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
 143 DEFINE_MUTEX(uuid_mutex);
 144 static LIST_HEAD(fs_uuids);
 145 struct list_head *btrfs_get_fs_uuids(void)
 146 {
 147         return &fs_uuids;
 148 }
 149
 150 static struct btrfs_fs_devices *__alloc_fs_devices(void)
 151 {
 152         struct btrfs_fs_devices *fs_devs;
 153
 154         fs_devs = kzalloc(sizeof(*fs_devs), GFP_KERNEL);
 155         if (!fs_devs)
 156                 return ERR_PTR(-ENOMEM);
 157
 158         mutex_init(&fs_devs->device_list_mutex);
 159
 160         INIT_LIST_HEAD(&fs_devs->devices);
 161         INIT_LIST_HEAD(&fs_devs->resized_devices);
 162         INIT_LIST_HEAD(&fs_devs->alloc_list);
 163         INIT_LIST_HEAD(&fs_devs->list);
 164
 165         return fs_devs;
 166 }
 167
 168 /**
 169  * alloc_fs_devices - allocate struct btrfs_fs_devices
 170  * @fsid:       a pointer to UUID for this FS.  If NULL a new UUID is
 171  *              generated.
 172  *
 173  * Return: a pointer to a new &struct btrfs_fs_devices on success;
 174  * ERR_PTR() on error.  Returned struct is not linked onto any lists and
 175  * can be destroyed with kfree() right away.
 176  */
 177 static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid)
 178 {
 179         struct btrfs_fs_devices *fs_devs;
 180
 181         fs_devs = __alloc_fs_devices();
 182         if (IS_ERR(fs_devs))
 183                 return fs_devs;
 184
 185         if (fsid)
 186                 memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE);
 187         else
 188                 generate_random_uuid(fs_devs->fsid);
 189
 190         return fs_devs;
 191 }
 192
 193 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
 194 {
 195         struct btrfs_device *device;
 196         WARN_ON(fs_devices->opened);
 197         while (!list_empty(&fs_devices->devices)) {
 198                 device = list_entry(fs_devices->devices.next,
 199                                     struct btrfs_device, dev_list);
 200                 list_del(&device->dev_list);
 201                 rcu_string_free(device->name);
 202                 kfree(device);
 203         }
 204         kfree(fs_devices);
 205 }
 206
 207 static void btrfs_kobject_uevent(struct block_device *bdev,
 208                                  enum kobject_action action)
 209 {
 210         int ret;
 211
 212         ret = kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, action);
 213         if (ret)
 214                 pr_warn("BTRFS: Sending event '%d' to kobject: '%s' (%p): failed\n",
 215                         action,
 216                         kobject_name(&disk_to_dev(bdev->bd_disk)->kobj),
 217                         &disk_to_dev(bdev->bd_disk)->kobj);
 218 }
 219
 220 void btrfs_cleanup_fs_uuids(void)
 221 {
 222         struct btrfs_fs_devices *fs_devices;
 223
 224         while (!list_empty(&fs_uuids)) {
 225                 fs_devices = list_entry(fs_uuids.next,
 226                                         struct btrfs_fs_devices, list);
 227                 list_del(&fs_devices->list);
 228                 free_fs_devices(fs_devices);
 229         }
 230 }
 231
 232 static struct btrfs_device *__alloc_device(void)
 233 {
 234         struct btrfs_device *dev;
 235
 236         dev = kzalloc(sizeof(*dev), GFP_KERNEL);
 237         if (!dev)
 238                 return ERR_PTR(-ENOMEM);
 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 static noinline struct btrfs_device *__find_device(struct list_head *head,
 257                                                    u64 devid, u8 *uuid)
 258 {
 259         struct btrfs_device *dev;
 260
 261         list_for_each_entry(dev, head, dev_list) {
 262                 if (dev->devid == devid &&
 263                     (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
 264                         return dev;
 265                 }
 266         }
 267         return NULL;
 268 }
 269
 270 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
 271 {
 272         struct btrfs_fs_devices *fs_devices;
 273
 274         list_for_each_entry(fs_devices, &fs_uuids, list) {
 275                 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
 276                         return fs_devices;
 277         }
 278         return NULL;
 279 }
 280
 281 static int
 282 btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder,
