fs/btrfs/volumes.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * Copyright (C) 2007 Oracle.  All rights reserved.
   4  */
   5
   6 #include <linux/sched.h>
   7 #include <linux/bio.h>
   8 #include <linux/slab.h>
   9 #include <linux/buffer_head.h>
  10 #include <linux/blkdev.h>
  11 #include <linux/iocontext.h>
  12 #include <linux/capability.h>
  13 #include <linux/ratelimit.h>
  14 #include <linux/kthread.h>
  15 #include <linux/raid/pq.h>
  16 #include <linux/semaphore.h>
  17 #include <linux/uuid.h>
  18 #include <linux/list_sort.h>
  19 #include <asm/div64.h>
  20 #include "ctree.h"
  21 #include "extent_map.h"
  22 #include "disk-io.h"
  23 #include "transaction.h"
  24 #include "print-tree.h"
  25 #include "volumes.h"
  26 #include "raid56.h"
  27 #include "async-thread.h"
  28 #include "check-integrity.h"
  29 #include "rcu-string.h"
  30 #include "math.h"
  31 #include "dev-replace.h"
  32 #include "sysfs.h"
  33
  34 const struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = {
  35         [BTRFS_RAID_RAID10] = {
  36                 .sub_stripes    = 2,
  37                 .dev_stripes    = 1,
  38                 .devs_max       = 0,    /* 0 == as many as possible */
  39                 .devs_min       = 4,
  40                 .tolerated_failures = 1,
  41                 .devs_increment = 2,
  42                 .ncopies        = 2,
  43         },
  44         [BTRFS_RAID_RAID1] = {
  45                 .sub_stripes    = 1,
  46                 .dev_stripes    = 1,
  47                 .devs_max       = 2,
  48                 .devs_min       = 2,
  49                 .tolerated_failures = 1,
  50                 .devs_increment = 2,
  51                 .ncopies        = 2,
  52         },
  53         [BTRFS_RAID_DUP] = {
  54                 .sub_stripes    = 1,
  55                 .dev_stripes    = 2,
  56                 .devs_max       = 1,
  57                 .devs_min       = 1,
  58                 .tolerated_failures = 0,
  59                 .devs_increment = 1,
  60                 .ncopies        = 2,
  61         },
  62         [BTRFS_RAID_RAID0] = {
  63                 .sub_stripes    = 1,
  64                 .dev_stripes    = 1,
  65                 .devs_max       = 0,
  66                 .devs_min       = 2,
  67                 .tolerated_failures = 0,
  68                 .devs_increment = 1,
  69                 .ncopies        = 1,
  70         },
  71         [BTRFS_RAID_SINGLE] = {
  72                 .sub_stripes    = 1,
  73                 .dev_stripes    = 1,
  74                 .devs_max       = 1,
  75                 .devs_min       = 1,
  76                 .tolerated_failures = 0,
  77                 .devs_increment = 1,
  78                 .ncopies        = 1,
  79         },
  80         [BTRFS_RAID_RAID5] = {
  81                 .sub_stripes    = 1,
  82                 .dev_stripes    = 1,
  83                 .devs_max       = 0,
  84                 .devs_min       = 2,
  85                 .tolerated_failures = 1,
  86                 .devs_increment = 1,
  87                 .ncopies        = 2,
  88         },
  89         [BTRFS_RAID_RAID6] = {
  90                 .sub_stripes    = 1,
  91                 .dev_stripes    = 1,
  92                 .devs_max       = 0,
  93                 .devs_min       = 3,
  94                 .tolerated_failures = 2,
  95                 .devs_increment = 1,
  96                 .ncopies        = 3,
  97         },
  98 };
  99
 100 const u64 btrfs_raid_group[BTRFS_NR_RAID_TYPES] = {
 101         [BTRFS_RAID_RAID10] = BTRFS_BLOCK_GROUP_RAID10,
 102         [BTRFS_RAID_RAID1]  = BTRFS_BLOCK_GROUP_RAID1,
 103         [BTRFS_RAID_DUP]    = BTRFS_BLOCK_GROUP_DUP,
 104         [BTRFS_RAID_RAID0]  = BTRFS_BLOCK_GROUP_RAID0,
 105         [BTRFS_RAID_SINGLE] = 0,
 106         [BTRFS_RAID_RAID5]  = BTRFS_BLOCK_GROUP_RAID5,
 107         [BTRFS_RAID_RAID6]  = BTRFS_BLOCK_GROUP_RAID6,
 108 };
 109
 110 /*
 111  * Table to convert BTRFS_RAID_* to the error code if minimum number of devices
 112  * condition is not met. Zero means there's no corresponding
 113  * BTRFS_ERROR_DEV_*_NOT_MET value.
 114  */
 115 const int btrfs_raid_mindev_error[BTRFS_NR_RAID_TYPES] = {
 116         [BTRFS_RAID_RAID10] = BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET,
 117         [BTRFS_RAID_RAID1]  = BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET,
 118         [BTRFS_RAID_DUP]    = 0,
 119         [BTRFS_RAID_RAID0]  = 0,
 120         [BTRFS_RAID_SINGLE] = 0,
 121         [BTRFS_RAID_RAID5]  = BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET,
 122         [BTRFS_RAID_RAID6]  = BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET,
 123 };
 124
 125 static int init_first_rw_device(struct btrfs_trans_handle *trans,
 126                                 struct btrfs_fs_info *fs_info);
 127 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info);
 128 static void __btrfs_reset_dev_stats(struct btrfs_device *dev);
 129 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev);
 130 static void btrfs_dev_stat_print_on_load(struct btrfs_device *device);
 131 static int __btrfs_map_block(struct btrfs_fs_info *fs_info,
 132                              enum btrfs_map_op op,
 133                              u64 logical, u64 *length,
 134                              struct btrfs_bio **bbio_ret,
 135                              int mirror_num, int need_raid_map);
 136
 137 /*
 138  * Device locking
 139  * ==============
 140  *
 141  * There are several mutexes that protect manipulation of devices and low-level
 142  * structures like chunks but not block groups, extents or files
 143  *
 144  * uuid_mutex (global lock)
 145  * ------------------------
 146  * protects the fs_uuids list that tracks all per-fs fs_devices, resulting from
 147  * the SCAN_DEV ioctl registration or from mount either implicitly (the first
 148  * device) or requested by the device= mount option
 149  *
 150  * the mutex can be very coarse and can cover long-running operations
 151  *
 152  * protects: updates to fs_devices counters like missing devices, rw devices,
 153  * seeding, structure cloning, openning/closing devices at mount/umount time
 154  *
 155  * global::fs_devs - add, remove, updates to the global list
 156  *
 157  * does not protect: manipulation of the fs_devices::devices list!
 158  *
 159  * btrfs_device::name - renames (write side), read is RCU
 160  *
 161  * fs_devices::device_list_mutex (per-fs, with RCU)
 162  * ------------------------------------------------
 163  * protects updates to fs_devices::devices, ie. adding and deleting
 164  *
 165  * simple list traversal with read-only actions can be done with RCU protection
 166  *
 167  * may be used to exclude some operations from running concurrently without any
 168  * modifications to the list (see write_all_supers)
 169  *
 170  * volume_mutex
 171  * ------------
 172  * coarse lock owned by a mounted filesystem; used to exclude some operations
 173  * that cannot run in parallel and affect the higher-level properties of the
 174  * filesystem like: device add/deleting/resize/replace, or balance
 175  *
 176  * balance_mutex
 177  * -------------
 178  * protects balance structures (status, state) and context accessed from
 179  * several places (internally, ioctl)
 180  *
 181  * chunk_mutex
 182  * -----------
 183  * protects chunks, adding or removing during allocation, trim or when a new
 184  * device is added/removed
 185  *
 186  * cleaner_mutex
 187  * -------------
 188  * a big lock that is held by the cleaner thread and prevents running subvolume
 189  * cleaning together with relocation or delayed iputs
 190  *
 191  *
 192  * Lock nesting
 193  * ============
 194  *
 195  * uuid_mutex
 196  *   volume_mutex
 197  *     device_list_mutex
 198  *       chunk_mutex
 199  *     balance_mutex
 200  */
 201
 202 DEFINE_MUTEX(uuid_mutex);
 203 static LIST_HEAD(fs_uuids);
 204 struct list_head *btrfs_get_fs_uuids(void)
 205 {
 206         return &fs_uuids;
 207 }
 208
 209 /*
 210  * alloc_fs_devices - allocate struct btrfs_fs_devices
 211  * @fsid:       if not NULL, copy the uuid to fs_devices::fsid
 212  *
 213  * Return a pointer to a new struct btrfs_fs_devices on success, or ERR_PTR().
 214  * The returned struct is not linked onto any lists and can be destroyed with
 215  * kfree() right away.
 216  */
 217 static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid)
 218 {
 219         struct btrfs_fs_devices *fs_devs;
 220
 221         fs_devs = kzalloc(sizeof(*fs_devs), GFP_KERNEL);
 222         if (!fs_devs)
 223                 return ERR_PTR(-ENOMEM);
 224
 225         mutex_init(&fs_devs->device_list_mutex);
 226
 227         INIT_LIST_HEAD(&fs_devs->devices);
 228         INIT_LIST_HEAD(&fs_devs->resized_devices);
 229         INIT_LIST_HEAD(&fs_devs->alloc_list);
 230         INIT_LIST_HEAD(&fs_devs->list);
 231         if (fsid)
 232                 memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE);
 233
 234         return fs_devs;
 235 }
 236
 237 static void free_device(struct btrfs_device *device)
 238 {
 239         rcu_string_free(device->name);
 240         bio_put(device->flush_bio);
 241         kfree(device);
 242 }
 243
 244 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
 245 {
 246         struct btrfs_device *device;
 247         WARN_ON(fs_devices->opened);
 248         while (!list_empty(&fs_devices->devices)) {
 249                 device = list_entry(fs_devices->devices.next,
 250                                     struct btrfs_device, dev_list);
 251                 list_del(&device->dev_list);
 252                 free_device(device);
 253         }
 254         kfree(fs_devices);
 255 }
 256
 257 static void btrfs_kobject_uevent(struct block_device *bdev,
 258                                  enum kobject_action action)
 259 {
 260         int ret;
 261
 262         ret = kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, action);
 263         if (ret)
 264                 pr_warn("BTRFS: Sending event '%d' to kobject: '%s' (%p): failed\n",
 265                         action,
 266                         kobject_name(&disk_to_dev(bdev->bd_disk)->kobj),
 267                         &disk_to_dev(bdev->bd_disk)->kobj);
 268 }
 269
 270 void __exit btrfs_cleanup_fs_uuids(void)
 271 {
 272         struct btrfs_fs_devices *fs_devices;
 273
 274         while (!list_empty(&fs_uuids)) {
 275                 fs_devices = list_entry(fs_uuids.next,
 276                                         struct btrfs_fs_devices, list);
 277                 list_del(&fs_devices->list);
 278                 free_fs_devices(fs_devices);
 279         }
 280 }
 281
 282 /*
 283  * Returns a pointer to a new btrfs_device on success; ERR_PTR() on error.
 284  * Returned struct is not linked onto any lists and must be destroyed using
 285  * free_device.
 286  */
 287 static struct btrfs_device *__alloc_device(void)
 288 {
 289         struct btrfs_device *dev;
 290
 291         dev = kzalloc(sizeof(*dev), GFP_KERNEL);
 292         if (!dev)
 293                 return ERR_PTR(-ENOMEM);
 294
 295         /*
 296          * Preallocate a bio that's always going to be used for flushing device
 297          * barriers and matches the device lifespan
 298          */
 299         dev->flush_bio = bio_alloc_bioset(GFP_KERNEL, 0, NULL);
 300         if (!dev->flush_bio) {
 301                 kfree(dev);
 302                 return ERR_PTR(-ENOMEM);
 303         }
 304
 305         INIT_LIST_HEAD(&dev->dev_list);
 306         INIT_LIST_HEAD(&dev->dev_alloc_list);
 307         INIT_LIST_HEAD(&dev->resized_list);
 308
 309         spin_lock_init(&dev->io_lock);
 310
 311         atomic_set(&dev->reada_in_flight, 0);
 312         atomic_set(&dev->dev_stats_ccnt, 0);
 313         btrfs_device_data_ordered_init(dev);
 314         INIT_RADIX_TREE(&dev->reada_zones, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
 315         INIT_RADIX_TREE(&dev->reada_extents, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
 316
 317         return dev;
 318 }
 319
 320 /*
 321  * Find a device specified by @devid or @uuid in the list of @fs_devices, or
 322  * return NULL.
 323  *
 324  * If devid and uuid are both specified, the match must be exact, otherwise
 325  * only devid is used.
 326  */
 327 static struct btrfs_device *find_device(struct btrfs_fs_devices *fs_devices,
 328                 u64 devid, const u8 *uuid)
 329 {
 330         struct list_head *head = &fs_devices->devices;
 331         struct btrfs_device *dev;
 332
 333         list_for_each_entry(dev, head, dev_list) {
 334                 if (dev->devid == devid &&
 335                     (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
 336                         return dev;
 337                 }
 338         }
 339         return NULL;
 340 }
 341
 342 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
 343 {
 344         struct btrfs_fs_devices *fs_devices;
 345
 346         list_for_each_entry(fs_devices, &fs_uuids, list) {
 347                 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
 348                         return fs_devices;
 349         }
 350         return NULL;
 351 }
 352
 353 static int
 354 btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder,
 355                       int flush, struct block_device **bdev,
 356                       struct buffer_head **bh)
 357 {
 358         int ret;
 359
 360         *bdev = blkdev_get_by_path(device_path, flags, holder);
 361
 362         if (IS_ERR(*bdev)) {
 363                 ret = PTR_ERR(*bdev);
 364                 goto error;
 365         }
 366
 367         if (flush)
 368                 filemap_write_and_wait((*bdev)->bd_inode->i_mapping);
 369         ret = set_blocksize(*bdev, BTRFS_BDEV_BLOCKSIZE);
 370         if (ret) {
 371                 blkdev_put(*bdev, flags);
 372                 goto error;
 373         }
 374         invalidate_bdev(*bdev);
 375         *bh = btrfs_read_dev_super(*bdev);
 376         if (IS_ERR(*bh)) {
 377                 ret = PTR_ERR(*bh);
 378                 blkdev_put(*bdev, flags);
 379                 goto error;
 380         }
 381
 382         return 0;
 383
 384 error:
 385         *bdev = NULL;
 386         *bh = NULL;
 387         return ret;
 388 }
 389
 390 static void requeue_list(struct btrfs_pending_bios *pending_bios,
 391                         struct bio *head, struct bio *tail)
 392 {
 393
 394         struct bio *old_head;
 395
 396         old_head = pending_bios->head;
 397         pending_bios->head = head;
 398         if (pending_bios->tail)
 399                 tail->bi_next = old_head;
 400         else
 401                 pending_bios->tail = tail;
 402 }
 403
 404 /*
 405  * we try to collect pending bios for a device so we don't get a large
 406  * number of procs sending bios down to the same device.  This greatly
 407  * improves the schedulers ability to collect and merge the bios.
 408  *
 409  * But, it also turns into a long list of bios to process and that is sure
 410  * to eventually make the worker thread block.  The solution here is to
 411  * make some progress and then put this work struct back at the end of
 412  * the list if the block device is congested.  This way, multiple devices
 413  * can make progress from a single worker thread.
 414  */
 415 static noinline void run_scheduled_bios(struct btrfs_device *device)
 416 {
 417         struct btrfs_fs_info *fs_info = device->fs_info;
 418         struct bio *pending;
 419         struct backing_dev_info *bdi;
 420         struct btrfs_pending_bios *pending_bios;
 421         struct bio *tail;
 422         struct bio *cur;
 423         int again = 0;
 424         unsigned long num_run;
 425         unsigned long batch_run = 0;
 426         unsigned long last_waited = 0;
 427         int force_reg = 0;
 428         int sync_pending = 0;
 429         struct blk_plug plug;
 430
 431         /*
 432          * this function runs all the bios we've collected for
 433          * a particular device.  We don't want to wander off to
 434          * another device without first sending all of these down.
 435          * So, setup a plug here and finish it off before we return
 436          */
 437         blk_start_plug(&plug);
 438
 439         bdi = device->bdev->bd_bdi;
 440
 441 loop:
 442         spin_lock(&device->io_lock);
 443
 444 loop_lock:
 445         num_run = 0;
 446
 447         /* take all the bios off the list at once and process them
 448          * later on (without the lock held).  But, remember the
 449          * tail and other pointers so the bios can be properly reinserted
 450          * into the list if we hit congestion
 451          */
 452         if (!force_reg && device->pending_sync_bios.head) {
 453                 pending_bios = &device->pending_sync_bios;
 454                 force_reg = 1;
 455         } else {
 456                 pending_bios = &device->pending_bios;
 457                 force_reg = 0;
 458         }
 459
 460         pending = pending_bios->head;
 461         tail = pending_bios->tail;
 462         WARN_ON(pending && !tail);
 463
 464         /*
 465          * if pending was null this time around, no bios need processing
 466          * at all and we can stop.  Otherwise it'll loop back up again
 467          * and do an additional check so no bios are missed.
 468          *
 469          * device->running_pending is used to synchronize with the
 470          * schedule_bio code.
 471          */
 472         if (device->pending_sync_bios.head == NULL &&
 473             device->pending_bios.head == NULL) {
 474                 again = 0;
 475                 device->running_pending = 0;
 476         } else {
 477                 again = 1;
 478                 device->running_pending = 1;
 479         }
 480
 481         pending_bios->head = NULL;
 482         pending_bios->tail = NULL;
 483
 484         spin_unlock(&device->io_lock);
 485
 486         while (pending) {
 487
 488                 rmb();
 489                 /* we want to work on both lists, but do more bios on the
 490                  * sync list than the regular list
 491                  */
 492                 if ((num_run > 32 &&
 493                     pending_bios != &device->pending_sync_bios &&
 494                     device->pending_sync_bios.head) ||
 495                    (num_run > 64 && pending_bios == &device->pending_sync_bios &&
 496                     device->pending_bios.head)) {
 497                         spin_lock(&device->io_lock);
 498                         requeue_list(pending_bios, pending, tail);
 499                         goto loop_lock;
 500                 }
 501
 502                 cur = pending;
 503                 pending = pending->bi_next;
 504                 cur->bi_next = NULL;
 505
 506                 BUG_ON(atomic_read(&cur->__bi_cnt) == 0);
 507
 508                 /*
 509                  * if we're doing the sync list, record that our
 510                  * plug has some sync requests on it
 511                  *
 512                  * If we're doing the regular list and there are
 513                  * sync requests sitting around, unplug before
 514                  * we add more
 515                  */
 516                 if (pending_bios == &device->pending_sync_bios) {
 517                         sync_pending = 1;
 518                 } else if (sync_pending) {
 519                         blk_finish_plug(&plug);
 520                         blk_start_plug(&plug);
 521                         sync_pending = 0;
 522                 }
 523
 524                 btrfsic_submit_bio(cur);
 525                 num_run++;
 526                 batch_run++;
 527
 528                 cond_resched();
 529
 530                 /*
 531                  * we made progress, there is more work to do and the bdi
 532                  * is now congested.  Back off and let other work structs
 533                  * run instead
 534                  */
 535                 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
 536                     fs_info->fs_devices->open_devices > 1) {
 537                         struct io_context *ioc;
 538
 539                         ioc = current->io_context;
 540
 541                         /*
 542                          * the main goal here is that we don't want to
 543                          * block if we're going to be able to submit
 544                          * more requests without blocking.
 545                          *
 546                          * This code does two great things, it pokes into
 547                          * the elevator code from a filesystem _and_
 548                          * it makes assumptions about how batching works.
 549                          */
 550                         if (ioc && ioc->nr_batch_requests > 0 &&
 551                             time_before(jiffies, ioc->last_waited + HZ/50UL) &&
 552                             (last_waited == 0 ||
 553                              ioc->last_waited == last_waited)) {
 554                                 /*
 555                                  * we want to go through our batch of
 556                                  * requests and stop.  So, we copy out
 557                                  * the ioc->last_waited time and test
 558                                  * against it before looping
 559                                  */
 560                                 last_waited = ioc->last_waited;
 561                                 cond_resched();
 562                                 continue;
 563                         }
 564                         spin_lock(&device->io_lock);
 565                         requeue_list(pending_bios, pending, tail);
 566                         device->running_pending = 1;
 567
 568                         spin_unlock(&device->io_lock);
 569                         btrfs_queue_work(fs_info->submit_workers,
 570                                          &device->work);
 571                         goto done;
 572                 }
 573         }
 574
 575         cond_resched();
 576         if (again)
 577                 goto loop;
 578
 579         spin_lock(&device->io_lock);
 580         if (device->pending_bios.head || device->pending_sync_bios.head)
 581                 goto loop_lock;
 582         spin_unlock(&device->io_lock);
 583
 584 done:
 585         blk_finish_plug(&plug);
 586 }
 587
 588 static void pending_bios_fn(struct btrfs_work *work)
 589 {
 590         struct btrfs_device *device;
 591
 592         device = container_of(work, struct btrfs_device, work);
 593         run_scheduled_bios(device);
 594 }
 595
 596 /*
 597  *  Search and remove all stale (devices which are not mounted) devices.
 598  *  When both inputs are NULL, it will search and release all stale devices.
 599  *  path:       Optional. When provided will it release all unmounted devices
 600  *              matching this path only.
 601  *  skip_dev:   Optional. Will skip this device when searching for the stale
 602  *              devices.
 603  */
 604 static void btrfs_free_stale_devices(const char *path,
 605                                      struct btrfs_device *skip_dev)
 606 {
 607         struct btrfs_fs_devices *fs_devs, *tmp_fs_devs;
 608         struct btrfs_device *dev, *tmp_dev;
 609
 610         list_for_each_entry_safe(fs_devs, tmp_fs_devs, &fs_uuids, list) {
 611
 612                 if (fs_devs->opened)
 613                         continue;
 614
 615                 list_for_each_entry_safe(dev, tmp_dev,
 616                                          &fs_devs->devices, dev_list) {
 617                         int not_found = 0;
 618
 619                         if (skip_dev && skip_dev == dev)
 620                                 continue;
 621                         if (path && !dev->name)
 622                                 continue;
 623
 624                         rcu_read_lock();
 625                         if (path)
 626                                 not_found = strcmp(rcu_str_deref(dev->name),
 627                                                    path);
 628                         rcu_read_unlock();
 629                         if (not_found)
 630                                 continue;
 631
 632                         /* delete the stale device */
 633                         if (fs_devs->num_devices == 1) {
 634                                 btrfs_sysfs_remove_fsid(fs_devs);
 635                                 list_del(&fs_devs->list);
 636                                 free_fs_devices(fs_devs);
 637                                 break;
 638                         } else {
 639                                 fs_devs->num_devices--;
 640                                 list_del(&dev->dev_list);
 641                                 free_device(dev);
 642                         }
 643                 }
 644         }
 645 }
 646
 647 static int btrfs_open_one_device(struct btrfs_fs_devices *fs_devices,
 648                         struct btrfs_device *device, fmode_t flags,
 649                         void *holder)
 650 {
 651         struct request_queue *q;
 652         struct block_device *bdev;
 653         struct buffer_head *bh;
 654         struct btrfs_super_block *disk_super;
 655         u64 devid;
 656         int ret;
 657
 658         if (device->bdev)
 659                 return -EINVAL;
 660         if (!device->name)
 661                 return -EINVAL;
 662
 663         ret = btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1,
 664                                     &bdev, &bh);
 665         if (ret)
 666                 return ret;
 667
 668         disk_super = (struct btrfs_super_block *)bh->b_data;
 669         devid = btrfs_stack_device_id(&disk_super->dev_item);
 670         if (devid != device->devid)
 671                 goto error_brelse;
 672
 673         if (memcmp(device->uuid, disk_super->dev_item.uuid, BTRFS_UUID_SIZE))
 674                 goto error_brelse;
 675
 676         device->generation = btrfs_super_generation(disk_super);
 677
 678         if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
 679                 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
 680                 fs_devices->seeding = 1;
 681         } else {
 682                 if (bdev_read_only(bdev))
 683                         clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
 684                 else
 685                         set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
 686         }
 687
 688         q = bdev_get_queue(bdev);
 689         if (!blk_queue_nonrot(q))
 690                 fs_devices->rotating = 1;
 691
 692         device->bdev = bdev;
 693         clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
 694         device->mode = flags;
 695
 696         fs_devices->open_devices++;
 697         if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
 698             device->devid != BTRFS_DEV_REPLACE_DEVID) {
 699                 fs_devices->rw_devices++;
 700                 list_add_tail(&device->dev_alloc_list, &fs_devices->alloc_list);
 701         }
 702         brelse(bh);
 703
 704         return 0;
 705
 706 error_brelse:
 707         brelse(bh);
 708         blkdev_put(bdev, flags);
 709
 710         return -EINVAL;
 711 }
 712
 713 /*
 714  * Add new device to list of registered devices
 715  *
 716  * Returns:
 717  * device pointer which was just added or updated when successful
 718  * error pointer when failed
 719  */
 720 static noinline struct btrfs_device *device_list_add(const char *path,
 721                            struct btrfs_super_block *disk_super)
 722 {
 723         struct btrfs_device *device;
 724         struct btrfs_fs_devices *fs_devices;
 725         struct rcu_string *name;
 726         u64 found_transid = btrfs_super_generation(disk_super);
 727         u64 devid = btrfs_stack_device_id(&disk_super->dev_item);
 728
 729         fs_devices = find_fsid(disk_super->fsid);
 730         if (!fs_devices) {
 731                 fs_devices = alloc_fs_devices(disk_super->fsid);
 732                 if (IS_ERR(fs_devices))
 733                         return ERR_CAST(fs_devices);
 734
 735                 list_add(&fs_devices->list, &fs_uuids);
 736
 737                 device = NULL;
 738         } else {
 739                 device = find_device(fs_devices, devid,
 740                                 disk_super->dev_item.uuid);
 741         }
 742
 743         if (!device) {
 744                 if (fs_devices->opened)
 745                         return ERR_PTR(-EBUSY);
 746
 747                 device = btrfs_alloc_device(NULL, &devid,
 748                                             disk_super->dev_item.uuid);
 749                 if (IS_ERR(device)) {
 750                         /* we can safely leave the fs_devices entry around */
 751                         return device;
 752                 }
 753
 754                 name = rcu_string_strdup(path, GFP_NOFS);
 755                 if (!name) {
 756                         free_device(device);
 757                         return ERR_PTR(-ENOMEM);
 758                 }
 759                 rcu_assign_pointer(device->name, name);
 760
 761                 mutex_lock(&fs_devices->device_list_mutex);
 762                 list_add_rcu(&device->dev_list, &fs_devices->devices);
 763                 fs_devices->num_devices++;
 764                 mutex_unlock(&fs_devices->device_list_mutex);
 765
 766                 device->fs_devices = fs_devices;
 767                 btrfs_free_stale_devices(path, device);
 768
 769                 if (disk_super->label[0])
 770                         pr_info("BTRFS: device label %s devid %llu transid %llu %s\n",
 771                                 disk_super->label, devid, found_transid, path);
 772                 else
 773                         pr_info("BTRFS: device fsid %pU devid %llu transid %llu %s\n",
 774                                 disk_super->fsid, devid, found_transid, path);
 775
 776         } else if (!device->name || strcmp(device->name->str, path)) {
 777                 /*
 778                  * When FS is already mounted.
 779                  * 1. If you are here and if the device->name is NULL that
 780                  *    means this device was missing at time of FS mount.
 781                  * 2. If you are here and if the device->name is different
 782                  *    from 'path' that means either
 783                  *      a. The same device disappeared and reappeared with
 784                  *         different name. or
 785                  *      b. The missing-disk-which-was-replaced, has
 786                  *         reappeared now.
 787                  *
 788                  * We must allow 1 and 2a above. But 2b would be a spurious
 789                  * and unintentional.
 790                  *
 791                  * Further in case of 1 and 2a above, the disk at 'path'
 792                  * would have missed some transaction when it was away and
 793                  * in case of 2a the stale bdev has to be updated as well.
 794                  * 2b must not be allowed at all time.
 795                  */
 796
 797                 /*
 798                  * For now, we do allow update to btrfs_fs_device through the
 799                  * btrfs dev scan cli after FS has been mounted.  We're still
 800                  * tracking a problem where systems fail mount by subvolume id
 801                  * when we reject replacement on a mounted FS.
 802                  */
 803                 if (!fs_devices->opened && found_transid < device->generation) {
 804                         /*
 805                          * That is if the FS is _not_ mounted and if you
 806                          * are here, that means there is more than one
 807                          * disk with same uuid and devid.We keep the one
 808                          * with larger generation number or the last-in if
 809                          * generation are equal.
 810                          */
 811                         return ERR_PTR(-EEXIST);
 812                 }
 813
 814                 name = rcu_string_strdup(path, GFP_NOFS);
 815                 if (!name)
 816                         return ERR_PTR(-ENOMEM);
 817                 rcu_string_free(device->name);
 818                 rcu_assign_pointer(device->name, name);
 819                 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) {
 820                         fs_devices->missing_devices--;
 821                         clear_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
 822                 }
 823         }
 824
 825         /*
 826          * Unmount does not free the btrfs_device struct but would zero
 827          * generation along with most of the other members. So just update
 828          * it back. We need it to pick the disk with largest generation
 829          * (as above).
 830          */
 831         if (!fs_devices->opened)
 832                 device->generation = found_transid;
 833
 834         fs_devices->total_devices = btrfs_super_num_devices(disk_super);
 835
 836         return device;
 837 }
 838
 839 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
 840 {
 841         struct btrfs_fs_devices *fs_devices;
 842         struct btrfs_device *device;
 843         struct btrfs_device *orig_dev;
 844
 845         fs_devices = alloc_fs_devices(orig->fsid);
 846         if (IS_ERR(fs_devices))
 847                 return fs_devices;
 848
 849         mutex_lock(&orig->device_list_mutex);
 850         fs_devices->total_devices = orig->total_devices;
 851
 852         /* We have held the volume lock, it is safe to get the devices. */
 853         list_for_each_entry(orig_dev, &orig->devices, dev_list) {
 854                 struct rcu_string *name;
 855
 856                 device = btrfs_alloc_device(NULL, &orig_dev->devid,
 857                                             orig_dev->uuid);
 858                 if (IS_ERR(device))
 859                         goto error;
 860
 861                 /*
 862                  * This is ok to do without rcu read locked because we hold the
 863                  * uuid mutex so nothing we touch in here is going to disappear.
 864                  */
 865                 if (orig_dev->name) {
 866                         name = rcu_string_strdup(orig_dev->name->str,
 867                                         GFP_KERNEL);
 868                         if (!name) {
 869                                 free_device(device);
 870                                 goto error;
 871                         }
 872                         rcu_assign_pointer(device->name, name);
 873                 }
 874
 875                 list_add(&device->dev_list, &fs_devices->devices);
 876                 device->fs_devices = fs_devices;
 877                 fs_devices->num_devices++;
 878         }
 879         mutex_unlock(&orig->device_list_mutex);
 880         return fs_devices;
 881 error:
 882         mutex_unlock(&orig->device_list_mutex);
 883         free_fs_devices(fs_devices);
 884         return ERR_PTR(-ENOMEM);
 885 }
 886
 887 /*
 888  * After we have read the system tree and know devids belonging to
 889  * this filesystem, remove the device which does not belong there.
 890  */
 891 void btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices, int step)
 892 {
 893         struct btrfs_device *device, *next;
 894         struct btrfs_device *latest_dev = NULL;
 895
 896         mutex_lock(&uuid_mutex);
 897 again:
 898         /* This is the initialized path, it is safe to release the devices. */
 899         list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
 900                 if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
 901                                                         &device->dev_state)) {
 902                         if (!test_bit(BTRFS_DEV_STATE_REPLACE_TGT,
 903                              &device->dev_state) &&
 904                              (!latest_dev ||
 905                               device->generation > latest_dev->generation)) {
 906                                 latest_dev = device;
 907                         }
 908                         continue;
 909                 }
 910
 911                 if (device->devid == BTRFS_DEV_REPLACE_DEVID) {
 912                         /*
 913                          * In the first step, keep the device which has
 914                          * the correct fsid and the devid that is used
 915                          * for the dev_replace procedure.
