pointer(device->name, name);
 660
 661                 mutex_lock(&fs_devices->device_list_mutex);
 662                 list_add_rcu(&device->dev_list, &fs_devices->devices);
 663                 fs_devices->num_devices++;
 664                 mutex_unlock(&fs_devices->device_list_mutex);
 665
 666                 ret = 1;
 667                 device->fs_devices = fs_devices;
 668         } else if (!device->name || strcmp(device->name->str, path)) {
 669                 /*
 670                  * When FS is already mounted.
 671                  * 1. If you are here and if the device->name is NULL that
 672                  *    means this device was missing at time of FS mount.
 673                  * 2. If you are here and if the device->name is different
 674                  *    from 'path' that means either
 675                  *      a. The same device disappeared and reappeared with
 676                  *         different name. or
 677                  *      b. The missing-disk-which-was-replaced, has
 678                  *         reappeared now.
 679                  *
 680                  * We must allow 1 and 2a above. But 2b would be a spurious
 681                  * and unintentional.
 682                  *
 683                  * Further in case of 1 and 2a above, the disk at 'path'
 684                  * would have missed some transaction when it was away and
 685                  * in case of 2a the stale bdev has to be updated as well.
 686                  * 2b must not be allowed at all time.
 687                  */
 688
 689                 /*
 690                  * For now, we do allow update to btrfs_fs_device through the
 691                  * btrfs dev scan cli after FS has been mounted.  We're still
 692                  * tracking a problem where systems fail mount by subvolume id
 693                  * when we reject replacement on a mounted FS.
 694                  */
 695                 if (!fs_devices->opened && found_transid < device->generation) {
 696                         /*
 697                          * That is if the FS is _not_ mounted and if you
 698                          * are here, that means there is more than one
 699                          * disk with same uuid and devid.We keep the one
 700                          * with larger generation number or the last-in if
 701                          * generation are equal.
 702                          */
 703                         return -EEXIST;
 704                 }
 705
 706                 name = rcu_string_strdup(path, GFP_NOFS);
 707                 if (!name)
 708                         return -ENOMEM;
 709                 rcu_string_free(device->name);
 710                 rcu_assign_pointer(device->name, name);
 711                 if (device->missing) {
 712                         fs_devices->missing_devices--;
 713                         device->missing = 0;
 714                 }
 715         }
 716
 717         /*
 718          * Unmount does not free the btrfs_device struct but would zero
 719          * generation along with most of the other members. So just update
 720          * it back. We need it to pick the disk with largest generation
 721          * (as above).
 722          */
 723         if (!fs_devices->opened)
 724                 device->generation = found_transid;
 725
 726         /*
 727          * if there is new btrfs on an already registered device,
 728          * then remove the stale device entry.
 729          */
 730         if (ret > 0)
 731                 btrfs_free_stale_device(device);
 732
 733         *fs_devices_ret = fs_devices;
 734
 735         return ret;
 736 }
 737
 738 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
 739 {
 740         struct btrfs_fs_devices *fs_devices;
 741         struct btrfs_device *device;
 742         struct btrfs_device *orig_dev;
 743
 744         fs_devices = alloc_fs_devices(orig->fsid);
 745         if (IS_ERR(fs_devices))
 746                 return fs_devices;
 747
 748         mutex_lock(&orig->device_list_mutex);
 749         fs_devices->total_devices = orig->total_devices;
 750
 751         /* We have held the volume lock, it is safe to get the devices. */
 752         list_for_each_entry(orig_dev, &orig->devices, dev_list) {
 753                 struct rcu_string *name;
 754
 755                 device = btrfs_alloc_device(NULL, &orig_dev->devid,
 756                                             orig_dev->uuid);
 757                 if (IS_ERR(device))
 758                         goto error;
 759
 760                 /*
 761                  * This is ok to do without rcu read locked because we hold the
 762                  * uuid mutex so nothing we touch in here is going to disappear.
 763                  */
 764                 if (orig_dev->name) {
 765                         name = rcu_string_strdup(orig_dev->name->str,
 766                                         GFP_KERNEL);
 767                         if (!name) {
 768                                 kfree(device);
 769                                 goto error;
 770                         }
 771                         rcu_assign_pointer(device->name, name);
 772                 }
 773
 774                 list_add(&device->dev_list, &fs_devices->devices);
 775                 device->fs_devices = fs_devices;
 776                 fs_devices->num_devices++;
 777         }
 778         mutex_unlock(&orig->device_list_mutex);
 779         return fs_devices;
 780 error:
 781         mutex_unlock(&orig->device_list_mutex);
 782         free_fs_devices(fs_devices);
 783         return ERR_PTR(-ENOMEM);
 784 }
 785
 786 void btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices, int step)
 787 {
 788         struct btrfs_device *device, *next;
 789         struct btrfs_device *latest_dev = NULL;
 790
 791         mutex_lock(&uuid_mutex);
 792 again:
 793         /* This is the initialized path, it is safe to release the devices. */
 794         list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
 795                 if (device->in_fs_metadata) {
 796                         if (!device->is_tgtdev_for_dev_replace &&
 797                             (!latest_dev ||
 798                              device->generation > latest_dev->generation)) {
 799                                 latest_dev = device;
 800                         }
 801                         continue;
 802                 }
 803
 804                 if (device->devid == BTRFS_DEV_REPLACE_DEVID) {
 805                         /*
 806                          * In the first step, keep the device which has
 807                          * the correct fsid and the devid that is used
 808                          * for the dev_replace procedure.
 809                          * In the second step, the dev_replace state is
 810                          * read from the device tree and it is known
 811                          * whether the procedure is really active or
 812                          * not, which means whether this device is
 813                          * used or whether it should be removed.
 814                          */
 815                         if (step == 0 || device->is_tgtdev_for_dev_replace) {
 816                                 continue;
 817                         }
 818                 }
 819                 if (device->bdev) {
 820                         blkdev_put(device->bdev, device->mode);
 821                         device->bdev = NULL;
 822                         fs_devices->open_devices--;
 823                 }
 824                 if (device->writeable) {
 825                         list_del_init(&device->dev_alloc_list);
 826                         device->writeable = 0;
 827                         if (!device->is_tgtdev_for_dev_replace)
 828                                 fs_devices->rw_devices--;
 829                 }
 830                 list_del_init(&device->dev_list);
 831                 fs_devices->num_devices--;
 832                 rcu_string_free(device->name);
 833                 kfree(device);
 834         }
 835
 836         if (fs_devices->seed) {
 837                 fs_devices = fs_devices->seed;
 838                 goto again;
 839         }
 840
 841         fs_devices->latest_bdev = latest_dev->bdev;
 842
 843         mutex_unlock(&uuid_mutex);
 844 }
 845
 846 static void __free_device(struct work_struct *work)
 847 {
 848         struct btrfs_device *device;
 849
 850         device = container_of(work, struct btrfs_device, rcu_work);
 851         rcu_string_free(device->name);
 852         bio_put(device->flush_bio);
 853         kfree(device);
 854 }
 855
 856 static void free_device(struct rcu_head *head)
 857 {
 858         struct btrfs_device *device;
 859
 860         device = container_of(head, struct btrfs_device, rcu);
 861
 862         INIT_WORK(&device->rcu_work, __free_device);
 863         schedule_work(&device->rcu_work);
 864 }
 865
 866 static void btrfs_close_bdev(struct btrfs_device *device)
 867 {
 868         if (device->bdev && device->writeable) {
 869                 sync_blockdev(device->bdev);
 870                 invalidate_bdev(device->bdev);
 871         }
 872
 873         if (device->bdev)
 874                 blkdev_put(device->bdev, device->mode);
 875 }
 876
 877 static void btrfs_prepare_close_one_device(struct btrfs_device *device)
 878 {
 879         struct btrfs_fs_devices *fs_devices = device->fs_devices;
 880         struct btrfs_device *new_device;
 881         struct rcu_string *name;
 882
 883         if (device->bdev)
 884                 fs_devices->open_devices--;
 885
 886         if (device->writeable &&
 887             device->devid != BTRFS_DEV_REPLACE_DEVID) {
 888                 list_del_init(&device->dev_alloc_list);
 889                 fs_devices->rw_devices--;
 890         }
 891
 892         if (device->missing)
 893                 fs_devices->missing_devices--;
 894
 895         new_device = btrfs_alloc_device(NULL, &device->devid,
 896                                         device->uuid);
 897         BUG_ON(IS_ERR(new_device)); /* -ENOMEM */
 898
 899         /* Safe because we are under uuid_mutex */
 900         if (device->name) {
 901                 name = rcu_string_strdup(device->name->str, GFP_NOFS);
 902                 BUG_ON(!name); /* -ENOMEM */
 903                 rcu_assign_pointer(new_device->name, name);
 904         }
 905
 906         list_replace_rcu(&device->dev_list, &new_device->dev_list);
 907         new_device->fs_devices = device->fs_devices;
 908 }
 909
 910 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
 911 {
 912         struct btrfs_device *device, *tmp;
 913         struct list_head pending_put;
 914
 915         INIT_LIST_HEAD(&pending_put);
 916
 917         if (--fs_devices->opened > 0)
 918                 return 0;
 919
 920         mutex_lock(&fs_devices->device_list_mutex);
 921         list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list) {
 922                 btrfs_prepare_close_one_device(device);
 923                 list_add(&device->dev_list, &pending_put);
 924         }
 925         mutex_unlock(&fs_devices->device_list_mutex);
 926
 927         /*
 928          * btrfs_show_devname() is using the device_list_mutex,
 929          * sometimes call to blkdev_put() leads vfs calling
 930          * into this func. So do put outside of device_list_mutex,
 931          * as of now.
 932          */
 933         while (!list_empty(&pending_put)) {
 934                 device = list_first_entry(&pending_put,
 935                                 struct btrfs_device, dev_list);
 936                 list_del(&device->dev_list);
 937                 btrfs_close_bdev(device);
 938                 call_rcu(&device->rcu, free_device);
 939         }
 940
 941         WARN_ON(fs_devices->open_devices);
 942         WARN_ON(fs_devices->rw_devices);
 943         fs_devices->opened = 0;
 944         fs_devices->seeding = 0;
 945
 946         return 0;
 947 }
 948
 949 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
 950 {
 951         struct btrfs_fs_devices *seed_devices = NULL;
 952         int ret;
 953
 954         mutex_lock(&uuid_mutex);
 955         ret = __btrfs_close_devices(fs_devices);
 956         if (!fs_devices->opened) {
 957                 seed_devices = fs_devices->seed;
 958                 fs_devices->seed = NULL;
 959         }
 960         mutex_unlock(&uuid_mutex);
 961
 962         while (seed_devices) {
 963                 fs_devices = seed_devices;
 964                 seed_devices = fs_devices->seed;
 965                 __btrfs_close_devices(fs_devices);
 966                 free_fs_devices(fs_devices);
 967         }
 968         /*
 969          * Wait for rcu kworkers under __btrfs_close_devices
 970          * to finish all blkdev_puts so device is really
 971          * free when umount is done.
 972          */
 973         rcu_barrier();
 974         return ret;
 975 }
 976
 977 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
 978                                 fmode_t flags, void *holder)
 979 {
 980         struct request_queue *q;
 981         struct block_device *bdev;
 982         struct list_head *head = &fs_devices->devices;
 983         struct btrfs_device *device;
 984         struct btrfs_device *latest_dev = NULL;
 985         struct buffer_head *bh;
 986         struct btrfs_super_block *disk_super;
 987         u64 devid;
 988         int seeding = 1;
 989         int ret = 0;
 990
 991         flags |= FMODE_EXCL;
 992
 993         list_for_each_entry(device, head, dev_list) {
 994                 if (device->bdev)
 995                         continue;
 996                 if (!device->name)
 997                         continue;
 998
 999                 /* Just open everything we can; ignore failures here */
1000                 if (btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1,
1001                                             &bdev, &bh))
1002                         continue;
1003
1004                 disk_super = (struct btrfs_super_block *)bh->b_data;
1005                 devid = btrfs_stack_device_id(&disk_super->dev_item);
1006                 if (devid != device->devid)
1007                         goto error_brelse;
1008
1009                 if (memcmp(device->uuid, disk_super->dev_item.uuid,
1010                            BTRFS_UUID_SIZE))
1011                         goto error_brelse;
1012
1013                 device->generation = btrfs_super_generation(disk_super);
1014                 if (!latest_dev ||
1015                     device->generation > latest_dev->generation)
1016                         latest_dev = device;
1017
1018                 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
1019                         device->writeable = 0;
1020                 } else {
1021                         device->writeable = !bdev_read_only(bdev);
1022                         seeding = 0;
1023                 }
1024
1025                 q = bdev_get_queue(bdev);
1026                 if (blk_queue_discard(q))
1027                         device->can_discard = 1;
1028                 if (!blk_queue_nonrot(q))
1029                         fs_devices->rotating = 1;
1030
1031                 device->bdev = bdev;
1032                 device->in_fs_metadata = 0;
1033                 device->mode = flags;
1034
1035                 fs_devices->open_devices++;
1036                 if (device->writeable &&
1037                     device->devid != BTRFS_DEV_REPLACE_DEVID) {
1038                         fs_devices->rw_devices++;
1039                         list_add(&device->dev_alloc_list,
1040                                  &fs_devices->alloc_list);
1041                 }
1042                 brelse(bh);
1043                 continue;
1044
1045 error_brelse:
1046                 brelse(bh);
1047                 blkdev_put(bdev, flags);
1048                 continue;
1049         }
1050         if (fs_devices->open_devices == 0) {
1051                 ret = -EINVAL;
1052                 goto out;
1053         }
1054         fs_devices->seeding = seeding;
1055         fs_devices->opened = 1;
1056         fs_devices->latest_bdev = latest_dev->bdev;
1057         fs_devices->total_rw_bytes = 0;
1058 out:
1059         return ret;
1060 }
1061
1062 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
1063                        fmode_t flags, void *holder)
1064 {
1065         int ret;
1066
1067         mutex_lock(&uuid_mutex);
1068         if (fs_devices->opened) {
1069                 fs_devices->opened++;
1070                 ret = 0;
1071         } else {
1072                 ret = __btrfs_open_devices(fs_devices, flags, holder);
1073         }
1074         mutex_unlock(&uuid_mutex);
1075         return ret;
1076 }
1077
1078 void btrfs_release_disk_super(struct page *page)
1079 {
1080         kunmap(page);
1081         put_page(page);
1082 }
1083
1084 int btrfs_read_disk_super(struct block_device *bdev, u64 bytenr,
1085                 struct page **page, struct btrfs_super_block **disk_super)
1086 {
1087         void *p;
1088         pgoff_t index;
1089
1090         /* make sure our super fits in the device */
1091         if (bytenr + PAGE_SIZE >= i_size_read(bdev->bd_inode))
1092                 return 1;
1093
1094         /* make sure our super fits in the page */
1095         if (sizeof(**disk_super) > PAGE_SIZE)
1096                 return 1;
1097
1098         /* make sure our super doesn't straddle pages on disk */
1099         index = bytenr >> PAGE_SHIFT;
1100         if ((bytenr + sizeof(**disk_super) - 1) >> PAGE_SHIFT != index)
1101                 return 1;
1102
1103         /* pull in the page with our super */
1104         *page = read_cache_page_gfp(bdev->bd_inode->i_mapping,
1105                                    index, GFP_KERNEL);
1106
1107         if (IS_ERR_OR_NULL(*page))
1108                 return 1;
1109
1110         p = kmap(*page);
1111
1112         /* align our pointer to the offset of the super block */
1113         *disk_super = p + (bytenr & ~PAGE_MASK);
1114
1115         if (btrfs_super_bytenr(*disk_super) != bytenr ||
1116             btrfs_super_magic(*disk_super) != BTRFS_MAGIC) {
1117                 btrfs_release_disk_super(*page);
1118                 return 1;
1119         }
1120
1121         if ((*disk_super)->label[0] &&
1122                 (*disk_super)->label[BTRFS_LABEL_SIZE - 1])
1123                 (*disk_super)->label[BTRFS_LABEL_SIZE - 1] = '\0';
1124
1125         return 0;
1126 }
1127
1128 /*
1129  * Look for a btrfs signature on a device. This may be called out of the mount path
1130  * and we are not allowed to call set_blocksize during the scan. The superblock
1131  * is read via pagecache
1132  */
1133 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
1134                           struct btrfs_fs_devices **fs_devices_ret)
1135 {
1136         struct btrfs_super_block *disk_super;
1137         struct block_device *bdev;
1138         struct page *page;
1139         int ret = -EINVAL;
1140         u64 devid;
1141         u64 transid;
1142         u64 total_devices;
1143         u64 bytenr;
1144
1145         /*
1146          * we would like to check all the supers, but that would make
1147          * a btrfs mount succeed after a mkfs from a different FS.
1148          * So, we need to add a special mount option to scan for
1149          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
1150          */
1151         bytenr = btrfs_sb_offset(0);
1152         flags |= FMODE_EXCL;
1153         mutex_lock(&uuid_mutex);
1154
1155         bdev = blkdev_get_by_path(path, flags, holder);
1156         if (IS_ERR(bdev)) {
1157                 ret = PTR_ERR(bdev);
1158                 goto error;
1159         }
1160
1161         if (btrfs_read_disk_super(bdev, bytenr, &page, &disk_super))
1162                 goto error_bdev_put;
1163
1164         devid = btrfs_stack_device_id(&disk_super->dev_item);
1165         transid = btrfs_super_generation(disk_super);
1166         total_devices = btrfs_super_num_devices(disk_super);
1167
1168         ret = device_list_add(path, disk_super, devid, fs_devices_ret);
1169         if (ret > 0) {
1170                 if (disk_super->label[0]) {
1171                         pr_info("BTRFS: device label %s ", disk_super->label);
1172                 } else {
1173                         pr_info("BTRFS: device fsid %pU ", disk_super->fsid);
1174                 }
1175
1176                 pr_cont("devid %llu transid %llu %s\n", devid, transid, path);
1177                 ret = 0;
1178         }
1179         if (!ret && fs_devices_ret)
1180                 (*fs_devices_ret)->total_devices = total_devices;
1181
1182         btrfs_release_disk_super(page);
1183
1184 error_bdev_put:
1185         blkdev_put(bdev, flags);
1186 error:
1187         mutex_unlock(&uuid_mutex);
1188         return ret;
1189 }
1190
1191 /* helper to account the used device space in the range */
1192 int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
1193                                    u64 end, u64 *length)
1194 {
1195         struct btrfs_key key;
1196         struct btrfs_root *root = device->fs_info->dev_root;
1197         struct btrfs_dev_extent *dev_extent;
1198         struct btrfs_path *path;
1199         u64 extent_end;
1200         int ret;
1201         int slot;
1202         struct extent_buffer *l;
1203
1204         *length = 0;
1205
1206         if (start >= device->total_bytes || device->is_tgtdev_for_dev_replace)
1207                 return 0;
1208
1209         path = btrfs_alloc_path();
1210         if (!path)
1211                 return -ENOMEM;
1212         path->reada = READA_FORWARD;
1213
1214         key.objectid = device->devid;
1215         key.offset = start;
1216         key.type = BTRFS_DEV_EXTENT_KEY;
1217
1218         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1219         if (ret < 0)
1220                 goto out;
1221         if (ret > 0) {
1222                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1223                 if (ret < 0)
1224                         goto out;
1225         }
1226
1227         while (1) {
1228                 l = path->nodes[0];
1229                 slot = path->slots[0];
1230                 if (slot >= btrfs_header_nritems(l)) {
1231                         ret = btrfs_next_leaf(root, path);
1232                         if (ret == 0)
1233                                 continue;
1234                         if (ret < 0)
1235                                 goto out;
1236
1237                         break;
1238                 }
1239                 btrfs_item_key_to_cpu(l, &key, slot);
1240
1241                 if (key.objectid < device->devid)
1242                         goto next;
1243
1244                 if (key.objectid > device->devid)
1245                         break;
1246
1247                 if (key.type != BTRFS_DEV_EXTENT_KEY)
1248                         goto next;
1249
1250                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1251                 extent_end = key.offset + btrfs_dev_extent_length(l,
1252                                                                   dev_extent);
1253                 if (key.offset <= start && extent_end > end) {
1254                         *length = end - start + 1;
1255                         break;
1256                 } else if (key.offset <= start && extent_end > start)
1257                         *length += extent_end - start;
1258                 else if (key.offset > start && extent_end <= end)
1259                         *length += extent_end - key.offset;
1260                 else if (key.offset > start && key.offset <= end) {
1261                         *length += end - key.offset + 1;
1262                         break;
1263                 } else if (key.offset > end)
1264                         break;
1265
1266 next:
1267                 path->slots[0]++;
1268         }
1269         ret = 0;
1270 out:
1271         btrfs_free_path(path);
1272         return ret;
1273 }
1274
1275 static int contains_pending_extent(struct btrfs_transaction *transaction,
1276                                    struct btrfs_device *device,
1277                                    u64 *start, u64 len)
1278 {
1279         struct btrfs_fs_info *fs_info = device->fs_info;
1280         struct extent_map *em;
1281         struct list_head *search_list = &fs_info->pinned_chunks;
1282         int ret = 0;
1283         u64 physical_start = *start;
1284
1285         if (transaction)
1286                 search_list = &transaction->pending_chunks;
1287 again:
1288         list_for_each_entry(em, search_list, list) {
1289                 struct map_lookup *map;
1290                 int i;
1291
1292                 map = em->map_lookup;
1293                 for (i = 0; i < map->num_stripes; i++) {
1294                         u64 end;
1295
1296                         if (map->stripes[i].dev != device)
1297                                 continue;
1298                         if (map->stripes[i].physical >= physical_start + len ||
1299                             map->stripes[i].physical + em->orig_block_len <=
1300                             physical_start)
1301                                 continue;
1302                         /*
1303                          * Make sure that while processing the pinned list we do
1304                          * not override our *start with a lower value, because
1305                          * we can have pinned chunks that fall within this
1306                          * device hole and that have lower physical addresses
1307                          * than the pending chunks we processed before. If we
1308                          * do not take this special care we can end up getting
1309                          * 2 pending chunks that start at the same physical
1310                          * device offsets because the end offset of a pinned
1311                          * chunk can be equal to the start offset of some
1312                          * pending chunk.
1313                          */
1314                         end = map->stripes[i].physical + em->orig_block_len;
1315                         if (end > *start) {
1316                                 *start = end;
1317                                 ret = 1;
1318                         }
1319                 }
1320         }
1321         if (search_list != &fs_info->pinned_chunks) {
1322                 search_list = &fs_info->pinned_chunks;
1323                 goto again;
1324         }
1325
1326         return ret;
1327 }
1328
1329
1330 /*
1331  * find_free_dev_extent_start - find free space in the specified device
1332  * @device:       the device which we search the free space in
1333  * @num_bytes:    the size of the free space that we need
1334  * @search_start: the position from which to begin the search
1335  * @start:        store the start of the free space.
1336  * @len:          the size of the free space. that we find, or the size
1337  *                of the max free space if we don't find suitable free space
1338  *
1339  * this uses a pretty simple search, the expectation is that it is
1340  * called very infrequently and that a given device has a small number
1341  * of extents
1342  *
1343  * @start is used to store the start of the free space if we find. But if we
1344  * don't find suitable free space, it will be used to store the start position
1345  * of the max free space.
1346  *
1347  * @len is used to store the size of the free space that we find.
1348  * But if we don't find suitable free space, it is used to store the size of
1349  * the max free space.
1350  */
1351 int find_free_dev_extent_start(struct btrfs_transaction *transaction,
1352                                struct btrfs_device *device, u64 num_bytes,
1353                                u64 search_start, u64 *start, u64 *len)
1354 {
1355         struct btrfs_fs_info *fs_info = device->fs_info;
1356         struct btrfs_root *root = fs_info->dev_root;
1357         struct btrfs_key key;
1358         struct btrfs_dev_extent *dev_extent;
1359         struct btrfs_path *path;
1360         u64 hole_size;
1361         u64 max_hole_start;
1362         u64 max_hole_size;
1363         u64 extent_end;
1364         u64 search_end = device->total_bytes;
1365         int ret;
1366         int slot;
1367         struct extent_buffer *l;
1368
1369         /*
1370          * We don't want to overwrite the superblock on the drive nor any area
1371          * used by the boot loader (grub for example), so we make sure to start
1372          * at an offset of at least 1MB.
1373          */
1374         search_start = max_t(u64, search_start, SZ_1M);
1375
1376         path = btrfs_alloc_path();
1377         if (!path)
1378                 return -ENOMEM;
1379
1380         max_hole_start = search_start;
1381         max_hole_size = 0;
1382
1383 again:
1384         if (search_start >= search_end || device->is_tgtdev_for_dev_replace) {
1385                 ret = -ENOSPC;
1386                 goto out;
1387         }
1388
1389         path->reada = READA_FORWARD;
1390         path->search_commit_root = 1;
1391         path->skip_locking = 1;
1392
1393         key.objectid = device->devid;
1394         key.offset = search_start;
1395         key.type = BTRFS_DEV_EXTENT_KEY;
1396
1397         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1398         if (ret < 0)
1399                 goto out;
1400         if (ret > 0) {
1401                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1402                 if (ret < 0)
1403                         goto out;
1404         }
1405
1406         while (1) {
1407                 l = path->nodes[0];
1408                 slot = path->slots[0];
1409                 if (slot >= btrfs_header_nritems(l)) {
1410                         ret = btrfs_next_leaf(root, path);
1411                         if (ret == 0)
1412                                 continue;
1413                         if (ret < 0)
1414                                 goto out;
1415
1416                         break;
1417                 }
1418                 btrfs_item_key_to_cpu(l, &key, slot);
1419
1420                 if (key.objectid < device->devid)
1421                         goto next;
1422
1423                 if (key.objectid > device->devid)
1424                         break;
1425
1426                 if (key.type != BTRFS_DEV_EXTENT_KEY)
1427                         goto next;
1428
1429                 if (key.offset > search_start) {
1430                         hole_size = key.offset - search_start;
1431
1432                         /*
1433                          * Have to check before we set max_hole_start, otherwise
1434                          * we could end up sending back this offset anyway.
1435                          */
1436                         if (contains_pending_extent(transaction, device,
1437                                                     &search_start,
1438                                                     hole_size)) {
1439                                 if (key.offset >= search_start) {
1440                                         hole_size = key.offset - search_start;
1441                                 } else {
1442                                         WARN_ON_ONCE(1);
1443                                         hole_size = 0;
1444                                 }
1445                         }
1446
1447                         if (hole_size > max_hole_size) {
1448                                 max_hole_start = search_start;
1449                                 max_hole_size = hole_size;
1450                         }
1451
1452                         /*
1453                          * If this free space is greater than which we need,
1454                          * it must be the max free space that we have found
1455                          * until now, so max_hole_start must point to the start
1456                          * of this free space and the length of this free space
1457                          * is stored in max_hole_size. Thus, we return
1458                          * max_hole_start and max_hole_size and go back to the
1459                          * caller.
1460                          */
1461                         if (hole_size >= num_bytes) {
1462                                 ret = 0;
1463                                 goto out;
1464                         }
1465                 }
1466
1467                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1468                 extent_end = key.offset + btrfs_dev_extent_length(l,
1469                                                                   dev_extent);
1470                 if (extent_end > search_start)
1471                         search_start = extent_end;
1472 next:
1473                 path->slots[0]++;
1474                 cond_resched();
1475         }
1476
1477         /*
1478          * At this point, search_start should be the end of
1479          * allocated dev extents, and when shrinking the device,
1480          * search_end may be smaller than search_start.
