fs/btrfs/volumes.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18 #include <linux/sched.h>
  19 #include <linux/bio.h>
  20 #include <linux/slab.h>
  21 #include <linux/buffer_head.h>
  22 #include <linux/blkdev.h>
  23 #include <linux/iocontext.h>
  24 #include <linux/capability.h>
  25 #include <linux/ratelimit.h>
  26 #include <linux/kthread.h>
  27 #include <linux/raid/pq.h>
  28 #include <linux/semaphore.h>
  29 #include <linux/uuid.h>
  30 #include <asm/div64.h>
  31 #include "ctree.h"
  32 #include "extent_map.h"
  33 #include "disk-io.h"
  34 #include "transaction.h"
  35 #include "print-tree.h"
  36 #include "volumes.h"
  37 #include "raid56.h"
  38 #include "async-thread.h"
  39 #include "check-integrity.h"
  40 #include "rcu-string.h"
  41 #include "math.h"
  42 #include "dev-replace.h"
  43 #include "sysfs.h"
  44
  45 const struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = {
  46         [BTRFS_RAID_RAID10] = {
  47                 .sub_stripes    = 2,
  48                 .dev_stripes    = 1,
  49                 .devs_max       = 0,    /* 0 == as many as possible */
  50                 .devs_min       = 4,
  51                 .tolerated_failures = 1,
  52                 .devs_increment = 2,
  53                 .ncopies        = 2,
  54         },
  55         [BTRFS_RAID_RAID1] = {
  56                 .sub_stripes    = 1,
  57                 .dev_stripes    = 1,
  58                 .devs_max       = 2,
  59                 .devs_min       = 2,
  60                 .tolerated_failures = 1,
  61                 .devs_increment = 2,
  62                 .ncopies        = 2,
  63         },
  64         [BTRFS_RAID_DUP] = {
  65                 .sub_stripes    = 1,
  66                 .dev_stripes    = 2,
  67                 .devs_max       = 1,
  68                 .devs_min       = 1,
  69                 .tolerated_failures = 0,
  70                 .devs_increment = 1,
  71                 .ncopies        = 2,
  72         },
  73         [BTRFS_RAID_RAID0] = {
  74                 .sub_stripes    = 1,
  75                 .dev_stripes    = 1,
  76                 .devs_max       = 0,
  77                 .devs_min       = 2,
  78                 .tolerated_failures = 0,
  79                 .devs_increment = 1,
  80                 .ncopies        = 1,
  81         },
  82         [BTRFS_RAID_SINGLE] = {
  83                 .sub_stripes    = 1,
  84                 .dev_stripes    = 1,
  85                 .devs_max       = 1,
  86                 .devs_min       = 1,
  87                 .tolerated_failures = 0,
  88                 .devs_increment = 1,
  89                 .ncopies        = 1,
  90         },
  91         [BTRFS_RAID_RAID5] = {
  92                 .sub_stripes    = 1,
  93                 .dev_stripes    = 1,
  94                 .devs_max       = 0,
  95                 .devs_min       = 2,
  96                 .tolerated_failures = 1,
  97                 .devs_increment = 1,
  98                 .ncopies        = 2,
  99         },
 100         [BTRFS_RAID_RAID6] = {
 101                 .sub_stripes    = 1,
 102                 .dev_stripes    = 1,
 103                 .devs_max       = 0,
 104                 .devs_min       = 3,
 105                 .tolerated_failures = 2,
 106                 .devs_increment = 1,
 107                 .ncopies        = 3,
 108         },
 109 };
 110
 111 const u64 btrfs_raid_group[BTRFS_NR_RAID_TYPES] = {
 112         [BTRFS_RAID_RAID10] = BTRFS_BLOCK_GROUP_RAID10,
 113         [BTRFS_RAID_RAID1]  = BTRFS_BLOCK_GROUP_RAID1,
 114         [BTRFS_RAID_DUP]    = BTRFS_BLOCK_GROUP_DUP,
 115         [BTRFS_RAID_RAID0]  = BTRFS_BLOCK_GROUP_RAID0,
 116         [BTRFS_RAID_SINGLE] = 0,
 117         [BTRFS_RAID_RAID5]  = BTRFS_BLOCK_GROUP_RAID5,
 118         [BTRFS_RAID_RAID6]  = BTRFS_BLOCK_GROUP_RAID6,
 119 };
 120
 121 /*
 122  * Table to convert BTRFS_RAID_* to the error code if minimum number of devices
 123  * condition is not met. Zero means there's no corresponding
 124  * BTRFS_ERROR_DEV_*_NOT_MET value.
 125  */
 126 const int btrfs_raid_mindev_error[BTRFS_NR_RAID_TYPES] = {
 127         [BTRFS_RAID_RAID10] = BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET,
 128         [BTRFS_RAID_RAID1]  = BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET,
 129         [BTRFS_RAID_DUP]    = 0,
 130         [BTRFS_RAID_RAID0]  = 0,
 131         [BTRFS_RAID_SINGLE] = 0,
 132         [BTRFS_RAID_RAID5]  = BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET,
 133         [BTRFS_RAID_RAID6]  = BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET,
 134 };
 135
 136 static int init_first_rw_device(struct btrfs_trans_handle *trans,
 137                                 struct btrfs_fs_info *fs_info);
 138 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info);
 139 static void __btrfs_reset_dev_stats(struct btrfs_device *dev);
 140 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev);
 141 static void btrfs_dev_stat_print_on_load(struct btrfs_device *device);
 142 static int __btrfs_map_block(struct btrfs_fs_info *fs_info,
 143                              enum btrfs_map_op op,
 144                              u64 logical, u64 *length,
 145                              struct btrfs_bio **bbio_ret,
 146                              int mirror_num, int need_raid_map);
 147
 148 DEFINE_MUTEX(uuid_mutex);
 149 static LIST_HEAD(fs_uuids);
 150 struct list_head *btrfs_get_fs_uuids(void)
 151 {
 152         return &fs_uuids;
 153 }
 154
 155 /*
 156  * alloc_fs_devices - allocate struct btrfs_fs_devices
 157  * @fsid:       if not NULL, copy the uuid to fs_devices::fsid
 158  *
 159  * Return a pointer to a new struct btrfs_fs_devices on success, or ERR_PTR().
 160  * The returned struct is not linked onto any lists and can be destroyed with
 161  * kfree() right away.
 162  */
 163 static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid)
 164 {
 165         struct btrfs_fs_devices *fs_devs;
 166
 167         fs_devs = kzalloc(sizeof(*fs_devs), GFP_KERNEL);
 168         if (!fs_devs)
 169                 return ERR_PTR(-ENOMEM);
 170
 171         mutex_init(&fs_devs->device_list_mutex);
 172
 173         INIT_LIST_HEAD(&fs_devs->devices);
 174         INIT_LIST_HEAD(&fs_devs->resized_devices);
 175         INIT_LIST_HEAD(&fs_devs->alloc_list);
 176         INIT_LIST_HEAD(&fs_devs->list);
 177         if (fsid)
 178                 memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE);
 179
 180         return fs_devs;
 181 }
 182
 183 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
 184 {
 185         struct btrfs_device *device;
 186         WARN_ON(fs_devices->opened);
 187         while (!list_empty(&fs_devices->devices)) {
 188                 device = list_entry(fs_devices->devices.next,
 189                                     struct btrfs_device, dev_list);
 190                 list_del(&device->dev_list);
 191                 rcu_string_free(device->name);
 192                 kfree(device);
 193         }
 194         kfree(fs_devices);
 195 }
 196
 197 static void btrfs_kobject_uevent(struct block_device *bdev,
 198                                  enum kobject_action action)
 199 {
 200         int ret;
 201
 202         ret = kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, action);
 203         if (ret)
 204                 pr_warn("BTRFS: Sending event '%d' to kobject: '%s' (%p): failed\n",
 205                         action,
 206                         kobject_name(&disk_to_dev(bdev->bd_disk)->kobj),
 207                         &disk_to_dev(bdev->bd_disk)->kobj);
 208 }
 209
 210 void btrfs_cleanup_fs_uuids(void)
 211 {
 212         struct btrfs_fs_devices *fs_devices;
 213
 214         while (!list_empty(&fs_uuids)) {
 215                 fs_devices = list_entry(fs_uuids.next,
 216                                         struct btrfs_fs_devices, list);
 217                 list_del(&fs_devices->list);
 218                 free_fs_devices(fs_devices);
 219         }
 220 }
 221
 222 static struct btrfs_device *__alloc_device(void)
 223 {
 224         struct btrfs_device *dev;
 225
 226         dev = kzalloc(sizeof(*dev), GFP_KERNEL);
 227         if (!dev)
 228                 return ERR_PTR(-ENOMEM);
 229
 230         /*
 231          * Preallocate a bio that's always going to be used for flushing device
 232          * barriers and matches the device lifespan
 233          */
 234         dev->flush_bio = bio_alloc_bioset(GFP_KERNEL, 0, NULL);
 235         if (!dev->flush_bio) {
 236                 kfree(dev);
 237                 return ERR_PTR(-ENOMEM);
 238         }
 239
 240         INIT_LIST_HEAD(&dev->dev_list);
 241         INIT_LIST_HEAD(&dev->dev_alloc_list);
 242         INIT_LIST_HEAD(&dev->resized_list);
 243
 244         spin_lock_init(&dev->io_lock);
 245
 246         spin_lock_init(&dev->reada_lock);
 247         atomic_set(&dev->reada_in_flight, 0);
 248         atomic_set(&dev->dev_stats_ccnt, 0);
 249         btrfs_device_data_ordered_init(dev);
 250         INIT_RADIX_TREE(&dev->reada_zones, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
 251         INIT_RADIX_TREE(&dev->reada_extents, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
 252
 253         return dev;
 254 }
 255
 256 /*
 257  * Find a device specified by @devid or @uuid in the list of @fs_devices, or
 258  * return NULL.
 259  *
 260  * If devid and uuid are both specified, the match must be exact, otherwise
 261  * only devid is used.
 262  */
 263 static struct btrfs_device *find_device(struct btrfs_fs_devices *fs_devices,
 264                 u64 devid, const u8 *uuid)
 265 {
 266         struct list_head *head = &fs_devices->devices;
 267         struct btrfs_device *dev;
 268
 269         list_for_each_entry(dev, head, dev_list) {
 270                 if (dev->devid == devid &&
 271                     (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
 272                         return dev;
 273                 }
 274         }
 275         return NULL;
 276 }
 277
 278 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
 279 {
 280         struct btrfs_fs_devices *fs_devices;
 281
 282         list_for_each_entry(fs_devices, &fs_uuids, list) {
 283                 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
 284                         return fs_devices;
 285         }
 286         return NULL;
 287 }
 288
 289 static int
 290 btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder,
 291                       int flush, struct block_device **bdev,
 292                       struct buffer_head **bh)
 293 {
 294         int ret;
 295
 296         *bdev = blkdev_get_by_path(device_path, flags, holder);
 297
 298         if (IS_ERR(*bdev)) {
 299                 ret = PTR_ERR(*bdev);
 300                 goto error;
 301         }
 302
 303         if (flush)
 304                 filemap_write_and_wait((*bdev)->bd_inode->i_mapping);
 305         ret = set_blocksize(*bdev, BTRFS_BDEV_BLOCKSIZE);
 306         if (ret) {
 307                 blkdev_put(*bdev, flags);
 308                 goto error;
 309         }
 310         invalidate_bdev(*bdev);
 311         *bh = btrfs_read_dev_super(*bdev);
 312         if (IS_ERR(*bh)) {
 313                 ret = PTR_ERR(*bh);
 314                 blkdev_put(*bdev, flags);
 315                 goto error;
 316         }
 317
 318         return 0;
 319
 320 error:
 321         *bdev = NULL;
 322         *bh = NULL;
 323         return ret;
 324 }
 325
 326 static void requeue_list(struct btrfs_pending_bios *pending_bios,
 327                         struct bio *head, struct bio *tail)
 328 {
 329
 330         struct bio *old_head;
 331
 332         old_head = pending_bios->head;
 333         pending_bios->head = head;
 334         if (pending_bios->tail)
 335                 tail->bi_next = old_head;
 336         else
 337                 pending_bios->tail = tail;
 338 }
 339
 340 /*
 341  * we try to collect pending bios for a device so we don't get a large
 342  * number of procs sending bios down to the same device.  This greatly
 343  * improves the schedulers ability to collect and merge the bios.
 344  *
 345  * But, it also turns into a long list of bios to process and that is sure
 346  * to eventually make the worker thread block.  The solution here is to
 347  * make some progress and then put this work struct back at the end of
 348  * the list if the block device is congested.  This way, multiple devices
 349  * can make progress from a single worker thread.
 350  */
 351 static noinline void run_scheduled_bios(struct btrfs_device *device)
 352 {
 353         struct btrfs_fs_info *fs_info = device->fs_info;
 354         struct bio *pending;
 355         struct backing_dev_info *bdi;
 356         struct btrfs_pending_bios *pending_bios;
 357         struct bio *tail;
 358         struct bio *cur;
 359         int again = 0;
 360         unsigned long num_run;
 361         unsigned long batch_run = 0;
 362         unsigned long limit;
 363         unsigned long last_waited = 0;
 364         int force_reg = 0;
 365         int sync_pending = 0;
 366         struct blk_plug plug;
 367
 368         /*
 369          * this function runs all the bios we've collected for
 370          * a particular device.  We don't want to wander off to
 371          * another device without first sending all of these down.
 372          * So, setup a plug here and finish it off before we return
 373          */
 374         blk_start_plug(&plug);
 375
 376         bdi = device->bdev->bd_bdi;
 377         limit = btrfs_async_submit_limit(fs_info);
 378         limit = limit * 2 / 3;
 379
 380 loop:
 381         spin_lock(&device->io_lock);
 382
 383 loop_lock:
 384         num_run = 0;
 385
 386         /* take all the bios off the list at once and process them
 387          * later on (without the lock held).  But, remember the
 388          * tail and other pointers so the bios can be properly reinserted
 389          * into the list if we hit congestion
 390          */
 391         if (!force_reg && device->pending_sync_bios.head) {
 392                 pending_bios = &device->pending_sync_bios;
 393                 force_reg = 1;
 394         } else {
 395                 pending_bios = &device->pending_bios;
 396                 force_reg = 0;
 397         }
 398
 399         pending = pending_bios->head;
 400         tail = pending_bios->tail;
 401         WARN_ON(pending && !tail);
 402
 403         /*
 404          * if pending was null this time around, no bios need processing
 405          * at all and we can stop.  Otherwise it'll loop back up again
 406          * and do an additional check so no bios are missed.
 407          *
 408          * device->running_pending is used to synchronize with the
 409          * schedule_bio code.
 410          */
 411         if (device->pending_sync_bios.head == NULL &&
 412             device->pending_bios.head == NULL) {
 413                 again = 0;
 414                 device->running_pending = 0;
 415         } else {
 416                 again = 1;
 417                 device->running_pending = 1;
 418         }
 419
 420         pending_bios->head = NULL;
 421         pending_bios->tail = NULL;
 422
 423         spin_unlock(&device->io_lock);
 424
 425         while (pending) {
 426
 427                 rmb();
 428                 /* we want to work on both lists, but do more bios on the
 429                  * sync list than the regular list
 430                  */
 431                 if ((num_run > 32 &&
 432                     pending_bios != &device->pending_sync_bios &&
 433                     device->pending_sync_bios.head) ||
 434                    (num_run > 64 && pending_bios == &device->pending_sync_bios &&
 435                     device->pending_bios.head)) {
 436                         spin_lock(&device->io_lock);
 437                         requeue_list(pending_bios, pending, tail);
 438                         goto loop_lock;
 439                 }
 440
 441                 cur = pending;
 442                 pending = pending->bi_next;
 443                 cur->bi_next = NULL;
 444
 445                 /*
 446                  * atomic_dec_return implies a barrier for waitqueue_active
 447                  */
 448                 if (atomic_dec_return(&fs_info->nr_async_bios) < limit &&
 449                     waitqueue_active(&fs_info->async_submit_wait))
 450                         wake_up(&fs_info->async_submit_wait);
 451
 452                 BUG_ON(atomic_read(&cur->__bi_cnt) == 0);
 453
 454                 /*
 455                  * if we're doing the sync list, record that our
 456                  * plug has some sync requests on it
 457                  *
 458                  * If we're doing the regular list and there are
 459                  * sync requests sitting around, unplug before
 460                  * we add more
 461                  */
 462                 if (pending_bios == &device->pending_sync_bios) {
 463                         sync_pending = 1;
 464                 } else if (sync_pending) {
 465                         blk_finish_plug(&plug);
 466                         blk_start_plug(&plug);
 467                         sync_pending = 0;
 468                 }
 469
 470                 btrfsic_submit_bio(cur);
 471                 num_run++;
 472                 batch_run++;
 473
 474                 cond_resched();
 475
 476                 /*
 477                  * we made progress, there is more work to do and the bdi
 478                  * is now congested.  Back off and let other work structs
 479                  * run instead
 480                  */
 481                 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
 482                     fs_info->fs_devices->open_devices > 1) {
 483                         struct io_context *ioc;
 484
 485                         ioc = current->io_context;
 486
 487                         /*
 488                          * the main goal here is that we don't want to
 489                          * block if we're going to be able to submit
 490                          * more requests without blocking.
 491                          *
 492                          * This code does two great things, it pokes into
 493                          * the elevator code from a filesystem _and_
 494                          * it makes assumptions about how batching works.
 495                          */
 496                         if (ioc && ioc->nr_batch_requests > 0 &&
 497                             time_before(jiffies, ioc->last_waited + HZ/50UL) &&
 498                             (last_waited == 0 ||
 499                              ioc->last_waited == last_waited)) {
 500                                 /*
 501                                  * we want to go through our batch of
 502                                  * requests and stop.  So, we copy out
 503                                  * the ioc->last_waited time and test
 504                                  * against it before looping
 505                                  */
 506                                 last_waited = ioc->last_waited;
 507                                 cond_resched();
 508                                 continue;
 509                         }
 510                         spin_lock(&device->io_lock);
 511                         requeue_list(pending_bios, pending, tail);
 512                         device->running_pending = 1;
 513
 514                         spin_unlock(&device->io_lock);
 515                         btrfs_queue_work(fs_info->submit_workers,
 516                                          &device->work);
 517                         goto done;
 518                 }
 519                 /* unplug every 64 requests just for good measure */
 520                 if (batch_run % 64 == 0) {
 521                         blk_finish_plug(&plug);
 522                         blk_start_plug(&plug);
 523                         sync_pending = 0;
 524                 }
 525         }
 526
 527         cond_resched();
 528         if (again)
 529                 goto loop;
 530
 531         spin_lock(&device->io_lock);
 532         if (device->pending_bios.head || device->pending_sync_bios.head)
 533                 goto loop_lock;
 534         spin_unlock(&device->io_lock);
 535
 536 done:
 537         blk_finish_plug(&plug);
 538 }
 539
 540 static void pending_bios_fn(struct btrfs_work *work)
 541 {
 542         struct btrfs_device *device;
 543
 544         device = container_of(work, struct btrfs_device, work);
 545         run_scheduled_bios(device);
 546 }
 547
 548
 549 void btrfs_free_stale_device(struct btrfs_device *cur_dev)
 550 {
 551         struct btrfs_fs_devices *fs_devs;
 552         struct btrfs_device *dev;
 553
 554         if (!cur_dev->name)
 555                 return;
 556
 557         list_for_each_entry(fs_devs, &fs_uuids, list) {
 558                 int del = 1;
 559
 560                 if (fs_devs->opened)
 561                         continue;
 562                 if (fs_devs->seeding)
 563                         continue;
 564
 565                 list_for_each_entry(dev, &fs_devs->devices, dev_list) {
 566
 567                         if (dev == cur_dev)
 568                                 continue;
 569                         if (!dev->name)
 570                                 continue;
 571
 572                         /*
 573                          * Todo: This won't be enough. What if the same device
 574                          * comes back (with new uuid and) with its mapper path?
 575                          * But for now, this does help as mostly an admin will
 576                          * either use mapper or non mapper path throughout.
 577                          */
 578                         rcu_read_lock();
 579                         del = strcmp(rcu_str_deref(dev->name),
 580                                                 rcu_str_deref(cur_dev->name));
 581                         rcu_read_unlock();
 582                         if (!del)
 583                                 break;
 584                 }
 585
 586                 if (!del) {
 587                         /* delete the stale device */
 588                         if (fs_devs->num_devices == 1) {
 589                                 btrfs_sysfs_remove_fsid(fs_devs);
 590                                 list_del(&fs_devs->list);
 591                                 free_fs_devices(fs_devs);
 592                         } else {
 593                                 fs_devs->num_devices--;
 594                                 list_del(&dev->dev_list);
 595                                 rcu_string_free(dev->name);
 596                                 kfree(dev);
 597                         }
 598                         break;
 599                 }
 600         }
 601 }
 602
 603 /*
 604  * Add new device to list of registered devices
 605  *
 606  * Returns:
 607  * 1   - first time device is seen
 608  * 0   - device already known
 609  * < 0 - error
 610  */
 611 static noinline int device_list_add(const char *path,
 612                            struct btrfs_super_block *disk_super,
 613                            u64 devid, struct btrfs_fs_devices **fs_devices_ret)
 614 {
 615         struct btrfs_device *device;
 616         struct btrfs_fs_devices *fs_devices;
 617         struct rcu_string *name;
 618         int ret = 0;
 619         u64 found_transid = btrfs_super_generation(disk_super);
 620
 621         fs_devices = find_fsid(disk_super->fsid);
 622         if (!fs_devices) {
 623                 fs_devices = alloc_fs_devices(disk_super->fsid);
 624                 if (IS_ERR(fs_devices))
 625                         return PTR_ERR(fs_devices);
 626
 627                 list_add(&fs_devices->list, &fs_uuids);
 628
 629                 device = NULL;
 630         } else {
 631                 device = find_device(fs_devices, devid,
 632                                 disk_super->dev_item.uuid);
 633         }
 634
 635         if (!device) {
 636                 if (fs_devices->opened)
 637                         return -EBUSY;
 638
 639                 device = btrfs_alloc_device(NULL, &devid,
 640                                             disk_super->dev_item.uuid);
 641                 if (IS_ERR(device)) {
 642                         /* we can safely leave the fs_devices entry around */
 643                         return PTR_ERR(device);
 644                 }
 645
 646                 name = rcu_string_strdup(path, GFP_NOFS);
 647                 if (!name) {
 648                         kfree(device);
 649                         return -ENOMEM;
 650                 }
 651                 rcu_assign_pointer(device->name, name);
 652
 653                 mutex_lock(&fs_devices->device_list_mutex);
 654                 list_add_rcu(&device->dev_list, &fs_devices->devices);
 655                 fs_devices->num_devices++;
 656                 mutex_unlock(&fs_devices->device_list_mutex);
 657
 658                 ret = 1;
 659                 device->fs_devices = fs_devices;
 660         } else if (!device->name || strcmp(device->name->str, path)) {
 661                 /*
 662                  * When FS is already mounted.
 663                  * 1. If you are here and if the device->name is NULL that
 664                  *    means this device was missing at time of FS mount.
 665                  * 2. If you are here and if the device->name is different
 666                  *    from 'path' that means either
 667                  *      a. The same device disappeared and reappeared with
 668                  *         different name. or
 669                  *      b. The missing-disk-which-was-replaced, has
 670                  *         reappeared now.
 671                  *
 672                  * We must allow 1 and 2a above. But 2b would be a spurious
 673                  * and unintentional.
 674                  *
 675                  * Further in case of 1 and 2a above, the disk at 'path'
 676                  * would have missed some transaction when it was away and
 677                  * in case of 2a the stale bdev has to be updated as well.
 678                  * 2b must not be allowed at all time.
 679                  */
 680
 681                 /*
 682                  * For now, we do allow update to btrfs_fs_device through the
 683                  * btrfs dev scan cli after FS has been mounted.  We're still
 684                  * tracking a problem where systems fail mount by subvolume id
 685                  * when we reject replacement on a mounted FS.
 686                  */
 687                 if (!fs_devices->opened && found_transid < device->generation) {
 688                         /*
 689                          * That is if the FS is _not_ mounted and if you
 690                          * are here, that means there is more than one
 691                          * disk with same uuid and devid.We keep the one
 692                          * with larger generation number or the last-in if
 693                          * generation are equal.
 694                          */
 695                         return -EEXIST;
 696                 }
 697
 698                 name = rcu_string_strdup(path, GFP_NOFS);
 699                 if (!name)
 700                         return -ENOMEM;
 701                 rcu_string_free(device->name);
 702                 rcu_assign_pointer(device->name, name);
 703                 if (device->missing) {
 704                         fs_devices->missing_devices--;
 705                         device->missing = 0;
 706                 }
 707         }
 708
 709         /*
 710          * Unmount does not free the btrfs_device struct but would zero
 711          * generation along with most of the other members. So just update
 712          * it back. We need it to pick the disk with largest generation
 713          * (as above).
 714          */
 715         if (!fs_devices->opened)
 716                 device->generation = found_transid;
 717
 718         /*
 719          * if there is new btrfs on an already registered device,
 720          * then remove the stale device entry.
 721          */
 722         if (ret > 0)
 723                 btrfs_free_stale_device(device);
 724
 725         *fs_devices_ret = fs_devices;
 726
 727         return ret;
 728 }
 729
 730 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
 731 {
 732         struct btrfs_fs_devices *fs_devices;
 733         struct btrfs_device *device;
 734         struct btrfs_device *orig_dev;
 735
 736         fs_devices = alloc_fs_devices(orig->fsid);
 737         if (IS_ERR(fs_devices))
 738                 return fs_devices;
 739
 740         mutex_lock(&orig->device_list_mutex);
 741         fs_devices->total_devices = orig->total_devices;
 742
 743         /* We have held the volume lock, it is safe to get the devices. */
 744         list_for_each_entry(orig_dev, &orig->devices, dev_list) {
 745                 struct rcu_string *name;
 746
 747                 device = btrfs_alloc_device(NULL, &orig_dev->devid,
 748                                             orig_dev->uuid);
 749                 if (IS_ERR(device))
 750                         goto error;
 751
 752                 /*
 753                  * This is ok to do without rcu read locked because we hold the
 754                  * uuid mutex so nothing we touch in here is going to disappear.
 755                  */
 756                 if (orig_dev->name) {
 757                         name = rcu_string_strdup(orig_dev->name->str,
 758                                         GFP_KERNEL);
 759                         if (!name) {
 760                                 kfree(device);
 761                                 goto error;
 762                         }
 763                         rcu_assign_pointer(device->name, name);
 764                 }
 765
 766                 list_add(&device->dev_list, &fs_devices->devices);
 767                 device->fs_devices = fs_devices;
 768                 fs_devices->num_devices++;
 769         }
 770         mutex_unlock(&orig->device_list_mutex);
 771         return fs_devices;
 772 error:
 773         mutex_unlock(&orig->device_list_mutex);
 774         free_fs_devices(fs_devices);
 775         return ERR_PTR(-ENOMEM);
 776 }
 777
 778 void btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices, int step)
 779 {
 780         struct btrfs_device *device, *next;
 781         struct btrfs_device *latest_dev = NULL;
 782
 783         mutex_lock(&uuid_mutex);
 784 again:
 785         /* This is the initialized path, it is safe to release the devices. */
 786         list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
 787                 if (device->in_fs_metadata) {
 788                         if (!device->is_tgtdev_for_dev_replace &&
 789                             (!latest_dev ||
 790                              device->generation > latest_dev->generation)) {
 791                                 latest_dev = device;
 792                         }
 793                         continue;
 794                 }
 795
 796                 if (device->devid == BTRFS_DEV_REPLACE_DEVID) {
 797                         /*
 798                          * In the first step, keep the device which has
 799                          * the correct fsid and the devid that is used
 800                          * for the dev_replace procedure.
 801                          * In the second step, the dev_replace state is
 802                          * read from the device tree and it is known
 803                          * whether the procedure is really active or
 804                          * not, which means whether this device is
 805                          * used or whether it should be removed.
 806                          */
 807                         if (step == 0 || device->is_tgtdev_for_dev_replace) {
 808                                 continue;
 809                         }
 810                 }
 811                 if (device->bdev) {
 812                         blkdev_put(device->bdev, device->mode);
 813                         device->bdev = NULL;
 814                         fs_devices->open_devices--;
 815                 }
 816                 if (device->writeable) {
 817                         list_del_init(&device->dev_alloc_list);
 818                         device->writeable = 0;
 819                         if (!device->is_tgtdev_for_dev_replace)
 820                                 fs_devices->rw_devices--;
 821                 }
 822                 list_del_init(&device->dev_list);
 823                 fs_devices->num_devices--;
 824                 rcu_string_free(device->name);
 825                 kfree(device);
 826         }
 827
 828         if (fs_devices->seed) {
 829                 fs_devices = fs_devices->seed;
 830                 goto again;
 831         }
 832
 833         fs_devices->latest_bdev = latest_dev->bdev;
 834
 835         mutex_unlock(&uuid_mutex);
 836 }
 837
 838 static void __free_device(struct work_struct *work)
 839 {
 840         struct btrfs_device *device;
 841
 842         device = container_of(work, struct btrfs_device, rcu_work);
 843         rcu_string_free(device->name);
 844         bio_put(device->flush_bio);
 845         kfree(device);
 846 }
 847
 848 static void free_device(struct rcu_head *head)
 849 {
 850         struct btrfs_device *device;
 851
 852         device = container_of(head, struct btrfs_device, rcu);
 853
 854         INIT_WORK(&device->rcu_work, __free_device);
 855         schedule_work(&device->rcu_work);
 856 }
 857
 858 static void btrfs_close_bdev(struct btrfs_device *device)
 859 {
 860         if (device->bdev && device->writeable) {
 861                 sync_blockdev(device->bdev);
 862                 invalidate_bdev(device->bdev);
 863         }
 864
 865         if (device->bdev)
 866                 blkdev_put(device->bdev, device->mode);
 867 }
 868
 869 static void btrfs_prepare_close_one_device(struct btrfs_device *device)
 870 {
 871         struct btrfs_fs_devices *fs_devices = device->fs_devices;
 872         struct btrfs_device *new_device;
 873         struct rcu_string *name;
 874
 875         if (device->bdev)
 876                 fs_devices->open_devices--;
 877
 878         if (device->writeable &&
 879             device->devid != BTRFS_DEV_REPLACE_DEVID) {
 880                 list_del_init(&device->dev_alloc_list);
 881                 fs_devices->rw_devices--;
 882         }
 883
 884         if (device->missing)
 885                 fs_devices->missing_devices--;
 886
 887         new_device = btrfs_alloc_device(NULL, &device->devid,
 888                                         device->uuid);
 889         BUG_ON(IS_ERR(new_device)); /* -ENOMEM */
 890
 891         /* Safe because we are under uuid_mutex */
 892         if (device->name) {
 893                 name = rcu_string_strdup(device->name->str, GFP_NOFS);
 894                 BUG_ON(!name); /* -ENOMEM */
 895                 rcu_assign_pointer(new_device->name, name);
 896         }
 897
 898         list_replace_rcu(&device->dev_list, &new_device->dev_list);
 899         new_device->fs_devices = device->fs_devices;
 900 }
 901
 902 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
 903 {
 904         struct btrfs_device *device, *tmp;
 905         struct list_head pending_put;
 906
 907         INIT_LIST_HEAD(&pending_put);
 908
 909         if (--fs_devices->opened > 0)
 910                 return 0;
 911
 912         mutex_lock(&fs_devices->device_list_mutex);
 913         list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list) {
 914                 btrfs_prepare_close_one_device(device);
 915                 list_add(&device->dev_list, &pending_put);
 916         }
 917         mutex_unlock(&fs_devices->device_list_mutex);
 918
 919         /*
 920          * btrfs_show_devname() is using the device_list_mutex,
 921          * sometimes call to blkdev_put() leads vfs calling
 922          * into this func. So do put outside of device_list_mutex,
 923          * as of now.