 283                       int flush, struct block_device **bdev,
 284                       struct buffer_head **bh)
 285 {
 286         int ret;
 287
 288         *bdev = blkdev_get_by_path(device_path, flags, holder);
 289
 290         if (IS_ERR(*bdev)) {
 291                 ret = PTR_ERR(*bdev);
 292                 goto error;
 293         }
 294
 295         if (flush)
 296                 filemap_write_and_wait((*bdev)->bd_inode->i_mapping);
 297         ret = set_blocksize(*bdev, 4096);
 298         if (ret) {
 299                 blkdev_put(*bdev, flags);
 300                 goto error;
 301         }
 302         invalidate_bdev(*bdev);
 303         *bh = btrfs_read_dev_super(*bdev);
 304         if (IS_ERR(*bh)) {
 305                 ret = PTR_ERR(*bh);
 306                 blkdev_put(*bdev, flags);
 307                 goto error;
 308         }
 309
 310         return 0;
 311
 312 error:
 313         *bdev = NULL;
 314         *bh = NULL;
 315         return ret;
 316 }
 317
 318 static void requeue_list(struct btrfs_pending_bios *pending_bios,
 319                         struct bio *head, struct bio *tail)
 320 {
 321
 322         struct bio *old_head;
 323
 324         old_head = pending_bios->head;
 325         pending_bios->head = head;
 326         if (pending_bios->tail)
 327                 tail->bi_next = old_head;
 328         else
 329                 pending_bios->tail = tail;
 330 }
 331
 332 /*
 333  * we try to collect pending bios for a device so we don't get a large
 334  * number of procs sending bios down to the same device.  This greatly
 335  * improves the schedulers ability to collect and merge the bios.
 336  *
 337  * But, it also turns into a long list of bios to process and that is sure
 338  * to eventually make the worker thread block.  The solution here is to
 339  * make some progress and then put this work struct back at the end of
 340  * the list if the block device is congested.  This way, multiple devices
 341  * can make progress from a single worker thread.
 342  */
 343 static noinline void run_scheduled_bios(struct btrfs_device *device)
 344 {
 345         struct btrfs_fs_info *fs_info = device->fs_info;
 346         struct bio *pending;
 347         struct backing_dev_info *bdi;
 348         struct btrfs_pending_bios *pending_bios;
 349         struct bio *tail;
 350         struct bio *cur;
 351         int again = 0;
 352         unsigned long num_run;
 353         unsigned long batch_run = 0;
 354         unsigned long limit;
 355         unsigned long last_waited = 0;
 356         int force_reg = 0;
 357         int sync_pending = 0;
 358         struct blk_plug plug;
 359
 360         /*
 361          * this function runs all the bios we've collected for
 362          * a particular device.  We don't want to wander off to
 363          * another device without first sending all of these down.
 364          * So, setup a plug here and finish it off before we return
 365          */
 366         blk_start_plug(&plug);
 367
 368         bdi = device->bdev->bd_bdi;
 369         limit = btrfs_async_submit_limit(fs_info);
 370         limit = limit * 2 / 3;
 371
 372 loop:
 373         spin_lock(&device->io_lock);
 374
 375 loop_lock:
 376         num_run = 0;
 377
 378         /* take all the bios off the list at once and process them
 379          * later on (without the lock held).  But, remember the
 380          * tail and other pointers so the bios can be properly reinserted
 381          * into the list if we hit congestion
 382          */
 383         if (!force_reg && device->pending_sync_bios.head) {
 384                 pending_bios = &device->pending_sync_bios;
 385                 force_reg = 1;
 386         } else {
 387                 pending_bios = &device->pending_bios;
 388                 force_reg = 0;
 389         }
 390
 391         pending = pending_bios->head;
 392         tail = pending_bios->tail;
 393         WARN_ON(pending && !tail);
 394
 395         /*
 396          * if pending was null this time around, no bios need processing
 397          * at all and we can stop.  Otherwise it'll loop back up again
 398          * and do an additional check so no bios are missed.
 399          *
 400          * device->running_pending is used to synchronize with the
 401          * schedule_bio code.