1481          */
1482         if (search_end > search_start) {
1483                 hole_size = search_end - search_start;
1484
1485                 if (contains_pending_extent(transaction, device, &search_start,
1486                                             hole_size)) {
1487                         btrfs_release_path(path);
1488                         goto again;
1489                 }
1490
1491                 if (hole_size > max_hole_size) {
1492                         max_hole_start = search_start;
1493                         max_hole_size = hole_size;
1494                 }
1495         }
1496
1497         /* See above. */
1498         if (max_hole_size < num_bytes)
1499                 ret = -ENOSPC;
1500         else
1501                 ret = 0;
1502
1503 out:
1504         btrfs_free_path(path);
1505         *start = max_hole_start;
1506         if (len)
1507                 *len = max_hole_size;
1508         return ret;
1509 }
1510
1511 int find_free_dev_extent(struct btrfs_trans_handle *trans,
1512                          struct btrfs_device *device, u64 num_bytes,
1513                          u64 *start, u64 *len)
1514 {
1515         /* FIXME use last free of some kind */
1516         return find_free_dev_extent_start(trans->transaction, device,
1517                                           num_bytes, 0, start, len);
1518 }
1519
1520 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1521                           struct btrfs_device *device,
1522                           u64 start, u64 *dev_extent_len)
1523 {
1524         struct btrfs_fs_info *fs_info = device->fs_info;
1525         struct btrfs_root *root = fs_info->dev_root;
1526         int ret;
1527         struct btrfs_path *path;
1528         struct btrfs_key key;
1529         struct btrfs_key found_key;
1530         struct extent_buffer *leaf = NULL;
1531         struct btrfs_dev_extent *extent = NULL;
1532
1533         path = btrfs_alloc_path();
1534         if (!path)
1535                 return -ENOMEM;
1536
1537         key.objectid = device->devid;
1538         key.offset = start;
1539         key.type = BTRFS_DEV_EXTENT_KEY;
1540 again:
1541         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1542         if (ret > 0) {
1543                 ret = btrfs_previous_item(root, path, key.objectid,
1544                                           BTRFS_DEV_EXTENT_KEY);
1545                 if (ret)
1546                         goto out;
1547                 leaf = path->nodes[0];
1548                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1549                 extent = btrfs_item_ptr(leaf, path->slots[0],
1550                                         struct btrfs_dev_extent);
1551                 BUG_ON(found_key.offset > start || found_key.offset +
1552                        btrfs_dev_extent_length(leaf, extent) < start);
1553                 key = found_key;
1554                 btrfs_release_path(path);
1555                 goto again;
1556         } else if (ret == 0) {
1557                 leaf = path->nodes[0];
1558                 extent = btrfs_item_ptr(leaf, path->slots[0],
1559                                         struct btrfs_dev_extent);
1560         } else {
1561                 btrfs_handle_fs_error(fs_info, ret, "Slot search failed");
1562                 goto out;
1563         }
1564
1565         *dev_extent_len = btrfs_dev_extent_length(leaf, extent);
1566
1567         ret = btrfs_del_item(trans, root, path);
1568         if (ret) {
1569                 btrfs_handle_fs_error(fs_info, ret,
1570                                       "Failed to remove dev extent item");
1571         } else {
1572                 set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags);
1573         }
1574 out:
1575         btrfs_free_path(path);
1576         return ret;
1577 }
1578
1579 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1580                                   struct btrfs_device *device,
1581                                   u64 chunk_tree, u64 chunk_objectid,
1582                                   u64 chunk_offset, u64 start, u64 num_bytes)
1583 {
1584         int ret;
1585         struct btrfs_path *path;
1586         struct btrfs_fs_info *fs_info = device->fs_info;
1587         struct btrfs_root *root = fs_info->dev_root;
1588         struct btrfs_dev_extent *extent;
1589         struct extent_buffer *leaf;
1590         struct btrfs_key key;
1591
1592         WARN_ON(!device->in_fs_metadata);
1593         WARN_ON(device->is_tgtdev_for_dev_replace);
1594         path = btrfs_alloc_path();
1595         if (!path)
1596                 return -ENOMEM;
1597
1598         key.objectid = device->devid;
1599         key.offset = start;
1600         key.type = BTRFS_DEV_EXTENT_KEY;
1601         ret = btrfs_insert_empty_item(trans, root, path, &key,
1602                                       sizeof(*extent));
1603         if (ret)
1604                 goto out;
1605
1606         leaf = path->nodes[0];
1607         extent = btrfs_item_ptr(leaf, path->slots[0],
1608                                 struct btrfs_dev_extent);
1609         btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
1610         btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
1611         btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1612
1613         write_extent_buffer_chunk_tree_uuid(leaf, fs_info->chunk_tree_uuid);
1614
1615         btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1616         btrfs_mark_buffer_dirty(leaf);
1617 out:
1618         btrfs_free_path(path);
1619         return ret;
1620 }
1621
1622 static u64 find_next_chunk(struct btrfs_fs_info *fs_info)
1623 {
1624         struct extent_map_tree *em_tree;
1625         struct extent_map *em;
1626         struct rb_node *n;
1627         u64 ret = 0;
1628
1629         em_tree = &fs_info->mapping_tree.map_tree;
1630         read_lock(&em_tree->lock);
1631         n = rb_last(&em_tree->map);
1632         if (n) {
1633                 em = rb_entry(n, struct extent_map, rb_node);
1634                 ret = em->start + em->len;
1635         }
1636         read_unlock(&em_tree->lock);
1637
1638         return ret;
1639 }
1640
1641 static noinline int find_next_devid(struct btrfs_fs_info *fs_info,
1642                                     u64 *devid_ret)
1643 {
1644         int ret;
1645         struct btrfs_key key;
1646         struct btrfs_key found_key;
1647         struct btrfs_path *path;
1648
1649         path = btrfs_alloc_path();
1650         if (!path)
1651                 return -ENOMEM;
1652
1653         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1654         key.type = BTRFS_DEV_ITEM_KEY;
1655         key.offset = (u64)-1;
1656
1657         ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
1658         if (ret < 0)
1659                 goto error;
1660
1661         BUG_ON(ret == 0); /* Corruption */
1662
1663         ret = btrfs_previous_item(fs_info->chunk_root, path,
1664                                   BTRFS_DEV_ITEMS_OBJECTID,
1665                                   BTRFS_DEV_ITEM_KEY);
1666         if (ret) {
1667                 *devid_ret = 1;
1668         } else {
1669                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1670                                       path->slots[0]);
1671                 *devid_ret = found_key.offset + 1;
1672         }
1673         ret = 0;
1674 error:
1675         btrfs_free_path(path);
1676         return ret;
1677 }
1678
1679 /*
1680  * the device information is stored in the chunk root
1681  * the btrfs_device struct should be fully filled in
1682  */
1683 static int btrfs_add_device(struct btrfs_trans_handle *trans,
1684                             struct btrfs_fs_info *fs_info,
1685                             struct btrfs_device *device)
1686 {
1687         struct btrfs_root *root = fs_info->chunk_root;
1688         int ret;
1689         struct btrfs_path *path;
1690         struct btrfs_dev_item *dev_item;
1691         struct extent_buffer *leaf;
1692         struct btrfs_key key;
1693         unsigned long ptr;
1694
1695         path = btrfs_alloc_path();
1696         if (!path)
1697                 return -ENOMEM;
1698
1699         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1700         key.type = BTRFS_DEV_ITEM_KEY;
1701         key.offset = device->devid;
1702
1703         ret = btrfs_insert_empty_item(trans, root, path, &key,
1704                                       sizeof(*dev_item));
1705         if (ret)
1706                 goto out;
1707
1708         leaf = path->nodes[0];
1709         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1710
1711         btrfs_set_device_id(leaf, dev_item, device->devid);
1712         btrfs_set_device_generation(leaf, dev_item, 0);
1713         btrfs_set_device_type(leaf, dev_item, device->type);
1714         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1715         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1716         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1717         btrfs_set_device_total_bytes(leaf, dev_item,
1718                                      btrfs_device_get_disk_total_bytes(device));
1719         btrfs_set_device_bytes_used(leaf, dev_item,
1720                                     btrfs_device_get_bytes_used(device));
1721         btrfs_set_device_group(leaf, dev_item, 0);
1722         btrfs_set_device_seek_speed(leaf, dev_item, 0);
1723         btrfs_set_device_bandwidth(leaf, dev_item, 0);
1724         btrfs_set_device_start_offset(leaf, dev_item, 0);
1725
1726         ptr = btrfs_device_uuid(dev_item);
1727         write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1728         ptr = btrfs_device_fsid(dev_item);
1729         write_extent_buffer(leaf, fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1730         btrfs_mark_buffer_dirty(leaf);
1731
1732         ret = 0;
1733 out:
1734         btrfs_free_path(path);
1735         return ret;
1736 }
1737
1738 /*
1739  * Function to update ctime/mtime for a given device path.
1740  * Mainly used for ctime/mtime based probe like libblkid.
1741  */
1742 static void update_dev_time(const char *path_name)
1743 {
1744         struct file *filp;
1745
1746         filp = filp_open(path_name, O_RDWR, 0);
1747         if (IS_ERR(filp))
1748                 return;
1749         file_update_time(filp);
1750         filp_close(filp, NULL);
1751 }
1752
1753 static int btrfs_rm_dev_item(struct btrfs_fs_info *fs_info,
1754                              struct btrfs_device *device)
1755 {
1756         struct btrfs_root *root = fs_info->chunk_root;
1757         int ret;
1758         struct btrfs_path *path;
1759         struct btrfs_key key;
1760         struct btrfs_trans_handle *trans;
1761
1762         path = btrfs_alloc_path();
1763         if (!path)
1764                 return -ENOMEM;
1765
1766         trans = btrfs_start_transaction(root, 0);
1767         if (IS_ERR(trans)) {
1768                 btrfs_free_path(path);
1769                 return PTR_ERR(trans);
1770         }
1771         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1772         key.type = BTRFS_DEV_ITEM_KEY;
1773         key.offset = device->devid;
1774
1775         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1776         if (ret < 0)
1777                 goto out;
1778
1779         if (ret > 0) {
1780                 ret = -ENOENT;
1781                 goto out;
1782         }
1783
1784         ret = btrfs_del_item(trans, root, path);
1785         if (ret)
1786                 goto out;
1787 out:
1788         btrfs_free_path(path);
1789         btrfs_commit_transaction(trans);
1790         return ret;
1791 }
1792
1793 /*
1794  * Verify that @num_devices satisfies the RAID profile constraints in the whole
1795  * filesystem. It's up to the caller to adjust that number regarding eg. device
1796  * replace.
1797  */
1798 static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info,
1799                 u64 num_devices)
1800 {
1801         u64 all_avail;
1802         unsigned seq;
1803         int i;
1804
1805         do {
1806                 seq = read_seqbegin(&fs_info->profiles_lock);
1807
1808                 all_avail = fs_info->avail_data_alloc_bits |
1809                             fs_info->avail_system_alloc_bits |
1810                             fs_info->avail_metadata_alloc_bits;
1811         } while (read_seqretry(&fs_info->profiles_lock, seq));
1812
1813         for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
1814                 if (!(all_avail & btrfs_raid_group[i]))
1815                         continue;
1816
1817                 if (num_devices < btrfs_raid_array[i].devs_min) {
1818                         int ret = btrfs_raid_mindev_error[i];
1819
1820                         if (ret)
1821                                 return ret;
1822                 }
1823         }
1824
1825         return 0;
1826 }
1827
1828 struct btrfs_device *btrfs_find_next_active_device(struct btrfs_fs_devices *fs_devs,
1829                                         struct btrfs_device *device)
1830 {
1831         struct btrfs_device *next_device;
1832
1833         list_for_each_entry(next_device, &fs_devs->devices, dev_list) {
1834                 if (next_device != device &&
1835                         !next_device->missing && next_device->bdev)
1836                         return next_device;
1837         }
1838
1839         return NULL;
1840 }
1841
1842 /*
1843  * Helper function to check if the given device is part of s_bdev / latest_bdev
1844  * and replace it with the provided or the next active device, in the context
1845  * where this function called, there should be always be another device (or
1846  * this_dev) which is active.
1847  */
1848 void btrfs_assign_next_active_device(struct btrfs_fs_info *fs_info,
1849                 struct btrfs_device *device, struct btrfs_device *this_dev)
1850 {
1851         struct btrfs_device *next_device;
1852
1853         if (this_dev)
1854                 next_device = this_dev;
1855         else
1856                 next_device = btrfs_find_next_active_device(fs_info->fs_devices,
1857                                                                 device);
1858         ASSERT(next_device);
1859
1860         if (fs_info->sb->s_bdev &&
1861                         (fs_info->sb->s_bdev == device->bdev))
1862                 fs_info->sb->s_bdev = next_device->bdev;
1863
1864         if (fs_info->fs_devices->latest_bdev == device->bdev)
1865                 fs_info->fs_devices->latest_bdev = next_device->bdev;
1866 }
1867
1868 int btrfs_rm_device(struct btrfs_fs_info *fs_info, const char *device_path,
1869                 u64 devid)
1870 {
1871         struct btrfs_device *device;
1872         struct btrfs_fs_devices *cur_devices;
1873         u64 num_devices;
1874         int ret = 0;
1875         bool clear_super = false;
1876
1877         mutex_lock(&uuid_mutex);
1878
1879         num_devices = fs_info->fs_devices->num_devices;
1880         btrfs_dev_replace_lock(&fs_info->dev_replace, 0);
1881         if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
1882                 WARN_ON(num_devices < 1);
1883                 num_devices--;
1884         }
1885         btrfs_dev_replace_unlock(&fs_info->dev_replace, 0);
1886
1887         ret = btrfs_check_raid_min_devices(fs_info, num_devices - 1);
1888         if (ret)
1889                 goto out;
1890
1891         ret = btrfs_find_device_by_devspec(fs_info, devid, device_path,
1892                                            &device);
1893         if (ret)
1894                 goto out;
1895
1896         if (device->is_tgtdev_for_dev_replace) {
1897                 ret = BTRFS_ERROR_DEV_TGT_REPLACE;
1898                 goto out;
1899         }
1900
1901         if (device->writeable && fs_info->fs_devices->rw_devices == 1) {
1902                 ret = BTRFS_ERROR_DEV_ONLY_WRITABLE;
1903                 goto out;
1904         }
1905
1906         if (device->writeable) {
1907                 mutex_lock(&fs_info->chunk_mutex);
1908                 list_del_init(&device->dev_alloc_list);
1909                 device->fs_devices->rw_devices--;
1910                 mutex_unlock(&fs_info->chunk_mutex);
1911                 clear_super = true;
1912         }
1913
1914         mutex_unlock(&uuid_mutex);
1915         ret = btrfs_shrink_device(device, 0);
1916         mutex_lock(&uuid_mutex);
1917         if (ret)
1918                 goto error_undo;
1919
1920         /*
1921          * TODO: the superblock still includes this device in its num_devices
1922          * counter although write_all_supers() is not locked out. This
1923          * could give a filesystem state which requires a degraded mount.
1924          */
1925         ret = btrfs_rm_dev_item(fs_info, device);
1926         if (ret)
1927                 goto error_undo;
1928
1929         device->in_fs_metadata = 0;
1930         btrfs_scrub_cancel_dev(fs_info, device);
1931
1932         /*
1933          * the device list mutex makes sure that we don't change
1934          * the device list while someone else is writing out all
1935          * the device supers. Whoever is writing all supers, should
1936          * lock the device list mutex before getting the number of
1937          * devices in the super block (super_copy). Conversely,
1938          * whoever updates the number of devices in the super block
1939          * (super_copy) should hold the device list mutex.
1940          */
1941
1942         cur_devices = device->fs_devices;
1943         mutex_lock(&fs_info->fs_devices->device_list_mutex);
1944         list_del_rcu(&device->dev_list);
1945
1946         device->fs_devices->num_devices--;
1947         device->fs_devices->total_devices--;
1948
1949         if (device->missing)
1950                 device->fs_devices->missing_devices--;
1951
1952         btrfs_assign_next_active_device(fs_info, device, NULL);
1953
1954         if (device->bdev) {
1955                 device->fs_devices->open_devices--;
1956                 /* remove sysfs entry */
1957                 btrfs_sysfs_rm_device_link(fs_info->fs_devices, device);
1958         }
1959
1960         num_devices = btrfs_super_num_devices(fs_info->super_copy) - 1;
1961         btrfs_set_super_num_devices(fs_info->super_copy, num_devices);
1962         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
1963
1964         /*
1965          * at this point, the device is zero sized and detached from
1966          * the devices list.  All that's left is to zero out the old
1967          * supers and free the device.
1968          */
1969         if (device->writeable)
1970                 btrfs_scratch_superblocks(device->bdev, device->name->str);
1971
1972         btrfs_close_bdev(device);
1973         call_rcu(&device->rcu, free_device);
1974
1975         if (cur_devices->open_devices == 0) {
1976                 struct btrfs_fs_devices *fs_devices;
1977                 fs_devices = fs_info->fs_devices;
1978                 while (fs_devices) {
1979                         if (fs_devices->seed == cur_devices) {
1980                                 fs_devices->seed = cur_devices->seed;
1981                                 break;
1982                         }
1983                         fs_devices = fs_devices->seed;
1984                 }
1985                 cur_devices->seed = NULL;
1986                 __btrfs_close_devices(cur_devices);
1987                 free_fs_devices(cur_devices);
1988         }
1989
1990         fs_info->num_tolerated_disk_barrier_failures =
1991                 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
1992
1993 out:
1994         mutex_unlock(&uuid_mutex);
1995         return ret;
1996
1997 error_undo:
1998         if (device->writeable) {
1999                 mutex_lock(&fs_info->chunk_mutex);
2000                 list_add(&device->dev_alloc_list,
2001                          &fs_info->fs_devices->alloc_list);
2002                 device->fs_devices->rw_devices++;
2003                 mutex_unlock(&fs_info->chunk_mutex);
2004         }
2005         goto out;
2006 }
2007
2008 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_fs_info *fs_info,
2009                                         struct btrfs_device *srcdev)
2010 {
2011         struct btrfs_fs_devices *fs_devices;
2012
2013         WARN_ON(!mutex_is_locked(&fs_info->fs_devices->device_list_mutex));
2014
2015         /*
2016          * in case of fs with no seed, srcdev->fs_devices will point
2017          * to fs_devices of fs_info. However when the dev being replaced is
2018          * a seed dev it will point to the seed's local fs_devices. In short
2019          * srcdev will have its correct fs_devices in both the cases.
2020          */
2021         fs_devices = srcdev->fs_devices;
2022
2023         list_del_rcu(&srcdev->dev_list);
2024         list_del_rcu(&srcdev->dev_alloc_list);
2025         fs_devices->num_devices--;
2026         if (srcdev->missing)
2027                 fs_devices->missing_devices--;
2028
2029         if (srcdev->writeable)
2030                 fs_devices->rw_devices--;
2031
2032         if (srcdev->bdev)
2033                 fs_devices->open_devices--;
2034 }
2035
2036 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_fs_info *fs_info,
2037                                       struct btrfs_device *srcdev)
2038 {
2039         struct btrfs_fs_devices *fs_devices = srcdev->fs_devices;
2040
2041         if (srcdev->writeable) {
2042                 /* zero out the old super if it is writable */
2043                 btrfs_scratch_superblocks(srcdev->bdev, srcdev->name->str);
2044         }
2045
2046         btrfs_close_bdev(srcdev);
2047
2048         call_rcu(&srcdev->rcu, free_device);
2049
2050         /*
2051          * unless fs_devices is seed fs, num_devices shouldn't go
2052          * zero
2053          */
2054         BUG_ON(!fs_devices->num_devices && !fs_devices->seeding);
2055
2056         /* if this is no devs we rather delete the fs_devices */
2057         if (!fs_devices->num_devices) {
2058                 struct btrfs_fs_devices *tmp_fs_devices;
2059
2060                 tmp_fs_devices = fs_info->fs_devices;
2061                 while (tmp_fs_devices) {
2062                         if (tmp_fs_devices->seed == fs_devices) {
2063                                 tmp_fs_devices->seed = fs_devices->seed;
2064                                 break;
2065                         }
2066                         tmp_fs_devices = tmp_fs_devices->seed;
2067                 }
2068                 fs_devices->seed = NULL;
2069                 __btrfs_close_devices(fs_devices);
2070                 free_fs_devices(fs_devices);
2071         }
2072 }
2073
2074 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
2075                                       struct btrfs_device *tgtdev)
2076 {
2077         mutex_lock(&uuid_mutex);
2078         WARN_ON(!tgtdev);
2079         mutex_lock(&fs_info->fs_devices->device_list_mutex);
2080
2081         btrfs_sysfs_rm_device_link(fs_info->fs_devices, tgtdev);
2082
2083         if (tgtdev->bdev)
2084                 fs_info->fs_devices->open_devices--;
2085
2086         fs_info->fs_devices->num_devices--;
2087
2088         btrfs_assign_next_active_device(fs_info, tgtdev, NULL);
2089
2090         list_del_rcu(&tgtdev->dev_list);
2091
2092         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2093         mutex_unlock(&uuid_mutex);
2094
2095         /*
2096          * The update_dev_time() with in btrfs_scratch_superblocks()
2097          * may lead to a call to btrfs_show_devname() which will try
2098          * to hold device_list_mutex. And here this device
2099          * is already out of device list, so we don't have to hold
2100          * the device_list_mutex lock.
2101          */
2102         btrfs_scratch_superblocks(tgtdev->bdev, tgtdev->name->str);
2103
2104         btrfs_close_bdev(tgtdev);
2105         call_rcu(&tgtdev->rcu, free_device);
2106 }
2107
2108 static int btrfs_find_device_by_path(struct btrfs_fs_info *fs_info,
2109                                      const char *device_path,
2110                                      struct btrfs_device **device)
2111 {
2112         int ret = 0;
2113         struct btrfs_super_block *disk_super;
2114         u64 devid;
2115         u8 *dev_uuid;
2116         struct block_device *bdev;
2117         struct buffer_head *bh;
2118
2119         *device = NULL;
2120         ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
2121                                     fs_info->bdev_holder, 0, &bdev, &bh);
2122         if (ret)
2123                 return ret;
2124         disk_super = (struct btrfs_super_block *)bh->b_data;
2125         devid = btrfs_stack_device_id(&disk_super->dev_item);
2126         dev_uuid = disk_super->dev_item.uuid;
2127         *device = btrfs_find_device(fs_info, devid, dev_uuid, disk_super->fsid);
2128         brelse(bh);
2129         if (!*device)
2130                 ret = -ENOENT;
2131         blkdev_put(bdev, FMODE_READ);
2132         return ret;
2133 }
2134
2135 int btrfs_find_device_missing_or_by_path(struct btrfs_fs_info *fs_info,
2136                                          const char *device_path,
2137                                          struct btrfs_device **device)
2138 {
2139         *device = NULL;
2140         if (strcmp(device_path, "missing") == 0) {
2141                 struct list_head *devices;
2142                 struct btrfs_device *tmp;
2143
2144                 devices = &fs_info->fs_devices->devices;
2145                 /*
2146                  * It is safe to read the devices since the volume_mutex
2147                  * is held by the caller.
2148                  */
2149                 list_for_each_entry(tmp, devices, dev_list) {
2150                         if (tmp->in_fs_metadata && !tmp->bdev) {
2151                                 *device = tmp;
2152                                 break;
2153                         }
2154                 }
2155
2156                 if (!*device)
2157                         return BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
2158
2159                 return 0;
2160         } else {
2161                 return btrfs_find_device_by_path(fs_info, device_path, device);
2162         }
2163 }
2164
2165 /*
2166  * Lookup a device given by device id, or the path if the id is 0.
2167  */
2168 int btrfs_find_device_by_devspec(struct btrfs_fs_info *fs_info, u64 devid,
2169                                  const char *devpath,
2170                                  struct btrfs_device **device)
2171 {
2172         int ret;
2173
2174         if (devid) {
2175                 ret = 0;
2176                 *device = btrfs_find_device(fs_info, devid, NULL, NULL);
2177                 if (!*device)
2178                         ret = -ENOENT;
2179         } else {
2180                 if (!devpath || !devpath[0])
2181                         return -EINVAL;
2182
2183                 ret = btrfs_find_device_missing_or_by_path(fs_info, devpath,
2184                                                            device);
2185         }
2186         return ret;
2187 }
2188
2189 /*
2190  * does all the dirty work required for changing file system's UUID.
2191  */
2192 static int btrfs_prepare_sprout(struct btrfs_fs_info *fs_info)
2193 {
2194         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2195         struct btrfs_fs_devices *old_devices;
2196         struct btrfs_fs_devices *seed_devices;
2197         struct btrfs_super_block *disk_super = fs_info->super_copy;
2198         struct btrfs_device *device;
2199         u64 super_flags;
2200
2201         BUG_ON(!mutex_is_locked(&uuid_mutex));
2202         if (!fs_devices->seeding)
2203                 return -EINVAL;
2204
2205         seed_devices = __alloc_fs_devices();
2206         if (IS_ERR(seed_devices))
2207                 return PTR_ERR(seed_devices);
2208
2209         old_devices = clone_fs_devices(fs_devices);
2210         if (IS_ERR(old_devices)) {
2211                 kfree(seed_devices);
2212                 return PTR_ERR(old_devices);
2213         }
2214
2215         list_add(&old_devices->list, &fs_uuids);
2216
2217         memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
2218         seed_devices->opened = 1;
2219         INIT_LIST_HEAD(&seed_devices->devices);
2220         INIT_LIST_HEAD(&seed_devices->alloc_list);
2221         mutex_init(&seed_devices->device_list_mutex);
2222
2223         mutex_lock(&fs_info->fs_devices->device_list_mutex);
2224         list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
2225                               synchronize_rcu);
2226         list_for_each_entry(device, &seed_devices->devices, dev_list)
2227                 device->fs_devices = seed_devices;
2228
2229         mutex_lock(&fs_info->chunk_mutex);
2230         list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
2231         mutex_unlock(&fs_info->chunk_mutex);
2232
2233         fs_devices->seeding = 0;
2234         fs_devices->num_devices = 0;
2235         fs_devices->open_devices = 0;
2236         fs_devices->missing_devices = 0;
2237         fs_devices->rotating = 0;
2238         fs_devices->seed = seed_devices;
2239
2240         generate_random_uuid(fs_devices->fsid);
2241         memcpy(fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2242         memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2243         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2244
2245         super_flags = btrfs_super_flags(disk_super) &
2246                       ~BTRFS_SUPER_FLAG_SEEDING;
2247         btrfs_set_super_flags(disk_super, super_flags);
2248
2249         return 0;
2250 }
2251
2252 /*
2253  * Store the expected generation for seed devices in device items.