 916                          * In the second step, the dev_replace state is
 917                          * read from the device tree and it is known
 918                          * whether the procedure is really active or
 919                          * not, which means whether this device is
 920                          * used or whether it should be removed.
 921                          */
 922                         if (step == 0 || test_bit(BTRFS_DEV_STATE_REPLACE_TGT,
 923                                                   &device->dev_state)) {
 924                                 continue;
 925                         }
 926                 }
 927                 if (device->bdev) {
 928                         blkdev_put(device->bdev, device->mode);
 929                         device->bdev = NULL;
 930                         fs_devices->open_devices--;
 931                 }
 932                 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
 933                         list_del_init(&device->dev_alloc_list);
 934                         clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
 935                         if (!test_bit(BTRFS_DEV_STATE_REPLACE_TGT,
 936                                       &device->dev_state))
 937                                 fs_devices->rw_devices--;
 938                 }
 939                 list_del_init(&device->dev_list);
 940                 fs_devices->num_devices--;
 941                 free_device(device);
 942         }
 943
 944         if (fs_devices->seed) {
 945                 fs_devices = fs_devices->seed;
 946                 goto again;
 947         }
 948
 949         fs_devices->latest_bdev = latest_dev->bdev;
 950
 951         mutex_unlock(&uuid_mutex);
 952 }
 953
 954 static void free_device_rcu(struct rcu_head *head)
 955 {
 956         struct btrfs_device *device;
 957
 958         device = container_of(head, struct btrfs_device, rcu);
 959         free_device(device);
 960 }
 961
 962 static void btrfs_close_bdev(struct btrfs_device *device)
 963 {
 964         if (!device->bdev)
 965                 return;
 966
 967         if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
 968                 sync_blockdev(device->bdev);
 969                 invalidate_bdev(device->bdev);
 970         }
 971
 972         blkdev_put(device->bdev, device->mode);
 973 }
 974
 975 static void btrfs_prepare_close_one_device(struct btrfs_device *device)
 976 {
 977         struct btrfs_fs_devices *fs_devices = device->fs_devices;
 978         struct btrfs_device *new_device;
 979         struct rcu_string *name;
 980
 981         if (device->bdev)
 982                 fs_devices->open_devices--;
 983
 984         if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
 985             device->devid != BTRFS_DEV_REPLACE_DEVID) {
 986                 list_del_init(&device->dev_alloc_list);
 987                 fs_devices->rw_devices--;
 988         }
 989
 990         if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state))
 991                 fs_devices->missing_devices--;
 992
 993         new_device = btrfs_alloc_device(NULL, &device->devid,
 994                                         device->uuid);
 995         BUG_ON(IS_ERR(new_device)); /* -ENOMEM */
 996
 997         /* Safe because we are under uuid_mutex */
 998         if (device->name) {
 999                 name = rcu_string_strdup(device->name->str, GFP_NOFS);
1000                 BUG_ON(!name); /* -ENOMEM */
1001                 rcu_assign_pointer(new_device->name, name);
1002         }
1003
1004         list_replace_rcu(&device->dev_list, &new_device->dev_list);
1005         new_device->fs_devices = device->fs_devices;
1006 }
1007
1008 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
1009 {
1010         struct btrfs_device *device, *tmp;
1011         struct list_head pending_put;
1012
1013         INIT_LIST_HEAD(&pending_put);
1014
1015         if (--fs_devices->opened > 0)
1016                 return 0;
1017
1018         mutex_lock(&fs_devices->device_list_mutex);
1019         list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list) {
1020                 btrfs_prepare_close_one_device(device);
1021                 list_add(&device->dev_list, &pending_put);
1022         }
1023         mutex_unlock(&fs_devices->device_list_mutex);
1024
1025         /*
1026          * btrfs_show_devname() is using the device_list_mutex,
1027          * sometimes call to blkdev_put() leads vfs calling
1028          * into this func. So do put outside of device_list_mutex,
1029          * as of now.
1030          */
1031         while (!list_empty(&pending_put)) {
1032                 device = list_first_entry(&pending_put,
1033                                 struct btrfs_device, dev_list);
1034                 list_del(&device->dev_list);
1035                 btrfs_close_bdev(device);
1036                 call_rcu(&device->rcu, free_device_rcu);
1037         }
1038
1039         WARN_ON(fs_devices->open_devices);
1040         WARN_ON(fs_devices->rw_devices);
1041         fs_devices->opened = 0;
1042         fs_devices->seeding = 0;
1043
1044         return 0;
1045 }
1046
1047 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
1048 {
1049         struct btrfs_fs_devices *seed_devices = NULL;
1050         int ret;
1051
1052         mutex_lock(&uuid_mutex);
1053         ret = __btrfs_close_devices(fs_devices);
1054         if (!fs_devices->opened) {
1055                 seed_devices = fs_devices->seed;
1056                 fs_devices->seed = NULL;
1057         }
1058         mutex_unlock(&uuid_mutex);
1059
1060         while (seed_devices) {
1061                 fs_devices = seed_devices;
1062                 seed_devices = fs_devices->seed;
1063                 __btrfs_close_devices(fs_devices);
1064                 free_fs_devices(fs_devices);
1065         }
1066         return ret;
1067 }
1068
1069 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
1070                                 fmode_t flags, void *holder)
1071 {
1072         struct list_head *head = &fs_devices->devices;
1073         struct btrfs_device *device;
1074         struct btrfs_device *latest_dev = NULL;
1075         int ret = 0;
1076
1077         flags |= FMODE_EXCL;
1078
1079         list_for_each_entry(device, head, dev_list) {
1080                 /* Just open everything we can; ignore failures here */
1081                 if (btrfs_open_one_device(fs_devices, device, flags, holder))
1082                         continue;
1083
1084                 if (!latest_dev ||
1085                     device->generation > latest_dev->generation)
1086                         latest_dev = device;
1087         }
1088         if (fs_devices->open_devices == 0) {
1089                 ret = -EINVAL;
1090                 goto out;
1091         }
1092         fs_devices->opened = 1;
1093         fs_devices->latest_bdev = latest_dev->bdev;
1094         fs_devices->total_rw_bytes = 0;
1095 out:
1096         return ret;
1097 }
1098
1099 static int devid_cmp(void *priv, struct list_head *a, struct list_head *b)
1100 {
1101         struct btrfs_device *dev1, *dev2;
1102
1103         dev1 = list_entry(a, struct btrfs_device, dev_list);
1104         dev2 = list_entry(b, struct btrfs_device, dev_list);
1105
1106         if (dev1->devid < dev2->devid)
1107                 return -1;
1108         else if (dev1->devid > dev2->devid)
1109                 return 1;
1110         return 0;
1111 }
1112
1113 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
1114                        fmode_t flags, void *holder)
1115 {
1116         int ret;
1117
1118         mutex_lock(&uuid_mutex);
1119         if (fs_devices->opened) {
1120                 fs_devices->opened++;
1121                 ret = 0;
1122         } else {
1123                 list_sort(NULL, &fs_devices->devices, devid_cmp);
1124                 ret = __btrfs_open_devices(fs_devices, flags, holder);
1125         }
1126         mutex_unlock(&uuid_mutex);
1127         return ret;
1128 }
1129
1130 static void btrfs_release_disk_super(struct page *page)
1131 {
1132         kunmap(page);
1133         put_page(page);
1134 }
1135
1136 static int btrfs_read_disk_super(struct block_device *bdev, u64 bytenr,
1137                                  struct page **page,
1138                                  struct btrfs_super_block **disk_super)
1139 {
1140         void *p;
1141         pgoff_t index;
1142
1143         /* make sure our super fits in the device */
1144         if (bytenr + PAGE_SIZE >= i_size_read(bdev->bd_inode))
1145                 return 1;
1146
1147         /* make sure our super fits in the page */
1148         if (sizeof(**disk_super) > PAGE_SIZE)
1149                 return 1;
1150
1151         /* make sure our super doesn't straddle pages on disk */
1152         index = bytenr >> PAGE_SHIFT;
1153         if ((bytenr + sizeof(**disk_super) - 1) >> PAGE_SHIFT != index)
1154                 return 1;
1155
1156         /* pull in the page with our super */
1157         *page = read_cache_page_gfp(bdev->bd_inode->i_mapping,
1158                                    index, GFP_KERNEL);
1159
1160         if (IS_ERR_OR_NULL(*page))
1161                 return 1;
1162
1163         p = kmap(*page);
1164
1165         /* align our pointer to the offset of the super block */
1166         *disk_super = p + (bytenr & ~PAGE_MASK);
1167
1168         if (btrfs_super_bytenr(*disk_super) != bytenr ||
1169             btrfs_super_magic(*disk_super) != BTRFS_MAGIC) {
1170                 btrfs_release_disk_super(*page);
1171                 return 1;
1172         }
1173
1174         if ((*disk_super)->label[0] &&
1175                 (*disk_super)->label[BTRFS_LABEL_SIZE - 1])
1176                 (*disk_super)->label[BTRFS_LABEL_SIZE - 1] = '\0';
1177
1178         return 0;
1179 }
1180
1181 /*
1182  * Look for a btrfs signature on a device. This may be called out of the mount path
1183  * and we are not allowed to call set_blocksize during the scan. The superblock
1184  * is read via pagecache
1185  */
1186 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
1187                           struct btrfs_fs_devices **fs_devices_ret)
1188 {
1189         struct btrfs_super_block *disk_super;
1190         struct btrfs_device *device;
1191         struct block_device *bdev;
1192         struct page *page;
1193         int ret = 0;
1194         u64 bytenr;
1195
1196         /*
1197          * we would like to check all the supers, but that would make
1198          * a btrfs mount succeed after a mkfs from a different FS.
1199          * So, we need to add a special mount option to scan for
1200          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
1201          */
1202         bytenr = btrfs_sb_offset(0);
1203         flags |= FMODE_EXCL;
1204         mutex_lock(&uuid_mutex);
1205
1206         bdev = blkdev_get_by_path(path, flags, holder);
1207         if (IS_ERR(bdev)) {
1208                 ret = PTR_ERR(bdev);
1209                 goto error;
1210         }
1211
1212         if (btrfs_read_disk_super(bdev, bytenr, &page, &disk_super)) {
1213                 ret = -EINVAL;
1214                 goto error_bdev_put;
1215         }
1216
1217         device = device_list_add(path, disk_super);
1218         if (IS_ERR(device))
1219                 ret = PTR_ERR(device);
1220         else
1221                 *fs_devices_ret = device->fs_devices;
1222
1223         btrfs_release_disk_super(page);
1224
1225 error_bdev_put:
1226         blkdev_put(bdev, flags);
1227 error:
1228         mutex_unlock(&uuid_mutex);
1229         return ret;
1230 }
1231
1232 /* helper to account the used device space in the range */
1233 int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
1234                                    u64 end, u64 *length)
1235 {
1236         struct btrfs_key key;
1237         struct btrfs_root *root = device->fs_info->dev_root;
1238         struct btrfs_dev_extent *dev_extent;
1239         struct btrfs_path *path;
1240         u64 extent_end;
1241         int ret;
1242         int slot;
1243         struct extent_buffer *l;
1244
1245         *length = 0;
1246
1247         if (start >= device->total_bytes ||
1248                 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
1249                 return 0;
1250
1251         path = btrfs_alloc_path();
1252         if (!path)
1253                 return -ENOMEM;
1254         path->reada = READA_FORWARD;
1255
1256         key.objectid = device->devid;
1257         key.offset = start;
1258         key.type = BTRFS_DEV_EXTENT_KEY;
1259
1260         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1261         if (ret < 0)
1262                 goto out;
1263         if (ret > 0) {
1264                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1265                 if (ret < 0)
1266                         goto out;
1267         }
1268
1269         while (1) {
1270                 l = path->nodes[0];
1271                 slot = path->slots[0];
1272                 if (slot >= btrfs_header_nritems(l)) {
1273                         ret = btrfs_next_leaf(root, path);
1274                         if (ret == 0)
1275                                 continue;
1276                         if (ret < 0)
1277                                 goto out;
1278
1279                         break;
1280                 }
1281                 btrfs_item_key_to_cpu(l, &key, slot);
1282
1283                 if (key.objectid < device->devid)
1284                         goto next;
1285
1286                 if (key.objectid > device->devid)
1287                         break;
1288
1289                 if (key.type != BTRFS_DEV_EXTENT_KEY)
1290                         goto next;
1291
1292                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1293                 extent_end = key.offset + btrfs_dev_extent_length(l,
1294                                                                   dev_extent);
1295                 if (key.offset <= start && extent_end > end) {
1296                         *length = end - start + 1;
1297                         break;
1298                 } else if (key.offset <= start && extent_end > start)
1299                         *length += extent_end - start;
1300                 else if (key.offset > start && extent_end <= end)
1301                         *length += extent_end - key.offset;
1302                 else if (key.offset > start && key.offset <= end) {
1303                         *length += end - key.offset + 1;
1304                         break;
1305                 } else if (key.offset > end)
1306                         break;
1307
1308 next:
1309                 path->slots[0]++;
1310         }
1311         ret = 0;
1312 out:
1313         btrfs_free_path(path);
1314         return ret;
1315 }
1316
1317 static int contains_pending_extent(struct btrfs_transaction *transaction,
1318                                    struct btrfs_device *device,
1319                                    u64 *start, u64 len)
1320 {
1321         struct btrfs_fs_info *fs_info = device->fs_info;
1322         struct extent_map *em;
1323         struct list_head *search_list = &fs_info->pinned_chunks;
1324         int ret = 0;
1325         u64 physical_start = *start;
1326
1327         if (transaction)
1328                 search_list = &transaction->pending_chunks;
1329 again:
1330         list_for_each_entry(em, search_list, list) {
1331                 struct map_lookup *map;
1332                 int i;
1333
1334                 map = em->map_lookup;
1335                 for (i = 0; i < map->num_stripes; i++) {
1336                         u64 end;
1337
1338                         if (map->stripes[i].dev != device)
1339                                 continue;
1340                         if (map->stripes[i].physical >= physical_start + len ||
1341                             map->stripes[i].physical + em->orig_block_len <=
1342                             physical_start)
1343                                 continue;
1344                         /*
1345                          * Make sure that while processing the pinned list we do
1346                          * not override our *start with a lower value, because
1347                          * we can have pinned chunks that fall within this
1348                          * device hole and that have lower physical addresses
1349                          * than the pending chunks we processed before. If we
1350                          * do not take this special care we can end up getting
1351                          * 2 pending chunks that start at the same physical
1352                          * device offsets because the end offset of a pinned
1353                          * chunk can be equal to the start offset of some
1354                          * pending chunk.
1355                          */
1356                         end = map->stripes[i].physical + em->orig_block_len;
1357                         if (end > *start) {
1358                                 *start = end;
1359                                 ret = 1;
1360                         }
1361                 }
1362         }
1363         if (search_list != &fs_info->pinned_chunks) {
1364                 search_list = &fs_info->pinned_chunks;
1365                 goto again;
1366         }
1367
1368         return ret;
1369 }
1370
1371
1372 /*
1373  * find_free_dev_extent_start - find free space in the specified device
1374  * @device:       the device which we search the free space in
1375  * @num_bytes:    the size of the free space that we need
1376  * @search_start: the position from which to begin the search
1377  * @start:        store the start of the free space.
1378  * @len:          the size of the free space. that we find, or the size
1379  *                of the max free space if we don't find suitable free space
1380  *
1381  * this uses a pretty simple search, the expectation is that it is
1382  * called very infrequently and that a given device has a small number
1383  * of extents
1384  *
1385  * @start is used to store the start of the free space if we find. But if we
1386  * don't find suitable free space, it will be used to store the start position
1387  * of the max free space.
1388  *
1389  * @len is used to store the size of the free space that we find.
1390  * But if we don't find suitable free space, it is used to store the size of
1391  * the max free space.
1392  */
1393 int find_free_dev_extent_start(struct btrfs_transaction *transaction,
1394                                struct btrfs_device *device, u64 num_bytes,
1395                                u64 search_start, u64 *start, u64 *len)
1396 {
1397         struct btrfs_fs_info *fs_info = device->fs_info;
1398         struct btrfs_root *root = fs_info->dev_root;
1399         struct btrfs_key key;
1400         struct btrfs_dev_extent *dev_extent;
1401         struct btrfs_path *path;
1402         u64 hole_size;
1403         u64 max_hole_start;
1404         u64 max_hole_size;
1405         u64 extent_end;
1406         u64 search_end = device->total_bytes;
1407         int ret;
1408         int slot;
1409         struct extent_buffer *l;
1410
1411         /*
1412          * We don't want to overwrite the superblock on the drive nor any area
1413          * used by the boot loader (grub for example), so we make sure to start
1414          * at an offset of at least 1MB.
1415          */
1416         search_start = max_t(u64, search_start, SZ_1M);
1417
1418         path = btrfs_alloc_path();
1419         if (!path)
1420                 return -ENOMEM;
1421
1422         max_hole_start = search_start;
1423         max_hole_size = 0;
1424
1425 again:
1426         if (search_start >= search_end ||
1427                 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
1428                 ret = -ENOSPC;
1429                 goto out;
1430         }
1431
1432         path->reada = READA_FORWARD;
1433         path->search_commit_root = 1;
1434         path->skip_locking = 1;
1435
1436         key.objectid = device->devid;
1437         key.offset = search_start;
1438         key.type = BTRFS_DEV_EXTENT_KEY;
1439
1440         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1441         if (ret < 0)
1442                 goto out;
1443         if (ret > 0) {
1444                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1445                 if (ret < 0)
1446                         goto out;
1447         }
1448
1449         while (1) {
1450                 l = path->nodes[0];
1451                 slot = path->slots[0];
1452                 if (slot >= btrfs_header_nritems(l)) {
1453                         ret = btrfs_next_leaf(root, path);
1454                         if (ret == 0)
1455                                 continue;
1456                         if (ret < 0)
1457                                 goto out;
1458
1459                         break;
1460                 }
1461                 btrfs_item_key_to_cpu(l, &key, slot);
1462
1463                 if (key.objectid < device->devid)
1464                         goto next;
1465
1466                 if (key.objectid > device->devid)
1467                         break;
1468
1469                 if (key.type != BTRFS_DEV_EXTENT_KEY)
1470                         goto next;
1471
1472                 if (key.offset > search_start) {
1473                         hole_size = key.offset - search_start;
1474
1475                         /*
1476                          * Have to check before we set max_hole_start, otherwise
1477                          * we could end up sending back this offset anyway.
1478                          */
1479                         if (contains_pending_extent(transaction, device,
1480                                                     &search_start,
1481                                                     hole_size)) {
1482                                 if (key.offset >= search_start) {
1483                                         hole_size = key.offset - search_start;
1484                                 } else {
1485                                         WARN_ON_ONCE(1);
1486                                         hole_size = 0;
1487                                 }
1488                         }
1489
1490                         if (hole_size > max_hole_size) {
1491                                 max_hole_start = search_start;
1492                                 max_hole_size = hole_size;
1493                         }
1494
1495                         /*
1496                          * If this free space is greater than which we need,
1497                          * it must be the max free space that we have found
1498                          * until now, so max_hole_start must point to the start
1499                          * of this free space and the length of this free space
1500                          * is stored in max_hole_size. Thus, we return
1501                          * max_hole_start and max_hole_size and go back to the
1502                          * caller.
1503                          */
1504                         if (hole_size >= num_bytes) {
1505                                 ret = 0;
1506                                 goto out;
1507                         }
1508                 }
1509
1510                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1511                 extent_end = key.offset + btrfs_dev_extent_length(l,
1512                                                                   dev_extent);
1513                 if (extent_end > search_start)
1514                         search_start = extent_end;
1515 next:
1516                 path->slots[0]++;
1517                 cond_resched();
1518         }
1519
1520         /*
1521          * At this point, search_start should be the end of
1522          * allocated dev extents, and when shrinking the device,
1523          * search_end may be smaller than search_start.
1524          */
1525         if (search_end > search_start) {
1526                 hole_size = search_end - search_start;
1527
1528                 if (contains_pending_extent(transaction, device, &search_start,
1529                                             hole_size)) {
1530                         btrfs_release_path(path);
1531                         goto again;
1532                 }
1533
1534                 if (hole_size > max_hole_size) {
1535                         max_hole_start = search_start;
1536                         max_hole_size = hole_size;
1537                 }
1538         }
1539
1540         /* See above. */
1541         if (max_hole_size < num_bytes)
1542                 ret = -ENOSPC;
1543         else
1544                 ret = 0;
1545
1546 out:
1547         btrfs_free_path(path);
1548         *start = max_hole_start;
1549         if (len)
1550                 *len = max_hole_size;
1551         return ret;
1552 }
1553
1554 int find_free_dev_extent(struct btrfs_trans_handle *trans,
1555                          struct btrfs_device *device, u64 num_bytes,
1556                          u64 *start, u64 *len)
1557 {
1558         /* FIXME use last free of some kind */
1559         return find_free_dev_extent_start(trans->transaction, device,
1560                                           num_bytes, 0, start, len);
1561 }
1562
1563 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1564                           struct btrfs_device *device,
1565                           u64 start, u64 *dev_extent_len)
1566 {
1567         struct btrfs_fs_info *fs_info = device->fs_info;
1568         struct btrfs_root *root = fs_info->dev_root;
1569         int ret;
1570         struct btrfs_path *path;
1571         struct btrfs_key key;
1572         struct btrfs_key found_key;
1573         struct extent_buffer *leaf = NULL;
1574         struct btrfs_dev_extent *extent = NULL;
1575
1576         path = btrfs_alloc_path();
1577         if (!path)
1578                 return -ENOMEM;
1579
1580         key.objectid = device->devid;
1581         key.offset = start;
1582         key.type = BTRFS_DEV_EXTENT_KEY;
1583 again:
1584         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1585         if (ret > 0) {
1586                 ret = btrfs_previous_item(root, path, key.objectid,
1587                                           BTRFS_DEV_EXTENT_KEY);
1588                 if (ret)
1589                         goto out;
1590                 leaf = path->nodes[0];
1591                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1592                 extent = btrfs_item_ptr(leaf, path->slots[0],
1593                                         struct btrfs_dev_extent);
1594                 BUG_ON(found_key.offset > start || found_key.offset +
1595                        btrfs_dev_extent_length(leaf, extent) < start);
1596                 key = found_key;
1597                 btrfs_release_path(path);
1598                 goto again;
1599         } else if (ret == 0) {
1600                 leaf = path->nodes[0];
1601                 extent = btrfs_item_ptr(leaf, path->slots[0],
1602                                         struct btrfs_dev_extent);
1603         } else {
1604                 btrfs_handle_fs_error(fs_info, ret, "Slot search failed");
1605                 goto out;
1606         }
1607
1608         *dev_extent_len = btrfs_dev_extent_length(leaf, extent);
1609
1610         ret = btrfs_del_item(trans, root, path);
1611         if (ret) {
1612                 btrfs_handle_fs_error(fs_info, ret,
1613                                       "Failed to remove dev extent item");
1614         } else {
1615                 set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags);
1616         }
1617 out:
1618         btrfs_free_path(path);
1619         return ret;
1620 }
1621
1622 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1623                                   struct btrfs_device *device,
1624                                   u64 chunk_offset, u64 start, u64 num_bytes)
1625 {
1626         int ret;
1627         struct btrfs_path *path;
1628         struct btrfs_fs_info *fs_info = device->fs_info;
1629         struct btrfs_root *root = fs_info->dev_root;
1630         struct btrfs_dev_extent *extent;
1631         struct extent_buffer *leaf;
1632         struct btrfs_key key;
1633
1634         WARN_ON(!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state));
1635         WARN_ON(test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state));
1636         path = btrfs_alloc_path();
1637         if (!path)
1638                 return -ENOMEM;
1639
1640         key.objectid = device->devid;
1641         key.offset = start;
1642         key.type = BTRFS_DEV_EXTENT_KEY;
1643         ret = btrfs_insert_empty_item(trans, root, path, &key,
1644                                       sizeof(*extent));
1645         if (ret)
1646                 goto out;
1647
1648         leaf = path->nodes[0];
1649         extent = btrfs_item_ptr(leaf, path->slots[0],
1650                                 struct btrfs_dev_extent);
1651         btrfs_set_dev_extent_chunk_tree(leaf, extent,
1652                                         BTRFS_CHUNK_TREE_OBJECTID);
1653         btrfs_set_dev_extent_chunk_objectid(leaf, extent,
1654                                             BTRFS_FIRST_CHUNK_TREE_OBJECTID);
1655         btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1656
1657         btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1658         btrfs_mark_buffer_dirty(leaf);
1659 out:
1660         btrfs_free_path(path);
1661         return ret;
1662 }
1663
1664 static u64 find_next_chunk(struct btrfs_fs_info *fs_info)
1665 {
1666         struct extent_map_tree *em_tree;
1667         struct extent_map *em;
1668         struct rb_node *n;
1669         u64 ret = 0;
1670
1671         em_tree = &fs_info->mapping_tree.map_tree;
1672         read_lock(&em_tree->lock);
1673         n = rb_last(&em_tree->map);
1674         if (n) {
1675                 em = rb_entry(n, struct extent_map, rb_node);
1676                 ret = em->start + em->len;
1677         }
1678         read_unlock(&em_tree->lock);
1679
1680         return ret;
1681 }
1682
1683 static noinline int find_next_devid(struct btrfs_fs_info *fs_info,
1684                                     u64 *devid_ret)
1685 {
1686         int ret;
1687         struct btrfs_key key;
1688         struct btrfs_key found_key;
1689         struct btrfs_path *path;
1690
1691         path = btrfs_alloc_path();
1692         if (!path)
1693                 return -ENOMEM;
1694
1695         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1696         key.type = BTRFS_DEV_ITEM_KEY;
1697         key.offset = (u64)-1;
1698
1699         ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
1700         if (ret < 0)
1701                 goto error;
1702
1703         BUG_ON(ret == 0); /* Corruption */
1704
1705         ret = btrfs_previous_item(fs_info->chunk_root, path,
1706                                   BTRFS_DEV_ITEMS_OBJECTID,
1707                                   BTRFS_DEV_ITEM_KEY);
1708         if (ret) {
1709                 *devid_ret = 1;
1710         } else {
1711                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1712                                       path->slots[0]);
1713                 *devid_ret = found_key.offset + 1;
1714         }
1715         ret = 0;
1716 error:
1717         btrfs_free_path(path);
1718         return ret;
1719 }
1720
1721 /*
1722  * the device information is stored in the chunk root
1723  * the btrfs_device struct should be fully filled in
1724  */
1725 static int btrfs_add_dev_item(struct btrfs_trans_handle *trans,
1726                             struct btrfs_fs_info *fs_info,
1727                             struct btrfs_device *device)
1728 {
1729         struct btrfs_root *root = fs_info->chunk_root;
1730         int ret;
1731         struct btrfs_path *path;
1732         struct btrfs_dev_item *dev_item;
1733         struct extent_buffer *leaf;
1734         struct btrfs_key key;
1735         unsigned long ptr;
1736
1737         path = btrfs_alloc_path();
1738         if (!path)
1739                 return -ENOMEM;
1740
1741         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1742         key.type = BTRFS_DEV_ITEM_KEY;
1743         key.offset = device->devid;
1744
1745         ret = btrfs_insert_empty_item(trans, root, path, &key,
1746                                       sizeof(*dev_item));
1747         if (ret)
1748                 goto out;
1749
1750         leaf = path->nodes[0];
1751         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1752
1753         btrfs_set_device_id(leaf, dev_item, device->devid);
1754         btrfs_set_device_generation(leaf, dev_item, 0);
1755         btrfs_set_device_type(leaf, dev_item, device->type);
1756         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1757         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1758         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1759         btrfs_set_device_total_bytes(leaf, dev_item,
1760                                      btrfs_device_get_disk_total_bytes(device));
1761         btrfs_set_device_bytes_used(leaf, dev_item,
1762                                     btrfs_device_get_bytes_used(device));
1763         btrfs_set_device_group(leaf, dev_item, 0);
1764         btrfs_set_device_seek_speed(leaf, dev_item, 0);
1765         btrfs_set_device_bandwidth(leaf, dev_item, 0);
1766         btrfs_set_device_start_offset(leaf, dev_item, 0);
1767
1768         ptr = btrfs_device_uuid(dev_item);
1769         write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1770         ptr = btrfs_device_fsid(dev_item);
1771         write_extent_buffer(leaf, fs_info->fsid, ptr, BTRFS_FSID_SIZE);
1772         btrfs_mark_buffer_dirty(leaf);
1773
1774         ret = 0;
1775 out:
1776         btrfs_free_path(path);
1777         return ret;
1778 }
1779
1780 /*
1781  * Function to update ctime/mtime for a given device path.
1782  * Mainly used for ctime/mtime based probe like libblkid.
1783  */
1784 static void update_dev_time(const char *path_name)
1785 {
1786         struct file *filp;
1787
1788         filp = filp_open(path_name, O_RDWR, 0);
1789         if (IS_ERR(filp))
1790                 return;
1791         file_update_time(filp);
1792         filp_close(filp, NULL);
1793 }
1794
1795 static int btrfs_rm_dev_item(struct btrfs_fs_info *fs_info,
1796                              struct btrfs_device *device)
1797 {
1798         struct btrfs_root *root = fs_info->chunk_root;
1799         int ret;
1800         struct btrfs_path *path;
1801         struct btrfs_key key;
1802         struct btrfs_trans_handle *trans;
1803
1804         path = btrfs_alloc_path();
1805         if (!path)
1806                 return -ENOMEM;
1807
1808         trans = btrfs_start_transaction(root, 0);
1809         if (IS_ERR(trans)) {
1810                 btrfs_free_path(path);
1811                 return PTR_ERR(trans);
1812         }
1813         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1814         key.type = BTRFS_DEV_ITEM_KEY;
1815         key.offset = device->devid;
1816
1817         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1818         if (ret) {
1819                 if (ret > 0)
1820                         ret = -ENOENT;
1821                 btrfs_abort_transaction(trans, ret);
1822                 btrfs_end_transaction(trans);
1823                 goto out;
1824         }
1825
1826         ret = btrfs_del_item(trans, root, path);
1827         if (ret) {
1828                 btrfs_abort_transaction(trans, ret);
1829                 btrfs_end_transaction(trans);
1830         }
1831
1832 out:
1833         btrfs_free_path(path);
1834         if (!ret)
1835                 ret = btrfs_commit_transaction(trans);
1836         return ret;
1837 }
1838
1839 /*
1840  * Verify that @num_devices satisfies the RAID profile constraints in the whole
1841  * filesystem. It's up to the caller to adjust that number regarding eg. device
1842  * replace.
1843  */
1844 static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info,
1845                 u64 num_devices)
1846 {
1847         u64 all_avail;
1848         unsigned seq;
1849         int i;
1850
1851         do {
1852                 seq = read_seqbegin(&fs_info->profiles_lock);
1853
1854                 all_avail = fs_info->avail_data_alloc_bits |
1855                             fs_info->avail_system_alloc_bits |
1856                             fs_info->avail_metadata_alloc_bits;
1857         } while (read_seqretry(&fs_info->profiles_lock, seq));
1858
1859         for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
1860                 if (!(all_avail & btrfs_raid_group[i]))
1861                         continue;
1862
1863                 if (num_devices < btrfs_raid_array[i].devs_min) {
1864                         int ret = btrfs_raid_mindev_error[i];
1865
1866                         if (ret)
1867                                 return ret;
1868                 }
1869         }
1870
1871         return 0;
1872 }
1873
1874 static struct btrfs_device * btrfs_find_next_active_device(
1875                 struct btrfs_fs_devices *fs_devs, struct btrfs_device *device)
1876 {
1877         struct btrfs_device *next_device;
1878
1879         list_for_each_entry(next_device, &fs_devs->devices, dev_list) {
1880                 if (next_device != device &&
1881                     !test_bit(BTRFS_DEV_STATE_MISSING, &next_device->dev_state)
1882                     && next_device->bdev)
1883                         return next_device;
1884         }
1885
1886         return NULL;
1887 }
1888
1889 /*
1890  * Helper function to check if the given device is part of s_bdev / latest_bdev
1891  * and replace it with the provided or the next active device, in the context
1892  * where this function called, there should be always be another device (or
1893  * this_dev) which is active.