 402          */
 403         if (device->pending_sync_bios.head == NULL &&
 404             device->pending_bios.head == NULL) {
 405                 again = 0;
 406                 device->running_pending = 0;
 407         } else {
 408                 again = 1;
 409                 device->running_pending = 1;
 410         }
 411
 412         pending_bios->head = NULL;
 413         pending_bios->tail = NULL;
 414
 415         spin_unlock(&device->io_lock);
 416
 417         while (pending) {
 418
 419                 rmb();
 420                 /* we want to work on both lists, but do more bios on the
 421                  * sync list than the regular list
 422                  */
 423                 if ((num_run > 32 &&
 424                     pending_bios != &device->pending_sync_bios &&
 425                     device->pending_sync_bios.head) ||
 426                    (num_run > 64 && pending_bios == &device->pending_sync_bios &&
 427                     device->pending_bios.head)) {
 428                         spin_lock(&device->io_lock);
 429                         requeue_list(pending_bios, pending, tail);
 430                         goto loop_lock;
 431                 }
 432
 433                 cur = pending;
 434                 pending = pending->bi_next;
 435                 cur->bi_next = NULL;
 436
 437                 /*
 438                  * atomic_dec_return implies a barrier for waitqueue_active
 439                  */
 440                 if (atomic_dec_return(&fs_info->nr_async_bios) < limit &&
 441                     waitqueue_active(&fs_info->async_submit_wait))
 442                         wake_up(&fs_info->async_submit_wait);
 443
 444                 BUG_ON(atomic_read(&cur->__bi_cnt) == 0);
 445
 446                 /*
 447                  * if we're doing the sync list, record that our
 448                  * plug has some sync requests on it
 449                  *
 450                  * If we're doing the regular list and there are
 451                  * sync requests sitting around, unplug before
 452                  * we add more
 453                  */
 454                 if (pending_bios == &device->pending_sync_bios) {
 455                         sync_pending = 1;
 456                 } else if (sync_pending) {
 457                         blk_finish_plug(&plug);
 458                         blk_start_plug(&plug);
 459                         sync_pending = 0;
 460                 }
 461
 462                 btrfsic_submit_bio(cur);
 463                 num_run++;
 464                 batch_run++;
 465
 466                 cond_resched();
 467
 468                 /*
 469                  * we made progress, there is more work to do and the bdi
 470                  * is now congested.  Back off and let other work structs
 471                  * run instead
 472                  */
 473                 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
 474                     fs_info->fs_devices->open_devices > 1) {
 475                         struct io_context *ioc;
 476
 477                         ioc = current->io_context;
 478
 479                         /*
 480                          * the main goal here is that we don't want to
 481                          * block if we're going to be able to submit
 482                          * more requests without blocking.
 483                          *
 484                          * This code does two great things, it pokes into
 485                          * the elevator code from a filesystem _and_
 486                          * it makes assumptions about how batching works.
 487                          */
 488                         if (ioc && ioc->nr_batch_requests > 0 &&
 489                             time_before(jiffies, ioc->last_waited + HZ/50UL) &&
 490                             (last_waited == 0 ||
 491                              ioc->last_waited == last_waited)) {
 492                                 /*
 493                                  * we want to go through our batch of
 494                                  * requests and stop.  So, we copy out
 495                                  * the ioc->last_waited time and test
 496                                  * against it before looping
 497                                  */
 498                                 last_waited = ioc->last_waited;
 499                                 cond_resched();
 500                                 continue;
 501                         }
 502                         spin_lock(&device->io_lock);
 503                         requeue_list(pending_bios, pending, tail);
 504                         device->running_pending = 1;
 505
 506                         spin_unlock(&device->io_lock);
 507                         btrfs_queue_work(fs_info->submit_workers,
 508                                          &device->work);
 509                         goto done;
 510                 }
 511                 /* unplug every 64 requests just for good measure */
 512                 if (batch_run % 64 == 0) {
 513                         blk_finish_plug(&plug);
 514                         blk_start_plug(&plug);
 515                         sync_pending = 0;
 516                 }
 517         }
 518
 519         cond_resched();
 520         if (again)
 521                 goto loop;
 522
 523         spin_lock(&device->io_lock);
 524         if (device->pending_bios.head || device->pending_sync_bios.head)
 525                 goto loop_lock;
 526         spin_unlock(&device->io_lock);
 527
 528 done:
 529         blk_finish_plug(&plug);
 530 }
 531
 532 static void pending_bios_fn(struct btrfs_work *work)
 533 {
 534         struct btrfs_device *device;
 535
 536         device = container_of(work, struct btrfs_device, work);
 537         run_scheduled_bios(device);
 538 }
 539
 540
 541 void btrfs_free_stale_device(struct btrfs_device *cur_dev)
 542 {
 543         struct btrfs_fs_devices *fs_devs;
 544         struct btrfs_device *dev;
 545
 546         if (!cur_dev->name)
 547                 return;
 548
 549         list_for_each_entry(fs_devs, &fs_uuids, list) {
 550                 int del = 1;
 551
 552                 if (fs_devs->opened)
 553                         continue;
 554                 if (fs_devs->seeding)
 555                         continue;
 556
 557                 list_for_each_entry(dev, &fs_devs->devices, dev_list) {
 558
 559                         if (dev == cur_dev)
 560                                 continue;
 561                         if (!dev->name)
 562                                 continue;
 563
 564                         /*
 565                          * Todo: This won't be enough. What if the same device
 566                          * comes back (with new uuid and) with its mapper path?