 924          */
 925         while (!list_empty(&pending_put)) {
 926                 device = list_first_entry(&pending_put,
 927                                 struct btrfs_device, dev_list);
 928                 list_del(&device->dev_list);
 929                 btrfs_close_bdev(device);
 930                 call_rcu(&device->rcu, free_device);
 931         }
 932
 933         WARN_ON(fs_devices->open_devices);
 934         WARN_ON(fs_devices->rw_devices);
 935         fs_devices->opened = 0;
 936         fs_devices->seeding = 0;
 937
 938         return 0;
 939 }
 940
 941 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
 942 {
 943         struct btrfs_fs_devices *seed_devices = NULL;
 944         int ret;
 945
 946         mutex_lock(&uuid_mutex);
 947         ret = __btrfs_close_devices(fs_devices);
 948         if (!fs_devices->opened) {
 949                 seed_devices = fs_devices->seed;
 950                 fs_devices->seed = NULL;
 951         }
 952         mutex_unlock(&uuid_mutex);
 953
 954         while (seed_devices) {
 955                 fs_devices = seed_devices;
 956                 seed_devices = fs_devices->seed;
 957                 __btrfs_close_devices(fs_devices);
 958                 free_fs_devices(fs_devices);
 959         }
 960         /*
 961          * Wait for rcu kworkers under __btrfs_close_devices
 962          * to finish all blkdev_puts so device is really
 963          * free when umount is done.
 964          */
 965         rcu_barrier();
 966         return ret;
 967 }
 968
 969 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
 970                                 fmode_t flags, void *holder)
 971 {
 972         struct request_queue *q;
 973         struct block_device *bdev;
 974         struct list_head *head = &fs_devices->devices;
 975         struct btrfs_device *device;
 976         struct btrfs_device *latest_dev = NULL;
 977         struct buffer_head *bh;
 978         struct btrfs_super_block *disk_super;
 979         u64 devid;
 980         int seeding = 1;
 981         int ret = 0;
 982
 983         flags |= FMODE_EXCL;
 984
 985         list_for_each_entry(device, head, dev_list) {
 986                 if (device->bdev)
 987                         continue;
 988                 if (!device->name)
 989                         continue;
 990
 991                 /* Just open everything we can; ignore failures here */
 992                 if (btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1,
 993                                             &bdev, &bh))
 994                         continue;
 995
 996                 disk_super = (struct btrfs_super_block *)bh->b_data;
 997                 devid = btrfs_stack_device_id(&disk_super->dev_item);
 998                 if (devid != device->devid)
 999                         goto error_brelse;
1000
1001                 if (memcmp(device->uuid, disk_super->dev_item.uuid,
1002                            BTRFS_UUID_SIZE))
1003                         goto error_brelse;
1004
1005                 device->generation = btrfs_super_generation(disk_super);
1006                 if (!latest_dev ||
1007                     device->generation > latest_dev->generation)
1008                         latest_dev = device;
1009
1010                 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
1011                         device->writeable = 0;
1012                 } else {
1013                         device->writeable = !bdev_read_only(bdev);
1014                         seeding = 0;
1015                 }
1016
1017                 q = bdev_get_queue(bdev);
1018                 if (blk_queue_discard(q))
1019                         device->can_discard = 1;
1020                 if (!blk_queue_nonrot(q))
1021                         fs_devices->rotating = 1;
1022
1023                 device->bdev = bdev;
1024                 device->in_fs_metadata = 0;
1025                 device->mode = flags;
1026
1027                 fs_devices->open_devices++;
1028                 if (device->writeable &&
1029                     device->devid != BTRFS_DEV_REPLACE_DEVID) {
1030                         fs_devices->rw_devices++;
1031                         list_add(&device->dev_alloc_list,
1032                                  &fs_devices->alloc_list);
1033                 }
1034                 brelse(bh);
1035                 continue;
1036
1037 error_brelse:
1038                 brelse(bh);
1039                 blkdev_put(bdev, flags);
1040                 continue;
1041         }
1042         if (fs_devices->open_devices == 0) {
1043                 ret = -EINVAL;
1044                 goto out;
1045         }
1046         fs_devices->seeding = seeding;
1047         fs_devices->opened = 1;
1048         fs_devices->latest_bdev = latest_dev->bdev;
1049         fs_devices->total_rw_bytes = 0;
1050 out:
1051         return ret;
1052 }
1053
1054 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
1055                        fmode_t flags, void *holder)
1056 {
1057         int ret;
1058
1059         mutex_lock(&uuid_mutex);
1060         if (fs_devices->opened) {
1061                 fs_devices->opened++;
1062                 ret = 0;
1063         } else {
1064                 ret = __btrfs_open_devices(fs_devices, flags, holder);
1065         }
1066         mutex_unlock(&uuid_mutex);
1067         return ret;
1068 }
1069
1070 void btrfs_release_disk_super(struct page *page)
1071 {
1072         kunmap(page);
1073         put_page(page);
1074 }
1075
1076 int btrfs_read_disk_super(struct block_device *bdev, u64 bytenr,
1077                 struct page **page, struct btrfs_super_block **disk_super)
1078 {
1079         void *p;
1080         pgoff_t index;
1081
1082         /* make sure our super fits in the device */
1083         if (bytenr + PAGE_SIZE >= i_size_read(bdev->bd_inode))
1084                 return 1;
1085
1086         /* make sure our super fits in the page */
1087         if (sizeof(**disk_super) > PAGE_SIZE)
1088                 return 1;
1089
1090         /* make sure our super doesn't straddle pages on disk */
1091         index = bytenr >> PAGE_SHIFT;
1092         if ((bytenr + sizeof(**disk_super) - 1) >> PAGE_SHIFT != index)
1093                 return 1;
1094
1095         /* pull in the page with our super */
1096         *page = read_cache_page_gfp(bdev->bd_inode->i_mapping,
1097                                    index, GFP_KERNEL);
1098
1099         if (IS_ERR_OR_NULL(*page))
1100                 return 1;
1101
1102         p = kmap(*page);
1103
1104         /* align our pointer to the offset of the super block */
1105         *disk_super = p + (bytenr & ~PAGE_MASK);
1106
1107         if (btrfs_super_bytenr(*disk_super) != bytenr ||
1108             btrfs_super_magic(*disk_super) != BTRFS_MAGIC) {
1109                 btrfs_release_disk_super(*page);
1110                 return 1;
1111         }
1112
1113         if ((*disk_super)->label[0] &&
1114                 (*disk_super)->label[BTRFS_LABEL_SIZE - 1])
1115                 (*disk_super)->label[BTRFS_LABEL_SIZE - 1] = '\0';
1116
1117         return 0;
1118 }
1119
1120 /*
1121  * Look for a btrfs signature on a device. This may be called out of the mount path
1122  * and we are not allowed to call set_blocksize during the scan. The superblock
1123  * is read via pagecache
1124  */
1125 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
1126                           struct btrfs_fs_devices **fs_devices_ret)
1127 {
1128         struct btrfs_super_block *disk_super;
1129         struct block_device *bdev;
1130         struct page *page;
1131         int ret = -EINVAL;
1132         u64 devid;
1133         u64 transid;
1134         u64 total_devices;
1135         u64 bytenr;
1136
1137         /*
1138          * we would like to check all the supers, but that would make
1139          * a btrfs mount succeed after a mkfs from a different FS.
1140          * So, we need to add a special mount option to scan for
1141          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
1142          */
1143         bytenr = btrfs_sb_offset(0);
1144         flags |= FMODE_EXCL;
1145         mutex_lock(&uuid_mutex);
1146
1147         bdev = blkdev_get_by_path(path, flags, holder);
1148         if (IS_ERR(bdev)) {
1149                 ret = PTR_ERR(bdev);
1150                 goto error;
1151         }
1152
1153         if (btrfs_read_disk_super(bdev, bytenr, &page, &disk_super))
1154                 goto error_bdev_put;
1155
1156         devid = btrfs_stack_device_id(&disk_super->dev_item);
1157         transid = btrfs_super_generation(disk_super);
1158         total_devices = btrfs_super_num_devices(disk_super);
1159
1160         ret = device_list_add(path, disk_super, devid, fs_devices_ret);
1161         if (ret > 0) {
1162                 if (disk_super->label[0]) {
1163                         pr_info("BTRFS: device label %s ", disk_super->label);
1164                 } else {
1165                         pr_info("BTRFS: device fsid %pU ", disk_super->fsid);
1166                 }
1167
1168                 pr_cont("devid %llu transid %llu %s\n", devid, transid, path);
1169                 ret = 0;
1170         }
1171         if (!ret && fs_devices_ret)
1172                 (*fs_devices_ret)->total_devices = total_devices;
1173
1174         btrfs_release_disk_super(page);
1175
1176 error_bdev_put:
1177         blkdev_put(bdev, flags);
1178 error:
1179         mutex_unlock(&uuid_mutex);
1180         return ret;
1181 }
1182
1183 /* helper to account the used device space in the range */
1184 int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
1185                                    u64 end, u64 *length)
1186 {
1187         struct btrfs_key key;
1188         struct btrfs_root *root = device->fs_info->dev_root;
1189         struct btrfs_dev_extent *dev_extent;
1190         struct btrfs_path *path;
1191         u64 extent_end;
1192         int ret;
1193         int slot;
1194         struct extent_buffer *l;
1195
1196         *length = 0;
1197
1198         if (start >= device->total_bytes || device->is_tgtdev_for_dev_replace)
1199                 return 0;
1200
1201         path = btrfs_alloc_path();
1202         if (!path)
1203                 return -ENOMEM;
1204         path->reada = READA_FORWARD;
1205
1206         key.objectid = device->devid;
1207         key.offset = start;
1208         key.type = BTRFS_DEV_EXTENT_KEY;
1209
1210         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1211         if (ret < 0)
1212                 goto out;
1213         if (ret > 0) {
1214                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1215                 if (ret < 0)
1216                         goto out;
1217         }
1218
1219         while (1) {
1220                 l = path->nodes[0];
1221                 slot = path->slots[0];
1222                 if (slot >= btrfs_header_nritems(l)) {
1223                         ret = btrfs_next_leaf(root, path);
1224                         if (ret == 0)
1225                                 continue;
1226                         if (ret < 0)
1227                                 goto out;
1228
1229                         break;
1230                 }
1231                 btrfs_item_key_to_cpu(l, &key, slot);
1232
1233                 if (key.objectid < device->devid)
1234                         goto next;
1235
1236                 if (key.objectid > device->devid)
1237                         break;
1238
1239                 if (key.type != BTRFS_DEV_EXTENT_KEY)
1240                         goto next;
1241
1242                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1243                 extent_end = key.offset + btrfs_dev_extent_length(l,
1244                                                                   dev_extent);
1245                 if (key.offset <= start && extent_end > end) {
1246                         *length = end - start + 1;
1247                         break;
1248                 } else if (key.offset <= start && extent_end > start)
1249                         *length += extent_end - start;
1250                 else if (key.offset > start && extent_end <= end)
1251                         *length += extent_end - key.offset;
1252                 else if (key.offset > start && key.offset <= end) {
1253                         *length += end - key.offset + 1;
1254                         break;
1255                 } else if (key.offset > end)
1256                         break;
1257
1258 next:
1259                 path->slots[0]++;
1260         }
1261         ret = 0;
1262 out:
1263         btrfs_free_path(path);
1264         return ret;
1265 }
1266
1267 static int contains_pending_extent(struct btrfs_transaction *transaction,
1268                                    struct btrfs_device *device,
1269                                    u64 *start, u64 len)
1270 {
1271         struct btrfs_fs_info *fs_info = device->fs_info;
1272         struct extent_map *em;
1273         struct list_head *search_list = &fs_info->pinned_chunks;
1274         int ret = 0;
1275         u64 physical_start = *start;
1276
1277         if (transaction)
1278                 search_list = &transaction->pending_chunks;
1279 again:
1280         list_for_each_entry(em, search_list, list) {
1281                 struct map_lookup *map;
1282                 int i;
1283
1284                 map = em->map_lookup;
1285                 for (i = 0; i < map->num_stripes; i++) {
1286                         u64 end;
1287
1288                         if (map->stripes[i].dev != device)
1289                                 continue;
1290                         if (map->stripes[i].physical >= physical_start + len ||
1291                             map->stripes[i].physical + em->orig_block_len <=
1292                             physical_start)
1293                                 continue;
1294                         /*
1295                          * Make sure that while processing the pinned list we do
1296                          * not override our *start with a lower value, because
1297                          * we can have pinned chunks that fall within this
1298                          * device hole and that have lower physical addresses
1299                          * than the pending chunks we processed before. If we
1300                          * do not take this special care we can end up getting
1301                          * 2 pending chunks that start at the same physical
1302                          * device offsets because the end offset of a pinned
1303                          * chunk can be equal to the start offset of some
1304                          * pending chunk.
1305                          */
1306                         end = map->stripes[i].physical + em->orig_block_len;
1307                         if (end > *start) {
1308                                 *start = end;
1309                                 ret = 1;
1310                         }
1311                 }
1312         }
1313         if (search_list != &fs_info->pinned_chunks) {
1314                 search_list = &fs_info->pinned_chunks;
1315                 goto again;
1316         }
1317
1318         return ret;
1319 }
1320
1321
1322 /*
1323  * find_free_dev_extent_start - find free space in the specified device
1324  * @device:       the device which we search the free space in
1325  * @num_bytes:    the size of the free space that we need
1326  * @search_start: the position from which to begin the search
1327  * @start:        store the start of the free space.
1328  * @len:          the size of the free space. that we find, or the size
1329  *                of the max free space if we don't find suitable free space
1330  *
1331  * this uses a pretty simple search, the expectation is that it is
1332  * called very infrequently and that a given device has a small number
1333  * of extents
1334  *
1335  * @start is used to store the start of the free space if we find. But if we
1336  * don't find suitable free space, it will be used to store the start position
1337  * of the max free space.
1338  *
1339  * @len is used to store the size of the free space that we find.
1340  * But if we don't find suitable free space, it is used to store the size of
1341  * the max free space.
1342  */
1343 int find_free_dev_extent_start(struct btrfs_transaction *transaction,
1344                                struct btrfs_device *device, u64 num_bytes,
1345                                u64 search_start, u64 *start, u64 *len)
1346 {
1347         struct btrfs_fs_info *fs_info = device->fs_info;
1348         struct btrfs_root *root = fs_info->dev_root;
1349         struct btrfs_key key;
1350         struct btrfs_dev_extent *dev_extent;
1351         struct btrfs_path *path;
1352         u64 hole_size;
1353         u64 max_hole_start;
1354         u64 max_hole_size;
1355         u64 extent_end;
1356         u64 search_end = device->total_bytes;
1357         int ret;
1358         int slot;
1359         struct extent_buffer *l;
1360
1361         /*
1362          * We don't want to overwrite the superblock on the drive nor any area
1363          * used by the boot loader (grub for example), so we make sure to start
1364          * at an offset of at least 1MB.
1365          */
1366         search_start = max_t(u64, search_start, SZ_1M);
1367
1368         path = btrfs_alloc_path();
1369         if (!path)
1370                 return -ENOMEM;
1371
1372         max_hole_start = search_start;
1373         max_hole_size = 0;
1374
1375 again:
1376         if (search_start >= search_end || device->is_tgtdev_for_dev_replace) {
1377                 ret = -ENOSPC;
1378                 goto out;
1379         }
1380
1381         path->reada = READA_FORWARD;
1382         path->search_commit_root = 1;
1383         path->skip_locking = 1;
1384
1385         key.objectid = device->devid;
1386         key.offset = search_start;
1387         key.type = BTRFS_DEV_EXTENT_KEY;
1388
1389         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1390         if (ret < 0)
1391                 goto out;
1392         if (ret > 0) {
1393                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1394                 if (ret < 0)
1395                         goto out;
1396         }
1397
1398         while (1) {
1399                 l = path->nodes[0];
1400                 slot = path->slots[0];
1401                 if (slot >= btrfs_header_nritems(l)) {
1402                         ret = btrfs_next_leaf(root, path);
1403                         if (ret == 0)
1404                                 continue;
1405                         if (ret < 0)
1406                                 goto out;
1407
1408                         break;
1409                 }
1410                 btrfs_item_key_to_cpu(l, &key, slot);
1411
1412                 if (key.objectid < device->devid)
1413                         goto next;
1414
1415                 if (key.objectid > device->devid)
1416                         break;
1417
1418                 if (key.type != BTRFS_DEV_EXTENT_KEY)
1419                         goto next;
1420
1421                 if (key.offset > search_start) {
1422                         hole_size = key.offset - search_start;
1423
1424                         /*
1425                          * Have to check before we set max_hole_start, otherwise
1426                          * we could end up sending back this offset anyway.
1427                          */
1428                         if (contains_pending_extent(transaction, device,
1429                                                     &search_start,
1430                                                     hole_size)) {
1431                                 if (key.offset >= search_start) {
1432                                         hole_size = key.offset - search_start;
1433                                 } else {
1434                                         WARN_ON_ONCE(1);
1435                                         hole_size = 0;
1436                                 }
1437                         }
1438
1439                         if (hole_size > max_hole_size) {
1440                                 max_hole_start = search_start;
1441                                 max_hole_size = hole_size;
1442                         }
1443
1444                         /*
1445                          * If this free space is greater than which we need,
1446                          * it must be the max free space that we have found
1447                          * until now, so max_hole_start must point to the start
1448                          * of this free space and the length of this free space
1449                          * is stored in max_hole_size. Thus, we return
1450                          * max_hole_start and max_hole_size and go back to the
1451                          * caller.
1452                          */
1453                         if (hole_size >= num_bytes) {
1454                                 ret = 0;
1455                                 goto out;
1456                         }
1457                 }
1458
1459                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1460                 extent_end = key.offset + btrfs_dev_extent_length(l,
1461                                                                   dev_extent);
1462                 if (extent_end > search_start)
1463                         search_start = extent_end;
1464 next:
1465                 path->slots[0]++;
1466                 cond_resched();
1467         }
1468
1469         /*
1470          * At this point, search_start should be the end of
1471          * allocated dev extents, and when shrinking the device,
1472          * search_end may be smaller than search_start.
1473          */
1474         if (search_end > search_start) {
1475                 hole_size = search_end - search_start;
1476
1477                 if (contains_pending_extent(transaction, device, &search_start,
1478                                             hole_size)) {
1479                         btrfs_release_path(path);
1480                         goto again;
1481                 }
1482
1483                 if (hole_size > max_hole_size) {
1484                         max_hole_start = search_start;
1485                         max_hole_size = hole_size;
1486                 }
1487         }
1488
1489         /* See above. */
1490         if (max_hole_size < num_bytes)
1491                 ret = -ENOSPC;
1492         else
1493                 ret = 0;
1494
1495 out:
1496         btrfs_free_path(path);
1497         *start = max_hole_start;
1498         if (len)
1499                 *len = max_hole_size;
1500         return ret;
1501 }
1502
1503 int find_free_dev_extent(struct btrfs_trans_handle *trans,
1504                          struct btrfs_device *device, u64 num_bytes,
1505                          u64 *start, u64 *len)
1506 {
1507         /* FIXME use last free of some kind */
1508         return find_free_dev_extent_start(trans->transaction, device,
1509                                           num_bytes, 0, start, len);
1510 }
1511
1512 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1513                           struct btrfs_device *device,
1514                           u64 start, u64 *dev_extent_len)
1515 {
1516         struct btrfs_fs_info *fs_info = device->fs_info;
1517         struct btrfs_root *root = fs_info->dev_root;
1518         int ret;
1519         struct btrfs_path *path;
1520         struct btrfs_key key;
1521         struct btrfs_key found_key;
1522         struct extent_buffer *leaf = NULL;
1523         struct btrfs_dev_extent *extent = NULL;
1524
1525         path = btrfs_alloc_path();
1526         if (!path)
1527                 return -ENOMEM;
1528
1529         key.objectid = device->devid;
1530         key.offset = start;
1531         key.type = BTRFS_DEV_EXTENT_KEY;
1532 again:
1533         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1534         if (ret > 0) {
1535                 ret = btrfs_previous_item(root, path, key.objectid,
1536                                           BTRFS_DEV_EXTENT_KEY);
1537                 if (ret)
1538                         goto out;
1539                 leaf = path->nodes[0];
1540                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1541                 extent = btrfs_item_ptr(leaf, path->slots[0],
1542                                         struct btrfs_dev_extent);
1543                 BUG_ON(found_key.offset > start || found_key.offset +
1544                        btrfs_dev_extent_length(leaf, extent) < start);
1545                 key = found_key;
1546                 btrfs_release_path(path);
1547                 goto again;
1548         } else if (ret == 0) {
1549                 leaf = path->nodes[0];
1550                 extent = btrfs_item_ptr(leaf, path->slots[0],
1551                                         struct btrfs_dev_extent);
1552         } else {
1553                 btrfs_handle_fs_error(fs_info, ret, "Slot search failed");
1554                 goto out;
1555         }
1556
1557         *dev_extent_len = btrfs_dev_extent_length(leaf, extent);
1558
1559         ret = btrfs_del_item(trans, root, path);
1560         if (ret) {
1561                 btrfs_handle_fs_error(fs_info, ret,
1562                                       "Failed to remove dev extent item");
1563         } else {
1564                 set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags);
1565         }
1566 out:
1567         btrfs_free_path(path);
1568         return ret;
1569 }
1570
1571 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1572                                   struct btrfs_device *device,
1573                                   u64 chunk_offset, u64 start, u64 num_bytes)
1574 {
1575         int ret;
1576         struct btrfs_path *path;
1577         struct btrfs_fs_info *fs_info = device->fs_info;
1578         struct btrfs_root *root = fs_info->dev_root;
1579         struct btrfs_dev_extent *extent;
1580         struct extent_buffer *leaf;
1581         struct btrfs_key key;
1582
1583         WARN_ON(!device->in_fs_metadata);
1584         WARN_ON(device->is_tgtdev_for_dev_replace);
1585         path = btrfs_alloc_path();
1586         if (!path)
1587                 return -ENOMEM;
1588
1589         key.objectid = device->devid;
1590         key.offset = start;
1591         key.type = BTRFS_DEV_EXTENT_KEY;
1592         ret = btrfs_insert_empty_item(trans, root, path, &key,
1593                                       sizeof(*extent));
1594         if (ret)
1595                 goto out;
1596
1597         leaf = path->nodes[0];
1598         extent = btrfs_item_ptr(leaf, path->slots[0],
1599                                 struct btrfs_dev_extent);
1600         btrfs_set_dev_extent_chunk_tree(leaf, extent,
1601                                         BTRFS_CHUNK_TREE_OBJECTID);
1602         btrfs_set_dev_extent_chunk_objectid(leaf, extent,
1603                                             BTRFS_FIRST_CHUNK_TREE_OBJECTID);
1604         btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1605
1606         btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1607         btrfs_mark_buffer_dirty(leaf);
1608 out:
1609         btrfs_free_path(path);
1610         return ret;
1611 }
1612
1613 static u64 find_next_chunk(struct btrfs_fs_info *fs_info)
1614 {
1615         struct extent_map_tree *em_tree;
1616         struct extent_map *em;
1617         struct rb_node *n;
1618         u64 ret = 0;
1619
1620         em_tree = &fs_info->mapping_tree.map_tree;
1621         read_lock(&em_tree->lock);
1622         n = rb_last(&em_tree->map);
1623         if (n) {
1624                 em = rb_entry(n, struct extent_map, rb_node);
1625                 ret = em->start + em->len;
1626         }
1627         read_unlock(&em_tree->lock);
1628
1629         return ret;
1630 }
1631
1632 static noinline int find_next_devid(struct btrfs_fs_info *fs_info,
1633                                     u64 *devid_ret)
1634 {
1635         int ret;
1636         struct btrfs_key key;
1637         struct btrfs_key found_key;
1638         struct btrfs_path *path;
1639
1640         path = btrfs_alloc_path();
1641         if (!path)
1642                 return -ENOMEM;
1643
1644         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1645         key.type = BTRFS_DEV_ITEM_KEY;
1646         key.offset = (u64)-1;
1647
1648         ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
1649         if (ret < 0)
1650                 goto error;
1651
1652         BUG_ON(ret == 0); /* Corruption */
1653
1654         ret = btrfs_previous_item(fs_info->chunk_root, path,
1655                                   BTRFS_DEV_ITEMS_OBJECTID,
1656                                   BTRFS_DEV_ITEM_KEY);
1657         if (ret) {
1658                 *devid_ret = 1;
1659         } else {
1660                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1661                                       path->slots[0]);
1662                 *devid_ret = found_key.offset + 1;
1663         }
1664         ret = 0;
1665 error:
1666         btrfs_free_path(path);
1667         return ret;
1668 }
1669
1670 /*
1671  * the device information is stored in the chunk root
1672  * the btrfs_device struct should be fully filled in
1673  */
1674 static int btrfs_add_device(struct btrfs_trans_handle *trans,
1675                             struct btrfs_fs_info *fs_info,
1676                             struct btrfs_device *device)
1677 {
1678         struct btrfs_root *root = fs_info->chunk_root;
1679         int ret;
1680         struct btrfs_path *path;
1681         struct btrfs_dev_item *dev_item;
1682         struct extent_buffer *leaf;
1683         struct btrfs_key key;
1684         unsigned long ptr;
1685
1686         path = btrfs_alloc_path();
1687         if (!path)
1688                 return -ENOMEM;
1689
1690         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1691         key.type = BTRFS_DEV_ITEM_KEY;
1692         key.offset = device->devid;
1693
1694         ret = btrfs_insert_empty_item(trans, root, path, &key,
1695                                       sizeof(*dev_item));
1696         if (ret)
1697                 goto out;
1698
1699         leaf = path->nodes[0];
1700         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1701
1702         btrfs_set_device_id(leaf, dev_item, device->devid);
1703         btrfs_set_device_generation(leaf, dev_item, 0);
1704         btrfs_set_device_type(leaf, dev_item, device->type);
1705         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1706         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1707         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1708         btrfs_set_device_total_bytes(leaf, dev_item,
1709                                      btrfs_device_get_disk_total_bytes(device));
1710         btrfs_set_device_bytes_used(leaf, dev_item,
1711                                     btrfs_device_get_bytes_used(device));
1712         btrfs_set_device_group(leaf, dev_item, 0);
1713         btrfs_set_device_seek_speed(leaf, dev_item, 0);
1714         btrfs_set_device_bandwidth(leaf, dev_item, 0);
1715         btrfs_set_device_start_offset(leaf, dev_item, 0);
1716
1717         ptr = btrfs_device_uuid(dev_item);
1718         write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1719         ptr = btrfs_device_fsid(dev_item);
1720         write_extent_buffer(leaf, fs_info->fsid, ptr, BTRFS_FSID_SIZE);
1721         btrfs_mark_buffer_dirty(leaf);
1722
1723         ret = 0;
1724 out:
1725         btrfs_free_path(path);
1726         return ret;
1727 }
1728
1729 /*
1730  * Function to update ctime/mtime for a given device path.
1731  * Mainly used for ctime/mtime based probe like libblkid.
1732  */
1733 static void update_dev_time(const char *path_name)
1734 {
1735         struct file *filp;
1736
1737         filp = filp_open(path_name, O_RDWR, 0);
1738         if (IS_ERR(filp))
1739                 return;
1740         file_update_time(filp);
1741         filp_close(filp, NULL);
1742 }
1743
1744 static int btrfs_rm_dev_item(struct btrfs_fs_info *fs_info,
1745                              struct btrfs_device *device)
1746 {
1747         struct btrfs_root *root = fs_info->chunk_root;
1748         int ret;
1749         struct btrfs_path *path;
1750         struct btrfs_key key;
1751         struct btrfs_trans_handle *trans;
1752
1753         path = btrfs_alloc_path();
1754         if (!path)
1755                 return -ENOMEM;
1756
1757         trans = btrfs_start_transaction(root, 0);
1758         if (IS_ERR(trans)) {
1759                 btrfs_free_path(path);
1760                 return PTR_ERR(trans);
1761         }
1762         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1763         key.type = BTRFS_DEV_ITEM_KEY;
1764         key.offset = device->devid;
1765
1766         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1767         if (ret) {
1768                 if (ret > 0)
1769                         ret = -ENOENT;
1770                 btrfs_abort_transaction(trans, ret);
1771                 btrfs_end_transaction(trans);
1772                 goto out;
1773         }
1774
1775         ret = btrfs_del_item(trans, root, path);
1776         if (ret) {
1777                 btrfs_abort_transaction(trans, ret);
1778                 btrfs_end_transaction(trans);
1779         }
1780
1781 out:
1782         btrfs_free_path(path);
1783         if (!ret)
1784                 ret = btrfs_commit_transaction(trans);
1785         return ret;
1786 }
1787
1788 /*
1789  * Verify that @num_devices satisfies the RAID profile constraints in the whole
1790  * filesystem. It's up to the caller to adjust that number regarding eg. device
1791  * replace.
1792  */
1793 static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info,
1794                 u64 num_devices)
1795 {
1796         u64 all_avail;
1797         unsigned seq;
1798         int i;
1799
1800         do {
1801                 seq = read_seqbegin(&fs_info->profiles_lock);
1802
1803                 all_avail = fs_info->avail_data_alloc_bits |
1804                             fs_info->avail_system_alloc_bits |
1805                             fs_info->avail_metadata_alloc_bits;
1806         } while (read_seqretry(&fs_info->profiles_lock, seq));
1807
1808         for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
1809                 if (!(all_avail & btrfs_raid_group[i]))
1810                         continue;
1811
1812                 if (num_devices < btrfs_raid_array[i].devs_min) {
1813                         int ret = btrfs_raid_mindev_error[i];
1814
1815                         if (ret)
1816                                 return ret;
1817                 }
1818         }
1819
1820         return 0;
1821 }
1822
1823 struct btrfs_device *btrfs_find_next_active_device(struct btrfs_fs_devices *fs_devs,
1824                                         struct btrfs_device *device)
1825 {
1826         struct btrfs_device *next_device;
1827
1828         list_for_each_entry(next_device, &fs_devs->devices, dev_list) {
1829                 if (next_device != device &&
1830                         !next_device->missing && next_device->bdev)
1831                         return next_device;
1832         }
1833
1834         return NULL;
1835 }
1836
1837 /*
1838  * Helper function to check if the given device is part of s_bdev / latest_bdev
1839  * and replace it with the provided or the next active device, in the context
1840  * where this function called, there should be always be another device (or
1841  * this_dev) which is active.