1894  */
1895 void btrfs_assign_next_active_device(struct btrfs_fs_info *fs_info,
1896                 struct btrfs_device *device, struct btrfs_device *this_dev)
1897 {
1898         struct btrfs_device *next_device;
1899
1900         if (this_dev)
1901                 next_device = this_dev;
1902         else
1903                 next_device = btrfs_find_next_active_device(fs_info->fs_devices,
1904                                                                 device);
1905         ASSERT(next_device);
1906
1907         if (fs_info->sb->s_bdev &&
1908                         (fs_info->sb->s_bdev == device->bdev))
1909                 fs_info->sb->s_bdev = next_device->bdev;
1910
1911         if (fs_info->fs_devices->latest_bdev == device->bdev)
1912                 fs_info->fs_devices->latest_bdev = next_device->bdev;
1913 }
1914
1915 int btrfs_rm_device(struct btrfs_fs_info *fs_info, const char *device_path,
1916                 u64 devid)
1917 {
1918         struct btrfs_device *device;
1919         struct btrfs_fs_devices *cur_devices;
1920         u64 num_devices;
1921         int ret = 0;
1922
1923         mutex_lock(&fs_info->volume_mutex);
1924         mutex_lock(&uuid_mutex);
1925
1926         num_devices = fs_info->fs_devices->num_devices;
1927         btrfs_dev_replace_read_lock(&fs_info->dev_replace);
1928         if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
1929                 WARN_ON(num_devices < 1);
1930                 num_devices--;
1931         }
1932         btrfs_dev_replace_read_unlock(&fs_info->dev_replace);
1933
1934         ret = btrfs_check_raid_min_devices(fs_info, num_devices - 1);
1935         if (ret)
1936                 goto out;
1937
1938         ret = btrfs_find_device_by_devspec(fs_info, devid, device_path,
1939                                            &device);
1940         if (ret)
1941                 goto out;
1942
1943         if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
1944                 ret = BTRFS_ERROR_DEV_TGT_REPLACE;
1945                 goto out;
1946         }
1947
1948         if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
1949             fs_info->fs_devices->rw_devices == 1) {
1950                 ret = BTRFS_ERROR_DEV_ONLY_WRITABLE;
1951                 goto out;
1952         }
1953
1954         if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
1955                 mutex_lock(&fs_info->chunk_mutex);
1956                 list_del_init(&device->dev_alloc_list);
1957                 device->fs_devices->rw_devices--;
1958                 mutex_unlock(&fs_info->chunk_mutex);
1959         }
1960
1961         mutex_unlock(&uuid_mutex);
1962         ret = btrfs_shrink_device(device, 0);
1963         mutex_lock(&uuid_mutex);
1964         if (ret)
1965                 goto error_undo;
1966
1967         /*
1968          * TODO: the superblock still includes this device in its num_devices
1969          * counter although write_all_supers() is not locked out. This
1970          * could give a filesystem state which requires a degraded mount.
1971          */
1972         ret = btrfs_rm_dev_item(fs_info, device);
1973         if (ret)
1974                 goto error_undo;
1975
1976         clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
1977         btrfs_scrub_cancel_dev(fs_info, device);
1978
1979         /*
1980          * the device list mutex makes sure that we don't change
1981          * the device list while someone else is writing out all
1982          * the device supers. Whoever is writing all supers, should
1983          * lock the device list mutex before getting the number of
1984          * devices in the super block (super_copy). Conversely,
1985          * whoever updates the number of devices in the super block
1986          * (super_copy) should hold the device list mutex.
1987          */
1988
1989         cur_devices = device->fs_devices;
1990         mutex_lock(&fs_info->fs_devices->device_list_mutex);
1991         list_del_rcu(&device->dev_list);
1992
1993         device->fs_devices->num_devices--;
1994         device->fs_devices->total_devices--;
1995
1996         if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state))
1997                 device->fs_devices->missing_devices--;
1998
1999         btrfs_assign_next_active_device(fs_info, device, NULL);
2000
2001         if (device->bdev) {
2002                 device->fs_devices->open_devices--;
2003                 /* remove sysfs entry */
2004                 btrfs_sysfs_rm_device_link(fs_info->fs_devices, device);
2005         }
2006
2007         num_devices = btrfs_super_num_devices(fs_info->super_copy) - 1;
2008         btrfs_set_super_num_devices(fs_info->super_copy, num_devices);
2009         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2010
2011         /*
2012          * at this point, the device is zero sized and detached from
2013          * the devices list.  All that's left is to zero out the old
2014          * supers and free the device.
2015          */
2016         if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
2017                 btrfs_scratch_superblocks(device->bdev, device->name->str);
2018
2019         btrfs_close_bdev(device);
2020         call_rcu(&device->rcu, free_device_rcu);
2021
2022         if (cur_devices->open_devices == 0) {
2023                 struct btrfs_fs_devices *fs_devices;
2024                 fs_devices = fs_info->fs_devices;
2025                 while (fs_devices) {
2026                         if (fs_devices->seed == cur_devices) {
2027                                 fs_devices->seed = cur_devices->seed;
2028                                 break;
2029                         }
2030                         fs_devices = fs_devices->seed;
2031                 }
2032                 cur_devices->seed = NULL;
2033                 __btrfs_close_devices(cur_devices);
2034                 free_fs_devices(cur_devices);
2035         }
2036
2037 out:
2038         mutex_unlock(&uuid_mutex);
2039         mutex_unlock(&fs_info->volume_mutex);
2040         return ret;
2041
2042 error_undo:
2043         if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
2044                 mutex_lock(&fs_info->chunk_mutex);
2045                 list_add(&device->dev_alloc_list,
2046                          &fs_info->fs_devices->alloc_list);
2047                 device->fs_devices->rw_devices++;
2048                 mutex_unlock(&fs_info->chunk_mutex);
2049         }
2050         goto out;
2051 }
2052
2053 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_fs_info *fs_info,
2054                                         struct btrfs_device *srcdev)
2055 {
2056         struct btrfs_fs_devices *fs_devices;
2057
2058         lockdep_assert_held(&fs_info->fs_devices->device_list_mutex);
2059
2060         /*
2061          * in case of fs with no seed, srcdev->fs_devices will point
2062          * to fs_devices of fs_info. However when the dev being replaced is
2063          * a seed dev it will point to the seed's local fs_devices. In short
2064          * srcdev will have its correct fs_devices in both the cases.
2065          */
2066         fs_devices = srcdev->fs_devices;
2067
2068         list_del_rcu(&srcdev->dev_list);
2069         list_del(&srcdev->dev_alloc_list);
2070         fs_devices->num_devices--;
2071         if (test_bit(BTRFS_DEV_STATE_MISSING, &srcdev->dev_state))
2072                 fs_devices->missing_devices--;
2073
2074         if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state))
2075                 fs_devices->rw_devices--;
2076
2077         if (srcdev->bdev)
2078                 fs_devices->open_devices--;
2079 }
2080
2081 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_fs_info *fs_info,
2082                                       struct btrfs_device *srcdev)
2083 {
2084         struct btrfs_fs_devices *fs_devices = srcdev->fs_devices;
2085
2086         if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state)) {
2087                 /* zero out the old super if it is writable */
2088                 btrfs_scratch_superblocks(srcdev->bdev, srcdev->name->str);
2089         }
2090
2091         btrfs_close_bdev(srcdev);
2092         call_rcu(&srcdev->rcu, free_device_rcu);
2093
2094         /* if this is no devs we rather delete the fs_devices */
2095         if (!fs_devices->num_devices) {
2096                 struct btrfs_fs_devices *tmp_fs_devices;
2097
2098                 /*
2099                  * On a mounted FS, num_devices can't be zero unless it's a
2100                  * seed. In case of a seed device being replaced, the replace
2101                  * target added to the sprout FS, so there will be no more
2102                  * device left under the seed FS.
2103                  */
2104                 ASSERT(fs_devices->seeding);
2105
2106                 tmp_fs_devices = fs_info->fs_devices;
2107                 while (tmp_fs_devices) {
2108                         if (tmp_fs_devices->seed == fs_devices) {
2109                                 tmp_fs_devices->seed = fs_devices->seed;
2110                                 break;
2111                         }
2112                         tmp_fs_devices = tmp_fs_devices->seed;
2113                 }
2114                 fs_devices->seed = NULL;
2115                 __btrfs_close_devices(fs_devices);
2116                 free_fs_devices(fs_devices);
2117         }
2118 }
2119
2120 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
2121                                       struct btrfs_device *tgtdev)
2122 {
2123         mutex_lock(&uuid_mutex);
2124         WARN_ON(!tgtdev);
2125         mutex_lock(&fs_info->fs_devices->device_list_mutex);
2126
2127         btrfs_sysfs_rm_device_link(fs_info->fs_devices, tgtdev);
2128
2129         if (tgtdev->bdev)
2130                 fs_info->fs_devices->open_devices--;
2131
2132         fs_info->fs_devices->num_devices--;
2133
2134         btrfs_assign_next_active_device(fs_info, tgtdev, NULL);
2135
2136         list_del_rcu(&tgtdev->dev_list);
2137
2138         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2139         mutex_unlock(&uuid_mutex);
2140
2141         /*
2142          * The update_dev_time() with in btrfs_scratch_superblocks()
2143          * may lead to a call to btrfs_show_devname() which will try
2144          * to hold device_list_mutex. And here this device
2145          * is already out of device list, so we don't have to hold
2146          * the device_list_mutex lock.
2147          */
2148         btrfs_scratch_superblocks(tgtdev->bdev, tgtdev->name->str);
2149
2150         btrfs_close_bdev(tgtdev);
2151         call_rcu(&tgtdev->rcu, free_device_rcu);
2152 }
2153
2154 static int btrfs_find_device_by_path(struct btrfs_fs_info *fs_info,
2155                                      const char *device_path,
2156                                      struct btrfs_device **device)
2157 {
2158         int ret = 0;
2159         struct btrfs_super_block *disk_super;
2160         u64 devid;
2161         u8 *dev_uuid;
2162         struct block_device *bdev;
2163         struct buffer_head *bh;
2164
2165         *device = NULL;
2166         ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
2167                                     fs_info->bdev_holder, 0, &bdev, &bh);
2168         if (ret)
2169                 return ret;
2170         disk_super = (struct btrfs_super_block *)bh->b_data;
2171         devid = btrfs_stack_device_id(&disk_super->dev_item);
2172         dev_uuid = disk_super->dev_item.uuid;
2173         *device = btrfs_find_device(fs_info, devid, dev_uuid, disk_super->fsid);
2174         brelse(bh);
2175         if (!*device)
2176                 ret = -ENOENT;
2177         blkdev_put(bdev, FMODE_READ);
2178         return ret;
2179 }
2180
2181 int btrfs_find_device_missing_or_by_path(struct btrfs_fs_info *fs_info,
2182                                          const char *device_path,
2183                                          struct btrfs_device **device)
2184 {
2185         *device = NULL;
2186         if (strcmp(device_path, "missing") == 0) {
2187                 struct list_head *devices;
2188                 struct btrfs_device *tmp;
2189
2190                 devices = &fs_info->fs_devices->devices;
2191                 /*
2192                  * It is safe to read the devices since the volume_mutex
2193                  * is held by the caller.
2194                  */
2195                 list_for_each_entry(tmp, devices, dev_list) {
2196                         if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
2197                                         &tmp->dev_state) && !tmp->bdev) {
2198                                 *device = tmp;
2199                                 break;
2200                         }
2201                 }
2202
2203                 if (!*device)
2204                         return BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
2205
2206                 return 0;
2207         } else {
2208                 return btrfs_find_device_by_path(fs_info, device_path, device);
2209         }
2210 }
2211
2212 /*
2213  * Lookup a device given by device id, or the path if the id is 0.
2214  */
2215 int btrfs_find_device_by_devspec(struct btrfs_fs_info *fs_info, u64 devid,
2216                                  const char *devpath,
2217                                  struct btrfs_device **device)
2218 {
2219         int ret;
2220
2221         if (devid) {
2222                 ret = 0;
2223                 *device = btrfs_find_device(fs_info, devid, NULL, NULL);
2224                 if (!*device)
2225                         ret = -ENOENT;
2226         } else {
2227                 if (!devpath || !devpath[0])
2228                         return -EINVAL;
2229
2230                 ret = btrfs_find_device_missing_or_by_path(fs_info, devpath,
2231                                                            device);
2232         }
2233         return ret;
2234 }
2235
2236 /*
2237  * does all the dirty work required for changing file system's UUID.
2238  */
2239 static int btrfs_prepare_sprout(struct btrfs_fs_info *fs_info)
2240 {
2241         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2242         struct btrfs_fs_devices *old_devices;
2243         struct btrfs_fs_devices *seed_devices;
2244         struct btrfs_super_block *disk_super = fs_info->super_copy;
2245         struct btrfs_device *device;
2246         u64 super_flags;
2247
2248         lockdep_assert_held(&uuid_mutex);
2249         if (!fs_devices->seeding)
2250                 return -EINVAL;
2251
2252         seed_devices = alloc_fs_devices(NULL);
2253         if (IS_ERR(seed_devices))
2254                 return PTR_ERR(seed_devices);
2255
2256         old_devices = clone_fs_devices(fs_devices);
2257         if (IS_ERR(old_devices)) {
2258                 kfree(seed_devices);
2259                 return PTR_ERR(old_devices);
2260         }
2261
2262         list_add(&old_devices->list, &fs_uuids);
2263
2264         memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
2265         seed_devices->opened = 1;
2266         INIT_LIST_HEAD(&seed_devices->devices);
2267         INIT_LIST_HEAD(&seed_devices->alloc_list);
2268         mutex_init(&seed_devices->device_list_mutex);
2269
2270         mutex_lock(&fs_info->fs_devices->device_list_mutex);
2271         list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
2272                               synchronize_rcu);
2273         list_for_each_entry(device, &seed_devices->devices, dev_list)
2274                 device->fs_devices = seed_devices;
2275
2276         mutex_lock(&fs_info->chunk_mutex);
2277         list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
2278         mutex_unlock(&fs_info->chunk_mutex);
2279
2280         fs_devices->seeding = 0;
2281         fs_devices->num_devices = 0;
2282         fs_devices->open_devices = 0;
2283         fs_devices->missing_devices = 0;
2284         fs_devices->rotating = 0;
2285         fs_devices->seed = seed_devices;
2286
2287         generate_random_uuid(fs_devices->fsid);
2288         memcpy(fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2289         memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2290         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2291
2292         super_flags = btrfs_super_flags(disk_super) &
2293                       ~BTRFS_SUPER_FLAG_SEEDING;
2294         btrfs_set_super_flags(disk_super, super_flags);
2295
2296         return 0;
2297 }
2298
2299 /*
2300  * Store the expected generation for seed devices in device items.
2301  */
2302 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
2303                                struct btrfs_fs_info *fs_info)
2304 {
2305         struct btrfs_root *root = fs_info->chunk_root;
2306         struct btrfs_path *path;
2307         struct extent_buffer *leaf;
2308         struct btrfs_dev_item *dev_item;
2309         struct btrfs_device *device;
2310         struct btrfs_key key;
2311         u8 fs_uuid[BTRFS_FSID_SIZE];
2312         u8 dev_uuid[BTRFS_UUID_SIZE];
2313         u64 devid;
2314         int ret;
2315
2316         path = btrfs_alloc_path();
2317         if (!path)
2318                 return -ENOMEM;
2319
2320         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2321         key.offset = 0;
2322         key.type = BTRFS_DEV_ITEM_KEY;
2323
2324         while (1) {
2325                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2326                 if (ret < 0)
2327                         goto error;
2328
2329                 leaf = path->nodes[0];
2330 next_slot:
2331                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2332                         ret = btrfs_next_leaf(root, path);
2333                         if (ret > 0)
2334                                 break;
2335                         if (ret < 0)
2336                                 goto error;
2337                         leaf = path->nodes[0];
2338                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2339                         btrfs_release_path(path);
2340                         continue;
2341                 }
2342
2343                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2344                 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
2345                     key.type != BTRFS_DEV_ITEM_KEY)
2346                         break;
2347
2348                 dev_item = btrfs_item_ptr(leaf, path->slots[0],
2349                                           struct btrfs_dev_item);
2350                 devid = btrfs_device_id(leaf, dev_item);
2351                 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
2352                                    BTRFS_UUID_SIZE);
2353                 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
2354                                    BTRFS_FSID_SIZE);
2355                 device = btrfs_find_device(fs_info, devid, dev_uuid, fs_uuid);
2356                 BUG_ON(!device); /* Logic error */
2357
2358                 if (device->fs_devices->seeding) {
2359                         btrfs_set_device_generation(leaf, dev_item,
2360                                                     device->generation);
2361                         btrfs_mark_buffer_dirty(leaf);
2362                 }
2363
2364                 path->slots[0]++;
2365                 goto next_slot;
2366         }
2367         ret = 0;
2368 error:
2369         btrfs_free_path(path);
2370         return ret;
2371 }
2372
2373 int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path)
2374 {
2375         struct btrfs_root *root = fs_info->dev_root;
2376         struct request_queue *q;
2377         struct btrfs_trans_handle *trans;
2378         struct btrfs_device *device;
2379         struct block_device *bdev;
2380         struct list_head *devices;
2381         struct super_block *sb = fs_info->sb;
2382         struct rcu_string *name;
2383         u64 tmp;
2384         int seeding_dev = 0;
2385         int ret = 0;
2386         bool unlocked = false;
2387
2388         if (sb_rdonly(sb) && !fs_info->fs_devices->seeding)
2389                 return -EROFS;
2390
2391         bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2392                                   fs_info->bdev_holder);
2393         if (IS_ERR(bdev))
2394                 return PTR_ERR(bdev);
2395
2396         if (fs_info->fs_devices->seeding) {
2397                 seeding_dev = 1;
2398                 down_write(&sb->s_umount);
2399                 mutex_lock(&uuid_mutex);
2400         }
2401
2402         filemap_write_and_wait(bdev->bd_inode->i_mapping);
2403
2404         devices = &fs_info->fs_devices->devices;
2405
2406         mutex_lock(&fs_info->fs_devices->device_list_mutex);
2407         list_for_each_entry(device, devices, dev_list) {
2408                 if (device->bdev == bdev) {
2409                         ret = -EEXIST;
2410                         mutex_unlock(
2411                                 &fs_info->fs_devices->device_list_mutex);
2412                         goto error;
2413                 }
2414         }
2415         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2416
2417         device = btrfs_alloc_device(fs_info, NULL, NULL);
2418         if (IS_ERR(device)) {
2419                 /* we can safely leave the fs_devices entry around */
2420                 ret = PTR_ERR(device);
2421                 goto error;
2422         }
2423
2424         name = rcu_string_strdup(device_path, GFP_KERNEL);
2425         if (!name) {
2426                 ret = -ENOMEM;
2427                 goto error_free_device;
2428         }
2429         rcu_assign_pointer(device->name, name);
2430
2431         trans = btrfs_start_transaction(root, 0);
2432         if (IS_ERR(trans)) {
2433                 ret = PTR_ERR(trans);
2434                 goto error_free_device;
2435         }
2436
2437         q = bdev_get_queue(bdev);
2438         set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
2439         device->generation = trans->transid;
2440         device->io_width = fs_info->sectorsize;
2441         device->io_align = fs_info->sectorsize;
2442         device->sector_size = fs_info->sectorsize;
2443         device->total_bytes = round_down(i_size_read(bdev->bd_inode),
2444                                          fs_info->sectorsize);
2445         device->disk_total_bytes = device->total_bytes;
2446         device->commit_total_bytes = device->total_bytes;
2447         device->fs_info = fs_info;
2448         device->bdev = bdev;
2449         set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
2450         clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
2451         device->mode = FMODE_EXCL;
2452         device->dev_stats_valid = 1;
2453         set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE);
2454
2455         if (seeding_dev) {
2456                 sb->s_flags &= ~SB_RDONLY;
2457                 ret = btrfs_prepare_sprout(fs_info);
2458                 if (ret) {
2459                         btrfs_abort_transaction(trans, ret);
2460                         goto error_trans;
2461                 }
2462         }
2463
2464         device->fs_devices = fs_info->fs_devices;
2465
2466         mutex_lock(&fs_info->fs_devices->device_list_mutex);
2467         mutex_lock(&fs_info->chunk_mutex);
2468         list_add_rcu(&device->dev_list, &fs_info->fs_devices->devices);
2469         list_add(&device->dev_alloc_list,
2470                  &fs_info->fs_devices->alloc_list);
2471         fs_info->fs_devices->num_devices++;
2472         fs_info->fs_devices->open_devices++;
2473         fs_info->fs_devices->rw_devices++;
2474         fs_info->fs_devices->total_devices++;
2475         fs_info->fs_devices->total_rw_bytes += device->total_bytes;
2476
2477         atomic64_add(device->total_bytes, &fs_info->free_chunk_space);
2478
2479         if (!blk_queue_nonrot(q))
2480                 fs_info->fs_devices->rotating = 1;
2481
2482         tmp = btrfs_super_total_bytes(fs_info->super_copy);
2483         btrfs_set_super_total_bytes(fs_info->super_copy,
2484                 round_down(tmp + device->total_bytes, fs_info->sectorsize));
2485
2486         tmp = btrfs_super_num_devices(fs_info->super_copy);
2487         btrfs_set_super_num_devices(fs_info->super_copy, tmp + 1);
2488
2489         /* add sysfs device entry */
2490         btrfs_sysfs_add_device_link(fs_info->fs_devices, device);
2491
2492         /*
2493          * we've got more storage, clear any full flags on the space
2494          * infos
2495          */
2496         btrfs_clear_space_info_full(fs_info);
2497
2498         mutex_unlock(&fs_info->chunk_mutex);
2499         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2500
2501         if (seeding_dev) {
2502                 mutex_lock(&fs_info->chunk_mutex);
2503                 ret = init_first_rw_device(trans, fs_info);
2504                 mutex_unlock(&fs_info->chunk_mutex);
2505                 if (ret) {
2506                         btrfs_abort_transaction(trans, ret);
2507                         goto error_sysfs;
2508                 }
2509         }
2510
2511         ret = btrfs_add_dev_item(trans, fs_info, device);
2512         if (ret) {
2513                 btrfs_abort_transaction(trans, ret);
2514                 goto error_sysfs;
2515         }
2516
2517         if (seeding_dev) {
2518                 char fsid_buf[BTRFS_UUID_UNPARSED_SIZE];
2519
2520                 ret = btrfs_finish_sprout(trans, fs_info);
2521                 if (ret) {
2522                         btrfs_abort_transaction(trans, ret);
2523                         goto error_sysfs;
2524                 }
2525
2526                 /* Sprouting would change fsid of the mounted root,
2527                  * so rename the fsid on the sysfs
2528                  */
2529                 snprintf(fsid_buf, BTRFS_UUID_UNPARSED_SIZE, "%pU",
2530                                                 fs_info->fsid);
2531                 if (kobject_rename(&fs_info->fs_devices->fsid_kobj, fsid_buf))
2532                         btrfs_warn(fs_info,
2533                                    "sysfs: failed to create fsid for sprout");
2534         }
2535
2536         ret = btrfs_commit_transaction(trans);
2537
2538         if (seeding_dev) {
2539                 mutex_unlock(&uuid_mutex);
2540                 up_write(&sb->s_umount);
2541                 unlocked = true;
2542
2543                 if (ret) /* transaction commit */
2544                         return ret;
2545
2546                 ret = btrfs_relocate_sys_chunks(fs_info);
2547                 if (ret < 0)
2548                         btrfs_handle_fs_error(fs_info, ret,
2549                                     "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command.");
2550                 trans = btrfs_attach_transaction(root);
2551                 if (IS_ERR(trans)) {
2552                         if (PTR_ERR(trans) == -ENOENT)
2553                                 return 0;
2554                         ret = PTR_ERR(trans);
2555                         trans = NULL;
2556                         goto error_sysfs;
2557                 }
2558                 ret = btrfs_commit_transaction(trans);
2559         }
2560
2561         /* Update ctime/mtime for libblkid */
2562         update_dev_time(device_path);
2563         return ret;
2564
2565 error_sysfs:
2566         btrfs_sysfs_rm_device_link(fs_info->fs_devices, device);
2567 error_trans:
2568         if (seeding_dev)
2569                 sb->s_flags |= SB_RDONLY;
2570         if (trans)
2571                 btrfs_end_transaction(trans);
2572 error_free_device:
2573         free_device(device);
2574 error:
2575         blkdev_put(bdev, FMODE_EXCL);
2576         if (seeding_dev && !unlocked) {
2577                 mutex_unlock(&uuid_mutex);
2578                 up_write(&sb->s_umount);
2579         }
2580         return ret;
2581 }
2582
2583 int btrfs_init_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
2584                                   const char *device_path,
2585                                   struct btrfs_device *srcdev,
2586                                   struct btrfs_device **device_out)
2587 {
2588         struct btrfs_device *device;
2589         struct block_device *bdev;
2590         struct list_head *devices;
2591         struct rcu_string *name;
2592         u64 devid = BTRFS_DEV_REPLACE_DEVID;
2593         int ret = 0;
2594
2595         *device_out = NULL;
2596         if (fs_info->fs_devices->seeding) {
2597                 btrfs_err(fs_info, "the filesystem is a seed filesystem!");
2598                 return -EINVAL;
2599         }
2600
2601         bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2602                                   fs_info->bdev_holder);
2603         if (IS_ERR(bdev)) {
2604                 btrfs_err(fs_info, "target device %s is invalid!", device_path);
2605                 return PTR_ERR(bdev);
2606         }
2607
2608         filemap_write_and_wait(bdev->bd_inode->i_mapping);
2609
2610         devices = &fs_info->fs_devices->devices;
2611         list_for_each_entry(device, devices, dev_list) {
2612                 if (device->bdev == bdev) {
2613                         btrfs_err(fs_info,
2614                                   "target device is in the filesystem!");
2615                         ret = -EEXIST;
2616                         goto error;
2617                 }
2618         }
2619
2620
2621         if (i_size_read(bdev->bd_inode) <
2622             btrfs_device_get_total_bytes(srcdev)) {
2623                 btrfs_err(fs_info,
2624                           "target device is smaller than source device!");
2625                 ret = -EINVAL;
2626                 goto error;
2627         }
2628
2629
2630         device = btrfs_alloc_device(NULL, &devid, NULL);
2631         if (IS_ERR(device)) {
2632                 ret = PTR_ERR(device);
2633                 goto error;
2634         }
2635
2636         name = rcu_string_strdup(device_path, GFP_KERNEL);
2637         if (!name) {
2638                 free_device(device);
2639                 ret = -ENOMEM;
2640                 goto error;
2641         }
2642         rcu_assign_pointer(device->name, name);
2643
2644         mutex_lock(&fs_info->fs_devices->device_list_mutex);
2645         set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
2646         device->generation = 0;
2647         device->io_width = fs_info->sectorsize;
2648         device->io_align = fs_info->sectorsize;
2649         device->sector_size = fs_info->sectorsize;
2650         device->total_bytes = btrfs_device_get_total_bytes(srcdev);
2651         device->disk_total_bytes = btrfs_device_get_disk_total_bytes(srcdev);
2652         device->bytes_used = btrfs_device_get_bytes_used(srcdev);
2653         device->commit_total_bytes = srcdev->commit_total_bytes;
2654         device->commit_bytes_used = device->bytes_used;
2655         device->fs_info = fs_info;
2656         device->bdev = bdev;
2657         set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
2658         set_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
2659         device->mode = FMODE_EXCL;
2660         device->dev_stats_valid = 1;
2661         set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE);
2662         device->fs_devices = fs_info->fs_devices;
2663         list_add(&device->dev_list, &fs_info->fs_devices->devices);
2664         fs_info->fs_devices->num_devices++;
2665         fs_info->fs_devices->open_devices++;
2666         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2667
2668         *device_out = device;
2669         return ret;
2670
2671 error:
2672         blkdev_put(bdev, FMODE_EXCL);
2673         return ret;
2674 }
2675
2676 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2677                                         struct btrfs_device *device)
2678 {
2679         int ret;
2680         struct btrfs_path *path;
2681         struct btrfs_root *root = device->fs_info->chunk_root;
2682         struct btrfs_dev_item *dev_item;
2683         struct extent_buffer *leaf;
2684         struct btrfs_key key;
2685
2686         path = btrfs_alloc_path();
2687         if (!path)
2688                 return -ENOMEM;
2689
2690         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2691         key.type = BTRFS_DEV_ITEM_KEY;
2692         key.offset = device->devid;
2693
2694         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2695         if (ret < 0)
2696                 goto out;
2697
2698         if (ret > 0) {
2699                 ret = -ENOENT;
2700                 goto out;
2701         }
2702
2703         leaf = path->nodes[0];
2704         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2705
2706         btrfs_set_device_id(leaf, dev_item, device->devid);
2707         btrfs_set_device_type(leaf, dev_item, device->type);
2708         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2709         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2710         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2711         btrfs_set_device_total_bytes(leaf, dev_item,
2712                                      btrfs_device_get_disk_total_bytes(device));
2713         btrfs_set_device_bytes_used(leaf, dev_item,
2714                                     btrfs_device_get_bytes_used(device));
2715         btrfs_mark_buffer_dirty(leaf);
2716
2717 out:
2718         btrfs_free_path(path);
2719         return ret;
2720 }
2721
2722 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2723                       struct btrfs_device *device, u64 new_size)
2724 {
2725         struct btrfs_fs_info *fs_info = device->fs_info;
2726         struct btrfs_super_block *super_copy = fs_info->super_copy;
2727         struct btrfs_fs_devices *fs_devices;
2728         u64 old_total;
2729         u64 diff;
2730
2731         if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
2732                 return -EACCES;
2733
2734         new_size = round_down(new_size, fs_info->sectorsize);
2735
2736         mutex_lock(&fs_info->chunk_mutex);
2737         old_total = btrfs_super_total_bytes(super_copy);
2738         diff = round_down(new_size - device->total_bytes, fs_info->sectorsize);
2739
2740         if (new_size <= device->total_bytes ||
2741             test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
2742                 mutex_unlock(&fs_info->chunk_mutex);
2743                 return -EINVAL;
2744         }
2745
2746         fs_devices = fs_info->fs_devices;
2747
2748         btrfs_set_super_total_bytes(super_copy,
2749                         round_down(old_total + diff, fs_info->sectorsize));
2750         device->fs_devices->total_rw_bytes += diff;
2751
2752         btrfs_device_set_total_bytes(device, new_size);
2753         btrfs_device_set_disk_total_bytes(device, new_size);
2754         btrfs_clear_space_info_full(device->fs_info);
2755         if (list_empty(&device->resized_list))
2756                 list_add_tail(&device->resized_list,
2757                               &fs_devices->resized_devices);
2758         mutex_unlock(&fs_info->chunk_mutex);
2759
2760         return btrfs_update_device(trans, device);
2761 }
2762
2763 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
2764                             struct btrfs_fs_info *fs_info, u64 chunk_offset)
2765 {
2766         struct btrfs_root *root = fs_info->chunk_root;
2767         int ret;
2768         struct btrfs_path *path;
2769         struct btrfs_key key;
2770
2771         path = btrfs_alloc_path();
2772         if (!path)
2773                 return -ENOMEM;
2774
2775         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2776         key.offset = chunk_offset;
2777         key.type = BTRFS_CHUNK_ITEM_KEY;
2778
2779         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2780         if (ret < 0)
2781                 goto out;
2782         else if (ret > 0) { /* Logic error or corruption */
2783                 btrfs_handle_fs_error(fs_info, -ENOENT,
2784                                       "Failed lookup while freeing chunk.");
2785                 ret = -ENOENT;
2786                 goto out;
2787         }
2788
2789         ret = btrfs_del_item(trans, root, path);
2790         if (ret < 0)
2791                 btrfs_handle_fs_error(fs_info, ret,
2792                                       "Failed to delete chunk item.");
2793 out:
2794         btrfs_free_path(path);
2795         return ret;
2796 }
2797
2798 static int btrfs_del_sys_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
2799 {
2800         struct btrfs_super_block *super_copy = fs_info->super_copy;
2801         struct btrfs_disk_key *disk_key;
2802         struct btrfs_chunk *chunk;
2803         u8 *ptr;
2804         int ret = 0;
2805         u32 num_stripes;
2806         u32 array_size;
2807         u32 len = 0;
2808         u32 cur;
2809         struct btrfs_key key;
2810
2811         mutex_lock(&fs_info->chunk_mutex);
2812         array_size = btrfs_super_sys_array_size(super_copy);
2813
2814         ptr = super_copy->sys_chunk_array;
2815         cur = 0;
2816
2817         while (cur < array_size) {
2818                 disk_key = (struct btrfs_disk_key *)ptr;
2819                 btrfs_disk_key_to_cpu(&key, disk_key);
2820
2821                 len = sizeof(*disk_key);
2822
2823                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2824                         chunk = (struct btrfs_chunk *)(ptr + len);
2825                         num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2826                         len += btrfs_chunk_item_size(num_stripes);
2827                 } else {
2828                         ret = -EIO;
2829                         break;
2830                 }
2831                 if (key.objectid == BTRFS_FIRST_CHUNK_TREE_OBJECTID &&
2832                     key.offset == chunk_offset) {
2833                         memmove(ptr, ptr + len, array_size - (cur + len));
2834                         array_size -= len;
2835                         btrfs_set_super_sys_array_size(super_copy, array_size);
2836                 } else {
2837                         ptr += len;
2838                         cur += len;
2839                 }
2840         }
2841         mutex_unlock(&fs_info->chunk_mutex);
2842         return ret;
2843 }
2844
2845 static struct extent_map *get_chunk_map(struct btrfs_fs_info *fs_info,
2846                                         u64 logical, u64 length)
2847 {
2848         struct extent_map_tree *em_tree;
2849         struct extent_map *em;
2850
2851         em_tree = &fs_info->mapping_tree.map_tree;
2852         read_lock(&em_tree->lock);
2853         em = lookup_extent_mapping(em_tree, logical, length);
2854         read_unlock(&em_tree->lock);
2855
2856         if (!em) {
2857                 btrfs_crit(fs_info, "unable to find logical %llu length %llu",
2858                            logical, length);
2859                 return ERR_PTR(-EINVAL);
2860         }
2861
2862         if (em->start > logical || em->start + em->len < logical) {
2863                 btrfs_crit(fs_info,
2864                            "found a bad mapping, wanted %llu-%llu, found %llu-%llu",
2865                            logical, length, em->start, em->start + em->len);
2866                 free_extent_map(em);
2867                 return ERR_PTR(-EINVAL);
2868         }
2869
2870         /* callers are responsible for dropping em's ref. */
2871         return em;
2872 }
2873
2874 int btrfs_remove_chunk(struct btrfs_trans_handle *trans,
2875                        struct btrfs_fs_info *fs_info, u64 chunk_offset)
2876 {
2877         struct extent_map *em;
2878         struct map_lookup *map;
2879         u64 dev_extent_len = 0;
2880         int i, ret = 0;
2881         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2882
2883         em = get_chunk_map(fs_info, chunk_offset, 1);
2884         if (IS_ERR(em)) {
2885                 /*
2886                  * This is a logic error, but we don't want to just rely on the
2887                  * user having built with ASSERT enabled, so if ASSERT doesn't
2888                  * do anything we still error out.