 567                          * But for now, this does help as mostly an admin will
 568                          * either use mapper or non mapper path throughout.
 569                          */
 570                         rcu_read_lock();
 571                         del = strcmp(rcu_str_deref(dev->name),
 572                                                 rcu_str_deref(cur_dev->name));
 573                         rcu_read_unlock();
 574                         if (!del)
 575                                 break;
 576                 }
 577
 578                 if (!del) {
 579                         /* delete the stale device */
 580                         if (fs_devs->num_devices == 1) {
 581                                 btrfs_sysfs_remove_fsid(fs_devs);
 582                                 list_del(&fs_devs->list);
 583                                 free_fs_devices(fs_devs);
 584                         } else {
 585                                 fs_devs->num_devices--;
 586                                 list_del(&dev->dev_list);
 587                                 rcu_string_free(dev->name);
 588                                 kfree(dev);
 589                         }
 590                         break;
 591                 }
 592         }
 593 }
 594
 595 /*
 596  * Add new device to list of registered devices
 597  *
 598  * Returns:
 599  * 1   - first time device is seen
 600  * 0   - device already known
 601  * < 0 - error
 602  */
 603 static noinline int device_list_add(const char *path,
 604                            struct btrfs_super_block *disk_super,
 605                            u64 devid, struct btrfs_fs_devices **fs_devices_ret)
 606 {
 607         struct btrfs_device *device;
 608         struct btrfs_fs_devices *fs_devices;
 609         struct rcu_string *name;
 610         int ret = 0;
 611         u64 found_transid = btrfs_super_generation(disk_super);
 612
 613         fs_devices = find_fsid(disk_super->fsid);
 614         if (!fs_devices) {
 615                 fs_devices = alloc_fs_devices(disk_super->fsid);
 616                 if (IS_ERR(fs_devices))
 617                         return PTR_ERR(fs_devices);
 618
 619                 list_add(&fs_devices->list, &fs_uuids);
 620
 621                 device = NULL;
 622         } else {
 623                 device = __find_device(&fs_devices->devices, devid,
 624                                        disk_super->dev_item.uuid);
 625         }
 626
 627         if (!device) {
 628                 if (fs_devices->opened)
 629                         return -EBUSY;
 630
 631                 device = btrfs_alloc_device(NULL, &devid,
 632                                             disk_super->dev_item.uuid);
 633                 if (IS_ERR(device)) {
 634                         /* we can safely leave the fs_devices entry around */
 635                         return PTR_ERR(device);
 636                 }
 637
 638                 name = rcu_string_strdup(path, GFP_NOFS);
 639                 if (!name) {
 640                         kfree(device);
 641                         return -ENOMEM;
 642                 }
 643                 rcu_assign_pointer(device->name, name);
 644
 645                 mutex_lock(&fs_devices->device_list_mutex);
 646                 list_add_rcu(&device->dev_list, &fs_devices->devices);
 647                 fs_devices->num_devices++;
 648                 mutex_unlock(&fs_devices->device_list_mutex);
 649
 650                 ret = 1;
 651                 device->fs_devices = fs_devices;
 652         } else if (!device->name || strcmp(device->name->str, path)) {
 653                 /*
 654                  * When FS is already mounted.