2254  */
2255 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
2256                                struct btrfs_fs_info *fs_info)
2257 {
2258         struct btrfs_root *root = fs_info->chunk_root;
2259         struct btrfs_path *path;
2260         struct extent_buffer *leaf;
2261         struct btrfs_dev_item *dev_item;
2262         struct btrfs_device *device;
2263         struct btrfs_key key;
2264         u8 fs_uuid[BTRFS_UUID_SIZE];
2265         u8 dev_uuid[BTRFS_UUID_SIZE];
2266         u64 devid;
2267         int ret;
2268
2269         path = btrfs_alloc_path();
2270         if (!path)
2271                 return -ENOMEM;
2272
2273         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2274         key.offset = 0;
2275         key.type = BTRFS_DEV_ITEM_KEY;
2276
2277         while (1) {
2278                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2279                 if (ret < 0)
2280                         goto error;
2281
2282                 leaf = path->nodes[0];
2283 next_slot:
2284                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2285                         ret = btrfs_next_leaf(root, path);
2286                         if (ret > 0)
2287                                 break;
2288                         if (ret < 0)
2289                                 goto error;
2290                         leaf = path->nodes[0];
2291                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2292                         btrfs_release_path(path);
2293                         continue;
2294                 }
2295
2296                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2297                 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
2298                     key.type != BTRFS_DEV_ITEM_KEY)
2299                         break;
2300
2301                 dev_item = btrfs_item_ptr(leaf, path->slots[0],
2302                                           struct btrfs_dev_item);
2303                 devid = btrfs_device_id(leaf, dev_item);
2304                 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
2305                                    BTRFS_UUID_SIZE);
2306                 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
2307                                    BTRFS_UUID_SIZE);
2308                 device = btrfs_find_device(fs_info, devid, dev_uuid, fs_uuid);
2309                 BUG_ON(!device); /* Logic error */
2310
2311                 if (device->fs_devices->seeding) {
2312                         btrfs_set_device_generation(leaf, dev_item,
2313                                                     device->generation);
2314                         btrfs_mark_buffer_dirty(leaf);
2315                 }
2316
2317                 path->slots[0]++;
2318                 goto next_slot;
2319         }
2320         ret = 0;
2321 error:
2322         btrfs_free_path(path);
2323         return ret;
2324 }
2325
2326 int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path)
2327 {
2328         struct btrfs_root *root = fs_info->dev_root;
2329         struct request_queue *q;
2330         struct btrfs_trans_handle *trans;
2331         struct btrfs_device *device;
2332         struct block_device *bdev;
2333         struct list_head *devices;
2334         struct super_block *sb = fs_info->sb;
2335         struct rcu_string *name;
2336         u64 tmp;
2337         int seeding_dev = 0;
2338         int ret = 0;
2339
2340         if ((sb->s_flags & MS_RDONLY) && !fs_info->fs_devices->seeding)
2341                 return -EROFS;
2342
2343         bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2344                                   fs_info->bdev_holder);
2345         if (IS_ERR(bdev))
2346                 return PTR_ERR(bdev);
2347
2348         if (fs_info->fs_devices->seeding) {
2349                 seeding_dev = 1;
2350                 down_write(&sb->s_umount);
2351                 mutex_lock(&uuid_mutex);
2352         }
2353
2354         filemap_write_and_wait(bdev->bd_inode->i_mapping);
2355
2356         devices = &fs_info->fs_devices->devices;
2357
2358         mutex_lock(&fs_info->fs_devices->device_list_mutex);
2359         list_for_each_entry(device, devices, dev_list) {
2360                 if (device->bdev == bdev) {
2361                         ret = -EEXIST;
2362                         mutex_unlock(
2363                                 &fs_info->fs_devices->device_list_mutex);
2364                         goto error;
2365                 }
2366         }
2367         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2368
2369         device = btrfs_alloc_device(fs_info, NULL, NULL);
2370         if (IS_ERR(device)) {
2371                 /* we can safely leave the fs_devices entry around */
2372                 ret = PTR_ERR(device);
2373                 goto error;
2374         }
2375
2376         name = rcu_string_strdup(device_path, GFP_KERNEL);
2377         if (!name) {
2378                 kfree(device);
2379                 ret = -ENOMEM;
2380                 goto error;
2381         }
2382         rcu_assign_pointer(device->name, name);
2383
2384         trans = btrfs_start_transaction(root, 0);
2385         if (IS_ERR(trans)) {
2386                 rcu_string_free(device->name);
2387                 kfree(device);
2388                 ret = PTR_ERR(trans);
2389                 goto error;
2390         }
2391
2392         q = bdev_get_queue(bdev);
2393         if (blk_queue_discard(q))
2394                 device->can_discard = 1;
2395         device->writeable = 1;
2396         device->generation = trans->transid;
2397         device->io_width = fs_info->sectorsize;
2398         device->io_align = fs_info->sectorsize;
2399         device->sector_size = fs_info->sectorsize;
2400         device->total_bytes = round_down(i_size_read(bdev->bd_inode),
2401                                          fs_info->sectorsize);
2402         device->disk_total_bytes = device->total_bytes;
2403         device->commit_total_bytes = device->total_bytes;
2404         device->fs_info = fs_info;
2405         device->bdev = bdev;
2406         device->in_fs_metadata = 1;
2407         device->is_tgtdev_for_dev_replace = 0;
2408         device->mode = FMODE_EXCL;
2409         device->dev_stats_valid = 1;
2410         set_blocksize(device->bdev, 4096);
2411
2412         if (seeding_dev) {
2413                 sb->s_flags &= ~MS_RDONLY;
2414                 ret = btrfs_prepare_sprout(fs_info);
2415                 BUG_ON(ret); /* -ENOMEM */
2416         }
2417
2418         device->fs_devices = fs_info->fs_devices;
2419
2420         mutex_lock(&fs_info->fs_devices->device_list_mutex);
2421         mutex_lock(&fs_info->chunk_mutex);
2422         list_add_rcu(&device->dev_list, &fs_info->fs_devices->devices);
2423         list_add(&device->dev_alloc_list,
2424                  &fs_info->fs_devices->alloc_list);
2425         fs_info->fs_devices->num_devices++;
2426         fs_info->fs_devices->open_devices++;
2427         fs_info->fs_devices->rw_devices++;
2428         fs_info->fs_devices->total_devices++;
2429         fs_info->fs_devices->total_rw_bytes += device->total_bytes;
2430
2431         atomic64_add(device->total_bytes, &fs_info->free_chunk_space);
2432
2433         if (!blk_queue_nonrot(q))
2434                 fs_info->fs_devices->rotating = 1;
2435
2436         tmp = btrfs_super_total_bytes(fs_info->super_copy);
2437         btrfs_set_super_total_bytes(fs_info->super_copy,
2438                 round_down(tmp + device->total_bytes, fs_info->sectorsize));
2439
2440         tmp = btrfs_super_num_devices(fs_info->super_copy);
2441         btrfs_set_super_num_devices(fs_info->super_copy, tmp + 1);
2442
2443         /* add sysfs device entry */
2444         btrfs_sysfs_add_device_link(fs_info->fs_devices, device);
2445
2446         /*
2447          * we've got more storage, clear any full flags on the space
2448          * infos
2449          */
2450         btrfs_clear_space_info_full(fs_info);
2451
2452         mutex_unlock(&fs_info->chunk_mutex);
2453         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2454
2455         if (seeding_dev) {
2456                 mutex_lock(&fs_info->chunk_mutex);
2457                 ret = init_first_rw_device(trans, fs_info);
2458                 mutex_unlock(&fs_info->chunk_mutex);
2459                 if (ret) {
2460                         btrfs_abort_transaction(trans, ret);
2461                         goto error_trans;
2462                 }
2463         }
2464
2465         ret = btrfs_add_device(trans, fs_info, device);
2466         if (ret) {
2467                 btrfs_abort_transaction(trans, ret);
2468                 goto error_trans;
2469         }
2470
2471         if (seeding_dev) {
2472                 char fsid_buf[BTRFS_UUID_UNPARSED_SIZE];
2473
2474                 ret = btrfs_finish_sprout(trans, fs_info);
2475                 if (ret) {
2476                         btrfs_abort_transaction(trans, ret);
2477                         goto error_trans;
2478                 }
2479
2480                 /* Sprouting would change fsid of the mounted root,
2481                  * so rename the fsid on the sysfs
2482                  */
2483                 snprintf(fsid_buf, BTRFS_UUID_UNPARSED_SIZE, "%pU",
2484                                                 fs_info->fsid);
2485                 if (kobject_rename(&fs_info->fs_devices->fsid_kobj, fsid_buf))
2486                         btrfs_warn(fs_info,
2487                                    "sysfs: failed to create fsid for sprout");
2488         }
2489
2490         fs_info->num_tolerated_disk_barrier_failures =
2491                 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
2492         ret = btrfs_commit_transaction(trans);
2493
2494         if (seeding_dev) {
2495                 mutex_unlock(&uuid_mutex);
2496                 up_write(&sb->s_umount);
2497
2498                 if (ret) /* transaction commit */
2499                         return ret;
2500
2501                 ret = btrfs_relocate_sys_chunks(fs_info);
2502                 if (ret < 0)
2503                         btrfs_handle_fs_error(fs_info, ret,
2504                                     "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command.");
2505                 trans = btrfs_attach_transaction(root);
2506                 if (IS_ERR(trans)) {
2507                         if (PTR_ERR(trans) == -ENOENT)
2508                                 return 0;
2509                         return PTR_ERR(trans);
2510                 }
2511                 ret = btrfs_commit_transaction(trans);
2512         }
2513
2514         /* Update ctime/mtime for libblkid */
2515         update_dev_time(device_path);
2516         return ret;
2517
2518 error_trans:
2519         btrfs_end_transaction(trans);
2520         rcu_string_free(device->name);
2521         btrfs_sysfs_rm_device_link(fs_info->fs_devices, device);
2522         kfree(device);
2523 error:
2524         blkdev_put(bdev, FMODE_EXCL);
2525         if (seeding_dev) {
2526                 mutex_unlock(&uuid_mutex);
2527                 up_write(&sb->s_umount);
2528         }
2529         return ret;
2530 }
2531
2532 int btrfs_init_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
2533                                   const char *device_path,
2534                                   struct btrfs_device *srcdev,
2535                                   struct btrfs_device **device_out)
2536 {
2537         struct request_queue *q;
2538         struct btrfs_device *device;
2539         struct block_device *bdev;
2540         struct list_head *devices;
2541         struct rcu_string *name;
2542         u64 devid = BTRFS_DEV_REPLACE_DEVID;
2543         int ret = 0;
2544
2545         *device_out = NULL;
2546         if (fs_info->fs_devices->seeding) {
2547                 btrfs_err(fs_info, "the filesystem is a seed filesystem!");
2548                 return -EINVAL;
2549         }
2550
2551         bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2552                                   fs_info->bdev_holder);
2553         if (IS_ERR(bdev)) {
2554                 btrfs_err(fs_info, "target device %s is invalid!", device_path);
2555                 return PTR_ERR(bdev);
2556         }
2557
2558         filemap_write_and_wait(bdev->bd_inode->i_mapping);
2559
2560         devices = &fs_info->fs_devices->devices;
2561         list_for_each_entry(device, devices, dev_list) {
2562                 if (device->bdev == bdev) {
2563                         btrfs_err(fs_info,
2564                                   "target device is in the filesystem!");
2565                         ret = -EEXIST;
2566                         goto error;
2567                 }
2568         }
2569
2570
2571         if (i_size_read(bdev->bd_inode) <
2572             btrfs_device_get_total_bytes(srcdev)) {
2573                 btrfs_err(fs_info,
2574                           "target device is smaller than source device!");
2575                 ret = -EINVAL;
2576                 goto error;
2577         }
2578
2579
2580         device = btrfs_alloc_device(NULL, &devid, NULL);
2581         if (IS_ERR(device)) {
2582                 ret = PTR_ERR(device);
2583                 goto error;
2584         }
2585
2586         name = rcu_string_strdup(device_path, GFP_KERNEL);
2587         if (!name) {
2588                 kfree(device);
2589                 ret = -ENOMEM;
2590                 goto error;
2591         }
2592         rcu_assign_pointer(device->name, name);
2593
2594         q = bdev_get_queue(bdev);
2595         if (blk_queue_discard(q))
2596                 device->can_discard = 1;
2597         mutex_lock(&fs_info->fs_devices->device_list_mutex);
2598         device->writeable = 1;
2599         device->generation = 0;
2600         device->io_width = fs_info->sectorsize;
2601         device->io_align = fs_info->sectorsize;
2602         device->sector_size = fs_info->sectorsize;
2603         device->total_bytes = btrfs_device_get_total_bytes(srcdev);
2604         device->disk_total_bytes = btrfs_device_get_disk_total_bytes(srcdev);
2605         device->bytes_used = btrfs_device_get_bytes_used(srcdev);
2606         ASSERT(list_empty(&srcdev->resized_list));
2607         device->commit_total_bytes = srcdev->commit_total_bytes;
2608         device->commit_bytes_used = device->bytes_used;
2609         device->fs_info = fs_info;
2610         device->bdev = bdev;
2611         device->in_fs_metadata = 1;
2612         device->is_tgtdev_for_dev_replace = 1;
2613         device->mode = FMODE_EXCL;
2614         device->dev_stats_valid = 1;
2615         set_blocksize(device->bdev, 4096);
2616         device->fs_devices = fs_info->fs_devices;
2617         list_add(&device->dev_list, &fs_info->fs_devices->devices);
2618         fs_info->fs_devices->num_devices++;
2619         fs_info->fs_devices->open_devices++;
2620         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2621
2622         *device_out = device;
2623         return ret;
2624
2625 error:
2626         blkdev_put(bdev, FMODE_EXCL);
2627         return ret;
2628 }
2629
2630 void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info *fs_info,
2631                                               struct btrfs_device *tgtdev)
2632 {
2633         u32 sectorsize = fs_info->sectorsize;
2634
2635         WARN_ON(fs_info->fs_devices->rw_devices == 0);
2636         tgtdev->io_width = sectorsize;
2637         tgtdev->io_align = sectorsize;
2638         tgtdev->sector_size = sectorsize;
2639         tgtdev->fs_info = fs_info;
2640         tgtdev->in_fs_metadata = 1;
2641 }
2642
2643 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2644                                         struct btrfs_device *device)
2645 {
2646         int ret;
2647         struct btrfs_path *path;
2648         struct btrfs_root *root = device->fs_info->chunk_root;
2649         struct btrfs_dev_item *dev_item;
2650         struct extent_buffer *leaf;
2651         struct btrfs_key key;
2652
2653         path = btrfs_alloc_path();
2654         if (!path)
2655                 return -ENOMEM;
2656
2657         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2658         key.type = BTRFS_DEV_ITEM_KEY;
2659         key.offset = device->devid;
2660
2661         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2662         if (ret < 0)
2663                 goto out;
2664
2665         if (ret > 0) {
2666                 ret = -ENOENT;
2667                 goto out;
2668         }
2669
2670         leaf = path->nodes[0];
2671         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2672
2673         btrfs_set_device_id(leaf, dev_item, device->devid);
2674         btrfs_set_device_type(leaf, dev_item, device->type);
2675         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2676         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2677         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2678         btrfs_set_device_total_bytes(leaf, dev_item,
2679                                      btrfs_device_get_disk_total_bytes(device));
2680         btrfs_set_device_bytes_used(leaf, dev_item,
2681                                     btrfs_device_get_bytes_used(device));
2682         btrfs_mark_buffer_dirty(leaf);
2683
2684 out:
2685         btrfs_free_path(path);
2686         return ret;
2687 }
2688
2689 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2690                       struct btrfs_device *device, u64 new_size)
2691 {
2692         struct btrfs_fs_info *fs_info = device->fs_info;
2693         struct btrfs_super_block *super_copy = fs_info->super_copy;
2694         struct btrfs_fs_devices *fs_devices;
2695         u64 old_total;
2696         u64 diff;
2697
2698         if (!device->writeable)
2699                 return -EACCES;
2700
2701         new_size = round_down(new_size, fs_info->sectorsize);
2702
2703         mutex_lock(&fs_info->chunk_mutex);
2704         old_total = btrfs_super_total_bytes(super_copy);
2705         diff = new_size - device->total_bytes;
2706
2707         if (new_size <= device->total_bytes ||
2708             device->is_tgtdev_for_dev_replace) {
2709                 mutex_unlock(&fs_info->chunk_mutex);
2710                 return -EINVAL;
2711         }
2712
2713         fs_devices = fs_info->fs_devices;
2714
2715         btrfs_set_super_total_bytes(super_copy,
2716                         round_down(old_total + diff, fs_info->sectorsize));
2717         device->fs_devices->total_rw_bytes += diff;
2718
2719         btrfs_device_set_total_bytes(device, new_size);
2720         btrfs_device_set_disk_total_bytes(device, new_size);
2721         btrfs_clear_space_info_full(device->fs_info);
2722         if (list_empty(&device->resized_list))
2723                 list_add_tail(&device->resized_list,
2724                               &fs_devices->resized_devices);
2725         mutex_unlock(&fs_info->chunk_mutex);
2726
2727         return btrfs_update_device(trans, device);
2728 }
2729
2730 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
2731                             struct btrfs_fs_info *fs_info, u64 chunk_objectid,
2732                             u64 chunk_offset)
2733 {
2734         struct btrfs_root *root = fs_info->chunk_root;
2735         int ret;
2736         struct btrfs_path *path;
2737         struct btrfs_key key;
2738
2739         path = btrfs_alloc_path();
2740         if (!path)
2741                 return -ENOMEM;
2742
2743         key.objectid = chunk_objectid;
2744         key.offset = chunk_offset;
2745         key.type = BTRFS_CHUNK_ITEM_KEY;
2746
2747         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2748         if (ret < 0)
2749                 goto out;
2750         else if (ret > 0) { /* Logic error or corruption */
2751                 btrfs_handle_fs_error(fs_info, -ENOENT,
2752                                       "Failed lookup while freeing chunk.");
2753                 ret = -ENOENT;
2754                 goto out;
2755         }
2756
2757         ret = btrfs_del_item(trans, root, path);
2758         if (ret < 0)
2759                 btrfs_handle_fs_error(fs_info, ret,
2760                                       "Failed to delete chunk item.");
2761 out:
2762         btrfs_free_path(path);
2763         return ret;
2764 }
2765
2766 static int btrfs_del_sys_chunk(struct btrfs_fs_info *fs_info,
2767                                u64 chunk_objectid, u64 chunk_offset)
2768 {
2769         struct btrfs_super_block *super_copy = fs_info->super_copy;
2770         struct btrfs_disk_key *disk_key;
2771         struct btrfs_chunk *chunk;
2772         u8 *ptr;
2773         int ret = 0;
2774         u32 num_stripes;
2775         u32 array_size;
2776         u32 len = 0;
2777         u32 cur;
2778         struct btrfs_key key;
2779
2780         mutex_lock(&fs_info->chunk_mutex);
2781         array_size = btrfs_super_sys_array_size(super_copy);
2782
2783         ptr = super_copy->sys_chunk_array;
2784         cur = 0;
2785
2786         while (cur < array_size) {
2787                 disk_key = (struct btrfs_disk_key *)ptr;
2788                 btrfs_disk_key_to_cpu(&key, disk_key);
2789
2790                 len = sizeof(*disk_key);
2791
2792                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2793                         chunk = (struct btrfs_chunk *)(ptr + len);
2794                         num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2795                         len += btrfs_chunk_item_size(num_stripes);
2796                 } else {
2797                         ret = -EIO;
2798                         break;
2799                 }
2800                 if (key.objectid == chunk_objectid &&
2801                     key.offset == chunk_offset) {
2802                         memmove(ptr, ptr + len, array_size - (cur + len));
2803                         array_size -= len;
2804                         btrfs_set_super_sys_array_size(super_copy, array_size);
2805                 } else {
2806                         ptr += len;
2807                         cur += len;
2808                 }
2809         }
2810         mutex_unlock(&fs_info->chunk_mutex);
2811         return ret;
2812 }
2813
2814 static struct extent_map *get_chunk_map(struct btrfs_fs_info *fs_info,
2815                                         u64 logical, u64 length)
2816 {
2817         struct extent_map_tree *em_tree;
2818         struct extent_map *em;
2819
2820         em_tree = &fs_info->mapping_tree.map_tree;
2821         read_lock(&em_tree->lock);
2822         em = lookup_extent_mapping(em_tree, logical, length);
2823         read_unlock(&em_tree->lock);
2824
2825         if (!em) {
2826                 btrfs_crit(fs_info, "unable to find logical %llu length %llu",
2827                            logical, length);
2828                 return ERR_PTR(-EINVAL);
2829         }
2830
2831         if (em->start > logical || em->start + em->len < logical) {
2832                 btrfs_crit(fs_info,
2833                            "found a bad mapping, wanted %llu-%llu, found %llu-%llu",
2834                            logical, length, em->start, em->start + em->len);
2835                 free_extent_map(em);
2836                 return ERR_PTR(-EINVAL);
2837         }
2838
2839         /* callers are responsible for dropping em's ref. */
2840         return em;
2841 }
2842
2843 int btrfs_remove_chunk(struct btrfs_trans_handle *trans,
2844                        struct btrfs_fs_info *fs_info, u64 chunk_offset)
2845 {
2846         struct extent_map *em;
2847         struct map_lookup *map;
2848         u64 dev_extent_len = 0;
2849         u64 chunk_objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2850         int i, ret = 0;
2851         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2852
2853         em = get_chunk_map(fs_info, chunk_offset, 1);
2854         if (IS_ERR(em)) {
2855                 /*
2856                  * This is a logic error, but we don't want to just rely on the
2857                  * user having built with ASSERT enabled, so if ASSERT doesn't
2858                  * do anything we still error out.
2859                  */
2860                 ASSERT(0);
2861                 return PTR_ERR(em);
2862         }
2863         map = em->map_lookup;
2864         mutex_lock(&fs_info->chunk_mutex);
2865         check_system_chunk(trans, fs_info, map->type);
2866         mutex_unlock(&fs_info->chunk_mutex);
2867
2868         /*
2869          * Take the device list mutex to prevent races with the final phase of
2870          * a device replace operation that replaces the device object associated
2871          * with map stripes (dev-replace.c:btrfs_dev_replace_finishing()).
2872          */
2873         mutex_lock(&fs_devices->device_list_mutex);
2874         for (i = 0; i < map->num_stripes; i++) {
2875                 struct btrfs_device *device = map->stripes[i].dev;
2876                 ret = btrfs_free_dev_extent(trans, device,
2877                                             map->stripes[i].physical,
2878                                             &dev_extent_len);
2879                 if (ret) {
2880                         mutex_unlock(&fs_devices->device_list_mutex);
2881                         btrfs_abort_transaction(trans, ret);
2882                         goto out;
2883                 }
2884
2885                 if (device->bytes_used > 0) {
2886                         mutex_lock(&fs_info->chunk_mutex);
2887                         btrfs_device_set_bytes_used(device,
2888                                         device->bytes_used - dev_extent_len);
2889                         atomic64_add(dev_extent_len, &fs_info->free_chunk_space);
2890                         btrfs_clear_space_info_full(fs_info);
2891                         mutex_unlock(&fs_info->chunk_mutex);
2892                 }
2893
2894                 if (map->stripes[i].dev) {
2895                         ret = btrfs_update_device(trans, map->stripes[i].dev);
2896                         if (ret) {
2897                                 mutex_unlock(&fs_devices->device_list_mutex);
2898                                 btrfs_abort_transaction(trans, ret);
2899                                 goto out;
2900                         }
2901                 }
2902         }
2903         mutex_unlock(&fs_devices->device_list_mutex);
2904
2905         ret = btrfs_free_chunk(trans, fs_info, chunk_objectid, chunk_offset);
2906         if (ret) {
2907                 btrfs_abort_transaction(trans, ret);
2908                 goto out;
2909         }
2910
2911         trace_btrfs_chunk_free(fs_info, map, chunk_offset, em->len);
2912
2913         if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2914                 ret = btrfs_del_sys_chunk(fs_info, chunk_objectid,
2915                                           chunk_offset);
2916                 if (ret) {
2917                         btrfs_abort_transaction(trans, ret);
2918                         goto out;
2919                 }
2920         }
2921
2922         ret = btrfs_remove_block_group(trans, fs_info, chunk_offset, em);
2923         if (ret) {
2924                 btrfs_abort_transaction(trans, ret);
2925                 goto out;
2926         }
2927
2928 out:
2929         /* once for us */
2930         free_extent_map(em);
2931         return ret;
2932 }
2933
2934 static int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
2935 {
2936         struct btrfs_root *root = fs_info->chunk_root;
2937         struct btrfs_trans_handle *trans;
2938         int ret;
2939
2940         /*
2941          * Prevent races with automatic removal of unused block groups.
2942          * After we relocate and before we remove the chunk with offset
2943          * chunk_offset, automatic removal of the block group can kick in,
2944          * resulting in a failure when calling btrfs_remove_chunk() below.
2945          *
2946          * Make sure to acquire this mutex before doing a tree search (dev
2947          * or chunk trees) to find chunks. Otherwise the cleaner kthread might
2948          * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after
2949          * we release the path used to search the chunk/dev tree and before
2950          * the current task acquires this mutex and calls us.
2951          */
2952         ASSERT(mutex_is_locked(&fs_info->delete_unused_bgs_mutex));
2953
2954         ret = btrfs_can_relocate(fs_info, chunk_offset);
2955         if (ret)
2956                 return -ENOSPC;
2957
2958         /* step one, relocate all the extents inside this chunk */
2959         btrfs_scrub_pause(fs_info);
2960         ret = btrfs_relocate_block_group(fs_info, chunk_offset);
2961         btrfs_scrub_continue(fs_info);
2962         if (ret)
2963                 return ret;
2964
2965         trans = btrfs_start_trans_remove_block_group(root->fs_info,
2966                                                      chunk_offset);
2967         if (IS_ERR(trans)) {
2968                 ret = PTR_ERR(trans);
2969                 btrfs_handle_fs_error(root->fs_info, ret, NULL);
2970                 return ret;
2971         }
2972
2973         /*
2974          * step two, delete the device extents and the
2975          * chunk tree entries
2976          */
2977         ret = btrfs_remove_chunk(trans, fs_info, chunk_offset);
2978         btrfs_end_transaction(trans);
2979         return ret;
2980 }
2981
2982 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info)
2983 {
2984         struct btrfs_root *chunk_root = fs_info->chunk_root;
2985         struct btrfs_path *path;
2986         struct extent_buffer *leaf;
2987         struct btrfs_chunk *chunk;
2988         struct btrfs_key key;
2989         struct btrfs_key found_key;
2990         u64 chunk_type;
2991         bool retried = false;
2992         int failed = 0;
2993         int ret;
2994
2995         path = btrfs_alloc_path();
2996         if (!path)
2997                 return -ENOMEM;
2998
2999 again:
3000         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3001         key.offset = (u64)-1;
3002         key.type = BTRFS_CHUNK_ITEM_KEY;
3003
3004         while (1) {
3005                 mutex_lock(&fs_info->delete_unused_bgs_mutex);
3006                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3007                 if (ret < 0) {
3008                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3009                         goto error;
3010                 }
3011                 BUG_ON(ret == 0); /* Corruption */
3012
3013                 ret = btrfs_previous_item(chunk_root, path, key.objectid,
3014                                           key.type);
3015                 if (ret)
3016                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3017                 if (ret < 0)
3018                         goto error;
3019                 if (ret > 0)
3020                         break;
3021
3022                 leaf = path->nodes[0];
3023                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3024
3025                 chunk = btrfs_item_ptr(leaf, path->slots[0],
3026                                        struct btrfs_chunk);
3027                 chunk_type = btrfs_chunk_type(leaf, chunk);
3028                 btrfs_release_path(path);
3029
3030                 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
3031                         ret = btrfs_relocate_chunk(fs_info, found_key.offset);
3032                         if (ret == -ENOSPC)
3033                                 failed++;
3034                         else
3035                                 BUG_ON(ret);
3036                 }
3037                 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3038
3039                 if (found_key.offset == 0)
3040                         break;
3041                 key.offset = found_key.offset - 1;
3042         }
3043         ret = 0;
3044         if (failed && !retried) {
3045                 failed = 0;
3046                 retried = true;
3047                 goto again;
3048         } else if (WARN_ON(failed && retried)) {
3049                 ret = -ENOSPC;
3050         }
3051 error:
3052         btrfs_free_path(path);
3053         return ret;
3054 }
3055
3056 static int insert_balance_item(struct btrfs_fs_info *fs_info,
3057                                struct btrfs_balance_control *bctl)
3058 {
3059         struct btrfs_root *root = fs_info->tree_root;
3060         struct btrfs_trans_handle *trans;
3061         struct btrfs_balance_item *item;
3062         struct btrfs_disk_balance_args disk_bargs;
3063         struct btrfs_path *path;
3064         struct extent_buffer *leaf;
3065         struct btrfs_key key;
3066         int ret, err;
3067
3068         path = btrfs_alloc_path();
3069         if (!path)
3070                 return -ENOMEM;
3071
3072         trans = btrfs_start_transaction(root, 0);
3073         if (IS_ERR(trans)) {
3074                 btrfs_free_path(path);
3075                 return PTR_ERR(trans);
3076         }
3077
3078         key.objectid = BTRFS_BALANCE_OBJECTID;
3079         key.type = BTRFS_TEMPORARY_ITEM_KEY;
3080         key.offset = 0;
3081
3082         ret = btrfs_insert_empty_item(trans, root, path, &key,
3083                                       sizeof(*item));
3084         if (ret)
3085                 goto out;
3086
3087         leaf = path->nodes[0];
3088         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3089
3090         memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item));
3091
3092         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
3093         btrfs_set_balance_data(leaf, item, &disk_bargs);
3094         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
3095         btrfs_set_balance_meta(leaf, item, &disk_bargs);
3096         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
3097         btrfs_set_balance_sys(leaf, item, &disk_bargs);
3098
3099         btrfs_set_balance_flags(leaf, item, bctl->flags);
3100
3101         btrfs_mark_buffer_dirty(leaf);
3102 out:
3103         btrfs_free_path(path);
3104         err = btrfs_commit_transaction(trans);
3105         if (err && !ret)
3106                 ret = err;
3107         return ret;
3108 }
3109
3110 static int del_balance_item(struct btrfs_fs_info *fs_info)
3111 {
3112         struct btrfs_root *root = fs_info->tree_root;
3113         struct btrfs_trans_handle *trans;
3114         struct btrfs_path *path;
3115         struct btrfs_key key;
3116         int ret, err;
3117
3118         path = btrfs_alloc_path();
3119         if (!path)
3120                 return -ENOMEM;
3121
3122         trans = btrfs_start_transaction(root, 0);
3123         if (IS_ERR(trans)) {
3124                 btrfs_free_path(path);
3125                 return PTR_ERR(trans);
3126         }
3127
3128         key.objectid = BTRFS_BALANCE_OBJECTID;
3129         key.type = BTRFS_TEMPORARY_ITEM_KEY;
3130         key.offset = 0;
3131
3132         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3133         if (ret < 0)
3134                 goto out;
3135         if (ret > 0) {
3136                 ret = -ENOENT;
3137                 goto out;
3138         }
3139
3140         ret = btrfs_del_item(trans, root, path);
3141 out:
3142         btrfs_free_path(path);
3143         err = btrfs_commit_transaction(trans);
3144         if (err && !ret)
3145                 ret = err;
3146         return ret;
3147 }
3148
3149 /*
3150  * This is a heuristic used to reduce the number of chunks balanced on
3151  * resume after balance was interrupted.