1842  */
1843 void btrfs_assign_next_active_device(struct btrfs_fs_info *fs_info,
1844                 struct btrfs_device *device, struct btrfs_device *this_dev)
1845 {
1846         struct btrfs_device *next_device;
1847
1848         if (this_dev)
1849                 next_device = this_dev;
1850         else
1851                 next_device = btrfs_find_next_active_device(fs_info->fs_devices,
1852                                                                 device);
1853         ASSERT(next_device);
1854
1855         if (fs_info->sb->s_bdev &&
1856                         (fs_info->sb->s_bdev == device->bdev))
1857                 fs_info->sb->s_bdev = next_device->bdev;
1858
1859         if (fs_info->fs_devices->latest_bdev == device->bdev)
1860                 fs_info->fs_devices->latest_bdev = next_device->bdev;
1861 }
1862
1863 int btrfs_rm_device(struct btrfs_fs_info *fs_info, const char *device_path,
1864                 u64 devid)
1865 {
1866         struct btrfs_device *device;
1867         struct btrfs_fs_devices *cur_devices;
1868         u64 num_devices;
1869         int ret = 0;
1870
1871         mutex_lock(&uuid_mutex);
1872
1873         num_devices = fs_info->fs_devices->num_devices;
1874         btrfs_dev_replace_lock(&fs_info->dev_replace, 0);
1875         if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
1876                 WARN_ON(num_devices < 1);
1877                 num_devices--;
1878         }
1879         btrfs_dev_replace_unlock(&fs_info->dev_replace, 0);
1880
1881         ret = btrfs_check_raid_min_devices(fs_info, num_devices - 1);
1882         if (ret)
1883                 goto out;
1884
1885         ret = btrfs_find_device_by_devspec(fs_info, devid, device_path,
1886                                            &device);
1887         if (ret)
1888                 goto out;
1889
1890         if (device->is_tgtdev_for_dev_replace) {
1891                 ret = BTRFS_ERROR_DEV_TGT_REPLACE;
1892                 goto out;
1893         }
1894
1895         if (device->writeable && fs_info->fs_devices->rw_devices == 1) {
1896                 ret = BTRFS_ERROR_DEV_ONLY_WRITABLE;
1897                 goto out;
1898         }
1899
1900         if (device->writeable) {
1901                 mutex_lock(&fs_info->chunk_mutex);
1902                 list_del_init(&device->dev_alloc_list);
1903                 device->fs_devices->rw_devices--;
1904                 mutex_unlock(&fs_info->chunk_mutex);
1905         }
1906
1907         mutex_unlock(&uuid_mutex);
1908         ret = btrfs_shrink_device(device, 0);
1909         mutex_lock(&uuid_mutex);
1910         if (ret)
1911                 goto error_undo;
1912
1913         /*
1914          * TODO: the superblock still includes this device in its num_devices
1915          * counter although write_all_supers() is not locked out. This
1916          * could give a filesystem state which requires a degraded mount.
1917          */
1918         ret = btrfs_rm_dev_item(fs_info, device);
1919         if (ret)
1920                 goto error_undo;
1921
1922         device->in_fs_metadata = 0;
1923         btrfs_scrub_cancel_dev(fs_info, device);
1924
1925         /*
1926          * the device list mutex makes sure that we don't change
1927          * the device list while someone else is writing out all
1928          * the device supers. Whoever is writing all supers, should
1929          * lock the device list mutex before getting the number of
1930          * devices in the super block (super_copy). Conversely,
1931          * whoever updates the number of devices in the super block
1932          * (super_copy) should hold the device list mutex.
1933          */
1934
1935         cur_devices = device->fs_devices;
1936         mutex_lock(&fs_info->fs_devices->device_list_mutex);
1937         list_del_rcu(&device->dev_list);
1938
1939         device->fs_devices->num_devices--;
1940         device->fs_devices->total_devices--;
1941
1942         if (device->missing)
1943                 device->fs_devices->missing_devices--;
1944
1945         btrfs_assign_next_active_device(fs_info, device, NULL);
1946
1947         if (device->bdev) {
1948                 device->fs_devices->open_devices--;
1949                 /* remove sysfs entry */
1950                 btrfs_sysfs_rm_device_link(fs_info->fs_devices, device);
1951         }
1952
1953         num_devices = btrfs_super_num_devices(fs_info->super_copy) - 1;
1954         btrfs_set_super_num_devices(fs_info->super_copy, num_devices);
1955         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
1956
1957         /*
1958          * at this point, the device is zero sized and detached from
1959          * the devices list.  All that's left is to zero out the old
1960          * supers and free the device.
1961          */
1962         if (device->writeable)
1963                 btrfs_scratch_superblocks(device->bdev, device->name->str);
1964
1965         btrfs_close_bdev(device);
1966         call_rcu(&device->rcu, free_device);
1967
1968         if (cur_devices->open_devices == 0) {
1969                 struct btrfs_fs_devices *fs_devices;
1970                 fs_devices = fs_info->fs_devices;
1971                 while (fs_devices) {
1972                         if (fs_devices->seed == cur_devices) {
1973                                 fs_devices->seed = cur_devices->seed;
1974                                 break;
1975                         }
1976                         fs_devices = fs_devices->seed;
1977                 }
1978                 cur_devices->seed = NULL;
1979                 __btrfs_close_devices(cur_devices);
1980                 free_fs_devices(cur_devices);
1981         }
1982
1983 out:
1984         mutex_unlock(&uuid_mutex);
1985         return ret;
1986
1987 error_undo:
1988         if (device->writeable) {
1989                 mutex_lock(&fs_info->chunk_mutex);
1990                 list_add(&device->dev_alloc_list,
1991                          &fs_info->fs_devices->alloc_list);
1992                 device->fs_devices->rw_devices++;
1993                 mutex_unlock(&fs_info->chunk_mutex);
1994         }
1995         goto out;
1996 }
1997
1998 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_fs_info *fs_info,
1999                                         struct btrfs_device *srcdev)
2000 {
2001         struct btrfs_fs_devices *fs_devices;
2002
2003         WARN_ON(!mutex_is_locked(&fs_info->fs_devices->device_list_mutex));
2004
2005         /*
2006          * in case of fs with no seed, srcdev->fs_devices will point
2007          * to fs_devices of fs_info. However when the dev being replaced is
2008          * a seed dev it will point to the seed's local fs_devices. In short
2009          * srcdev will have its correct fs_devices in both the cases.
2010          */
2011         fs_devices = srcdev->fs_devices;
2012
2013         list_del_rcu(&srcdev->dev_list);
2014         list_del_rcu(&srcdev->dev_alloc_list);
2015         fs_devices->num_devices--;
2016         if (srcdev->missing)
2017                 fs_devices->missing_devices--;
2018
2019         if (srcdev->writeable)
2020                 fs_devices->rw_devices--;
2021
2022         if (srcdev->bdev)
2023                 fs_devices->open_devices--;
2024 }
2025
2026 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_fs_info *fs_info,
2027                                       struct btrfs_device *srcdev)
2028 {
2029         struct btrfs_fs_devices *fs_devices = srcdev->fs_devices;
2030
2031         if (srcdev->writeable) {
2032                 /* zero out the old super if it is writable */
2033                 btrfs_scratch_superblocks(srcdev->bdev, srcdev->name->str);
2034         }
2035
2036         btrfs_close_bdev(srcdev);
2037
2038         call_rcu(&srcdev->rcu, free_device);
2039
2040         /*
2041          * unless fs_devices is seed fs, num_devices shouldn't go
2042          * zero
2043          */
2044         BUG_ON(!fs_devices->num_devices && !fs_devices->seeding);
2045
2046         /* if this is no devs we rather delete the fs_devices */
2047         if (!fs_devices->num_devices) {
2048                 struct btrfs_fs_devices *tmp_fs_devices;
2049
2050                 tmp_fs_devices = fs_info->fs_devices;
2051                 while (tmp_fs_devices) {
2052                         if (tmp_fs_devices->seed == fs_devices) {
2053                                 tmp_fs_devices->seed = fs_devices->seed;
2054                                 break;
2055                         }
2056                         tmp_fs_devices = tmp_fs_devices->seed;
2057                 }
2058                 fs_devices->seed = NULL;
2059                 __btrfs_close_devices(fs_devices);
2060                 free_fs_devices(fs_devices);
2061         }
2062 }
2063
2064 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
2065                                       struct btrfs_device *tgtdev)
2066 {
2067         mutex_lock(&uuid_mutex);
2068         WARN_ON(!tgtdev);
2069         mutex_lock(&fs_info->fs_devices->device_list_mutex);
2070
2071         btrfs_sysfs_rm_device_link(fs_info->fs_devices, tgtdev);
2072
2073         if (tgtdev->bdev)
2074                 fs_info->fs_devices->open_devices--;
2075
2076         fs_info->fs_devices->num_devices--;
2077
2078         btrfs_assign_next_active_device(fs_info, tgtdev, NULL);
2079
2080         list_del_rcu(&tgtdev->dev_list);
2081
2082         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2083         mutex_unlock(&uuid_mutex);
2084
2085         /*
2086          * The update_dev_time() with in btrfs_scratch_superblocks()
2087          * may lead to a call to btrfs_show_devname() which will try
2088          * to hold device_list_mutex. And here this device
2089          * is already out of device list, so we don't have to hold
2090          * the device_list_mutex lock.
2091          */
2092         btrfs_scratch_superblocks(tgtdev->bdev, tgtdev->name->str);
2093
2094         btrfs_close_bdev(tgtdev);
2095         call_rcu(&tgtdev->rcu, free_device);
2096 }
2097
2098 static int btrfs_find_device_by_path(struct btrfs_fs_info *fs_info,
2099                                      const char *device_path,
2100                                      struct btrfs_device **device)
2101 {
2102         int ret = 0;
2103         struct btrfs_super_block *disk_super;
2104         u64 devid;
2105         u8 *dev_uuid;
2106         struct block_device *bdev;
2107         struct buffer_head *bh;
2108
2109         *device = NULL;
2110         ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
2111                                     fs_info->bdev_holder, 0, &bdev, &bh);
2112         if (ret)
2113                 return ret;
2114         disk_super = (struct btrfs_super_block *)bh->b_data;
2115         devid = btrfs_stack_device_id(&disk_super->dev_item);
2116         dev_uuid = disk_super->dev_item.uuid;
2117         *device = btrfs_find_device(fs_info, devid, dev_uuid, disk_super->fsid);
2118         brelse(bh);
2119         if (!*device)
2120                 ret = -ENOENT;
2121         blkdev_put(bdev, FMODE_READ);
2122         return ret;
2123 }
2124
2125 int btrfs_find_device_missing_or_by_path(struct btrfs_fs_info *fs_info,
2126                                          const char *device_path,
2127                                          struct btrfs_device **device)
2128 {
2129         *device = NULL;
2130         if (strcmp(device_path, "missing") == 0) {
2131                 struct list_head *devices;
2132                 struct btrfs_device *tmp;
2133
2134                 devices = &fs_info->fs_devices->devices;
2135                 /*
2136                  * It is safe to read the devices since the volume_mutex
2137                  * is held by the caller.
2138                  */
2139                 list_for_each_entry(tmp, devices, dev_list) {
2140                         if (tmp->in_fs_metadata && !tmp->bdev) {
2141                                 *device = tmp;
2142                                 break;
2143                         }
2144                 }
2145
2146                 if (!*device)
2147                         return BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
2148
2149                 return 0;
2150         } else {
2151                 return btrfs_find_device_by_path(fs_info, device_path, device);
2152         }
2153 }
2154
2155 /*
2156  * Lookup a device given by device id, or the path if the id is 0.
2157  */
2158 int btrfs_find_device_by_devspec(struct btrfs_fs_info *fs_info, u64 devid,
2159                                  const char *devpath,
2160                                  struct btrfs_device **device)
2161 {
2162         int ret;
2163
2164         if (devid) {
2165                 ret = 0;
2166                 *device = btrfs_find_device(fs_info, devid, NULL, NULL);
2167                 if (!*device)
2168                         ret = -ENOENT;
2169         } else {
2170                 if (!devpath || !devpath[0])
2171                         return -EINVAL;
2172
2173                 ret = btrfs_find_device_missing_or_by_path(fs_info, devpath,
2174                                                            device);
2175         }
2176         return ret;
2177 }
2178
2179 /*
2180  * does all the dirty work required for changing file system's UUID.
2181  */
2182 static int btrfs_prepare_sprout(struct btrfs_fs_info *fs_info)
2183 {
2184         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2185         struct btrfs_fs_devices *old_devices;
2186         struct btrfs_fs_devices *seed_devices;
2187         struct btrfs_super_block *disk_super = fs_info->super_copy;
2188         struct btrfs_device *device;
2189         u64 super_flags;
2190
2191         BUG_ON(!mutex_is_locked(&uuid_mutex));
2192         if (!fs_devices->seeding)
2193                 return -EINVAL;
2194
2195         seed_devices = alloc_fs_devices(NULL);
2196         if (IS_ERR(seed_devices))
2197                 return PTR_ERR(seed_devices);
2198
2199         old_devices = clone_fs_devices(fs_devices);
2200         if (IS_ERR(old_devices)) {
2201                 kfree(seed_devices);
2202                 return PTR_ERR(old_devices);
2203         }
2204
2205         list_add(&old_devices->list, &fs_uuids);
2206
2207         memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
2208         seed_devices->opened = 1;
2209         INIT_LIST_HEAD(&seed_devices->devices);
2210         INIT_LIST_HEAD(&seed_devices->alloc_list);
2211         mutex_init(&seed_devices->device_list_mutex);
2212
2213         mutex_lock(&fs_info->fs_devices->device_list_mutex);
2214         list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
2215                               synchronize_rcu);
2216         list_for_each_entry(device, &seed_devices->devices, dev_list)
2217                 device->fs_devices = seed_devices;
2218
2219         mutex_lock(&fs_info->chunk_mutex);
2220         list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
2221         mutex_unlock(&fs_info->chunk_mutex);
2222
2223         fs_devices->seeding = 0;
2224         fs_devices->num_devices = 0;
2225         fs_devices->open_devices = 0;
2226         fs_devices->missing_devices = 0;
2227         fs_devices->rotating = 0;
2228         fs_devices->seed = seed_devices;
2229
2230         generate_random_uuid(fs_devices->fsid);
2231         memcpy(fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2232         memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2233         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2234
2235         super_flags = btrfs_super_flags(disk_super) &
2236                       ~BTRFS_SUPER_FLAG_SEEDING;
2237         btrfs_set_super_flags(disk_super, super_flags);
2238
2239         return 0;
2240 }
2241
2242 /*
2243  * Store the expected generation for seed devices in device items.
2244  */
2245 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
2246                                struct btrfs_fs_info *fs_info)
2247 {
2248         struct btrfs_root *root = fs_info->chunk_root;
2249         struct btrfs_path *path;
2250         struct extent_buffer *leaf;
2251         struct btrfs_dev_item *dev_item;
2252         struct btrfs_device *device;
2253         struct btrfs_key key;
2254         u8 fs_uuid[BTRFS_FSID_SIZE];
2255         u8 dev_uuid[BTRFS_UUID_SIZE];
2256         u64 devid;
2257         int ret;
2258
2259         path = btrfs_alloc_path();
2260         if (!path)
2261                 return -ENOMEM;
2262
2263         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2264         key.offset = 0;
2265         key.type = BTRFS_DEV_ITEM_KEY;
2266
2267         while (1) {
2268                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2269                 if (ret < 0)
2270                         goto error;
2271
2272                 leaf = path->nodes[0];
2273 next_slot:
2274                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2275                         ret = btrfs_next_leaf(root, path);
2276                         if (ret > 0)
2277                                 break;
2278                         if (ret < 0)
2279                                 goto error;
2280                         leaf = path->nodes[0];
2281                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2282                         btrfs_release_path(path);
2283                         continue;
2284                 }
2285
2286                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2287                 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
2288                     key.type != BTRFS_DEV_ITEM_KEY)
2289                         break;
2290
2291                 dev_item = btrfs_item_ptr(leaf, path->slots[0],
2292                                           struct btrfs_dev_item);
2293                 devid = btrfs_device_id(leaf, dev_item);
2294                 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
2295                                    BTRFS_UUID_SIZE);
2296                 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
2297                                    BTRFS_FSID_SIZE);
2298                 device = btrfs_find_device(fs_info, devid, dev_uuid, fs_uuid);
2299                 BUG_ON(!device); /* Logic error */
2300
2301                 if (device->fs_devices->seeding) {
2302                         btrfs_set_device_generation(leaf, dev_item,
2303                                                     device->generation);
2304                         btrfs_mark_buffer_dirty(leaf);
2305                 }
2306
2307                 path->slots[0]++;
2308                 goto next_slot;
2309         }
2310         ret = 0;
2311 error:
2312         btrfs_free_path(path);
2313         return ret;
2314 }
2315
2316 int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path)
2317 {
2318         struct btrfs_root *root = fs_info->dev_root;
2319         struct request_queue *q;
2320         struct btrfs_trans_handle *trans;
2321         struct btrfs_device *device;
2322         struct block_device *bdev;
2323         struct list_head *devices;
2324         struct super_block *sb = fs_info->sb;
2325         struct rcu_string *name;
2326         u64 tmp;
2327         int seeding_dev = 0;
2328         int ret = 0;
2329
2330         if (sb_rdonly(sb) && !fs_info->fs_devices->seeding)
2331                 return -EROFS;
2332
2333         bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2334                                   fs_info->bdev_holder);
2335         if (IS_ERR(bdev))
2336                 return PTR_ERR(bdev);
2337
2338         if (fs_info->fs_devices->seeding) {
2339                 seeding_dev = 1;
2340                 down_write(&sb->s_umount);
2341                 mutex_lock(&uuid_mutex);
2342         }
2343
2344         filemap_write_and_wait(bdev->bd_inode->i_mapping);
2345
2346         devices = &fs_info->fs_devices->devices;
2347
2348         mutex_lock(&fs_info->fs_devices->device_list_mutex);
2349         list_for_each_entry(device, devices, dev_list) {
2350                 if (device->bdev == bdev) {
2351                         ret = -EEXIST;
2352                         mutex_unlock(
2353                                 &fs_info->fs_devices->device_list_mutex);
2354                         goto error;
2355                 }
2356         }
2357         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2358
2359         device = btrfs_alloc_device(fs_info, NULL, NULL);
2360         if (IS_ERR(device)) {
2361                 /* we can safely leave the fs_devices entry around */
2362                 ret = PTR_ERR(device);
2363                 goto error;
2364         }
2365
2366         name = rcu_string_strdup(device_path, GFP_KERNEL);
2367         if (!name) {
2368                 kfree(device);
2369                 ret = -ENOMEM;
2370                 goto error;
2371         }
2372         rcu_assign_pointer(device->name, name);
2373
2374         trans = btrfs_start_transaction(root, 0);
2375         if (IS_ERR(trans)) {
2376                 rcu_string_free(device->name);
2377                 kfree(device);
2378                 ret = PTR_ERR(trans);
2379                 goto error;
2380         }
2381
2382         q = bdev_get_queue(bdev);
2383         if (blk_queue_discard(q))
2384                 device->can_discard = 1;
2385         device->writeable = 1;
2386         device->generation = trans->transid;
2387         device->io_width = fs_info->sectorsize;
2388         device->io_align = fs_info->sectorsize;
2389         device->sector_size = fs_info->sectorsize;
2390         device->total_bytes = round_down(i_size_read(bdev->bd_inode),
2391                                          fs_info->sectorsize);
2392         device->disk_total_bytes = device->total_bytes;
2393         device->commit_total_bytes = device->total_bytes;
2394         device->fs_info = fs_info;
2395         device->bdev = bdev;
2396         device->in_fs_metadata = 1;
2397         device->is_tgtdev_for_dev_replace = 0;
2398         device->mode = FMODE_EXCL;
2399         device->dev_stats_valid = 1;
2400         set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE);
2401
2402         if (seeding_dev) {
2403                 sb->s_flags &= ~MS_RDONLY;
2404                 ret = btrfs_prepare_sprout(fs_info);
2405                 BUG_ON(ret); /* -ENOMEM */
2406         }
2407
2408         device->fs_devices = fs_info->fs_devices;
2409
2410         mutex_lock(&fs_info->fs_devices->device_list_mutex);
2411         mutex_lock(&fs_info->chunk_mutex);
2412         list_add_rcu(&device->dev_list, &fs_info->fs_devices->devices);
2413         list_add(&device->dev_alloc_list,
2414                  &fs_info->fs_devices->alloc_list);
2415         fs_info->fs_devices->num_devices++;
2416         fs_info->fs_devices->open_devices++;
2417         fs_info->fs_devices->rw_devices++;
2418         fs_info->fs_devices->total_devices++;
2419         fs_info->fs_devices->total_rw_bytes += device->total_bytes;
2420
2421         atomic64_add(device->total_bytes, &fs_info->free_chunk_space);
2422
2423         if (!blk_queue_nonrot(q))
2424                 fs_info->fs_devices->rotating = 1;
2425
2426         tmp = btrfs_super_total_bytes(fs_info->super_copy);
2427         btrfs_set_super_total_bytes(fs_info->super_copy,
2428                 round_down(tmp + device->total_bytes, fs_info->sectorsize));
2429
2430         tmp = btrfs_super_num_devices(fs_info->super_copy);
2431         btrfs_set_super_num_devices(fs_info->super_copy, tmp + 1);
2432
2433         /* add sysfs device entry */
2434         btrfs_sysfs_add_device_link(fs_info->fs_devices, device);
2435
2436         /*
2437          * we've got more storage, clear any full flags on the space
2438          * infos
2439          */
2440         btrfs_clear_space_info_full(fs_info);
2441
2442         mutex_unlock(&fs_info->chunk_mutex);
2443         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2444
2445         if (seeding_dev) {
2446                 mutex_lock(&fs_info->chunk_mutex);
2447                 ret = init_first_rw_device(trans, fs_info);
2448                 mutex_unlock(&fs_info->chunk_mutex);
2449                 if (ret) {
2450                         btrfs_abort_transaction(trans, ret);
2451                         goto error_trans;
2452                 }
2453         }
2454
2455         ret = btrfs_add_device(trans, fs_info, device);
2456         if (ret) {
2457                 btrfs_abort_transaction(trans, ret);
2458                 goto error_trans;
2459         }
2460
2461         if (seeding_dev) {
2462                 char fsid_buf[BTRFS_UUID_UNPARSED_SIZE];
2463
2464                 ret = btrfs_finish_sprout(trans, fs_info);
2465                 if (ret) {
2466                         btrfs_abort_transaction(trans, ret);
2467                         goto error_trans;
2468                 }
2469
2470                 /* Sprouting would change fsid of the mounted root,
2471                  * so rename the fsid on the sysfs
2472                  */
2473                 snprintf(fsid_buf, BTRFS_UUID_UNPARSED_SIZE, "%pU",
2474                                                 fs_info->fsid);
2475                 if (kobject_rename(&fs_info->fs_devices->fsid_kobj, fsid_buf))
2476                         btrfs_warn(fs_info,
2477                                    "sysfs: failed to create fsid for sprout");
2478         }
2479
2480         ret = btrfs_commit_transaction(trans);
2481
2482         if (seeding_dev) {
2483                 mutex_unlock(&uuid_mutex);
2484                 up_write(&sb->s_umount);
2485
2486                 if (ret) /* transaction commit */
2487                         return ret;
2488
2489                 ret = btrfs_relocate_sys_chunks(fs_info);
2490                 if (ret < 0)
2491                         btrfs_handle_fs_error(fs_info, ret,
2492                                     "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command.");
2493                 trans = btrfs_attach_transaction(root);
2494                 if (IS_ERR(trans)) {
2495                         if (PTR_ERR(trans) == -ENOENT)
2496                                 return 0;
2497                         return PTR_ERR(trans);
2498                 }
2499                 ret = btrfs_commit_transaction(trans);
2500         }
2501
2502         /* Update ctime/mtime for libblkid */
2503         update_dev_time(device_path);
2504         return ret;
2505
2506 error_trans:
2507         if (seeding_dev)
2508                 sb->s_flags |= MS_RDONLY;
2509         btrfs_end_transaction(trans);
2510         rcu_string_free(device->name);
2511         btrfs_sysfs_rm_device_link(fs_info->fs_devices, device);
2512         kfree(device);
2513 error:
2514         blkdev_put(bdev, FMODE_EXCL);
2515         if (seeding_dev) {
2516                 mutex_unlock(&uuid_mutex);
2517                 up_write(&sb->s_umount);
2518         }
2519         return ret;
2520 }
2521
2522 int btrfs_init_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
2523                                   const char *device_path,
2524                                   struct btrfs_device *srcdev,
2525                                   struct btrfs_device **device_out)
2526 {
2527         struct request_queue *q;
2528         struct btrfs_device *device;
2529         struct block_device *bdev;
2530         struct list_head *devices;
2531         struct rcu_string *name;
2532         u64 devid = BTRFS_DEV_REPLACE_DEVID;
2533         int ret = 0;
2534
2535         *device_out = NULL;
2536         if (fs_info->fs_devices->seeding) {
2537                 btrfs_err(fs_info, "the filesystem is a seed filesystem!");
2538                 return -EINVAL;
2539         }
2540
2541         bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2542                                   fs_info->bdev_holder);
2543         if (IS_ERR(bdev)) {
2544                 btrfs_err(fs_info, "target device %s is invalid!", device_path);
2545                 return PTR_ERR(bdev);
2546         }
2547
2548         filemap_write_and_wait(bdev->bd_inode->i_mapping);
2549
2550         devices = &fs_info->fs_devices->devices;
2551         list_for_each_entry(device, devices, dev_list) {
2552                 if (device->bdev == bdev) {
2553                         btrfs_err(fs_info,
2554                                   "target device is in the filesystem!");
2555                         ret = -EEXIST;
2556                         goto error;
2557                 }
2558         }
2559
2560
2561         if (i_size_read(bdev->bd_inode) <
2562             btrfs_device_get_total_bytes(srcdev)) {
2563                 btrfs_err(fs_info,
2564                           "target device is smaller than source device!");
2565                 ret = -EINVAL;
2566                 goto error;
2567         }
2568
2569
2570         device = btrfs_alloc_device(NULL, &devid, NULL);
2571         if (IS_ERR(device)) {
2572                 ret = PTR_ERR(device);
2573                 goto error;
2574         }
2575
2576         name = rcu_string_strdup(device_path, GFP_KERNEL);
2577         if (!name) {
2578                 kfree(device);
2579                 ret = -ENOMEM;
2580                 goto error;
2581         }
2582         rcu_assign_pointer(device->name, name);
2583
2584         q = bdev_get_queue(bdev);
2585         if (blk_queue_discard(q))
2586                 device->can_discard = 1;
2587         mutex_lock(&fs_info->fs_devices->device_list_mutex);
2588         device->writeable = 1;
2589         device->generation = 0;
2590         device->io_width = fs_info->sectorsize;
2591         device->io_align = fs_info->sectorsize;
2592         device->sector_size = fs_info->sectorsize;
2593         device->total_bytes = btrfs_device_get_total_bytes(srcdev);
2594         device->disk_total_bytes = btrfs_device_get_disk_total_bytes(srcdev);
2595         device->bytes_used = btrfs_device_get_bytes_used(srcdev);
2596         ASSERT(list_empty(&srcdev->resized_list));
2597         device->commit_total_bytes = srcdev->commit_total_bytes;
2598         device->commit_bytes_used = device->bytes_used;
2599         device->fs_info = fs_info;
2600         device->bdev = bdev;
2601         device->in_fs_metadata = 1;
2602         device->is_tgtdev_for_dev_replace = 1;
2603         device->mode = FMODE_EXCL;
2604         device->dev_stats_valid = 1;
2605         set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE);
2606         device->fs_devices = fs_info->fs_devices;
2607         list_add(&device->dev_list, &fs_info->fs_devices->devices);
2608         fs_info->fs_devices->num_devices++;
2609         fs_info->fs_devices->open_devices++;
2610         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2611
2612         *device_out = device;
2613         return ret;
2614
2615 error:
2616         blkdev_put(bdev, FMODE_EXCL);
2617         return ret;
2618 }
2619
2620 void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info *fs_info,
2621                                               struct btrfs_device *tgtdev)
2622 {
2623         u32 sectorsize = fs_info->sectorsize;
2624
2625         WARN_ON(fs_info->fs_devices->rw_devices == 0);
2626         tgtdev->io_width = sectorsize;
2627         tgtdev->io_align = sectorsize;
2628         tgtdev->sector_size = sectorsize;
2629         tgtdev->fs_info = fs_info;
2630         tgtdev->in_fs_metadata = 1;
2631 }
2632
2633 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2634                                         struct btrfs_device *device)
2635 {
2636         int ret;
2637         struct btrfs_path *path;
2638         struct btrfs_root *root = device->fs_info->chunk_root;
2639         struct btrfs_dev_item *dev_item;
2640         struct extent_buffer *leaf;
2641         struct btrfs_key key;
2642
2643         path = btrfs_alloc_path();
2644         if (!path)
2645                 return -ENOMEM;
2646
2647         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2648         key.type = BTRFS_DEV_ITEM_KEY;
2649         key.offset = device->devid;
2650
2651         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2652         if (ret < 0)
2653                 goto out;
2654
2655         if (ret > 0) {
2656                 ret = -ENOENT;
2657                 goto out;
2658         }
2659
2660         leaf = path->nodes[0];
2661         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2662
2663         btrfs_set_device_id(leaf, dev_item, device->devid);
2664         btrfs_set_device_type(leaf, dev_item, device->type);
2665         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2666         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2667         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2668         btrfs_set_device_total_bytes(leaf, dev_item,
2669                                      btrfs_device_get_disk_total_bytes(device));
2670         btrfs_set_device_bytes_used(leaf, dev_item,
2671                                     btrfs_device_get_bytes_used(device));
2672         btrfs_mark_buffer_dirty(leaf);
2673
2674 out:
2675         btrfs_free_path(path);
2676         return ret;
2677 }
2678
2679 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2680                       struct btrfs_device *device, u64 new_size)
2681 {
2682         struct btrfs_fs_info *fs_info = device->fs_info;
2683         struct btrfs_super_block *super_copy = fs_info->super_copy;
2684         struct btrfs_fs_devices *fs_devices;
2685         u64 old_total;
2686         u64 diff;
2687
2688         if (!device->writeable)
2689                 return -EACCES;
2690
2691         new_size = round_down(new_size, fs_info->sectorsize);
2692
2693         mutex_lock(&fs_info->chunk_mutex);
2694         old_total = btrfs_super_total_bytes(super_copy);
2695         diff = round_down(new_size - device->total_bytes, fs_info->sectorsize);
2696
2697         if (new_size <= device->total_bytes ||
2698             device->is_tgtdev_for_dev_replace) {
2699                 mutex_unlock(&fs_info->chunk_mutex);
2700                 return -EINVAL;
2701         }
2702
2703         fs_devices = fs_info->fs_devices;
2704
2705         btrfs_set_super_total_bytes(super_copy,
2706                         round_down(old_total + diff, fs_info->sectorsize));
2707         device->fs_devices->total_rw_bytes += diff;
2708
2709         btrfs_device_set_total_bytes(device, new_size);
2710         btrfs_device_set_disk_total_bytes(device, new_size);
2711         btrfs_clear_space_info_full(device->fs_info);
2712         if (list_empty(&device->resized_list))
2713                 list_add_tail(&device->resized_list,
2714                               &fs_devices->resized_devices);
2715         mutex_unlock(&fs_info->chunk_mutex);
2716
2717         return btrfs_update_device(trans, device);
2718 }
2719
2720 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
2721                             struct btrfs_fs_info *fs_info, u64 chunk_offset)
2722 {
2723         struct btrfs_root *root = fs_info->chunk_root;
2724         int ret;
2725         struct btrfs_path *path;
2726         struct btrfs_key key;
2727
2728         path = btrfs_alloc_path();
2729         if (!path)
2730                 return -ENOMEM;
2731
2732         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2733         key.offset = chunk_offset;
2734         key.type = BTRFS_CHUNK_ITEM_KEY;
2735
2736         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2737         if (ret < 0)
2738                 goto out;
2739         else if (ret > 0) { /* Logic error or corruption */
2740                 btrfs_handle_fs_error(fs_info, -ENOENT,
2741                                       "Failed lookup while freeing chunk.");
2742                 ret = -ENOENT;
2743                 goto out;
2744         }
2745
2746         ret = btrfs_del_item(trans, root, path);
2747         if (ret < 0)
2748                 btrfs_handle_fs_error(fs_info, ret,
2749                                       "Failed to delete chunk item.");
2750 out:
2751         btrfs_free_path(path);
2752         return ret;
2753 }
2754
2755 static int btrfs_del_sys_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
2756 {
2757         struct btrfs_super_block *super_copy = fs_info->super_copy;
2758         struct btrfs_disk_key *disk_key;
2759         struct btrfs_chunk *chunk;
2760         u8 *ptr;
2761         int ret = 0;
2762         u32 num_stripes;
2763         u32 array_size;
2764         u32 len = 0;
2765         u32 cur;
2766         struct btrfs_key key;
2767
2768         mutex_lock(&fs_info->chunk_mutex);
2769         array_size = btrfs_super_sys_array_size(super_copy);
2770
2771         ptr = super_copy->sys_chunk_array;
2772         cur = 0;
2773
2774         while (cur < array_size) {
2775                 disk_key = (struct btrfs_disk_key *)ptr;
2776                 btrfs_disk_key_to_cpu(&key, disk_key);
2777
2778                 len = sizeof(*disk_key);
2779
2780                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2781                         chunk = (struct btrfs_chunk *)(ptr + len);
2782                         num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2783                         len += btrfs_chunk_item_size(num_stripes);
2784                 } else {
2785                         ret = -EIO;
2786                         break;
2787                 }
2788                 if (key.objectid == BTRFS_FIRST_CHUNK_TREE_OBJECTID &&
2789                     key.offset == chunk_offset) {
2790                         memmove(ptr, ptr + len, array_size - (cur + len));
2791                         array_size -= len;
2792                         btrfs_set_super_sys_array_size(super_copy, array_size);
2793                 } else {
2794                         ptr += len;
2795                         cur += len;
2796                 }
2797         }
2798         mutex_unlock(&fs_info->chunk_mutex);
2799         return ret;
2800 }
2801
2802 static struct extent_map *get_chunk_map(struct btrfs_fs_info *fs_info,
2803                                         u64 logical, u64 length)
2804 {
2805         struct extent_map_tree *em_tree;
2806         struct extent_map *em;
2807
2808         em_tree = &fs_info->mapping_tree.map_tree;
2809         read_lock(&em_tree->lock);
2810         em = lookup_extent_mapping(em_tree, logical, length);
2811         read_unlock(&em_tree->lock);
2812
2813         if (!em) {
2814                 btrfs_crit(fs_info, "unable to find logical %llu length %llu",
2815                            logical, length);
2816                 return ERR_PTR(-EINVAL);
2817         }
2818
2819         if (em->start > logical || em->start + em->len < logical) {
2820                 btrfs_crit(fs_info,
2821                            "found a bad mapping, wanted %llu-%llu, found %llu-%llu",
2822                            logical, length, em->start, em->start + em->len);
2823                 free_extent_map(em);
2824                 return ERR_PTR(-EINVAL);
2825         }
2826
2827         /* callers are responsible for dropping em's ref. */
2828         return em;
2829 }
2830
2831 int btrfs_remove_chunk(struct btrfs_trans_handle *trans,
2832                        struct btrfs_fs_info *fs_info, u64 chunk_offset)
2833 {
2834         struct extent_map *em;
2835         struct map_lookup *map;
2836         u64 dev_extent_len = 0;
2837         int i, ret = 0;
2838         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2839
2840         em = get_chunk_map(fs_info, chunk_offset, 1);
2841         if (IS_ERR(em)) {
2842                 /*
2843                  * This is a logic error, but we don't want to just rely on the
2844                  * user having built with ASSERT enabled, so if ASSERT doesn't
2845                  * do anything we still error out.