2889                  */
2890                 ASSERT(0);
2891                 return PTR_ERR(em);
2892         }
2893         map = em->map_lookup;
2894         mutex_lock(&fs_info->chunk_mutex);
2895         check_system_chunk(trans, fs_info, map->type);
2896         mutex_unlock(&fs_info->chunk_mutex);
2897
2898         /*
2899          * Take the device list mutex to prevent races with the final phase of
2900          * a device replace operation that replaces the device object associated
2901          * with map stripes (dev-replace.c:btrfs_dev_replace_finishing()).
2902          */
2903         mutex_lock(&fs_devices->device_list_mutex);
2904         for (i = 0; i < map->num_stripes; i++) {
2905                 struct btrfs_device *device = map->stripes[i].dev;
2906                 ret = btrfs_free_dev_extent(trans, device,
2907                                             map->stripes[i].physical,
2908                                             &dev_extent_len);
2909                 if (ret) {
2910                         mutex_unlock(&fs_devices->device_list_mutex);
2911                         btrfs_abort_transaction(trans, ret);
2912                         goto out;
2913                 }
2914
2915                 if (device->bytes_used > 0) {
2916                         mutex_lock(&fs_info->chunk_mutex);
2917                         btrfs_device_set_bytes_used(device,
2918                                         device->bytes_used - dev_extent_len);
2919                         atomic64_add(dev_extent_len, &fs_info->free_chunk_space);
2920                         btrfs_clear_space_info_full(fs_info);
2921                         mutex_unlock(&fs_info->chunk_mutex);
2922                 }
2923
2924                 if (map->stripes[i].dev) {
2925                         ret = btrfs_update_device(trans, map->stripes[i].dev);
2926                         if (ret) {
2927                                 mutex_unlock(&fs_devices->device_list_mutex);
2928                                 btrfs_abort_transaction(trans, ret);
2929                                 goto out;
2930                         }
2931                 }
2932         }
2933         mutex_unlock(&fs_devices->device_list_mutex);
2934
2935         ret = btrfs_free_chunk(trans, fs_info, chunk_offset);
2936         if (ret) {
2937                 btrfs_abort_transaction(trans, ret);
2938                 goto out;
2939         }
2940
2941         trace_btrfs_chunk_free(fs_info, map, chunk_offset, em->len);
2942
2943         if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2944                 ret = btrfs_del_sys_chunk(fs_info, chunk_offset);
2945                 if (ret) {
2946                         btrfs_abort_transaction(trans, ret);
2947                         goto out;
2948                 }
2949         }
2950
2951         ret = btrfs_remove_block_group(trans, fs_info, chunk_offset, em);
2952         if (ret) {
2953                 btrfs_abort_transaction(trans, ret);
2954                 goto out;
2955         }
2956
2957 out:
2958         /* once for us */
2959         free_extent_map(em);
2960         return ret;
2961 }
2962
2963 static int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
2964 {
2965         struct btrfs_root *root = fs_info->chunk_root;
2966         struct btrfs_trans_handle *trans;
2967         int ret;
2968
2969         /*
2970          * Prevent races with automatic removal of unused block groups.
2971          * After we relocate and before we remove the chunk with offset
2972          * chunk_offset, automatic removal of the block group can kick in,
2973          * resulting in a failure when calling btrfs_remove_chunk() below.
2974          *
2975          * Make sure to acquire this mutex before doing a tree search (dev
2976          * or chunk trees) to find chunks. Otherwise the cleaner kthread might
2977          * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after
2978          * we release the path used to search the chunk/dev tree and before
2979          * the current task acquires this mutex and calls us.
2980          */
2981         lockdep_assert_held(&fs_info->delete_unused_bgs_mutex);
2982
2983         ret = btrfs_can_relocate(fs_info, chunk_offset);
2984         if (ret)
2985                 return -ENOSPC;
2986
2987         /* step one, relocate all the extents inside this chunk */
2988         btrfs_scrub_pause(fs_info);
2989         ret = btrfs_relocate_block_group(fs_info, chunk_offset);
2990         btrfs_scrub_continue(fs_info);
2991         if (ret)
2992                 return ret;
2993
2994         /*
2995          * We add the kobjects here (and after forcing data chunk creation)
2996          * since relocation is the only place we'll create chunks of a new
2997          * type at runtime.  The only place where we'll remove the last
2998          * chunk of a type is the call immediately below this one.  Even
2999          * so, we're protected against races with the cleaner thread since
3000          * we're covered by the delete_unused_bgs_mutex.
3001          */
3002         btrfs_add_raid_kobjects(fs_info);
3003
3004         trans = btrfs_start_trans_remove_block_group(root->fs_info,
3005                                                      chunk_offset);
3006         if (IS_ERR(trans)) {
3007                 ret = PTR_ERR(trans);
3008                 btrfs_handle_fs_error(root->fs_info, ret, NULL);
3009                 return ret;
3010         }
3011
3012         /*
3013          * step two, delete the device extents and the
3014          * chunk tree entries
3015          */
3016         ret = btrfs_remove_chunk(trans, fs_info, chunk_offset);
3017         btrfs_end_transaction(trans);
3018         return ret;
3019 }
3020
3021 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info)
3022 {
3023         struct btrfs_root *chunk_root = fs_info->chunk_root;
3024         struct btrfs_path *path;
3025         struct extent_buffer *leaf;
3026         struct btrfs_chunk *chunk;
3027         struct btrfs_key key;
3028         struct btrfs_key found_key;
3029         u64 chunk_type;
3030         bool retried = false;
3031         int failed = 0;
3032         int ret;
3033
3034         path = btrfs_alloc_path();
3035         if (!path)
3036                 return -ENOMEM;
3037
3038 again:
3039         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3040         key.offset = (u64)-1;
3041         key.type = BTRFS_CHUNK_ITEM_KEY;
3042
3043         while (1) {
3044                 mutex_lock(&fs_info->delete_unused_bgs_mutex);
3045                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3046                 if (ret < 0) {
3047                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3048                         goto error;
3049                 }
3050                 BUG_ON(ret == 0); /* Corruption */
3051
3052                 ret = btrfs_previous_item(chunk_root, path, key.objectid,
3053                                           key.type);
3054                 if (ret)
3055                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3056                 if (ret < 0)
3057                         goto error;
3058                 if (ret > 0)
3059                         break;
3060
3061                 leaf = path->nodes[0];
3062                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3063
3064                 chunk = btrfs_item_ptr(leaf, path->slots[0],
3065                                        struct btrfs_chunk);
3066                 chunk_type = btrfs_chunk_type(leaf, chunk);
3067                 btrfs_release_path(path);
3068
3069                 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
3070                         ret = btrfs_relocate_chunk(fs_info, found_key.offset);
3071                         if (ret == -ENOSPC)
3072                                 failed++;
3073                         else
3074                                 BUG_ON(ret);
3075                 }
3076                 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3077
3078                 if (found_key.offset == 0)
3079                         break;
3080                 key.offset = found_key.offset - 1;
3081         }
3082         ret = 0;
3083         if (failed && !retried) {
3084                 failed = 0;
3085                 retried = true;
3086                 goto again;
3087         } else if (WARN_ON(failed && retried)) {
3088                 ret = -ENOSPC;
3089         }
3090 error:
3091         btrfs_free_path(path);
3092         return ret;
3093 }
3094
3095 /*
3096  * return 1 : allocate a data chunk successfully,
3097  * return <0: errors during allocating a data chunk,
3098  * return 0 : no need to allocate a data chunk.
3099  */
3100 static int btrfs_may_alloc_data_chunk(struct btrfs_fs_info *fs_info,
3101                                       u64 chunk_offset)
3102 {
3103         struct btrfs_block_group_cache *cache;
3104         u64 bytes_used;
3105         u64 chunk_type;
3106
3107         cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3108         ASSERT(cache);
3109         chunk_type = cache->flags;
3110         btrfs_put_block_group(cache);
3111
3112         if (chunk_type & BTRFS_BLOCK_GROUP_DATA) {
3113                 spin_lock(&fs_info->data_sinfo->lock);
3114                 bytes_used = fs_info->data_sinfo->bytes_used;
3115                 spin_unlock(&fs_info->data_sinfo->lock);
3116
3117                 if (!bytes_used) {
3118                         struct btrfs_trans_handle *trans;
3119                         int ret;
3120
3121                         trans = btrfs_join_transaction(fs_info->tree_root);
3122                         if (IS_ERR(trans))
3123                                 return PTR_ERR(trans);
3124
3125                         ret = btrfs_force_chunk_alloc(trans, fs_info,
3126                                                       BTRFS_BLOCK_GROUP_DATA);
3127                         btrfs_end_transaction(trans);
3128                         if (ret < 0)
3129                                 return ret;
3130
3131                         btrfs_add_raid_kobjects(fs_info);
3132
3133                         return 1;
3134                 }
3135         }
3136         return 0;
3137 }
3138
3139 static int insert_balance_item(struct btrfs_fs_info *fs_info,
3140                                struct btrfs_balance_control *bctl)
3141 {
3142         struct btrfs_root *root = fs_info->tree_root;
3143         struct btrfs_trans_handle *trans;
3144         struct btrfs_balance_item *item;
3145         struct btrfs_disk_balance_args disk_bargs;
3146         struct btrfs_path *path;
3147         struct extent_buffer *leaf;
3148         struct btrfs_key key;
3149         int ret, err;
3150
3151         path = btrfs_alloc_path();
3152         if (!path)
3153                 return -ENOMEM;
3154
3155         trans = btrfs_start_transaction(root, 0);
3156         if (IS_ERR(trans)) {
3157                 btrfs_free_path(path);
3158                 return PTR_ERR(trans);
3159         }
3160
3161         key.objectid = BTRFS_BALANCE_OBJECTID;
3162         key.type = BTRFS_TEMPORARY_ITEM_KEY;
3163         key.offset = 0;
3164
3165         ret = btrfs_insert_empty_item(trans, root, path, &key,
3166                                       sizeof(*item));
3167         if (ret)
3168                 goto out;
3169
3170         leaf = path->nodes[0];
3171         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3172
3173         memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item));
3174
3175         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
3176         btrfs_set_balance_data(leaf, item, &disk_bargs);
3177         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
3178         btrfs_set_balance_meta(leaf, item, &disk_bargs);
3179         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
3180         btrfs_set_balance_sys(leaf, item, &disk_bargs);
3181
3182         btrfs_set_balance_flags(leaf, item, bctl->flags);
3183
3184         btrfs_mark_buffer_dirty(leaf);
3185 out:
3186         btrfs_free_path(path);
3187         err = btrfs_commit_transaction(trans);
3188         if (err && !ret)
3189                 ret = err;
3190         return ret;
3191 }
3192
3193 static int del_balance_item(struct btrfs_fs_info *fs_info)
3194 {
3195         struct btrfs_root *root = fs_info->tree_root;
3196         struct btrfs_trans_handle *trans;
3197         struct btrfs_path *path;
3198         struct btrfs_key key;
3199         int ret, err;
3200
3201         path = btrfs_alloc_path();
3202         if (!path)
3203                 return -ENOMEM;
3204
3205         trans = btrfs_start_transaction(root, 0);
3206         if (IS_ERR(trans)) {
3207                 btrfs_free_path(path);
3208                 return PTR_ERR(trans);
3209         }
3210
3211         key.objectid = BTRFS_BALANCE_OBJECTID;
3212         key.type = BTRFS_TEMPORARY_ITEM_KEY;
3213         key.offset = 0;
3214
3215         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3216         if (ret < 0)
3217                 goto out;
3218         if (ret > 0) {
3219                 ret = -ENOENT;
3220                 goto out;
3221         }
3222
3223         ret = btrfs_del_item(trans, root, path);
3224 out:
3225         btrfs_free_path(path);
3226         err = btrfs_commit_transaction(trans);
3227         if (err && !ret)
3228                 ret = err;
3229         return ret;
3230 }
3231
3232 /*
3233  * This is a heuristic used to reduce the number of chunks balanced on
3234  * resume after balance was interrupted.
3235  */
3236 static void update_balance_args(struct btrfs_balance_control *bctl)
3237 {
3238         /*
3239          * Turn on soft mode for chunk types that were being converted.
3240          */
3241         if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
3242                 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
3243         if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
3244                 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
3245         if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
3246                 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
3247
3248         /*
3249          * Turn on usage filter if is not already used.  The idea is
3250          * that chunks that we have already balanced should be
3251          * reasonably full.  Don't do it for chunks that are being
3252          * converted - that will keep us from relocating unconverted
3253          * (albeit full) chunks.
3254          */
3255         if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3256             !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3257             !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3258                 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
3259                 bctl->data.usage = 90;
3260         }
3261         if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3262             !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3263             !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3264                 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
3265                 bctl->sys.usage = 90;
3266         }
3267         if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3268             !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3269             !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3270                 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
3271                 bctl->meta.usage = 90;
3272         }
3273 }
3274
3275 /*
3276  * Should be called with both balance and volume mutexes held to
3277  * serialize other volume operations (add_dev/rm_dev/resize) with
3278  * restriper.  Same goes for unset_balance_control.
3279  */
3280 static void set_balance_control(struct btrfs_balance_control *bctl)
3281 {
3282         struct btrfs_fs_info *fs_info = bctl->fs_info;
3283
3284         BUG_ON(fs_info->balance_ctl);
3285
3286         spin_lock(&fs_info->balance_lock);
3287         fs_info->balance_ctl = bctl;
3288         spin_unlock(&fs_info->balance_lock);
3289 }
3290
3291 static void unset_balance_control(struct btrfs_fs_info *fs_info)
3292 {
3293         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3294
3295         BUG_ON(!fs_info->balance_ctl);
3296
3297         spin_lock(&fs_info->balance_lock);
3298         fs_info->balance_ctl = NULL;
3299         spin_unlock(&fs_info->balance_lock);
3300
3301         kfree(bctl);
3302 }
3303
3304 /*
3305  * Balance filters.  Return 1 if chunk should be filtered out
3306  * (should not be balanced).
3307  */
3308 static int chunk_profiles_filter(u64 chunk_type,
3309                                  struct btrfs_balance_args *bargs)
3310 {
3311         chunk_type = chunk_to_extended(chunk_type) &
3312                                 BTRFS_EXTENDED_PROFILE_MASK;
3313
3314         if (bargs->profiles & chunk_type)
3315                 return 0;
3316
3317         return 1;
3318 }
3319
3320 static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
3321                               struct btrfs_balance_args *bargs)
3322 {
3323         struct btrfs_block_group_cache *cache;
3324         u64 chunk_used;
3325         u64 user_thresh_min;
3326         u64 user_thresh_max;
3327         int ret = 1;
3328
3329         cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3330         chunk_used = btrfs_block_group_used(&cache->item);
3331
3332         if (bargs->usage_min == 0)
3333                 user_thresh_min = 0;
3334         else
3335                 user_thresh_min = div_factor_fine(cache->key.offset,
3336                                         bargs->usage_min);
3337
3338         if (bargs->usage_max == 0)
3339                 user_thresh_max = 1;
3340         else if (bargs->usage_max > 100)
3341                 user_thresh_max = cache->key.offset;
3342         else
3343                 user_thresh_max = div_factor_fine(cache->key.offset,
3344                                         bargs->usage_max);
3345
3346         if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max)
3347                 ret = 0;
3348
3349         btrfs_put_block_group(cache);
3350         return ret;
3351 }
3352
3353 static int chunk_usage_filter(struct btrfs_fs_info *fs_info,
3354                 u64 chunk_offset, struct btrfs_balance_args *bargs)
3355 {
3356         struct btrfs_block_group_cache *cache;
3357         u64 chunk_used, user_thresh;
3358         int ret = 1;
3359
3360         cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3361         chunk_used = btrfs_block_group_used(&cache->item);
3362
3363         if (bargs->usage_min == 0)
3364                 user_thresh = 1;
3365         else if (bargs->usage > 100)
3366                 user_thresh = cache->key.offset;
3367         else
3368                 user_thresh = div_factor_fine(cache->key.offset,
3369                                               bargs->usage);
3370
3371         if (chunk_used < user_thresh)
3372                 ret = 0;
3373
3374         btrfs_put_block_group(cache);
3375         return ret;
3376 }
3377
3378 static int chunk_devid_filter(struct extent_buffer *leaf,
3379                               struct btrfs_chunk *chunk,
3380                               struct btrfs_balance_args *bargs)
3381 {
3382         struct btrfs_stripe *stripe;
3383         int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3384         int i;
3385
3386         for (i = 0; i < num_stripes; i++) {
3387                 stripe = btrfs_stripe_nr(chunk, i);
3388                 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
3389                         return 0;
3390         }
3391
3392         return 1;
3393 }
3394
3395 /* [pstart, pend) */
3396 static int chunk_drange_filter(struct extent_buffer *leaf,
3397                                struct btrfs_chunk *chunk,
3398                                struct btrfs_balance_args *bargs)
3399 {
3400         struct btrfs_stripe *stripe;
3401         int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3402         u64 stripe_offset;
3403         u64 stripe_length;
3404         int factor;
3405         int i;
3406
3407         if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
3408                 return 0;
3409
3410         if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP |
3411              BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) {
3412                 factor = num_stripes / 2;
3413         } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) {
3414                 factor = num_stripes - 1;
3415         } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) {
3416                 factor = num_stripes - 2;
3417         } else {
3418                 factor = num_stripes;
3419         }
3420
3421         for (i = 0; i < num_stripes; i++) {
3422                 stripe = btrfs_stripe_nr(chunk, i);
3423                 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
3424                         continue;
3425
3426                 stripe_offset = btrfs_stripe_offset(leaf, stripe);
3427                 stripe_length = btrfs_chunk_length(leaf, chunk);
3428                 stripe_length = div_u64(stripe_length, factor);
3429
3430                 if (stripe_offset < bargs->pend &&
3431                     stripe_offset + stripe_length > bargs->pstart)
3432                         return 0;
3433         }
3434
3435         return 1;
3436 }
3437
3438 /* [vstart, vend) */
3439 static int chunk_vrange_filter(struct extent_buffer *leaf,
3440                                struct btrfs_chunk *chunk,
3441                                u64 chunk_offset,
3442                                struct btrfs_balance_args *bargs)
3443 {
3444         if (chunk_offset < bargs->vend &&
3445             chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
3446                 /* at least part of the chunk is inside this vrange */
3447                 return 0;
3448
3449         return 1;
3450 }
3451
3452 static int chunk_stripes_range_filter(struct extent_buffer *leaf,
3453                                struct btrfs_chunk *chunk,
3454                                struct btrfs_balance_args *bargs)
3455 {
3456         int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3457
3458         if (bargs->stripes_min <= num_stripes
3459                         && num_stripes <= bargs->stripes_max)
3460                 return 0;
3461
3462         return 1;
3463 }
3464
3465 static int chunk_soft_convert_filter(u64 chunk_type,
3466                                      struct btrfs_balance_args *bargs)
3467 {
3468         if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
3469                 return 0;
3470
3471         chunk_type = chunk_to_extended(chunk_type) &
3472                                 BTRFS_EXTENDED_PROFILE_MASK;
3473
3474         if (bargs->target == chunk_type)
3475                 return 1;
3476
3477         return 0;
3478 }
3479
3480 static int should_balance_chunk(struct btrfs_fs_info *fs_info,
3481                                 struct extent_buffer *leaf,
3482                                 struct btrfs_chunk *chunk, u64 chunk_offset)
3483 {
3484         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3485         struct btrfs_balance_args *bargs = NULL;
3486         u64 chunk_type = btrfs_chunk_type(leaf, chunk);
3487
3488         /* type filter */
3489         if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
3490               (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
3491                 return 0;
3492         }
3493
3494         if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3495                 bargs = &bctl->data;
3496         else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3497                 bargs = &bctl->sys;
3498         else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3499                 bargs = &bctl->meta;
3500
3501         /* profiles filter */
3502         if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
3503             chunk_profiles_filter(chunk_type, bargs)) {
3504                 return 0;
3505         }
3506
3507         /* usage filter */
3508         if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
3509             chunk_usage_filter(fs_info, chunk_offset, bargs)) {
3510                 return 0;
3511         } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3512             chunk_usage_range_filter(fs_info, chunk_offset, bargs)) {
3513                 return 0;
3514         }
3515
3516         /* devid filter */
3517         if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
3518             chunk_devid_filter(leaf, chunk, bargs)) {
3519                 return 0;
3520         }
3521
3522         /* drange filter, makes sense only with devid filter */
3523         if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
3524             chunk_drange_filter(leaf, chunk, bargs)) {
3525                 return 0;
3526         }
3527
3528         /* vrange filter */
3529         if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
3530             chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
3531                 return 0;
3532         }
3533
3534         /* stripes filter */
3535         if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) &&
3536             chunk_stripes_range_filter(leaf, chunk, bargs)) {
3537                 return 0;
3538         }
3539
3540         /* soft profile changing mode */
3541         if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
3542             chunk_soft_convert_filter(chunk_type, bargs)) {
3543                 return 0;
3544         }
3545
3546         /*
3547          * limited by count, must be the last filter
3548          */
3549         if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) {
3550                 if (bargs->limit == 0)
3551                         return 0;
3552                 else
3553                         bargs->limit--;
3554         } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) {
3555                 /*
3556                  * Same logic as the 'limit' filter; the minimum cannot be
3557                  * determined here because we do not have the global information
3558                  * about the count of all chunks that satisfy the filters.
3559                  */
3560                 if (bargs->limit_max == 0)
3561                         return 0;
3562                 else
3563                         bargs->limit_max--;
3564         }
3565
3566         return 1;
3567 }
3568
3569 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
3570 {
3571         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3572         struct btrfs_root *chunk_root = fs_info->chunk_root;
3573         struct btrfs_root *dev_root = fs_info->dev_root;
3574         struct list_head *devices;
3575         struct btrfs_device *device;
3576         u64 old_size;
3577         u64 size_to_free;
3578         u64 chunk_type;
3579         struct btrfs_chunk *chunk;
3580         struct btrfs_path *path = NULL;
3581         struct btrfs_key key;
3582         struct btrfs_key found_key;
3583         struct btrfs_trans_handle *trans;
3584         struct extent_buffer *leaf;
3585         int slot;
3586         int ret;
3587         int enospc_errors = 0;
3588         bool counting = true;
3589         /* The single value limit and min/max limits use the same bytes in the */
3590         u64 limit_data = bctl->data.limit;
3591         u64 limit_meta = bctl->meta.limit;
3592         u64 limit_sys = bctl->sys.limit;
3593         u32 count_data = 0;
3594         u32 count_meta = 0;
3595         u32 count_sys = 0;
3596         int chunk_reserved = 0;
3597
3598         /* step one make some room on all the devices */
3599         devices = &fs_info->fs_devices->devices;
3600         list_for_each_entry(device, devices, dev_list) {
3601                 old_size = btrfs_device_get_total_bytes(device);
3602                 size_to_free = div_factor(old_size, 1);
3603                 size_to_free = min_t(u64, size_to_free, SZ_1M);
3604                 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) ||
3605                     btrfs_device_get_total_bytes(device) -
3606                     btrfs_device_get_bytes_used(device) > size_to_free ||
3607                     test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
3608                         continue;
3609
3610                 ret = btrfs_shrink_device(device, old_size - size_to_free);
3611                 if (ret == -ENOSPC)
3612                         break;
3613                 if (ret) {
3614                         /* btrfs_shrink_device never returns ret > 0 */
3615                         WARN_ON(ret > 0);
3616                         goto error;
3617                 }
3618
3619                 trans = btrfs_start_transaction(dev_root, 0);
3620                 if (IS_ERR(trans)) {
3621                         ret = PTR_ERR(trans);
3622                         btrfs_info_in_rcu(fs_info,
3623                  "resize: unable to start transaction after shrinking device %s (error %d), old size %llu, new size %llu",
3624                                           rcu_str_deref(device->name), ret,
3625                                           old_size, old_size - size_to_free);
3626                         goto error;
3627                 }
3628
3629                 ret = btrfs_grow_device(trans, device, old_size);
3630                 if (ret) {
3631                         btrfs_end_transaction(trans);
3632                         /* btrfs_grow_device never returns ret > 0 */
3633                         WARN_ON(ret > 0);
3634                         btrfs_info_in_rcu(fs_info,
3635                  "resize: unable to grow device after shrinking device %s (error %d), old size %llu, new size %llu",
3636                                           rcu_str_deref(device->name), ret,
3637                                           old_size, old_size - size_to_free);
3638                         goto error;
3639                 }
3640
3641                 btrfs_end_transaction(trans);
3642         }
3643
3644         /* step two, relocate all the chunks */
3645         path = btrfs_alloc_path();
3646         if (!path) {
3647                 ret = -ENOMEM;
3648                 goto error;
3649         }
3650
3651         /* zero out stat counters */
3652         spin_lock(&fs_info->balance_lock);
3653         memset(&bctl->stat, 0, sizeof(bctl->stat));
3654         spin_unlock(&fs_info->balance_lock);
3655 again:
3656         if (!counting) {
3657                 /*
3658                  * The single value limit and min/max limits use the same bytes
3659                  * in the
3660                  */
3661                 bctl->data.limit = limit_data;
3662                 bctl->meta.limit = limit_meta;
3663                 bctl->sys.limit = limit_sys;
3664         }
3665         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3666         key.offset = (u64)-1;
3667         key.type = BTRFS_CHUNK_ITEM_KEY;
3668
3669         while (1) {
3670                 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
3671                     atomic_read(&fs_info->balance_cancel_req)) {
3672                         ret = -ECANCELED;
3673                         goto error;
3674                 }
3675
3676                 mutex_lock(&fs_info->delete_unused_bgs_mutex);
3677                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3678                 if (ret < 0) {
3679                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3680                         goto error;
3681                 }
3682
3683                 /*
3684                  * this shouldn't happen, it means the last relocate
3685                  * failed
3686                  */
3687                 if (ret == 0)
3688                         BUG(); /* FIXME break ? */
3689
3690                 ret = btrfs_previous_item(chunk_root, path, 0,
3691                                           BTRFS_CHUNK_ITEM_KEY);
3692                 if (ret) {
3693                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3694                         ret = 0;
3695                         break;
3696                 }
3697
3698                 leaf = path->nodes[0];
3699                 slot = path->slots[0];
3700                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3701
3702                 if (found_key.objectid != key.objectid) {
3703                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3704                         break;
3705                 }
3706
3707                 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3708                 chunk_type = btrfs_chunk_type(leaf, chunk);
3709
3710                 if (!counting) {
3711                         spin_lock(&fs_info->balance_lock);
3712                         bctl->stat.considered++;
3713                         spin_unlock(&fs_info->balance_lock);
3714                 }
3715
3716                 ret = should_balance_chunk(fs_info, leaf, chunk,
3717                                            found_key.offset);
3718
3719                 btrfs_release_path(path);
3720                 if (!ret) {
3721                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3722                         goto loop;
3723                 }
3724
3725                 if (counting) {
3726                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3727                         spin_lock(&fs_info->balance_lock);
3728                         bctl->stat.expected++;
3729                         spin_unlock(&fs_info->balance_lock);
3730
3731                         if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3732                                 count_data++;
3733                         else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3734                                 count_sys++;
3735                         else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3736                                 count_meta++;
3737
3738                         goto loop;
3739                 }
3740
3741                 /*
3742                  * Apply limit_min filter, no need to check if the LIMITS
3743                  * filter is used, limit_min is 0 by default
3744                  */
3745                 if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
3746                                         count_data < bctl->data.limit_min)
3747                                 || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) &&
3748                                         count_meta < bctl->meta.limit_min)
3749                                 || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) &&
3750                                         count_sys < bctl->sys.limit_min)) {
3751                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3752                         goto loop;
3753                 }
3754
3755                 if (!chunk_reserved) {
3756                         /*
3757                          * We may be relocating the only data chunk we have,
3758                          * which could potentially end up with losing data's
3759                          * raid profile, so lets allocate an empty one in
3760                          * advance.