 655                  * 1. If you are here and if the device->name is NULL that
 656                  *    means this device was missing at time of FS mount.
 657                  * 2. If you are here and if the device->name is different
 658                  *    from 'path' that means either
 659                  *      a. The same device disappeared and reappeared with
 660                  *         different name. or
 661                  *      b. The missing-disk-which-was-replaced, has
 662                  *         reappeared now.
 663                  *
 664                  * We must allow 1 and 2a above. But 2b would be a spurious
 665                  * and unintentional.
 666                  *
 667                  * Further in case of 1 and 2a above, the disk at 'path'
 668                  * would have missed some transaction when it was away and
 669                  * in case of 2a the stale bdev has to be updated as well.
 670                  * 2b must not be allowed at all time.
 671                  */
 672
 673                 /*
 674                  * For now, we do allow update to btrfs_fs_device through the
 675                  * btrfs dev scan cli after FS has been mounted.  We're still
 676                  * tracking a problem where systems fail mount by subvolume id
 677                  * when we reject replacement on a mounted FS.
 678                  */
 679                 if (!fs_devices->opened && found_transid < device->generation) {
 680                         /*
 681                          * That is if the FS is _not_ mounted and if you
 682                          * are here, that means there is more than one
 683                          * disk with same uuid and devid.We keep the one
 684                          * with larger generation number or the last-in if
 685                          * generation are equal.
 686                          */
 687                         return -EEXIST;
 688                 }
 689
 690                 name = rcu_string_strdup(path, GFP_NOFS);
 691                 if (!name)
 692                         return -ENOMEM;
 693                 rcu_string_free(device->name);
 694                 rcu_assign_pointer(device->name, name);
 695                 if (device->missing) {
 696                         fs_devices->missing_devices--;
 697                         device->missing = 0;
 698                 }
 699         }
 700
 701         /*
 702          * Unmount does not free the btrfs_device struct but would zero
 703          * generation along with most of the other members. So just update
 704          * it back. We need it to pick the disk with largest generation
 705          * (as above).
 706          */
 707         if (!fs_devices->opened)
 708                 device->generation = found_transid;
 709
 710         /*
 711          * if there is new btrfs on an already registered device,
 712          * then remove the stale device entry.
 713          */
 714         if (ret > 0)
 715                 btrfs_free_stale_device(device);
 716
 717         *fs_devices_ret = fs_devices;
 718
 719         return ret;
 720 }
 721
 722 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
 723 {
 724         struct btrfs_fs_devices *fs_devices;
 725         struct btrfs_device *device;
 726         struct btrfs_device *orig_dev;
 727
 728         fs_devices = alloc_fs_devices(orig->fsid);
 729         if (IS_ERR(fs_devices))
 730                 return fs_devices;
 731
 732         mutex_lock(&orig->device_list_mutex);
 733         fs_devices->total_devices = orig->total_devices;
 734
 735         /* We have held the volume lock, it is safe to get the devices. */
 736         list_for_each_entry(orig_dev, &orig->devices, dev_list) {
 737                 struct rcu_string *name;
 738
 739                 device = btrfs_alloc_device(NULL, &orig_dev->devid,
 740                                             orig_dev->uuid);
 741                 if (IS_ERR(device))
 742                         goto error;
 743
 744                 /*
 745                  * This is ok to do without rcu read locked because we hold the
 746                  * uuid mutex so nothing we touch in here is going to disappear.