2846                  */
2847                 ASSERT(0);
2848                 return PTR_ERR(em);
2849         }
2850         map = em->map_lookup;
2851         mutex_lock(&fs_info->chunk_mutex);
2852         check_system_chunk(trans, fs_info, map->type);
2853         mutex_unlock(&fs_info->chunk_mutex);
2854
2855         /*
2856          * Take the device list mutex to prevent races with the final phase of
2857          * a device replace operation that replaces the device object associated
2858          * with map stripes (dev-replace.c:btrfs_dev_replace_finishing()).
2859          */
2860         mutex_lock(&fs_devices->device_list_mutex);
2861         for (i = 0; i < map->num_stripes; i++) {
2862                 struct btrfs_device *device = map->stripes[i].dev;
2863                 ret = btrfs_free_dev_extent(trans, device,
2864                                             map->stripes[i].physical,
2865                                             &dev_extent_len);
2866                 if (ret) {
2867                         mutex_unlock(&fs_devices->device_list_mutex);
2868                         btrfs_abort_transaction(trans, ret);
2869                         goto out;
2870                 }
2871
2872                 if (device->bytes_used > 0) {
2873                         mutex_lock(&fs_info->chunk_mutex);
2874                         btrfs_device_set_bytes_used(device,
2875                                         device->bytes_used - dev_extent_len);
2876                         atomic64_add(dev_extent_len, &fs_info->free_chunk_space);
2877                         btrfs_clear_space_info_full(fs_info);
2878                         mutex_unlock(&fs_info->chunk_mutex);
2879                 }
2880
2881                 if (map->stripes[i].dev) {
2882                         ret = btrfs_update_device(trans, map->stripes[i].dev);
2883                         if (ret) {
2884                                 mutex_unlock(&fs_devices->device_list_mutex);
2885                                 btrfs_abort_transaction(trans, ret);
2886                                 goto out;
2887                         }
2888                 }
2889         }
2890         mutex_unlock(&fs_devices->device_list_mutex);
2891
2892         ret = btrfs_free_chunk(trans, fs_info, chunk_offset);
2893         if (ret) {
2894                 btrfs_abort_transaction(trans, ret);
2895                 goto out;
2896         }
2897
2898         trace_btrfs_chunk_free(fs_info, map, chunk_offset, em->len);
2899
2900         if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2901                 ret = btrfs_del_sys_chunk(fs_info, chunk_offset);
2902                 if (ret) {
2903                         btrfs_abort_transaction(trans, ret);
2904                         goto out;
2905                 }
2906         }
2907
2908         ret = btrfs_remove_block_group(trans, fs_info, chunk_offset, em);
2909         if (ret) {
2910                 btrfs_abort_transaction(trans, ret);
2911                 goto out;
2912         }
2913
2914 out:
2915         /* once for us */
2916         free_extent_map(em);
2917         return ret;
2918 }
2919
2920 static int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
2921 {
2922         struct btrfs_root *root = fs_info->chunk_root;
2923         struct btrfs_trans_handle *trans;
2924         int ret;
2925
2926         /*
2927          * Prevent races with automatic removal of unused block groups.
2928          * After we relocate and before we remove the chunk with offset
2929          * chunk_offset, automatic removal of the block group can kick in,
2930          * resulting in a failure when calling btrfs_remove_chunk() below.
2931          *
2932          * Make sure to acquire this mutex before doing a tree search (dev
2933          * or chunk trees) to find chunks. Otherwise the cleaner kthread might
2934          * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after
2935          * we release the path used to search the chunk/dev tree and before
2936          * the current task acquires this mutex and calls us.
2937          */
2938         ASSERT(mutex_is_locked(&fs_info->delete_unused_bgs_mutex));
2939
2940         ret = btrfs_can_relocate(fs_info, chunk_offset);
2941         if (ret)
2942                 return -ENOSPC;
2943
2944         /* step one, relocate all the extents inside this chunk */
2945         btrfs_scrub_pause(fs_info);
2946         ret = btrfs_relocate_block_group(fs_info, chunk_offset);
2947         btrfs_scrub_continue(fs_info);
2948         if (ret)
2949                 return ret;
2950
2951         trans = btrfs_start_trans_remove_block_group(root->fs_info,
2952                                                      chunk_offset);
2953         if (IS_ERR(trans)) {
2954                 ret = PTR_ERR(trans);
2955                 btrfs_handle_fs_error(root->fs_info, ret, NULL);
2956                 return ret;
2957         }
2958
2959         /*
2960          * step two, delete the device extents and the
2961          * chunk tree entries
2962          */
2963         ret = btrfs_remove_chunk(trans, fs_info, chunk_offset);
2964         btrfs_end_transaction(trans);
2965         return ret;
2966 }
2967
2968 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info)
2969 {
2970         struct btrfs_root *chunk_root = fs_info->chunk_root;
2971         struct btrfs_path *path;
2972         struct extent_buffer *leaf;
2973         struct btrfs_chunk *chunk;
2974         struct btrfs_key key;
2975         struct btrfs_key found_key;
2976         u64 chunk_type;
2977         bool retried = false;
2978         int failed = 0;
2979         int ret;
2980
2981         path = btrfs_alloc_path();
2982         if (!path)
2983                 return -ENOMEM;
2984
2985 again:
2986         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2987         key.offset = (u64)-1;
2988         key.type = BTRFS_CHUNK_ITEM_KEY;
2989
2990         while (1) {
2991                 mutex_lock(&fs_info->delete_unused_bgs_mutex);
2992                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2993                 if (ret < 0) {
2994                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
2995                         goto error;
2996                 }
2997                 BUG_ON(ret == 0); /* Corruption */
2998
2999                 ret = btrfs_previous_item(chunk_root, path, key.objectid,
3000                                           key.type);
3001                 if (ret)
3002                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3003                 if (ret < 0)
3004                         goto error;
3005                 if (ret > 0)
3006                         break;
3007
3008                 leaf = path->nodes[0];
3009                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3010
3011                 chunk = btrfs_item_ptr(leaf, path->slots[0],
3012                                        struct btrfs_chunk);
3013                 chunk_type = btrfs_chunk_type(leaf, chunk);
3014                 btrfs_release_path(path);
3015
3016                 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
3017                         ret = btrfs_relocate_chunk(fs_info, found_key.offset);
3018                         if (ret == -ENOSPC)
3019                                 failed++;
3020                         else
3021                                 BUG_ON(ret);
3022                 }
3023                 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3024
3025                 if (found_key.offset == 0)
3026                         break;
3027                 key.offset = found_key.offset - 1;
3028         }
3029         ret = 0;
3030         if (failed && !retried) {
3031                 failed = 0;
3032                 retried = true;
3033                 goto again;
3034         } else if (WARN_ON(failed && retried)) {
3035                 ret = -ENOSPC;
3036         }
3037 error:
3038         btrfs_free_path(path);
3039         return ret;
3040 }
3041
3042 static int insert_balance_item(struct btrfs_fs_info *fs_info,
3043                                struct btrfs_balance_control *bctl)
3044 {
3045         struct btrfs_root *root = fs_info->tree_root;
3046         struct btrfs_trans_handle *trans;
3047         struct btrfs_balance_item *item;
3048         struct btrfs_disk_balance_args disk_bargs;
3049         struct btrfs_path *path;
3050         struct extent_buffer *leaf;
3051         struct btrfs_key key;
3052         int ret, err;
3053
3054         path = btrfs_alloc_path();
3055         if (!path)
3056                 return -ENOMEM;
3057
3058         trans = btrfs_start_transaction(root, 0);
3059         if (IS_ERR(trans)) {
3060                 btrfs_free_path(path);
3061                 return PTR_ERR(trans);
3062         }
3063
3064         key.objectid = BTRFS_BALANCE_OBJECTID;
3065         key.type = BTRFS_TEMPORARY_ITEM_KEY;
3066         key.offset = 0;
3067
3068         ret = btrfs_insert_empty_item(trans, root, path, &key,
3069                                       sizeof(*item));
3070         if (ret)
3071                 goto out;
3072
3073         leaf = path->nodes[0];
3074         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3075
3076         memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item));
3077
3078         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
3079         btrfs_set_balance_data(leaf, item, &disk_bargs);
3080         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
3081         btrfs_set_balance_meta(leaf, item, &disk_bargs);
3082         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
3083         btrfs_set_balance_sys(leaf, item, &disk_bargs);
3084
3085         btrfs_set_balance_flags(leaf, item, bctl->flags);
3086
3087         btrfs_mark_buffer_dirty(leaf);
3088 out:
3089         btrfs_free_path(path);
3090         err = btrfs_commit_transaction(trans);
3091         if (err && !ret)
3092                 ret = err;
3093         return ret;
3094 }
3095
3096 static int del_balance_item(struct btrfs_fs_info *fs_info)
3097 {
3098         struct btrfs_root *root = fs_info->tree_root;
3099         struct btrfs_trans_handle *trans;
3100         struct btrfs_path *path;
3101         struct btrfs_key key;
3102         int ret, err;
3103
3104         path = btrfs_alloc_path();
3105         if (!path)
3106                 return -ENOMEM;
3107
3108         trans = btrfs_start_transaction(root, 0);
3109         if (IS_ERR(trans)) {
3110                 btrfs_free_path(path);
3111                 return PTR_ERR(trans);
3112         }
3113
3114         key.objectid = BTRFS_BALANCE_OBJECTID;
3115         key.type = BTRFS_TEMPORARY_ITEM_KEY;
3116         key.offset = 0;
3117
3118         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3119         if (ret < 0)
3120                 goto out;
3121         if (ret > 0) {
3122                 ret = -ENOENT;
3123                 goto out;
3124         }
3125
3126         ret = btrfs_del_item(trans, root, path);
3127 out:
3128         btrfs_free_path(path);
3129         err = btrfs_commit_transaction(trans);
3130         if (err && !ret)
3131                 ret = err;
3132         return ret;
3133 }
3134
3135 /*
3136  * This is a heuristic used to reduce the number of chunks balanced on
3137  * resume after balance was interrupted.
3138  */
3139 static void update_balance_args(struct btrfs_balance_control *bctl)
3140 {
3141         /*
3142          * Turn on soft mode for chunk types that were being converted.
3143          */
3144         if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
3145                 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
3146         if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
3147                 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
3148         if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
3149                 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
3150
3151         /*
3152          * Turn on usage filter if is not already used.  The idea is
3153          * that chunks that we have already balanced should be
3154          * reasonably full.  Don't do it for chunks that are being
3155          * converted - that will keep us from relocating unconverted
3156          * (albeit full) chunks.
3157          */
3158         if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3159             !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3160             !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3161                 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
3162                 bctl->data.usage = 90;
3163         }
3164         if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3165             !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3166             !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3167                 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
3168                 bctl->sys.usage = 90;
3169         }
3170         if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3171             !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3172             !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3173                 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
3174                 bctl->meta.usage = 90;
3175         }
3176 }
3177
3178 /*
3179  * Should be called with both balance and volume mutexes held to
3180  * serialize other volume operations (add_dev/rm_dev/resize) with
3181  * restriper.  Same goes for unset_balance_control.
3182  */
3183 static void set_balance_control(struct btrfs_balance_control *bctl)
3184 {
3185         struct btrfs_fs_info *fs_info = bctl->fs_info;
3186
3187         BUG_ON(fs_info->balance_ctl);
3188
3189         spin_lock(&fs_info->balance_lock);
3190         fs_info->balance_ctl = bctl;
3191         spin_unlock(&fs_info->balance_lock);
3192 }
3193
3194 static void unset_balance_control(struct btrfs_fs_info *fs_info)
3195 {
3196         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3197
3198         BUG_ON(!fs_info->balance_ctl);
3199
3200         spin_lock(&fs_info->balance_lock);
3201         fs_info->balance_ctl = NULL;
3202         spin_unlock(&fs_info->balance_lock);
3203
3204         kfree(bctl);
3205 }
3206
3207 /*
3208  * Balance filters.  Return 1 if chunk should be filtered out
3209  * (should not be balanced).
3210  */
3211 static int chunk_profiles_filter(u64 chunk_type,
3212                                  struct btrfs_balance_args *bargs)
3213 {
3214         chunk_type = chunk_to_extended(chunk_type) &
3215                                 BTRFS_EXTENDED_PROFILE_MASK;
3216
3217         if (bargs->profiles & chunk_type)
3218                 return 0;
3219
3220         return 1;
3221 }
3222
3223 static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
3224                               struct btrfs_balance_args *bargs)
3225 {
3226         struct btrfs_block_group_cache *cache;
3227         u64 chunk_used;
3228         u64 user_thresh_min;
3229         u64 user_thresh_max;
3230         int ret = 1;
3231
3232         cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3233         chunk_used = btrfs_block_group_used(&cache->item);
3234
3235         if (bargs->usage_min == 0)
3236                 user_thresh_min = 0;
3237         else
3238                 user_thresh_min = div_factor_fine(cache->key.offset,
3239                                         bargs->usage_min);
3240
3241         if (bargs->usage_max == 0)
3242                 user_thresh_max = 1;
3243         else if (bargs->usage_max > 100)
3244                 user_thresh_max = cache->key.offset;
3245         else
3246                 user_thresh_max = div_factor_fine(cache->key.offset,
3247                                         bargs->usage_max);
3248
3249         if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max)
3250                 ret = 0;
3251
3252         btrfs_put_block_group(cache);
3253         return ret;
3254 }
3255
3256 static int chunk_usage_filter(struct btrfs_fs_info *fs_info,
3257                 u64 chunk_offset, struct btrfs_balance_args *bargs)
3258 {
3259         struct btrfs_block_group_cache *cache;
3260         u64 chunk_used, user_thresh;
3261         int ret = 1;
3262
3263         cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3264         chunk_used = btrfs_block_group_used(&cache->item);
3265
3266         if (bargs->usage_min == 0)
3267                 user_thresh = 1;
3268         else if (bargs->usage > 100)
3269                 user_thresh = cache->key.offset;
3270         else
3271                 user_thresh = div_factor_fine(cache->key.offset,
3272                                               bargs->usage);
3273
3274         if (chunk_used < user_thresh)
3275                 ret = 0;
3276
3277         btrfs_put_block_group(cache);
3278         return ret;
3279 }
3280
3281 static int chunk_devid_filter(struct extent_buffer *leaf,
3282                               struct btrfs_chunk *chunk,
3283                               struct btrfs_balance_args *bargs)
3284 {
3285         struct btrfs_stripe *stripe;
3286         int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3287         int i;
3288
3289         for (i = 0; i < num_stripes; i++) {
3290                 stripe = btrfs_stripe_nr(chunk, i);
3291                 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
3292                         return 0;
3293         }
3294
3295         return 1;
3296 }
3297
3298 /* [pstart, pend) */
3299 static int chunk_drange_filter(struct extent_buffer *leaf,
3300                                struct btrfs_chunk *chunk,
3301                                struct btrfs_balance_args *bargs)
3302 {
3303         struct btrfs_stripe *stripe;
3304         int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3305         u64 stripe_offset;
3306         u64 stripe_length;
3307         int factor;
3308         int i;
3309
3310         if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
3311                 return 0;
3312
3313         if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP |
3314              BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) {
3315                 factor = num_stripes / 2;
3316         } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) {
3317                 factor = num_stripes - 1;
3318         } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) {
3319                 factor = num_stripes - 2;
3320         } else {
3321                 factor = num_stripes;
3322         }
3323
3324         for (i = 0; i < num_stripes; i++) {
3325                 stripe = btrfs_stripe_nr(chunk, i);
3326                 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
3327                         continue;
3328
3329                 stripe_offset = btrfs_stripe_offset(leaf, stripe);
3330                 stripe_length = btrfs_chunk_length(leaf, chunk);
3331                 stripe_length = div_u64(stripe_length, factor);
3332
3333                 if (stripe_offset < bargs->pend &&
3334                     stripe_offset + stripe_length > bargs->pstart)
3335                         return 0;
3336         }
3337
3338         return 1;
3339 }
3340
3341 /* [vstart, vend) */
3342 static int chunk_vrange_filter(struct extent_buffer *leaf,
3343                                struct btrfs_chunk *chunk,
3344                                u64 chunk_offset,
3345                                struct btrfs_balance_args *bargs)
3346 {
3347         if (chunk_offset < bargs->vend &&
3348             chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
3349                 /* at least part of the chunk is inside this vrange */
3350                 return 0;
3351
3352         return 1;
3353 }
3354
3355 static int chunk_stripes_range_filter(struct extent_buffer *leaf,
3356                                struct btrfs_chunk *chunk,
3357                                struct btrfs_balance_args *bargs)
3358 {
3359         int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3360
3361         if (bargs->stripes_min <= num_stripes
3362                         && num_stripes <= bargs->stripes_max)
3363                 return 0;
3364
3365         return 1;
3366 }
3367
3368 static int chunk_soft_convert_filter(u64 chunk_type,
3369                                      struct btrfs_balance_args *bargs)
3370 {
3371         if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
3372                 return 0;
3373
3374         chunk_type = chunk_to_extended(chunk_type) &
3375                                 BTRFS_EXTENDED_PROFILE_MASK;
3376
3377         if (bargs->target == chunk_type)
3378                 return 1;
3379
3380         return 0;
3381 }
3382
3383 static int should_balance_chunk(struct btrfs_fs_info *fs_info,
3384                                 struct extent_buffer *leaf,
3385                                 struct btrfs_chunk *chunk, u64 chunk_offset)
3386 {
3387         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3388         struct btrfs_balance_args *bargs = NULL;
3389         u64 chunk_type = btrfs_chunk_type(leaf, chunk);
3390
3391         /* type filter */
3392         if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
3393               (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
3394                 return 0;
3395         }
3396
3397         if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3398                 bargs = &bctl->data;
3399         else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3400                 bargs = &bctl->sys;
3401         else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3402                 bargs = &bctl->meta;
3403
3404         /* profiles filter */
3405         if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
3406             chunk_profiles_filter(chunk_type, bargs)) {
3407                 return 0;
3408         }
3409
3410         /* usage filter */
3411         if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
3412             chunk_usage_filter(fs_info, chunk_offset, bargs)) {
3413                 return 0;
3414         } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3415             chunk_usage_range_filter(fs_info, chunk_offset, bargs)) {
3416                 return 0;
3417         }
3418
3419         /* devid filter */
3420         if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
3421             chunk_devid_filter(leaf, chunk, bargs)) {
3422                 return 0;
3423         }
3424
3425         /* drange filter, makes sense only with devid filter */
3426         if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
3427             chunk_drange_filter(leaf, chunk, bargs)) {
3428                 return 0;
3429         }
3430
3431         /* vrange filter */
3432         if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
3433             chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
3434                 return 0;
3435         }
3436
3437         /* stripes filter */
3438         if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) &&
3439             chunk_stripes_range_filter(leaf, chunk, bargs)) {
3440                 return 0;
3441         }
3442
3443         /* soft profile changing mode */
3444         if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
3445             chunk_soft_convert_filter(chunk_type, bargs)) {
3446                 return 0;
3447         }
3448
3449         /*
3450          * limited by count, must be the last filter
3451          */
3452         if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) {
3453                 if (bargs->limit == 0)
3454                         return 0;
3455                 else
3456                         bargs->limit--;
3457         } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) {
3458                 /*
3459                  * Same logic as the 'limit' filter; the minimum cannot be
3460                  * determined here because we do not have the global information
3461                  * about the count of all chunks that satisfy the filters.
3462                  */
3463                 if (bargs->limit_max == 0)
3464                         return 0;
3465                 else
3466                         bargs->limit_max--;
3467         }
3468
3469         return 1;
3470 }
3471
3472 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
3473 {
3474         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3475         struct btrfs_root *chunk_root = fs_info->chunk_root;
3476         struct btrfs_root *dev_root = fs_info->dev_root;
3477         struct list_head *devices;
3478         struct btrfs_device *device;
3479         u64 old_size;
3480         u64 size_to_free;
3481         u64 chunk_type;
3482         struct btrfs_chunk *chunk;
3483         struct btrfs_path *path = NULL;
3484         struct btrfs_key key;
3485         struct btrfs_key found_key;
3486         struct btrfs_trans_handle *trans;
3487         struct extent_buffer *leaf;
3488         int slot;
3489         int ret;
3490         int enospc_errors = 0;
3491         bool counting = true;
3492         /* The single value limit and min/max limits use the same bytes in the */
3493         u64 limit_data = bctl->data.limit;
3494         u64 limit_meta = bctl->meta.limit;
3495         u64 limit_sys = bctl->sys.limit;
3496         u32 count_data = 0;
3497         u32 count_meta = 0;
3498         u32 count_sys = 0;
3499         int chunk_reserved = 0;
3500         u64 bytes_used = 0;
3501
3502         /* step one make some room on all the devices */
3503         devices = &fs_info->fs_devices->devices;
3504         list_for_each_entry(device, devices, dev_list) {
3505                 old_size = btrfs_device_get_total_bytes(device);
3506                 size_to_free = div_factor(old_size, 1);
3507                 size_to_free = min_t(u64, size_to_free, SZ_1M);
3508                 if (!device->writeable ||
3509                     btrfs_device_get_total_bytes(device) -
3510                     btrfs_device_get_bytes_used(device) > size_to_free ||
3511                     device->is_tgtdev_for_dev_replace)
3512                         continue;
3513
3514                 ret = btrfs_shrink_device(device, old_size - size_to_free);
3515                 if (ret == -ENOSPC)
3516                         break;
3517                 if (ret) {
3518                         /* btrfs_shrink_device never returns ret > 0 */
3519                         WARN_ON(ret > 0);
3520                         goto error;
3521                 }
3522
3523                 trans = btrfs_start_transaction(dev_root, 0);
3524                 if (IS_ERR(trans)) {
3525                         ret = PTR_ERR(trans);
3526                         btrfs_info_in_rcu(fs_info,
3527                  "resize: unable to start transaction after shrinking device %s (error %d), old size %llu, new size %llu",
3528                                           rcu_str_deref(device->name), ret,
3529                                           old_size, old_size - size_to_free);
3530                         goto error;
3531                 }
3532
3533                 ret = btrfs_grow_device(trans, device, old_size);
3534                 if (ret) {
3535                         btrfs_end_transaction(trans);
3536                         /* btrfs_grow_device never returns ret > 0 */
3537                         WARN_ON(ret > 0);
3538                         btrfs_info_in_rcu(fs_info,
3539                  "resize: unable to grow device after shrinking device %s (error %d), old size %llu, new size %llu",
3540                                           rcu_str_deref(device->name), ret,
3541                                           old_size, old_size - size_to_free);
3542                         goto error;
3543                 }
3544
3545                 btrfs_end_transaction(trans);
3546         }
3547
3548         /* step two, relocate all the chunks */
3549         path = btrfs_alloc_path();
3550         if (!path) {
3551                 ret = -ENOMEM;
3552                 goto error;
3553         }
3554
3555         /* zero out stat counters */
3556         spin_lock(&fs_info->balance_lock);
3557         memset(&bctl->stat, 0, sizeof(bctl->stat));
3558         spin_unlock(&fs_info->balance_lock);
3559 again:
3560         if (!counting) {
3561                 /*
3562                  * The single value limit and min/max limits use the same bytes
3563                  * in the
3564                  */
3565                 bctl->data.limit = limit_data;
3566                 bctl->meta.limit = limit_meta;
3567                 bctl->sys.limit = limit_sys;
3568         }
3569         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3570         key.offset = (u64)-1;
3571         key.type = BTRFS_CHUNK_ITEM_KEY;
3572
3573         while (1) {
3574                 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
3575                     atomic_read(&fs_info->balance_cancel_req)) {
3576                         ret = -ECANCELED;
3577                         goto error;
3578                 }
3579
3580                 mutex_lock(&fs_info->delete_unused_bgs_mutex);
3581                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3582                 if (ret < 0) {
3583                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3584                         goto error;
3585                 }
3586
3587                 /*
3588                  * this shouldn't happen, it means the last relocate
3589                  * failed
3590                  */
3591                 if (ret == 0)
3592                         BUG(); /* FIXME break ? */
3593
3594                 ret = btrfs_previous_item(chunk_root, path, 0,
3595                                           BTRFS_CHUNK_ITEM_KEY);
3596                 if (ret) {
3597                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3598                         ret = 0;
3599                         break;
3600                 }
3601
3602                 leaf = path->nodes[0];
3603                 slot = path->slots[0];
3604                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3605
3606                 if (found_key.objectid != key.objectid) {
3607                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3608                         break;
3609                 }
3610
3611                 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3612                 chunk_type = btrfs_chunk_type(leaf, chunk);
3613
3614                 if (!counting) {
3615                         spin_lock(&fs_info->balance_lock);
3616                         bctl->stat.considered++;
3617                         spin_unlock(&fs_info->balance_lock);
3618                 }
3619
3620                 ret = should_balance_chunk(fs_info, leaf, chunk,
3621                                            found_key.offset);
3622
3623                 btrfs_release_path(path);
3624                 if (!ret) {
3625                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3626                         goto loop;
3627                 }
3628
3629                 if (counting) {
3630                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3631                         spin_lock(&fs_info->balance_lock);
3632                         bctl->stat.expected++;
3633                         spin_unlock(&fs_info->balance_lock);
3634
3635                         if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3636                                 count_data++;
3637                         else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3638                                 count_sys++;
3639                         else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3640                                 count_meta++;
3641
3642                         goto loop;
3643                 }
3644
3645                 /*
3646                  * Apply limit_min filter, no need to check if the LIMITS
3647                  * filter is used, limit_min is 0 by default
3648                  */
3649                 if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
3650                                         count_data < bctl->data.limit_min)
3651                                 || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) &&
3652                                         count_meta < bctl->meta.limit_min)
3653                                 || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) &&
3654                                         count_sys < bctl->sys.limit_min)) {
3655                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3656                         goto loop;
3657                 }
3658
3659                 ASSERT(fs_info->data_sinfo);
3660                 spin_lock(&fs_info->data_sinfo->lock);
3661                 bytes_used = fs_info->data_sinfo->bytes_used;
3662                 spin_unlock(&fs_info->data_sinfo->lock);
3663
3664                 if ((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
3665                     !chunk_reserved && !bytes_used) {
3666                         trans = btrfs_start_transaction(chunk_root, 0);
3667                         if (IS_ERR(trans)) {
3668                                 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3669                                 ret = PTR_ERR(trans);
3670                                 goto error;
3671                         }
3672
3673                         ret = btrfs_force_chunk_alloc(trans, fs_info,
3674                                                       BTRFS_BLOCK_GROUP_DATA);
3675                         btrfs_end_transaction(trans);
3676                         if (ret < 0) {
3677                                 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3678                                 goto error;
3679                         }
3680                         chunk_reserved = 1;
3681                 }
3682
3683                 ret = btrfs_relocate_chunk(fs_info, found_key.offset);
3684                 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3685                 if (ret && ret != -ENOSPC)
3686                         goto error;
3687                 if (ret == -ENOSPC) {
3688                         enospc_errors++;
3689                 } else {
3690                         spin_lock(&fs_info->balance_lock);
3691                         bctl->stat.completed++;
3692                         spin_unlock(&fs_info->balance_lock);
3693                 }
3694 loop:
3695                 if (found_key.offset == 0)
3696                         break;
3697                 key.offset = found_key.offset - 1;
3698         }
3699
3700         if (counting) {
3701                 btrfs_release_path(path);
3702                 counting = false;
3703                 goto again;
3704         }
3705 error:
3706         btrfs_free_path(path);
3707         if (enospc_errors) {
3708                 btrfs_info(fs_info, "%d enospc errors during balance",
3709                            enospc_errors);
3710                 if (!ret)
3711                         ret = -ENOSPC;
3712         }
3713
3714         return ret;
3715 }
3716
3717 /**
3718  * alloc_profile_is_valid - see if a given profile is valid and reduced
3719  * @flags: profile to validate
3720  * @extended: if true @flags is treated as an extended profile
3721  */
3722 static int alloc_profile_is_valid(u64 flags, int extended)
3723 {
3724         u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
3725                                BTRFS_BLOCK_GROUP_PROFILE_MASK);
3726
3727         flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
3728
3729         /* 1) check that all other bits are zeroed */
3730         if (flags & ~mask)
3731                 return 0;
3732
3733         /* 2) see if profile is reduced */
3734         if (flags == 0)
3735                 return !extended; /* "0" is valid for usual profiles */
3736
3737         /* true if exactly one bit set */
3738         return (flags & (flags - 1)) == 0;
3739 }
3740
3741 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
3742 {
3743         /* cancel requested || normal exit path */
3744         return atomic_read(&fs_info->balance_cancel_req) ||
3745                 (atomic_read(&fs_info->balance_pause_req) == 0 &&
3746                  atomic_read(&fs_info->balance_cancel_req) == 0);
3747 }
3748
3749 static void __cancel_balance(struct btrfs_fs_info *fs_info)
3750 {
3751         int ret;
3752
3753         unset_balance_control(fs_info);
3754         ret = del_balance_item(fs_info);
3755         if (ret)
3756                 btrfs_handle_fs_error(fs_info, ret, NULL);
3757
3758         clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
3759 }
3760
3761 /* Non-zero return value signifies invalidity */
3762 static inline int validate_convert_profile(struct btrfs_balance_args *bctl_arg,
3763                 u64 allowed)
3764 {
3765         return ((bctl_arg->flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3766                 (!alloc_profile_is_valid(bctl_arg->target, 1) ||
3767                  (bctl_arg->target & ~allowed)));
3768 }
3769
3770 /*
3771  * Should be called with both balance and volume mutexes held
3772  */
3773 int btrfs_balance(struct btrfs_balance_control *bctl,
3774                   struct btrfs_ioctl_balance_args *bargs)
3775 {
3776         struct btrfs_fs_info *fs_info = bctl->fs_info;
3777         u64 meta_target, data_target;
3778         u64 allowed;
3779         int mixed = 0;
3780         int ret;
3781         u64 num_devices;
3782         unsigned seq;
3783
3784         if (btrfs_fs_closing(fs_info) ||
3785             atomic_read(&fs_info->balance_pause_req) ||
3786             atomic_read(&fs_info->balance_cancel_req)) {
3787                 ret = -EINVAL;
3788                 goto out;
3789         }
3790
3791         allowed = btrfs_super_incompat_flags(fs_info->super_copy);
3792         if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
3793                 mixed = 1;
3794
3795         /*
3796          * In case of mixed groups both data and meta should be picked,
3797          * and identical options should be given for both of them.