3152  */
3153 static void update_balance_args(struct btrfs_balance_control *bctl)
3154 {
3155         /*
3156          * Turn on soft mode for chunk types that were being converted.
3157          */
3158         if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
3159                 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
3160         if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
3161                 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
3162         if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
3163                 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
3164
3165         /*
3166          * Turn on usage filter if is not already used.  The idea is
3167          * that chunks that we have already balanced should be
3168          * reasonably full.  Don't do it for chunks that are being
3169          * converted - that will keep us from relocating unconverted
3170          * (albeit full) chunks.
3171          */
3172         if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3173             !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3174             !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3175                 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
3176                 bctl->data.usage = 90;
3177         }
3178         if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3179             !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3180             !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3181                 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
3182                 bctl->sys.usage = 90;
3183         }
3184         if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3185             !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3186             !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3187                 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
3188                 bctl->meta.usage = 90;
3189         }
3190 }
3191
3192 /*
3193  * Should be called with both balance and volume mutexes held to
3194  * serialize other volume operations (add_dev/rm_dev/resize) with
3195  * restriper.  Same goes for unset_balance_control.
3196  */
3197 static void set_balance_control(struct btrfs_balance_control *bctl)
3198 {
3199         struct btrfs_fs_info *fs_info = bctl->fs_info;
3200
3201         BUG_ON(fs_info->balance_ctl);
3202
3203         spin_lock(&fs_info->balance_lock);
3204         fs_info->balance_ctl = bctl;
3205         spin_unlock(&fs_info->balance_lock);
3206 }
3207
3208 static void unset_balance_control(struct btrfs_fs_info *fs_info)
3209 {
3210         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3211
3212         BUG_ON(!fs_info->balance_ctl);
3213
3214         spin_lock(&fs_info->balance_lock);
3215         fs_info->balance_ctl = NULL;
3216         spin_unlock(&fs_info->balance_lock);
3217
3218         kfree(bctl);
3219 }
3220
3221 /*
3222  * Balance filters.  Return 1 if chunk should be filtered out
3223  * (should not be balanced).
3224  */
3225 static int chunk_profiles_filter(u64 chunk_type,
3226                                  struct btrfs_balance_args *bargs)
3227 {
3228         chunk_type = chunk_to_extended(chunk_type) &
3229                                 BTRFS_EXTENDED_PROFILE_MASK;
3230
3231         if (bargs->profiles & chunk_type)
3232                 return 0;
3233
3234         return 1;
3235 }
3236
3237 static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
3238                               struct btrfs_balance_args *bargs)
3239 {
3240         struct btrfs_block_group_cache *cache;
3241         u64 chunk_used;
3242         u64 user_thresh_min;
3243         u64 user_thresh_max;
3244         int ret = 1;
3245
3246         cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3247         chunk_used = btrfs_block_group_used(&cache->item);
3248
3249         if (bargs->usage_min == 0)
3250                 user_thresh_min = 0;
3251         else
3252                 user_thresh_min = div_factor_fine(cache->key.offset,
3253                                         bargs->usage_min);
3254
3255         if (bargs->usage_max == 0)
3256                 user_thresh_max = 1;
3257         else if (bargs->usage_max > 100)
3258                 user_thresh_max = cache->key.offset;
3259         else
3260                 user_thresh_max = div_factor_fine(cache->key.offset,
3261                                         bargs->usage_max);
3262
3263         if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max)
3264                 ret = 0;
3265
3266         btrfs_put_block_group(cache);
3267         return ret;
3268 }
3269
3270 static int chunk_usage_filter(struct btrfs_fs_info *fs_info,
3271                 u64 chunk_offset, struct btrfs_balance_args *bargs)
3272 {
3273         struct btrfs_block_group_cache *cache;
3274         u64 chunk_used, user_thresh;
3275         int ret = 1;
3276
3277         cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3278         chunk_used = btrfs_block_group_used(&cache->item);
3279
3280         if (bargs->usage_min == 0)
3281                 user_thresh = 1;
3282         else if (bargs->usage > 100)
3283                 user_thresh = cache->key.offset;
3284         else
3285                 user_thresh = div_factor_fine(cache->key.offset,
3286                                               bargs->usage);
3287
3288         if (chunk_used < user_thresh)
3289                 ret = 0;
3290
3291         btrfs_put_block_group(cache);
3292         return ret;
3293 }
3294
3295 static int chunk_devid_filter(struct extent_buffer *leaf,
3296                               struct btrfs_chunk *chunk,
3297                               struct btrfs_balance_args *bargs)
3298 {
3299         struct btrfs_stripe *stripe;
3300         int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3301         int i;
3302
3303         for (i = 0; i < num_stripes; i++) {
3304                 stripe = btrfs_stripe_nr(chunk, i);
3305                 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
3306                         return 0;
3307         }
3308
3309         return 1;
3310 }
3311
3312 /* [pstart, pend) */
3313 static int chunk_drange_filter(struct extent_buffer *leaf,
3314                                struct btrfs_chunk *chunk,
3315                                u64 chunk_offset,
3316                                struct btrfs_balance_args *bargs)
3317 {
3318         struct btrfs_stripe *stripe;
3319         int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3320         u64 stripe_offset;
3321         u64 stripe_length;
3322         int factor;
3323         int i;
3324
3325         if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
3326                 return 0;
3327
3328         if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP |
3329              BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) {
3330                 factor = num_stripes / 2;
3331         } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) {
3332                 factor = num_stripes - 1;
3333         } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) {
3334                 factor = num_stripes - 2;
3335         } else {
3336                 factor = num_stripes;
3337         }
3338
3339         for (i = 0; i < num_stripes; i++) {
3340                 stripe = btrfs_stripe_nr(chunk, i);
3341                 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
3342                         continue;
3343
3344                 stripe_offset = btrfs_stripe_offset(leaf, stripe);
3345                 stripe_length = btrfs_chunk_length(leaf, chunk);
3346                 stripe_length = div_u64(stripe_length, factor);
3347
3348                 if (stripe_offset < bargs->pend &&
3349                     stripe_offset + stripe_length > bargs->pstart)
3350                         return 0;
3351         }
3352
3353         return 1;
3354 }
3355
3356 /* [vstart, vend) */
3357 static int chunk_vrange_filter(struct extent_buffer *leaf,
3358                                struct btrfs_chunk *chunk,
3359                                u64 chunk_offset,
3360                                struct btrfs_balance_args *bargs)
3361 {
3362         if (chunk_offset < bargs->vend &&
3363             chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
3364                 /* at least part of the chunk is inside this vrange */
3365                 return 0;
3366
3367         return 1;
3368 }
3369
3370 static int chunk_stripes_range_filter(struct extent_buffer *leaf,
3371                                struct btrfs_chunk *chunk,
3372                                struct btrfs_balance_args *bargs)
3373 {
3374         int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3375
3376         if (bargs->stripes_min <= num_stripes
3377                         && num_stripes <= bargs->stripes_max)
3378                 return 0;
3379
3380         return 1;
3381 }
3382
3383 static int chunk_soft_convert_filter(u64 chunk_type,
3384                                      struct btrfs_balance_args *bargs)
3385 {
3386         if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
3387                 return 0;
3388
3389         chunk_type = chunk_to_extended(chunk_type) &
3390                                 BTRFS_EXTENDED_PROFILE_MASK;
3391
3392         if (bargs->target == chunk_type)
3393                 return 1;
3394
3395         return 0;
3396 }
3397
3398 static int should_balance_chunk(struct btrfs_fs_info *fs_info,
3399                                 struct extent_buffer *leaf,
3400                                 struct btrfs_chunk *chunk, u64 chunk_offset)
3401 {
3402         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3403         struct btrfs_balance_args *bargs = NULL;
3404         u64 chunk_type = btrfs_chunk_type(leaf, chunk);
3405
3406         /* type filter */
3407         if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
3408               (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
3409                 return 0;
3410         }
3411
3412         if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3413                 bargs = &bctl->data;
3414         else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3415                 bargs = &bctl->sys;
3416         else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3417                 bargs = &bctl->meta;
3418
3419         /* profiles filter */
3420         if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
3421             chunk_profiles_filter(chunk_type, bargs)) {
3422                 return 0;
3423         }
3424
3425         /* usage filter */
3426         if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
3427             chunk_usage_filter(fs_info, chunk_offset, bargs)) {
3428                 return 0;
3429         } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3430             chunk_usage_range_filter(fs_info, chunk_offset, bargs)) {
3431                 return 0;
3432         }
3433
3434         /* devid filter */
3435         if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
3436             chunk_devid_filter(leaf, chunk, bargs)) {
3437                 return 0;
3438         }
3439
3440         /* drange filter, makes sense only with devid filter */
3441         if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
3442             chunk_drange_filter(leaf, chunk, chunk_offset, bargs)) {
3443                 return 0;
3444         }
3445
3446         /* vrange filter */
3447         if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
3448             chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
3449                 return 0;
3450         }
3451
3452         /* stripes filter */
3453         if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) &&
3454             chunk_stripes_range_filter(leaf, chunk, bargs)) {
3455                 return 0;
3456         }
3457
3458         /* soft profile changing mode */
3459         if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
3460             chunk_soft_convert_filter(chunk_type, bargs)) {
3461                 return 0;
3462         }
3463
3464         /*
3465          * limited by count, must be the last filter
3466          */
3467         if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) {
3468                 if (bargs->limit == 0)
3469                         return 0;
3470                 else
3471                         bargs->limit--;
3472         } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) {
3473                 /*
3474                  * Same logic as the 'limit' filter; the minimum cannot be
3475                  * determined here because we do not have the global information
3476                  * about the count of all chunks that satisfy the filters.
3477                  */
3478                 if (bargs->limit_max == 0)
3479                         return 0;
3480                 else
3481                         bargs->limit_max--;
3482         }
3483
3484         return 1;
3485 }
3486
3487 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
3488 {
3489         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3490         struct btrfs_root *chunk_root = fs_info->chunk_root;
3491         struct btrfs_root *dev_root = fs_info->dev_root;
3492         struct list_head *devices;
3493         struct btrfs_device *device;
3494         u64 old_size;
3495         u64 size_to_free;
3496         u64 chunk_type;
3497         struct btrfs_chunk *chunk;
3498         struct btrfs_path *path = NULL;
3499         struct btrfs_key key;
3500         struct btrfs_key found_key;
3501         struct btrfs_trans_handle *trans;
3502         struct extent_buffer *leaf;
3503         int slot;
3504         int ret;
3505         int enospc_errors = 0;
3506         bool counting = true;
3507         /* The single value limit and min/max limits use the same bytes in the */
3508         u64 limit_data = bctl->data.limit;
3509         u64 limit_meta = bctl->meta.limit;
3510         u64 limit_sys = bctl->sys.limit;
3511         u32 count_data = 0;
3512         u32 count_meta = 0;
3513         u32 count_sys = 0;
3514         int chunk_reserved = 0;
3515         u64 bytes_used = 0;
3516
3517         /* step one make some room on all the devices */
3518         devices = &fs_info->fs_devices->devices;
3519         list_for_each_entry(device, devices, dev_list) {
3520                 old_size = btrfs_device_get_total_bytes(device);
3521                 size_to_free = div_factor(old_size, 1);
3522                 size_to_free = min_t(u64, size_to_free, SZ_1M);
3523                 if (!device->writeable ||
3524                     btrfs_device_get_total_bytes(device) -
3525                     btrfs_device_get_bytes_used(device) > size_to_free ||
3526                     device->is_tgtdev_for_dev_replace)
3527                         continue;
3528
3529                 ret = btrfs_shrink_device(device, old_size - size_to_free);
3530                 if (ret == -ENOSPC)
3531                         break;
3532                 if (ret) {
3533                         /* btrfs_shrink_device never returns ret > 0 */
3534                         WARN_ON(ret > 0);
3535                         goto error;
3536                 }
3537
3538                 trans = btrfs_start_transaction(dev_root, 0);
3539                 if (IS_ERR(trans)) {
3540                         ret = PTR_ERR(trans);
3541                         btrfs_info_in_rcu(fs_info,
3542                  "resize: unable to start transaction after shrinking device %s (error %d), old size %llu, new size %llu",
3543                                           rcu_str_deref(device->name), ret,
3544                                           old_size, old_size - size_to_free);
3545                         goto error;
3546                 }
3547
3548                 ret = btrfs_grow_device(trans, device, old_size);
3549                 if (ret) {
3550                         btrfs_end_transaction(trans);
3551                         /* btrfs_grow_device never returns ret > 0 */
3552                         WARN_ON(ret > 0);
3553                         btrfs_info_in_rcu(fs_info,
3554                  "resize: unable to grow device after shrinking device %s (error %d), old size %llu, new size %llu",
3555                                           rcu_str_deref(device->name), ret,
3556                                           old_size, old_size - size_to_free);
3557                         goto error;
3558                 }
3559
3560                 btrfs_end_transaction(trans);
3561         }
3562
3563         /* step two, relocate all the chunks */
3564         path = btrfs_alloc_path();
3565         if (!path) {
3566                 ret = -ENOMEM;
3567                 goto error;
3568         }
3569
3570         /* zero out stat counters */
3571         spin_lock(&fs_info->balance_lock);
3572         memset(&bctl->stat, 0, sizeof(bctl->stat));
3573         spin_unlock(&fs_info->balance_lock);
3574 again:
3575         if (!counting) {
3576                 /*
3577                  * The single value limit and min/max limits use the same bytes
3578                  * in the
3579                  */
3580                 bctl->data.limit = limit_data;
3581                 bctl->meta.limit = limit_meta;
3582                 bctl->sys.limit = limit_sys;
3583         }
3584         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3585         key.offset = (u64)-1;
3586         key.type = BTRFS_CHUNK_ITEM_KEY;
3587
3588         while (1) {
3589                 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
3590                     atomic_read(&fs_info->balance_cancel_req)) {
3591                         ret = -ECANCELED;
3592                         goto error;
3593                 }
3594
3595                 mutex_lock(&fs_info->delete_unused_bgs_mutex);
3596                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3597                 if (ret < 0) {
3598                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3599                         goto error;
3600                 }
3601
3602                 /*
3603                  * this shouldn't happen, it means the last relocate
3604                  * failed
3605                  */
3606                 if (ret == 0)
3607                         BUG(); /* FIXME break ? */
3608
3609                 ret = btrfs_previous_item(chunk_root, path, 0,
3610                                           BTRFS_CHUNK_ITEM_KEY);
3611                 if (ret) {
3612                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3613                         ret = 0;
3614                         break;
3615                 }
3616
3617                 leaf = path->nodes[0];
3618                 slot = path->slots[0];
3619                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3620
3621                 if (found_key.objectid != key.objectid) {
3622                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3623                         break;
3624                 }
3625
3626                 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3627                 chunk_type = btrfs_chunk_type(leaf, chunk);
3628
3629                 if (!counting) {
3630                         spin_lock(&fs_info->balance_lock);
3631                         bctl->stat.considered++;
3632                         spin_unlock(&fs_info->balance_lock);
3633                 }
3634
3635                 ret = should_balance_chunk(fs_info, leaf, chunk,
3636                                            found_key.offset);
3637
3638                 btrfs_release_path(path);
3639                 if (!ret) {
3640                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3641                         goto loop;
3642                 }
3643
3644                 if (counting) {
3645                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3646                         spin_lock(&fs_info->balance_lock);
3647                         bctl->stat.expected++;
3648                         spin_unlock(&fs_info->balance_lock);
3649
3650                         if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3651                                 count_data++;
3652                         else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3653                                 count_sys++;
3654                         else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3655                                 count_meta++;
3656
3657                         goto loop;
3658                 }
3659
3660                 /*
3661                  * Apply limit_min filter, no need to check if the LIMITS
3662                  * filter is used, limit_min is 0 by default
3663                  */
3664                 if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
3665                                         count_data < bctl->data.limit_min)
3666                                 || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) &&
3667                                         count_meta < bctl->meta.limit_min)
3668                                 || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) &&
3669                                         count_sys < bctl->sys.limit_min)) {
3670                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3671                         goto loop;
3672                 }
3673
3674                 ASSERT(fs_info->data_sinfo);
3675                 spin_lock(&fs_info->data_sinfo->lock);
3676                 bytes_used = fs_info->data_sinfo->bytes_used;
3677                 spin_unlock(&fs_info->data_sinfo->lock);
3678
3679                 if ((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
3680                     !chunk_reserved && !bytes_used) {
3681                         trans = btrfs_start_transaction(chunk_root, 0);
3682                         if (IS_ERR(trans)) {
3683                                 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3684                                 ret = PTR_ERR(trans);
3685                                 goto error;
3686                         }
3687
3688                         ret = btrfs_force_chunk_alloc(trans, fs_info,
3689                                                       BTRFS_BLOCK_GROUP_DATA);
3690                         btrfs_end_transaction(trans);
3691                         if (ret < 0) {
3692                                 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3693                                 goto error;
3694                         }
3695                         chunk_reserved = 1;
3696                 }
3697
3698                 ret = btrfs_relocate_chunk(fs_info, found_key.offset);
3699                 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3700                 if (ret && ret != -ENOSPC)
3701                         goto error;
3702                 if (ret == -ENOSPC) {
3703                         enospc_errors++;
3704                 } else {
3705                         spin_lock(&fs_info->balance_lock);
3706                         bctl->stat.completed++;
3707                         spin_unlock(&fs_info->balance_lock);
3708                 }
3709 loop:
3710                 if (found_key.offset == 0)
3711                         break;
3712                 key.offset = found_key.offset - 1;
3713         }
3714
3715         if (counting) {
3716                 btrfs_release_path(path);
3717                 counting = false;
3718                 goto again;
3719         }
3720 error:
3721         btrfs_free_path(path);
3722         if (enospc_errors) {
3723                 btrfs_info(fs_info, "%d enospc errors during balance",
3724                            enospc_errors);
3725                 if (!ret)
3726                         ret = -ENOSPC;
3727         }
3728
3729         return ret;
3730 }
3731
3732 /**
3733  * alloc_profile_is_valid - see if a given profile is valid and reduced
3734  * @flags: profile to validate
3735  * @extended: if true @flags is treated as an extended profile
3736  */
3737 static int alloc_profile_is_valid(u64 flags, int extended)
3738 {
3739         u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
3740                                BTRFS_BLOCK_GROUP_PROFILE_MASK);
3741
3742         flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
3743
3744         /* 1) check that all other bits are zeroed */
3745         if (flags & ~mask)
3746                 return 0;
3747
3748         /* 2) see if profile is reduced */
3749         if (flags == 0)
3750                 return !extended; /* "0" is valid for usual profiles */
3751
3752         /* true if exactly one bit set */
3753         return (flags & (flags - 1)) == 0;
3754 }
3755
3756 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
3757 {
3758         /* cancel requested || normal exit path */
3759         return atomic_read(&fs_info->balance_cancel_req) ||
3760                 (atomic_read(&fs_info->balance_pause_req) == 0 &&
3761                  atomic_read(&fs_info->balance_cancel_req) == 0);
3762 }
3763
3764 static void __cancel_balance(struct btrfs_fs_info *fs_info)
3765 {
3766         int ret;
3767
3768         unset_balance_control(fs_info);
3769         ret = del_balance_item(fs_info);
3770         if (ret)
3771                 btrfs_handle_fs_error(fs_info, ret, NULL);
3772
3773         clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
3774 }
3775
3776 /* Non-zero return value signifies invalidity */
3777 static inline int validate_convert_profile(struct btrfs_balance_args *bctl_arg,
3778                 u64 allowed)
3779 {
3780         return ((bctl_arg->flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3781                 (!alloc_profile_is_valid(bctl_arg->target, 1) ||
3782                  (bctl_arg->target & ~allowed)));
3783 }
3784
3785 /*
3786  * Should be called with both balance and volume mutexes held
3787  */
3788 int btrfs_balance(struct btrfs_balance_control *bctl,
3789                   struct btrfs_ioctl_balance_args *bargs)
3790 {
3791         struct btrfs_fs_info *fs_info = bctl->fs_info;
3792         u64 meta_target, data_target;
3793         u64 allowed;
3794         int mixed = 0;
3795         int ret;
3796         u64 num_devices;
3797         unsigned seq;
3798
3799         if (btrfs_fs_closing(fs_info) ||
3800             atomic_read(&fs_info->balance_pause_req) ||
3801             atomic_read(&fs_info->balance_cancel_req)) {
3802                 ret = -EINVAL;
3803                 goto out;
3804         }
3805
3806         allowed = btrfs_super_incompat_flags(fs_info->super_copy);
3807         if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
3808                 mixed = 1;
3809
3810         /*
3811          * In case of mixed groups both data and meta should be picked,
3812          * and identical options should be given for both of them.