3761                          */
3762                         ret = btrfs_may_alloc_data_chunk(fs_info,
3763                                                          found_key.offset);
3764                         if (ret < 0) {
3765                                 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3766                                 goto error;
3767                         } else if (ret == 1) {
3768                                 chunk_reserved = 1;
3769                         }
3770                 }
3771
3772                 ret = btrfs_relocate_chunk(fs_info, found_key.offset);
3773                 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3774                 if (ret && ret != -ENOSPC)
3775                         goto error;
3776                 if (ret == -ENOSPC) {
3777                         enospc_errors++;
3778                 } else {
3779                         spin_lock(&fs_info->balance_lock);
3780                         bctl->stat.completed++;
3781                         spin_unlock(&fs_info->balance_lock);
3782                 }
3783 loop:
3784                 if (found_key.offset == 0)
3785                         break;
3786                 key.offset = found_key.offset - 1;
3787         }
3788
3789         if (counting) {
3790                 btrfs_release_path(path);
3791                 counting = false;
3792                 goto again;
3793         }
3794 error:
3795         btrfs_free_path(path);
3796         if (enospc_errors) {
3797                 btrfs_info(fs_info, "%d enospc errors during balance",
3798                            enospc_errors);
3799                 if (!ret)
3800                         ret = -ENOSPC;
3801         }
3802
3803         return ret;
3804 }
3805
3806 /**
3807  * alloc_profile_is_valid - see if a given profile is valid and reduced
3808  * @flags: profile to validate
3809  * @extended: if true @flags is treated as an extended profile
3810  */
3811 static int alloc_profile_is_valid(u64 flags, int extended)
3812 {
3813         u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
3814                                BTRFS_BLOCK_GROUP_PROFILE_MASK);
3815
3816         flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
3817
3818         /* 1) check that all other bits are zeroed */
3819         if (flags & ~mask)
3820                 return 0;
3821
3822         /* 2) see if profile is reduced */
3823         if (flags == 0)
3824                 return !extended; /* "0" is valid for usual profiles */
3825
3826         /* true if exactly one bit set */
3827         return (flags & (flags - 1)) == 0;
3828 }
3829
3830 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
3831 {
3832         /* cancel requested || normal exit path */
3833         return atomic_read(&fs_info->balance_cancel_req) ||
3834                 (atomic_read(&fs_info->balance_pause_req) == 0 &&
3835                  atomic_read(&fs_info->balance_cancel_req) == 0);
3836 }
3837
3838 static void __cancel_balance(struct btrfs_fs_info *fs_info)
3839 {
3840         int ret;
3841
3842         unset_balance_control(fs_info);
3843         ret = del_balance_item(fs_info);
3844         if (ret)
3845                 btrfs_handle_fs_error(fs_info, ret, NULL);
3846
3847         clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
3848 }
3849
3850 /* Non-zero return value signifies invalidity */
3851 static inline int validate_convert_profile(struct btrfs_balance_args *bctl_arg,
3852                 u64 allowed)
3853 {
3854         return ((bctl_arg->flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3855                 (!alloc_profile_is_valid(bctl_arg->target, 1) ||
3856                  (bctl_arg->target & ~allowed)));
3857 }
3858
3859 /*
3860  * Should be called with both balance and volume mutexes held
3861  */
3862 int btrfs_balance(struct btrfs_balance_control *bctl,
3863                   struct btrfs_ioctl_balance_args *bargs)
3864 {
3865         struct btrfs_fs_info *fs_info = bctl->fs_info;
3866         u64 meta_target, data_target;
3867         u64 allowed;
3868         int mixed = 0;
3869         int ret;
3870         u64 num_devices;
3871         unsigned seq;
3872
3873         if (btrfs_fs_closing(fs_info) ||
3874             atomic_read(&fs_info->balance_pause_req) ||
3875             atomic_read(&fs_info->balance_cancel_req)) {
3876                 ret = -EINVAL;
3877                 goto out;
3878         }
3879
3880         allowed = btrfs_super_incompat_flags(fs_info->super_copy);
3881         if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
3882                 mixed = 1;
3883
3884         /*
3885          * In case of mixed groups both data and meta should be picked,
3886          * and identical options should be given for both of them.
3887          */
3888         allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
3889         if (mixed && (bctl->flags & allowed)) {
3890                 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
3891                     !(bctl->flags & BTRFS_BALANCE_METADATA) ||
3892                     memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
3893                         btrfs_err(fs_info,
3894                                   "with mixed groups data and metadata balance options must be the same");
3895                         ret = -EINVAL;
3896                         goto out;
3897                 }
3898         }
3899
3900         num_devices = fs_info->fs_devices->num_devices;
3901         btrfs_dev_replace_read_lock(&fs_info->dev_replace);
3902         if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
3903                 BUG_ON(num_devices < 1);
3904                 num_devices--;
3905         }
3906         btrfs_dev_replace_read_unlock(&fs_info->dev_replace);
3907         allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE | BTRFS_BLOCK_GROUP_DUP;
3908         if (num_devices > 1)
3909                 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1);
3910         if (num_devices > 2)
3911                 allowed |= BTRFS_BLOCK_GROUP_RAID5;
3912         if (num_devices > 3)
3913                 allowed |= (BTRFS_BLOCK_GROUP_RAID10 |
3914                             BTRFS_BLOCK_GROUP_RAID6);
3915         if (validate_convert_profile(&bctl->data, allowed)) {
3916                 btrfs_err(fs_info,
3917                           "unable to start balance with target data profile %llu",
3918                           bctl->data.target);
3919                 ret = -EINVAL;
3920                 goto out;
3921         }
3922         if (validate_convert_profile(&bctl->meta, allowed)) {
3923                 btrfs_err(fs_info,
3924                           "unable to start balance with target metadata profile %llu",
3925                           bctl->meta.target);
3926                 ret = -EINVAL;
3927                 goto out;
3928         }
3929         if (validate_convert_profile(&bctl->sys, allowed)) {
3930                 btrfs_err(fs_info,
3931                           "unable to start balance with target system profile %llu",
3932                           bctl->sys.target);
3933                 ret = -EINVAL;
3934                 goto out;
3935         }
3936
3937         /* allow to reduce meta or sys integrity only if force set */
3938         allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3939                         BTRFS_BLOCK_GROUP_RAID10 |
3940                         BTRFS_BLOCK_GROUP_RAID5 |
3941                         BTRFS_BLOCK_GROUP_RAID6;
3942         do {
3943                 seq = read_seqbegin(&fs_info->profiles_lock);
3944
3945                 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3946                      (fs_info->avail_system_alloc_bits & allowed) &&
3947                      !(bctl->sys.target & allowed)) ||
3948                     ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3949                      (fs_info->avail_metadata_alloc_bits & allowed) &&
3950                      !(bctl->meta.target & allowed))) {
3951                         if (bctl->flags & BTRFS_BALANCE_FORCE) {
3952                                 btrfs_info(fs_info,
3953                                            "force reducing metadata integrity");
3954                         } else {
3955                                 btrfs_err(fs_info,
3956                                           "balance will reduce metadata integrity, use force if you want this");
3957                                 ret = -EINVAL;
3958                                 goto out;
3959                         }
3960                 }
3961         } while (read_seqretry(&fs_info->profiles_lock, seq));
3962
3963         /* if we're not converting, the target field is uninitialized */
3964         meta_target = (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
3965                 bctl->meta.target : fs_info->avail_metadata_alloc_bits;
3966         data_target = (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
3967                 bctl->data.target : fs_info->avail_data_alloc_bits;
3968         if (btrfs_get_num_tolerated_disk_barrier_failures(meta_target) <
3969                 btrfs_get_num_tolerated_disk_barrier_failures(data_target)) {
3970                 btrfs_warn(fs_info,
3971                            "metadata profile 0x%llx has lower redundancy than data profile 0x%llx",
3972                            meta_target, data_target);
3973         }
3974
3975         ret = insert_balance_item(fs_info, bctl);
3976         if (ret && ret != -EEXIST)
3977                 goto out;
3978
3979         if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
3980                 BUG_ON(ret == -EEXIST);
3981                 set_balance_control(bctl);
3982         } else {
3983                 BUG_ON(ret != -EEXIST);
3984                 spin_lock(&fs_info->balance_lock);
3985                 update_balance_args(bctl);
3986                 spin_unlock(&fs_info->balance_lock);
3987         }
3988
3989         atomic_inc(&fs_info->balance_running);
3990         mutex_unlock(&fs_info->balance_mutex);
3991
3992         ret = __btrfs_balance(fs_info);
3993
3994         mutex_lock(&fs_info->balance_mutex);
3995         atomic_dec(&fs_info->balance_running);
3996
3997         if (bargs) {
3998                 memset(bargs, 0, sizeof(*bargs));
3999                 update_ioctl_balance_args(fs_info, 0, bargs);
4000         }
4001
4002         if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
4003             balance_need_close(fs_info)) {
4004                 __cancel_balance(fs_info);
4005         }
4006
4007         wake_up(&fs_info->balance_wait_q);
4008
4009         return ret;
4010 out:
4011         if (bctl->flags & BTRFS_BALANCE_RESUME)
4012                 __cancel_balance(fs_info);
4013         else {
4014                 kfree(bctl);
4015                 clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
4016         }
4017         return ret;
4018 }
4019
4020 static int balance_kthread(void *data)
4021 {
4022         struct btrfs_fs_info *fs_info = data;
4023         int ret = 0;
4024
4025         mutex_lock(&fs_info->volume_mutex);
4026         mutex_lock(&fs_info->balance_mutex);
4027
4028         if (fs_info->balance_ctl) {
4029                 btrfs_info(fs_info, "continuing balance");
4030                 ret = btrfs_balance(fs_info->balance_ctl, NULL);
4031         }
4032
4033         mutex_unlock(&fs_info->balance_mutex);
4034         mutex_unlock(&fs_info->volume_mutex);
4035
4036         return ret;
4037 }
4038
4039 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
4040 {
4041         struct task_struct *tsk;
4042
4043         spin_lock(&fs_info->balance_lock);
4044         if (!fs_info->balance_ctl) {
4045                 spin_unlock(&fs_info->balance_lock);
4046                 return 0;
4047         }
4048         spin_unlock(&fs_info->balance_lock);
4049
4050         if (btrfs_test_opt(fs_info, SKIP_BALANCE)) {
4051                 btrfs_info(fs_info, "force skipping balance");
4052                 return 0;
4053         }
4054
4055         /*
4056          * A ro->rw remount sequence should continue with the paused balance
4057          * regardless of who pauses it, system or the user as of now, so set
4058          * the resume flag.
4059          */
4060         spin_lock(&fs_info->balance_lock);
4061         fs_info->balance_ctl->flags |= BTRFS_BALANCE_RESUME;
4062         spin_unlock(&fs_info->balance_lock);
4063
4064         tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
4065         return PTR_ERR_OR_ZERO(tsk);
4066 }
4067
4068 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
4069 {
4070         struct btrfs_balance_control *bctl;
4071         struct btrfs_balance_item *item;
4072         struct btrfs_disk_balance_args disk_bargs;
4073         struct btrfs_path *path;
4074         struct extent_buffer *leaf;
4075         struct btrfs_key key;
4076         int ret;
4077
4078         path = btrfs_alloc_path();
4079         if (!path)
4080                 return -ENOMEM;
4081
4082         key.objectid = BTRFS_BALANCE_OBJECTID;
4083         key.type = BTRFS_TEMPORARY_ITEM_KEY;
4084         key.offset = 0;
4085
4086         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
4087         if (ret < 0)
4088                 goto out;
4089         if (ret > 0) { /* ret = -ENOENT; */
4090                 ret = 0;
4091                 goto out;
4092         }
4093
4094         bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
4095         if (!bctl) {
4096                 ret = -ENOMEM;
4097                 goto out;
4098         }
4099
4100         leaf = path->nodes[0];
4101         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
4102
4103         bctl->fs_info = fs_info;
4104         bctl->flags = btrfs_balance_flags(leaf, item);
4105         bctl->flags |= BTRFS_BALANCE_RESUME;
4106
4107         btrfs_balance_data(leaf, item, &disk_bargs);
4108         btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
4109         btrfs_balance_meta(leaf, item, &disk_bargs);
4110         btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
4111         btrfs_balance_sys(leaf, item, &disk_bargs);
4112         btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
4113
4114         WARN_ON(test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags));
4115
4116         mutex_lock(&fs_info->volume_mutex);
4117         mutex_lock(&fs_info->balance_mutex);
4118
4119         set_balance_control(bctl);
4120
4121         mutex_unlock(&fs_info->balance_mutex);
4122         mutex_unlock(&fs_info->volume_mutex);
4123 out:
4124         btrfs_free_path(path);
4125         return ret;
4126 }
4127
4128 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
4129 {
4130         int ret = 0;
4131
4132         mutex_lock(&fs_info->balance_mutex);
4133         if (!fs_info->balance_ctl) {
4134                 mutex_unlock(&fs_info->balance_mutex);
4135                 return -ENOTCONN;
4136         }
4137
4138         if (atomic_read(&fs_info->balance_running)) {
4139                 atomic_inc(&fs_info->balance_pause_req);
4140                 mutex_unlock(&fs_info->balance_mutex);
4141
4142                 wait_event(fs_info->balance_wait_q,
4143                            atomic_read(&fs_info->balance_running) == 0);
4144
4145                 mutex_lock(&fs_info->balance_mutex);
4146                 /* we are good with balance_ctl ripped off from under us */
4147                 BUG_ON(atomic_read(&fs_info->balance_running));
4148                 atomic_dec(&fs_info->balance_pause_req);
4149         } else {
4150                 ret = -ENOTCONN;
4151         }
4152
4153         mutex_unlock(&fs_info->balance_mutex);
4154         return ret;
4155 }
4156
4157 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
4158 {
4159         if (sb_rdonly(fs_info->sb))
4160                 return -EROFS;
4161
4162         mutex_lock(&fs_info->balance_mutex);
4163         if (!fs_info->balance_ctl) {
4164                 mutex_unlock(&fs_info->balance_mutex);
4165                 return -ENOTCONN;
4166         }
4167
4168         atomic_inc(&fs_info->balance_cancel_req);
4169         /*
4170          * if we are running just wait and return, balance item is
4171          * deleted in btrfs_balance in this case
4172          */
4173         if (atomic_read(&fs_info->balance_running)) {
4174                 mutex_unlock(&fs_info->balance_mutex);
4175                 wait_event(fs_info->balance_wait_q,
4176                            atomic_read(&fs_info->balance_running) == 0);
4177                 mutex_lock(&fs_info->balance_mutex);
4178         } else {
4179                 /* __cancel_balance needs volume_mutex */
4180                 mutex_unlock(&fs_info->balance_mutex);
4181                 mutex_lock(&fs_info->volume_mutex);
4182                 mutex_lock(&fs_info->balance_mutex);
4183
4184                 if (fs_info->balance_ctl)
4185                         __cancel_balance(fs_info);
4186
4187                 mutex_unlock(&fs_info->volume_mutex);
4188         }
4189
4190         BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running));
4191         atomic_dec(&fs_info->balance_cancel_req);
4192         mutex_unlock(&fs_info->balance_mutex);
4193         return 0;
4194 }
4195
4196 static int btrfs_uuid_scan_kthread(void *data)
4197 {
4198         struct btrfs_fs_info *fs_info = data;
4199         struct btrfs_root *root = fs_info->tree_root;
4200         struct btrfs_key key;
4201         struct btrfs_path *path = NULL;
4202         int ret = 0;
4203         struct extent_buffer *eb;
4204         int slot;
4205         struct btrfs_root_item root_item;
4206         u32 item_size;
4207         struct btrfs_trans_handle *trans = NULL;
4208
4209         path = btrfs_alloc_path();
4210         if (!path) {
4211                 ret = -ENOMEM;
4212                 goto out;
4213         }
4214
4215         key.objectid = 0;
4216         key.type = BTRFS_ROOT_ITEM_KEY;
4217         key.offset = 0;
4218
4219         while (1) {
4220                 ret = btrfs_search_forward(root, &key, path,
4221                                 BTRFS_OLDEST_GENERATION);
4222                 if (ret) {
4223                         if (ret > 0)
4224                                 ret = 0;
4225                         break;
4226                 }
4227
4228                 if (key.type != BTRFS_ROOT_ITEM_KEY ||
4229                     (key.objectid < BTRFS_FIRST_FREE_OBJECTID &&
4230                      key.objectid != BTRFS_FS_TREE_OBJECTID) ||
4231                     key.objectid > BTRFS_LAST_FREE_OBJECTID)
4232                         goto skip;
4233
4234                 eb = path->nodes[0];
4235                 slot = path->slots[0];
4236                 item_size = btrfs_item_size_nr(eb, slot);
4237                 if (item_size < sizeof(root_item))
4238                         goto skip;
4239
4240                 read_extent_buffer(eb, &root_item,
4241                                    btrfs_item_ptr_offset(eb, slot),
4242                                    (int)sizeof(root_item));
4243                 if (btrfs_root_refs(&root_item) == 0)
4244                         goto skip;
4245
4246                 if (!btrfs_is_empty_uuid(root_item.uuid) ||
4247                     !btrfs_is_empty_uuid(root_item.received_uuid)) {
4248                         if (trans)
4249                                 goto update_tree;
4250
4251                         btrfs_release_path(path);
4252                         /*
4253                          * 1 - subvol uuid item
4254                          * 1 - received_subvol uuid item
4255                          */
4256                         trans = btrfs_start_transaction(fs_info->uuid_root, 2);
4257                         if (IS_ERR(trans)) {
4258                                 ret = PTR_ERR(trans);
4259                                 break;
4260                         }
4261                         continue;
4262                 } else {
4263                         goto skip;
4264                 }
4265 update_tree:
4266                 if (!btrfs_is_empty_uuid(root_item.uuid)) {
4267                         ret = btrfs_uuid_tree_add(trans, fs_info,
4268                                                   root_item.uuid,
4269                                                   BTRFS_UUID_KEY_SUBVOL,
4270                                                   key.objectid);
4271                         if (ret < 0) {
4272                                 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4273                                         ret);
4274                                 break;
4275                         }
4276                 }
4277
4278                 if (!btrfs_is_empty_uuid(root_item.received_uuid)) {
4279                         ret = btrfs_uuid_tree_add(trans, fs_info,
4280                                                   root_item.received_uuid,
4281                                                  BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4282                                                   key.objectid);
4283                         if (ret < 0) {
4284                                 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4285                                         ret);
4286                                 break;
4287                         }
4288                 }
4289
4290 skip:
4291                 if (trans) {
4292                         ret = btrfs_end_transaction(trans);
4293                         trans = NULL;
4294                         if (ret)
4295                                 break;
4296                 }
4297
4298                 btrfs_release_path(path);
4299                 if (key.offset < (u64)-1) {
4300                         key.offset++;
4301                 } else if (key.type < BTRFS_ROOT_ITEM_KEY) {
4302                         key.offset = 0;
4303                         key.type = BTRFS_ROOT_ITEM_KEY;
4304                 } else if (key.objectid < (u64)-1) {
4305                         key.offset = 0;
4306                         key.type = BTRFS_ROOT_ITEM_KEY;
4307                         key.objectid++;
4308                 } else {
4309                         break;
4310                 }
4311                 cond_resched();
4312         }
4313
4314 out:
4315         btrfs_free_path(path);
4316         if (trans && !IS_ERR(trans))
4317                 btrfs_end_transaction(trans);
4318         if (ret)
4319                 btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret);
4320         else
4321                 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
4322         up(&fs_info->uuid_tree_rescan_sem);
4323         return 0;
4324 }
4325
4326 /*
4327  * Callback for btrfs_uuid_tree_iterate().
4328  * returns:
4329  * 0    check succeeded, the entry is not outdated.
4330  * < 0  if an error occurred.
4331  * > 0  if the check failed, which means the caller shall remove the entry.
4332  */
4333 static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info *fs_info,
4334                                        u8 *uuid, u8 type, u64 subid)
4335 {
4336         struct btrfs_key key;
4337         int ret = 0;
4338         struct btrfs_root *subvol_root;
4339
4340         if (type != BTRFS_UUID_KEY_SUBVOL &&
4341             type != BTRFS_UUID_KEY_RECEIVED_SUBVOL)
4342                 goto out;
4343
4344         key.objectid = subid;
4345         key.type = BTRFS_ROOT_ITEM_KEY;
4346         key.offset = (u64)-1;
4347         subvol_root = btrfs_read_fs_root_no_name(fs_info, &key);
4348         if (IS_ERR(subvol_root)) {
4349                 ret = PTR_ERR(subvol_root);
4350                 if (ret == -ENOENT)
4351                         ret = 1;
4352                 goto out;
4353         }
4354
4355         switch (type) {
4356         case BTRFS_UUID_KEY_SUBVOL:
4357                 if (memcmp(uuid, subvol_root->root_item.uuid, BTRFS_UUID_SIZE))
4358                         ret = 1;
4359                 break;
4360         case BTRFS_UUID_KEY_RECEIVED_SUBVOL:
4361                 if (memcmp(uuid, subvol_root->root_item.received_uuid,
4362                            BTRFS_UUID_SIZE))
4363                         ret = 1;
4364                 break;
4365         }
4366
4367 out:
4368         return ret;
4369 }
4370
4371 static int btrfs_uuid_rescan_kthread(void *data)
4372 {
4373         struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data;
4374         int ret;
4375
4376         /*
4377          * 1st step is to iterate through the existing UUID tree and
4378          * to delete all entries that contain outdated data.
4379          * 2nd step is to add all missing entries to the UUID tree.
4380          */
4381         ret = btrfs_uuid_tree_iterate(fs_info, btrfs_check_uuid_tree_entry);
4382         if (ret < 0) {
4383                 btrfs_warn(fs_info, "iterating uuid_tree failed %d", ret);
4384                 up(&fs_info->uuid_tree_rescan_sem);
4385                 return ret;
4386         }
4387         return btrfs_uuid_scan_kthread(data);
4388 }
4389
4390 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info)
4391 {
4392         struct btrfs_trans_handle *trans;
4393         struct btrfs_root *tree_root = fs_info->tree_root;
4394         struct btrfs_root *uuid_root;
4395         struct task_struct *task;
4396         int ret;
4397
4398         /*
4399          * 1 - root node
4400          * 1 - root item
4401          */
4402         trans = btrfs_start_transaction(tree_root, 2);
4403         if (IS_ERR(trans))
4404                 return PTR_ERR(trans);
4405
4406         uuid_root = btrfs_create_tree(trans, fs_info,
4407                                       BTRFS_UUID_TREE_OBJECTID);
4408         if (IS_ERR(uuid_root)) {
4409                 ret = PTR_ERR(uuid_root);
4410                 btrfs_abort_transaction(trans, ret);
4411                 btrfs_end_transaction(trans);
4412                 return ret;
4413         }
4414
4415         fs_info->uuid_root = uuid_root;
4416
4417         ret = btrfs_commit_transaction(trans);
4418         if (ret)
4419                 return ret;
4420
4421         down(&fs_info->uuid_tree_rescan_sem);
4422         task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid");
4423         if (IS_ERR(task)) {
4424                 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4425                 btrfs_warn(fs_info, "failed to start uuid_scan task");
4426                 up(&fs_info->uuid_tree_rescan_sem);
4427                 return PTR_ERR(task);
4428         }
4429
4430         return 0;
4431 }
4432
4433 int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
4434 {
4435         struct task_struct *task;
4436
4437         down(&fs_info->uuid_tree_rescan_sem);
4438         task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
4439         if (IS_ERR(task)) {
4440                 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4441                 btrfs_warn(fs_info, "failed to start uuid_rescan task");
4442                 up(&fs_info->uuid_tree_rescan_sem);
4443                 return PTR_ERR(task);
4444         }
4445
4446         return 0;
4447 }
4448
4449 /*
4450  * shrinking a device means finding all of the device extents past
4451  * the new size, and then following the back refs to the chunks.
4452  * The chunk relocation code actually frees the device extent
4453  */
4454 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
4455 {
4456         struct btrfs_fs_info *fs_info = device->fs_info;
4457         struct btrfs_root *root = fs_info->dev_root;
4458         struct btrfs_trans_handle *trans;
4459         struct btrfs_dev_extent *dev_extent = NULL;
4460         struct btrfs_path *path;
4461         u64 length;
4462         u64 chunk_offset;
4463         int ret;
4464         int slot;
4465         int failed = 0;
4466         bool retried = false;
4467         bool checked_pending_chunks = false;
4468         struct extent_buffer *l;
4469         struct btrfs_key key;
4470         struct btrfs_super_block *super_copy = fs_info->super_copy;
4471         u64 old_total = btrfs_super_total_bytes(super_copy);
4472         u64 old_size = btrfs_device_get_total_bytes(device);
4473         u64 diff;
4474
4475         new_size = round_down(new_size, fs_info->sectorsize);
4476         diff = round_down(old_size - new_size, fs_info->sectorsize);
4477
4478         if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
4479                 return -EINVAL;
4480
4481         path = btrfs_alloc_path();
4482         if (!path)
4483                 return -ENOMEM;
4484
4485         path->reada = READA_FORWARD;
4486
4487         mutex_lock(&fs_info->chunk_mutex);
4488
4489         btrfs_device_set_total_bytes(device, new_size);
4490         if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
4491                 device->fs_devices->total_rw_bytes -= diff;
4492                 atomic64_sub(diff, &fs_info->free_chunk_space);
4493         }
4494         mutex_unlock(&fs_info->chunk_mutex);
4495
4496 again:
4497         key.objectid = device->devid;
4498         key.offset = (u64)-1;
4499         key.type = BTRFS_DEV_EXTENT_KEY;
4500
4501         do {
4502                 mutex_lock(&fs_info->delete_unused_bgs_mutex);
4503                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4504                 if (ret < 0) {
4505                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4506                         goto done;
4507                 }
4508
4509                 ret = btrfs_previous_item(root, path, 0, key.type);
4510                 if (ret)
4511                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4512                 if (ret < 0)
4513                         goto done;
4514                 if (ret) {
4515                         ret = 0;
4516                         btrfs_release_path(path);
4517                         break;
4518                 }
4519
4520                 l = path->nodes[0];
4521                 slot = path->slots[0];
4522                 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
4523
4524                 if (key.objectid != device->devid) {
4525                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4526                         btrfs_release_path(path);
4527                         break;
4528                 }
4529
4530                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
4531                 length = btrfs_dev_extent_length(l, dev_extent);
4532
4533                 if (key.offset + length <= new_size) {
4534                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4535                         btrfs_release_path(path);
4536                         break;
4537                 }
4538
4539                 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
4540                 btrfs_release_path(path);
4541
4542                 /*
4543                  * We may be relocating the only data chunk we have,
4544                  * which could potentially end up with losing data's
4545                  * raid profile, so lets allocate an empty one in
4546                  * advance.
4547                  */
4548                 ret = btrfs_may_alloc_data_chunk(fs_info, chunk_offset);
4549                 if (ret < 0) {
4550                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4551                         goto done;
4552                 }
4553
4554                 ret = btrfs_relocate_chunk(fs_info, chunk_offset);
4555                 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4556                 if (ret && ret != -ENOSPC)
4557                         goto done;
4558                 if (ret == -ENOSPC)
4559                         failed++;
4560         } while (key.offset-- > 0);
4561
4562         if (failed && !retried) {
4563                 failed = 0;
4564                 retried = true;
4565                 goto again;
4566         } else if (failed && retried) {
4567                 ret = -ENOSPC;
4568                 goto done;
4569         }
4570
4571         /* Shrinking succeeded, else we would be at "done". */
4572         trans = btrfs_start_transaction(root, 0);
4573         if (IS_ERR(trans)) {
4574                 ret = PTR_ERR(trans);
4575                 goto done;
4576         }
4577
4578         mutex_lock(&fs_info->chunk_mutex);
4579
4580         /*
4581          * We checked in the above loop all device extents that were already in
4582          * the device tree. However before we have updated the device's
4583          * total_bytes to the new size, we might have had chunk allocations that
4584          * have not complete yet (new block groups attached to transaction
4585          * handles), and therefore their device extents were not yet in the
4586          * device tree and we missed them in the loop above. So if we have any
4587          * pending chunk using a device extent that overlaps the device range
4588          * that we can not use anymore, commit the current transaction and
4589          * repeat the search on the device tree - this way we guarantee we will
4590          * not have chunks using device extents that end beyond 'new_size'.
4591          */
4592         if (!checked_pending_chunks) {
4593                 u64 start = new_size;
4594                 u64 len = old_size - new_size;
4595
4596                 if (contains_pending_extent(trans->transaction, device,
4597                                             &start, len)) {
4598                         mutex_unlock(&fs_info->chunk_mutex);
4599                         checked_pending_chunks = true;
4600                         failed = 0;
4601                         retried = false;
4602                         ret = btrfs_commit_transaction(trans);
4603                         if (ret)
4604                                 goto done;
4605                         goto again;
4606                 }
4607         }
4608
4609         btrfs_device_set_disk_total_bytes(device, new_size);
4610         if (list_empty(&device->resized_list))
4611                 list_add_tail(&device->resized_list,
4612                               &fs_info->fs_devices->resized_devices);
4613
4614         WARN_ON(diff > old_total);
4615         btrfs_set_super_total_bytes(super_copy,
4616                         round_down(old_total - diff, fs_info->sectorsize));
4617         mutex_unlock(&fs_info->chunk_mutex);
4618
4619         /* Now btrfs_update_device() will change the on-disk size. */
4620         ret = btrfs_update_device(trans, device);
4621         btrfs_end_transaction(trans);
4622 done:
4623         btrfs_free_path(path);
4624         if (ret) {
4625                 mutex_lock(&fs_info->chunk_mutex);
4626                 btrfs_device_set_total_bytes(device, old_size);
4627                 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
4628                         device->fs_devices->total_rw_bytes += diff;
4629                 atomic64_add(diff, &fs_info->free_chunk_space);
4630                 mutex_unlock(&fs_info->chunk_mutex);
4631         }
4632         return ret;
4633 }
4634
4635 static int btrfs_add_system_chunk(struct btrfs_fs_info *fs_info,
4636                            struct btrfs_key *key,
4637                            struct btrfs_chunk *chunk, int item_size)
4638 {
4639         struct btrfs_super_block *super_copy = fs_info->super_copy;
4640         struct btrfs_disk_key disk_key;
4641         u32 array_size;
4642         u8 *ptr;
4643
4644         mutex_lock(&fs_info->chunk_mutex);
4645         array_size = btrfs_super_sys_array_size(super_copy);
4646         if (array_size + item_size + sizeof(disk_key)
4647                         > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
4648                 mutex_unlock(&fs_info->chunk_mutex);
4649                 return -EFBIG;
4650         }
4651
4652         ptr = super_copy->sys_chunk_array + array_size;
4653         btrfs_cpu_key_to_disk(&disk_key, key);
4654         memcpy(ptr, &disk_key, sizeof(disk_key));
4655         ptr += sizeof(disk_key);
4656         memcpy(ptr, chunk, item_size);
4657         item_size += sizeof(disk_key);
4658         btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
4659         mutex_unlock(&fs_info->chunk_mutex);
4660
4661         return 0;
4662 }
4663
4664 /*
4665  * sort the devices in descending order by max_avail, total_avail
4666  */
4667 static int btrfs_cmp_device_info(const void *a, const void *b)
4668 {
4669         const struct btrfs_device_info *di_a = a;
4670         const struct btrfs_device_info *di_b = b;
4671
4672         if (di_a->max_avail > di_b->max_avail)
4673                 return -1;
4674         if (di_a->max_avail < di_b->max_avail)
4675                 return 1;
4676         if (di_a->total_avail > di_b->total_avail)
4677                 return -1;
4678         if (di_a->total_avail < di_b->total_avail)
4679                 return 1;
4680         return 0;
4681 }
4682
4683 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
4684 {
4685         if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK))
4686                 return;
4687
4688         btrfs_set_fs_incompat(info, RAID56);
4689 }
4690
4691 #define BTRFS_MAX_DEVS(info) ((BTRFS_MAX_ITEM_SIZE(info)        \
4692                         - sizeof(struct btrfs_chunk))           \
4693                         / sizeof(struct btrfs_stripe) + 1)
4694
4695 #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE        \
4696                                 - 2 * sizeof(struct btrfs_disk_key)     \
4697                                 - 2 * sizeof(struct btrfs_chunk))       \
4698                                 / sizeof(struct btrfs_stripe) + 1)
4699
4700 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4701                                u64 start, u64 type)
4702 {
4703         struct btrfs_fs_info *info = trans->fs_info;
4704         struct btrfs_fs_devices *fs_devices = info->fs_devices;
4705         struct btrfs_device *device;
4706         struct map_lookup *map = NULL;
4707         struct extent_map_tree *em_tree;
4708         struct extent_map *em;
4709         struct btrfs_device_info *devices_info = NULL;
4710         u64 total_avail;
4711         int num_stripes;        /* total number of stripes to allocate */
4712         int data_stripes;       /* number of stripes that count for
4713                                    block group size */
4714         int sub_stripes;        /* sub_stripes info for map */
4715         int dev_stripes;        /* stripes per dev */
4716         int devs_max;           /* max devs to use */
4717         int devs_min;           /* min devs needed */
4718         int devs_increment;     /* ndevs has to be a multiple of this */
4719         int ncopies;            /* how many copies to data has */
4720         int ret;
4721         u64 max_stripe_size;
4722         u64 max_chunk_size;
4723         u64 stripe_size;
4724         u64 num_bytes;
4725         int ndevs;
4726         int i;
4727         int j;
4728         int index;
4729
4730         BUG_ON(!alloc_profile_is_valid(type, 0));
4731
4732         if (list_empty(&fs_devices->alloc_list)) {
4733                 if (btrfs_test_opt(info, ENOSPC_DEBUG))
4734                         btrfs_debug(info, "%s: no writable device", __func__);
4735                 return -ENOSPC;
4736         }
4737
4738         index = btrfs_bg_flags_to_raid_index(type);
4739
4740         sub_stripes = btrfs_raid_array[index].sub_stripes;
4741         dev_stripes = btrfs_raid_array[index].dev_stripes;
4742         devs_max = btrfs_raid_array[index].devs_max;
4743         devs_min = btrfs_raid_array[index].devs_min;
4744         devs_increment = btrfs_raid_array[index].devs_increment;
4745         ncopies = btrfs_raid_array[index].ncopies;
4746
4747         if (type & BTRFS_BLOCK_GROUP_DATA) {
4748                 max_stripe_size = SZ_1G;
4749                 max_chunk_size = 10 * max_stripe_size;
4750                 if (!devs_max)
4751                         devs_max = BTRFS_MAX_DEVS(info);
4752         } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
4753                 /* for larger filesystems, use larger metadata chunks */
4754                 if (fs_devices->total_rw_bytes > 50ULL * SZ_1G)
4755                         max_stripe_size = SZ_1G;
4756                 else
4757                         max_stripe_size = SZ_256M;
4758                 max_chunk_size = max_stripe_size;
4759                 if (!devs_max)
4760                         devs_max = BTRFS_MAX_DEVS(info);
4761         } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
4762                 max_stripe_size = SZ_32M;
4763                 max_chunk_size = 2 * max_stripe_size;
4764                 if (!devs_max)
4765                         devs_max = BTRFS_MAX_DEVS_SYS_CHUNK;
4766         } else {
4767                 btrfs_err(info, "invalid chunk type 0x%llx requested",
4768                        type);
4769                 BUG_ON(1);
4770         }
4771
4772         /* we don't want a chunk larger than 10% of writeable space */
4773         max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
4774                              max_chunk_size);
4775
4776         devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info),
4777                                GFP_NOFS);
4778         if (!devices_info)
4779                 return -ENOMEM;
4780
4781         /*
4782          * in the first pass through the devices list, we gather information
4783          * about the available holes on each device.