3798          */
3799         allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
3800         if (mixed && (bctl->flags & allowed)) {
3801                 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
3802                     !(bctl->flags & BTRFS_BALANCE_METADATA) ||
3803                     memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
3804                         btrfs_err(fs_info,
3805                                   "with mixed groups data and metadata balance options must be the same");
3806                         ret = -EINVAL;
3807                         goto out;
3808                 }
3809         }
3810
3811         num_devices = fs_info->fs_devices->num_devices;
3812         btrfs_dev_replace_lock(&fs_info->dev_replace, 0);
3813         if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
3814                 BUG_ON(num_devices < 1);
3815                 num_devices--;
3816         }
3817         btrfs_dev_replace_unlock(&fs_info->dev_replace, 0);
3818         allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE | BTRFS_BLOCK_GROUP_DUP;
3819         if (num_devices > 1)
3820                 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1);
3821         if (num_devices > 2)
3822                 allowed |= BTRFS_BLOCK_GROUP_RAID5;
3823         if (num_devices > 3)
3824                 allowed |= (BTRFS_BLOCK_GROUP_RAID10 |
3825                             BTRFS_BLOCK_GROUP_RAID6);
3826         if (validate_convert_profile(&bctl->data, allowed)) {
3827                 btrfs_err(fs_info,
3828                           "unable to start balance with target data profile %llu",
3829                           bctl->data.target);
3830                 ret = -EINVAL;
3831                 goto out;
3832         }
3833         if (validate_convert_profile(&bctl->meta, allowed)) {
3834                 btrfs_err(fs_info,
3835                           "unable to start balance with target metadata profile %llu",
3836                           bctl->meta.target);
3837                 ret = -EINVAL;
3838                 goto out;
3839         }
3840         if (validate_convert_profile(&bctl->sys, allowed)) {
3841                 btrfs_err(fs_info,
3842                           "unable to start balance with target system profile %llu",
3843                           bctl->sys.target);
3844                 ret = -EINVAL;
3845                 goto out;
3846         }
3847
3848         /* allow to reduce meta or sys integrity only if force set */
3849         allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3850                         BTRFS_BLOCK_GROUP_RAID10 |
3851                         BTRFS_BLOCK_GROUP_RAID5 |
3852                         BTRFS_BLOCK_GROUP_RAID6;
3853         do {
3854                 seq = read_seqbegin(&fs_info->profiles_lock);
3855
3856                 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3857                      (fs_info->avail_system_alloc_bits & allowed) &&
3858                      !(bctl->sys.target & allowed)) ||
3859                     ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3860                      (fs_info->avail_metadata_alloc_bits & allowed) &&
3861                      !(bctl->meta.target & allowed))) {
3862                         if (bctl->flags & BTRFS_BALANCE_FORCE) {
3863                                 btrfs_info(fs_info,
3864                                            "force reducing metadata integrity");
3865                         } else {
3866                                 btrfs_err(fs_info,
3867                                           "balance will reduce metadata integrity, use force if you want this");
3868                                 ret = -EINVAL;
3869                                 goto out;
3870                         }
3871                 }
3872         } while (read_seqretry(&fs_info->profiles_lock, seq));
3873
3874         /* if we're not converting, the target field is uninitialized */
3875         meta_target = (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
3876                 bctl->meta.target : fs_info->avail_metadata_alloc_bits;
3877         data_target = (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
3878                 bctl->data.target : fs_info->avail_data_alloc_bits;
3879         if (btrfs_get_num_tolerated_disk_barrier_failures(meta_target) <
3880                 btrfs_get_num_tolerated_disk_barrier_failures(data_target)) {
3881                 btrfs_warn(fs_info,
3882                            "metadata profile 0x%llx has lower redundancy than data profile 0x%llx",
3883                            meta_target, data_target);
3884         }
3885
3886         ret = insert_balance_item(fs_info, bctl);
3887         if (ret && ret != -EEXIST)
3888                 goto out;
3889
3890         if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
3891                 BUG_ON(ret == -EEXIST);
3892                 set_balance_control(bctl);
3893         } else {
3894                 BUG_ON(ret != -EEXIST);
3895                 spin_lock(&fs_info->balance_lock);
3896                 update_balance_args(bctl);
3897                 spin_unlock(&fs_info->balance_lock);
3898         }
3899
3900         atomic_inc(&fs_info->balance_running);
3901         mutex_unlock(&fs_info->balance_mutex);
3902
3903         ret = __btrfs_balance(fs_info);
3904
3905         mutex_lock(&fs_info->balance_mutex);
3906         atomic_dec(&fs_info->balance_running);
3907
3908         if (bargs) {
3909                 memset(bargs, 0, sizeof(*bargs));
3910                 update_ioctl_balance_args(fs_info, 0, bargs);
3911         }
3912
3913         if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
3914             balance_need_close(fs_info)) {
3915                 __cancel_balance(fs_info);
3916         }
3917
3918         wake_up(&fs_info->balance_wait_q);
3919
3920         return ret;
3921 out:
3922         if (bctl->flags & BTRFS_BALANCE_RESUME)
3923                 __cancel_balance(fs_info);
3924         else {
3925                 kfree(bctl);
3926                 clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
3927         }
3928         return ret;
3929 }
3930
3931 static int balance_kthread(void *data)
3932 {
3933         struct btrfs_fs_info *fs_info = data;
3934         int ret = 0;
3935
3936         mutex_lock(&fs_info->volume_mutex);
3937         mutex_lock(&fs_info->balance_mutex);
3938
3939         if (fs_info->balance_ctl) {
3940                 btrfs_info(fs_info, "continuing balance");
3941                 ret = btrfs_balance(fs_info->balance_ctl, NULL);
3942         }
3943
3944         mutex_unlock(&fs_info->balance_mutex);
3945         mutex_unlock(&fs_info->volume_mutex);
3946
3947         return ret;
3948 }
3949
3950 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
3951 {
3952         struct task_struct *tsk;
3953
3954         spin_lock(&fs_info->balance_lock);
3955         if (!fs_info->balance_ctl) {
3956                 spin_unlock(&fs_info->balance_lock);
3957                 return 0;
3958         }
3959         spin_unlock(&fs_info->balance_lock);
3960
3961         if (btrfs_test_opt(fs_info, SKIP_BALANCE)) {
3962                 btrfs_info(fs_info, "force skipping balance");
3963                 return 0;
3964         }
3965
3966         tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
3967         return PTR_ERR_OR_ZERO(tsk);
3968 }
3969
3970 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
3971 {
3972         struct btrfs_balance_control *bctl;
3973         struct btrfs_balance_item *item;
3974         struct btrfs_disk_balance_args disk_bargs;
3975         struct btrfs_path *path;
3976         struct extent_buffer *leaf;
3977         struct btrfs_key key;
3978         int ret;
3979
3980         path = btrfs_alloc_path();
3981         if (!path)
3982                 return -ENOMEM;
3983
3984         key.objectid = BTRFS_BALANCE_OBJECTID;
3985         key.type = BTRFS_TEMPORARY_ITEM_KEY;
3986         key.offset = 0;
3987
3988         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
3989         if (ret < 0)
3990                 goto out;
3991         if (ret > 0) { /* ret = -ENOENT; */
3992                 ret = 0;
3993                 goto out;
3994         }
3995
3996         bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
3997         if (!bctl) {
3998                 ret = -ENOMEM;
3999                 goto out;
4000         }
4001
4002         leaf = path->nodes[0];
4003         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
4004
4005         bctl->fs_info = fs_info;
4006         bctl->flags = btrfs_balance_flags(leaf, item);
4007         bctl->flags |= BTRFS_BALANCE_RESUME;
4008
4009         btrfs_balance_data(leaf, item, &disk_bargs);
4010         btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
4011         btrfs_balance_meta(leaf, item, &disk_bargs);
4012         btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
4013         btrfs_balance_sys(leaf, item, &disk_bargs);
4014         btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
4015
4016         WARN_ON(test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags));
4017
4018         mutex_lock(&fs_info->volume_mutex);
4019         mutex_lock(&fs_info->balance_mutex);
4020
4021         set_balance_control(bctl);
4022
4023         mutex_unlock(&fs_info->balance_mutex);
4024         mutex_unlock(&fs_info->volume_mutex);
4025 out:
4026         btrfs_free_path(path);
4027         return ret;
4028 }
4029
4030 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
4031 {
4032         int ret = 0;
4033
4034         mutex_lock(&fs_info->balance_mutex);
4035         if (!fs_info->balance_ctl) {
4036                 mutex_unlock(&fs_info->balance_mutex);
4037                 return -ENOTCONN;
4038         }
4039
4040         if (atomic_read(&fs_info->balance_running)) {
4041                 atomic_inc(&fs_info->balance_pause_req);
4042                 mutex_unlock(&fs_info->balance_mutex);
4043
4044                 wait_event(fs_info->balance_wait_q,
4045                            atomic_read(&fs_info->balance_running) == 0);
4046
4047                 mutex_lock(&fs_info->balance_mutex);
4048                 /* we are good with balance_ctl ripped off from under us */
4049                 BUG_ON(atomic_read(&fs_info->balance_running));
4050                 atomic_dec(&fs_info->balance_pause_req);
4051         } else {
4052                 ret = -ENOTCONN;
4053         }
4054
4055         mutex_unlock(&fs_info->balance_mutex);
4056         return ret;
4057 }
4058
4059 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
4060 {
4061         if (sb_rdonly(fs_info->sb))
4062                 return -EROFS;
4063
4064         mutex_lock(&fs_info->balance_mutex);
4065         if (!fs_info->balance_ctl) {
4066                 mutex_unlock(&fs_info->balance_mutex);
4067                 return -ENOTCONN;
4068         }
4069
4070         atomic_inc(&fs_info->balance_cancel_req);
4071         /*
4072          * if we are running just wait and return, balance item is
4073          * deleted in btrfs_balance in this case
4074          */
4075         if (atomic_read(&fs_info->balance_running)) {
4076                 mutex_unlock(&fs_info->balance_mutex);
4077                 wait_event(fs_info->balance_wait_q,
4078                            atomic_read(&fs_info->balance_running) == 0);
4079                 mutex_lock(&fs_info->balance_mutex);
4080         } else {
4081                 /* __cancel_balance needs volume_mutex */
4082                 mutex_unlock(&fs_info->balance_mutex);
4083                 mutex_lock(&fs_info->volume_mutex);
4084                 mutex_lock(&fs_info->balance_mutex);
4085
4086                 if (fs_info->balance_ctl)
4087                         __cancel_balance(fs_info);
4088
4089                 mutex_unlock(&fs_info->volume_mutex);
4090         }
4091
4092         BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running));
4093         atomic_dec(&fs_info->balance_cancel_req);
4094         mutex_unlock(&fs_info->balance_mutex);
4095         return 0;
4096 }
4097
4098 static int btrfs_uuid_scan_kthread(void *data)
4099 {
4100         struct btrfs_fs_info *fs_info = data;
4101         struct btrfs_root *root = fs_info->tree_root;
4102         struct btrfs_key key;
4103         struct btrfs_path *path = NULL;
4104         int ret = 0;
4105         struct extent_buffer *eb;
4106         int slot;
4107         struct btrfs_root_item root_item;
4108         u32 item_size;
4109         struct btrfs_trans_handle *trans = NULL;
4110
4111         path = btrfs_alloc_path();
4112         if (!path) {
4113                 ret = -ENOMEM;
4114                 goto out;
4115         }
4116
4117         key.objectid = 0;
4118         key.type = BTRFS_ROOT_ITEM_KEY;
4119         key.offset = 0;
4120
4121         while (1) {
4122                 ret = btrfs_search_forward(root, &key, path, 0);
4123                 if (ret) {
4124                         if (ret > 0)
4125                                 ret = 0;
4126                         break;
4127                 }
4128
4129                 if (key.type != BTRFS_ROOT_ITEM_KEY ||
4130                     (key.objectid < BTRFS_FIRST_FREE_OBJECTID &&
4131                      key.objectid != BTRFS_FS_TREE_OBJECTID) ||
4132                     key.objectid > BTRFS_LAST_FREE_OBJECTID)
4133                         goto skip;
4134
4135                 eb = path->nodes[0];
4136                 slot = path->slots[0];
4137                 item_size = btrfs_item_size_nr(eb, slot);
4138                 if (item_size < sizeof(root_item))
4139                         goto skip;
4140
4141                 read_extent_buffer(eb, &root_item,
4142                                    btrfs_item_ptr_offset(eb, slot),
4143                                    (int)sizeof(root_item));
4144                 if (btrfs_root_refs(&root_item) == 0)
4145                         goto skip;
4146
4147                 if (!btrfs_is_empty_uuid(root_item.uuid) ||
4148                     !btrfs_is_empty_uuid(root_item.received_uuid)) {
4149                         if (trans)
4150                                 goto update_tree;
4151
4152                         btrfs_release_path(path);
4153                         /*
4154                          * 1 - subvol uuid item
4155                          * 1 - received_subvol uuid item
4156                          */
4157                         trans = btrfs_start_transaction(fs_info->uuid_root, 2);
4158                         if (IS_ERR(trans)) {
4159                                 ret = PTR_ERR(trans);
4160                                 break;
4161                         }
4162                         continue;
4163                 } else {
4164                         goto skip;
4165                 }
4166 update_tree:
4167                 if (!btrfs_is_empty_uuid(root_item.uuid)) {
4168                         ret = btrfs_uuid_tree_add(trans, fs_info,
4169                                                   root_item.uuid,
4170                                                   BTRFS_UUID_KEY_SUBVOL,
4171                                                   key.objectid);
4172                         if (ret < 0) {
4173                                 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4174                                         ret);
4175                                 break;
4176                         }
4177                 }
4178
4179                 if (!btrfs_is_empty_uuid(root_item.received_uuid)) {
4180                         ret = btrfs_uuid_tree_add(trans, fs_info,
4181                                                   root_item.received_uuid,
4182                                                  BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4183                                                   key.objectid);
4184                         if (ret < 0) {
4185                                 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4186                                         ret);
4187                                 break;
4188                         }
4189                 }
4190
4191 skip:
4192                 if (trans) {
4193                         ret = btrfs_end_transaction(trans);
4194                         trans = NULL;
4195                         if (ret)
4196                                 break;
4197                 }
4198
4199                 btrfs_release_path(path);
4200                 if (key.offset < (u64)-1) {
4201                         key.offset++;
4202                 } else if (key.type < BTRFS_ROOT_ITEM_KEY) {
4203                         key.offset = 0;
4204                         key.type = BTRFS_ROOT_ITEM_KEY;
4205                 } else if (key.objectid < (u64)-1) {
4206                         key.offset = 0;
4207                         key.type = BTRFS_ROOT_ITEM_KEY;
4208                         key.objectid++;
4209                 } else {
4210                         break;
4211                 }
4212                 cond_resched();
4213         }
4214
4215 out:
4216         btrfs_free_path(path);
4217         if (trans && !IS_ERR(trans))
4218                 btrfs_end_transaction(trans);
4219         if (ret)
4220                 btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret);
4221         else
4222                 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
4223         up(&fs_info->uuid_tree_rescan_sem);
4224         return 0;
4225 }
4226
4227 /*
4228  * Callback for btrfs_uuid_tree_iterate().
4229  * returns:
4230  * 0    check succeeded, the entry is not outdated.
4231  * < 0  if an error occurred.
4232  * > 0  if the check failed, which means the caller shall remove the entry.
4233  */
4234 static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info *fs_info,
4235                                        u8 *uuid, u8 type, u64 subid)
4236 {
4237         struct btrfs_key key;
4238         int ret = 0;
4239         struct btrfs_root *subvol_root;
4240
4241         if (type != BTRFS_UUID_KEY_SUBVOL &&
4242             type != BTRFS_UUID_KEY_RECEIVED_SUBVOL)
4243                 goto out;
4244
4245         key.objectid = subid;
4246         key.type = BTRFS_ROOT_ITEM_KEY;
4247         key.offset = (u64)-1;
4248         subvol_root = btrfs_read_fs_root_no_name(fs_info, &key);
4249         if (IS_ERR(subvol_root)) {
4250                 ret = PTR_ERR(subvol_root);
4251                 if (ret == -ENOENT)
4252                         ret = 1;
4253                 goto out;
4254         }
4255
4256         switch (type) {
4257         case BTRFS_UUID_KEY_SUBVOL:
4258                 if (memcmp(uuid, subvol_root->root_item.uuid, BTRFS_UUID_SIZE))
4259                         ret = 1;
4260                 break;
4261         case BTRFS_UUID_KEY_RECEIVED_SUBVOL:
4262                 if (memcmp(uuid, subvol_root->root_item.received_uuid,
4263                            BTRFS_UUID_SIZE))
4264                         ret = 1;
4265                 break;
4266         }
4267
4268 out:
4269         return ret;
4270 }
4271
4272 static int btrfs_uuid_rescan_kthread(void *data)
4273 {
4274         struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data;
4275         int ret;
4276
4277         /*
4278          * 1st step is to iterate through the existing UUID tree and
4279          * to delete all entries that contain outdated data.
4280          * 2nd step is to add all missing entries to the UUID tree.
4281          */
4282         ret = btrfs_uuid_tree_iterate(fs_info, btrfs_check_uuid_tree_entry);
4283         if (ret < 0) {
4284                 btrfs_warn(fs_info, "iterating uuid_tree failed %d", ret);
4285                 up(&fs_info->uuid_tree_rescan_sem);
4286                 return ret;
4287         }
4288         return btrfs_uuid_scan_kthread(data);
4289 }
4290
4291 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info)
4292 {
4293         struct btrfs_trans_handle *trans;
4294         struct btrfs_root *tree_root = fs_info->tree_root;
4295         struct btrfs_root *uuid_root;
4296         struct task_struct *task;
4297         int ret;
4298
4299         /*
4300          * 1 - root node
4301          * 1 - root item
4302          */
4303         trans = btrfs_start_transaction(tree_root, 2);
4304         if (IS_ERR(trans))
4305                 return PTR_ERR(trans);
4306
4307         uuid_root = btrfs_create_tree(trans, fs_info,
4308                                       BTRFS_UUID_TREE_OBJECTID);
4309         if (IS_ERR(uuid_root)) {
4310                 ret = PTR_ERR(uuid_root);
4311                 btrfs_abort_transaction(trans, ret);
4312                 btrfs_end_transaction(trans);
4313                 return ret;
4314         }
4315
4316         fs_info->uuid_root = uuid_root;
4317
4318         ret = btrfs_commit_transaction(trans);
4319         if (ret)
4320                 return ret;
4321
4322         down(&fs_info->uuid_tree_rescan_sem);
4323         task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid");
4324         if (IS_ERR(task)) {
4325                 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4326                 btrfs_warn(fs_info, "failed to start uuid_scan task");
4327                 up(&fs_info->uuid_tree_rescan_sem);
4328                 return PTR_ERR(task);
4329         }
4330
4331         return 0;
4332 }
4333
4334 int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
4335 {
4336         struct task_struct *task;
4337
4338         down(&fs_info->uuid_tree_rescan_sem);
4339         task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
4340         if (IS_ERR(task)) {
4341                 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4342                 btrfs_warn(fs_info, "failed to start uuid_rescan task");
4343                 up(&fs_info->uuid_tree_rescan_sem);
4344                 return PTR_ERR(task);
4345         }
4346
4347         return 0;
4348 }
4349
4350 /*
4351  * shrinking a device means finding all of the device extents past
4352  * the new size, and then following the back refs to the chunks.
4353  * The chunk relocation code actually frees the device extent
4354  */
4355 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
4356 {
4357         struct btrfs_fs_info *fs_info = device->fs_info;
4358         struct btrfs_root *root = fs_info->dev_root;
4359         struct btrfs_trans_handle *trans;
4360         struct btrfs_dev_extent *dev_extent = NULL;
4361         struct btrfs_path *path;
4362         u64 length;
4363         u64 chunk_offset;
4364         int ret;
4365         int slot;
4366         int failed = 0;
4367         bool retried = false;
4368         bool checked_pending_chunks = false;
4369         struct extent_buffer *l;
4370         struct btrfs_key key;
4371         struct btrfs_super_block *super_copy = fs_info->super_copy;
4372         u64 old_total = btrfs_super_total_bytes(super_copy);
4373         u64 old_size = btrfs_device_get_total_bytes(device);
4374         u64 diff;
4375
4376         new_size = round_down(new_size, fs_info->sectorsize);
4377         diff = round_down(old_size - new_size, fs_info->sectorsize);
4378
4379         if (device->is_tgtdev_for_dev_replace)
4380                 return -EINVAL;
4381
4382         path = btrfs_alloc_path();
4383         if (!path)
4384                 return -ENOMEM;
4385
4386         path->reada = READA_FORWARD;
4387
4388         mutex_lock(&fs_info->chunk_mutex);
4389
4390         btrfs_device_set_total_bytes(device, new_size);
4391         if (device->writeable) {
4392                 device->fs_devices->total_rw_bytes -= diff;
4393                 atomic64_sub(diff, &fs_info->free_chunk_space);
4394         }
4395         mutex_unlock(&fs_info->chunk_mutex);
4396
4397 again:
4398         key.objectid = device->devid;
4399         key.offset = (u64)-1;
4400         key.type = BTRFS_DEV_EXTENT_KEY;
4401
4402         do {
4403                 mutex_lock(&fs_info->delete_unused_bgs_mutex);
4404                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4405                 if (ret < 0) {
4406                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4407                         goto done;
4408                 }
4409
4410                 ret = btrfs_previous_item(root, path, 0, key.type);
4411                 if (ret)
4412                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4413                 if (ret < 0)
4414                         goto done;
4415                 if (ret) {
4416                         ret = 0;
4417                         btrfs_release_path(path);
4418                         break;
4419                 }
4420
4421                 l = path->nodes[0];
4422                 slot = path->slots[0];
4423                 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
4424
4425                 if (key.objectid != device->devid) {
4426                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4427                         btrfs_release_path(path);
4428                         break;
4429                 }
4430
4431                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
4432                 length = btrfs_dev_extent_length(l, dev_extent);
4433
4434                 if (key.offset + length <= new_size) {
4435                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4436                         btrfs_release_path(path);
4437                         break;
4438                 }
4439
4440                 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
4441                 btrfs_release_path(path);
4442
4443                 ret = btrfs_relocate_chunk(fs_info, chunk_offset);
4444                 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4445                 if (ret && ret != -ENOSPC)
4446                         goto done;
4447                 if (ret == -ENOSPC)
4448                         failed++;
4449         } while (key.offset-- > 0);
4450
4451         if (failed && !retried) {
4452                 failed = 0;
4453                 retried = true;
4454                 goto again;
4455         } else if (failed && retried) {
4456                 ret = -ENOSPC;
4457                 goto done;
4458         }
4459
4460         /* Shrinking succeeded, else we would be at "done". */
4461         trans = btrfs_start_transaction(root, 0);
4462         if (IS_ERR(trans)) {
4463                 ret = PTR_ERR(trans);
4464                 goto done;
4465         }
4466
4467         mutex_lock(&fs_info->chunk_mutex);
4468
4469         /*
4470          * We checked in the above loop all device extents that were already in
4471          * the device tree. However before we have updated the device's
4472          * total_bytes to the new size, we might have had chunk allocations that
4473          * have not complete yet (new block groups attached to transaction
4474          * handles), and therefore their device extents were not yet in the
4475          * device tree and we missed them in the loop above. So if we have any
4476          * pending chunk using a device extent that overlaps the device range
4477          * that we can not use anymore, commit the current transaction and
4478          * repeat the search on the device tree - this way we guarantee we will
4479          * not have chunks using device extents that end beyond 'new_size'.
4480          */
4481         if (!checked_pending_chunks) {
4482                 u64 start = new_size;
4483                 u64 len = old_size - new_size;
4484
4485                 if (contains_pending_extent(trans->transaction, device,
4486                                             &start, len)) {
4487                         mutex_unlock(&fs_info->chunk_mutex);
4488                         checked_pending_chunks = true;
4489                         failed = 0;
4490                         retried = false;
4491                         ret = btrfs_commit_transaction(trans);
4492                         if (ret)
4493                                 goto done;
4494                         goto again;
4495                 }
4496         }
4497
4498         btrfs_device_set_disk_total_bytes(device, new_size);
4499         if (list_empty(&device->resized_list))
4500                 list_add_tail(&device->resized_list,
4501                               &fs_info->fs_devices->resized_devices);
4502
4503         WARN_ON(diff > old_total);
4504         btrfs_set_super_total_bytes(super_copy,
4505                         round_down(old_total - diff, fs_info->sectorsize));
4506         mutex_unlock(&fs_info->chunk_mutex);
4507
4508         /* Now btrfs_update_device() will change the on-disk size. */
4509         ret = btrfs_update_device(trans, device);
4510         btrfs_end_transaction(trans);
4511 done:
4512         btrfs_free_path(path);
4513         if (ret) {
4514                 mutex_lock(&fs_info->chunk_mutex);
4515                 btrfs_device_set_total_bytes(device, old_size);
4516                 if (device->writeable)
4517                         device->fs_devices->total_rw_bytes += diff;
4518                 atomic64_add(diff, &fs_info->free_chunk_space);
4519                 mutex_unlock(&fs_info->chunk_mutex);
4520         }
4521         return ret;
4522 }
4523
4524 static int btrfs_add_system_chunk(struct btrfs_fs_info *fs_info,
4525                            struct btrfs_key *key,
4526                            struct btrfs_chunk *chunk, int item_size)
4527 {
4528         struct btrfs_super_block *super_copy = fs_info->super_copy;
4529         struct btrfs_disk_key disk_key;
4530         u32 array_size;
4531         u8 *ptr;
4532
4533         mutex_lock(&fs_info->chunk_mutex);
4534         array_size = btrfs_super_sys_array_size(super_copy);
4535         if (array_size + item_size + sizeof(disk_key)
4536                         > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
4537                 mutex_unlock(&fs_info->chunk_mutex);
4538                 return -EFBIG;
4539         }
4540
4541         ptr = super_copy->sys_chunk_array + array_size;
4542         btrfs_cpu_key_to_disk(&disk_key, key);
4543         memcpy(ptr, &disk_key, sizeof(disk_key));
4544         ptr += sizeof(disk_key);
4545         memcpy(ptr, chunk, item_size);
4546         item_size += sizeof(disk_key);
4547         btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
4548         mutex_unlock(&fs_info->chunk_mutex);
4549
4550         return 0;
4551 }
4552
4553 /*
4554  * sort the devices in descending order by max_avail, total_avail
4555  */
4556 static int btrfs_cmp_device_info(const void *a, const void *b)
4557 {
4558         const struct btrfs_device_info *di_a = a;
4559         const struct btrfs_device_info *di_b = b;
4560
4561         if (di_a->max_avail > di_b->max_avail)
4562                 return -1;
4563         if (di_a->max_avail < di_b->max_avail)
4564                 return 1;
4565         if (di_a->total_avail > di_b->total_avail)
4566                 return -1;
4567         if (di_a->total_avail < di_b->total_avail)
4568                 return 1;
4569         return 0;
4570 }
4571
4572 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
4573 {
4574         if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK))
4575                 return;
4576
4577         btrfs_set_fs_incompat(info, RAID56);
4578 }
4579
4580 #define BTRFS_MAX_DEVS(r) ((BTRFS_MAX_ITEM_SIZE(r->fs_info)             \
4581                         - sizeof(struct btrfs_chunk))           \
4582                         / sizeof(struct btrfs_stripe) + 1)
4583
4584 #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE        \
4585                                 - 2 * sizeof(struct btrfs_disk_key)     \
4586                                 - 2 * sizeof(struct btrfs_chunk))       \
4587                                 / sizeof(struct btrfs_stripe) + 1)
4588
4589 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4590                                u64 start, u64 type)
4591 {
4592         struct btrfs_fs_info *info = trans->fs_info;
4593         struct btrfs_fs_devices *fs_devices = info->fs_devices;
4594         struct btrfs_device *device;
4595         struct map_lookup *map = NULL;
4596         struct extent_map_tree *em_tree;
4597         struct extent_map *em;
4598         struct btrfs_device_info *devices_info = NULL;
4599         u64 total_avail;
4600         int num_stripes;        /* total number of stripes to allocate */
4601         int data_stripes;       /* number of stripes that count for
4602                                    block group size */
4603         int sub_stripes;        /* sub_stripes info for map */
4604         int dev_stripes;        /* stripes per dev */
4605         int devs_max;           /* max devs to use */
4606         int devs_min;           /* min devs needed */
4607         int devs_increment;     /* ndevs has to be a multiple of this */
4608         int ncopies;            /* how many copies to data has */
4609         int ret;
4610         u64 max_stripe_size;
4611         u64 max_chunk_size;
4612         u64 stripe_size;
4613         u64 num_bytes;
4614         int ndevs;
4615         int i;
4616         int j;
4617         int index;
4618
4619         BUG_ON(!alloc_profile_is_valid(type, 0));
4620
4621         if (list_empty(&fs_devices->alloc_list))
4622                 return -ENOSPC;
4623
4624         index = __get_raid_index(type);
4625
4626         sub_stripes = btrfs_raid_array[index].sub_stripes;
4627         dev_stripes = btrfs_raid_array[index].dev_stripes;
4628         devs_max = btrfs_raid_array[index].devs_max;
4629         devs_min = btrfs_raid_array[index].devs_min;
4630         devs_increment = btrfs_raid_array[index].devs_increment;
4631         ncopies = btrfs_raid_array[index].ncopies;
4632
4633         if (type & BTRFS_BLOCK_GROUP_DATA) {
4634                 max_stripe_size = SZ_1G;
4635                 max_chunk_size = 10 * max_stripe_size;
4636                 if (!devs_max)
4637                         devs_max = BTRFS_MAX_DEVS(info->chunk_root);
4638         } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
4639                 /* for larger filesystems, use larger metadata chunks */
4640                 if (fs_devices->total_rw_bytes > 50ULL * SZ_1G)
4641                         max_stripe_size = SZ_1G;
4642                 else
4643                         max_stripe_size = SZ_256M;
4644                 max_chunk_size = max_stripe_size;
4645                 if (!devs_max)
4646                         devs_max = BTRFS_MAX_DEVS(info->chunk_root);
4647         } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
4648                 max_stripe_size = SZ_32M;
4649                 max_chunk_size = 2 * max_stripe_size;
4650                 if (!devs_max)
4651                         devs_max = BTRFS_MAX_DEVS_SYS_CHUNK;
4652         } else {
4653                 btrfs_err(info, "invalid chunk type 0x%llx requested",
4654                        type);
4655                 BUG_ON(1);
4656         }
4657
4658         /* we don't want a chunk larger than 10% of writeable space */
4659         max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
4660                              max_chunk_size);
4661
4662         devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info),
4663                                GFP_NOFS);
4664         if (!devices_info)
4665                 return -ENOMEM;
4666
4667         /*
4668          * in the first pass through the devices list, we gather information
4669          * about the available holes on each device.