3813          */
3814         allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
3815         if (mixed && (bctl->flags & allowed)) {
3816                 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
3817                     !(bctl->flags & BTRFS_BALANCE_METADATA) ||
3818                     memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
3819                         btrfs_err(fs_info,
3820                                   "with mixed groups data and metadata balance options must be the same");
3821                         ret = -EINVAL;
3822                         goto out;
3823                 }
3824         }
3825
3826         num_devices = fs_info->fs_devices->num_devices;
3827         btrfs_dev_replace_lock(&fs_info->dev_replace, 0);
3828         if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
3829                 BUG_ON(num_devices < 1);
3830                 num_devices--;
3831         }
3832         btrfs_dev_replace_unlock(&fs_info->dev_replace, 0);
3833         allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE | BTRFS_BLOCK_GROUP_DUP;
3834         if (num_devices > 1)
3835                 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1);
3836         if (num_devices > 2)
3837                 allowed |= BTRFS_BLOCK_GROUP_RAID5;
3838         if (num_devices > 3)
3839                 allowed |= (BTRFS_BLOCK_GROUP_RAID10 |
3840                             BTRFS_BLOCK_GROUP_RAID6);
3841         if (validate_convert_profile(&bctl->data, allowed)) {
3842                 btrfs_err(fs_info,
3843                           "unable to start balance with target data profile %llu",
3844                           bctl->data.target);
3845                 ret = -EINVAL;
3846                 goto out;
3847         }
3848         if (validate_convert_profile(&bctl->meta, allowed)) {
3849                 btrfs_err(fs_info,
3850                           "unable to start balance with target metadata profile %llu",
3851                           bctl->meta.target);
3852                 ret = -EINVAL;
3853                 goto out;
3854         }
3855         if (validate_convert_profile(&bctl->sys, allowed)) {
3856                 btrfs_err(fs_info,
3857                           "unable to start balance with target system profile %llu",
3858                           bctl->sys.target);
3859                 ret = -EINVAL;
3860                 goto out;
3861         }
3862
3863         /* allow to reduce meta or sys integrity only if force set */
3864         allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3865                         BTRFS_BLOCK_GROUP_RAID10 |
3866                         BTRFS_BLOCK_GROUP_RAID5 |
3867                         BTRFS_BLOCK_GROUP_RAID6;
3868         do {
3869                 seq = read_seqbegin(&fs_info->profiles_lock);
3870
3871                 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3872                      (fs_info->avail_system_alloc_bits & allowed) &&
3873                      !(bctl->sys.target & allowed)) ||
3874                     ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3875                      (fs_info->avail_metadata_alloc_bits & allowed) &&
3876                      !(bctl->meta.target & allowed))) {
3877                         if (bctl->flags & BTRFS_BALANCE_FORCE) {
3878                                 btrfs_info(fs_info,
3879                                            "force reducing metadata integrity");
3880                         } else {
3881                                 btrfs_err(fs_info,
3882                                           "balance will reduce metadata integrity, use force if you want this");
3883                                 ret = -EINVAL;
3884                                 goto out;
3885                         }
3886                 }
3887         } while (read_seqretry(&fs_info->profiles_lock, seq));
3888
3889         /* if we're not converting, the target field is uninitialized */
3890         meta_target = (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
3891                 bctl->meta.target : fs_info->avail_metadata_alloc_bits;
3892         data_target = (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
3893                 bctl->data.target : fs_info->avail_data_alloc_bits;
3894         if (btrfs_get_num_tolerated_disk_barrier_failures(meta_target) <
3895                 btrfs_get_num_tolerated_disk_barrier_failures(data_target)) {
3896                 btrfs_warn(fs_info,
3897                            "metadata profile 0x%llx has lower redundancy than data profile 0x%llx",
3898                            meta_target, data_target);
3899         }
3900
3901         if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3902                 fs_info->num_tolerated_disk_barrier_failures = min(
3903                         btrfs_calc_num_tolerated_disk_barrier_failures(fs_info),
3904                         btrfs_get_num_tolerated_disk_barrier_failures(
3905                                 bctl->sys.target));
3906         }
3907
3908         ret = insert_balance_item(fs_info, bctl);
3909         if (ret && ret != -EEXIST)
3910                 goto out;
3911
3912         if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
3913                 BUG_ON(ret == -EEXIST);
3914                 set_balance_control(bctl);
3915         } else {
3916                 BUG_ON(ret != -EEXIST);
3917                 spin_lock(&fs_info->balance_lock);
3918                 update_balance_args(bctl);
3919                 spin_unlock(&fs_info->balance_lock);
3920         }
3921
3922         atomic_inc(&fs_info->balance_running);
3923         mutex_unlock(&fs_info->balance_mutex);
3924
3925         ret = __btrfs_balance(fs_info);
3926
3927         mutex_lock(&fs_info->balance_mutex);
3928         atomic_dec(&fs_info->balance_running);
3929
3930         if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3931                 fs_info->num_tolerated_disk_barrier_failures =
3932                         btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3933         }
3934
3935         if (bargs) {
3936                 memset(bargs, 0, sizeof(*bargs));
3937                 update_ioctl_balance_args(fs_info, 0, bargs);
3938         }
3939
3940         if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
3941             balance_need_close(fs_info)) {
3942                 __cancel_balance(fs_info);
3943         }
3944
3945         wake_up(&fs_info->balance_wait_q);
3946
3947         return ret;
3948 out:
3949         if (bctl->flags & BTRFS_BALANCE_RESUME)
3950                 __cancel_balance(fs_info);
3951         else {
3952                 kfree(bctl);
3953                 clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
3954         }
3955         return ret;
3956 }
3957
3958 static int balance_kthread(void *data)
3959 {
3960         struct btrfs_fs_info *fs_info = data;
3961         int ret = 0;
3962
3963         mutex_lock(&fs_info->volume_mutex);
3964         mutex_lock(&fs_info->balance_mutex);
3965
3966         if (fs_info->balance_ctl) {
3967                 btrfs_info(fs_info, "continuing balance");
3968                 ret = btrfs_balance(fs_info->balance_ctl, NULL);
3969         }
3970
3971         mutex_unlock(&fs_info->balance_mutex);
3972         mutex_unlock(&fs_info->volume_mutex);
3973
3974         return ret;
3975 }
3976
3977 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
3978 {
3979         struct task_struct *tsk;
3980
3981         spin_lock(&fs_info->balance_lock);
3982         if (!fs_info->balance_ctl) {
3983                 spin_unlock(&fs_info->balance_lock);
3984                 return 0;
3985         }
3986         spin_unlock(&fs_info->balance_lock);
3987
3988         if (btrfs_test_opt(fs_info, SKIP_BALANCE)) {
3989                 btrfs_info(fs_info, "force skipping balance");
3990                 return 0;
3991         }
3992
3993         tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
3994         return PTR_ERR_OR_ZERO(tsk);
3995 }
3996
3997 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
3998 {
3999         struct btrfs_balance_control *bctl;
4000         struct btrfs_balance_item *item;
4001         struct btrfs_disk_balance_args disk_bargs;
4002         struct btrfs_path *path;
4003         struct extent_buffer *leaf;
4004         struct btrfs_key key;
4005         int ret;
4006
4007         path = btrfs_alloc_path();
4008         if (!path)
4009                 return -ENOMEM;
4010
4011         key.objectid = BTRFS_BALANCE_OBJECTID;
4012         key.type = BTRFS_TEMPORARY_ITEM_KEY;
4013         key.offset = 0;
4014
4015         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
4016         if (ret < 0)
4017                 goto out;
4018         if (ret > 0) { /* ret = -ENOENT; */
4019                 ret = 0;
4020                 goto out;
4021         }
4022
4023         bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
4024         if (!bctl) {
4025                 ret = -ENOMEM;
4026                 goto out;
4027         }
4028
4029         leaf = path->nodes[0];
4030         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
4031
4032         bctl->fs_info = fs_info;
4033         bctl->flags = btrfs_balance_flags(leaf, item);
4034         bctl->flags |= BTRFS_BALANCE_RESUME;
4035
4036         btrfs_balance_data(leaf, item, &disk_bargs);
4037         btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
4038         btrfs_balance_meta(leaf, item, &disk_bargs);
4039         btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
4040         btrfs_balance_sys(leaf, item, &disk_bargs);
4041         btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
4042
4043         WARN_ON(test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags));
4044
4045         mutex_lock(&fs_info->volume_mutex);
4046         mutex_lock(&fs_info->balance_mutex);
4047
4048         set_balance_control(bctl);
4049
4050         mutex_unlock(&fs_info->balance_mutex);
4051         mutex_unlock(&fs_info->volume_mutex);
4052 out:
4053         btrfs_free_path(path);
4054         return ret;
4055 }
4056
4057 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
4058 {
4059         int ret = 0;
4060
4061         mutex_lock(&fs_info->balance_mutex);
4062         if (!fs_info->balance_ctl) {
4063                 mutex_unlock(&fs_info->balance_mutex);
4064                 return -ENOTCONN;
4065         }
4066
4067         if (atomic_read(&fs_info->balance_running)) {
4068                 atomic_inc(&fs_info->balance_pause_req);
4069                 mutex_unlock(&fs_info->balance_mutex);
4070
4071                 wait_event(fs_info->balance_wait_q,
4072                            atomic_read(&fs_info->balance_running) == 0);
4073
4074                 mutex_lock(&fs_info->balance_mutex);
4075                 /* we are good with balance_ctl ripped off from under us */
4076                 BUG_ON(atomic_read(&fs_info->balance_running));
4077                 atomic_dec(&fs_info->balance_pause_req);
4078         } else {
4079                 ret = -ENOTCONN;
4080         }
4081
4082         mutex_unlock(&fs_info->balance_mutex);
4083         return ret;
4084 }
4085
4086 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
4087 {
4088         if (fs_info->sb->s_flags & MS_RDONLY)
4089                 return -EROFS;
4090
4091         mutex_lock(&fs_info->balance_mutex);
4092         if (!fs_info->balance_ctl) {
4093                 mutex_unlock(&fs_info->balance_mutex);
4094                 return -ENOTCONN;
4095         }
4096
4097         atomic_inc(&fs_info->balance_cancel_req);
4098         /*
4099          * if we are running just wait and return, balance item is
4100          * deleted in btrfs_balance in this case
4101          */
4102         if (atomic_read(&fs_info->balance_running)) {
4103                 mutex_unlock(&fs_info->balance_mutex);
4104                 wait_event(fs_info->balance_wait_q,
4105                            atomic_read(&fs_info->balance_running) == 0);
4106                 mutex_lock(&fs_info->balance_mutex);
4107         } else {
4108                 /* __cancel_balance needs volume_mutex */
4109                 mutex_unlock(&fs_info->balance_mutex);
4110                 mutex_lock(&fs_info->volume_mutex);
4111                 mutex_lock(&fs_info->balance_mutex);
4112
4113                 if (fs_info->balance_ctl)
4114                         __cancel_balance(fs_info);
4115
4116                 mutex_unlock(&fs_info->volume_mutex);
4117         }
4118
4119         BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running));
4120         atomic_dec(&fs_info->balance_cancel_req);
4121         mutex_unlock(&fs_info->balance_mutex);
4122         return 0;
4123 }
4124
4125 static int btrfs_uuid_scan_kthread(void *data)
4126 {
4127         struct btrfs_fs_info *fs_info = data;
4128         struct btrfs_root *root = fs_info->tree_root;
4129         struct btrfs_key key;
4130         struct btrfs_key max_key;
4131         struct btrfs_path *path = NULL;
4132         int ret = 0;
4133         struct extent_buffer *eb;
4134         int slot;
4135         struct btrfs_root_item root_item;
4136         u32 item_size;
4137         struct btrfs_trans_handle *trans = NULL;
4138
4139         path = btrfs_alloc_path();
4140         if (!path) {
4141                 ret = -ENOMEM;
4142                 goto out;
4143         }
4144
4145         key.objectid = 0;
4146         key.type = BTRFS_ROOT_ITEM_KEY;
4147         key.offset = 0;
4148
4149         max_key.objectid = (u64)-1;
4150         max_key.type = BTRFS_ROOT_ITEM_KEY;
4151         max_key.offset = (u64)-1;
4152
4153         while (1) {
4154                 ret = btrfs_search_forward(root, &key, path, 0);
4155                 if (ret) {
4156                         if (ret > 0)
4157                                 ret = 0;
4158                         break;
4159                 }
4160
4161                 if (key.type != BTRFS_ROOT_ITEM_KEY ||
4162                     (key.objectid < BTRFS_FIRST_FREE_OBJECTID &&
4163                      key.objectid != BTRFS_FS_TREE_OBJECTID) ||
4164                     key.objectid > BTRFS_LAST_FREE_OBJECTID)
4165                         goto skip;
4166
4167                 eb = path->nodes[0];
4168                 slot = path->slots[0];
4169                 item_size = btrfs_item_size_nr(eb, slot);
4170                 if (item_size < sizeof(root_item))
4171                         goto skip;
4172
4173                 read_extent_buffer(eb, &root_item,
4174                                    btrfs_item_ptr_offset(eb, slot),
4175                                    (int)sizeof(root_item));
4176                 if (btrfs_root_refs(&root_item) == 0)
4177                         goto skip;
4178
4179                 if (!btrfs_is_empty_uuid(root_item.uuid) ||
4180                     !btrfs_is_empty_uuid(root_item.received_uuid)) {
4181                         if (trans)
4182                                 goto update_tree;
4183
4184                         btrfs_release_path(path);
4185                         /*
4186                          * 1 - subvol uuid item
4187                          * 1 - received_subvol uuid item
4188                          */
4189                         trans = btrfs_start_transaction(fs_info->uuid_root, 2);
4190                         if (IS_ERR(trans)) {
4191                                 ret = PTR_ERR(trans);
4192                                 break;
4193                         }
4194                         continue;
4195                 } else {
4196                         goto skip;
4197                 }
4198 update_tree:
4199                 if (!btrfs_is_empty_uuid(root_item.uuid)) {
4200                         ret = btrfs_uuid_tree_add(trans, fs_info,
4201                                                   root_item.uuid,
4202                                                   BTRFS_UUID_KEY_SUBVOL,
4203                                                   key.objectid);
4204                         if (ret < 0) {
4205                                 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4206                                         ret);
4207                                 break;
4208                         }
4209                 }
4210
4211                 if (!btrfs_is_empty_uuid(root_item.received_uuid)) {
4212                         ret = btrfs_uuid_tree_add(trans, fs_info,
4213                                                   root_item.received_uuid,
4214                                                  BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4215                                                   key.objectid);
4216                         if (ret < 0) {
4217                                 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4218                                         ret);
4219                                 break;
4220                         }
4221                 }
4222
4223 skip:
4224                 if (trans) {
4225                         ret = btrfs_end_transaction(trans);
4226                         trans = NULL;
4227                         if (ret)
4228                                 break;
4229                 }
4230
4231                 btrfs_release_path(path);
4232                 if (key.offset < (u64)-1) {
4233                         key.offset++;
4234                 } else if (key.type < BTRFS_ROOT_ITEM_KEY) {
4235                         key.offset = 0;
4236                         key.type = BTRFS_ROOT_ITEM_KEY;
4237                 } else if (key.objectid < (u64)-1) {
4238                         key.offset = 0;
4239                         key.type = BTRFS_ROOT_ITEM_KEY;
4240                         key.objectid++;
4241                 } else {
4242                         break;
4243                 }
4244                 cond_resched();
4245         }
4246
4247 out:
4248         btrfs_free_path(path);
4249         if (trans && !IS_ERR(trans))
4250                 btrfs_end_transaction(trans);
4251         if (ret)
4252                 btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret);
4253         else
4254                 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
4255         up(&fs_info->uuid_tree_rescan_sem);
4256         return 0;
4257 }
4258
4259 /*
4260  * Callback for btrfs_uuid_tree_iterate().
4261  * returns:
4262  * 0    check succeeded, the entry is not outdated.
4263  * < 0  if an error occurred.
4264  * > 0  if the check failed, which means the caller shall remove the entry.
4265  */
4266 static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info *fs_info,
4267                                        u8 *uuid, u8 type, u64 subid)
4268 {
4269         struct btrfs_key key;
4270         int ret = 0;
4271         struct btrfs_root *subvol_root;
4272
4273         if (type != BTRFS_UUID_KEY_SUBVOL &&
4274             type != BTRFS_UUID_KEY_RECEIVED_SUBVOL)
4275                 goto out;
4276
4277         key.objectid = subid;
4278         key.type = BTRFS_ROOT_ITEM_KEY;
4279         key.offset = (u64)-1;
4280         subvol_root = btrfs_read_fs_root_no_name(fs_info, &key);
4281         if (IS_ERR(subvol_root)) {
4282                 ret = PTR_ERR(subvol_root);
4283                 if (ret == -ENOENT)
4284                         ret = 1;
4285                 goto out;
4286         }
4287
4288         switch (type) {
4289         case BTRFS_UUID_KEY_SUBVOL:
4290                 if (memcmp(uuid, subvol_root->root_item.uuid, BTRFS_UUID_SIZE))
4291                         ret = 1;
4292                 break;
4293         case BTRFS_UUID_KEY_RECEIVED_SUBVOL:
4294                 if (memcmp(uuid, subvol_root->root_item.received_uuid,
4295                            BTRFS_UUID_SIZE))
4296                         ret = 1;
4297                 break;
4298         }
4299
4300 out:
4301         return ret;
4302 }
4303
4304 static int btrfs_uuid_rescan_kthread(void *data)
4305 {
4306         struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data;
4307         int ret;
4308
4309         /*
4310          * 1st step is to iterate through the existing UUID tree and
4311          * to delete all entries that contain outdated data.
4312          * 2nd step is to add all missing entries to the UUID tree.
4313          */
4314         ret = btrfs_uuid_tree_iterate(fs_info, btrfs_check_uuid_tree_entry);
4315         if (ret < 0) {
4316                 btrfs_warn(fs_info, "iterating uuid_tree failed %d", ret);
4317                 up(&fs_info->uuid_tree_rescan_sem);
4318                 return ret;
4319         }
4320         return btrfs_uuid_scan_kthread(data);
4321 }
4322
4323 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info)
4324 {
4325         struct btrfs_trans_handle *trans;
4326         struct btrfs_root *tree_root = fs_info->tree_root;
4327         struct btrfs_root *uuid_root;
4328         struct task_struct *task;
4329         int ret;
4330
4331         /*
4332          * 1 - root node
4333          * 1 - root item
4334          */
4335         trans = btrfs_start_transaction(tree_root, 2);
4336         if (IS_ERR(trans))
4337                 return PTR_ERR(trans);
4338
4339         uuid_root = btrfs_create_tree(trans, fs_info,
4340                                       BTRFS_UUID_TREE_OBJECTID);
4341         if (IS_ERR(uuid_root)) {
4342                 ret = PTR_ERR(uuid_root);
4343                 btrfs_abort_transaction(trans, ret);
4344                 btrfs_end_transaction(trans);
4345                 return ret;
4346         }
4347
4348         fs_info->uuid_root = uuid_root;
4349
4350         ret = btrfs_commit_transaction(trans);
4351         if (ret)
4352                 return ret;
4353
4354         down(&fs_info->uuid_tree_rescan_sem);
4355         task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid");
4356         if (IS_ERR(task)) {
4357                 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4358                 btrfs_warn(fs_info, "failed to start uuid_scan task");
4359                 up(&fs_info->uuid_tree_rescan_sem);
4360                 return PTR_ERR(task);
4361         }
4362
4363         return 0;
4364 }
4365
4366 int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
4367 {
4368         struct task_struct *task;
4369
4370         down(&fs_info->uuid_tree_rescan_sem);
4371         task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
4372         if (IS_ERR(task)) {
4373                 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4374                 btrfs_warn(fs_info, "failed to start uuid_rescan task");
4375                 up(&fs_info->uuid_tree_rescan_sem);
4376                 return PTR_ERR(task);
4377         }
4378
4379         return 0;
4380 }
4381
4382 /*
4383  * shrinking a device means finding all of the device extents past
4384  * the new size, and then following the back refs to the chunks.
4385  * The chunk relocation code actually frees the device extent
4386  */
4387 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
4388 {
4389         struct btrfs_fs_info *fs_info = device->fs_info;
4390         struct btrfs_root *root = fs_info->dev_root;
4391         struct btrfs_trans_handle *trans;
4392         struct btrfs_dev_extent *dev_extent = NULL;
4393         struct btrfs_path *path;
4394         u64 length;
4395         u64 chunk_offset;
4396         int ret;
4397         int slot;
4398         int failed = 0;
4399         bool retried = false;
4400         bool checked_pending_chunks = false;
4401         struct extent_buffer *l;
4402         struct btrfs_key key;
4403         struct btrfs_super_block *super_copy = fs_info->super_copy;
4404         u64 old_total = btrfs_super_total_bytes(super_copy);
4405         u64 old_size = btrfs_device_get_total_bytes(device);
4406         u64 diff;
4407
4408         new_size = round_down(new_size, fs_info->sectorsize);
4409         diff = old_size - new_size;
4410
4411         if (device->is_tgtdev_for_dev_replace)
4412                 return -EINVAL;
4413
4414         path = btrfs_alloc_path();
4415         if (!path)
4416                 return -ENOMEM;
4417
4418         path->reada = READA_FORWARD;
4419
4420         mutex_lock(&fs_info->chunk_mutex);
4421
4422         btrfs_device_set_total_bytes(device, new_size);
4423         if (device->writeable) {
4424                 device->fs_devices->total_rw_bytes -= diff;
4425                 atomic64_sub(diff, &fs_info->free_chunk_space);
4426         }
4427         mutex_unlock(&fs_info->chunk_mutex);
4428
4429 again:
4430         key.objectid = device->devid;
4431         key.offset = (u64)-1;
4432         key.type = BTRFS_DEV_EXTENT_KEY;
4433
4434         do {
4435                 mutex_lock(&fs_info->delete_unused_bgs_mutex);
4436                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4437                 if (ret < 0) {
4438                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4439                         goto done;
4440                 }
4441
4442                 ret = btrfs_previous_item(root, path, 0, key.type);
4443                 if (ret)
4444                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4445                 if (ret < 0)
4446                         goto done;
4447                 if (ret) {
4448                         ret = 0;
4449                         btrfs_release_path(path);
4450                         break;
4451                 }
4452
4453                 l = path->nodes[0];
4454                 slot = path->slots[0];
4455                 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
4456
4457                 if (key.objectid != device->devid) {
4458                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4459                         btrfs_release_path(path);
4460                         break;
4461                 }
4462
4463                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
4464                 length = btrfs_dev_extent_length(l, dev_extent);
4465
4466                 if (key.offset + length <= new_size) {
4467                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4468                         btrfs_release_path(path);
4469                         break;
4470                 }
4471
4472                 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
4473                 btrfs_release_path(path);
4474
4475                 ret = btrfs_relocate_chunk(fs_info, chunk_offset);
4476                 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4477                 if (ret && ret != -ENOSPC)
4478                         goto done;
4479                 if (ret == -ENOSPC)
4480                         failed++;
4481         } while (key.offset-- > 0);
4482
4483         if (failed && !retried) {
4484                 failed = 0;
4485                 retried = true;
4486                 goto again;
4487         } else if (failed && retried) {
4488                 ret = -ENOSPC;
4489                 goto done;
4490         }
4491
4492         /* Shrinking succeeded, else we would be at "done". */
4493         trans = btrfs_start_transaction(root, 0);
4494         if (IS_ERR(trans)) {
4495                 ret = PTR_ERR(trans);
4496                 goto done;
4497         }
4498
4499         mutex_lock(&fs_info->chunk_mutex);
4500
4501         /*
4502          * We checked in the above loop all device extents that were already in
4503          * the device tree. However before we have updated the device's
4504          * total_bytes to the new size, we might have had chunk allocations that
4505          * have not complete yet (new block groups attached to transaction
4506          * handles), and therefore their device extents were not yet in the
4507          * device tree and we missed them in the loop above. So if we have any
4508          * pending chunk using a device extent that overlaps the device range
4509          * that we can not use anymore, commit the current transaction and
4510          * repeat the search on the device tree - this way we guarantee we will
4511          * not have chunks using device extents that end beyond 'new_size'.
4512          */
4513         if (!checked_pending_chunks) {
4514                 u64 start = new_size;
4515                 u64 len = old_size - new_size;
4516
4517                 if (contains_pending_extent(trans->transaction, device,
4518                                             &start, len)) {
4519                         mutex_unlock(&fs_info->chunk_mutex);
4520                         checked_pending_chunks = true;
4521                         failed = 0;
4522                         retried = false;
4523                         ret = btrfs_commit_transaction(trans);
4524                         if (ret)
4525                                 goto done;
4526                         goto again;
4527                 }
4528         }
4529
4530         btrfs_device_set_disk_total_bytes(device, new_size);
4531         if (list_empty(&device->resized_list))
4532                 list_add_tail(&device->resized_list,
4533                               &fs_info->fs_devices->resized_devices);
4534
4535         WARN_ON(diff > old_total);
4536         btrfs_set_super_total_bytes(super_copy,
4537                         round_down(old_total - diff, fs_info->sectorsize));
4538         mutex_unlock(&fs_info->chunk_mutex);
4539
4540         /* Now btrfs_update_device() will change the on-disk size. */
4541         ret = btrfs_update_device(trans, device);
4542         btrfs_end_transaction(trans);
4543 done:
4544         btrfs_free_path(path);
4545         if (ret) {
4546                 mutex_lock(&fs_info->chunk_mutex);
4547                 btrfs_device_set_total_bytes(device, old_size);
4548                 if (device->writeable)
4549                         device->fs_devices->total_rw_bytes += diff;
4550                 atomic64_add(diff, &fs_info->free_chunk_space);
4551                 mutex_unlock(&fs_info->chunk_mutex);
4552         }
4553         return ret;
4554 }
4555
4556 static int btrfs_add_system_chunk(struct btrfs_fs_info *fs_info,
4557                            struct btrfs_key *key,
4558                            struct btrfs_chunk *chunk, int item_size)
4559 {
4560         struct btrfs_super_block *super_copy = fs_info->super_copy;
4561         struct btrfs_disk_key disk_key;
4562         u32 array_size;
4563         u8 *ptr;
4564
4565         mutex_lock(&fs_info->chunk_mutex);
4566         array_size = btrfs_super_sys_array_size(super_copy);
4567         if (array_size + item_size + sizeof(disk_key)
4568                         > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
4569                 mutex_unlock(&fs_info->chunk_mutex);
4570                 return -EFBIG;
4571         }
4572
4573         ptr = super_copy->sys_chunk_array + array_size;
4574         btrfs_cpu_key_to_disk(&disk_key, key);
4575         memcpy(ptr, &disk_key, sizeof(disk_key));
4576         ptr += sizeof(disk_key);
4577         memcpy(ptr, chunk, item_size);
4578         item_size += sizeof(disk_key);
4579         btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
4580         mutex_unlock(&fs_info->chunk_mutex);
4581
4582         return 0;
4583 }
4584
4585 /*
4586  * sort the devices in descending order by max_avail, total_avail
4587  */
4588 static int btrfs_cmp_device_info(const void *a, const void *b)
4589 {
4590         const struct btrfs_device_info *di_a = a;
4591         const struct btrfs_device_info *di_b = b;
4592
4593         if (di_a->max_avail > di_b->max_avail)
4594                 return -1;
4595         if (di_a->max_avail < di_b->max_avail)
4596                 return 1;
4597         if (di_a->total_avail > di_b->total_avail)
4598                 return -1;
4599         if (di_a->total_avail < di_b->total_avail)
4600                 return 1;
4601         return 0;
4602 }
4603
4604 static u32 find_raid56_stripe_len(u32 data_devices, u32 dev_stripe_target)
4605 {
4606         /* TODO allow them to set a preferred stripe size */
4607         return SZ_64K;
4608 }
4609
4610 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
4611 {
4612         if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK))
4613                 return;
4614
4615         btrfs_set_fs_incompat(info, RAID56);
4616 }
4617
4618 #define BTRFS_MAX_DEVS(r) ((BTRFS_MAX_ITEM_SIZE(r->fs_info)             \
4619                         - sizeof(struct btrfs_chunk))           \
4620                         / sizeof(struct btrfs_stripe) + 1)
4621
4622 #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE        \
4623                                 - 2 * sizeof(struct btrfs_disk_key)     \
4624                                 - 2 * sizeof(struct btrfs_chunk))       \
4625                                 / sizeof(struct btrfs_stripe) + 1)
4626
4627 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4628                                u64 start, u64 type)
4629 {
4630         struct btrfs_fs_info *info = trans->fs_info;
4631         struct btrfs_fs_devices *fs_devices = info->fs_devices;
4632         struct list_head *cur;
4633         struct map_lookup *map = NULL;
4634         struct extent_map_tree *em_tree;
4635         struct extent_map *em;
4636         struct btrfs_device_info *devices_info = NULL;
4637         u64 total_avail;
4638         int num_stripes;        /* total number of stripes to allocate */
4639         int data_stripes;       /* number of stripes that count for
4640                                    block group size */
4641         int sub_stripes;        /* sub_stripes info for map */
4642         int dev_stripes;        /* stripes per dev */
4643         int devs_max;           /* max devs to use */
4644         int devs_min;           /* min devs needed */
4645         int devs_increment;     /* ndevs has to be a multiple of this */
4646         int ncopies;            /* how many copies to data has */
4647         int ret;
4648         u64 max_stripe_size;
4649         u64 max_chunk_size;
4650         u64 stripe_size;
4651         u64 num_bytes;
4652         u64 raid_stripe_len = BTRFS_STRIPE_LEN;
4653         int ndevs;
4654         int i;
4655         int j;
4656         int index;
4657
4658         BUG_ON(!alloc_profile_is_valid(type, 0));
4659
4660         if (list_empty(&fs_devices->alloc_list))
4661                 return -ENOSPC;
4662
4663         index = __get_raid_index(type);
4664
4665         sub_stripes = btrfs_raid_array[index].sub_stripes;
4666         dev_stripes = btrfs_raid_array[index].dev_stripes;
4667         devs_max = btrfs_raid_array[index].devs_max;
4668         devs_min = btrfs_raid_array[index].devs_min;
4669         devs_increment = btrfs_raid_array[index].devs_increment;
4670         ncopies = btrfs_raid_array[index].ncopies;
4671
4672         if (type & BTRFS_BLOCK_GROUP_DATA) {
4673                 max_stripe_size = SZ_1G;
4674                 max_chunk_size = 10 * max_stripe_size;
4675                 if (!devs_max)
4676                         devs_max = BTRFS_MAX_DEVS(info->chunk_root);
4677         } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
4678                 /* for larger filesystems, use larger metadata chunks */
4679                 if (fs_devices->total_rw_bytes > 50ULL * SZ_1G)
4680                         max_stripe_size = SZ_1G;
4681                 else
4682                         max_stripe_size = SZ_256M;
4683                 max_chunk_size = max_stripe_size;
4684                 if (!devs_max)
4685                         devs_max = BTRFS_MAX_DEVS(info->chunk_root);
4686         } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
4687                 max_stripe_size = SZ_32M;
4688                 max_chunk_size = 2 * max_stripe_size;
4689                 if (!devs_max)
4690                         devs_max = BTRFS_MAX_DEVS_SYS_CHUNK;
4691         } else {
4692                 btrfs_err(info, "invalid chunk type 0x%llx requested",
4693                        type);
4694                 BUG_ON(1);
4695         }
4696
4697         /* we don't want a chunk larger than 10% of writeable space */
4698         max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
4699                              max_chunk_size);
4700
4701         devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info),
4702                                GFP_NOFS);
4703         if (!devices_info)
4704                 return -ENOMEM;
4705
4706         cur = fs_devices->alloc_list.next;
4707
4708         /*
4709          * in the first pass through the devices list, we gather information
4710          * about the available holes on each device.
4711          */
4712         ndevs = 0;
4713         while (cur != &fs_devices->alloc_list) {
4714                 struct btrfs_device *device;
4715                 u64 max_avail;
4716                 u64 dev_offset;
4717
4718                 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
4719
4720                 cur = cur->next;
4721
4722                 if (!device->writeable) {
4723                         WARN(1, KERN_ERR
4724                                "BTRFS: read-only device in alloc_list\n");
4725                         continue;
4726                 }
4727
4728                 if (!device->in_fs_metadata ||
4729                     device->is_tgtdev_for_dev_replace)
4730                         continue;
4731
4732                 if (device->total_bytes > device->bytes_used)
4733                         total_avail = device->total_bytes - device->bytes_used;
4734                 else
4735                         total_avail = 0;
4736
4737                 /* If there is no space on this device, skip it. */
4738                 if (total_avail == 0)
4739                         continue;
4740
4741                 ret = find_free_dev_extent(trans, device,
4742                                            max_stripe_size * dev_stripes,
4743                                            &dev_offset, &max_avail);
4744                 if (ret && ret != -ENOSPC)
4745                         goto error;
4746
4747                 if (ret == 0)
4748                         max_avail = max_stripe_size * dev_stripes;
4749
4750                 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes)
4751                         continue;
4752
4753                 if (ndevs == fs_devices->rw_devices) {
4754                         WARN(1, "%s: found more than %llu devices\n",
4755                              __func__, fs_devices->rw_devices);
4756                         break;
4757                 }
4758                 devices_info[ndevs].dev_offset = dev_offset;
4759                 devices_info[ndevs].max_avail = max_avail;
4760                 devices_info[ndevs].total_avail = total_avail;
4761                 devices_info[ndevs].dev = device;
4762                 ++ndevs;
4763         }
4764
4765         /*
4766          * now sort the devices by hole size / available space
4767          */
4768         sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
4769              btrfs_cmp_device_info, NULL);
4770
4771         /* round down to number of usable stripes */
4772         ndevs -= ndevs % devs_increment;
4773
4774         if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) {
4775                 ret = -ENOSPC;
4776                 goto error;
4777         }
4778
4779         if (devs_max && ndevs > devs_max)
4780                 ndevs = devs_max;
4781         /*
4782          * the primary goal is to maximize the number of stripes, so use as many
4783          * devices as possible, even if the stripes are not maximum sized.