4784          */
4785         ndevs = 0;
4786         list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
4787                 u64 max_avail;
4788                 u64 dev_offset;
4789
4790                 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
4791                         WARN(1, KERN_ERR
4792                                "BTRFS: read-only device in alloc_list\n");
4793                         continue;
4794                 }
4795
4796                 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
4797                                         &device->dev_state) ||
4798                     test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
4799                         continue;
4800
4801                 if (device->total_bytes > device->bytes_used)
4802                         total_avail = device->total_bytes - device->bytes_used;
4803                 else
4804                         total_avail = 0;
4805
4806                 /* If there is no space on this device, skip it. */
4807                 if (total_avail == 0)
4808                         continue;
4809
4810                 ret = find_free_dev_extent(trans, device,
4811                                            max_stripe_size * dev_stripes,
4812                                            &dev_offset, &max_avail);
4813                 if (ret && ret != -ENOSPC)
4814                         goto error;
4815
4816                 if (ret == 0)
4817                         max_avail = max_stripe_size * dev_stripes;
4818
4819                 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes) {
4820                         if (btrfs_test_opt(info, ENOSPC_DEBUG))
4821                                 btrfs_debug(info,
4822                         "%s: devid %llu has no free space, have=%llu want=%u",
4823                                             __func__, device->devid, max_avail,
4824                                             BTRFS_STRIPE_LEN * dev_stripes);
4825                         continue;
4826                 }
4827
4828                 if (ndevs == fs_devices->rw_devices) {
4829                         WARN(1, "%s: found more than %llu devices\n",
4830                              __func__, fs_devices->rw_devices);
4831                         break;
4832                 }
4833                 devices_info[ndevs].dev_offset = dev_offset;
4834                 devices_info[ndevs].max_avail = max_avail;
4835                 devices_info[ndevs].total_avail = total_avail;
4836                 devices_info[ndevs].dev = device;
4837                 ++ndevs;
4838         }
4839
4840         /*
4841          * now sort the devices by hole size / available space
4842          */
4843         sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
4844              btrfs_cmp_device_info, NULL);
4845
4846         /* round down to number of usable stripes */
4847         ndevs = round_down(ndevs, devs_increment);
4848
4849         if (ndevs < devs_min) {
4850                 ret = -ENOSPC;
4851                 if (btrfs_test_opt(info, ENOSPC_DEBUG)) {
4852                         btrfs_debug(info,
4853         "%s: not enough devices with free space: have=%d minimum required=%d",
4854                                     __func__, ndevs, devs_min);
4855                 }
4856                 goto error;
4857         }
4858
4859         ndevs = min(ndevs, devs_max);
4860
4861         /*
4862          * The primary goal is to maximize the number of stripes, so use as
4863          * many devices as possible, even if the stripes are not maximum sized.
4864          *
4865          * The DUP profile stores more than one stripe per device, the
4866          * max_avail is the total size so we have to adjust.
4867          */
4868         stripe_size = div_u64(devices_info[ndevs - 1].max_avail, dev_stripes);
4869         num_stripes = ndevs * dev_stripes;
4870
4871         /*
4872          * this will have to be fixed for RAID1 and RAID10 over
4873          * more drives
4874          */
4875         data_stripes = num_stripes / ncopies;
4876
4877         if (type & BTRFS_BLOCK_GROUP_RAID5)
4878                 data_stripes = num_stripes - 1;
4879
4880         if (type & BTRFS_BLOCK_GROUP_RAID6)
4881                 data_stripes = num_stripes - 2;
4882
4883         /*
4884          * Use the number of data stripes to figure out how big this chunk
4885          * is really going to be in terms of logical address space,
4886          * and compare that answer with the max chunk size
4887          */
4888         if (stripe_size * data_stripes > max_chunk_size) {
4889                 stripe_size = div_u64(max_chunk_size, data_stripes);
4890
4891                 /* bump the answer up to a 16MB boundary */
4892                 stripe_size = round_up(stripe_size, SZ_16M);
4893
4894                 /*
4895                  * But don't go higher than the limits we found while searching
4896                  * for free extents
4897                  */
4898                 stripe_size = min(devices_info[ndevs - 1].max_avail,
4899                                   stripe_size);
4900         }
4901
4902         /* align to BTRFS_STRIPE_LEN */
4903         stripe_size = round_down(stripe_size, BTRFS_STRIPE_LEN);
4904
4905         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
4906         if (!map) {
4907                 ret = -ENOMEM;
4908                 goto error;
4909         }
4910         map->num_stripes = num_stripes;
4911
4912         for (i = 0; i < ndevs; ++i) {
4913                 for (j = 0; j < dev_stripes; ++j) {
4914                         int s = i * dev_stripes + j;
4915                         map->stripes[s].dev = devices_info[i].dev;
4916                         map->stripes[s].physical = devices_info[i].dev_offset +
4917                                                    j * stripe_size;
4918                 }
4919         }
4920         map->stripe_len = BTRFS_STRIPE_LEN;
4921         map->io_align = BTRFS_STRIPE_LEN;
4922         map->io_width = BTRFS_STRIPE_LEN;
4923         map->type = type;
4924         map->sub_stripes = sub_stripes;
4925
4926         num_bytes = stripe_size * data_stripes;
4927
4928         trace_btrfs_chunk_alloc(info, map, start, num_bytes);
4929
4930         em = alloc_extent_map();
4931         if (!em) {
4932                 kfree(map);
4933                 ret = -ENOMEM;
4934                 goto error;
4935         }
4936         set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
4937         em->map_lookup = map;
4938         em->start = start;
4939         em->len = num_bytes;
4940         em->block_start = 0;
4941         em->block_len = em->len;
4942         em->orig_block_len = stripe_size;
4943
4944         em_tree = &info->mapping_tree.map_tree;
4945         write_lock(&em_tree->lock);
4946         ret = add_extent_mapping(em_tree, em, 0);
4947         if (ret) {
4948                 write_unlock(&em_tree->lock);
4949                 free_extent_map(em);
4950                 goto error;
4951         }
4952
4953         list_add_tail(&em->list, &trans->transaction->pending_chunks);
4954         refcount_inc(&em->refs);
4955         write_unlock(&em_tree->lock);
4956
4957         ret = btrfs_make_block_group(trans, info, 0, type, start, num_bytes);
4958         if (ret)
4959                 goto error_del_extent;
4960
4961         for (i = 0; i < map->num_stripes; i++) {
4962                 num_bytes = map->stripes[i].dev->bytes_used + stripe_size;
4963                 btrfs_device_set_bytes_used(map->stripes[i].dev, num_bytes);
4964         }
4965
4966         atomic64_sub(stripe_size * map->num_stripes, &info->free_chunk_space);
4967
4968         free_extent_map(em);
4969         check_raid56_incompat_flag(info, type);
4970
4971         kfree(devices_info);
4972         return 0;
4973
4974 error_del_extent:
4975         write_lock(&em_tree->lock);
4976         remove_extent_mapping(em_tree, em);
4977         write_unlock(&em_tree->lock);
4978
4979         /* One for our allocation */
4980         free_extent_map(em);
4981         /* One for the tree reference */
4982         free_extent_map(em);
4983         /* One for the pending_chunks list reference */
4984         free_extent_map(em);
4985 error:
4986         kfree(devices_info);
4987         return ret;
4988 }
4989
4990 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans,
4991                                 struct btrfs_fs_info *fs_info,
4992                                 u64 chunk_offset, u64 chunk_size)
4993 {
4994         struct btrfs_root *extent_root = fs_info->extent_root;
4995         struct btrfs_root *chunk_root = fs_info->chunk_root;
4996         struct btrfs_key key;
4997         struct btrfs_device *device;
4998         struct btrfs_chunk *chunk;
4999         struct btrfs_stripe *stripe;
5000         struct extent_map *em;
5001         struct map_lookup *map;
5002         size_t item_size;
5003         u64 dev_offset;
5004         u64 stripe_size;
5005         int i = 0;
5006         int ret = 0;
5007
5008         em = get_chunk_map(fs_info, chunk_offset, chunk_size);
5009         if (IS_ERR(em))
5010                 return PTR_ERR(em);
5011
5012         map = em->map_lookup;
5013         item_size = btrfs_chunk_item_size(map->num_stripes);
5014         stripe_size = em->orig_block_len;
5015
5016         chunk = kzalloc(item_size, GFP_NOFS);
5017         if (!chunk) {
5018                 ret = -ENOMEM;
5019                 goto out;
5020         }
5021
5022         /*
5023          * Take the device list mutex to prevent races with the final phase of
5024          * a device replace operation that replaces the device object associated
5025          * with the map's stripes, because the device object's id can change
5026          * at any time during that final phase of the device replace operation
5027          * (dev-replace.c:btrfs_dev_replace_finishing()).
5028          */
5029         mutex_lock(&fs_info->fs_devices->device_list_mutex);
5030         for (i = 0; i < map->num_stripes; i++) {
5031                 device = map->stripes[i].dev;
5032                 dev_offset = map->stripes[i].physical;
5033
5034                 ret = btrfs_update_device(trans, device);
5035                 if (ret)
5036                         break;
5037                 ret = btrfs_alloc_dev_extent(trans, device, chunk_offset,
5038                                              dev_offset, stripe_size);
5039                 if (ret)
5040                         break;
5041         }
5042         if (ret) {
5043                 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
5044                 goto out;
5045         }
5046
5047         stripe = &chunk->stripe;
5048         for (i = 0; i < map->num_stripes; i++) {
5049                 device = map->stripes[i].dev;
5050                 dev_offset = map->stripes[i].physical;
5051
5052                 btrfs_set_stack_stripe_devid(stripe, device->devid);
5053                 btrfs_set_stack_stripe_offset(stripe, dev_offset);
5054                 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
5055                 stripe++;
5056         }
5057         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
5058
5059         btrfs_set_stack_chunk_length(chunk, chunk_size);
5060         btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
5061         btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
5062         btrfs_set_stack_chunk_type(chunk, map->type);
5063         btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
5064         btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
5065         btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
5066         btrfs_set_stack_chunk_sector_size(chunk, fs_info->sectorsize);
5067         btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
5068
5069         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
5070         key.type = BTRFS_CHUNK_ITEM_KEY;
5071         key.offset = chunk_offset;
5072
5073         ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
5074         if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
5075                 /*
5076                  * TODO: Cleanup of inserted chunk root in case of
5077                  * failure.
5078                  */
5079                 ret = btrfs_add_system_chunk(fs_info, &key, chunk, item_size);
5080         }
5081
5082 out:
5083         kfree(chunk);
5084         free_extent_map(em);
5085         return ret;
5086 }
5087
5088 /*
5089  * Chunk allocation falls into two parts. The first part does works
5090  * that make the new allocated chunk useable, but not do any operation
5091  * that modifies the chunk tree. The second part does the works that
5092  * require modifying the chunk tree. This division is important for the
5093  * bootstrap process of adding storage to a seed btrfs.
5094  */
5095 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
5096                       struct btrfs_fs_info *fs_info, u64 type)
5097 {
5098         u64 chunk_offset;
5099
5100         lockdep_assert_held(&fs_info->chunk_mutex);
5101         chunk_offset = find_next_chunk(fs_info);
5102         return __btrfs_alloc_chunk(trans, chunk_offset, type);
5103 }
5104
5105 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
5106                                          struct btrfs_fs_info *fs_info)
5107 {
5108         u64 chunk_offset;
5109         u64 sys_chunk_offset;
5110         u64 alloc_profile;
5111         int ret;
5112
5113         chunk_offset = find_next_chunk(fs_info);
5114         alloc_profile = btrfs_metadata_alloc_profile(fs_info);
5115         ret = __btrfs_alloc_chunk(trans, chunk_offset, alloc_profile);
5116         if (ret)
5117                 return ret;
5118
5119         sys_chunk_offset = find_next_chunk(fs_info);
5120         alloc_profile = btrfs_system_alloc_profile(fs_info);
5121         ret = __btrfs_alloc_chunk(trans, sys_chunk_offset, alloc_profile);
5122         return ret;
5123 }
5124
5125 static inline int btrfs_chunk_max_errors(struct map_lookup *map)
5126 {
5127         int max_errors;
5128
5129         if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
5130                          BTRFS_BLOCK_GROUP_RAID10 |
5131                          BTRFS_BLOCK_GROUP_RAID5 |
5132                          BTRFS_BLOCK_GROUP_DUP)) {
5133                 max_errors = 1;
5134         } else if (map->type & BTRFS_BLOCK_GROUP_RAID6) {
5135                 max_errors = 2;
5136         } else {
5137                 max_errors = 0;
5138         }
5139
5140         return max_errors;
5141 }
5142
5143 int btrfs_chunk_readonly(struct btrfs_fs_info *fs_info, u64 chunk_offset)
5144 {
5145         struct extent_map *em;
5146         struct map_lookup *map;
5147         int readonly = 0;
5148         int miss_ndevs = 0;
5149         int i;
5150
5151         em = get_chunk_map(fs_info, chunk_offset, 1);
5152         if (IS_ERR(em))
5153                 return 1;
5154
5155         map = em->map_lookup;
5156         for (i = 0; i < map->num_stripes; i++) {
5157                 if (test_bit(BTRFS_DEV_STATE_MISSING,
5158                                         &map->stripes[i].dev->dev_state)) {
5159                         miss_ndevs++;
5160                         continue;
5161                 }
5162                 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE,
5163                                         &map->stripes[i].dev->dev_state)) {
5164                         readonly = 1;
5165                         goto end;
5166                 }
5167         }
5168
5169         /*
5170          * If the number of missing devices is larger than max errors,
5171          * we can not write the data into that chunk successfully, so
5172          * set it readonly.
5173          */
5174         if (miss_ndevs > btrfs_chunk_max_errors(map))
5175                 readonly = 1;
5176 end:
5177         free_extent_map(em);
5178         return readonly;
5179 }
5180
5181 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
5182 {
5183         extent_map_tree_init(&tree->map_tree);
5184 }
5185
5186 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
5187 {
5188         struct extent_map *em;
5189
5190         while (1) {
5191                 write_lock(&tree->map_tree.lock);
5192                 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
5193                 if (em)
5194                         remove_extent_mapping(&tree->map_tree, em);
5195                 write_unlock(&tree->map_tree.lock);
5196                 if (!em)
5197                         break;
5198                 /* once for us */
5199                 free_extent_map(em);
5200                 /* once for the tree */
5201                 free_extent_map(em);
5202         }
5203 }
5204
5205 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5206 {
5207         struct extent_map *em;
5208         struct map_lookup *map;
5209         int ret;
5210
5211         em = get_chunk_map(fs_info, logical, len);
5212         if (IS_ERR(em))
5213                 /*
5214                  * We could return errors for these cases, but that could get
5215                  * ugly and we'd probably do the same thing which is just not do
5216                  * anything else and exit, so return 1 so the callers don't try
5217                  * to use other copies.
5218                  */
5219                 return 1;
5220
5221         map = em->map_lookup;
5222         if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
5223                 ret = map->num_stripes;
5224         else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5225                 ret = map->sub_stripes;
5226         else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
5227                 ret = 2;
5228         else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5229                 /*
5230                  * There could be two corrupted data stripes, we need
5231                  * to loop retry in order to rebuild the correct data.
5232                  *
5233                  * Fail a stripe at a time on every retry except the
5234                  * stripe under reconstruction.
5235                  */
5236                 ret = map->num_stripes;
5237         else
5238                 ret = 1;
5239         free_extent_map(em);
5240
5241         btrfs_dev_replace_read_lock(&fs_info->dev_replace);
5242         if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace) &&
5243             fs_info->dev_replace.tgtdev)
5244                 ret++;
5245         btrfs_dev_replace_read_unlock(&fs_info->dev_replace);
5246
5247         return ret;
5248 }
5249
5250 unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info,
5251                                     u64 logical)
5252 {
5253         struct extent_map *em;
5254         struct map_lookup *map;
5255         unsigned long len = fs_info->sectorsize;
5256
5257         em = get_chunk_map(fs_info, logical, len);
5258
5259         if (!WARN_ON(IS_ERR(em))) {
5260                 map = em->map_lookup;
5261                 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5262                         len = map->stripe_len * nr_data_stripes(map);
5263                 free_extent_map(em);
5264         }
5265         return len;
5266 }
5267
5268 int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5269 {
5270         struct extent_map *em;
5271         struct map_lookup *map;
5272         int ret = 0;
5273
5274         em = get_chunk_map(fs_info, logical, len);
5275
5276         if(!WARN_ON(IS_ERR(em))) {
5277                 map = em->map_lookup;
5278                 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5279                         ret = 1;
5280                 free_extent_map(em);
5281         }
5282         return ret;
5283 }
5284
5285 static int find_live_mirror(struct btrfs_fs_info *fs_info,
5286                             struct map_lookup *map, int first,
5287                             int dev_replace_is_ongoing)
5288 {
5289         int i;
5290         int num_stripes;
5291         int preferred_mirror;
5292         int tolerance;
5293         struct btrfs_device *srcdev;
5294
5295         ASSERT((map->type &
5296                  (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)));
5297
5298         if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5299                 num_stripes = map->sub_stripes;
5300         else
5301                 num_stripes = map->num_stripes;
5302
5303         preferred_mirror = first + current->pid % num_stripes;
5304
5305         if (dev_replace_is_ongoing &&
5306             fs_info->dev_replace.cont_reading_from_srcdev_mode ==
5307              BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
5308                 srcdev = fs_info->dev_replace.srcdev;
5309         else
5310                 srcdev = NULL;
5311
5312         /*
5313          * try to avoid the drive that is the source drive for a
5314          * dev-replace procedure, only choose it if no other non-missing
5315          * mirror is available
5316          */
5317         for (tolerance = 0; tolerance < 2; tolerance++) {
5318                 if (map->stripes[preferred_mirror].dev->bdev &&
5319                     (tolerance || map->stripes[preferred_mirror].dev != srcdev))
5320                         return preferred_mirror;
5321                 for (i = first; i < first + num_stripes; i++) {
5322                         if (map->stripes[i].dev->bdev &&
5323                             (tolerance || map->stripes[i].dev != srcdev))
5324                                 return i;
5325                 }
5326         }
5327
5328         /* we couldn't find one that doesn't fail.  Just return something
5329          * and the io error handling code will clean up eventually
5330          */
5331         return preferred_mirror;
5332 }
5333
5334 static inline int parity_smaller(u64 a, u64 b)
5335 {
5336         return a > b;
5337 }
5338
5339 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
5340 static void sort_parity_stripes(struct btrfs_bio *bbio, int num_stripes)
5341 {
5342         struct btrfs_bio_stripe s;
5343         int i;
5344         u64 l;
5345         int again = 1;
5346
5347         while (again) {
5348                 again = 0;
5349                 for (i = 0; i < num_stripes - 1; i++) {
5350                         if (parity_smaller(bbio->raid_map[i],
5351                                            bbio->raid_map[i+1])) {
5352                                 s = bbio->stripes[i];
5353                                 l = bbio->raid_map[i];
5354                                 bbio->stripes[i] = bbio->stripes[i+1];
5355                                 bbio->raid_map[i] = bbio->raid_map[i+1];
5356                                 bbio->stripes[i+1] = s;
5357                                 bbio->raid_map[i+1] = l;
5358
5359                                 again = 1;
5360                         }
5361                 }
5362         }
5363 }
5364
5365 static struct btrfs_bio *alloc_btrfs_bio(int total_stripes, int real_stripes)
5366 {
5367         struct btrfs_bio *bbio = kzalloc(
5368                  /* the size of the btrfs_bio */
5369                 sizeof(struct btrfs_bio) +
5370                 /* plus the variable array for the stripes */
5371                 sizeof(struct btrfs_bio_stripe) * (total_stripes) +
5372                 /* plus the variable array for the tgt dev */
5373                 sizeof(int) * (real_stripes) +
5374                 /*
5375                  * plus the raid_map, which includes both the tgt dev
5376                  * and the stripes
5377                  */
5378                 sizeof(u64) * (total_stripes),
5379                 GFP_NOFS|__GFP_NOFAIL);
5380
5381         atomic_set(&bbio->error, 0);
5382         refcount_set(&bbio->refs, 1);
5383
5384         return bbio;
5385 }
5386
5387 void btrfs_get_bbio(struct btrfs_bio *bbio)
5388 {
5389         WARN_ON(!refcount_read(&bbio->refs));
5390         refcount_inc(&bbio->refs);
5391 }
5392
5393 void btrfs_put_bbio(struct btrfs_bio *bbio)
5394 {
5395         if (!bbio)
5396                 return;
5397         if (refcount_dec_and_test(&bbio->refs))
5398                 kfree(bbio);
5399 }
5400
5401 /* can REQ_OP_DISCARD be sent with other REQ like REQ_OP_WRITE? */
5402 /*
5403  * Please note that, discard won't be sent to target device of device
5404  * replace.
5405  */
5406 static int __btrfs_map_block_for_discard(struct btrfs_fs_info *fs_info,
5407                                          u64 logical, u64 length,
5408                                          struct btrfs_bio **bbio_ret)
5409 {
5410         struct extent_map *em;
5411         struct map_lookup *map;
5412         struct btrfs_bio *bbio;
5413         u64 offset;
5414         u64 stripe_nr;
5415         u64 stripe_nr_end;
5416         u64 stripe_end_offset;
5417         u64 stripe_cnt;
5418         u64 stripe_len;
5419         u64 stripe_offset;
5420         u64 num_stripes;
5421         u32 stripe_index;
5422         u32 factor = 0;
5423         u32 sub_stripes = 0;
5424         u64 stripes_per_dev = 0;
5425         u32 remaining_stripes = 0;
5426         u32 last_stripe = 0;
5427         int ret = 0;
5428         int i;
5429
5430         /* discard always return a bbio */
5431         ASSERT(bbio_ret);
5432
5433         em = get_chunk_map(fs_info, logical, length);
5434         if (IS_ERR(em))
5435                 return PTR_ERR(em);
5436
5437         map = em->map_lookup;
5438         /* we don't discard raid56 yet */
5439         if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5440                 ret = -EOPNOTSUPP;
5441                 goto out;
5442         }
5443
5444         offset = logical - em->start;
5445         length = min_t(u64, em->len - offset, length);
5446
5447         stripe_len = map->stripe_len;
5448         /*
5449          * stripe_nr counts the total number of stripes we have to stride
5450          * to get to this block
5451          */
5452         stripe_nr = div64_u64(offset, stripe_len);
5453
5454         /* stripe_offset is the offset of this block in its stripe */
5455         stripe_offset = offset - stripe_nr * stripe_len;
5456
5457         stripe_nr_end = round_up(offset + length, map->stripe_len);
5458         stripe_nr_end = div64_u64(stripe_nr_end, map->stripe_len);
5459         stripe_cnt = stripe_nr_end - stripe_nr;
5460         stripe_end_offset = stripe_nr_end * map->stripe_len -
5461                             (offset + length);
5462         /*
5463          * after this, stripe_nr is the number of stripes on this
5464          * device we have to walk to find the data, and stripe_index is
5465          * the number of our device in the stripe array
5466          */
5467         num_stripes = 1;
5468         stripe_index = 0;
5469         if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5470                          BTRFS_BLOCK_GROUP_RAID10)) {
5471                 if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5472                         sub_stripes = 1;
5473                 else
5474                         sub_stripes = map->sub_stripes;
5475
5476                 factor = map->num_stripes / sub_stripes;
5477                 num_stripes = min_t(u64, map->num_stripes,
5478                                     sub_stripes * stripe_cnt);
5479                 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
5480                 stripe_index *= sub_stripes;
5481                 stripes_per_dev = div_u64_rem(stripe_cnt, factor,
5482                                               &remaining_stripes);
5483                 div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
5484                 last_stripe *= sub_stripes;
5485         } else if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
5486                                 BTRFS_BLOCK_GROUP_DUP)) {
5487                 num_stripes = map->num_stripes;
5488         } else {
5489                 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5490                                         &stripe_index);
5491         }
5492
5493         bbio = alloc_btrfs_bio(num_stripes, 0);
5494         if (!bbio) {
5495                 ret = -ENOMEM;
5496                 goto out;
5497         }
5498
5499         for (i = 0; i < num_stripes; i++) {
5500                 bbio->stripes[i].physical =
5501                         map->stripes[stripe_index].physical +
5502                         stripe_offset + stripe_nr * map->stripe_len;
5503                 bbio->stripes[i].dev = map->stripes[stripe_index].dev;
5504
5505                 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5506                                  BTRFS_BLOCK_GROUP_RAID10)) {
5507                         bbio->stripes[i].length = stripes_per_dev *
5508                                 map->stripe_len;
5509
5510                         if (i / sub_stripes < remaining_stripes)
5511                                 bbio->stripes[i].length +=
5512                                         map->stripe_len;
5513
5514                         /*
5515                          * Special for the first stripe and
5516                          * the last stripe:
5517                          *
5518                          * |-------|...|-------|
5519                          *     |----------|
5520                          *    off     end_off
5521                          */
5522                         if (i < sub_stripes)
5523                                 bbio->stripes[i].length -=
5524                                         stripe_offset;
5525
5526                         if (stripe_index >= last_stripe &&
5527                             stripe_index <= (last_stripe +
5528                                              sub_stripes - 1))
5529                                 bbio->stripes[i].length -=
5530                                         stripe_end_offset;
5531
5532                         if (i == sub_stripes - 1)
5533                                 stripe_offset = 0;
5534                 } else {
5535                         bbio->stripes[i].length = length;
5536                 }
5537
5538                 stripe_index++;
5539                 if (stripe_index == map->num_stripes) {
5540                         stripe_index = 0;
5541                         stripe_nr++;
5542                 }
5543         }
5544
5545         *bbio_ret = bbio;
5546         bbio->map_type = map->type;
5547         bbio->num_stripes = num_stripes;
5548 out:
5549         free_extent_map(em);
5550         return ret;
5551 }
5552
5553 /*
5554  * In dev-replace case, for repair case (that's the only case where the mirror
5555  * is selected explicitly when calling btrfs_map_block), blocks left of the
5556  * left cursor can also be read from the target drive.
5557  *
5558  * For REQ_GET_READ_MIRRORS, the target drive is added as the last one to the
5559  * array of stripes.
5560  * For READ, it also needs to be supported using the same mirror number.
5561  *
5562  * If the requested block is not left of the left cursor, EIO is returned. This
5563  * can happen because btrfs_num_copies() returns one more in the dev-replace
5564  * case.
5565  */
5566 static int get_extra_mirror_from_replace(struct btrfs_fs_info *fs_info,
5567                                          u64 logical, u64 length,
5568                                          u64 srcdev_devid, int *mirror_num,
5569                                          u64 *physical)
5570 {
5571         struct btrfs_bio *bbio = NULL;
5572         int num_stripes;
5573         int index_srcdev = 0;
5574         int found = 0;
5575         u64 physical_of_found = 0;
5576         int i;
5577         int ret = 0;
5578
5579         ret = __btrfs_map_block(fs_info, BTRFS_MAP_GET_READ_MIRRORS,
5580                                 logical, &length, &bbio, 0, 0);
5581         if (ret) {
5582                 ASSERT(bbio == NULL);
5583                 return ret;
5584         }
5585
5586         num_stripes = bbio->num_stripes;
5587         if (*mirror_num > num_stripes) {
5588                 /*
5589                  * BTRFS_MAP_GET_READ_MIRRORS does not contain this mirror,
5590                  * that means that the requested area is not left of the left
5591                  * cursor
5592                  */
5593                 btrfs_put_bbio(bbio);
5594                 return -EIO;
5595         }
5596
5597         /*
5598          * process the rest of the function using the mirror_num of the source
5599          * drive. Therefore look it up first.  At the end, patch the device
5600          * pointer to the one of the target drive.
5601          */
5602         for (i = 0; i < num_stripes; i++) {
5603                 if (bbio->stripes[i].dev->devid != srcdev_devid)
5604                         continue;
5605
5606                 /*
5607                  * In case of DUP, in order to keep it simple, only add the
5608                  * mirror with the lowest physical address
5609                  */
5610                 if (found &&
5611                     physical_of_found <= bbio->stripes[i].physical)
5612                         continue;
5613
5614                 index_srcdev = i;
5615                 found = 1;
5616                 physical_of_found = bbio->stripes[i].physical;
5617         }
5618
5619         btrfs_put_bbio(bbio);
5620
5621         ASSERT(found);
5622         if (!found)
5623                 return -EIO;
5624
5625         *mirror_num = index_srcdev + 1;
5626         *physical = physical_of_found;
5627         return ret;
5628 }
5629
5630 static void handle_ops_on_dev_replace(enum btrfs_map_op op,
5631                                       struct btrfs_bio **bbio_ret,
5632                                       struct btrfs_dev_replace *dev_replace,
5633                                       int *num_stripes_ret, int *max_errors_ret)
5634 {
5635         struct btrfs_bio *bbio = *bbio_ret;
5636         u64 srcdev_devid = dev_replace->srcdev->devid;
5637         int tgtdev_indexes = 0;
5638         int num_stripes = *num_stripes_ret;
5639         int max_errors = *max_errors_ret;
5640         int i;
5641
5642         if (op == BTRFS_MAP_WRITE) {
5643                 int index_where_to_add;
5644
5645                 /*
5646                  * duplicate the write operations while the dev replace
5647                  * procedure is running. Since the copying of the old disk to
5648                  * the new disk takes place at run time while the filesystem is
5649                  * mounted writable, the regular write operations to the old
5650                  * disk have to be duplicated to go to the new disk as well.
5651                  *
5652                  * Note that device->missing is handled by the caller, and that
5653                  * the write to the old disk is already set up in the stripes
5654                  * array.