4670          */
4671         ndevs = 0;
4672         list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
4673                 u64 max_avail;
4674                 u64 dev_offset;
4675
4676                 if (!device->writeable) {
4677                         WARN(1, KERN_ERR
4678                                "BTRFS: read-only device in alloc_list\n");
4679                         continue;
4680                 }
4681
4682                 if (!device->in_fs_metadata ||
4683                     device->is_tgtdev_for_dev_replace)
4684                         continue;
4685
4686                 if (device->total_bytes > device->bytes_used)
4687                         total_avail = device->total_bytes - device->bytes_used;
4688                 else
4689                         total_avail = 0;
4690
4691                 /* If there is no space on this device, skip it. */
4692                 if (total_avail == 0)
4693                         continue;
4694
4695                 ret = find_free_dev_extent(trans, device,
4696                                            max_stripe_size * dev_stripes,
4697                                            &dev_offset, &max_avail);
4698                 if (ret && ret != -ENOSPC)
4699                         goto error;
4700
4701                 if (ret == 0)
4702                         max_avail = max_stripe_size * dev_stripes;
4703
4704                 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes)
4705                         continue;
4706
4707                 if (ndevs == fs_devices->rw_devices) {
4708                         WARN(1, "%s: found more than %llu devices\n",
4709                              __func__, fs_devices->rw_devices);
4710                         break;
4711                 }
4712                 devices_info[ndevs].dev_offset = dev_offset;
4713                 devices_info[ndevs].max_avail = max_avail;
4714                 devices_info[ndevs].total_avail = total_avail;
4715                 devices_info[ndevs].dev = device;
4716                 ++ndevs;
4717         }
4718
4719         /*
4720          * now sort the devices by hole size / available space
4721          */
4722         sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
4723              btrfs_cmp_device_info, NULL);
4724
4725         /* round down to number of usable stripes */
4726         ndevs = round_down(ndevs, devs_increment);
4727
4728         if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) {
4729                 ret = -ENOSPC;
4730                 goto error;
4731         }
4732
4733         ndevs = min(ndevs, devs_max);
4734
4735         /*
4736          * the primary goal is to maximize the number of stripes, so use as many
4737          * devices as possible, even if the stripes are not maximum sized.
4738          */
4739         stripe_size = devices_info[ndevs-1].max_avail;
4740         num_stripes = ndevs * dev_stripes;
4741
4742         /*
4743          * this will have to be fixed for RAID1 and RAID10 over
4744          * more drives
4745          */
4746         data_stripes = num_stripes / ncopies;
4747
4748         if (type & BTRFS_BLOCK_GROUP_RAID5)
4749                 data_stripes = num_stripes - 1;
4750
4751         if (type & BTRFS_BLOCK_GROUP_RAID6)
4752                 data_stripes = num_stripes - 2;
4753
4754         /*
4755          * Use the number of data stripes to figure out how big this chunk
4756          * is really going to be in terms of logical address space,
4757          * and compare that answer with the max chunk size
4758          */
4759         if (stripe_size * data_stripes > max_chunk_size) {
4760                 u64 mask = (1ULL << 24) - 1;
4761
4762                 stripe_size = div_u64(max_chunk_size, data_stripes);
4763
4764                 /* bump the answer up to a 16MB boundary */
4765                 stripe_size = (stripe_size + mask) & ~mask;
4766
4767                 /* but don't go higher than the limits we found
4768                  * while searching for free extents
4769                  */
4770                 if (stripe_size > devices_info[ndevs-1].max_avail)
4771                         stripe_size = devices_info[ndevs-1].max_avail;
4772         }
4773
4774         stripe_size = div_u64(stripe_size, dev_stripes);
4775
4776         /* align to BTRFS_STRIPE_LEN */
4777         stripe_size = round_down(stripe_size, BTRFS_STRIPE_LEN);
4778
4779         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
4780         if (!map) {
4781                 ret = -ENOMEM;
4782                 goto error;
4783         }
4784         map->num_stripes = num_stripes;
4785
4786         for (i = 0; i < ndevs; ++i) {
4787                 for (j = 0; j < dev_stripes; ++j) {
4788                         int s = i * dev_stripes + j;
4789                         map->stripes[s].dev = devices_info[i].dev;
4790                         map->stripes[s].physical = devices_info[i].dev_offset +
4791                                                    j * stripe_size;
4792                 }
4793         }
4794         map->stripe_len = BTRFS_STRIPE_LEN;
4795         map->io_align = BTRFS_STRIPE_LEN;
4796         map->io_width = BTRFS_STRIPE_LEN;
4797         map->type = type;
4798         map->sub_stripes = sub_stripes;
4799
4800         num_bytes = stripe_size * data_stripes;
4801
4802         trace_btrfs_chunk_alloc(info, map, start, num_bytes);
4803
4804         em = alloc_extent_map();
4805         if (!em) {
4806                 kfree(map);
4807                 ret = -ENOMEM;
4808                 goto error;
4809         }
4810         set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
4811         em->map_lookup = map;
4812         em->start = start;
4813         em->len = num_bytes;
4814         em->block_start = 0;
4815         em->block_len = em->len;
4816         em->orig_block_len = stripe_size;
4817
4818         em_tree = &info->mapping_tree.map_tree;
4819         write_lock(&em_tree->lock);
4820         ret = add_extent_mapping(em_tree, em, 0);
4821         if (!ret) {
4822                 list_add_tail(&em->list, &trans->transaction->pending_chunks);
4823                 refcount_inc(&em->refs);
4824         }
4825         write_unlock(&em_tree->lock);
4826         if (ret) {
4827                 free_extent_map(em);
4828                 goto error;
4829         }
4830
4831         ret = btrfs_make_block_group(trans, info, 0, type, start, num_bytes);
4832         if (ret)
4833                 goto error_del_extent;
4834
4835         for (i = 0; i < map->num_stripes; i++) {
4836                 num_bytes = map->stripes[i].dev->bytes_used + stripe_size;
4837                 btrfs_device_set_bytes_used(map->stripes[i].dev, num_bytes);
4838         }
4839
4840         atomic64_sub(stripe_size * map->num_stripes, &info->free_chunk_space);
4841
4842         free_extent_map(em);
4843         check_raid56_incompat_flag(info, type);
4844
4845         kfree(devices_info);
4846         return 0;
4847
4848 error_del_extent:
4849         write_lock(&em_tree->lock);
4850         remove_extent_mapping(em_tree, em);
4851         write_unlock(&em_tree->lock);
4852
4853         /* One for our allocation */
4854         free_extent_map(em);
4855         /* One for the tree reference */
4856         free_extent_map(em);
4857         /* One for the pending_chunks list reference */
4858         free_extent_map(em);
4859 error:
4860         kfree(devices_info);
4861         return ret;
4862 }
4863
4864 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans,
4865                                 struct btrfs_fs_info *fs_info,
4866                                 u64 chunk_offset, u64 chunk_size)
4867 {
4868         struct btrfs_root *extent_root = fs_info->extent_root;
4869         struct btrfs_root *chunk_root = fs_info->chunk_root;
4870         struct btrfs_key key;
4871         struct btrfs_device *device;
4872         struct btrfs_chunk *chunk;
4873         struct btrfs_stripe *stripe;
4874         struct extent_map *em;
4875         struct map_lookup *map;
4876         size_t item_size;
4877         u64 dev_offset;
4878         u64 stripe_size;
4879         int i = 0;
4880         int ret = 0;
4881
4882         em = get_chunk_map(fs_info, chunk_offset, chunk_size);
4883         if (IS_ERR(em))
4884                 return PTR_ERR(em);
4885
4886         map = em->map_lookup;
4887         item_size = btrfs_chunk_item_size(map->num_stripes);
4888         stripe_size = em->orig_block_len;
4889
4890         chunk = kzalloc(item_size, GFP_NOFS);
4891         if (!chunk) {
4892                 ret = -ENOMEM;
4893                 goto out;
4894         }
4895
4896         /*
4897          * Take the device list mutex to prevent races with the final phase of
4898          * a device replace operation that replaces the device object associated
4899          * with the map's stripes, because the device object's id can change
4900          * at any time during that final phase of the device replace operation
4901          * (dev-replace.c:btrfs_dev_replace_finishing()).
4902          */
4903         mutex_lock(&fs_info->fs_devices->device_list_mutex);
4904         for (i = 0; i < map->num_stripes; i++) {
4905                 device = map->stripes[i].dev;
4906                 dev_offset = map->stripes[i].physical;
4907
4908                 ret = btrfs_update_device(trans, device);
4909                 if (ret)
4910                         break;
4911                 ret = btrfs_alloc_dev_extent(trans, device, chunk_offset,
4912                                              dev_offset, stripe_size);
4913                 if (ret)
4914                         break;
4915         }
4916         if (ret) {
4917                 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4918                 goto out;
4919         }
4920
4921         stripe = &chunk->stripe;
4922         for (i = 0; i < map->num_stripes; i++) {
4923                 device = map->stripes[i].dev;
4924                 dev_offset = map->stripes[i].physical;
4925
4926                 btrfs_set_stack_stripe_devid(stripe, device->devid);
4927                 btrfs_set_stack_stripe_offset(stripe, dev_offset);
4928                 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
4929                 stripe++;
4930         }
4931         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4932
4933         btrfs_set_stack_chunk_length(chunk, chunk_size);
4934         btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
4935         btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
4936         btrfs_set_stack_chunk_type(chunk, map->type);
4937         btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
4938         btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
4939         btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
4940         btrfs_set_stack_chunk_sector_size(chunk, fs_info->sectorsize);
4941         btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
4942
4943         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
4944         key.type = BTRFS_CHUNK_ITEM_KEY;
4945         key.offset = chunk_offset;
4946
4947         ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
4948         if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
4949                 /*
4950                  * TODO: Cleanup of inserted chunk root in case of
4951                  * failure.
4952                  */
4953                 ret = btrfs_add_system_chunk(fs_info, &key, chunk, item_size);
4954         }
4955
4956 out:
4957         kfree(chunk);
4958         free_extent_map(em);
4959         return ret;
4960 }
4961
4962 /*
4963  * Chunk allocation falls into two parts. The first part does works
4964  * that make the new allocated chunk useable, but not do any operation
4965  * that modifies the chunk tree. The second part does the works that
4966  * require modifying the chunk tree. This division is important for the
4967  * bootstrap process of adding storage to a seed btrfs.
4968  */
4969 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4970                       struct btrfs_fs_info *fs_info, u64 type)
4971 {
4972         u64 chunk_offset;
4973
4974         ASSERT(mutex_is_locked(&fs_info->chunk_mutex));
4975         chunk_offset = find_next_chunk(fs_info);
4976         return __btrfs_alloc_chunk(trans, chunk_offset, type);
4977 }
4978
4979 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
4980                                          struct btrfs_fs_info *fs_info)
4981 {
4982         u64 chunk_offset;
4983         u64 sys_chunk_offset;
4984         u64 alloc_profile;
4985         int ret;
4986
4987         chunk_offset = find_next_chunk(fs_info);
4988         alloc_profile = btrfs_metadata_alloc_profile(fs_info);
4989         ret = __btrfs_alloc_chunk(trans, chunk_offset, alloc_profile);
4990         if (ret)
4991                 return ret;
4992
4993         sys_chunk_offset = find_next_chunk(fs_info);
4994         alloc_profile = btrfs_system_alloc_profile(fs_info);
4995         ret = __btrfs_alloc_chunk(trans, sys_chunk_offset, alloc_profile);
4996         return ret;
4997 }
4998
4999 static inline int btrfs_chunk_max_errors(struct map_lookup *map)
5000 {
5001         int max_errors;
5002
5003         if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
5004                          BTRFS_BLOCK_GROUP_RAID10 |
5005                          BTRFS_BLOCK_GROUP_RAID5 |
5006                          BTRFS_BLOCK_GROUP_DUP)) {
5007                 max_errors = 1;
5008         } else if (map->type & BTRFS_BLOCK_GROUP_RAID6) {
5009                 max_errors = 2;
5010         } else {
5011                 max_errors = 0;
5012         }
5013
5014         return max_errors;
5015 }
5016
5017 int btrfs_chunk_readonly(struct btrfs_fs_info *fs_info, u64 chunk_offset)
5018 {
5019         struct extent_map *em;
5020         struct map_lookup *map;
5021         int readonly = 0;
5022         int miss_ndevs = 0;
5023         int i;
5024
5025         em = get_chunk_map(fs_info, chunk_offset, 1);
5026         if (IS_ERR(em))
5027                 return 1;
5028
5029         map = em->map_lookup;
5030         for (i = 0; i < map->num_stripes; i++) {
5031                 if (map->stripes[i].dev->missing) {
5032                         miss_ndevs++;
5033                         continue;
5034                 }
5035
5036                 if (!map->stripes[i].dev->writeable) {
5037                         readonly = 1;
5038                         goto end;
5039                 }
5040         }
5041
5042         /*
5043          * If the number of missing devices is larger than max errors,
5044          * we can not write the data into that chunk successfully, so
5045          * set it readonly.
5046          */
5047         if (miss_ndevs > btrfs_chunk_max_errors(map))
5048                 readonly = 1;
5049 end:
5050         free_extent_map(em);
5051         return readonly;
5052 }
5053
5054 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
5055 {
5056         extent_map_tree_init(&tree->map_tree);
5057 }
5058
5059 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
5060 {
5061         struct extent_map *em;
5062
5063         while (1) {
5064                 write_lock(&tree->map_tree.lock);
5065                 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
5066                 if (em)
5067                         remove_extent_mapping(&tree->map_tree, em);
5068                 write_unlock(&tree->map_tree.lock);
5069                 if (!em)
5070                         break;
5071                 /* once for us */
5072                 free_extent_map(em);
5073                 /* once for the tree */
5074                 free_extent_map(em);
5075         }
5076 }
5077
5078 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5079 {
5080         struct extent_map *em;
5081         struct map_lookup *map;
5082         int ret;
5083
5084         em = get_chunk_map(fs_info, logical, len);
5085         if (IS_ERR(em))
5086                 /*
5087                  * We could return errors for these cases, but that could get
5088                  * ugly and we'd probably do the same thing which is just not do
5089                  * anything else and exit, so return 1 so the callers don't try
5090                  * to use other copies.
5091                  */
5092                 return 1;
5093
5094         map = em->map_lookup;
5095         if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
5096                 ret = map->num_stripes;
5097         else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5098                 ret = map->sub_stripes;
5099         else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
5100                 ret = 2;
5101         else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5102                 ret = 3;
5103         else
5104                 ret = 1;
5105         free_extent_map(em);
5106
5107         btrfs_dev_replace_lock(&fs_info->dev_replace, 0);
5108         if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace) &&
5109             fs_info->dev_replace.tgtdev)
5110                 ret++;
5111         btrfs_dev_replace_unlock(&fs_info->dev_replace, 0);
5112
5113         return ret;
5114 }
5115
5116 unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info,
5117                                     u64 logical)
5118 {
5119         struct extent_map *em;
5120         struct map_lookup *map;
5121         unsigned long len = fs_info->sectorsize;
5122
5123         em = get_chunk_map(fs_info, logical, len);
5124
5125         if (!WARN_ON(IS_ERR(em))) {
5126                 map = em->map_lookup;
5127                 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5128                         len = map->stripe_len * nr_data_stripes(map);
5129                 free_extent_map(em);
5130         }
5131         return len;
5132 }
5133
5134 int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5135 {
5136         struct extent_map *em;
5137         struct map_lookup *map;
5138         int ret = 0;
5139
5140         em = get_chunk_map(fs_info, logical, len);
5141
5142         if(!WARN_ON(IS_ERR(em))) {
5143                 map = em->map_lookup;
5144                 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5145                         ret = 1;
5146                 free_extent_map(em);
5147         }
5148         return ret;
5149 }
5150
5151 static int find_live_mirror(struct btrfs_fs_info *fs_info,
5152                             struct map_lookup *map, int first, int num,
5153                             int optimal, int dev_replace_is_ongoing)
5154 {
5155         int i;
5156         int tolerance;
5157         struct btrfs_device *srcdev;
5158
5159         if (dev_replace_is_ongoing &&
5160             fs_info->dev_replace.cont_reading_from_srcdev_mode ==
5161              BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
5162                 srcdev = fs_info->dev_replace.srcdev;
5163         else
5164                 srcdev = NULL;
5165
5166         /*
5167          * try to avoid the drive that is the source drive for a
5168          * dev-replace procedure, only choose it if no other non-missing
5169          * mirror is available
5170          */
5171         for (tolerance = 0; tolerance < 2; tolerance++) {
5172                 if (map->stripes[optimal].dev->bdev &&
5173                     (tolerance || map->stripes[optimal].dev != srcdev))
5174                         return optimal;
5175                 for (i = first; i < first + num; i++) {
5176                         if (map->stripes[i].dev->bdev &&
5177                             (tolerance || map->stripes[i].dev != srcdev))
5178                                 return i;
5179                 }
5180         }
5181
5182         /* we couldn't find one that doesn't fail.  Just return something
5183          * and the io error handling code will clean up eventually
5184          */
5185         return optimal;
5186 }
5187
5188 static inline int parity_smaller(u64 a, u64 b)
5189 {
5190         return a > b;
5191 }
5192
5193 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
5194 static void sort_parity_stripes(struct btrfs_bio *bbio, int num_stripes)
5195 {
5196         struct btrfs_bio_stripe s;
5197         int i;
5198         u64 l;
5199         int again = 1;
5200
5201         while (again) {
5202                 again = 0;
5203                 for (i = 0; i < num_stripes - 1; i++) {
5204                         if (parity_smaller(bbio->raid_map[i],
5205                                            bbio->raid_map[i+1])) {
5206                                 s = bbio->stripes[i];
5207                                 l = bbio->raid_map[i];
5208                                 bbio->stripes[i] = bbio->stripes[i+1];
5209                                 bbio->raid_map[i] = bbio->raid_map[i+1];
5210                                 bbio->stripes[i+1] = s;
5211                                 bbio->raid_map[i+1] = l;
5212
5213                                 again = 1;
5214                         }
5215                 }
5216         }
5217 }
5218
5219 static struct btrfs_bio *alloc_btrfs_bio(int total_stripes, int real_stripes)
5220 {
5221         struct btrfs_bio *bbio = kzalloc(
5222                  /* the size of the btrfs_bio */
5223                 sizeof(struct btrfs_bio) +
5224                 /* plus the variable array for the stripes */
5225                 sizeof(struct btrfs_bio_stripe) * (total_stripes) +
5226                 /* plus the variable array for the tgt dev */
5227                 sizeof(int) * (real_stripes) +
5228                 /*
5229                  * plus the raid_map, which includes both the tgt dev
5230                  * and the stripes
5231                  */
5232                 sizeof(u64) * (total_stripes),
5233                 GFP_NOFS|__GFP_NOFAIL);
5234
5235         atomic_set(&bbio->error, 0);
5236         refcount_set(&bbio->refs, 1);
5237
5238         return bbio;
5239 }
5240
5241 void btrfs_get_bbio(struct btrfs_bio *bbio)
5242 {
5243         WARN_ON(!refcount_read(&bbio->refs));
5244         refcount_inc(&bbio->refs);
5245 }
5246
5247 void btrfs_put_bbio(struct btrfs_bio *bbio)
5248 {
5249         if (!bbio)
5250                 return;
5251         if (refcount_dec_and_test(&bbio->refs))
5252                 kfree(bbio);
5253 }
5254
5255 /* can REQ_OP_DISCARD be sent with other REQ like REQ_OP_WRITE? */
5256 /*
5257  * Please note that, discard won't be sent to target device of device
5258  * replace.
5259  */
5260 static int __btrfs_map_block_for_discard(struct btrfs_fs_info *fs_info,
5261                                          u64 logical, u64 length,
5262                                          struct btrfs_bio **bbio_ret)
5263 {
5264         struct extent_map *em;
5265         struct map_lookup *map;
5266         struct btrfs_bio *bbio;
5267         u64 offset;
5268         u64 stripe_nr;
5269         u64 stripe_nr_end;
5270         u64 stripe_end_offset;
5271         u64 stripe_cnt;
5272         u64 stripe_len;
5273         u64 stripe_offset;
5274         u64 num_stripes;
5275         u32 stripe_index;
5276         u32 factor = 0;
5277         u32 sub_stripes = 0;
5278         u64 stripes_per_dev = 0;
5279         u32 remaining_stripes = 0;
5280         u32 last_stripe = 0;
5281         int ret = 0;
5282         int i;
5283
5284         /* discard always return a bbio */
5285         ASSERT(bbio_ret);
5286
5287         em = get_chunk_map(fs_info, logical, length);
5288         if (IS_ERR(em))
5289                 return PTR_ERR(em);
5290
5291         map = em->map_lookup;
5292         /* we don't discard raid56 yet */
5293         if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5294                 ret = -EOPNOTSUPP;
5295                 goto out;
5296         }
5297
5298         offset = logical - em->start;
5299         length = min_t(u64, em->len - offset, length);
5300
5301         stripe_len = map->stripe_len;
5302         /*
5303          * stripe_nr counts the total number of stripes we have to stride
5304          * to get to this block
5305          */
5306         stripe_nr = div64_u64(offset, stripe_len);
5307
5308         /* stripe_offset is the offset of this block in its stripe */
5309         stripe_offset = offset - stripe_nr * stripe_len;
5310
5311         stripe_nr_end = round_up(offset + length, map->stripe_len);
5312         stripe_nr_end = div64_u64(stripe_nr_end, map->stripe_len);
5313         stripe_cnt = stripe_nr_end - stripe_nr;
5314         stripe_end_offset = stripe_nr_end * map->stripe_len -
5315                             (offset + length);
5316         /*
5317          * after this, stripe_nr is the number of stripes on this
5318          * device we have to walk to find the data, and stripe_index is
5319          * the number of our device in the stripe array
5320          */
5321         num_stripes = 1;
5322         stripe_index = 0;
5323         if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5324                          BTRFS_BLOCK_GROUP_RAID10)) {
5325                 if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5326                         sub_stripes = 1;
5327                 else
5328                         sub_stripes = map->sub_stripes;
5329
5330                 factor = map->num_stripes / sub_stripes;
5331                 num_stripes = min_t(u64, map->num_stripes,
5332                                     sub_stripes * stripe_cnt);
5333                 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
5334                 stripe_index *= sub_stripes;
5335                 stripes_per_dev = div_u64_rem(stripe_cnt, factor,
5336                                               &remaining_stripes);
5337                 div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
5338                 last_stripe *= sub_stripes;
5339         } else if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
5340                                 BTRFS_BLOCK_GROUP_DUP)) {
5341                 num_stripes = map->num_stripes;
5342         } else {
5343                 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5344                                         &stripe_index);
5345         }
5346
5347         bbio = alloc_btrfs_bio(num_stripes, 0);
5348         if (!bbio) {
5349                 ret = -ENOMEM;
5350                 goto out;
5351         }
5352
5353         for (i = 0; i < num_stripes; i++) {
5354                 bbio->stripes[i].physical =
5355                         map->stripes[stripe_index].physical +
5356                         stripe_offset + stripe_nr * map->stripe_len;
5357                 bbio->stripes[i].dev = map->stripes[stripe_index].dev;
5358
5359                 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5360                                  BTRFS_BLOCK_GROUP_RAID10)) {
5361                         bbio->stripes[i].length = stripes_per_dev *
5362                                 map->stripe_len;
5363
5364                         if (i / sub_stripes < remaining_stripes)
5365                                 bbio->stripes[i].length +=
5366                                         map->stripe_len;
5367
5368                         /*
5369                          * Special for the first stripe and
5370                          * the last stripe:
5371                          *
5372                          * |-------|...|-------|
5373                          *     |----------|
5374                          *    off     end_off
5375                          */
5376                         if (i < sub_stripes)
5377                                 bbio->stripes[i].length -=
5378                                         stripe_offset;
5379
5380                         if (stripe_index >= last_stripe &&
5381                             stripe_index <= (last_stripe +
5382                                              sub_stripes - 1))
5383                                 bbio->stripes[i].length -=
5384                                         stripe_end_offset;
5385
5386                         if (i == sub_stripes - 1)
5387                                 stripe_offset = 0;
5388                 } else {
5389                         bbio->stripes[i].length = length;
5390                 }
5391
5392                 stripe_index++;
5393                 if (stripe_index == map->num_stripes) {
5394                         stripe_index = 0;
5395                         stripe_nr++;
5396                 }
5397         }
5398
5399         *bbio_ret = bbio;
5400         bbio->map_type = map->type;
5401         bbio->num_stripes = num_stripes;
5402 out:
5403         free_extent_map(em);
5404         return ret;
5405 }
5406
5407 /*
5408  * In dev-replace case, for repair case (that's the only case where the mirror
5409  * is selected explicitly when calling btrfs_map_block), blocks left of the
5410  * left cursor can also be read from the target drive.
5411  *
5412  * For REQ_GET_READ_MIRRORS, the target drive is added as the last one to the
5413  * array of stripes.
5414  * For READ, it also needs to be supported using the same mirror number.
5415  *
5416  * If the requested block is not left of the left cursor, EIO is returned. This
5417  * can happen because btrfs_num_copies() returns one more in the dev-replace
5418  * case.
5419  */
5420 static int get_extra_mirror_from_replace(struct btrfs_fs_info *fs_info,
5421                                          u64 logical, u64 length,
5422                                          u64 srcdev_devid, int *mirror_num,
5423                                          u64 *physical)
5424 {
5425         struct btrfs_bio *bbio = NULL;
5426         int num_stripes;
5427         int index_srcdev = 0;
5428         int found = 0;
5429         u64 physical_of_found = 0;
5430         int i;
5431         int ret = 0;
5432
5433         ret = __btrfs_map_block(fs_info, BTRFS_MAP_GET_READ_MIRRORS,
5434                                 logical, &length, &bbio, 0, 0);
5435         if (ret) {
5436                 ASSERT(bbio == NULL);
5437                 return ret;
5438         }
5439
5440         num_stripes = bbio->num_stripes;
5441         if (*mirror_num > num_stripes) {
5442                 /*
5443                  * BTRFS_MAP_GET_READ_MIRRORS does not contain this mirror,
5444                  * that means that the requested area is not left of the left
5445                  * cursor
5446                  */
5447                 btrfs_put_bbio(bbio);
5448                 return -EIO;
5449         }
5450
5451         /*
5452          * process the rest of the function using the mirror_num of the source
5453          * drive. Therefore look it up first.  At the end, patch the device
5454          * pointer to the one of the target drive.
5455          */
5456         for (i = 0; i < num_stripes; i++) {
5457                 if (bbio->stripes[i].dev->devid != srcdev_devid)
5458                         continue;
5459
5460                 /*
5461                  * In case of DUP, in order to keep it simple, only add the
5462                  * mirror with the lowest physical address
5463                  */
5464                 if (found &&
5465                     physical_of_found <= bbio->stripes[i].physical)
5466                         continue;
5467
5468                 index_srcdev = i;
5469                 found = 1;
5470                 physical_of_found = bbio->stripes[i].physical;
5471         }
5472
5473         btrfs_put_bbio(bbio);
5474
5475         ASSERT(found);
5476         if (!found)
5477                 return -EIO;
5478
5479         *mirror_num = index_srcdev + 1;
5480         *physical = physical_of_found;
5481         return ret;
5482 }
5483
5484 static void handle_ops_on_dev_replace(enum btrfs_map_op op,
5485                                       struct btrfs_bio **bbio_ret,
5486                                       struct btrfs_dev_replace *dev_replace,
5487                                       int *num_stripes_ret, int *max_errors_ret)
5488 {
5489         struct btrfs_bio *bbio = *bbio_ret;
5490         u64 srcdev_devid = dev_replace->srcdev->devid;
5491         int tgtdev_indexes = 0;
5492         int num_stripes = *num_stripes_ret;
5493         int max_errors = *max_errors_ret;
5494         int i;
5495
5496         if (op == BTRFS_MAP_WRITE) {
5497                 int index_where_to_add;
5498
5499                 /*
5500                  * duplicate the write operations while the dev replace
5501                  * procedure is running. Since the copying of the old disk to
5502                  * the new disk takes place at run time while the filesystem is
5503                  * mounted writable, the regular write operations to the old
5504                  * disk have to be duplicated to go to the new disk as well.
5505                  *
5506                  * Note that device->missing is handled by the caller, and that
5507                  * the write to the old disk is already set up in the stripes
5508                  * array.
5509                  */
5510                 index_where_to_add = num_stripes;
5511                 for (i = 0; i < num_stripes; i++) {
5512                         if (bbio->stripes[i].dev->devid == srcdev_devid) {
5513                                 /* write to new disk, too */
5514                                 struct btrfs_bio_stripe *new =
5515                                         bbio->stripes + index_where_to_add;
5516                                 struct btrfs_bio_stripe *old =
5517                                         bbio->stripes + i;
5518
5519                                 new->physical = old->physical;
5520                                 new->length = old->length;
5521                                 new->dev = dev_replace->tgtdev;
5522                                 bbio->tgtdev_map[i] = index_where_to_add;
5523                                 index_where_to_add++;
5524                                 max_errors++;
5525                                 tgtdev_indexes++;
5526                         }
5527                 }
5528                 num_stripes = index_where_to_add;
5529         } else if (op == BTRFS_MAP_GET_READ_MIRRORS) {
5530                 int index_srcdev = 0;
5531                 int found = 0;
5532                 u64 physical_of_found = 0;
5533
5534                 /*
5535                  * During the dev-replace procedure, the target drive can also
5536                  * be used to read data in case it is needed to repair a corrupt
5537                  * block elsewhere. This is possible if the requested area is
5538                  * left of the left cursor. In this area, the target drive is a
5539                  * full copy of the source drive.