4784          */
4785         stripe_size = devices_info[ndevs-1].max_avail;
4786         num_stripes = ndevs * dev_stripes;
4787
4788         /*
4789          * this will have to be fixed for RAID1 and RAID10 over
4790          * more drives
4791          */
4792         data_stripes = num_stripes / ncopies;
4793
4794         if (type & BTRFS_BLOCK_GROUP_RAID5) {
4795                 raid_stripe_len = find_raid56_stripe_len(ndevs - 1,
4796                                                          info->stripesize);
4797                 data_stripes = num_stripes - 1;
4798         }
4799         if (type & BTRFS_BLOCK_GROUP_RAID6) {
4800                 raid_stripe_len = find_raid56_stripe_len(ndevs - 2,
4801                                                          info->stripesize);
4802                 data_stripes = num_stripes - 2;
4803         }
4804
4805         /*
4806          * Use the number of data stripes to figure out how big this chunk
4807          * is really going to be in terms of logical address space,
4808          * and compare that answer with the max chunk size
4809          */
4810         if (stripe_size * data_stripes > max_chunk_size) {
4811                 u64 mask = (1ULL << 24) - 1;
4812
4813                 stripe_size = div_u64(max_chunk_size, data_stripes);
4814
4815                 /* bump the answer up to a 16MB boundary */
4816                 stripe_size = (stripe_size + mask) & ~mask;
4817
4818                 /* but don't go higher than the limits we found
4819                  * while searching for free extents
4820                  */
4821                 if (stripe_size > devices_info[ndevs-1].max_avail)
4822                         stripe_size = devices_info[ndevs-1].max_avail;
4823         }
4824
4825         stripe_size = div_u64(stripe_size, dev_stripes);
4826
4827         /* align to BTRFS_STRIPE_LEN */
4828         stripe_size = div64_u64(stripe_size, raid_stripe_len);
4829         stripe_size *= raid_stripe_len;
4830
4831         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
4832         if (!map) {
4833                 ret = -ENOMEM;
4834                 goto error;
4835         }
4836         map->num_stripes = num_stripes;
4837
4838         for (i = 0; i < ndevs; ++i) {
4839                 for (j = 0; j < dev_stripes; ++j) {
4840                         int s = i * dev_stripes + j;
4841                         map->stripes[s].dev = devices_info[i].dev;
4842                         map->stripes[s].physical = devices_info[i].dev_offset +
4843                                                    j * stripe_size;
4844                 }
4845         }
4846         map->sector_size = info->sectorsize;
4847         map->stripe_len = raid_stripe_len;
4848         map->io_align = raid_stripe_len;
4849         map->io_width = raid_stripe_len;
4850         map->type = type;
4851         map->sub_stripes = sub_stripes;
4852
4853         num_bytes = stripe_size * data_stripes;
4854
4855         trace_btrfs_chunk_alloc(info, map, start, num_bytes);
4856
4857         em = alloc_extent_map();
4858         if (!em) {
4859                 kfree(map);
4860                 ret = -ENOMEM;
4861                 goto error;
4862         }
4863         set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
4864         em->map_lookup = map;
4865         em->start = start;
4866         em->len = num_bytes;
4867         em->block_start = 0;
4868         em->block_len = em->len;
4869         em->orig_block_len = stripe_size;
4870
4871         em_tree = &info->mapping_tree.map_tree;
4872         write_lock(&em_tree->lock);
4873         ret = add_extent_mapping(em_tree, em, 0);
4874         if (!ret) {
4875                 list_add_tail(&em->list, &trans->transaction->pending_chunks);
4876                 refcount_inc(&em->refs);
4877         }
4878         write_unlock(&em_tree->lock);
4879         if (ret) {
4880                 free_extent_map(em);
4881                 goto error;
4882         }
4883
4884         ret = btrfs_make_block_group(trans, info, 0, type,
4885                                      BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4886                                      start, num_bytes);
4887         if (ret)
4888                 goto error_del_extent;
4889
4890         for (i = 0; i < map->num_stripes; i++) {
4891                 num_bytes = map->stripes[i].dev->bytes_used + stripe_size;
4892                 btrfs_device_set_bytes_used(map->stripes[i].dev, num_bytes);
4893         }
4894
4895         atomic64_sub(stripe_size * map->num_stripes, &info->free_chunk_space);
4896
4897         free_extent_map(em);
4898         check_raid56_incompat_flag(info, type);
4899
4900         kfree(devices_info);
4901         return 0;
4902
4903 error_del_extent:
4904         write_lock(&em_tree->lock);
4905         remove_extent_mapping(em_tree, em);
4906         write_unlock(&em_tree->lock);
4907
4908         /* One for our allocation */
4909         free_extent_map(em);
4910         /* One for the tree reference */
4911         free_extent_map(em);
4912         /* One for the pending_chunks list reference */
4913         free_extent_map(em);
4914 error:
4915         kfree(devices_info);
4916         return ret;
4917 }
4918
4919 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans,
4920                                 struct btrfs_fs_info *fs_info,
4921                                 u64 chunk_offset, u64 chunk_size)
4922 {
4923         struct btrfs_root *extent_root = fs_info->extent_root;
4924         struct btrfs_root *chunk_root = fs_info->chunk_root;
4925         struct btrfs_key key;
4926         struct btrfs_device *device;
4927         struct btrfs_chunk *chunk;
4928         struct btrfs_stripe *stripe;
4929         struct extent_map *em;
4930         struct map_lookup *map;
4931         size_t item_size;
4932         u64 dev_offset;
4933         u64 stripe_size;
4934         int i = 0;
4935         int ret = 0;
4936
4937         em = get_chunk_map(fs_info, chunk_offset, chunk_size);
4938         if (IS_ERR(em))
4939                 return PTR_ERR(em);
4940
4941         map = em->map_lookup;
4942         item_size = btrfs_chunk_item_size(map->num_stripes);
4943         stripe_size = em->orig_block_len;
4944
4945         chunk = kzalloc(item_size, GFP_NOFS);
4946         if (!chunk) {
4947                 ret = -ENOMEM;
4948                 goto out;
4949         }
4950
4951         /*
4952          * Take the device list mutex to prevent races with the final phase of
4953          * a device replace operation that replaces the device object associated
4954          * with the map's stripes, because the device object's id can change
4955          * at any time during that final phase of the device replace operation
4956          * (dev-replace.c:btrfs_dev_replace_finishing()).
4957          */
4958         mutex_lock(&fs_info->fs_devices->device_list_mutex);
4959         for (i = 0; i < map->num_stripes; i++) {
4960                 device = map->stripes[i].dev;
4961                 dev_offset = map->stripes[i].physical;
4962
4963                 ret = btrfs_update_device(trans, device);
4964                 if (ret)
4965                         break;
4966                 ret = btrfs_alloc_dev_extent(trans, device,
4967                                              chunk_root->root_key.objectid,
4968                                              BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4969                                              chunk_offset, dev_offset,
4970                                              stripe_size);
4971                 if (ret)
4972                         break;
4973         }
4974         if (ret) {
4975                 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4976                 goto out;
4977         }
4978
4979         stripe = &chunk->stripe;
4980         for (i = 0; i < map->num_stripes; i++) {
4981                 device = map->stripes[i].dev;
4982                 dev_offset = map->stripes[i].physical;
4983
4984                 btrfs_set_stack_stripe_devid(stripe, device->devid);
4985                 btrfs_set_stack_stripe_offset(stripe, dev_offset);
4986                 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
4987                 stripe++;
4988         }
4989         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4990
4991         btrfs_set_stack_chunk_length(chunk, chunk_size);
4992         btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
4993         btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
4994         btrfs_set_stack_chunk_type(chunk, map->type);
4995         btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
4996         btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
4997         btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
4998         btrfs_set_stack_chunk_sector_size(chunk, fs_info->sectorsize);
4999         btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
5000
5001         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
5002         key.type = BTRFS_CHUNK_ITEM_KEY;
5003         key.offset = chunk_offset;
5004
5005         ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
5006         if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
5007                 /*
5008                  * TODO: Cleanup of inserted chunk root in case of
5009                  * failure.
5010                  */
5011                 ret = btrfs_add_system_chunk(fs_info, &key, chunk, item_size);
5012         }
5013
5014 out:
5015         kfree(chunk);
5016         free_extent_map(em);
5017         return ret;
5018 }
5019
5020 /*
5021  * Chunk allocation falls into two parts. The first part does works
5022  * that make the new allocated chunk useable, but not do any operation
5023  * that modifies the chunk tree. The second part does the works that
5024  * require modifying the chunk tree. This division is important for the
5025  * bootstrap process of adding storage to a seed btrfs.
5026  */
5027 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
5028                       struct btrfs_fs_info *fs_info, u64 type)
5029 {
5030         u64 chunk_offset;
5031
5032         ASSERT(mutex_is_locked(&fs_info->chunk_mutex));
5033         chunk_offset = find_next_chunk(fs_info);
5034         return __btrfs_alloc_chunk(trans, chunk_offset, type);
5035 }
5036
5037 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
5038                                          struct btrfs_fs_info *fs_info)
5039 {
5040         u64 chunk_offset;
5041         u64 sys_chunk_offset;
5042         u64 alloc_profile;
5043         int ret;
5044
5045         chunk_offset = find_next_chunk(fs_info);
5046         alloc_profile = btrfs_metadata_alloc_profile(fs_info);
5047         ret = __btrfs_alloc_chunk(trans, chunk_offset, alloc_profile);
5048         if (ret)
5049                 return ret;
5050
5051         sys_chunk_offset = find_next_chunk(fs_info);
5052         alloc_profile = btrfs_system_alloc_profile(fs_info);
5053         ret = __btrfs_alloc_chunk(trans, sys_chunk_offset, alloc_profile);
5054         return ret;
5055 }
5056
5057 static inline int btrfs_chunk_max_errors(struct map_lookup *map)
5058 {
5059         int max_errors;
5060
5061         if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
5062                          BTRFS_BLOCK_GROUP_RAID10 |
5063                          BTRFS_BLOCK_GROUP_RAID5 |
5064                          BTRFS_BLOCK_GROUP_DUP)) {
5065                 max_errors = 1;
5066         } else if (map->type & BTRFS_BLOCK_GROUP_RAID6) {
5067                 max_errors = 2;
5068         } else {
5069                 max_errors = 0;
5070         }
5071
5072         return max_errors;
5073 }
5074
5075 int btrfs_chunk_readonly(struct btrfs_fs_info *fs_info, u64 chunk_offset)
5076 {
5077         struct extent_map *em;
5078         struct map_lookup *map;
5079         int readonly = 0;
5080         int miss_ndevs = 0;
5081         int i;
5082
5083         em = get_chunk_map(fs_info, chunk_offset, 1);
5084         if (IS_ERR(em))
5085                 return 1;
5086
5087         map = em->map_lookup;
5088         for (i = 0; i < map->num_stripes; i++) {
5089                 if (map->stripes[i].dev->missing) {
5090                         miss_ndevs++;
5091                         continue;
5092                 }
5093
5094                 if (!map->stripes[i].dev->writeable) {
5095                         readonly = 1;
5096                         goto end;
5097                 }
5098         }
5099
5100         /*
5101          * If the number of missing devices is larger than max errors,
5102          * we can not write the data into that chunk successfully, so
5103          * set it readonly.
5104          */
5105         if (miss_ndevs > btrfs_chunk_max_errors(map))
5106                 readonly = 1;
5107 end:
5108         free_extent_map(em);
5109         return readonly;
5110 }
5111
5112 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
5113 {
5114         extent_map_tree_init(&tree->map_tree);
5115 }
5116
5117 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
5118 {
5119         struct extent_map *em;
5120
5121         while (1) {
5122                 write_lock(&tree->map_tree.lock);
5123                 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
5124                 if (em)
5125                         remove_extent_mapping(&tree->map_tree, em);
5126                 write_unlock(&tree->map_tree.lock);
5127                 if (!em)
5128                         break;
5129                 /* once for us */
5130                 free_extent_map(em);
5131                 /* once for the tree */
5132                 free_extent_map(em);
5133         }
5134 }
5135
5136 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5137 {
5138         struct extent_map *em;
5139         struct map_lookup *map;
5140         int ret;
5141
5142         em = get_chunk_map(fs_info, logical, len);
5143         if (IS_ERR(em))
5144                 /*
5145                  * We could return errors for these cases, but that could get
5146                  * ugly and we'd probably do the same thing which is just not do
5147                  * anything else and exit, so return 1 so the callers don't try
5148                  * to use other copies.
5149                  */
5150                 return 1;
5151
5152         map = em->map_lookup;
5153         if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
5154                 ret = map->num_stripes;
5155         else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5156                 ret = map->sub_stripes;
5157         else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
5158                 ret = 2;
5159         else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5160                 ret = 3;
5161         else
5162                 ret = 1;
5163         free_extent_map(em);
5164
5165         btrfs_dev_replace_lock(&fs_info->dev_replace, 0);
5166         if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace) &&
5167             fs_info->dev_replace.tgtdev)
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_fs_info *fs_info,
5175                                     struct btrfs_mapping_tree *map_tree,
5176                                     u64 logical)
5177 {
5178         struct extent_map *em;
5179         struct map_lookup *map;
5180         unsigned long len = fs_info->sectorsize;
5181
5182         em = get_chunk_map(fs_info, logical, len);
5183         WARN_ON(IS_ERR(em));
5184
5185         map = em->map_lookup;
5186         if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5187                 len = map->stripe_len * nr_data_stripes(map);
5188         free_extent_map(em);
5189         return len;
5190 }
5191
5192 int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info,
5193                            u64 logical, u64 len, int mirror_num)
5194 {
5195         struct extent_map *em;
5196         struct map_lookup *map;
5197         int ret = 0;
5198
5199         em = get_chunk_map(fs_info, logical, len);
5200         WARN_ON(IS_ERR(em));
5201
5202         map = em->map_lookup;
5203         if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5204                 ret = 1;
5205         free_extent_map(em);
5206         return ret;
5207 }
5208
5209 static int find_live_mirror(struct btrfs_fs_info *fs_info,
5210                             struct map_lookup *map, int first, int num,
5211                             int optimal, int dev_replace_is_ongoing)
5212 {
5213         int i;
5214         int tolerance;
5215         struct btrfs_device *srcdev;
5216
5217         if (dev_replace_is_ongoing &&
5218             fs_info->dev_replace.cont_reading_from_srcdev_mode ==
5219              BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
5220                 srcdev = fs_info->dev_replace.srcdev;
5221         else
5222                 srcdev = NULL;
5223
5224         /*
5225          * try to avoid the drive that is the source drive for a
5226          * dev-replace procedure, only choose it if no other non-missing
5227          * mirror is available
5228          */
5229         for (tolerance = 0; tolerance < 2; tolerance++) {
5230                 if (map->stripes[optimal].dev->bdev &&
5231                     (tolerance || map->stripes[optimal].dev != srcdev))
5232                         return optimal;
5233                 for (i = first; i < first + num; i++) {
5234                         if (map->stripes[i].dev->bdev &&
5235                             (tolerance || map->stripes[i].dev != srcdev))
5236                                 return i;
5237                 }
5238         }
5239
5240         /* we couldn't find one that doesn't fail.  Just return something
5241          * and the io error handling code will clean up eventually
5242          */
5243         return optimal;
5244 }
5245
5246 static inline int parity_smaller(u64 a, u64 b)
5247 {
5248         return a > b;
5249 }
5250
5251 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
5252 static void sort_parity_stripes(struct btrfs_bio *bbio, int num_stripes)
5253 {
5254         struct btrfs_bio_stripe s;
5255         int i;
5256         u64 l;
5257         int again = 1;
5258
5259         while (again) {
5260                 again = 0;
5261                 for (i = 0; i < num_stripes - 1; i++) {
5262                         if (parity_smaller(bbio->raid_map[i],
5263                                            bbio->raid_map[i+1])) {
5264                                 s = bbio->stripes[i];
5265                                 l = bbio->raid_map[i];
5266                                 bbio->stripes[i] = bbio->stripes[i+1];
5267                                 bbio->raid_map[i] = bbio->raid_map[i+1];
5268                                 bbio->stripes[i+1] = s;
5269                                 bbio->raid_map[i+1] = l;
5270
5271                                 again = 1;
5272                         }
5273                 }
5274         }
5275 }
5276
5277 static struct btrfs_bio *alloc_btrfs_bio(int total_stripes, int real_stripes)
5278 {
5279         struct btrfs_bio *bbio = kzalloc(
5280                  /* the size of the btrfs_bio */
5281                 sizeof(struct btrfs_bio) +
5282                 /* plus the variable array for the stripes */
5283                 sizeof(struct btrfs_bio_stripe) * (total_stripes) +
5284                 /* plus the variable array for the tgt dev */
5285                 sizeof(int) * (real_stripes) +
5286                 /*
5287                  * plus the raid_map, which includes both the tgt dev
5288                  * and the stripes
5289                  */
5290                 sizeof(u64) * (total_stripes),
5291                 GFP_NOFS|__GFP_NOFAIL);
5292
5293         atomic_set(&bbio->error, 0);
5294         refcount_set(&bbio->refs, 1);
5295
5296         return bbio;
5297 }
5298
5299 void btrfs_get_bbio(struct btrfs_bio *bbio)
5300 {
5301         WARN_ON(!refcount_read(&bbio->refs));
5302         refcount_inc(&bbio->refs);
5303 }
5304
5305 void btrfs_put_bbio(struct btrfs_bio *bbio)
5306 {
5307         if (!bbio)
5308                 return;
5309         if (refcount_dec_and_test(&bbio->refs))
5310                 kfree(bbio);
5311 }
5312
5313 /* can REQ_OP_DISCARD be sent with other REQ like REQ_OP_WRITE? */
5314 /*
5315  * Please note that, discard won't be sent to target device of device
5316  * replace.
5317  */
5318 static int __btrfs_map_block_for_discard(struct btrfs_fs_info *fs_info,
5319                                          u64 logical, u64 length,
5320                                          struct btrfs_bio **bbio_ret)
5321 {
5322         struct extent_map *em;
5323         struct map_lookup *map;
5324         struct btrfs_bio *bbio;
5325         u64 offset;
5326         u64 stripe_nr;
5327         u64 stripe_nr_end;
5328         u64 stripe_end_offset;
5329         u64 stripe_cnt;
5330         u64 stripe_len;
5331         u64 stripe_offset;
5332         u64 num_stripes;
5333         u32 stripe_index;
5334         u32 factor = 0;
5335         u32 sub_stripes = 0;
5336         u64 stripes_per_dev = 0;
5337         u32 remaining_stripes = 0;
5338         u32 last_stripe = 0;
5339         int ret = 0;
5340         int i;
5341
5342         /* discard always return a bbio */
5343         ASSERT(bbio_ret);
5344
5345         em = get_chunk_map(fs_info, logical, length);
5346         if (IS_ERR(em))
5347                 return PTR_ERR(em);
5348
5349         map = em->map_lookup;
5350         /* we don't discard raid56 yet */
5351         if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5352                 ret = -EOPNOTSUPP;
5353                 goto out;
5354         }
5355
5356         offset = logical - em->start;
5357         length = min_t(u64, em->len - offset, length);
5358
5359         stripe_len = map->stripe_len;
5360         /*
5361          * stripe_nr counts the total number of stripes we have to stride
5362          * to get to this block
5363          */
5364         stripe_nr = div64_u64(offset, stripe_len);
5365
5366         /* stripe_offset is the offset of this block in its stripe */
5367         stripe_offset = offset - stripe_nr * stripe_len;
5368
5369         stripe_nr_end = round_up(offset + length, map->stripe_len);
5370         stripe_nr_end = div64_u64(stripe_nr_end, map->stripe_len);
5371         stripe_cnt = stripe_nr_end - stripe_nr;
5372         stripe_end_offset = stripe_nr_end * map->stripe_len -
5373                             (offset + length);
5374         /*
5375          * after this, stripe_nr is the number of stripes on this
5376          * device we have to walk to find the data, and stripe_index is
5377          * the number of our device in the stripe array
5378          */
5379         num_stripes = 1;
5380         stripe_index = 0;
5381         if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5382                          BTRFS_BLOCK_GROUP_RAID10)) {
5383                 if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5384                         sub_stripes = 1;
5385                 else
5386                         sub_stripes = map->sub_stripes;
5387
5388                 factor = map->num_stripes / sub_stripes;
5389                 num_stripes = min_t(u64, map->num_stripes,
5390                                     sub_stripes * stripe_cnt);
5391                 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
5392                 stripe_index *= sub_stripes;
5393                 stripes_per_dev = div_u64_rem(stripe_cnt, factor,
5394                                               &remaining_stripes);
5395                 div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
5396                 last_stripe *= sub_stripes;
5397         } else if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
5398                                 BTRFS_BLOCK_GROUP_DUP)) {
5399                 num_stripes = map->num_stripes;
5400         } else {
5401                 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5402                                         &stripe_index);
5403         }
5404
5405         bbio = alloc_btrfs_bio(num_stripes, 0);
5406         if (!bbio) {
5407                 ret = -ENOMEM;
5408                 goto out;
5409         }
5410
5411         for (i = 0; i < num_stripes; i++) {
5412                 bbio->stripes[i].physical =
5413                         map->stripes[stripe_index].physical +
5414                         stripe_offset + stripe_nr * map->stripe_len;
5415                 bbio->stripes[i].dev = map->stripes[stripe_index].dev;
5416
5417                 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5418                                  BTRFS_BLOCK_GROUP_RAID10)) {
5419                         bbio->stripes[i].length = stripes_per_dev *
5420                                 map->stripe_len;
5421
5422                         if (i / sub_stripes < remaining_stripes)
5423                                 bbio->stripes[i].length +=
5424                                         map->stripe_len;
5425
5426                         /*
5427                          * Special for the first stripe and
5428                          * the last stripe:
5429                          *
5430                          * |-------|...|-------|
5431                          *     |----------|
5432                          *    off     end_off
5433                          */
5434                         if (i < sub_stripes)
5435                                 bbio->stripes[i].length -=
5436                                         stripe_offset;
5437
5438                         if (stripe_index >= last_stripe &&
5439                             stripe_index <= (last_stripe +
5440                                              sub_stripes - 1))
5441                                 bbio->stripes[i].length -=
5442                                         stripe_end_offset;
5443
5444                         if (i == sub_stripes - 1)
5445                                 stripe_offset = 0;
5446                 } else {
5447                         bbio->stripes[i].length = length;
5448                 }
5449
5450                 stripe_index++;
5451                 if (stripe_index == map->num_stripes) {
5452                         stripe_index = 0;
5453                         stripe_nr++;
5454                 }
5455         }
5456
5457         *bbio_ret = bbio;
5458         bbio->map_type = map->type;
5459         bbio->num_stripes = num_stripes;
5460 out:
5461         free_extent_map(em);
5462         return ret;
5463 }
5464
5465 /*
5466  * In dev-replace case, for repair case (that's the only case where the mirror
5467  * is selected explicitly when calling btrfs_map_block), blocks left of the
5468  * left cursor can also be read from the target drive.
5469  *
5470  * For REQ_GET_READ_MIRRORS, the target drive is added as the last one to the
5471  * array of stripes.
5472  * For READ, it also needs to be supported using the same mirror number.
5473  *
5474  * If the requested block is not left of the left cursor, EIO is returned. This
5475  * can happen because btrfs_num_copies() returns one more in the dev-replace
5476  * case.
5477  */
5478 static int get_extra_mirror_from_replace(struct btrfs_fs_info *fs_info,
5479                                          u64 logical, u64 length,
5480                                          u64 srcdev_devid, int *mirror_num,
5481                                          u64 *physical)
5482 {
5483         struct btrfs_bio *bbio = NULL;
5484         int num_stripes;
5485         int index_srcdev = 0;
5486         int found = 0;
5487         u64 physical_of_found = 0;
5488         int i;
5489         int ret = 0;
5490
5491         ret = __btrfs_map_block(fs_info, BTRFS_MAP_GET_READ_MIRRORS,
5492                                 logical, &length, &bbio, 0, 0);
5493         if (ret) {
5494                 ASSERT(bbio == NULL);
5495                 return ret;
5496         }
5497
5498         num_stripes = bbio->num_stripes;
5499         if (*mirror_num > num_stripes) {
5500                 /*
5501                  * BTRFS_MAP_GET_READ_MIRRORS does not contain this mirror,
5502                  * that means that the requested area is not left of the left
5503                  * cursor
5504                  */
5505                 btrfs_put_bbio(bbio);
5506                 return -EIO;
5507         }
5508
5509         /*
5510          * process the rest of the function using the mirror_num of the source
5511          * drive. Therefore look it up first.  At the end, patch the device
5512          * pointer to the one of the target drive.
5513          */
5514         for (i = 0; i < num_stripes; i++) {
5515                 if (bbio->stripes[i].dev->devid != srcdev_devid)
5516                         continue;
5517
5518                 /*
5519                  * In case of DUP, in order to keep it simple, only add the
5520                  * mirror with the lowest physical address
5521                  */
5522                 if (found &&
5523                     physical_of_found <= bbio->stripes[i].physical)
5524                         continue;
5525
5526                 index_srcdev = i;
5527                 found = 1;
5528                 physical_of_found = bbio->stripes[i].physical;
5529         }
5530
5531         btrfs_put_bbio(bbio);
5532
5533         ASSERT(found);
5534         if (!found)
5535                 return -EIO;
5536
5537         *mirror_num = index_srcdev + 1;
5538         *physical = physical_of_found;
5539         return ret;
5540 }
5541
5542 static void handle_ops_on_dev_replace(enum btrfs_map_op op,
5543                                       struct btrfs_bio **bbio_ret,
5544                                       struct btrfs_dev_replace *dev_replace,
5545                                       int *num_stripes_ret, int *max_errors_ret)
5546 {
5547         struct btrfs_bio *bbio = *bbio_ret;
5548         u64 srcdev_devid = dev_replace->srcdev->devid;
5549         int tgtdev_indexes = 0;
5550         int num_stripes = *num_stripes_ret;
5551         int max_errors = *max_errors_ret;
5552         int i;
5553
5554         if (op == BTRFS_MAP_WRITE) {
5555                 int index_where_to_add;
5556
5557                 /*
5558                  * duplicate the write operations while the dev replace
5559                  * procedure is running. Since the copying of the old disk to
5560                  * the new disk takes place at run time while the filesystem is
5561                  * mounted writable, the regular write operations to the old
5562                  * disk have to be duplicated to go to the new disk as well.
5563                  *
5564                  * Note that device->missing is handled by the caller, and that
5565                  * the write to the old disk is already set up in the stripes
5566                  * array.
5567                  */
5568                 index_where_to_add = num_stripes;
5569                 for (i = 0; i < num_stripes; i++) {
5570                         if (bbio->stripes[i].dev->devid == srcdev_devid) {
5571                                 /* write to new disk, too */
5572                                 struct btrfs_bio_stripe *new =
5573                                         bbio->stripes + index_where_to_add;
5574                                 struct btrfs_bio_stripe *old =
5575                                         bbio->stripes + i;
5576
5577                                 new->physical = old->physical;
5578                                 new->length = old->length;
5579                                 new->dev = dev_replace->tgtdev;
5580                                 bbio->tgtdev_map[i] = index_where_to_add;
5581                                 index_where_to_add++;
5582                                 max_errors++;
5583                                 tgtdev_indexes++;
5584                         }
5585                 }
5586                 num_stripes = index_where_to_add;
5587         } else if (op == BTRFS_MAP_GET_READ_MIRRORS) {
5588                 int index_srcdev = 0;
5589                 int found = 0;
5590                 u64 physical_of_found = 0;
5591
5592                 /*
5593                  * During the dev-replace procedure, the target drive can also
5594                  * be used to read data in case it is needed to repair a corrupt
5595                  * block elsewhere. This is possible if the requested area is
5596                  * left of the left cursor. In this area, the target drive is a
5597                  * full copy of the source drive.
5598                  */
5599                 for (i = 0; i < num_stripes; i++) {
5600                         if (bbio->stripes[i].dev->devid == srcdev_devid) {
5601                                 /*
5602                                  * In case of DUP, in order to keep it simple,
5603                                  * only add the mirror with the lowest physical
5604                                  * address
5605                                  */
5606                                 if (found &&
5607                                     physical_of_found <=
5608                                      bbio->stripes[i].physical)
5609                                         continue;
5610                                 index_srcdev = i;
5611                                 found = 1;
5612                                 physical_of_found = bbio->stripes[i].physical;
5613                         }
5614                 }
5615                 if (found) {
5616                         struct btrfs_bio_stripe *tgtdev_stripe =
5617                                 bbio->stripes + num_stripes;
5618
5619                         tgtdev_stripe->physical = physical_of_found;
5620                         tgtdev_stripe->length =
5621                                 bbio->stripes[index_srcdev].length;
5622                         tgtdev_stripe->dev = dev_replace->tgtdev;
5623                         bbio->tgtdev_map[index_srcdev] = num_stripes;
5624
5625                         tgtdev_indexes++;
5626                         num_stripes++;
5627                 }
5628         }
5629
5630         *num_stripes_ret = num_stripes;
5631         *max_errors_ret = max_errors;
5632         bbio->num_tgtdevs = tgtdev_indexes;
5633         *bbio_ret = bbio;
5634 }
5635
5636 static bool need_full_stripe(enum btrfs_map_op op)
5637 {
5638         return (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS);
5639 }
5640
5641 static int __btrfs_map_block(struct btrfs_fs_info *fs_info,
5642                              enum btrfs_map_op op,
5643                              u64 logical, u64 *length,
5644                              struct btrfs_bio **bbio_ret,
5645                              int mirror_num, int need_raid_map)
5646 {
5647         struct extent_map *em;
5648         struct map_lookup *map;
5649         u64 offset;
5650         u64 stripe_offset;
5651         u64 stripe_nr;
5652         u64 stripe_len;
5653         u32 stripe_index;
5654         int i;
5655         int ret = 0;
5656         int num_stripes;
5657         int max_errors = 0;
5658         int tgtdev_indexes = 0;
5659         struct btrfs_bio *bbio = NULL;
5660         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
5661         int dev_replace_is_ongoing = 0;
5662         int num_alloc_stripes;
5663         int patch_the_first_stripe_for_dev_replace = 0;
5664         u64 physical_to_patch_in_first_stripe = 0;
5665         u64 raid56_full_stripe_start = (u64)-1;
5666
5667         if (op == BTRFS_MAP_DISCARD)
5668                 return __btrfs_map_block_for_discard(fs_info, logical,
5669                                                      *length, bbio_ret);
5670
5671         em = get_chunk_map(fs_info, logical, *length);
5672         if (IS_ERR(em))
5673                 return PTR_ERR(em);
5674
5675         map = em->map_lookup;
5676         offset = logical - em->start;
5677
5678         stripe_len = map->stripe_len;
5679         stripe_nr = offset;
5680         /*
5681          * stripe_nr counts the total number of stripes we have to stride
5682          * to get to this block
5683          */
5684         stripe_nr = div64_u64(stripe_nr, stripe_len);
5685
5686         stripe_offset = stripe_nr * stripe_len;
5687         if (offset < stripe_offset) {
5688                 btrfs_crit(fs_info,
5689                            "stripe math has gone wrong, stripe_offset=%llu, offset=%llu, start=%llu, logical=%llu, stripe_len=%llu",
5690                            stripe_offset, offset, em->start, logical,
5691                            stripe_len);
5692                 free_extent_map(em);
5693                 return -EINVAL;
5694         }
5695
5696         /* stripe_offset is the offset of this block in its stripe*/
5697         stripe_offset = offset - stripe_offset;
5698
5699         /* if we're here for raid56, we need to know the stripe aligned start */
5700         if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5701                 unsigned long full_stripe_len = stripe_len * nr_data_stripes(map);
5702                 raid56_full_stripe_start = offset;
5703
5704                 /* allow a write of a full stripe, but make sure we don't
5705                  * allow straddling of stripes
5706                  */
5707                 raid56_full_stripe_start = div64_u64(raid56_full_stripe_start,
5708                                 full_stripe_len);
5709                 raid56_full_stripe_start *= full_stripe_len;
5710         }
5711
5712         if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
5713                 u64 max_len;
5714                 /* For writes to RAID[56], allow a full stripeset across all disks.