5655                  */
5656                 index_where_to_add = num_stripes;
5657                 for (i = 0; i < num_stripes; i++) {
5658                         if (bbio->stripes[i].dev->devid == srcdev_devid) {
5659                                 /* write to new disk, too */
5660                                 struct btrfs_bio_stripe *new =
5661                                         bbio->stripes + index_where_to_add;
5662                                 struct btrfs_bio_stripe *old =
5663                                         bbio->stripes + i;
5664
5665                                 new->physical = old->physical;
5666                                 new->length = old->length;
5667                                 new->dev = dev_replace->tgtdev;
5668                                 bbio->tgtdev_map[i] = index_where_to_add;
5669                                 index_where_to_add++;
5670                                 max_errors++;
5671                                 tgtdev_indexes++;
5672                         }
5673                 }
5674                 num_stripes = index_where_to_add;
5675         } else if (op == BTRFS_MAP_GET_READ_MIRRORS) {
5676                 int index_srcdev = 0;
5677                 int found = 0;
5678                 u64 physical_of_found = 0;
5679
5680                 /*
5681                  * During the dev-replace procedure, the target drive can also
5682                  * be used to read data in case it is needed to repair a corrupt
5683                  * block elsewhere. This is possible if the requested area is
5684                  * left of the left cursor. In this area, the target drive is a
5685                  * full copy of the source drive.
5686                  */
5687                 for (i = 0; i < num_stripes; i++) {
5688                         if (bbio->stripes[i].dev->devid == srcdev_devid) {
5689                                 /*
5690                                  * In case of DUP, in order to keep it simple,
5691                                  * only add the mirror with the lowest physical
5692                                  * address
5693                                  */
5694                                 if (found &&
5695                                     physical_of_found <=
5696                                      bbio->stripes[i].physical)
5697                                         continue;
5698                                 index_srcdev = i;
5699                                 found = 1;
5700                                 physical_of_found = bbio->stripes[i].physical;
5701                         }
5702                 }
5703                 if (found) {
5704                         struct btrfs_bio_stripe *tgtdev_stripe =
5705                                 bbio->stripes + num_stripes;
5706
5707                         tgtdev_stripe->physical = physical_of_found;
5708                         tgtdev_stripe->length =
5709                                 bbio->stripes[index_srcdev].length;
5710                         tgtdev_stripe->dev = dev_replace->tgtdev;
5711                         bbio->tgtdev_map[index_srcdev] = num_stripes;
5712
5713                         tgtdev_indexes++;
5714                         num_stripes++;
5715                 }
5716         }
5717
5718         *num_stripes_ret = num_stripes;
5719         *max_errors_ret = max_errors;
5720         bbio->num_tgtdevs = tgtdev_indexes;
5721         *bbio_ret = bbio;
5722 }
5723
5724 static bool need_full_stripe(enum btrfs_map_op op)
5725 {
5726         return (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS);
5727 }
5728
5729 static int __btrfs_map_block(struct btrfs_fs_info *fs_info,
5730                              enum btrfs_map_op op,
5731                              u64 logical, u64 *length,
5732                              struct btrfs_bio **bbio_ret,
5733                              int mirror_num, int need_raid_map)
5734 {
5735         struct extent_map *em;
5736         struct map_lookup *map;
5737         u64 offset;
5738         u64 stripe_offset;
5739         u64 stripe_nr;
5740         u64 stripe_len;
5741         u32 stripe_index;
5742         int i;
5743         int ret = 0;
5744         int num_stripes;
5745         int max_errors = 0;
5746         int tgtdev_indexes = 0;
5747         struct btrfs_bio *bbio = NULL;
5748         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
5749         int dev_replace_is_ongoing = 0;
5750         int num_alloc_stripes;
5751         int patch_the_first_stripe_for_dev_replace = 0;
5752         u64 physical_to_patch_in_first_stripe = 0;
5753         u64 raid56_full_stripe_start = (u64)-1;
5754
5755         if (op == BTRFS_MAP_DISCARD)
5756                 return __btrfs_map_block_for_discard(fs_info, logical,
5757                                                      *length, bbio_ret);
5758
5759         em = get_chunk_map(fs_info, logical, *length);
5760         if (IS_ERR(em))
5761                 return PTR_ERR(em);
5762
5763         map = em->map_lookup;
5764         offset = logical - em->start;
5765
5766         stripe_len = map->stripe_len;
5767         stripe_nr = offset;
5768         /*
5769          * stripe_nr counts the total number of stripes we have to stride
5770          * to get to this block
5771          */
5772         stripe_nr = div64_u64(stripe_nr, stripe_len);
5773
5774         stripe_offset = stripe_nr * stripe_len;
5775         if (offset < stripe_offset) {
5776                 btrfs_crit(fs_info,
5777                            "stripe math has gone wrong, stripe_offset=%llu, offset=%llu, start=%llu, logical=%llu, stripe_len=%llu",
5778                            stripe_offset, offset, em->start, logical,
5779                            stripe_len);
5780                 free_extent_map(em);
5781                 return -EINVAL;
5782         }
5783
5784         /* stripe_offset is the offset of this block in its stripe*/
5785         stripe_offset = offset - stripe_offset;
5786
5787         /* if we're here for raid56, we need to know the stripe aligned start */
5788         if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5789                 unsigned long full_stripe_len = stripe_len * nr_data_stripes(map);
5790                 raid56_full_stripe_start = offset;
5791
5792                 /* allow a write of a full stripe, but make sure we don't
5793                  * allow straddling of stripes
5794                  */
5795                 raid56_full_stripe_start = div64_u64(raid56_full_stripe_start,
5796                                 full_stripe_len);
5797                 raid56_full_stripe_start *= full_stripe_len;
5798         }
5799
5800         if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
5801                 u64 max_len;
5802                 /* For writes to RAID[56], allow a full stripeset across all disks.
5803                    For other RAID types and for RAID[56] reads, just allow a single
5804                    stripe (on a single disk). */
5805                 if ((map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
5806                     (op == BTRFS_MAP_WRITE)) {
5807                         max_len = stripe_len * nr_data_stripes(map) -
5808                                 (offset - raid56_full_stripe_start);
5809                 } else {
5810                         /* we limit the length of each bio to what fits in a stripe */
5811                         max_len = stripe_len - stripe_offset;
5812                 }
5813                 *length = min_t(u64, em->len - offset, max_len);
5814         } else {
5815                 *length = em->len - offset;
5816         }
5817
5818         /* This is for when we're called from btrfs_merge_bio_hook() and all
5819            it cares about is the length */
5820         if (!bbio_ret)
5821                 goto out;
5822
5823         btrfs_dev_replace_read_lock(dev_replace);
5824         dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
5825         if (!dev_replace_is_ongoing)
5826                 btrfs_dev_replace_read_unlock(dev_replace);
5827         else
5828                 btrfs_dev_replace_set_lock_blocking(dev_replace);
5829
5830         if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
5831             !need_full_stripe(op) && dev_replace->tgtdev != NULL) {
5832                 ret = get_extra_mirror_from_replace(fs_info, logical, *length,
5833                                                     dev_replace->srcdev->devid,
5834                                                     &mirror_num,
5835                                             &physical_to_patch_in_first_stripe);
5836                 if (ret)
5837                         goto out;
5838                 else
5839                         patch_the_first_stripe_for_dev_replace = 1;
5840         } else if (mirror_num > map->num_stripes) {
5841                 mirror_num = 0;
5842         }
5843
5844         num_stripes = 1;
5845         stripe_index = 0;
5846         if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5847                 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5848                                 &stripe_index);
5849                 if (!need_full_stripe(op))
5850                         mirror_num = 1;
5851         } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
5852                 if (need_full_stripe(op))
5853                         num_stripes = map->num_stripes;
5854                 else if (mirror_num)
5855                         stripe_index = mirror_num - 1;
5856                 else {
5857                         stripe_index = find_live_mirror(fs_info, map, 0,
5858                                             dev_replace_is_ongoing);
5859                         mirror_num = stripe_index + 1;
5860                 }
5861
5862         } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
5863                 if (need_full_stripe(op)) {
5864                         num_stripes = map->num_stripes;
5865                 } else if (mirror_num) {
5866                         stripe_index = mirror_num - 1;
5867                 } else {
5868                         mirror_num = 1;
5869                 }
5870
5871         } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5872                 u32 factor = map->num_stripes / map->sub_stripes;
5873
5874                 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
5875                 stripe_index *= map->sub_stripes;
5876
5877                 if (need_full_stripe(op))
5878                         num_stripes = map->sub_stripes;
5879                 else if (mirror_num)
5880                         stripe_index += mirror_num - 1;
5881                 else {
5882                         int old_stripe_index = stripe_index;
5883                         stripe_index = find_live_mirror(fs_info, map,
5884                                               stripe_index,
5885                                               dev_replace_is_ongoing);
5886                         mirror_num = stripe_index - old_stripe_index + 1;
5887                 }
5888
5889         } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5890                 if (need_raid_map && (need_full_stripe(op) || mirror_num > 1)) {
5891                         /* push stripe_nr back to the start of the full stripe */
5892                         stripe_nr = div64_u64(raid56_full_stripe_start,
5893                                         stripe_len * nr_data_stripes(map));
5894
5895                         /* RAID[56] write or recovery. Return all stripes */
5896                         num_stripes = map->num_stripes;
5897                         max_errors = nr_parity_stripes(map);
5898
5899                         *length = map->stripe_len;
5900                         stripe_index = 0;
5901                         stripe_offset = 0;
5902                 } else {
5903                         /*
5904                          * Mirror #0 or #1 means the original data block.
5905                          * Mirror #2 is RAID5 parity block.
5906                          * Mirror #3 is RAID6 Q block.
5907                          */
5908                         stripe_nr = div_u64_rem(stripe_nr,
5909                                         nr_data_stripes(map), &stripe_index);
5910                         if (mirror_num > 1)
5911                                 stripe_index = nr_data_stripes(map) +
5912                                                 mirror_num - 2;
5913
5914                         /* We distribute the parity blocks across stripes */
5915                         div_u64_rem(stripe_nr + stripe_index, map->num_stripes,
5916                                         &stripe_index);
5917                         if (!need_full_stripe(op) && mirror_num <= 1)
5918                                 mirror_num = 1;
5919                 }
5920         } else {
5921                 /*
5922                  * after this, stripe_nr is the number of stripes on this
5923                  * device we have to walk to find the data, and stripe_index is
5924                  * the number of our device in the stripe array
5925                  */
5926                 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5927                                 &stripe_index);
5928                 mirror_num = stripe_index + 1;
5929         }
5930         if (stripe_index >= map->num_stripes) {
5931                 btrfs_crit(fs_info,
5932                            "stripe index math went horribly wrong, got stripe_index=%u, num_stripes=%u",
5933                            stripe_index, map->num_stripes);
5934                 ret = -EINVAL;
5935                 goto out;
5936         }
5937
5938         num_alloc_stripes = num_stripes;
5939         if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL) {
5940                 if (op == BTRFS_MAP_WRITE)
5941                         num_alloc_stripes <<= 1;
5942                 if (op == BTRFS_MAP_GET_READ_MIRRORS)
5943                         num_alloc_stripes++;
5944                 tgtdev_indexes = num_stripes;
5945         }
5946
5947         bbio = alloc_btrfs_bio(num_alloc_stripes, tgtdev_indexes);
5948         if (!bbio) {
5949                 ret = -ENOMEM;
5950                 goto out;
5951         }
5952         if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL)
5953                 bbio->tgtdev_map = (int *)(bbio->stripes + num_alloc_stripes);
5954
5955         /* build raid_map */
5956         if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK && need_raid_map &&
5957             (need_full_stripe(op) || mirror_num > 1)) {
5958                 u64 tmp;
5959                 unsigned rot;
5960
5961                 bbio->raid_map = (u64 *)((void *)bbio->stripes +
5962                                  sizeof(struct btrfs_bio_stripe) *
5963                                  num_alloc_stripes +
5964                                  sizeof(int) * tgtdev_indexes);
5965
5966                 /* Work out the disk rotation on this stripe-set */
5967                 div_u64_rem(stripe_nr, num_stripes, &rot);
5968
5969                 /* Fill in the logical address of each stripe */
5970                 tmp = stripe_nr * nr_data_stripes(map);
5971                 for (i = 0; i < nr_data_stripes(map); i++)
5972                         bbio->raid_map[(i+rot) % num_stripes] =
5973                                 em->start + (tmp + i) * map->stripe_len;
5974
5975                 bbio->raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE;
5976                 if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5977                         bbio->raid_map[(i+rot+1) % num_stripes] =
5978                                 RAID6_Q_STRIPE;
5979         }
5980
5981
5982         for (i = 0; i < num_stripes; i++) {
5983                 bbio->stripes[i].physical =
5984                         map->stripes[stripe_index].physical +
5985                         stripe_offset +
5986                         stripe_nr * map->stripe_len;
5987                 bbio->stripes[i].dev =
5988                         map->stripes[stripe_index].dev;
5989                 stripe_index++;
5990         }
5991
5992         if (need_full_stripe(op))
5993                 max_errors = btrfs_chunk_max_errors(map);
5994
5995         if (bbio->raid_map)
5996                 sort_parity_stripes(bbio, num_stripes);
5997
5998         if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL &&
5999             need_full_stripe(op)) {
6000                 handle_ops_on_dev_replace(op, &bbio, dev_replace, &num_stripes,
6001                                           &max_errors);
6002         }
6003
6004         *bbio_ret = bbio;
6005         bbio->map_type = map->type;
6006         bbio->num_stripes = num_stripes;
6007         bbio->max_errors = max_errors;
6008         bbio->mirror_num = mirror_num;
6009
6010         /*
6011          * this is the case that REQ_READ && dev_replace_is_ongoing &&
6012          * mirror_num == num_stripes + 1 && dev_replace target drive is
6013          * available as a mirror
6014          */
6015         if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) {
6016                 WARN_ON(num_stripes > 1);
6017                 bbio->stripes[0].dev = dev_replace->tgtdev;
6018                 bbio->stripes[0].physical = physical_to_patch_in_first_stripe;
6019                 bbio->mirror_num = map->num_stripes + 1;
6020         }
6021 out:
6022         if (dev_replace_is_ongoing) {
6023                 btrfs_dev_replace_clear_lock_blocking(dev_replace);
6024                 btrfs_dev_replace_read_unlock(dev_replace);
6025         }
6026         free_extent_map(em);
6027         return ret;
6028 }
6029
6030 int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
6031                       u64 logical, u64 *length,
6032                       struct btrfs_bio **bbio_ret, int mirror_num)
6033 {
6034         return __btrfs_map_block(fs_info, op, logical, length, bbio_ret,
6035                                  mirror_num, 0);
6036 }
6037
6038 /* For Scrub/replace */
6039 int btrfs_map_sblock(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
6040                      u64 logical, u64 *length,
6041                      struct btrfs_bio **bbio_ret)
6042 {
6043         return __btrfs_map_block(fs_info, op, logical, length, bbio_ret, 0, 1);
6044 }
6045
6046 int btrfs_rmap_block(struct btrfs_fs_info *fs_info,
6047                      u64 chunk_start, u64 physical, u64 devid,
6048                      u64 **logical, int *naddrs, int *stripe_len)
6049 {
6050         struct extent_map *em;
6051         struct map_lookup *map;
6052         u64 *buf;
6053         u64 bytenr;
6054         u64 length;
6055         u64 stripe_nr;
6056         u64 rmap_len;
6057         int i, j, nr = 0;
6058
6059         em = get_chunk_map(fs_info, chunk_start, 1);
6060         if (IS_ERR(em))
6061                 return -EIO;
6062
6063         map = em->map_lookup;
6064         length = em->len;
6065         rmap_len = map->stripe_len;
6066
6067         if (map->type & BTRFS_BLOCK_GROUP_RAID10)
6068                 length = div_u64(length, map->num_stripes / map->sub_stripes);
6069         else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
6070                 length = div_u64(length, map->num_stripes);
6071         else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
6072                 length = div_u64(length, nr_data_stripes(map));
6073                 rmap_len = map->stripe_len * nr_data_stripes(map);
6074         }
6075
6076         buf = kcalloc(map->num_stripes, sizeof(u64), GFP_NOFS);
6077         BUG_ON(!buf); /* -ENOMEM */
6078
6079         for (i = 0; i < map->num_stripes; i++) {
6080                 if (devid && map->stripes[i].dev->devid != devid)
6081                         continue;
6082                 if (map->stripes[i].physical > physical ||
6083                     map->stripes[i].physical + length <= physical)
6084                         continue;
6085
6086                 stripe_nr = physical - map->stripes[i].physical;
6087                 stripe_nr = div64_u64(stripe_nr, map->stripe_len);
6088
6089                 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
6090                         stripe_nr = stripe_nr * map->num_stripes + i;
6091                         stripe_nr = div_u64(stripe_nr, map->sub_stripes);
6092                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
6093                         stripe_nr = stripe_nr * map->num_stripes + i;
6094                 } /* else if RAID[56], multiply by nr_data_stripes().
6095                    * Alternatively, just use rmap_len below instead of
6096                    * map->stripe_len */
6097
6098                 bytenr = chunk_start + stripe_nr * rmap_len;
6099                 WARN_ON(nr >= map->num_stripes);
6100                 for (j = 0; j < nr; j++) {
6101                         if (buf[j] == bytenr)
6102                                 break;
6103                 }
6104                 if (j == nr) {
6105                         WARN_ON(nr >= map->num_stripes);
6106                         buf[nr++] = bytenr;
6107                 }
6108         }
6109
6110         *logical = buf;
6111         *naddrs = nr;
6112         *stripe_len = rmap_len;
6113
6114         free_extent_map(em);
6115         return 0;
6116 }
6117
6118 static inline void btrfs_end_bbio(struct btrfs_bio *bbio, struct bio *bio)
6119 {
6120         bio->bi_private = bbio->private;
6121         bio->bi_end_io = bbio->end_io;
6122         bio_endio(bio);
6123
6124         btrfs_put_bbio(bbio);
6125 }
6126
6127 static void btrfs_end_bio(struct bio *bio)
6128 {
6129         struct btrfs_bio *bbio = bio->bi_private;
6130         int is_orig_bio = 0;
6131
6132         if (bio->bi_status) {
6133                 atomic_inc(&bbio->error);
6134                 if (bio->bi_status == BLK_STS_IOERR ||
6135                     bio->bi_status == BLK_STS_TARGET) {
6136                         unsigned int stripe_index =
6137                                 btrfs_io_bio(bio)->stripe_index;
6138                         struct btrfs_device *dev;
6139
6140                         BUG_ON(stripe_index >= bbio->num_stripes);
6141                         dev = bbio->stripes[stripe_index].dev;
6142                         if (dev->bdev) {
6143                                 if (bio_op(bio) == REQ_OP_WRITE)
6144                                         btrfs_dev_stat_inc_and_print(dev,
6145                                                 BTRFS_DEV_STAT_WRITE_ERRS);
6146                                 else
6147                                         btrfs_dev_stat_inc_and_print(dev,
6148                                                 BTRFS_DEV_STAT_READ_ERRS);
6149                                 if (bio->bi_opf & REQ_PREFLUSH)
6150                                         btrfs_dev_stat_inc_and_print(dev,
6151                                                 BTRFS_DEV_STAT_FLUSH_ERRS);
6152                         }
6153                 }
6154         }
6155
6156         if (bio == bbio->orig_bio)
6157                 is_orig_bio = 1;
6158
6159         btrfs_bio_counter_dec(bbio->fs_info);
6160
6161         if (atomic_dec_and_test(&bbio->stripes_pending)) {
6162                 if (!is_orig_bio) {
6163                         bio_put(bio);
6164                         bio = bbio->orig_bio;
6165                 }
6166
6167                 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
6168                 /* only send an error to the higher layers if it is
6169                  * beyond the tolerance of the btrfs bio
6170                  */
6171                 if (atomic_read(&bbio->error) > bbio->max_errors) {
6172                         bio->bi_status = BLK_STS_IOERR;
6173                 } else {
6174                         /*
6175                          * this bio is actually up to date, we didn't
6176                          * go over the max number of errors
6177                          */
6178                         bio->bi_status = BLK_STS_OK;
6179                 }
6180
6181                 btrfs_end_bbio(bbio, bio);
6182         } else if (!is_orig_bio) {
6183                 bio_put(bio);
6184         }
6185 }
6186
6187 /*
6188  * see run_scheduled_bios for a description of why bios are collected for
6189  * async submit.
6190  *
6191  * This will add one bio to the pending list for a device and make sure
6192  * the work struct is scheduled.
6193  */
6194 static noinline void btrfs_schedule_bio(struct btrfs_device *device,
6195                                         struct bio *bio)
6196 {
6197         struct btrfs_fs_info *fs_info = device->fs_info;
6198         int should_queue = 1;
6199         struct btrfs_pending_bios *pending_bios;
6200
6201         if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state) ||
6202             !device->bdev) {
6203                 bio_io_error(bio);
6204                 return;
6205         }
6206
6207         /* don't bother with additional async steps for reads, right now */
6208         if (bio_op(bio) == REQ_OP_READ) {
6209                 btrfsic_submit_bio(bio);
6210                 return;
6211         }
6212
6213         WARN_ON(bio->bi_next);
6214         bio->bi_next = NULL;
6215
6216         spin_lock(&device->io_lock);
6217         if (op_is_sync(bio->bi_opf))
6218                 pending_bios = &device->pending_sync_bios;
6219         else
6220                 pending_bios = &device->pending_bios;
6221
6222         if (pending_bios->tail)
6223                 pending_bios->tail->bi_next = bio;
6224
6225         pending_bios->tail = bio;
6226         if (!pending_bios->head)
6227                 pending_bios->head = bio;
6228         if (device->running_pending)
6229                 should_queue = 0;
6230
6231         spin_unlock(&device->io_lock);
6232
6233         if (should_queue)
6234                 btrfs_queue_work(fs_info->submit_workers, &device->work);
6235 }
6236
6237 static void submit_stripe_bio(struct btrfs_bio *bbio, struct bio *bio,
6238                               u64 physical, int dev_nr, int async)
6239 {
6240         struct btrfs_device *dev = bbio->stripes[dev_nr].dev;
6241         struct btrfs_fs_info *fs_info = bbio->fs_info;
6242
6243         bio->bi_private = bbio;
6244         btrfs_io_bio(bio)->stripe_index = dev_nr;
6245         bio->bi_end_io = btrfs_end_bio;
6246         bio->bi_iter.bi_sector = physical >> 9;
6247 #ifdef DEBUG
6248         {
6249                 struct rcu_string *name;
6250
6251                 rcu_read_lock();
6252                 name = rcu_dereference(dev->name);
6253                 btrfs_debug(fs_info,
6254                         "btrfs_map_bio: rw %d 0x%x, sector=%llu, dev=%lu (%s id %llu), size=%u",
6255                         bio_op(bio), bio->bi_opf,
6256                         (u64)bio->bi_iter.bi_sector,
6257                         (u_long)dev->bdev->bd_dev, name->str, dev->devid,
6258                         bio->bi_iter.bi_size);
6259                 rcu_read_unlock();
6260         }
6261 #endif
6262         bio_set_dev(bio, dev->bdev);
6263
6264         btrfs_bio_counter_inc_noblocked(fs_info);
6265
6266         if (async)
6267                 btrfs_schedule_bio(dev, bio);
6268         else
6269                 btrfsic_submit_bio(bio);
6270 }
6271
6272 static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical)
6273 {
6274         atomic_inc(&bbio->error);
6275         if (atomic_dec_and_test(&bbio->stripes_pending)) {
6276                 /* Should be the original bio. */
6277                 WARN_ON(bio != bbio->orig_bio);
6278
6279                 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
6280                 bio->bi_iter.bi_sector = logical >> 9;
6281                 if (atomic_read(&bbio->error) > bbio->max_errors)
6282                         bio->bi_status = BLK_STS_IOERR;
6283                 else
6284                         bio->bi_status = BLK_STS_OK;
6285                 btrfs_end_bbio(bbio, bio);
6286         }
6287 }
6288
6289 blk_status_t btrfs_map_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
6290                            int mirror_num, int async_submit)
6291 {
6292         struct btrfs_device *dev;
6293         struct bio *first_bio = bio;
6294         u64 logical = (u64)bio->bi_iter.bi_sector << 9;
6295         u64 length = 0;
6296         u64 map_length;
6297         int ret;
6298         int dev_nr;
6299         int total_devs;
6300         struct btrfs_bio *bbio = NULL;
6301
6302         length = bio->bi_iter.bi_size;
6303         map_length = length;
6304
6305         btrfs_bio_counter_inc_blocked(fs_info);
6306         ret = __btrfs_map_block(fs_info, btrfs_op(bio), logical,
6307                                 &map_length, &bbio, mirror_num, 1);
6308         if (ret) {
6309                 btrfs_bio_counter_dec(fs_info);
6310                 return errno_to_blk_status(ret);
6311         }
6312
6313         total_devs = bbio->num_stripes;
6314         bbio->orig_bio = first_bio;
6315         bbio->private = first_bio->bi_private;
6316         bbio->end_io = first_bio->bi_end_io;
6317         bbio->fs_info = fs_info;
6318         atomic_set(&bbio->stripes_pending, bbio->num_stripes);
6319
6320         if ((bbio->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
6321             ((bio_op(bio) == REQ_OP_WRITE) || (mirror_num > 1))) {
6322                 /* In this case, map_length has been set to the length of
6323                    a single stripe; not the whole write */
6324                 if (bio_op(bio) == REQ_OP_WRITE) {
6325                         ret = raid56_parity_write(fs_info, bio, bbio,
6326                                                   map_length);
6327                 } else {
6328                         ret = raid56_parity_recover(fs_info, bio, bbio,
6329                                                     map_length, mirror_num, 1);
6330                 }
6331
6332                 btrfs_bio_counter_dec(fs_info);
6333                 return errno_to_blk_status(ret);
6334         }
6335
6336         if (map_length < length) {
6337                 btrfs_crit(fs_info,
6338                            "mapping failed logical %llu bio len %llu len %llu",
6339                            logical, length, map_length);
6340                 BUG();
6341         }
6342
6343         for (dev_nr = 0; dev_nr < total_devs; dev_nr++) {
6344                 dev = bbio->stripes[dev_nr].dev;
6345                 if (!dev || !dev->bdev ||
6346                     (bio_op(first_bio) == REQ_OP_WRITE &&
6347                     !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))) {
6348                         bbio_error(bbio, first_bio, logical);
6349                         continue;
6350                 }
6351
6352                 if (dev_nr < total_devs - 1)
6353                         bio = btrfs_bio_clone(first_bio);
6354                 else
6355                         bio = first_bio;
6356
6357                 submit_stripe_bio(bbio, bio, bbio->stripes[dev_nr].physical,
6358                                   dev_nr, async_submit);
6359         }
6360         btrfs_bio_counter_dec(fs_info);
6361         return BLK_STS_OK;
6362 }
6363
6364 struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid,
6365                                        u8 *uuid, u8 *fsid)
6366 {
6367         struct btrfs_device *device;
6368         struct btrfs_fs_devices *cur_devices;
6369
6370         cur_devices = fs_info->fs_devices;
6371         while (cur_devices) {
6372                 if (!fsid ||
6373                     !memcmp(cur_devices->fsid, fsid, BTRFS_FSID_SIZE)) {
6374                         device = find_device(cur_devices, devid, uuid);
6375                         if (device)
6376                                 return device;
6377                 }
6378                 cur_devices = cur_devices->seed;
6379         }
6380         return NULL;
6381 }
6382
6383 static struct btrfs_device *add_missing_dev(struct btrfs_fs_devices *fs_devices,
6384                                             u64 devid, u8 *dev_uuid)
6385 {
6386         struct btrfs_device *device;
6387
6388         device = btrfs_alloc_device(NULL, &devid, dev_uuid);
6389         if (IS_ERR(device))
6390                 return device;
6391
6392         list_add(&device->dev_list, &fs_devices->devices);
6393         device->fs_devices = fs_devices;
6394         fs_devices->num_devices++;
6395
6396         set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
6397         fs_devices->missing_devices++;
6398
6399         return device;
6400 }
6401
6402 /**
6403  * btrfs_alloc_device - allocate struct btrfs_device
6404  * @fs_info:    used only for generating a new devid, can be NULL if
6405  *              devid is provided (i.e. @devid != NULL).
6406  * @devid:      a pointer to devid for this device.  If NULL a new devid
6407  *              is generated.
6408  * @uuid:       a pointer to UUID for this device.  If NULL a new UUID
6409  *              is generated.
6410  *
6411  * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
6412  * on error.  Returned struct is not linked onto any lists and must be
6413  * destroyed with free_device.