 747                  */
 748                 if (orig_dev->name) {
 749                         name = rcu_string_strdup(orig_dev->name->str,
 750                                         GFP_KERNEL);
 751                         if (!name) {
 752                                 kfree(device);
 753                                 goto error;
 754                         }
 755                         rcu_assign_pointer(device->name, name);
 756                 }
 757
 758                 list_add(&device->dev_list, &fs_devices->devices);
 759                 device->fs_devices = fs_devices;
 760                 fs_devices->num_devices++;
 761         }
 762         mutex_unlock(&orig->device_list_mutex);
 763         return fs_devices;
 764 error:
 765         mutex_unlock(&orig->device_list_mutex);
 766         free_fs_devices(fs_devices);
 767         return ERR_PTR(-ENOMEM);
 768 }
 769
 770 void btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices, int step)
 771 {
 772         struct btrfs_device *device, *next;
 773         struct btrfs_device *latest_dev = NULL;
 774
 775         mutex_lock(&uuid_mutex);
 776 again:
 777         /* This is the initialized path, it is safe to release the devices. */
 778         list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
 779                 if (device->in_fs_metadata) {
 780                         if (!device->is_tgtdev_for_dev_replace &&
 781                             (!latest_dev ||
 782                              device->generation > latest_dev->generation)) {
 783                                 latest_dev = device;
 784                         }
 785                         continue;
 786                 }
 787
 788                 if (device->devid == BTRFS_DEV_REPLACE_DEVID) {
 789                         /*
 790                          * In the first step, keep the device which has
 791                          * the correct fsid and the devid that is used
 792                          * for the dev_replace procedure.
 793                          * In the second step, the dev_replace state is
 794                          * read from the device tree and it is known
 795                          * whether the procedure is really active or
 796                          * not, which means whether this device is
 797                          * used or whether it should be removed.
 798                          */
 799                         if (step == 0 || device->is_tgtdev_for_dev_replace) {
 800                                 continue;
 801                         }
 802                 }
 803                 if (device->bdev) {
 804                         blkdev_put(device->bdev, device->mode);
 805                         device->bdev = NULL;
 806                         fs_devices->open_devices--;
 807                 }
 808                 if (device->writeable) {
 809                         list_del_init(&device->dev_alloc_list);
 810                         device->writeable = 0;
 811                         if (!device->is_tgtdev_for_dev_replace)
 812                                 fs_devices->rw_devices--;
 813                 }
 814                 list_del_init(&device->dev_list);
 815                 fs_devices->num_devices--;
 816                 rcu_string_free(device->name);
 817                 kfree(device);
 818         }
 819
 820         if (fs_devices->seed) {
 821                 fs_devices = fs_devices->seed;
 822                 goto again;
 823         }
 824
 825         fs_devices->latest_bdev = latest_dev->bdev;
 826
 827         mutex_unlock(&uuid_mutex);
 828 }
 829
 830 static void __free_device(struct work_struct *work)
 831 {
 832         struct btrfs_device *device;
 833
 834         device = container_of(work, struct btrfs_device, rcu_work);
 835         rcu_string_free(device->name);
 836         kfree(device);
 837 }
 838
 839 static void free_device(struct rcu_head *head)
 840 {
 841         struct btrfs_device *device;
 842
 843         device = container_of(head, struct btrfs_device, rcu);
 844
 845         INIT_WORK(&device->rcu_work, __free_device);
 846         schedule_work(&device->rcu_work);
 847 }
 848
 849 static void btrfs_close_bdev(struct btrfs_device *device)
 850 {
 851         if (device->bdev && device->writeable) {
 852                 sync_blockdev(device->bdev);
 853                 invalidate_bdev(device->bdev);
 854         }
 855
 856         if (device->bdev)
 857                 blkdev_put(device->bdev, device->mode);
 858 }
 859
 860 static void btrfs_prepare_close_one_device(struct btrfs_device *device)
 861 {
 862         struct btrfs_fs_devices *fs_devices = device->fs_devices;
 863         struct btrfs_device *new_device;
 864         struct rcu_string *name;
 865
 866         if (device->bdev)
 867                 fs_devices->open_devices--;
 868
 869         if (device->writeable &&
 870             device->devid != BTRFS_DEV_REPLACE_DEVID) {
 871                 list_del_init(&device->dev_alloc_list);
 872                 fs_devices->rw_devices--;
 873         }
 874
 875         if (device->missing)
 876                 fs_devices->missing_devices--;
 877
 878         new_device = btrfs_alloc_device(NULL, &device->devid,
 879                                         device->uuid);
 880         BUG_ON(IS_ERR(new_device)); /* -ENOMEM */
 881
 882         /* Safe because we are under uuid_mutex */
 883         if (device->name) {
 884                 name = rcu_string_strdup(device->name->str, GFP_NOFS);
 885                 BUG_ON(!name); /* -ENOMEM */
 886                 rcu_assign_pointer(new_device->name, name);
 887         }
 888
 889         list_replace_rcu(&device->dev_list, &new_device->dev_list);
 890         new_device->fs_devices = device->fs_devices;
 891 }
 892
 893 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
 894 {
 895         struct btrfs_device *device, *tmp;
 896         struct list_head pending_put;
 897
 898         INIT_LIST_HEAD(&pending_put);
 899
 900         if (--fs_devices->opened > 0)
 901                 return 0;
 902
 903         mutex_lock(&fs_devices->device_list_mutex);
 904         list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list) {
 905                 btrfs_prepare_close_one_device(device);
 906                 list_add(&device->dev_list, &pending_put);
 907         }
 908         mutex_unlock(&fs_devices->device_list_mutex);
 909
 910         /*
 911          * btrfs_show_devname() is using the device_list_mutex,
 912          * sometimes call to blkdev_put() leads vfs calling
 913          * into this func. So do put outside of device_list_mutex,
 914          * as of now.