5715                    For other RAID types and for RAID[56] reads, just allow a single
5716                    stripe (on a single disk). */
5717                 if ((map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
5718                     (op == BTRFS_MAP_WRITE)) {
5719                         max_len = stripe_len * nr_data_stripes(map) -
5720                                 (offset - raid56_full_stripe_start);
5721                 } else {
5722                         /* we limit the length of each bio to what fits in a stripe */
5723                         max_len = stripe_len - stripe_offset;
5724                 }
5725                 *length = min_t(u64, em->len - offset, max_len);
5726         } else {
5727                 *length = em->len - offset;
5728         }
5729
5730         /* This is for when we're called from btrfs_merge_bio_hook() and all
5731            it cares about is the length */
5732         if (!bbio_ret)
5733                 goto out;
5734
5735         btrfs_dev_replace_lock(dev_replace, 0);
5736         dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
5737         if (!dev_replace_is_ongoing)
5738                 btrfs_dev_replace_unlock(dev_replace, 0);
5739         else
5740                 btrfs_dev_replace_set_lock_blocking(dev_replace);
5741
5742         if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
5743             !need_full_stripe(op) && dev_replace->tgtdev != NULL) {
5744                 ret = get_extra_mirror_from_replace(fs_info, logical, *length,
5745                                                     dev_replace->srcdev->devid,
5746                                                     &mirror_num,
5747                                             &physical_to_patch_in_first_stripe);
5748                 if (ret)
5749                         goto out;
5750                 else
5751                         patch_the_first_stripe_for_dev_replace = 1;
5752         } else if (mirror_num > map->num_stripes) {
5753                 mirror_num = 0;
5754         }
5755
5756         num_stripes = 1;
5757         stripe_index = 0;
5758         if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5759                 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5760                                 &stripe_index);
5761                 if (op != BTRFS_MAP_WRITE && op != BTRFS_MAP_GET_READ_MIRRORS)
5762                         mirror_num = 1;
5763         } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
5764                 if (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS)
5765                         num_stripes = map->num_stripes;
5766                 else if (mirror_num)
5767                         stripe_index = mirror_num - 1;
5768                 else {
5769                         stripe_index = find_live_mirror(fs_info, map, 0,
5770                                             map->num_stripes,
5771                                             current->pid % map->num_stripes,
5772                                             dev_replace_is_ongoing);
5773                         mirror_num = stripe_index + 1;
5774                 }
5775
5776         } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
5777                 if (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS) {
5778                         num_stripes = map->num_stripes;
5779                 } else if (mirror_num) {
5780                         stripe_index = mirror_num - 1;
5781                 } else {
5782                         mirror_num = 1;
5783                 }
5784
5785         } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5786                 u32 factor = map->num_stripes / map->sub_stripes;
5787
5788                 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
5789                 stripe_index *= map->sub_stripes;
5790
5791                 if (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS)
5792                         num_stripes = map->sub_stripes;
5793                 else if (mirror_num)
5794                         stripe_index += mirror_num - 1;
5795                 else {
5796                         int old_stripe_index = stripe_index;
5797                         stripe_index = find_live_mirror(fs_info, map,
5798                                               stripe_index,
5799                                               map->sub_stripes, stripe_index +
5800                                               current->pid % map->sub_stripes,
5801                                               dev_replace_is_ongoing);
5802                         mirror_num = stripe_index - old_stripe_index + 1;
5803                 }
5804
5805         } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5806                 if (need_raid_map &&
5807                     (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS ||
5808                      mirror_num > 1)) {
5809                         /* push stripe_nr back to the start of the full stripe */
5810                         stripe_nr = div64_u64(raid56_full_stripe_start,
5811                                         stripe_len * nr_data_stripes(map));
5812
5813                         /* RAID[56] write or recovery. Return all stripes */
5814                         num_stripes = map->num_stripes;
5815                         max_errors = nr_parity_stripes(map);
5816
5817                         *length = map->stripe_len;
5818                         stripe_index = 0;
5819                         stripe_offset = 0;
5820                 } else {
5821                         /*
5822                          * Mirror #0 or #1 means the original data block.
5823                          * Mirror #2 is RAID5 parity block.
5824                          * Mirror #3 is RAID6 Q block.
5825                          */
5826                         stripe_nr = div_u64_rem(stripe_nr,
5827                                         nr_data_stripes(map), &stripe_index);
5828                         if (mirror_num > 1)
5829                                 stripe_index = nr_data_stripes(map) +
5830                                                 mirror_num - 2;
5831
5832                         /* We distribute the parity blocks across stripes */
5833                         div_u64_rem(stripe_nr + stripe_index, map->num_stripes,
5834                                         &stripe_index);
5835                         if ((op != BTRFS_MAP_WRITE &&
5836                              op != BTRFS_MAP_GET_READ_MIRRORS) &&
5837                             mirror_num <= 1)
5838                                 mirror_num = 1;
5839                 }
5840         } else {
5841                 /*
5842                  * after this, stripe_nr is the number of stripes on this
5843                  * device we have to walk to find the data, and stripe_index is
5844                  * the number of our device in the stripe array
5845                  */
5846                 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5847                                 &stripe_index);
5848                 mirror_num = stripe_index + 1;
5849         }
5850         if (stripe_index >= map->num_stripes) {
5851                 btrfs_crit(fs_info,
5852                            "stripe index math went horribly wrong, got stripe_index=%u, num_stripes=%u",
5853                            stripe_index, map->num_stripes);
5854                 ret = -EINVAL;
5855                 goto out;
5856         }
5857
5858         num_alloc_stripes = num_stripes;
5859         if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL) {
5860                 if (op == BTRFS_MAP_WRITE)
5861                         num_alloc_stripes <<= 1;
5862                 if (op == BTRFS_MAP_GET_READ_MIRRORS)
5863                         num_alloc_stripes++;
5864                 tgtdev_indexes = num_stripes;
5865         }
5866
5867         bbio = alloc_btrfs_bio(num_alloc_stripes, tgtdev_indexes);
5868         if (!bbio) {
5869                 ret = -ENOMEM;
5870                 goto out;
5871         }
5872         if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL)
5873                 bbio->tgtdev_map = (int *)(bbio->stripes + num_alloc_stripes);
5874
5875         /* build raid_map */
5876         if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK && need_raid_map &&
5877             (need_full_stripe(op) || mirror_num > 1)) {
5878                 u64 tmp;
5879                 unsigned rot;
5880
5881                 bbio->raid_map = (u64 *)((void *)bbio->stripes +
5882                                  sizeof(struct btrfs_bio_stripe) *
5883                                  num_alloc_stripes +
5884                                  sizeof(int) * tgtdev_indexes);
5885
5886                 /* Work out the disk rotation on this stripe-set */
5887                 div_u64_rem(stripe_nr, num_stripes, &rot);
5888
5889                 /* Fill in the logical address of each stripe */
5890                 tmp = stripe_nr * nr_data_stripes(map);
5891                 for (i = 0; i < nr_data_stripes(map); i++)
5892                         bbio->raid_map[(i+rot) % num_stripes] =
5893                                 em->start + (tmp + i) * map->stripe_len;
5894
5895                 bbio->raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE;
5896                 if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5897                         bbio->raid_map[(i+rot+1) % num_stripes] =
5898                                 RAID6_Q_STRIPE;
5899         }
5900
5901
5902         for (i = 0; i < num_stripes; i++) {
5903                 bbio->stripes[i].physical =
5904                         map->stripes[stripe_index].physical +
5905                         stripe_offset +
5906                         stripe_nr * map->stripe_len;
5907                 bbio->stripes[i].dev =
5908                         map->stripes[stripe_index].dev;
5909                 stripe_index++;
5910         }
5911
5912         if (need_full_stripe(op))
5913                 max_errors = btrfs_chunk_max_errors(map);
5914
5915         if (bbio->raid_map)
5916                 sort_parity_stripes(bbio, num_stripes);
5917
5918         if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL &&
5919             need_full_stripe(op)) {
5920                 handle_ops_on_dev_replace(op, &bbio, dev_replace, &num_stripes,
5921                                           &max_errors);
5922         }
5923
5924         *bbio_ret = bbio;
5925         bbio->map_type = map->type;
5926         bbio->num_stripes = num_stripes;
5927         bbio->max_errors = max_errors;
5928         bbio->mirror_num = mirror_num;
5929
5930         /*
5931          * this is the case that REQ_READ && dev_replace_is_ongoing &&
5932          * mirror_num == num_stripes + 1 && dev_replace target drive is
5933          * available as a mirror
5934          */
5935         if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) {
5936                 WARN_ON(num_stripes > 1);
5937                 bbio->stripes[0].dev = dev_replace->tgtdev;
5938                 bbio->stripes[0].physical = physical_to_patch_in_first_stripe;
5939                 bbio->mirror_num = map->num_stripes + 1;
5940         }
5941 out:
5942         if (dev_replace_is_ongoing) {
5943                 btrfs_dev_replace_clear_lock_blocking(dev_replace);
5944                 btrfs_dev_replace_unlock(dev_replace, 0);
5945         }
5946         free_extent_map(em);
5947         return ret;
5948 }
5949
5950 int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
5951                       u64 logical, u64 *length,
5952                       struct btrfs_bio **bbio_ret, int mirror_num)
5953 {
5954         return __btrfs_map_block(fs_info, op, logical, length, bbio_ret,
5955                                  mirror_num, 0);
5956 }
5957
5958 /* For Scrub/replace */
5959 int btrfs_map_sblock(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
5960                      u64 logical, u64 *length,
5961                      struct btrfs_bio **bbio_ret)
5962 {
5963         return __btrfs_map_block(fs_info, op, logical, length, bbio_ret, 0, 1);
5964 }
5965
5966 int btrfs_rmap_block(struct btrfs_fs_info *fs_info,
5967                      u64 chunk_start, u64 physical, u64 devid,
5968                      u64 **logical, int *naddrs, int *stripe_len)
5969 {
5970         struct extent_map *em;
5971         struct map_lookup *map;
5972         u64 *buf;
5973         u64 bytenr;
5974         u64 length;
5975         u64 stripe_nr;
5976         u64 rmap_len;
5977         int i, j, nr = 0;
5978
5979         em = get_chunk_map(fs_info, chunk_start, 1);
5980         if (IS_ERR(em))
5981                 return -EIO;
5982
5983         map = em->map_lookup;
5984         length = em->len;
5985         rmap_len = map->stripe_len;
5986
5987         if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5988                 length = div_u64(length, map->num_stripes / map->sub_stripes);
5989         else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5990                 length = div_u64(length, map->num_stripes);
5991         else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5992                 length = div_u64(length, nr_data_stripes(map));
5993                 rmap_len = map->stripe_len * nr_data_stripes(map);
5994         }
5995
5996         buf = kcalloc(map->num_stripes, sizeof(u64), GFP_NOFS);
5997         BUG_ON(!buf); /* -ENOMEM */
5998
5999         for (i = 0; i < map->num_stripes; i++) {
6000                 if (devid && map->stripes[i].dev->devid != devid)
6001                         continue;
6002                 if (map->stripes[i].physical > physical ||
6003                     map->stripes[i].physical + length <= physical)
6004                         continue;
6005
6006                 stripe_nr = physical - map->stripes[i].physical;
6007                 stripe_nr = div64_u64(stripe_nr, map->stripe_len);
6008
6009                 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
6010                         stripe_nr = stripe_nr * map->num_stripes + i;
6011                         stripe_nr = div_u64(stripe_nr, map->sub_stripes);
6012                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
6013                         stripe_nr = stripe_nr * map->num_stripes + i;
6014                 } /* else if RAID[56], multiply by nr_data_stripes().
6015                    * Alternatively, just use rmap_len below instead of
6016                    * map->stripe_len */
6017
6018                 bytenr = chunk_start + stripe_nr * rmap_len;
6019                 WARN_ON(nr >= map->num_stripes);
6020                 for (j = 0; j < nr; j++) {
6021                         if (buf[j] == bytenr)
6022                                 break;
6023                 }
6024                 if (j == nr) {
6025                         WARN_ON(nr >= map->num_stripes);
6026                         buf[nr++] = bytenr;
6027                 }
6028         }
6029
6030         *logical = buf;
6031         *naddrs = nr;
6032         *stripe_len = rmap_len;
6033
6034         free_extent_map(em);
6035         return 0;
6036 }
6037
6038 static inline void btrfs_end_bbio(struct btrfs_bio *bbio, struct bio *bio)
6039 {
6040         bio->bi_private = bbio->private;
6041         bio->bi_end_io = bbio->end_io;
6042         bio_endio(bio);
6043
6044         btrfs_put_bbio(bbio);
6045 }
6046
6047 static void btrfs_end_bio(struct bio *bio)
6048 {
6049         struct btrfs_bio *bbio = bio->bi_private;
6050         int is_orig_bio = 0;
6051
6052         if (bio->bi_error) {
6053                 atomic_inc(&bbio->error);
6054                 if (bio->bi_error == -EIO || bio->bi_error == -EREMOTEIO) {
6055                         unsigned int stripe_index =
6056                                 btrfs_io_bio(bio)->stripe_index;
6057                         struct btrfs_device *dev;
6058
6059                         BUG_ON(stripe_index >= bbio->num_stripes);
6060                         dev = bbio->stripes[stripe_index].dev;
6061                         if (dev->bdev) {
6062                                 if (bio_op(bio) == REQ_OP_WRITE)
6063                                         btrfs_dev_stat_inc(dev,
6064                                                 BTRFS_DEV_STAT_WRITE_ERRS);
6065                                 else
6066                                         btrfs_dev_stat_inc(dev,
6067                                                 BTRFS_DEV_STAT_READ_ERRS);
6068                                 if (bio->bi_opf & REQ_PREFLUSH)
6069                                         btrfs_dev_stat_inc(dev,
6070                                                 BTRFS_DEV_STAT_FLUSH_ERRS);
6071                                 btrfs_dev_stat_print_on_error(dev);
6072                         }
6073                 }
6074         }
6075
6076         if (bio == bbio->orig_bio)
6077                 is_orig_bio = 1;
6078
6079         btrfs_bio_counter_dec(bbio->fs_info);
6080
6081         if (atomic_dec_and_test(&bbio->stripes_pending)) {
6082                 if (!is_orig_bio) {
6083                         bio_put(bio);
6084                         bio = bbio->orig_bio;
6085                 }
6086
6087                 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
6088                 /* only send an error to the higher layers if it is
6089                  * beyond the tolerance of the btrfs bio
6090                  */
6091                 if (atomic_read(&bbio->error) > bbio->max_errors) {
6092                         bio->bi_error = -EIO;
6093                 } else {
6094                         /*
6095                          * this bio is actually up to date, we didn't
6096                          * go over the max number of errors
6097                          */
6098                         bio->bi_error = 0;
6099                 }
6100
6101                 btrfs_end_bbio(bbio, bio);
6102         } else if (!is_orig_bio) {
6103                 bio_put(bio);
6104         }
6105 }
6106
6107 /*
6108  * see run_scheduled_bios for a description of why bios are collected for
6109  * async submit.
6110  *
6111  * This will add one bio to the pending list for a device and make sure
6112  * the work struct is scheduled.
6113  */
6114 static noinline void btrfs_schedule_bio(struct btrfs_device *device,
6115                                         struct bio *bio)
6116 {
6117         struct btrfs_fs_info *fs_info = device->fs_info;
6118         int should_queue = 1;
6119         struct btrfs_pending_bios *pending_bios;
6120
6121         if (device->missing || !device->bdev) {
6122                 bio_io_error(bio);
6123                 return;
6124         }
6125
6126         /* don't bother with additional async steps for reads, right now */
6127         if (bio_op(bio) == REQ_OP_READ) {
6128                 bio_get(bio);
6129                 btrfsic_submit_bio(bio);
6130                 bio_put(bio);
6131                 return;
6132         }
6133
6134         /*
6135          * nr_async_bios allows us to reliably return congestion to the
6136          * higher layers.  Otherwise, the async bio makes it appear we have
6137          * made progress against dirty pages when we've really just put it
6138          * on a queue for later
6139          */
6140         atomic_inc(&fs_info->nr_async_bios);
6141         WARN_ON(bio->bi_next);
6142         bio->bi_next = NULL;
6143
6144         spin_lock(&device->io_lock);
6145         if (op_is_sync(bio->bi_opf))
6146                 pending_bios = &device->pending_sync_bios;
6147         else
6148                 pending_bios = &device->pending_bios;
6149
6150         if (pending_bios->tail)
6151                 pending_bios->tail->bi_next = bio;
6152
6153         pending_bios->tail = bio;
6154         if (!pending_bios->head)
6155                 pending_bios->head = bio;
6156         if (device->running_pending)
6157                 should_queue = 0;
6158
6159         spin_unlock(&device->io_lock);
6160
6161         if (should_queue)
6162                 btrfs_queue_work(fs_info->submit_workers, &device->work);
6163 }
6164
6165 static void submit_stripe_bio(struct btrfs_bio *bbio, struct bio *bio,
6166                               u64 physical, int dev_nr, int async)
6167 {
6168         struct btrfs_device *dev = bbio->stripes[dev_nr].dev;
6169         struct btrfs_fs_info *fs_info = bbio->fs_info;
6170
6171         bio->bi_private = bbio;
6172         btrfs_io_bio(bio)->stripe_index = dev_nr;
6173         bio->bi_end_io = btrfs_end_bio;
6174         bio->bi_iter.bi_sector = physical >> 9;
6175 #ifdef DEBUG
6176         {
6177                 struct rcu_string *name;
6178
6179                 rcu_read_lock();
6180                 name = rcu_dereference(dev->name);
6181                 btrfs_debug(fs_info,
6182                         "btrfs_map_bio: rw %d 0x%x, sector=%llu, dev=%lu (%s id %llu), size=%u",
6183                         bio_op(bio), bio->bi_opf,
6184                         (u64)bio->bi_iter.bi_sector,
6185                         (u_long)dev->bdev->bd_dev, name->str, dev->devid,
6186                         bio->bi_iter.bi_size);
6187                 rcu_read_unlock();
6188         }
6189 #endif
6190         bio->bi_bdev = dev->bdev;
6191
6192         btrfs_bio_counter_inc_noblocked(fs_info);
6193
6194         if (async)
6195                 btrfs_schedule_bio(dev, bio);
6196         else
6197                 btrfsic_submit_bio(bio);
6198 }
6199
6200 static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical)
6201 {
6202         atomic_inc(&bbio->error);
6203         if (atomic_dec_and_test(&bbio->stripes_pending)) {
6204                 /* Should be the original bio. */
6205                 WARN_ON(bio != bbio->orig_bio);
6206
6207                 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
6208                 bio->bi_iter.bi_sector = logical >> 9;
6209                 bio->bi_error = -EIO;
6210                 btrfs_end_bbio(bbio, bio);
6211         }
6212 }
6213
6214 int btrfs_map_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
6215                   int mirror_num, int async_submit)
6216 {
6217         struct btrfs_device *dev;
6218         struct bio *first_bio = bio;
6219         u64 logical = (u64)bio->bi_iter.bi_sector << 9;
6220         u64 length = 0;
6221         u64 map_length;
6222         int ret;
6223         int dev_nr;
6224         int total_devs;
6225         struct btrfs_bio *bbio = NULL;
6226
6227         length = bio->bi_iter.bi_size;
6228         map_length = length;
6229
6230         btrfs_bio_counter_inc_blocked(fs_info);
6231         ret = __btrfs_map_block(fs_info, bio_op(bio), logical,
6232                                 &map_length, &bbio, mirror_num, 1);
6233         if (ret) {
6234                 btrfs_bio_counter_dec(fs_info);
6235                 return ret;
6236         }
6237
6238         total_devs = bbio->num_stripes;
6239         bbio->orig_bio = first_bio;
6240         bbio->private = first_bio->bi_private;
6241         bbio->end_io = first_bio->bi_end_io;
6242         bbio->fs_info = fs_info;
6243         atomic_set(&bbio->stripes_pending, bbio->num_stripes);
6244
6245         if ((bbio->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
6246             ((bio_op(bio) == REQ_OP_WRITE) || (mirror_num > 1))) {
6247                 /* In this case, map_length has been set to the length of
6248                    a single stripe; not the whole write */
6249                 if (bio_op(bio) == REQ_OP_WRITE) {
6250                         ret = raid56_parity_write(fs_info, bio, bbio,
6251                                                   map_length);
6252                 } else {
6253                         ret = raid56_parity_recover(fs_info, bio, bbio,
6254                                                     map_length, mirror_num, 1);
6255                 }
6256
6257                 btrfs_bio_counter_dec(fs_info);
6258                 return ret;
6259         }
6260
6261         if (map_length < length) {
6262                 btrfs_crit(fs_info,
6263                            "mapping failed logical %llu bio len %llu len %llu",
6264                            logical, length, map_length);
6265                 BUG();
6266         }
6267
6268         for (dev_nr = 0; dev_nr < total_devs; dev_nr++) {
6269                 dev = bbio->stripes[dev_nr].dev;
6270                 if (!dev || !dev->bdev ||
6271                     (bio_op(first_bio) == REQ_OP_WRITE && !dev->writeable)) {
6272                         bbio_error(bbio, first_bio, logical);
6273                         continue;
6274                 }
6275
6276                 if (dev_nr < total_devs - 1)
6277                         bio = btrfs_bio_clone(first_bio);
6278                 else
6279                         bio = first_bio;
6280
6281                 submit_stripe_bio(bbio, bio, bbio->stripes[dev_nr].physical,
6282                                   dev_nr, async_submit);
6283         }
6284         btrfs_bio_counter_dec(fs_info);
6285         return 0;
6286 }
6287
6288 struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid,
6289                                        u8 *uuid, u8 *fsid)
6290 {
6291         struct btrfs_device *device;
6292         struct btrfs_fs_devices *cur_devices;
6293
6294         cur_devices = fs_info->fs_devices;
6295         while (cur_devices) {
6296                 if (!fsid ||
6297                     !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
6298                         device = __find_device(&cur_devices->devices,
6299                                                devid, uuid);
6300                         if (device)
6301                                 return device;
6302                 }
6303                 cur_devices = cur_devices->seed;
6304         }
6305         return NULL;
6306 }
6307
6308 static struct btrfs_device *add_missing_dev(struct btrfs_fs_devices *fs_devices,
6309                                             u64 devid, u8 *dev_uuid)
6310 {
6311         struct btrfs_device *device;
6312
6313         device = btrfs_alloc_device(NULL, &devid, dev_uuid);
6314         if (IS_ERR(device))
6315                 return NULL;
6316
6317         list_add(&device->dev_list, &fs_devices->devices);
6318         device->fs_devices = fs_devices;
6319         fs_devices->num_devices++;
6320
6321         device->missing = 1;
6322         fs_devices->missing_devices++;
6323
6324         return device;
6325 }
6326
6327 /**
6328  * btrfs_alloc_device - allocate struct btrfs_device
6329  * @fs_info:    used only for generating a new devid, can be NULL if
6330  *              devid is provided (i.e. @devid != NULL).
6331  * @devid:      a pointer to devid for this device.  If NULL a new devid
6332  *              is generated.
6333  * @uuid:       a pointer to UUID for this device.  If NULL a new UUID
6334  *              is generated.
6335  *
6336  * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
6337  * on error.  Returned struct is not linked onto any lists and can be
6338  * destroyed with kfree() right away.