5540                  */
5541                 for (i = 0; i < num_stripes; i++) {
5542                         if (bbio->stripes[i].dev->devid == srcdev_devid) {
5543                                 /*
5544                                  * In case of DUP, in order to keep it simple,
5545                                  * only add the mirror with the lowest physical
5546                                  * address
5547                                  */
5548                                 if (found &&
5549                                     physical_of_found <=
5550                                      bbio->stripes[i].physical)
5551                                         continue;
5552                                 index_srcdev = i;
5553                                 found = 1;
5554                                 physical_of_found = bbio->stripes[i].physical;
5555                         }
5556                 }
5557                 if (found) {
5558                         struct btrfs_bio_stripe *tgtdev_stripe =
5559                                 bbio->stripes + num_stripes;
5560
5561                         tgtdev_stripe->physical = physical_of_found;
5562                         tgtdev_stripe->length =
5563                                 bbio->stripes[index_srcdev].length;
5564                         tgtdev_stripe->dev = dev_replace->tgtdev;
5565                         bbio->tgtdev_map[index_srcdev] = num_stripes;
5566
5567                         tgtdev_indexes++;
5568                         num_stripes++;
5569                 }
5570         }
5571
5572         *num_stripes_ret = num_stripes;
5573         *max_errors_ret = max_errors;
5574         bbio->num_tgtdevs = tgtdev_indexes;
5575         *bbio_ret = bbio;
5576 }
5577
5578 static bool need_full_stripe(enum btrfs_map_op op)
5579 {
5580         return (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS);
5581 }
5582
5583 static int __btrfs_map_block(struct btrfs_fs_info *fs_info,
5584                              enum btrfs_map_op op,
5585                              u64 logical, u64 *length,
5586                              struct btrfs_bio **bbio_ret,
5587                              int mirror_num, int need_raid_map)
5588 {
5589         struct extent_map *em;
5590         struct map_lookup *map;
5591         u64 offset;
5592         u64 stripe_offset;
5593         u64 stripe_nr;
5594         u64 stripe_len;
5595         u32 stripe_index;
5596         int i;
5597         int ret = 0;
5598         int num_stripes;
5599         int max_errors = 0;
5600         int tgtdev_indexes = 0;
5601         struct btrfs_bio *bbio = NULL;
5602         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
5603         int dev_replace_is_ongoing = 0;
5604         int num_alloc_stripes;
5605         int patch_the_first_stripe_for_dev_replace = 0;
5606         u64 physical_to_patch_in_first_stripe = 0;
5607         u64 raid56_full_stripe_start = (u64)-1;
5608
5609         if (op == BTRFS_MAP_DISCARD)
5610                 return __btrfs_map_block_for_discard(fs_info, logical,
5611                                                      *length, bbio_ret);
5612
5613         em = get_chunk_map(fs_info, logical, *length);
5614         if (IS_ERR(em))
5615                 return PTR_ERR(em);
5616
5617         map = em->map_lookup;
5618         offset = logical - em->start;
5619
5620         stripe_len = map->stripe_len;
5621         stripe_nr = offset;
5622         /*
5623          * stripe_nr counts the total number of stripes we have to stride
5624          * to get to this block
5625          */
5626         stripe_nr = div64_u64(stripe_nr, stripe_len);
5627
5628         stripe_offset = stripe_nr * stripe_len;
5629         if (offset < stripe_offset) {
5630                 btrfs_crit(fs_info,
5631                            "stripe math has gone wrong, stripe_offset=%llu, offset=%llu, start=%llu, logical=%llu, stripe_len=%llu",
5632                            stripe_offset, offset, em->start, logical,
5633                            stripe_len);
5634                 free_extent_map(em);
5635                 return -EINVAL;
5636         }
5637
5638         /* stripe_offset is the offset of this block in its stripe*/
5639         stripe_offset = offset - stripe_offset;
5640
5641         /* if we're here for raid56, we need to know the stripe aligned start */
5642         if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5643                 unsigned long full_stripe_len = stripe_len * nr_data_stripes(map);
5644                 raid56_full_stripe_start = offset;
5645
5646                 /* allow a write of a full stripe, but make sure we don't
5647                  * allow straddling of stripes
5648                  */
5649                 raid56_full_stripe_start = div64_u64(raid56_full_stripe_start,
5650                                 full_stripe_len);
5651                 raid56_full_stripe_start *= full_stripe_len;
5652         }
5653
5654         if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
5655                 u64 max_len;
5656                 /* For writes to RAID[56], allow a full stripeset across all disks.
5657                    For other RAID types and for RAID[56] reads, just allow a single
5658                    stripe (on a single disk). */
5659                 if ((map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
5660                     (op == BTRFS_MAP_WRITE)) {
5661                         max_len = stripe_len * nr_data_stripes(map) -
5662                                 (offset - raid56_full_stripe_start);
5663                 } else {
5664                         /* we limit the length of each bio to what fits in a stripe */
5665                         max_len = stripe_len - stripe_offset;
5666                 }
5667                 *length = min_t(u64, em->len - offset, max_len);
5668         } else {
5669                 *length = em->len - offset;
5670         }
5671
5672         /* This is for when we're called from btrfs_merge_bio_hook() and all
5673            it cares about is the length */
5674         if (!bbio_ret)
5675                 goto out;
5676
5677         btrfs_dev_replace_lock(dev_replace, 0);
5678         dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
5679         if (!dev_replace_is_ongoing)
5680                 btrfs_dev_replace_unlock(dev_replace, 0);
5681         else
5682                 btrfs_dev_replace_set_lock_blocking(dev_replace);
5683
5684         if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
5685             !need_full_stripe(op) && dev_replace->tgtdev != NULL) {
5686                 ret = get_extra_mirror_from_replace(fs_info, logical, *length,
5687                                                     dev_replace->srcdev->devid,
5688                                                     &mirror_num,
5689                                             &physical_to_patch_in_first_stripe);
5690                 if (ret)
5691                         goto out;
5692                 else
5693                         patch_the_first_stripe_for_dev_replace = 1;
5694         } else if (mirror_num > map->num_stripes) {
5695                 mirror_num = 0;
5696         }
5697
5698         num_stripes = 1;
5699         stripe_index = 0;
5700         if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5701                 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5702                                 &stripe_index);
5703                 if (op != BTRFS_MAP_WRITE && op != BTRFS_MAP_GET_READ_MIRRORS)
5704                         mirror_num = 1;
5705         } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
5706                 if (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS)
5707                         num_stripes = map->num_stripes;
5708                 else if (mirror_num)
5709                         stripe_index = mirror_num - 1;
5710                 else {
5711                         stripe_index = find_live_mirror(fs_info, map, 0,
5712                                             map->num_stripes,
5713                                             current->pid % map->num_stripes,
5714                                             dev_replace_is_ongoing);
5715                         mirror_num = stripe_index + 1;
5716                 }
5717
5718         } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
5719                 if (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS) {
5720                         num_stripes = map->num_stripes;
5721                 } else if (mirror_num) {
5722                         stripe_index = mirror_num - 1;
5723                 } else {
5724                         mirror_num = 1;
5725                 }
5726
5727         } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5728                 u32 factor = map->num_stripes / map->sub_stripes;
5729
5730                 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
5731                 stripe_index *= map->sub_stripes;
5732
5733                 if (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS)
5734                         num_stripes = map->sub_stripes;
5735                 else if (mirror_num)
5736                         stripe_index += mirror_num - 1;
5737                 else {
5738                         int old_stripe_index = stripe_index;
5739                         stripe_index = find_live_mirror(fs_info, map,
5740                                               stripe_index,
5741                                               map->sub_stripes, stripe_index +
5742                                               current->pid % map->sub_stripes,
5743                                               dev_replace_is_ongoing);
5744                         mirror_num = stripe_index - old_stripe_index + 1;
5745                 }
5746
5747         } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5748                 if (need_raid_map &&
5749                     (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS ||
5750                      mirror_num > 1)) {
5751                         /* push stripe_nr back to the start of the full stripe */
5752                         stripe_nr = div64_u64(raid56_full_stripe_start,
5753                                         stripe_len * nr_data_stripes(map));
5754
5755                         /* RAID[56] write or recovery. Return all stripes */
5756                         num_stripes = map->num_stripes;
5757                         max_errors = nr_parity_stripes(map);
5758
5759                         *length = map->stripe_len;
5760                         stripe_index = 0;
5761                         stripe_offset = 0;
5762                 } else {
5763                         /*
5764                          * Mirror #0 or #1 means the original data block.
5765                          * Mirror #2 is RAID5 parity block.
5766                          * Mirror #3 is RAID6 Q block.
5767                          */
5768                         stripe_nr = div_u64_rem(stripe_nr,
5769                                         nr_data_stripes(map), &stripe_index);
5770                         if (mirror_num > 1)
5771                                 stripe_index = nr_data_stripes(map) +
5772                                                 mirror_num - 2;
5773
5774                         /* We distribute the parity blocks across stripes */
5775                         div_u64_rem(stripe_nr + stripe_index, map->num_stripes,
5776                                         &stripe_index);
5777                         if ((op != BTRFS_MAP_WRITE &&
5778                              op != BTRFS_MAP_GET_READ_MIRRORS) &&
5779                             mirror_num <= 1)
5780                                 mirror_num = 1;
5781                 }
5782         } else {
5783                 /*
5784                  * after this, stripe_nr is the number of stripes on this
5785                  * device we have to walk to find the data, and stripe_index is
5786                  * the number of our device in the stripe array
5787                  */
5788                 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5789                                 &stripe_index);
5790                 mirror_num = stripe_index + 1;
5791         }
5792         if (stripe_index >= map->num_stripes) {
5793                 btrfs_crit(fs_info,
5794                            "stripe index math went horribly wrong, got stripe_index=%u, num_stripes=%u",
5795                            stripe_index, map->num_stripes);
5796                 ret = -EINVAL;
5797                 goto out;
5798         }
5799
5800         num_alloc_stripes = num_stripes;
5801         if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL) {
5802                 if (op == BTRFS_MAP_WRITE)
5803                         num_alloc_stripes <<= 1;
5804                 if (op == BTRFS_MAP_GET_READ_MIRRORS)
5805                         num_alloc_stripes++;
5806                 tgtdev_indexes = num_stripes;
5807         }
5808
5809         bbio = alloc_btrfs_bio(num_alloc_stripes, tgtdev_indexes);
5810         if (!bbio) {
5811                 ret = -ENOMEM;
5812                 goto out;
5813         }
5814         if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL)
5815                 bbio->tgtdev_map = (int *)(bbio->stripes + num_alloc_stripes);
5816
5817         /* build raid_map */
5818         if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK && need_raid_map &&
5819             (need_full_stripe(op) || mirror_num > 1)) {
5820                 u64 tmp;
5821                 unsigned rot;
5822
5823                 bbio->raid_map = (u64 *)((void *)bbio->stripes +
5824                                  sizeof(struct btrfs_bio_stripe) *
5825                                  num_alloc_stripes +
5826                                  sizeof(int) * tgtdev_indexes);
5827
5828                 /* Work out the disk rotation on this stripe-set */
5829                 div_u64_rem(stripe_nr, num_stripes, &rot);
5830
5831                 /* Fill in the logical address of each stripe */
5832                 tmp = stripe_nr * nr_data_stripes(map);
5833                 for (i = 0; i < nr_data_stripes(map); i++)
5834                         bbio->raid_map[(i+rot) % num_stripes] =
5835                                 em->start + (tmp + i) * map->stripe_len;
5836
5837                 bbio->raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE;
5838                 if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5839                         bbio->raid_map[(i+rot+1) % num_stripes] =
5840                                 RAID6_Q_STRIPE;
5841         }
5842
5843
5844         for (i = 0; i < num_stripes; i++) {
5845                 bbio->stripes[i].physical =
5846                         map->stripes[stripe_index].physical +
5847                         stripe_offset +
5848                         stripe_nr * map->stripe_len;
5849                 bbio->stripes[i].dev =
5850                         map->stripes[stripe_index].dev;
5851                 stripe_index++;
5852         }
5853
5854         if (need_full_stripe(op))
5855                 max_errors = btrfs_chunk_max_errors(map);
5856
5857         if (bbio->raid_map)
5858                 sort_parity_stripes(bbio, num_stripes);
5859
5860         if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL &&
5861             need_full_stripe(op)) {
5862                 handle_ops_on_dev_replace(op, &bbio, dev_replace, &num_stripes,
5863                                           &max_errors);
5864         }
5865
5866         *bbio_ret = bbio;
5867         bbio->map_type = map->type;
5868         bbio->num_stripes = num_stripes;
5869         bbio->max_errors = max_errors;
5870         bbio->mirror_num = mirror_num;
5871
5872         /*
5873          * this is the case that REQ_READ && dev_replace_is_ongoing &&
5874          * mirror_num == num_stripes + 1 && dev_replace target drive is
5875          * available as a mirror
5876          */
5877         if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) {
5878                 WARN_ON(num_stripes > 1);
5879                 bbio->stripes[0].dev = dev_replace->tgtdev;
5880                 bbio->stripes[0].physical = physical_to_patch_in_first_stripe;
5881                 bbio->mirror_num = map->num_stripes + 1;
5882         }
5883 out:
5884         if (dev_replace_is_ongoing) {
5885                 btrfs_dev_replace_clear_lock_blocking(dev_replace);
5886                 btrfs_dev_replace_unlock(dev_replace, 0);
5887         }
5888         free_extent_map(em);
5889         return ret;
5890 }
5891
5892 int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
5893                       u64 logical, u64 *length,
5894                       struct btrfs_bio **bbio_ret, int mirror_num)
5895 {
5896         return __btrfs_map_block(fs_info, op, logical, length, bbio_ret,
5897                                  mirror_num, 0);
5898 }
5899
5900 /* For Scrub/replace */
5901 int btrfs_map_sblock(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
5902                      u64 logical, u64 *length,
5903                      struct btrfs_bio **bbio_ret)
5904 {
5905         return __btrfs_map_block(fs_info, op, logical, length, bbio_ret, 0, 1);
5906 }
5907
5908 int btrfs_rmap_block(struct btrfs_fs_info *fs_info,
5909                      u64 chunk_start, u64 physical, u64 devid,
5910                      u64 **logical, int *naddrs, int *stripe_len)
5911 {
5912         struct extent_map *em;
5913         struct map_lookup *map;
5914         u64 *buf;
5915         u64 bytenr;
5916         u64 length;
5917         u64 stripe_nr;
5918         u64 rmap_len;
5919         int i, j, nr = 0;
5920
5921         em = get_chunk_map(fs_info, chunk_start, 1);
5922         if (IS_ERR(em))
5923                 return -EIO;
5924
5925         map = em->map_lookup;
5926         length = em->len;
5927         rmap_len = map->stripe_len;
5928
5929         if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5930                 length = div_u64(length, map->num_stripes / map->sub_stripes);
5931         else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5932                 length = div_u64(length, map->num_stripes);
5933         else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5934                 length = div_u64(length, nr_data_stripes(map));
5935                 rmap_len = map->stripe_len * nr_data_stripes(map);
5936         }
5937
5938         buf = kcalloc(map->num_stripes, sizeof(u64), GFP_NOFS);
5939         BUG_ON(!buf); /* -ENOMEM */
5940
5941         for (i = 0; i < map->num_stripes; i++) {
5942                 if (devid && map->stripes[i].dev->devid != devid)
5943                         continue;
5944                 if (map->stripes[i].physical > physical ||
5945                     map->stripes[i].physical + length <= physical)
5946                         continue;
5947
5948                 stripe_nr = physical - map->stripes[i].physical;
5949                 stripe_nr = div64_u64(stripe_nr, map->stripe_len);
5950
5951                 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5952                         stripe_nr = stripe_nr * map->num_stripes + i;
5953                         stripe_nr = div_u64(stripe_nr, map->sub_stripes);
5954                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5955                         stripe_nr = stripe_nr * map->num_stripes + i;
5956                 } /* else if RAID[56], multiply by nr_data_stripes().
5957                    * Alternatively, just use rmap_len below instead of
5958                    * map->stripe_len */
5959
5960                 bytenr = chunk_start + stripe_nr * rmap_len;
5961                 WARN_ON(nr >= map->num_stripes);
5962                 for (j = 0; j < nr; j++) {
5963                         if (buf[j] == bytenr)
5964                                 break;
5965                 }
5966                 if (j == nr) {
5967                         WARN_ON(nr >= map->num_stripes);
5968                         buf[nr++] = bytenr;
5969                 }
5970         }
5971
5972         *logical = buf;
5973         *naddrs = nr;
5974         *stripe_len = rmap_len;
5975
5976         free_extent_map(em);
5977         return 0;
5978 }
5979
5980 static inline void btrfs_end_bbio(struct btrfs_bio *bbio, struct bio *bio)
5981 {
5982         bio->bi_private = bbio->private;
5983         bio->bi_end_io = bbio->end_io;
5984         bio_endio(bio);
5985
5986         btrfs_put_bbio(bbio);
5987 }
5988
5989 static void btrfs_end_bio(struct bio *bio)
5990 {
5991         struct btrfs_bio *bbio = bio->bi_private;
5992         int is_orig_bio = 0;
5993
5994         if (bio->bi_status) {
5995                 atomic_inc(&bbio->error);
5996                 if (bio->bi_status == BLK_STS_IOERR ||
5997                     bio->bi_status == BLK_STS_TARGET) {
5998                         unsigned int stripe_index =
5999                                 btrfs_io_bio(bio)->stripe_index;
6000                         struct btrfs_device *dev;
6001
6002                         BUG_ON(stripe_index >= bbio->num_stripes);
6003                         dev = bbio->stripes[stripe_index].dev;
6004                         if (dev->bdev) {
6005                                 if (bio_op(bio) == REQ_OP_WRITE)
6006                                         btrfs_dev_stat_inc(dev,
6007                                                 BTRFS_DEV_STAT_WRITE_ERRS);
6008                                 else
6009                                         btrfs_dev_stat_inc(dev,
6010                                                 BTRFS_DEV_STAT_READ_ERRS);
6011                                 if (bio->bi_opf & REQ_PREFLUSH)
6012                                         btrfs_dev_stat_inc(dev,
6013                                                 BTRFS_DEV_STAT_FLUSH_ERRS);
6014                                 btrfs_dev_stat_print_on_error(dev);
6015                         }
6016                 }
6017         }
6018
6019         if (bio == bbio->orig_bio)
6020                 is_orig_bio = 1;
6021
6022         btrfs_bio_counter_dec(bbio->fs_info);
6023
6024         if (atomic_dec_and_test(&bbio->stripes_pending)) {
6025                 if (!is_orig_bio) {
6026                         bio_put(bio);
6027                         bio = bbio->orig_bio;
6028                 }
6029
6030                 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
6031                 /* only send an error to the higher layers if it is
6032                  * beyond the tolerance of the btrfs bio
6033                  */
6034                 if (atomic_read(&bbio->error) > bbio->max_errors) {
6035                         bio->bi_status = BLK_STS_IOERR;
6036                 } else {
6037                         /*
6038                          * this bio is actually up to date, we didn't
6039                          * go over the max number of errors
6040                          */
6041                         bio->bi_status = 0;
6042                 }
6043
6044                 btrfs_end_bbio(bbio, bio);
6045         } else if (!is_orig_bio) {
6046                 bio_put(bio);
6047         }
6048 }
6049
6050 /*
6051  * see run_scheduled_bios for a description of why bios are collected for
6052  * async submit.
6053  *
6054  * This will add one bio to the pending list for a device and make sure
6055  * the work struct is scheduled.
6056  */
6057 static noinline void btrfs_schedule_bio(struct btrfs_device *device,
6058                                         struct bio *bio)
6059 {
6060         struct btrfs_fs_info *fs_info = device->fs_info;
6061         int should_queue = 1;
6062         struct btrfs_pending_bios *pending_bios;
6063
6064         if (device->missing || !device->bdev) {
6065                 bio_io_error(bio);
6066                 return;
6067         }
6068
6069         /* don't bother with additional async steps for reads, right now */
6070         if (bio_op(bio) == REQ_OP_READ) {
6071                 bio_get(bio);
6072                 btrfsic_submit_bio(bio);
6073                 bio_put(bio);
6074                 return;
6075         }
6076
6077         /*
6078          * nr_async_bios allows us to reliably return congestion to the
6079          * higher layers.  Otherwise, the async bio makes it appear we have
6080          * made progress against dirty pages when we've really just put it
6081          * on a queue for later
6082          */
6083         atomic_inc(&fs_info->nr_async_bios);
6084         WARN_ON(bio->bi_next);
6085         bio->bi_next = NULL;
6086
6087         spin_lock(&device->io_lock);
6088         if (op_is_sync(bio->bi_opf))
6089                 pending_bios = &device->pending_sync_bios;
6090         else
6091                 pending_bios = &device->pending_bios;
6092
6093         if (pending_bios->tail)
6094                 pending_bios->tail->bi_next = bio;
6095
6096         pending_bios->tail = bio;
6097         if (!pending_bios->head)
6098                 pending_bios->head = bio;
6099         if (device->running_pending)
6100                 should_queue = 0;
6101
6102         spin_unlock(&device->io_lock);
6103
6104         if (should_queue)
6105                 btrfs_queue_work(fs_info->submit_workers, &device->work);
6106 }
6107
6108 static void submit_stripe_bio(struct btrfs_bio *bbio, struct bio *bio,
6109                               u64 physical, int dev_nr, int async)
6110 {
6111         struct btrfs_device *dev = bbio->stripes[dev_nr].dev;
6112         struct btrfs_fs_info *fs_info = bbio->fs_info;
6113
6114         bio->bi_private = bbio;
6115         btrfs_io_bio(bio)->stripe_index = dev_nr;
6116         bio->bi_end_io = btrfs_end_bio;
6117         bio->bi_iter.bi_sector = physical >> 9;
6118 #ifdef DEBUG
6119         {
6120                 struct rcu_string *name;
6121
6122                 rcu_read_lock();
6123                 name = rcu_dereference(dev->name);
6124                 btrfs_debug(fs_info,
6125                         "btrfs_map_bio: rw %d 0x%x, sector=%llu, dev=%lu (%s id %llu), size=%u",
6126                         bio_op(bio), bio->bi_opf,
6127                         (u64)bio->bi_iter.bi_sector,
6128                         (u_long)dev->bdev->bd_dev, name->str, dev->devid,
6129                         bio->bi_iter.bi_size);
6130                 rcu_read_unlock();
6131         }
6132 #endif
6133         bio_set_dev(bio, dev->bdev);
6134
6135         btrfs_bio_counter_inc_noblocked(fs_info);
6136
6137         if (async)
6138                 btrfs_schedule_bio(dev, bio);
6139         else
6140                 btrfsic_submit_bio(bio);
6141 }
6142
6143 static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical)
6144 {
6145         atomic_inc(&bbio->error);
6146         if (atomic_dec_and_test(&bbio->stripes_pending)) {
6147                 /* Should be the original bio. */
6148                 WARN_ON(bio != bbio->orig_bio);
6149
6150                 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
6151                 bio->bi_iter.bi_sector = logical >> 9;
6152                 if (atomic_read(&bbio->error) > bbio->max_errors)
6153                         bio->bi_status = BLK_STS_IOERR;
6154                 else
6155                         bio->bi_status = BLK_STS_OK;
6156                 btrfs_end_bbio(bbio, bio);
6157         }
6158 }
6159
6160 blk_status_t btrfs_map_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
6161                            int mirror_num, int async_submit)
6162 {
6163         struct btrfs_device *dev;
6164         struct bio *first_bio = bio;
6165         u64 logical = (u64)bio->bi_iter.bi_sector << 9;
6166         u64 length = 0;
6167         u64 map_length;
6168         int ret;
6169         int dev_nr;
6170         int total_devs;
6171         struct btrfs_bio *bbio = NULL;
6172
6173         length = bio->bi_iter.bi_size;
6174         map_length = length;
6175
6176         btrfs_bio_counter_inc_blocked(fs_info);
6177         ret = __btrfs_map_block(fs_info, btrfs_op(bio), logical,
6178                                 &map_length, &bbio, mirror_num, 1);
6179         if (ret) {
6180                 btrfs_bio_counter_dec(fs_info);
6181                 return errno_to_blk_status(ret);
6182         }
6183
6184         total_devs = bbio->num_stripes;
6185         bbio->orig_bio = first_bio;
6186         bbio->private = first_bio->bi_private;
6187         bbio->end_io = first_bio->bi_end_io;
6188         bbio->fs_info = fs_info;
6189         atomic_set(&bbio->stripes_pending, bbio->num_stripes);
6190
6191         if ((bbio->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
6192             ((bio_op(bio) == REQ_OP_WRITE) || (mirror_num > 1))) {
6193                 /* In this case, map_length has been set to the length of
6194                    a single stripe; not the whole write */
6195                 if (bio_op(bio) == REQ_OP_WRITE) {
6196                         ret = raid56_parity_write(fs_info, bio, bbio,
6197                                                   map_length);
6198                 } else {
6199                         ret = raid56_parity_recover(fs_info, bio, bbio,
6200                                                     map_length, mirror_num, 1);
6201                 }
6202
6203                 btrfs_bio_counter_dec(fs_info);
6204                 return errno_to_blk_status(ret);
6205         }
6206
6207         if (map_length < length) {
6208                 btrfs_crit(fs_info,
6209                            "mapping failed logical %llu bio len %llu len %llu",
6210                            logical, length, map_length);
6211                 BUG();
6212         }
6213
6214         for (dev_nr = 0; dev_nr < total_devs; dev_nr++) {
6215                 dev = bbio->stripes[dev_nr].dev;
6216                 if (!dev || !dev->bdev ||
6217                     (bio_op(first_bio) == REQ_OP_WRITE && !dev->writeable)) {
6218                         bbio_error(bbio, first_bio, logical);
6219                         continue;
6220                 }
6221
6222                 if (dev_nr < total_devs - 1)
6223                         bio = btrfs_bio_clone(first_bio);
6224                 else
6225                         bio = first_bio;
6226
6227                 submit_stripe_bio(bbio, bio, bbio->stripes[dev_nr].physical,
6228                                   dev_nr, async_submit);
6229         }
6230         btrfs_bio_counter_dec(fs_info);
6231         return BLK_STS_OK;
6232 }
6233
6234 struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid,
6235                                        u8 *uuid, u8 *fsid)
6236 {
6237         struct btrfs_device *device;
6238         struct btrfs_fs_devices *cur_devices;
6239
6240         cur_devices = fs_info->fs_devices;
6241         while (cur_devices) {
6242                 if (!fsid ||
6243                     !memcmp(cur_devices->fsid, fsid, BTRFS_FSID_SIZE)) {
6244                         device = find_device(cur_devices, devid, uuid);
6245                         if (device)
6246                                 return device;
6247                 }
6248                 cur_devices = cur_devices->seed;
6249         }
6250         return NULL;
6251 }
6252
6253 static struct btrfs_device *add_missing_dev(struct btrfs_fs_devices *fs_devices,
6254                                             u64 devid, u8 *dev_uuid)
6255 {
6256         struct btrfs_device *device;
6257
6258         device = btrfs_alloc_device(NULL, &devid, dev_uuid);
6259         if (IS_ERR(device))
6260                 return NULL;
6261
6262         list_add(&device->dev_list, &fs_devices->devices);
6263         device->fs_devices = fs_devices;
6264         fs_devices->num_devices++;
6265
6266         device->missing = 1;
6267         fs_devices->missing_devices++;
6268
6269         return device;
6270 }
6271
6272 /**
6273  * btrfs_alloc_device - allocate struct btrfs_device
6274  * @fs_info:    used only for generating a new devid, can be NULL if
6275  *              devid is provided (i.e. @devid != NULL).
6276  * @devid:      a pointer to devid for this device.  If NULL a new devid
6277  *              is generated.
6278  * @uuid:       a pointer to UUID for this device.  If NULL a new UUID
6279  *              is generated.
6280  *
6281  * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
6282  * on error.  Returned struct is not linked onto any lists and can be
6283  * destroyed with kfree() right away.