 915          */
 916         while (!list_empty(&pending_put)) {
 917                 device = list_first_entry(&pending_put,
 918                                 struct btrfs_device, dev_list);
 919                 list_del(&device->dev_list);
 920                 btrfs_close_bdev(device);
 921                 call_rcu(&device->rcu, free_device);
 922         }
 923
 924         WARN_ON(fs_devices->open_devices);
 925         WARN_ON(fs_devices->rw_devices);
 926         fs_devices->opened = 0;
 927         fs_devices->seeding = 0;
 928
 929         return 0;
 930 }
 931
 932 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
 933 {
 934         struct btrfs_fs_devices *seed_devices = NULL;
 935         int ret;
 936
 937         mutex_lock(&uuid_mutex);
 938         ret = __btrfs_close_devices(fs_devices);
 939         if (!fs_devices->opened) {
 940                 seed_devices = fs_devices->seed;
 941                 fs_devices->seed = NULL;
 942         }
 943         mutex_unlock(&uuid_mutex);
 944
 945         while (seed_devices) {
 946                 fs_devices = seed_devices;
 947                 seed_devices = fs_devices->seed;
 948                 __btrfs_close_devices(fs_devices);
 949                 free_fs_devices(fs_devices);
 950         }
 951         /*
 952          * Wait for rcu kworkers under __btrfs_close_devices
 953          * to finish all blkdev_puts so device is really
 954          * free when umount is done.
 955          */
 956         rcu_barrier();
 957         return ret;
 958 }
 959
 960 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
 961                                 fmode_t flags, void *holder)
 962 {
 963         struct request_queue *q;
 964         struct block_device *bdev;
 965         struct list_head *head = &fs_devices->devices;
 966         struct btrfs_device *device;
 967         struct btrfs_device *latest_dev = NULL;
 968         struct buffer_head *bh;
 969         struct btrfs_super_block *disk_super;
 970         u64 devid;
 971         int seeding = 1;
 972         int ret = 0;
 973
 974         flags |= FMODE_EXCL;
 975
 976         list_for_each_entry(device, head, dev_list) {
 977                 if (device->bdev)
 978                         continue;
 979                 if (!device->name)
 980                         continue;
 981
 982                 /* Just open everything we can; ignore failures here */
 983                 if (btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1,
 984                                             &bdev, &bh))
 985                         continue;
 986
 987                 disk_super = (struct btrfs_super_block *)bh->b_data;
 988                 devid = btrfs_stack_device_id(&disk_super->dev_item);
 989                 if (devid != device->devid)
 990                         goto error_brelse;
 991
 992                 if (memcmp(device->uuid, disk_super->dev_item.uuid,
 993                            BTRFS_UUID_SIZE))
 994                         goto error_brelse;
 995
 996                 device->generation = btrfs_super_generation(disk_super);
 997                 if (!latest_dev ||
 998                     device->generation > latest_dev->generation)
 999                         latest_dev = device;
1000
1001                 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
1002                         device->writeable = 0;
1003                 } else {
1004                         device->writeable = !bdev_read_only(bdev);
1005                         seeding = 0;
1006                 }
1007
1008                 q = bdev_get_queue(bdev);
1009                 if (blk_queue_discard(q))
1010                         device->can_discard = 1;
1011
1012                 device->bdev = bdev;
1013                 device->in_fs_metadata = 0;
1014                 device->mode = flags;
1015
1016                 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
1017                         fs_devices->rotating = 1;