6339  */
6340 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
6341                                         const u64 *devid,
6342                                         const u8 *uuid)
6343 {
6344         struct btrfs_device *dev;
6345         u64 tmp;
6346
6347         if (WARN_ON(!devid && !fs_info))
6348                 return ERR_PTR(-EINVAL);
6349
6350         dev = __alloc_device();
6351         if (IS_ERR(dev))
6352                 return dev;
6353
6354         if (devid)
6355                 tmp = *devid;
6356         else {
6357                 int ret;
6358
6359                 ret = find_next_devid(fs_info, &tmp);
6360                 if (ret) {
6361                         kfree(dev);
6362                         return ERR_PTR(ret);
6363                 }
6364         }
6365         dev->devid = tmp;
6366
6367         if (uuid)
6368                 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE);
6369         else
6370                 generate_random_uuid(dev->uuid);
6371
6372         btrfs_init_work(&dev->work, btrfs_submit_helper,
6373                         pending_bios_fn, NULL, NULL);
6374
6375         return dev;
6376 }
6377
6378 /* Return -EIO if any error, otherwise return 0. */
6379 static int btrfs_check_chunk_valid(struct btrfs_fs_info *fs_info,
6380                                    struct extent_buffer *leaf,
6381                                    struct btrfs_chunk *chunk, u64 logical)
6382 {
6383         u64 length;
6384         u64 stripe_len;
6385         u16 num_stripes;
6386         u16 sub_stripes;
6387         u64 type;
6388
6389         length = btrfs_chunk_length(leaf, chunk);
6390         stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6391         num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6392         sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
6393         type = btrfs_chunk_type(leaf, chunk);
6394
6395         if (!num_stripes) {
6396                 btrfs_err(fs_info, "invalid chunk num_stripes: %u",
6397                           num_stripes);
6398                 return -EIO;
6399         }
6400         if (!IS_ALIGNED(logical, fs_info->sectorsize)) {
6401                 btrfs_err(fs_info, "invalid chunk logical %llu", logical);
6402                 return -EIO;
6403         }
6404         if (btrfs_chunk_sector_size(leaf, chunk) != fs_info->sectorsize) {
6405                 btrfs_err(fs_info, "invalid chunk sectorsize %u",
6406                           btrfs_chunk_sector_size(leaf, chunk));
6407                 return -EIO;
6408         }
6409         if (!length || !IS_ALIGNED(length, fs_info->sectorsize)) {
6410                 btrfs_err(fs_info, "invalid chunk length %llu", length);
6411                 return -EIO;
6412         }
6413         if (!is_power_of_2(stripe_len) || stripe_len != BTRFS_STRIPE_LEN) {
6414                 btrfs_err(fs_info, "invalid chunk stripe length: %llu",
6415                           stripe_len);
6416                 return -EIO;
6417         }
6418         if (~(BTRFS_BLOCK_GROUP_TYPE_MASK | BTRFS_BLOCK_GROUP_PROFILE_MASK) &
6419             type) {
6420                 btrfs_err(fs_info, "unrecognized chunk type: %llu",
6421                           ~(BTRFS_BLOCK_GROUP_TYPE_MASK |
6422                             BTRFS_BLOCK_GROUP_PROFILE_MASK) &
6423                           btrfs_chunk_type(leaf, chunk));
6424                 return -EIO;
6425         }
6426         if ((type & BTRFS_BLOCK_GROUP_RAID10 && sub_stripes != 2) ||
6427             (type & BTRFS_BLOCK_GROUP_RAID1 && num_stripes < 1) ||
6428             (type & BTRFS_BLOCK_GROUP_RAID5 && num_stripes < 2) ||
6429             (type & BTRFS_BLOCK_GROUP_RAID6 && num_stripes < 3) ||
6430             (type & BTRFS_BLOCK_GROUP_DUP && num_stripes > 2) ||
6431             ((type & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 &&
6432              num_stripes != 1)) {
6433                 btrfs_err(fs_info,
6434                         "invalid num_stripes:sub_stripes %u:%u for profile %llu",
6435                         num_stripes, sub_stripes,
6436                         type & BTRFS_BLOCK_GROUP_PROFILE_MASK);
6437                 return -EIO;
6438         }
6439
6440         return 0;
6441 }
6442
6443 static int read_one_chunk(struct btrfs_fs_info *fs_info, struct btrfs_key *key,
6444                           struct extent_buffer *leaf,
6445                           struct btrfs_chunk *chunk)
6446 {
6447         struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
6448         struct map_lookup *map;
6449         struct extent_map *em;
6450         u64 logical;
6451         u64 length;
6452         u64 stripe_len;
6453         u64 devid;
6454         u8 uuid[BTRFS_UUID_SIZE];
6455         int num_stripes;
6456         int ret;
6457         int i;
6458
6459         logical = key->offset;
6460         length = btrfs_chunk_length(leaf, chunk);
6461         stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6462         num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6463
6464         ret = btrfs_check_chunk_valid(fs_info, leaf, chunk, logical);
6465         if (ret)
6466                 return ret;
6467
6468         read_lock(&map_tree->map_tree.lock);
6469         em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
6470         read_unlock(&map_tree->map_tree.lock);
6471
6472         /* already mapped? */
6473         if (em && em->start <= logical && em->start + em->len > logical) {
6474                 free_extent_map(em);
6475                 return 0;
6476         } else if (em) {
6477                 free_extent_map(em);
6478         }
6479
6480         em = alloc_extent_map();
6481         if (!em)
6482                 return -ENOMEM;
6483         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
6484         if (!map) {
6485                 free_extent_map(em);
6486                 return -ENOMEM;
6487         }
6488
6489         set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
6490         em->map_lookup = map;
6491         em->start = logical;
6492         em->len = length;
6493         em->orig_start = 0;
6494         em->block_start = 0;
6495         em->block_len = em->len;
6496
6497         map->num_stripes = num_stripes;
6498         map->io_width = btrfs_chunk_io_width(leaf, chunk);
6499         map->io_align = btrfs_chunk_io_align(leaf, chunk);
6500         map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
6501         map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6502         map->type = btrfs_chunk_type(leaf, chunk);
6503         map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
6504         for (i = 0; i < num_stripes; i++) {
6505                 map->stripes[i].physical =
6506                         btrfs_stripe_offset_nr(leaf, chunk, i);
6507                 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
6508                 read_extent_buffer(leaf, uuid, (unsigned long)
6509                                    btrfs_stripe_dev_uuid_nr(chunk, i),
6510                                    BTRFS_UUID_SIZE);
6511                 map->stripes[i].dev = btrfs_find_device(fs_info, devid,
6512                                                         uuid, NULL);
6513                 if (!map->stripes[i].dev &&
6514                     !btrfs_test_opt(fs_info, DEGRADED)) {
6515                         free_extent_map(em);
6516                         return -EIO;
6517                 }
6518                 if (!map->stripes[i].dev) {
6519                         map->stripes[i].dev =
6520                                 add_missing_dev(fs_info->fs_devices, devid,
6521                                                 uuid);
6522                         if (!map->stripes[i].dev) {
6523                                 free_extent_map(em);
6524                                 return -EIO;
6525                         }
6526                         btrfs_warn(fs_info, "devid %llu uuid %pU is missing",
6527                                    devid, uuid);
6528                 }
6529                 map->stripes[i].dev->in_fs_metadata = 1;
6530         }
6531
6532         write_lock(&map_tree->map_tree.lock);
6533         ret = add_extent_mapping(&map_tree->map_tree, em, 0);
6534         write_unlock(&map_tree->map_tree.lock);
6535         BUG_ON(ret); /* Tree corruption */
6536         free_extent_map(em);
6537
6538         return 0;
6539 }
6540
6541 static void fill_device_from_item(struct extent_buffer *leaf,
6542                                  struct btrfs_dev_item *dev_item,
6543                                  struct btrfs_device *device)
6544 {
6545         unsigned long ptr;
6546
6547         device->devid = btrfs_device_id(leaf, dev_item);
6548         device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
6549         device->total_bytes = device->disk_total_bytes;
6550         device->commit_total_bytes = device->disk_total_bytes;
6551         device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
6552         device->commit_bytes_used = device->bytes_used;
6553         device->type = btrfs_device_type(leaf, dev_item);
6554         device->io_align = btrfs_device_io_align(leaf, dev_item);
6555         device->io_width = btrfs_device_io_width(leaf, dev_item);
6556         device->sector_size = btrfs_device_sector_size(leaf, dev_item);
6557         WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
6558         device->is_tgtdev_for_dev_replace = 0;
6559
6560         ptr = btrfs_device_uuid(dev_item);
6561         read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
6562 }
6563
6564 static struct btrfs_fs_devices *open_seed_devices(struct btrfs_fs_info *fs_info,
6565                                                   u8 *fsid)
6566 {
6567         struct btrfs_fs_devices *fs_devices;
6568         int ret;
6569
6570         BUG_ON(!mutex_is_locked(&uuid_mutex));
6571
6572         fs_devices = fs_info->fs_devices->seed;
6573         while (fs_devices) {
6574                 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE))
6575                         return fs_devices;
6576
6577                 fs_devices = fs_devices->seed;
6578         }
6579
6580         fs_devices = find_fsid(fsid);
6581         if (!fs_devices) {
6582                 if (!btrfs_test_opt(fs_info, DEGRADED))
6583                         return ERR_PTR(-ENOENT);
6584
6585                 fs_devices = alloc_fs_devices(fsid);
6586                 if (IS_ERR(fs_devices))
6587                         return fs_devices;
6588
6589                 fs_devices->seeding = 1;
6590                 fs_devices->opened = 1;
6591                 return fs_devices;
6592         }
6593
6594         fs_devices = clone_fs_devices(fs_devices);
6595         if (IS_ERR(fs_devices))
6596                 return fs_devices;
6597
6598         ret = __btrfs_open_devices(fs_devices, FMODE_READ,
6599                                    fs_info->bdev_holder);
6600         if (ret) {
6601                 free_fs_devices(fs_devices);
6602                 fs_devices = ERR_PTR(ret);
6603                 goto out;
6604         }
6605
6606         if (!fs_devices->seeding) {
6607                 __btrfs_close_devices(fs_devices);
6608                 free_fs_devices(fs_devices);
6609                 fs_devices = ERR_PTR(-EINVAL);
6610                 goto out;
6611         }
6612
6613         fs_devices->seed = fs_info->fs_devices->seed;
6614         fs_info->fs_devices->seed = fs_devices;
6615 out:
6616         return fs_devices;
6617 }
6618
6619 static int read_one_dev(struct btrfs_fs_info *fs_info,
6620                         struct extent_buffer *leaf,
6621                         struct btrfs_dev_item *dev_item)
6622 {
6623         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6624         struct btrfs_device *device;
6625         u64 devid;
6626         int ret;
6627         u8 fs_uuid[BTRFS_UUID_SIZE];
6628         u8 dev_uuid[BTRFS_UUID_SIZE];
6629
6630         devid = btrfs_device_id(leaf, dev_item);
6631         read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
6632                            BTRFS_UUID_SIZE);
6633         read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
6634                            BTRFS_UUID_SIZE);
6635
6636         if (memcmp(fs_uuid, fs_info->fsid, BTRFS_UUID_SIZE)) {
6637                 fs_devices = open_seed_devices(fs_info, fs_uuid);
6638                 if (IS_ERR(fs_devices))
6639                         return PTR_ERR(fs_devices);
6640         }
6641
6642         device = btrfs_find_device(fs_info, devid, dev_uuid, fs_uuid);
6643         if (!device) {
6644                 if (!btrfs_test_opt(fs_info, DEGRADED))
6645                         return -EIO;
6646
6647                 device = add_missing_dev(fs_devices, devid, dev_uuid);
6648                 if (!device)
6649                         return -ENOMEM;
6650                 btrfs_warn(fs_info, "devid %llu uuid %pU missing",
6651                                 devid, dev_uuid);
6652         } else {
6653                 if (!device->bdev && !btrfs_test_opt(fs_info, DEGRADED))
6654                         return -EIO;
6655
6656                 if(!device->bdev && !device->missing) {
6657                         /*
6658                          * this happens when a device that was properly setup
6659                          * in the device info lists suddenly goes bad.
6660                          * device->bdev is NULL, and so we have to set
6661                          * device->missing to one here
6662                          */
6663                         device->fs_devices->missing_devices++;
6664                         device->missing = 1;
6665                 }
6666
6667                 /* Move the device to its own fs_devices */
6668                 if (device->fs_devices != fs_devices) {
6669                         ASSERT(device->missing);
6670
6671                         list_move(&device->dev_list, &fs_devices->devices);
6672                         device->fs_devices->num_devices--;
6673                         fs_devices->num_devices++;
6674
6675                         device->fs_devices->missing_devices--;
6676                         fs_devices->missing_devices++;
6677
6678                         device->fs_devices = fs_devices;
6679                 }
6680         }
6681
6682         if (device->fs_devices != fs_info->fs_devices) {
6683                 BUG_ON(device->writeable);
6684                 if (device->generation !=
6685                     btrfs_device_generation(leaf, dev_item))
6686                         return -EINVAL;
6687         }
6688
6689         fill_device_from_item(leaf, dev_item, device);
6690         device->in_fs_metadata = 1;
6691         if (device->writeable && !device->is_tgtdev_for_dev_replace) {
6692                 device->fs_devices->total_rw_bytes += device->total_bytes;
6693                 atomic64_add(device->total_bytes - device->bytes_used,
6694                                 &fs_info->free_chunk_space);
6695         }
6696         ret = 0;
6697         return ret;
6698 }
6699
6700 int btrfs_read_sys_array(struct btrfs_fs_info *fs_info)
6701 {
6702         struct btrfs_root *root = fs_info->tree_root;
6703         struct btrfs_super_block *super_copy = fs_info->super_copy;
6704         struct extent_buffer *sb;
6705         struct btrfs_disk_key *disk_key;
6706         struct btrfs_chunk *chunk;
6707         u8 *array_ptr;
6708         unsigned long sb_array_offset;
6709         int ret = 0;
6710         u32 num_stripes;
6711         u32 array_size;
6712         u32 len = 0;
6713         u32 cur_offset;
6714         u64 type;
6715         struct btrfs_key key;
6716
6717         ASSERT(BTRFS_SUPER_INFO_SIZE <= fs_info->nodesize);
6718         /*
6719          * This will create extent buffer of nodesize, superblock size is
6720          * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will
6721          * overallocate but we can keep it as-is, only the first page is used.
6722          */
6723         sb = btrfs_find_create_tree_block(fs_info, BTRFS_SUPER_INFO_OFFSET);
6724         if (IS_ERR(sb))
6725                 return PTR_ERR(sb);
6726         set_extent_buffer_uptodate(sb);
6727         btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
6728         /*
6729          * The sb extent buffer is artificial and just used to read the system array.
6730          * set_extent_buffer_uptodate() call does not properly mark all it's
6731          * pages up-to-date when the page is larger: extent does not cover the
6732          * whole page and consequently check_page_uptodate does not find all
6733          * the page's extents up-to-date (the hole beyond sb),
6734          * write_extent_buffer then triggers a WARN_ON.
6735          *
6736          * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
6737          * but sb spans only this function. Add an explicit SetPageUptodate call
6738          * to silence the warning eg. on PowerPC 64.
6739          */
6740         if (PAGE_SIZE > BTRFS_SUPER_INFO_SIZE)
6741                 SetPageUptodate(sb->pages[0]);
6742
6743         write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
6744         array_size = btrfs_super_sys_array_size(super_copy);
6745
6746         array_ptr = super_copy->sys_chunk_array;
6747         sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array);
6748         cur_offset = 0;
6749
6750         while (cur_offset < array_size) {
6751                 disk_key = (struct btrfs_disk_key *)array_ptr;
6752                 len = sizeof(*disk_key);
6753                 if (cur_offset + len > array_size)
6754                         goto out_short_read;
6755
6756                 btrfs_disk_key_to_cpu(&key, disk_key);
6757
6758                 array_ptr += len;
6759                 sb_array_offset += len;
6760                 cur_offset += len;
6761
6762                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
6763                         chunk = (struct btrfs_chunk *)sb_array_offset;
6764                         /*
6765                          * At least one btrfs_chunk with one stripe must be
6766                          * present, exact stripe count check comes afterwards
6767                          */
6768                         len = btrfs_chunk_item_size(1);
6769                         if (cur_offset + len > array_size)
6770                                 goto out_short_read;
6771
6772                         num_stripes = btrfs_chunk_num_stripes(sb, chunk);
6773                         if (!num_stripes) {
6774                                 btrfs_err(fs_info,
6775                                         "invalid number of stripes %u in sys_array at offset %u",
6776                                         num_stripes, cur_offset);
6777                                 ret = -EIO;
6778                                 break;
6779                         }
6780
6781                         type = btrfs_chunk_type(sb, chunk);
6782                         if ((type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) {
6783                                 btrfs_err(fs_info,
6784                             "invalid chunk type %llu in sys_array at offset %u",
6785                                         type, cur_offset);
6786                                 ret = -EIO;
6787                                 break;
6788                         }
6789
6790                         len = btrfs_chunk_item_size(num_stripes);
6791                         if (cur_offset + len > array_size)
6792                                 goto out_short_read;
6793
6794                         ret = read_one_chunk(fs_info, &key, sb, chunk);
6795                         if (ret)
6796                                 break;
6797                 } else {
6798                         btrfs_err(fs_info,
6799                             "unexpected item type %u in sys_array at offset %u",
6800                                   (u32)key.type, cur_offset);
6801                         ret = -EIO;
6802                         break;
6803                 }
6804                 array_ptr += len;
6805                 sb_array_offset += len;
6806                 cur_offset += len;
6807         }
6808         clear_extent_buffer_uptodate(sb);
6809         free_extent_buffer_stale(sb);
6810         return ret;
6811
6812 out_short_read:
6813         btrfs_err(fs_info, "sys_array too short to read %u bytes at offset %u",
6814                         len, cur_offset);
6815         clear_extent_buffer_uptodate(sb);
6816         free_extent_buffer_stale(sb);
6817         return -EIO;
6818 }
6819
6820 int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info)
6821 {
6822         struct btrfs_root *root = fs_info->chunk_root;
6823         struct btrfs_path *path;
6824         struct extent_buffer *leaf;
6825         struct btrfs_key key;
6826         struct btrfs_key found_key;
6827         int ret;
6828         int slot;
6829         u64 total_dev = 0;
6830
6831         path = btrfs_alloc_path();
6832         if (!path)
6833                 return -ENOMEM;
6834
6835         mutex_lock(&uuid_mutex);
6836         mutex_lock(&fs_info->chunk_mutex);
6837
6838         /*
6839          * Read all device items, and then all the chunk items. All
6840          * device items are found before any chunk item (their object id
6841          * is smaller than the lowest possible object id for a chunk
6842          * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
6843          */
6844         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
6845         key.offset = 0;
6846         key.type = 0;
6847         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6848         if (ret < 0)
6849                 goto error;
6850         while (1) {
6851                 leaf = path->nodes[0];
6852                 slot = path->slots[0];
6853                 if (slot >= btrfs_header_nritems(leaf)) {
6854                         ret = btrfs_next_leaf(root, path);
6855                         if (ret == 0)
6856                                 continue;
6857                         if (ret < 0)
6858                                 goto error;
6859                         break;
6860                 }
6861                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6862                 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
6863                         struct btrfs_dev_item *dev_item;
6864                         dev_item = btrfs_item_ptr(leaf, slot,
6865                                                   struct btrfs_dev_item);
6866                         ret = read_one_dev(fs_info, leaf, dev_item);
6867                         if (ret)
6868                                 goto error;
6869                         total_dev++;
6870                 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
6871                         struct btrfs_chunk *chunk;
6872                         chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
6873                         ret = read_one_chunk(fs_info, &found_key, leaf, chunk);
6874                         if (ret)
6875                                 goto error;
6876                 }
6877                 path->slots[0]++;
6878         }
6879
6880         /*
6881          * After loading chunk tree, we've got all device information,
6882          * do another round of validation checks.
6883          */
6884         if (total_dev != fs_info->fs_devices->total_devices) {
6885                 btrfs_err(fs_info,
6886            "super_num_devices %llu mismatch with num_devices %llu found here",
6887                           btrfs_super_num_devices(fs_info->super_copy),
6888                           total_dev);
6889                 ret = -EINVAL;
6890                 goto error;
6891         }
6892         if (btrfs_super_total_bytes(fs_info->super_copy) <
6893             fs_info->fs_devices->total_rw_bytes) {
6894                 btrfs_err(fs_info,
6895         "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu",
6896                           btrfs_super_total_bytes(fs_info->super_copy),
6897                           fs_info->fs_devices->total_rw_bytes);
6898                 ret = -EINVAL;
6899                 goto error;
6900         }
6901         ret = 0;
6902 error:
6903         mutex_unlock(&fs_info->chunk_mutex);
6904         mutex_unlock(&uuid_mutex);
6905
6906         btrfs_free_path(path);
6907         return ret;
6908 }
6909
6910 void btrfs_init_devices_late(struct btrfs_fs_info *fs_info)
6911 {
6912         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6913         struct btrfs_device *device;
6914
6915         while (fs_devices) {
6916                 mutex_lock(&fs_devices->device_list_mutex);
6917                 list_for_each_entry(device, &fs_devices->devices, dev_list)
6918                         device->fs_info = fs_info;
6919                 mutex_unlock(&fs_devices->device_list_mutex);
6920
6921                 fs_devices = fs_devices->seed;
6922         }
6923 }
6924
6925 static void __btrfs_reset_dev_stats(struct btrfs_device *dev)
6926 {
6927         int i;
6928
6929         for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6930                 btrfs_dev_stat_reset(dev, i);
6931 }
6932
6933 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
6934 {
6935         struct btrfs_key key;
6936         struct btrfs_key found_key;
6937         struct btrfs_root *dev_root = fs_info->dev_root;
6938         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6939         struct extent_buffer *eb;
6940         int slot;
6941         int ret = 0;
6942         struct btrfs_device *device;
6943         struct btrfs_path *path = NULL;
6944         int i;
6945
6946         path = btrfs_alloc_path();
6947         if (!path) {
6948                 ret = -ENOMEM;
6949                 goto out;
6950         }
6951
6952         mutex_lock(&fs_devices->device_list_mutex);
6953         list_for_each_entry(device, &fs_devices->devices, dev_list) {
6954                 int item_size;
6955                 struct btrfs_dev_stats_item *ptr;
6956
6957                 key.objectid = BTRFS_DEV_STATS_OBJECTID;
6958                 key.type = BTRFS_PERSISTENT_ITEM_KEY;
6959                 key.offset = device->devid;
6960                 ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
6961                 if (ret) {
6962                         __btrfs_reset_dev_stats(device);
6963                         device->dev_stats_valid = 1;
6964                         btrfs_release_path(path);
6965                         continue;
6966                 }
6967                 slot = path->slots[0];
6968                 eb = path->nodes[0];
6969                 btrfs_item_key_to_cpu(eb, &found_key, slot);
6970                 item_size = btrfs_item_size_nr(eb, slot);
6971
6972                 ptr = btrfs_item_ptr(eb, slot,
6973                                      struct btrfs_dev_stats_item);
6974
6975                 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
6976                         if (item_size >= (1 + i) * sizeof(__le64))
6977                                 btrfs_dev_stat_set(device, i,
6978                                         btrfs_dev_stats_value(eb, ptr, i));
6979                         else
6980                                 btrfs_dev_stat_reset(device, i);
6981                 }
6982
6983                 device->dev_stats_valid = 1;
6984                 btrfs_dev_stat_print_on_load(device);
6985                 btrfs_release_path(path);
6986         }
6987         mutex_unlock(&fs_devices->device_list_mutex);
6988
6989 out:
6990         btrfs_free_path(path);
6991         return ret < 0 ? ret : 0;
6992 }
6993
6994 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
6995                                 struct btrfs_fs_info *fs_info,
6996                                 struct btrfs_device *device)
6997 {
6998         struct btrfs_root *dev_root = fs_info->dev_root;
6999         struct btrfs_path *path;
7000         struct btrfs_key key;
7001         struct extent_buffer *eb;
7002         struct btrfs_dev_stats_item *ptr;
7003         int ret;
7004         int i;
7005
7006         key.objectid = BTRFS_DEV_STATS_OBJECTID;
7007         key.type = BTRFS_PERSISTENT_ITEM_KEY;
7008         key.offset = device->devid;
7009
7010         path = btrfs_alloc_path();
7011         if (!path)
7012                 return -ENOMEM;
7013         ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
7014         if (ret < 0) {
7015                 btrfs_warn_in_rcu(fs_info,
7016                         "error %d while searching for dev_stats item for device %s",
7017                               ret, rcu_str_deref(device->name));
7018                 goto out;
7019         }
7020
7021         if (ret == 0 &&
7022             btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
7023                 /* need to delete old one and insert a new one */
7024                 ret = btrfs_del_item(trans, dev_root, path);
7025                 if (ret != 0) {
7026                         btrfs_warn_in_rcu(fs_info,
7027                                 "delete too small dev_stats item for device %s failed %d",
7028                                       rcu_str_deref(device->name), ret);
7029                         goto out;
7030                 }
7031                 ret = 1;
7032         }
7033
7034         if (ret == 1) {
7035                 /* need to insert a new item */
7036                 btrfs_release_path(path);
7037                 ret = btrfs_insert_empty_item(trans, dev_root, path,
7038                                               &key, sizeof(*ptr));
7039                 if (ret < 0) {
7040                         btrfs_warn_in_rcu(fs_info,
7041                                 "insert dev_stats item for device %s failed %d",
7042                                 rcu_str_deref(device->name), ret);
7043                         goto out;
7044                 }
7045         }
7046
7047         eb = path->nodes[0];
7048         ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
7049         for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7050                 btrfs_set_dev_stats_value(eb, ptr, i,
7051                                           btrfs_dev_stat_read(device, i));
7052         btrfs_mark_buffer_dirty(eb);
7053
7054 out:
7055         btrfs_free_path(path);
7056         return ret;
7057 }
7058
7059 /*
7060  * called from commit_transaction. Writes all changed device stats to disk.
7061  */
7062 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans,
7063                         struct btrfs_fs_info *fs_info)
7064 {
7065         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7066         struct btrfs_device *device;
7067         int stats_cnt;
7068         int ret = 0;
7069
7070         mutex_lock(&fs_devices->device_list_mutex);
7071         list_for_each_entry(device, &fs_devices->devices, dev_list) {
7072                 if (!device->dev_stats_valid || !btrfs_dev_stats_dirty(device))
7073                         continue;
7074
7075                 stats_cnt = atomic_read(&device->dev_stats_ccnt);
7076                 ret = update_dev_stat_item(trans, fs_info, device);
7077                 if (!ret)
7078                         atomic_sub(stats_cnt, &device->dev_stats_ccnt);
7079         }
7080         mutex_unlock(&fs_devices->device_list_mutex);
7081
7082         return ret;
7083 }
7084
7085 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
7086 {
7087         btrfs_dev_stat_inc(dev, index);
7088         btrfs_dev_stat_print_on_error(dev);
7089 }
7090
7091 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
7092 {
7093         if (!dev->dev_stats_valid)
7094                 return;
7095         btrfs_err_rl_in_rcu(dev->fs_info,
7096                 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7097                            rcu_str_deref(dev->name),
7098                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7099                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7100                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7101                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7102                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7103 }
7104
7105 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
7106 {
7107         int i;
7108
7109         for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7110                 if (btrfs_dev_stat_read(dev, i) != 0)
7111                         break;
7112         if (i == BTRFS_DEV_STAT_VALUES_MAX)
7113                 return; /* all values == 0, suppress message */
7114
7115         btrfs_info_in_rcu(dev->fs_info,
7116                 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7117                rcu_str_deref(dev->name),
7118                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7119                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7120                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7121                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7122                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7123 }
7124
7125 int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info,
7126                         struct btrfs_ioctl_get_dev_stats *stats)
7127 {
7128         struct btrfs_device *dev;
7129         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7130         int i;
7131
7132         mutex_lock(&fs_devices->device_list_mutex);
7133         dev = btrfs_find_device(fs_info, stats->devid, NULL, NULL);
7134         mutex_unlock(&fs_devices->device_list_mutex);
7135
7136         if (!dev) {
7137                 btrfs_warn(fs_info, "get dev_stats failed, device not found");
7138                 return -ENODEV;
7139         } else if (!dev->dev_stats_valid) {
7140                 btrfs_warn(fs_info, "get dev_stats failed, not yet valid");
7141                 return -ENODEV;
7142         } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
7143                 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7144                         if (stats->nr_items > i)
7145                                 stats->values[i] =
7146                                         btrfs_dev_stat_read_and_reset(dev, i);
7147                         else
7148                                 btrfs_dev_stat_reset(dev, i);
7149                 }
7150         } else {
7151                 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7152                         if (stats->nr_items > i)
7153                                 stats->values[i] = btrfs_dev_stat_read(dev, i);
7154         }
7155         if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
7156                 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
7157         return 0;
7158 }
7159
7160 void btrfs_scratch_superblocks(struct block_device *bdev, const char *device_path)
7161 {
7162         struct buffer_head *bh;
7163         struct btrfs_super_block *disk_super;
7164         int copy_num;
7165
7166         if (!bdev)
7167                 return;
7168
7169         for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX;
7170                 copy_num++) {
7171
7172                 if (btrfs_read_dev_one_super(bdev, copy_num, &bh))
7173                         continue;
7174
7175                 disk_super = (struct btrfs_super_block *)bh->b_data;
7176
7177                 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
7178                 set_buffer_dirty(bh);
7179                 sync_dirty_buffer(bh);
7180                 brelse(bh);
7181         }
7182
7183         /* Notify udev that device has changed */
7184         btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
7185
7186         /* Update ctime/mtime for device path for libblkid */
7187         update_dev_time(device_path);
7188 }
7189
7190 /*
7191  * Update the size of all devices, which is used for writing out the
7192  * super blocks.
7193  */
7194 void btrfs_update_commit_device_size(struct btrfs_fs_info *fs_info)
7195 {
7196         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7197         struct btrfs_device *curr, *next;
7198
7199         if (list_empty(&fs_devices->resized_devices))
7200                 return;
7201
7202         mutex_lock(&fs_devices->device_list_mutex);
7203         mutex_lock(&fs_info->chunk_mutex);
7204         list_for_each_entry_safe(curr, next, &fs_devices->resized_devices,
7205                                  resized_list) {
7206                 list_del_init(&curr->resized_list);
7207                 curr->commit_total_bytes = curr->disk_total_bytes;
7208         }
7209         mutex_unlock(&fs_info->chunk_mutex);
7210         mutex_unlock(&fs_devices->device_list_mutex);
7211 }
7212
7213 /* Must be invoked during the transaction commit */
7214 void btrfs_update_commit_device_bytes_used(struct btrfs_fs_info *fs_info,
7215                                         struct btrfs_transaction *transaction)
7216 {
7217         struct extent_map *em;
7218         struct map_lookup *map;
7219         struct btrfs_device *dev;
7220         int i;
7221
7222         if (list_empty(&transaction->pending_chunks))
7223                 return;
7224
7225         /* In order to kick the device replace finish process */
7226         mutex_lock(&fs_info->chunk_mutex);
7227         list_for_each_entry(em, &transaction->pending_chunks, list) {
7228                 map = em->map_lookup;
7229
7230                 for (i = 0; i < map->num_stripes; i++) {
7231                         dev = map->stripes[i].dev;
7232                         dev->commit_bytes_used = dev->bytes_used;
7233                 }
7234         }
7235         mutex_unlock(&fs_info->chunk_mutex);
7236 }
7237
7238 void btrfs_set_fs_info_ptr(struct btrfs_fs_info *fs_info)
7239 {
7240         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7241         while (fs_devices) {
7242                 fs_devices->fs_info = fs_info;
7243                 fs_devices = fs_devices->seed;
7244         }
7245 }
7246
7247 void btrfs_reset_fs_info_ptr(struct btrfs_fs_info *fs_info)
7248 {
7249         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7250         while (fs_devices) {
7251                 fs_devices->fs_info = NULL;
7252                 fs_devices = fs_devices->seed;
7253         }
7254 }