6284  */
6285 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
6286                                         const u64 *devid,
6287                                         const u8 *uuid)
6288 {
6289         struct btrfs_device *dev;
6290         u64 tmp;
6291
6292         if (WARN_ON(!devid && !fs_info))
6293                 return ERR_PTR(-EINVAL);
6294
6295         dev = __alloc_device();
6296         if (IS_ERR(dev))
6297                 return dev;
6298
6299         if (devid)
6300                 tmp = *devid;
6301         else {
6302                 int ret;
6303
6304                 ret = find_next_devid(fs_info, &tmp);
6305                 if (ret) {
6306                         kfree(dev);
6307                         return ERR_PTR(ret);
6308                 }
6309         }
6310         dev->devid = tmp;
6311
6312         if (uuid)
6313                 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE);
6314         else
6315                 generate_random_uuid(dev->uuid);
6316
6317         btrfs_init_work(&dev->work, btrfs_submit_helper,
6318                         pending_bios_fn, NULL, NULL);
6319
6320         return dev;
6321 }
6322
6323 /* Return -EIO if any error, otherwise return 0. */
6324 static int btrfs_check_chunk_valid(struct btrfs_fs_info *fs_info,
6325                                    struct extent_buffer *leaf,
6326                                    struct btrfs_chunk *chunk, u64 logical)
6327 {
6328         u64 length;
6329         u64 stripe_len;
6330         u16 num_stripes;
6331         u16 sub_stripes;
6332         u64 type;
6333
6334         length = btrfs_chunk_length(leaf, chunk);
6335         stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6336         num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6337         sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
6338         type = btrfs_chunk_type(leaf, chunk);
6339
6340         if (!num_stripes) {
6341                 btrfs_err(fs_info, "invalid chunk num_stripes: %u",
6342                           num_stripes);
6343                 return -EIO;
6344         }
6345         if (!IS_ALIGNED(logical, fs_info->sectorsize)) {
6346                 btrfs_err(fs_info, "invalid chunk logical %llu", logical);
6347                 return -EIO;
6348         }
6349         if (btrfs_chunk_sector_size(leaf, chunk) != fs_info->sectorsize) {
6350                 btrfs_err(fs_info, "invalid chunk sectorsize %u",
6351                           btrfs_chunk_sector_size(leaf, chunk));
6352                 return -EIO;
6353         }
6354         if (!length || !IS_ALIGNED(length, fs_info->sectorsize)) {
6355                 btrfs_err(fs_info, "invalid chunk length %llu", length);
6356                 return -EIO;
6357         }
6358         if (!is_power_of_2(stripe_len) || stripe_len != BTRFS_STRIPE_LEN) {
6359                 btrfs_err(fs_info, "invalid chunk stripe length: %llu",
6360                           stripe_len);
6361                 return -EIO;
6362         }
6363         if (~(BTRFS_BLOCK_GROUP_TYPE_MASK | BTRFS_BLOCK_GROUP_PROFILE_MASK) &
6364             type) {
6365                 btrfs_err(fs_info, "unrecognized chunk type: %llu",
6366                           ~(BTRFS_BLOCK_GROUP_TYPE_MASK |
6367                             BTRFS_BLOCK_GROUP_PROFILE_MASK) &
6368                           btrfs_chunk_type(leaf, chunk));
6369                 return -EIO;
6370         }
6371         if ((type & BTRFS_BLOCK_GROUP_RAID10 && sub_stripes != 2) ||
6372             (type & BTRFS_BLOCK_GROUP_RAID1 && num_stripes < 1) ||
6373             (type & BTRFS_BLOCK_GROUP_RAID5 && num_stripes < 2) ||
6374             (type & BTRFS_BLOCK_GROUP_RAID6 && num_stripes < 3) ||
6375             (type & BTRFS_BLOCK_GROUP_DUP && num_stripes > 2) ||
6376             ((type & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 &&
6377              num_stripes != 1)) {
6378                 btrfs_err(fs_info,
6379                         "invalid num_stripes:sub_stripes %u:%u for profile %llu",
6380                         num_stripes, sub_stripes,
6381                         type & BTRFS_BLOCK_GROUP_PROFILE_MASK);
6382                 return -EIO;
6383         }
6384
6385         return 0;
6386 }
6387
6388 static int read_one_chunk(struct btrfs_fs_info *fs_info, struct btrfs_key *key,
6389                           struct extent_buffer *leaf,
6390                           struct btrfs_chunk *chunk)
6391 {
6392         struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
6393         struct map_lookup *map;
6394         struct extent_map *em;
6395         u64 logical;
6396         u64 length;
6397         u64 devid;
6398         u8 uuid[BTRFS_UUID_SIZE];
6399         int num_stripes;
6400         int ret;
6401         int i;
6402
6403         logical = key->offset;
6404         length = btrfs_chunk_length(leaf, chunk);
6405         num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6406
6407         ret = btrfs_check_chunk_valid(fs_info, leaf, chunk, logical);
6408         if (ret)
6409                 return ret;
6410
6411         read_lock(&map_tree->map_tree.lock);
6412         em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
6413         read_unlock(&map_tree->map_tree.lock);
6414
6415         /* already mapped? */
6416         if (em && em->start <= logical && em->start + em->len > logical) {
6417                 free_extent_map(em);
6418                 return 0;
6419         } else if (em) {
6420                 free_extent_map(em);
6421         }
6422
6423         em = alloc_extent_map();
6424         if (!em)
6425                 return -ENOMEM;
6426         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
6427         if (!map) {
6428                 free_extent_map(em);
6429                 return -ENOMEM;
6430         }
6431
6432         set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
6433         em->map_lookup = map;
6434         em->start = logical;
6435         em->len = length;
6436         em->orig_start = 0;
6437         em->block_start = 0;
6438         em->block_len = em->len;
6439
6440         map->num_stripes = num_stripes;
6441         map->io_width = btrfs_chunk_io_width(leaf, chunk);
6442         map->io_align = btrfs_chunk_io_align(leaf, chunk);
6443         map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6444         map->type = btrfs_chunk_type(leaf, chunk);
6445         map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
6446         for (i = 0; i < num_stripes; i++) {
6447                 map->stripes[i].physical =
6448                         btrfs_stripe_offset_nr(leaf, chunk, i);
6449                 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
6450                 read_extent_buffer(leaf, uuid, (unsigned long)
6451                                    btrfs_stripe_dev_uuid_nr(chunk, i),
6452                                    BTRFS_UUID_SIZE);
6453                 map->stripes[i].dev = btrfs_find_device(fs_info, devid,
6454                                                         uuid, NULL);
6455                 if (!map->stripes[i].dev &&
6456                     !btrfs_test_opt(fs_info, DEGRADED)) {
6457                         free_extent_map(em);
6458                         btrfs_report_missing_device(fs_info, devid, uuid);
6459                         return -EIO;
6460                 }
6461                 if (!map->stripes[i].dev) {
6462                         map->stripes[i].dev =
6463                                 add_missing_dev(fs_info->fs_devices, devid,
6464                                                 uuid);
6465                         if (!map->stripes[i].dev) {
6466                                 free_extent_map(em);
6467                                 return -EIO;
6468                         }
6469                         btrfs_report_missing_device(fs_info, devid, uuid);
6470                 }
6471                 map->stripes[i].dev->in_fs_metadata = 1;
6472         }
6473
6474         write_lock(&map_tree->map_tree.lock);
6475         ret = add_extent_mapping(&map_tree->map_tree, em, 0);
6476         write_unlock(&map_tree->map_tree.lock);
6477         BUG_ON(ret); /* Tree corruption */
6478         free_extent_map(em);
6479
6480         return 0;
6481 }
6482
6483 static void fill_device_from_item(struct extent_buffer *leaf,
6484                                  struct btrfs_dev_item *dev_item,
6485                                  struct btrfs_device *device)
6486 {
6487         unsigned long ptr;
6488
6489         device->devid = btrfs_device_id(leaf, dev_item);
6490         device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
6491         device->total_bytes = device->disk_total_bytes;
6492         device->commit_total_bytes = device->disk_total_bytes;
6493         device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
6494         device->commit_bytes_used = device->bytes_used;
6495         device->type = btrfs_device_type(leaf, dev_item);
6496         device->io_align = btrfs_device_io_align(leaf, dev_item);
6497         device->io_width = btrfs_device_io_width(leaf, dev_item);
6498         device->sector_size = btrfs_device_sector_size(leaf, dev_item);
6499         WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
6500         device->is_tgtdev_for_dev_replace = 0;
6501
6502         ptr = btrfs_device_uuid(dev_item);
6503         read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
6504 }
6505
6506 static struct btrfs_fs_devices *open_seed_devices(struct btrfs_fs_info *fs_info,
6507                                                   u8 *fsid)
6508 {
6509         struct btrfs_fs_devices *fs_devices;
6510         int ret;
6511
6512         BUG_ON(!mutex_is_locked(&uuid_mutex));
6513         ASSERT(fsid);
6514
6515         fs_devices = fs_info->fs_devices->seed;
6516         while (fs_devices) {
6517                 if (!memcmp(fs_devices->fsid, fsid, BTRFS_FSID_SIZE))
6518                         return fs_devices;
6519
6520                 fs_devices = fs_devices->seed;
6521         }
6522
6523         fs_devices = find_fsid(fsid);
6524         if (!fs_devices) {
6525                 if (!btrfs_test_opt(fs_info, DEGRADED))
6526                         return ERR_PTR(-ENOENT);
6527
6528                 fs_devices = alloc_fs_devices(fsid);
6529                 if (IS_ERR(fs_devices))
6530                         return fs_devices;
6531
6532                 fs_devices->seeding = 1;
6533                 fs_devices->opened = 1;
6534                 return fs_devices;
6535         }
6536
6537         fs_devices = clone_fs_devices(fs_devices);
6538         if (IS_ERR(fs_devices))
6539                 return fs_devices;
6540
6541         ret = __btrfs_open_devices(fs_devices, FMODE_READ,
6542                                    fs_info->bdev_holder);
6543         if (ret) {
6544                 free_fs_devices(fs_devices);
6545                 fs_devices = ERR_PTR(ret);
6546                 goto out;
6547         }
6548
6549         if (!fs_devices->seeding) {
6550                 __btrfs_close_devices(fs_devices);
6551                 free_fs_devices(fs_devices);
6552                 fs_devices = ERR_PTR(-EINVAL);
6553                 goto out;
6554         }
6555
6556         fs_devices->seed = fs_info->fs_devices->seed;
6557         fs_info->fs_devices->seed = fs_devices;
6558 out:
6559         return fs_devices;
6560 }
6561
6562 static int read_one_dev(struct btrfs_fs_info *fs_info,
6563                         struct extent_buffer *leaf,
6564                         struct btrfs_dev_item *dev_item)
6565 {
6566         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6567         struct btrfs_device *device;
6568         u64 devid;
6569         int ret;
6570         u8 fs_uuid[BTRFS_FSID_SIZE];
6571         u8 dev_uuid[BTRFS_UUID_SIZE];
6572
6573         devid = btrfs_device_id(leaf, dev_item);
6574         read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
6575                            BTRFS_UUID_SIZE);
6576         read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
6577                            BTRFS_FSID_SIZE);
6578
6579         if (memcmp(fs_uuid, fs_info->fsid, BTRFS_FSID_SIZE)) {
6580                 fs_devices = open_seed_devices(fs_info, fs_uuid);
6581                 if (IS_ERR(fs_devices))
6582                         return PTR_ERR(fs_devices);
6583         }
6584
6585         device = btrfs_find_device(fs_info, devid, dev_uuid, fs_uuid);
6586         if (!device) {
6587                 if (!btrfs_test_opt(fs_info, DEGRADED)) {
6588                         btrfs_report_missing_device(fs_info, devid, dev_uuid);
6589                         return -EIO;
6590                 }
6591
6592                 device = add_missing_dev(fs_devices, devid, dev_uuid);
6593                 if (!device)
6594                         return -ENOMEM;
6595                 btrfs_report_missing_device(fs_info, devid, dev_uuid);
6596         } else {
6597                 if (!device->bdev) {
6598                         btrfs_report_missing_device(fs_info, devid, dev_uuid);
6599                         if (!btrfs_test_opt(fs_info, DEGRADED))
6600                                 return -EIO;
6601                 }
6602
6603                 if(!device->bdev && !device->missing) {
6604                         /*
6605                          * this happens when a device that was properly setup
6606                          * in the device info lists suddenly goes bad.
6607                          * device->bdev is NULL, and so we have to set
6608                          * device->missing to one here
6609                          */
6610                         device->fs_devices->missing_devices++;
6611                         device->missing = 1;
6612                 }
6613
6614                 /* Move the device to its own fs_devices */
6615                 if (device->fs_devices != fs_devices) {
6616                         ASSERT(device->missing);
6617
6618                         list_move(&device->dev_list, &fs_devices->devices);
6619                         device->fs_devices->num_devices--;
6620                         fs_devices->num_devices++;
6621
6622                         device->fs_devices->missing_devices--;
6623                         fs_devices->missing_devices++;
6624
6625                         device->fs_devices = fs_devices;
6626                 }
6627         }
6628
6629         if (device->fs_devices != fs_info->fs_devices) {
6630                 BUG_ON(device->writeable);
6631                 if (device->generation !=
6632                     btrfs_device_generation(leaf, dev_item))
6633                         return -EINVAL;
6634         }
6635
6636         fill_device_from_item(leaf, dev_item, device);
6637         device->in_fs_metadata = 1;
6638         if (device->writeable && !device->is_tgtdev_for_dev_replace) {
6639                 device->fs_devices->total_rw_bytes += device->total_bytes;
6640                 atomic64_add(device->total_bytes - device->bytes_used,
6641                                 &fs_info->free_chunk_space);
6642         }
6643         ret = 0;
6644         return ret;
6645 }
6646
6647 int btrfs_read_sys_array(struct btrfs_fs_info *fs_info)
6648 {
6649         struct btrfs_root *root = fs_info->tree_root;
6650         struct btrfs_super_block *super_copy = fs_info->super_copy;
6651         struct extent_buffer *sb;
6652         struct btrfs_disk_key *disk_key;
6653         struct btrfs_chunk *chunk;
6654         u8 *array_ptr;
6655         unsigned long sb_array_offset;
6656         int ret = 0;
6657         u32 num_stripes;
6658         u32 array_size;
6659         u32 len = 0;
6660         u32 cur_offset;
6661         u64 type;
6662         struct btrfs_key key;
6663
6664         ASSERT(BTRFS_SUPER_INFO_SIZE <= fs_info->nodesize);
6665         /*
6666          * This will create extent buffer of nodesize, superblock size is
6667          * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will
6668          * overallocate but we can keep it as-is, only the first page is used.
6669          */
6670         sb = btrfs_find_create_tree_block(fs_info, BTRFS_SUPER_INFO_OFFSET);
6671         if (IS_ERR(sb))
6672                 return PTR_ERR(sb);
6673         set_extent_buffer_uptodate(sb);
6674         btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
6675         /*
6676          * The sb extent buffer is artificial and just used to read the system array.
6677          * set_extent_buffer_uptodate() call does not properly mark all it's
6678          * pages up-to-date when the page is larger: extent does not cover the
6679          * whole page and consequently check_page_uptodate does not find all
6680          * the page's extents up-to-date (the hole beyond sb),
6681          * write_extent_buffer then triggers a WARN_ON.
6682          *
6683          * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
6684          * but sb spans only this function. Add an explicit SetPageUptodate call
6685          * to silence the warning eg. on PowerPC 64.
6686          */
6687         if (PAGE_SIZE > BTRFS_SUPER_INFO_SIZE)
6688                 SetPageUptodate(sb->pages[0]);
6689
6690         write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
6691         array_size = btrfs_super_sys_array_size(super_copy);
6692
6693         array_ptr = super_copy->sys_chunk_array;
6694         sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array);
6695         cur_offset = 0;
6696
6697         while (cur_offset < array_size) {
6698                 disk_key = (struct btrfs_disk_key *)array_ptr;
6699                 len = sizeof(*disk_key);
6700                 if (cur_offset + len > array_size)
6701                         goto out_short_read;
6702
6703                 btrfs_disk_key_to_cpu(&key, disk_key);
6704
6705                 array_ptr += len;
6706                 sb_array_offset += len;
6707                 cur_offset += len;
6708
6709                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
6710                         chunk = (struct btrfs_chunk *)sb_array_offset;
6711                         /*
6712                          * At least one btrfs_chunk with one stripe must be
6713                          * present, exact stripe count check comes afterwards
6714                          */
6715                         len = btrfs_chunk_item_size(1);
6716                         if (cur_offset + len > array_size)
6717                                 goto out_short_read;
6718
6719                         num_stripes = btrfs_chunk_num_stripes(sb, chunk);
6720                         if (!num_stripes) {
6721                                 btrfs_err(fs_info,
6722                                         "invalid number of stripes %u in sys_array at offset %u",
6723                                         num_stripes, cur_offset);
6724                                 ret = -EIO;
6725                                 break;
6726                         }
6727
6728                         type = btrfs_chunk_type(sb, chunk);
6729                         if ((type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) {
6730                                 btrfs_err(fs_info,
6731                             "invalid chunk type %llu in sys_array at offset %u",
6732                                         type, cur_offset);
6733                                 ret = -EIO;
6734                                 break;
6735                         }
6736
6737                         len = btrfs_chunk_item_size(num_stripes);
6738                         if (cur_offset + len > array_size)
6739                                 goto out_short_read;
6740
6741                         ret = read_one_chunk(fs_info, &key, sb, chunk);
6742                         if (ret)
6743                                 break;
6744                 } else {
6745                         btrfs_err(fs_info,
6746                             "unexpected item type %u in sys_array at offset %u",
6747                                   (u32)key.type, cur_offset);
6748                         ret = -EIO;
6749                         break;
6750                 }
6751                 array_ptr += len;
6752                 sb_array_offset += len;
6753                 cur_offset += len;
6754         }
6755         clear_extent_buffer_uptodate(sb);
6756         free_extent_buffer_stale(sb);
6757         return ret;
6758
6759 out_short_read:
6760         btrfs_err(fs_info, "sys_array too short to read %u bytes at offset %u",
6761                         len, cur_offset);
6762         clear_extent_buffer_uptodate(sb);
6763         free_extent_buffer_stale(sb);
6764         return -EIO;
6765 }
6766
6767 void btrfs_report_missing_device(struct btrfs_fs_info *fs_info, u64 devid,
6768                                  u8 *uuid)
6769 {
6770         btrfs_warn_rl(fs_info, "devid %llu uuid %pU is missing", devid, uuid);
6771 }
6772
6773 /*
6774  * Check if all chunks in the fs are OK for read-write degraded mount
6775  *
6776  * Return true if all chunks meet the minimal RW mount requirements.
6777  * Return false if any chunk doesn't meet the minimal RW mount requirements.
6778  */
6779 bool btrfs_check_rw_degradable(struct btrfs_fs_info *fs_info)
6780 {
6781         struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
6782         struct extent_map *em;
6783         u64 next_start = 0;
6784         bool ret = true;
6785
6786         read_lock(&map_tree->map_tree.lock);
6787         em = lookup_extent_mapping(&map_tree->map_tree, 0, (u64)-1);
6788         read_unlock(&map_tree->map_tree.lock);
6789         /* No chunk at all? Return false anyway */
6790         if (!em) {
6791                 ret = false;
6792                 goto out;
6793         }
6794         while (em) {
6795                 struct map_lookup *map;
6796                 int missing = 0;
6797                 int max_tolerated;
6798                 int i;
6799
6800                 map = em->map_lookup;
6801                 max_tolerated =
6802                         btrfs_get_num_tolerated_disk_barrier_failures(
6803                                         map->type);
6804                 for (i = 0; i < map->num_stripes; i++) {
6805                         struct btrfs_device *dev = map->stripes[i].dev;
6806
6807                         if (!dev || !dev->bdev || dev->missing ||
6808                             dev->last_flush_error)
6809                                 missing++;
6810                 }
6811                 if (missing > max_tolerated) {
6812                         btrfs_warn(fs_info,
6813         "chunk %llu missing %d devices, max tolerance is %d for writeable mount",
6814                                    em->start, missing, max_tolerated);
6815                         free_extent_map(em);
6816                         ret = false;
6817                         goto out;
6818                 }
6819                 next_start = extent_map_end(em);
6820                 free_extent_map(em);
6821
6822                 read_lock(&map_tree->map_tree.lock);
6823                 em = lookup_extent_mapping(&map_tree->map_tree, next_start,
6824                                            (u64)(-1) - next_start);
6825                 read_unlock(&map_tree->map_tree.lock);
6826         }
6827 out:
6828         return ret;
6829 }
6830
6831 int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info)
6832 {
6833         struct btrfs_root *root = fs_info->chunk_root;
6834         struct btrfs_path *path;
6835         struct extent_buffer *leaf;
6836         struct btrfs_key key;
6837         struct btrfs_key found_key;
6838         int ret;
6839         int slot;
6840         u64 total_dev = 0;
6841
6842         path = btrfs_alloc_path();
6843         if (!path)
6844                 return -ENOMEM;
6845
6846         mutex_lock(&uuid_mutex);
6847         mutex_lock(&fs_info->chunk_mutex);
6848
6849         /*
6850          * Read all device items, and then all the chunk items. All
6851          * device items are found before any chunk item (their object id
6852          * is smaller than the lowest possible object id for a chunk
6853          * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
6854          */
6855         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
6856         key.offset = 0;
6857         key.type = 0;
6858         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6859         if (ret < 0)
6860                 goto error;
6861         while (1) {
6862                 leaf = path->nodes[0];
6863                 slot = path->slots[0];
6864                 if (slot >= btrfs_header_nritems(leaf)) {
6865                         ret = btrfs_next_leaf(root, path);
6866                         if (ret == 0)
6867                                 continue;
6868                         if (ret < 0)
6869                                 goto error;
6870                         break;
6871                 }
6872                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6873                 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
6874                         struct btrfs_dev_item *dev_item;
6875                         dev_item = btrfs_item_ptr(leaf, slot,
6876                                                   struct btrfs_dev_item);
6877                         ret = read_one_dev(fs_info, leaf, dev_item);
6878                         if (ret)
6879                                 goto error;
6880                         total_dev++;
6881                 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
6882                         struct btrfs_chunk *chunk;
6883                         chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
6884                         ret = read_one_chunk(fs_info, &found_key, leaf, chunk);
6885                         if (ret)
6886                                 goto error;
6887                 }
6888                 path->slots[0]++;
6889         }
6890
6891         /*
6892          * After loading chunk tree, we've got all device information,
6893          * do another round of validation checks.
6894          */
6895         if (total_dev != fs_info->fs_devices->total_devices) {
6896                 btrfs_err(fs_info,
6897            "super_num_devices %llu mismatch with num_devices %llu found here",
6898                           btrfs_super_num_devices(fs_info->super_copy),
6899                           total_dev);
6900                 ret = -EINVAL;
6901                 goto error;
6902         }
6903         if (btrfs_super_total_bytes(fs_info->super_copy) <
6904             fs_info->fs_devices->total_rw_bytes) {
6905                 btrfs_err(fs_info,
6906         "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu",
6907                           btrfs_super_total_bytes(fs_info->super_copy),
6908                           fs_info->fs_devices->total_rw_bytes);
6909                 ret = -EINVAL;
6910                 goto error;
6911         }
6912         ret = 0;
6913 error:
6914         mutex_unlock(&fs_info->chunk_mutex);
6915         mutex_unlock(&uuid_mutex);
6916
6917         btrfs_free_path(path);
6918         return ret;
6919 }
6920
6921 void btrfs_init_devices_late(struct btrfs_fs_info *fs_info)
6922 {
6923         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6924         struct btrfs_device *device;
6925
6926         while (fs_devices) {
6927                 mutex_lock(&fs_devices->device_list_mutex);
6928                 list_for_each_entry(device, &fs_devices->devices, dev_list)
6929                         device->fs_info = fs_info;
6930                 mutex_unlock(&fs_devices->device_list_mutex);
6931
6932                 fs_devices = fs_devices->seed;
6933         }
6934 }
6935
6936 static void __btrfs_reset_dev_stats(struct btrfs_device *dev)
6937 {
6938         int i;
6939
6940         for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6941                 btrfs_dev_stat_reset(dev, i);
6942 }
6943
6944 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
6945 {
6946         struct btrfs_key key;
6947         struct btrfs_key found_key;
6948         struct btrfs_root *dev_root = fs_info->dev_root;
6949         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6950         struct extent_buffer *eb;
6951         int slot;
6952         int ret = 0;
6953         struct btrfs_device *device;
6954         struct btrfs_path *path = NULL;
6955         int i;
6956
6957         path = btrfs_alloc_path();
6958         if (!path) {
6959                 ret = -ENOMEM;
6960                 goto out;
6961         }
6962
6963         mutex_lock(&fs_devices->device_list_mutex);
6964         list_for_each_entry(device, &fs_devices->devices, dev_list) {
6965                 int item_size;
6966                 struct btrfs_dev_stats_item *ptr;
6967
6968                 key.objectid = BTRFS_DEV_STATS_OBJECTID;
6969                 key.type = BTRFS_PERSISTENT_ITEM_KEY;
6970                 key.offset = device->devid;
6971                 ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
6972                 if (ret) {
6973                         __btrfs_reset_dev_stats(device);
6974                         device->dev_stats_valid = 1;
6975                         btrfs_release_path(path);
6976                         continue;
6977                 }
6978                 slot = path->slots[0];
6979                 eb = path->nodes[0];
6980                 btrfs_item_key_to_cpu(eb, &found_key, slot);
6981                 item_size = btrfs_item_size_nr(eb, slot);
6982
6983                 ptr = btrfs_item_ptr(eb, slot,
6984                                      struct btrfs_dev_stats_item);
6985
6986                 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
6987                         if (item_size >= (1 + i) * sizeof(__le64))
6988                                 btrfs_dev_stat_set(device, i,
6989                                         btrfs_dev_stats_value(eb, ptr, i));
6990                         else
6991                                 btrfs_dev_stat_reset(device, i);
6992                 }
6993
6994                 device->dev_stats_valid = 1;
6995                 btrfs_dev_stat_print_on_load(device);
6996                 btrfs_release_path(path);
6997         }
6998         mutex_unlock(&fs_devices->device_list_mutex);
6999
7000 out:
7001         btrfs_free_path(path);
7002         return ret < 0 ? ret : 0;
7003 }
7004
7005 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
7006                                 struct btrfs_fs_info *fs_info,
7007                                 struct btrfs_device *device)
7008 {
7009         struct btrfs_root *dev_root = fs_info->dev_root;
7010         struct btrfs_path *path;
7011         struct btrfs_key key;
7012         struct extent_buffer *eb;
7013         struct btrfs_dev_stats_item *ptr;
7014         int ret;
7015         int i;
7016
7017         key.objectid = BTRFS_DEV_STATS_OBJECTID;
7018         key.type = BTRFS_PERSISTENT_ITEM_KEY;
7019         key.offset = device->devid;
7020
7021         path = btrfs_alloc_path();
7022         if (!path)
7023                 return -ENOMEM;
7024         ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
7025         if (ret < 0) {
7026                 btrfs_warn_in_rcu(fs_info,
7027                         "error %d while searching for dev_stats item for device %s",
7028                               ret, rcu_str_deref(device->name));
7029                 goto out;
7030         }
7031
7032         if (ret == 0 &&
7033             btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
7034                 /* need to delete old one and insert a new one */
7035                 ret = btrfs_del_item(trans, dev_root, path);
7036                 if (ret != 0) {
7037                         btrfs_warn_in_rcu(fs_info,
7038                                 "delete too small dev_stats item for device %s failed %d",
7039                                       rcu_str_deref(device->name), ret);
7040                         goto out;
7041                 }
7042                 ret = 1;
7043         }
7044
7045         if (ret == 1) {
7046                 /* need to insert a new item */
7047                 btrfs_release_path(path);
7048                 ret = btrfs_insert_empty_item(trans, dev_root, path,
7049                                               &key, sizeof(*ptr));
7050                 if (ret < 0) {
7051                         btrfs_warn_in_rcu(fs_info,
7052                                 "insert dev_stats item for device %s failed %d",
7053                                 rcu_str_deref(device->name), ret);
7054                         goto out;
7055                 }
7056         }
7057
7058         eb = path->nodes[0];
7059         ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
7060         for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7061                 btrfs_set_dev_stats_value(eb, ptr, i,
7062                                           btrfs_dev_stat_read(device, i));
7063         btrfs_mark_buffer_dirty(eb);
7064
7065 out:
7066         btrfs_free_path(path);
7067         return ret;
7068 }
7069
7070 /*
7071  * called from commit_transaction. Writes all changed device stats to disk.
7072  */
7073 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans,
7074                         struct btrfs_fs_info *fs_info)
7075 {
7076         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7077         struct btrfs_device *device;
7078         int stats_cnt;
7079         int ret = 0;
7080
7081         mutex_lock(&fs_devices->device_list_mutex);
7082         list_for_each_entry(device, &fs_devices->devices, dev_list) {
7083                 if (!device->dev_stats_valid || !btrfs_dev_stats_dirty(device))
7084                         continue;
7085
7086                 stats_cnt = atomic_read(&device->dev_stats_ccnt);
7087                 ret = update_dev_stat_item(trans, fs_info, device);
7088                 if (!ret)
7089                         atomic_sub(stats_cnt, &device->dev_stats_ccnt);
7090         }
7091         mutex_unlock(&fs_devices->device_list_mutex);
7092
7093         return ret;
7094 }
7095
7096 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
7097 {
7098         btrfs_dev_stat_inc(dev, index);
7099         btrfs_dev_stat_print_on_error(dev);
7100 }
7101
7102 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
7103 {
7104         if (!dev->dev_stats_valid)
7105                 return;
7106         btrfs_err_rl_in_rcu(dev->fs_info,
7107                 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7108                            rcu_str_deref(dev->name),
7109                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7110                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7111                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7112                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7113                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7114 }
7115
7116 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
7117 {
7118         int i;
7119
7120         for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7121                 if (btrfs_dev_stat_read(dev, i) != 0)
7122                         break;
7123         if (i == BTRFS_DEV_STAT_VALUES_MAX)
7124                 return; /* all values == 0, suppress message */
7125
7126         btrfs_info_in_rcu(dev->fs_info,
7127                 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7128                rcu_str_deref(dev->name),
7129                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7130                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7131                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7132                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7133                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7134 }
7135
7136 int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info,
7137                         struct btrfs_ioctl_get_dev_stats *stats)
7138 {
7139         struct btrfs_device *dev;
7140         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7141         int i;
7142
7143         mutex_lock(&fs_devices->device_list_mutex);
7144         dev = btrfs_find_device(fs_info, stats->devid, NULL, NULL);
7145         mutex_unlock(&fs_devices->device_list_mutex);
7146
7147         if (!dev) {
7148                 btrfs_warn(fs_info, "get dev_stats failed, device not found");
7149                 return -ENODEV;
7150         } else if (!dev->dev_stats_valid) {
7151                 btrfs_warn(fs_info, "get dev_stats failed, not yet valid");
7152                 return -ENODEV;
7153         } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
7154                 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7155                         if (stats->nr_items > i)
7156                                 stats->values[i] =
7157                                         btrfs_dev_stat_read_and_reset(dev, i);
7158                         else
7159                                 btrfs_dev_stat_reset(dev, i);
7160                 }
7161         } else {
7162                 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7163                         if (stats->nr_items > i)
7164                                 stats->values[i] = btrfs_dev_stat_read(dev, i);
7165         }
7166         if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
7167                 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
7168         return 0;
7169 }
7170
7171 void btrfs_scratch_superblocks(struct block_device *bdev, const char *device_path)
7172 {
7173         struct buffer_head *bh;
7174         struct btrfs_super_block *disk_super;
7175         int copy_num;
7176
7177         if (!bdev)
7178                 return;
7179
7180         for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX;
7181                 copy_num++) {
7182
7183                 if (btrfs_read_dev_one_super(bdev, copy_num, &bh))
7184                         continue;
7185
7186                 disk_super = (struct btrfs_super_block *)bh->b_data;
7187
7188                 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
7189                 set_buffer_dirty(bh);
7190                 sync_dirty_buffer(bh);
7191                 brelse(bh);
7192         }
7193
7194         /* Notify udev that device has changed */
7195         btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
7196
7197         /* Update ctime/mtime for device path for libblkid */
7198         update_dev_time(device_path);
7199 }
7200
7201 /*
7202  * Update the size of all devices, which is used for writing out the
7203  * super blocks.
7204  */
7205 void btrfs_update_commit_device_size(struct btrfs_fs_info *fs_info)
7206 {
7207         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7208         struct btrfs_device *curr, *next;
7209
7210         if (list_empty(&fs_devices->resized_devices))
7211                 return;
7212
7213         mutex_lock(&fs_devices->device_list_mutex);
7214         mutex_lock(&fs_info->chunk_mutex);
7215         list_for_each_entry_safe(curr, next, &fs_devices->resized_devices,
7216                                  resized_list) {
7217                 list_del_init(&curr->resized_list);
7218                 curr->commit_total_bytes = curr->disk_total_bytes;
7219         }
7220         mutex_unlock(&fs_info->chunk_mutex);
7221         mutex_unlock(&fs_devices->device_list_mutex);
7222 }
7223
7224 /* Must be invoked during the transaction commit */
7225 void btrfs_update_commit_device_bytes_used(struct btrfs_fs_info *fs_info,
7226                                         struct btrfs_transaction *transaction)
7227 {
7228         struct extent_map *em;
7229         struct map_lookup *map;
7230         struct btrfs_device *dev;
7231         int i;
7232
7233         if (list_empty(&transaction->pending_chunks))
7234                 return;
7235
7236         /* In order to kick the device replace finish process */
7237         mutex_lock(&fs_info->chunk_mutex);
7238         list_for_each_entry(em, &transaction->pending_chunks, list) {
7239                 map = em->map_lookup;
7240
7241                 for (i = 0; i < map->num_stripes; i++) {
7242                         dev = map->stripes[i].dev;
7243                         dev->commit_bytes_used = dev->bytes_used;
7244                 }
7245         }
7246         mutex_unlock(&fs_info->chunk_mutex);
7247 }
7248
7249 void btrfs_set_fs_info_ptr(struct btrfs_fs_info *fs_info)
7250 {
7251         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7252         while (fs_devices) {
7253                 fs_devices->fs_info = fs_info;
7254                 fs_devices = fs_devices->seed;
7255         }
7256 }
7257
7258 void btrfs_reset_fs_info_ptr(struct btrfs_fs_info *fs_info)
7259 {
7260         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7261         while (fs_devices) {
7262                 fs_devices->fs_info = NULL;
7263                 fs_devices = fs_devices->seed;
7264         }
7265 }