6414  */
6415 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
6416                                         const u64 *devid,
6417                                         const u8 *uuid)
6418 {
6419         struct btrfs_device *dev;
6420         u64 tmp;
6421
6422         if (WARN_ON(!devid && !fs_info))
6423                 return ERR_PTR(-EINVAL);
6424
6425         dev = __alloc_device();
6426         if (IS_ERR(dev))
6427                 return dev;
6428
6429         if (devid)
6430                 tmp = *devid;
6431         else {
6432                 int ret;
6433
6434                 ret = find_next_devid(fs_info, &tmp);
6435                 if (ret) {
6436                         free_device(dev);
6437                         return ERR_PTR(ret);
6438                 }
6439         }
6440         dev->devid = tmp;
6441
6442         if (uuid)
6443                 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE);
6444         else
6445                 generate_random_uuid(dev->uuid);
6446
6447         btrfs_init_work(&dev->work, btrfs_submit_helper,
6448                         pending_bios_fn, NULL, NULL);
6449
6450         return dev;
6451 }
6452
6453 /* Return -EIO if any error, otherwise return 0. */
6454 static int btrfs_check_chunk_valid(struct btrfs_fs_info *fs_info,
6455                                    struct extent_buffer *leaf,
6456                                    struct btrfs_chunk *chunk, u64 logical)
6457 {
6458         u64 length;
6459         u64 stripe_len;
6460         u16 num_stripes;
6461         u16 sub_stripes;
6462         u64 type;
6463
6464         length = btrfs_chunk_length(leaf, chunk);
6465         stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6466         num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6467         sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
6468         type = btrfs_chunk_type(leaf, chunk);
6469
6470         if (!num_stripes) {
6471                 btrfs_err(fs_info, "invalid chunk num_stripes: %u",
6472                           num_stripes);
6473                 return -EIO;
6474         }
6475         if (!IS_ALIGNED(logical, fs_info->sectorsize)) {
6476                 btrfs_err(fs_info, "invalid chunk logical %llu", logical);
6477                 return -EIO;
6478         }
6479         if (btrfs_chunk_sector_size(leaf, chunk) != fs_info->sectorsize) {
6480                 btrfs_err(fs_info, "invalid chunk sectorsize %u",
6481                           btrfs_chunk_sector_size(leaf, chunk));
6482                 return -EIO;
6483         }
6484         if (!length || !IS_ALIGNED(length, fs_info->sectorsize)) {
6485                 btrfs_err(fs_info, "invalid chunk length %llu", length);
6486                 return -EIO;
6487         }
6488         if (!is_power_of_2(stripe_len) || stripe_len != BTRFS_STRIPE_LEN) {
6489                 btrfs_err(fs_info, "invalid chunk stripe length: %llu",
6490                           stripe_len);
6491                 return -EIO;
6492         }
6493         if (~(BTRFS_BLOCK_GROUP_TYPE_MASK | BTRFS_BLOCK_GROUP_PROFILE_MASK) &
6494             type) {
6495                 btrfs_err(fs_info, "unrecognized chunk type: %llu",
6496                           ~(BTRFS_BLOCK_GROUP_TYPE_MASK |
6497                             BTRFS_BLOCK_GROUP_PROFILE_MASK) &
6498                           btrfs_chunk_type(leaf, chunk));
6499                 return -EIO;
6500         }
6501         if ((type & BTRFS_BLOCK_GROUP_RAID10 && sub_stripes != 2) ||
6502             (type & BTRFS_BLOCK_GROUP_RAID1 && num_stripes < 1) ||
6503             (type & BTRFS_BLOCK_GROUP_RAID5 && num_stripes < 2) ||
6504             (type & BTRFS_BLOCK_GROUP_RAID6 && num_stripes < 3) ||
6505             (type & BTRFS_BLOCK_GROUP_DUP && num_stripes > 2) ||
6506             ((type & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 &&
6507              num_stripes != 1)) {
6508                 btrfs_err(fs_info,
6509                         "invalid num_stripes:sub_stripes %u:%u for profile %llu",
6510                         num_stripes, sub_stripes,
6511                         type & BTRFS_BLOCK_GROUP_PROFILE_MASK);
6512                 return -EIO;
6513         }
6514
6515         return 0;
6516 }
6517
6518 static void btrfs_report_missing_device(struct btrfs_fs_info *fs_info,
6519                                         u64 devid, u8 *uuid, bool error)
6520 {
6521         if (error)
6522                 btrfs_err_rl(fs_info, "devid %llu uuid %pU is missing",
6523                               devid, uuid);
6524         else
6525                 btrfs_warn_rl(fs_info, "devid %llu uuid %pU is missing",
6526                               devid, uuid);
6527 }
6528
6529 static int read_one_chunk(struct btrfs_fs_info *fs_info, struct btrfs_key *key,
6530                           struct extent_buffer *leaf,
6531                           struct btrfs_chunk *chunk)
6532 {
6533         struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
6534         struct map_lookup *map;
6535         struct extent_map *em;
6536         u64 logical;
6537         u64 length;
6538         u64 devid;
6539         u8 uuid[BTRFS_UUID_SIZE];
6540         int num_stripes;
6541         int ret;
6542         int i;
6543
6544         logical = key->offset;
6545         length = btrfs_chunk_length(leaf, chunk);
6546         num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6547
6548         ret = btrfs_check_chunk_valid(fs_info, leaf, chunk, logical);
6549         if (ret)
6550                 return ret;
6551
6552         read_lock(&map_tree->map_tree.lock);
6553         em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
6554         read_unlock(&map_tree->map_tree.lock);
6555
6556         /* already mapped? */
6557         if (em && em->start <= logical && em->start + em->len > logical) {
6558                 free_extent_map(em);
6559                 return 0;
6560         } else if (em) {
6561                 free_extent_map(em);
6562         }
6563
6564         em = alloc_extent_map();
6565         if (!em)
6566                 return -ENOMEM;
6567         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
6568         if (!map) {
6569                 free_extent_map(em);
6570                 return -ENOMEM;
6571         }
6572
6573         set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
6574         em->map_lookup = map;
6575         em->start = logical;
6576         em->len = length;
6577         em->orig_start = 0;
6578         em->block_start = 0;
6579         em->block_len = em->len;
6580
6581         map->num_stripes = num_stripes;
6582         map->io_width = btrfs_chunk_io_width(leaf, chunk);
6583         map->io_align = btrfs_chunk_io_align(leaf, chunk);
6584         map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6585         map->type = btrfs_chunk_type(leaf, chunk);
6586         map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
6587         for (i = 0; i < num_stripes; i++) {
6588                 map->stripes[i].physical =
6589                         btrfs_stripe_offset_nr(leaf, chunk, i);
6590                 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
6591                 read_extent_buffer(leaf, uuid, (unsigned long)
6592                                    btrfs_stripe_dev_uuid_nr(chunk, i),
6593                                    BTRFS_UUID_SIZE);
6594                 map->stripes[i].dev = btrfs_find_device(fs_info, devid,
6595                                                         uuid, NULL);
6596                 if (!map->stripes[i].dev &&
6597                     !btrfs_test_opt(fs_info, DEGRADED)) {
6598                         free_extent_map(em);
6599                         btrfs_report_missing_device(fs_info, devid, uuid, true);
6600                         return -ENOENT;
6601                 }
6602                 if (!map->stripes[i].dev) {
6603                         map->stripes[i].dev =
6604                                 add_missing_dev(fs_info->fs_devices, devid,
6605                                                 uuid);
6606                         if (IS_ERR(map->stripes[i].dev)) {
6607                                 free_extent_map(em);
6608                                 btrfs_err(fs_info,
6609                                         "failed to init missing dev %llu: %ld",
6610                                         devid, PTR_ERR(map->stripes[i].dev));
6611                                 return PTR_ERR(map->stripes[i].dev);
6612                         }
6613                         btrfs_report_missing_device(fs_info, devid, uuid, false);
6614                 }
6615                 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
6616                                 &(map->stripes[i].dev->dev_state));
6617
6618         }
6619
6620         write_lock(&map_tree->map_tree.lock);
6621         ret = add_extent_mapping(&map_tree->map_tree, em, 0);
6622         write_unlock(&map_tree->map_tree.lock);
6623         BUG_ON(ret); /* Tree corruption */
6624         free_extent_map(em);
6625
6626         return 0;
6627 }
6628
6629 static void fill_device_from_item(struct extent_buffer *leaf,
6630                                  struct btrfs_dev_item *dev_item,
6631                                  struct btrfs_device *device)
6632 {
6633         unsigned long ptr;
6634
6635         device->devid = btrfs_device_id(leaf, dev_item);
6636         device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
6637         device->total_bytes = device->disk_total_bytes;
6638         device->commit_total_bytes = device->disk_total_bytes;
6639         device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
6640         device->commit_bytes_used = device->bytes_used;
6641         device->type = btrfs_device_type(leaf, dev_item);
6642         device->io_align = btrfs_device_io_align(leaf, dev_item);
6643         device->io_width = btrfs_device_io_width(leaf, dev_item);
6644         device->sector_size = btrfs_device_sector_size(leaf, dev_item);
6645         WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
6646         clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
6647
6648         ptr = btrfs_device_uuid(dev_item);
6649         read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
6650 }
6651
6652 static struct btrfs_fs_devices *open_seed_devices(struct btrfs_fs_info *fs_info,
6653                                                   u8 *fsid)
6654 {
6655         struct btrfs_fs_devices *fs_devices;
6656         int ret;
6657
6658         lockdep_assert_held(&uuid_mutex);
6659         ASSERT(fsid);
6660
6661         fs_devices = fs_info->fs_devices->seed;
6662         while (fs_devices) {
6663                 if (!memcmp(fs_devices->fsid, fsid, BTRFS_FSID_SIZE))
6664                         return fs_devices;
6665
6666                 fs_devices = fs_devices->seed;
6667         }
6668
6669         fs_devices = find_fsid(fsid);
6670         if (!fs_devices) {
6671                 if (!btrfs_test_opt(fs_info, DEGRADED))
6672                         return ERR_PTR(-ENOENT);
6673
6674                 fs_devices = alloc_fs_devices(fsid);
6675                 if (IS_ERR(fs_devices))
6676                         return fs_devices;
6677
6678                 fs_devices->seeding = 1;
6679                 fs_devices->opened = 1;
6680                 return fs_devices;
6681         }
6682
6683         fs_devices = clone_fs_devices(fs_devices);
6684         if (IS_ERR(fs_devices))
6685                 return fs_devices;
6686
6687         ret = __btrfs_open_devices(fs_devices, FMODE_READ,
6688                                    fs_info->bdev_holder);
6689         if (ret) {
6690                 free_fs_devices(fs_devices);
6691                 fs_devices = ERR_PTR(ret);
6692                 goto out;
6693         }
6694
6695         if (!fs_devices->seeding) {
6696                 __btrfs_close_devices(fs_devices);
6697                 free_fs_devices(fs_devices);
6698                 fs_devices = ERR_PTR(-EINVAL);
6699                 goto out;
6700         }
6701
6702         fs_devices->seed = fs_info->fs_devices->seed;
6703         fs_info->fs_devices->seed = fs_devices;
6704 out:
6705         return fs_devices;
6706 }
6707
6708 static int read_one_dev(struct btrfs_fs_info *fs_info,
6709                         struct extent_buffer *leaf,
6710                         struct btrfs_dev_item *dev_item)
6711 {
6712         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6713         struct btrfs_device *device;
6714         u64 devid;
6715         int ret;
6716         u8 fs_uuid[BTRFS_FSID_SIZE];
6717         u8 dev_uuid[BTRFS_UUID_SIZE];
6718
6719         devid = btrfs_device_id(leaf, dev_item);
6720         read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
6721                            BTRFS_UUID_SIZE);
6722         read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
6723                            BTRFS_FSID_SIZE);
6724
6725         if (memcmp(fs_uuid, fs_info->fsid, BTRFS_FSID_SIZE)) {
6726                 fs_devices = open_seed_devices(fs_info, fs_uuid);
6727                 if (IS_ERR(fs_devices))
6728                         return PTR_ERR(fs_devices);
6729         }
6730
6731         device = btrfs_find_device(fs_info, devid, dev_uuid, fs_uuid);
6732         if (!device) {
6733                 if (!btrfs_test_opt(fs_info, DEGRADED)) {
6734                         btrfs_report_missing_device(fs_info, devid,
6735                                                         dev_uuid, true);
6736                         return -ENOENT;
6737                 }
6738
6739                 device = add_missing_dev(fs_devices, devid, dev_uuid);
6740                 if (IS_ERR(device)) {
6741                         btrfs_err(fs_info,
6742                                 "failed to add missing dev %llu: %ld",
6743                                 devid, PTR_ERR(device));
6744                         return PTR_ERR(device);
6745                 }
6746                 btrfs_report_missing_device(fs_info, devid, dev_uuid, false);
6747         } else {
6748                 if (!device->bdev) {
6749                         if (!btrfs_test_opt(fs_info, DEGRADED)) {
6750                                 btrfs_report_missing_device(fs_info,
6751                                                 devid, dev_uuid, true);
6752                                 return -ENOENT;
6753                         }
6754                         btrfs_report_missing_device(fs_info, devid,
6755                                                         dev_uuid, false);
6756                 }
6757
6758                 if (!device->bdev &&
6759                     !test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) {
6760                         /*
6761                          * this happens when a device that was properly setup
6762                          * in the device info lists suddenly goes bad.
6763                          * device->bdev is NULL, and so we have to set
6764                          * device->missing to one here
6765                          */
6766                         device->fs_devices->missing_devices++;
6767                         set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
6768                 }
6769
6770                 /* Move the device to its own fs_devices */
6771                 if (device->fs_devices != fs_devices) {
6772                         ASSERT(test_bit(BTRFS_DEV_STATE_MISSING,
6773                                                         &device->dev_state));
6774
6775                         list_move(&device->dev_list, &fs_devices->devices);
6776                         device->fs_devices->num_devices--;
6777                         fs_devices->num_devices++;
6778
6779                         device->fs_devices->missing_devices--;
6780                         fs_devices->missing_devices++;
6781
6782                         device->fs_devices = fs_devices;
6783                 }
6784         }
6785
6786         if (device->fs_devices != fs_info->fs_devices) {
6787                 BUG_ON(test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state));
6788                 if (device->generation !=
6789                     btrfs_device_generation(leaf, dev_item))
6790                         return -EINVAL;
6791         }
6792
6793         fill_device_from_item(leaf, dev_item, device);
6794         set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
6795         if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
6796            !test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
6797                 device->fs_devices->total_rw_bytes += device->total_bytes;
6798                 atomic64_add(device->total_bytes - device->bytes_used,
6799                                 &fs_info->free_chunk_space);
6800         }
6801         ret = 0;
6802         return ret;
6803 }
6804
6805 int btrfs_read_sys_array(struct btrfs_fs_info *fs_info)
6806 {
6807         struct btrfs_root *root = fs_info->tree_root;
6808         struct btrfs_super_block *super_copy = fs_info->super_copy;
6809         struct extent_buffer *sb;
6810         struct btrfs_disk_key *disk_key;
6811         struct btrfs_chunk *chunk;
6812         u8 *array_ptr;
6813         unsigned long sb_array_offset;
6814         int ret = 0;
6815         u32 num_stripes;
6816         u32 array_size;
6817         u32 len = 0;
6818         u32 cur_offset;
6819         u64 type;
6820         struct btrfs_key key;
6821
6822         ASSERT(BTRFS_SUPER_INFO_SIZE <= fs_info->nodesize);
6823         /*
6824          * This will create extent buffer of nodesize, superblock size is
6825          * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will
6826          * overallocate but we can keep it as-is, only the first page is used.
6827          */
6828         sb = btrfs_find_create_tree_block(fs_info, BTRFS_SUPER_INFO_OFFSET);
6829         if (IS_ERR(sb))
6830                 return PTR_ERR(sb);
6831         set_extent_buffer_uptodate(sb);
6832         btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
6833         /*
6834          * The sb extent buffer is artificial and just used to read the system array.
6835          * set_extent_buffer_uptodate() call does not properly mark all it's
6836          * pages up-to-date when the page is larger: extent does not cover the
6837          * whole page and consequently check_page_uptodate does not find all
6838          * the page's extents up-to-date (the hole beyond sb),
6839          * write_extent_buffer then triggers a WARN_ON.
6840          *
6841          * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
6842          * but sb spans only this function. Add an explicit SetPageUptodate call
6843          * to silence the warning eg. on PowerPC 64.
6844          */
6845         if (PAGE_SIZE > BTRFS_SUPER_INFO_SIZE)
6846                 SetPageUptodate(sb->pages[0]);
6847
6848         write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
6849         array_size = btrfs_super_sys_array_size(super_copy);
6850
6851         array_ptr = super_copy->sys_chunk_array;
6852         sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array);
6853         cur_offset = 0;
6854
6855         while (cur_offset < array_size) {
6856                 disk_key = (struct btrfs_disk_key *)array_ptr;
6857                 len = sizeof(*disk_key);
6858                 if (cur_offset + len > array_size)
6859                         goto out_short_read;
6860
6861                 btrfs_disk_key_to_cpu(&key, disk_key);
6862
6863                 array_ptr += len;
6864                 sb_array_offset += len;
6865                 cur_offset += len;
6866
6867                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
6868                         chunk = (struct btrfs_chunk *)sb_array_offset;
6869                         /*
6870                          * At least one btrfs_chunk with one stripe must be
6871                          * present, exact stripe count check comes afterwards
6872                          */
6873                         len = btrfs_chunk_item_size(1);
6874                         if (cur_offset + len > array_size)
6875                                 goto out_short_read;
6876
6877                         num_stripes = btrfs_chunk_num_stripes(sb, chunk);
6878                         if (!num_stripes) {
6879                                 btrfs_err(fs_info,
6880                                         "invalid number of stripes %u in sys_array at offset %u",
6881                                         num_stripes, cur_offset);
6882                                 ret = -EIO;
6883                                 break;
6884                         }
6885
6886                         type = btrfs_chunk_type(sb, chunk);
6887                         if ((type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) {
6888                                 btrfs_err(fs_info,
6889                             "invalid chunk type %llu in sys_array at offset %u",
6890                                         type, cur_offset);
6891                                 ret = -EIO;
6892                                 break;
6893                         }
6894
6895                         len = btrfs_chunk_item_size(num_stripes);
6896                         if (cur_offset + len > array_size)
6897                                 goto out_short_read;
6898
6899                         ret = read_one_chunk(fs_info, &key, sb, chunk);
6900                         if (ret)
6901                                 break;
6902                 } else {
6903                         btrfs_err(fs_info,
6904                             "unexpected item type %u in sys_array at offset %u",
6905                                   (u32)key.type, cur_offset);
6906                         ret = -EIO;
6907                         break;
6908                 }
6909                 array_ptr += len;
6910                 sb_array_offset += len;
6911                 cur_offset += len;
6912         }
6913         clear_extent_buffer_uptodate(sb);
6914         free_extent_buffer_stale(sb);
6915         return ret;
6916
6917 out_short_read:
6918         btrfs_err(fs_info, "sys_array too short to read %u bytes at offset %u",
6919                         len, cur_offset);
6920         clear_extent_buffer_uptodate(sb);
6921         free_extent_buffer_stale(sb);
6922         return -EIO;
6923 }
6924
6925 /*
6926  * Check if all chunks in the fs are OK for read-write degraded mount
6927  *
6928  * If the @failing_dev is specified, it's accounted as missing.
6929  *
6930  * Return true if all chunks meet the minimal RW mount requirements.
6931  * Return false if any chunk doesn't meet the minimal RW mount requirements.
6932  */
6933 bool btrfs_check_rw_degradable(struct btrfs_fs_info *fs_info,
6934                                         struct btrfs_device *failing_dev)
6935 {
6936         struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
6937         struct extent_map *em;
6938         u64 next_start = 0;
6939         bool ret = true;
6940
6941         read_lock(&map_tree->map_tree.lock);
6942         em = lookup_extent_mapping(&map_tree->map_tree, 0, (u64)-1);
6943         read_unlock(&map_tree->map_tree.lock);
6944         /* No chunk at all? Return false anyway */
6945         if (!em) {
6946                 ret = false;
6947                 goto out;
6948         }
6949         while (em) {
6950                 struct map_lookup *map;
6951                 int missing = 0;
6952                 int max_tolerated;
6953                 int i;
6954
6955                 map = em->map_lookup;
6956                 max_tolerated =
6957                         btrfs_get_num_tolerated_disk_barrier_failures(
6958                                         map->type);
6959                 for (i = 0; i < map->num_stripes; i++) {
6960                         struct btrfs_device *dev = map->stripes[i].dev;
6961
6962                         if (!dev || !dev->bdev ||
6963                             test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state) ||
6964                             dev->last_flush_error)
6965                                 missing++;
6966                         else if (failing_dev && failing_dev == dev)
6967                                 missing++;
6968                 }
6969                 if (missing > max_tolerated) {
6970                         if (!failing_dev)
6971                                 btrfs_warn(fs_info,
6972         "chunk %llu missing %d devices, max tolerance is %d for writeable mount",
6973                                    em->start, missing, max_tolerated);
6974                         free_extent_map(em);
6975                         ret = false;
6976                         goto out;
6977                 }
6978                 next_start = extent_map_end(em);
6979                 free_extent_map(em);
6980
6981                 read_lock(&map_tree->map_tree.lock);
6982                 em = lookup_extent_mapping(&map_tree->map_tree, next_start,
6983                                            (u64)(-1) - next_start);
6984                 read_unlock(&map_tree->map_tree.lock);
6985         }
6986 out:
6987         return ret;
6988 }
6989
6990 int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info)
6991 {
6992         struct btrfs_root *root = fs_info->chunk_root;
6993         struct btrfs_path *path;
6994         struct extent_buffer *leaf;
6995         struct btrfs_key key;
6996         struct btrfs_key found_key;
6997         int ret;
6998         int slot;
6999         u64 total_dev = 0;
7000
7001         path = btrfs_alloc_path();
7002         if (!path)
7003                 return -ENOMEM;
7004
7005         mutex_lock(&uuid_mutex);
7006         mutex_lock(&fs_info->chunk_mutex);
7007
7008         /*
7009          * Read all device items, and then all the chunk items. All
7010          * device items are found before any chunk item (their object id
7011          * is smaller than the lowest possible object id for a chunk
7012          * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
7013          */
7014         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
7015         key.offset = 0;
7016         key.type = 0;
7017         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
7018         if (ret < 0)
7019                 goto error;
7020         while (1) {
7021                 leaf = path->nodes[0];
7022                 slot = path->slots[0];
7023                 if (slot >= btrfs_header_nritems(leaf)) {
7024                         ret = btrfs_next_leaf(root, path);
7025                         if (ret == 0)
7026                                 continue;
7027                         if (ret < 0)
7028                                 goto error;
7029                         break;
7030                 }
7031                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
7032                 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
7033                         struct btrfs_dev_item *dev_item;
7034                         dev_item = btrfs_item_ptr(leaf, slot,
7035                                                   struct btrfs_dev_item);
7036                         ret = read_one_dev(fs_info, leaf, dev_item);
7037                         if (ret)
7038                                 goto error;
7039                         total_dev++;
7040                 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
7041                         struct btrfs_chunk *chunk;
7042                         chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
7043                         ret = read_one_chunk(fs_info, &found_key, leaf, chunk);
7044                         if (ret)
7045                                 goto error;
7046                 }
7047                 path->slots[0]++;
7048         }
7049
7050         /*
7051          * After loading chunk tree, we've got all device information,
7052          * do another round of validation checks.
7053          */
7054         if (total_dev != fs_info->fs_devices->total_devices) {
7055                 btrfs_err(fs_info,
7056            "super_num_devices %llu mismatch with num_devices %llu found here",
7057                           btrfs_super_num_devices(fs_info->super_copy),
7058                           total_dev);
7059                 ret = -EINVAL;
7060                 goto error;
7061         }
7062         if (btrfs_super_total_bytes(fs_info->super_copy) <
7063             fs_info->fs_devices->total_rw_bytes) {
7064                 btrfs_err(fs_info,
7065         "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu",
7066                           btrfs_super_total_bytes(fs_info->super_copy),
7067                           fs_info->fs_devices->total_rw_bytes);
7068                 ret = -EINVAL;
7069                 goto error;
7070         }
7071         ret = 0;
7072 error:
7073         mutex_unlock(&fs_info->chunk_mutex);
7074         mutex_unlock(&uuid_mutex);
7075
7076         btrfs_free_path(path);
7077         return ret;
7078 }
7079
7080 void btrfs_init_devices_late(struct btrfs_fs_info *fs_info)
7081 {
7082         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7083         struct btrfs_device *device;
7084
7085         while (fs_devices) {
7086                 mutex_lock(&fs_devices->device_list_mutex);
7087                 list_for_each_entry(device, &fs_devices->devices, dev_list)
7088                         device->fs_info = fs_info;
7089                 mutex_unlock(&fs_devices->device_list_mutex);
7090
7091                 fs_devices = fs_devices->seed;
7092         }
7093 }
7094
7095 static void __btrfs_reset_dev_stats(struct btrfs_device *dev)
7096 {
7097         int i;
7098
7099         for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7100                 btrfs_dev_stat_reset(dev, i);
7101 }
7102
7103 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
7104 {
7105         struct btrfs_key key;
7106         struct btrfs_key found_key;
7107         struct btrfs_root *dev_root = fs_info->dev_root;
7108         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7109         struct extent_buffer *eb;
7110         int slot;
7111         int ret = 0;
7112         struct btrfs_device *device;
7113         struct btrfs_path *path = NULL;
7114         int i;
7115
7116         path = btrfs_alloc_path();
7117         if (!path) {
7118                 ret = -ENOMEM;
7119                 goto out;
7120         }
7121
7122         mutex_lock(&fs_devices->device_list_mutex);
7123         list_for_each_entry(device, &fs_devices->devices, dev_list) {
7124                 int item_size;
7125                 struct btrfs_dev_stats_item *ptr;
7126
7127                 key.objectid = BTRFS_DEV_STATS_OBJECTID;
7128                 key.type = BTRFS_PERSISTENT_ITEM_KEY;
7129                 key.offset = device->devid;
7130                 ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
7131                 if (ret) {
7132                         __btrfs_reset_dev_stats(device);
7133                         device->dev_stats_valid = 1;
7134                         btrfs_release_path(path);
7135                         continue;
7136                 }
7137                 slot = path->slots[0];
7138                 eb = path->nodes[0];
7139                 btrfs_item_key_to_cpu(eb, &found_key, slot);
7140                 item_size = btrfs_item_size_nr(eb, slot);
7141
7142                 ptr = btrfs_item_ptr(eb, slot,
7143                                      struct btrfs_dev_stats_item);
7144
7145                 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7146                         if (item_size >= (1 + i) * sizeof(__le64))
7147                                 btrfs_dev_stat_set(device, i,
7148                                         btrfs_dev_stats_value(eb, ptr, i));
7149                         else
7150                                 btrfs_dev_stat_reset(device, i);
7151                 }
7152
7153                 device->dev_stats_valid = 1;
7154                 btrfs_dev_stat_print_on_load(device);
7155                 btrfs_release_path(path);
7156         }
7157         mutex_unlock(&fs_devices->device_list_mutex);
7158
7159 out:
7160         btrfs_free_path(path);
7161         return ret < 0 ? ret : 0;
7162 }
7163
7164 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
7165                                 struct btrfs_fs_info *fs_info,
7166                                 struct btrfs_device *device)
7167 {
7168         struct btrfs_root *dev_root = fs_info->dev_root;
7169         struct btrfs_path *path;
7170         struct btrfs_key key;
7171         struct extent_buffer *eb;
7172         struct btrfs_dev_stats_item *ptr;
7173         int ret;
7174         int i;
7175
7176         key.objectid = BTRFS_DEV_STATS_OBJECTID;
7177         key.type = BTRFS_PERSISTENT_ITEM_KEY;
7178         key.offset = device->devid;
7179
7180         path = btrfs_alloc_path();
7181         if (!path)
7182                 return -ENOMEM;
7183         ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
7184         if (ret < 0) {
7185                 btrfs_warn_in_rcu(fs_info,
7186                         "error %d while searching for dev_stats item for device %s",
7187                               ret, rcu_str_deref(device->name));
7188                 goto out;
7189         }
7190
7191         if (ret == 0 &&
7192             btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
7193                 /* need to delete old one and insert a new one */
7194                 ret = btrfs_del_item(trans, dev_root, path);
7195                 if (ret != 0) {
7196                         btrfs_warn_in_rcu(fs_info,
7197                                 "delete too small dev_stats item for device %s failed %d",
7198                                       rcu_str_deref(device->name), ret);
7199                         goto out;
7200                 }
7201                 ret = 1;
7202         }
7203
7204         if (ret == 1) {
7205                 /* need to insert a new item */
7206                 btrfs_release_path(path);
7207                 ret = btrfs_insert_empty_item(trans, dev_root, path,
7208                                               &key, sizeof(*ptr));
7209                 if (ret < 0) {
7210                         btrfs_warn_in_rcu(fs_info,
7211                                 "insert dev_stats item for device %s failed %d",
7212                                 rcu_str_deref(device->name), ret);
7213                         goto out;
7214                 }
7215         }
7216
7217         eb = path->nodes[0];
7218         ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
7219         for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7220                 btrfs_set_dev_stats_value(eb, ptr, i,
7221                                           btrfs_dev_stat_read(device, i));
7222         btrfs_mark_buffer_dirty(eb);
7223
7224 out:
7225         btrfs_free_path(path);
7226         return ret;
7227 }
7228
7229 /*
7230  * called from commit_transaction. Writes all changed device stats to disk.
7231  */
7232 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans,
7233                         struct btrfs_fs_info *fs_info)
7234 {
7235         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7236         struct btrfs_device *device;
7237         int stats_cnt;
7238         int ret = 0;
7239
7240         mutex_lock(&fs_devices->device_list_mutex);
7241         list_for_each_entry(device, &fs_devices->devices, dev_list) {
7242                 stats_cnt = atomic_read(&device->dev_stats_ccnt);
7243                 if (!device->dev_stats_valid || stats_cnt == 0)
7244                         continue;
7245
7246
7247                 /*
7248                  * There is a LOAD-LOAD control dependency between the value of
7249                  * dev_stats_ccnt and updating the on-disk values which requires
7250                  * reading the in-memory counters. Such control dependencies
7251                  * require explicit read memory barriers.
7252                  *
7253                  * This memory barriers pairs with smp_mb__before_atomic in
7254                  * btrfs_dev_stat_inc/btrfs_dev_stat_set and with the full
7255                  * barrier implied by atomic_xchg in
7256                  * btrfs_dev_stats_read_and_reset
7257                  */
7258                 smp_rmb();
7259
7260                 ret = update_dev_stat_item(trans, fs_info, device);
7261                 if (!ret)
7262                         atomic_sub(stats_cnt, &device->dev_stats_ccnt);
7263         }
7264         mutex_unlock(&fs_devices->device_list_mutex);
7265
7266         return ret;
7267 }
7268
7269 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
7270 {
7271         btrfs_dev_stat_inc(dev, index);
7272         btrfs_dev_stat_print_on_error(dev);
7273 }
7274
7275 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
7276 {
7277         if (!dev->dev_stats_valid)
7278                 return;
7279         btrfs_err_rl_in_rcu(dev->fs_info,
7280                 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7281                            rcu_str_deref(dev->name),
7282                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7283                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7284                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7285                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7286                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7287 }
7288
7289 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
7290 {
7291         int i;
7292
7293         for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7294                 if (btrfs_dev_stat_read(dev, i) != 0)
7295                         break;
7296         if (i == BTRFS_DEV_STAT_VALUES_MAX)
7297                 return; /* all values == 0, suppress message */
7298
7299         btrfs_info_in_rcu(dev->fs_info,
7300                 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7301                rcu_str_deref(dev->name),
7302                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7303                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7304                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7305                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7306                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7307 }
7308
7309 int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info,
7310                         struct btrfs_ioctl_get_dev_stats *stats)
7311 {
7312         struct btrfs_device *dev;
7313         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7314         int i;
7315
7316         mutex_lock(&fs_devices->device_list_mutex);
7317         dev = btrfs_find_device(fs_info, stats->devid, NULL, NULL);
7318         mutex_unlock(&fs_devices->device_list_mutex);
7319
7320         if (!dev) {
7321                 btrfs_warn(fs_info, "get dev_stats failed, device not found");
7322                 return -ENODEV;
7323         } else if (!dev->dev_stats_valid) {
7324                 btrfs_warn(fs_info, "get dev_stats failed, not yet valid");
7325                 return -ENODEV;
7326         } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
7327                 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7328                         if (stats->nr_items > i)
7329                                 stats->values[i] =
7330                                         btrfs_dev_stat_read_and_reset(dev, i);
7331                         else
7332                                 btrfs_dev_stat_reset(dev, i);
7333                 }
7334         } else {
7335                 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7336                         if (stats->nr_items > i)
7337                                 stats->values[i] = btrfs_dev_stat_read(dev, i);
7338         }
7339         if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
7340                 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
7341         return 0;
7342 }
7343
7344 void btrfs_scratch_superblocks(struct block_device *bdev, const char *device_path)
7345 {
7346         struct buffer_head *bh;
7347         struct btrfs_super_block *disk_super;
7348         int copy_num;
7349
7350         if (!bdev)
7351                 return;
7352
7353         for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX;
7354                 copy_num++) {
7355
7356                 if (btrfs_read_dev_one_super(bdev, copy_num, &bh))
7357                         continue;
7358
7359                 disk_super = (struct btrfs_super_block *)bh->b_data;
7360
7361                 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
7362                 set_buffer_dirty(bh);
7363                 sync_dirty_buffer(bh);
7364                 brelse(bh);
7365         }
7366
7367         /* Notify udev that device has changed */
7368         btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
7369
7370         /* Update ctime/mtime for device path for libblkid */
7371         update_dev_time(device_path);
7372 }
7373
7374 /*
7375  * Update the size of all devices, which is used for writing out the
7376  * super blocks.
7377  */
7378 void btrfs_update_commit_device_size(struct btrfs_fs_info *fs_info)
7379 {
7380         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7381         struct btrfs_device *curr, *next;
7382
7383         if (list_empty(&fs_devices->resized_devices))
7384                 return;
7385
7386         mutex_lock(&fs_devices->device_list_mutex);
7387         mutex_lock(&fs_info->chunk_mutex);
7388         list_for_each_entry_safe(curr, next, &fs_devices->resized_devices,
7389                                  resized_list) {
7390                 list_del_init(&curr->resized_list);
7391                 curr->commit_total_bytes = curr->disk_total_bytes;
7392         }
7393         mutex_unlock(&fs_info->chunk_mutex);
7394         mutex_unlock(&fs_devices->device_list_mutex);
7395 }
7396
7397 /* Must be invoked during the transaction commit */
7398 void btrfs_update_commit_device_bytes_used(struct btrfs_transaction *trans)
7399 {
7400         struct btrfs_fs_info *fs_info = trans->fs_info;
7401         struct extent_map *em;
7402         struct map_lookup *map;
7403         struct btrfs_device *dev;
7404         int i;
7405
7406         if (list_empty(&trans->pending_chunks))
7407                 return;
7408
7409         /* In order to kick the device replace finish process */
7410         mutex_lock(&fs_info->chunk_mutex);
7411         list_for_each_entry(em, &trans->pending_chunks, list) {
7412                 map = em->map_lookup;
7413
7414                 for (i = 0; i < map->num_stripes; i++) {
7415                         dev = map->stripes[i].dev;
7416                         dev->commit_bytes_used = dev->bytes_used;
7417                 }
7418         }
7419         mutex_unlock(&fs_info->chunk_mutex);
7420 }
7421
7422 void btrfs_set_fs_info_ptr(struct btrfs_fs_info *fs_info)
7423 {
7424         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7425         while (fs_devices) {
7426                 fs_devices->fs_info = fs_info;
7427                 fs_devices = fs_devices->seed;
7428         }
7429 }
7430
7431 void btrfs_reset_fs_info_ptr(struct btrfs_fs_info *fs_info)
7432 {
7433         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7434         while (fs_devices) {
7435                 fs_devices->fs_info = NULL;
7436                 fs_devices = fs_devices->seed;
7437         }
7438 }