1018
1019                 fs_devices->open_devices++;
1020                 if (device->writeable &&
1021                     device->devid != BTRFS_DEV_REPLACE_DEVID) {
1022                         fs_devices->rw_devices++;
1023                         list_add(&device->dev_alloc_list,
1024                                  &fs_devices->alloc_list);
1025                 }
1026                 brelse(bh);
1027                 continue;
1028
1029 error_brelse:
1030                 brelse(bh);
1031                 blkdev_put(bdev, flags);
1032                 continue;
1033         }
1034         if (fs_devices->open_devices == 0) {
1035                 ret = -EINVAL;
1036                 goto out;
1037         }
1038         fs_devices->seeding = seeding;
1039         fs_devices->opened = 1;
1040         fs_devices->latest_bdev = latest_dev->bdev;
1041         fs_devices->total_rw_bytes = 0;
1042 out:
1043         return ret;
1044 }
1045
1046 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
1047                        fmode_t flags, void *holder)
1048 {
1049         int ret;
1050
1051         mutex_lock(&uuid_mutex);
1052         if (fs_devices->opened) {
1053                 fs_devices->opened++;
1054                 ret = 0;
1055         } else {
1056                 ret = __btrfs_open_devices(fs_devices, flags, holder);
1057         }
1058         mutex_unlock(&uuid_mutex);
1059         return ret;
1060 }
1061
1062 void btrfs_release_disk_super(struct page *page)
1063 {
1064         kunmap(page);
1065         put_page(page);
1066 }
1067
1068 int btrfs_read_disk_super(struct block_device *bdev, u64 bytenr,
1069                 struct page **page, struct btrfs_super_block **disk_super)
1070 {
1071         void *p;
1072         pgoff_t index;
1073
1074         /* make sure our super fits in the device */
1075         if (bytenr + PAGE_SIZE >= i_size_read(bdev->bd_inode))
1076                 return 1;
1077
1078         /* make sure our super fits in the page */
1079         if (sizeof(**disk_super) > PAGE_SIZE)
1080                 return 1;
1081
1082         /* make sure our super doesn't straddle pages on disk */
1083         index = bytenr >> PAGE_SHIFT;
1084         if ((bytenr + sizeof(**disk_super) - 1) >> PAGE_SHIFT != index)
1085                 return 1;
1086
1087         /* pull in the page with our super */
1088         *page = read_cache_page_gfp(bdev->bd_inode->i_mapping,
1089                                    index, GFP_KERNEL);
1090
1091         if (IS_ERR_OR_NULL(*page))
1092                 return 1;
1093
1094         p = kmap(*page);
1095
1096         /* align our pointer to the offset of the super block */
1097         *disk_super = p + (bytenr & ~PAGE_MASK);
1098
1099         if (btrfs_super_bytenr(*disk_super) != bytenr ||
1100             btrfs_super_magic(*disk_super) != BTRFS_MAGIC) {
1101                 btrfs_release_disk_super(*page);
1102                 return 1;
1103         }
1104
1105         if ((*disk_super)->label[0] &&
1106                 (*disk_super)->label[BTRFS_LABEL_SIZE - 1])
1107                 (*disk_super)->label[BTRFS_LABEL_SIZE - 1] = '\0';
1108
1109         return 0;
1110 }
1111
1112 /*
1113  * Look for a btrfs signature on a device. This may be called out of the mount path
1114  * and we are not allowed to call set_blocksize during the scan. The superblock
1115  * is read via pagecache
1116  */
1117 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
1118                           struct btrfs_fs_devices **fs_devices_ret)
1119 {
1120         struct btrfs_super_block *disk_super;
1121         struct block_device *bdev;
1122         struct page *page;
1123         int ret = -EINVAL;
1124         u64 devid;
1125         u64 transid;
1126         u64 total_devices;
1127         u64 bytenr;
1128
1129         /*
1130          * we would like to check all the supers, but that would make
1131          * a btrfs mount succeed after a mkfs from a different FS.
1132          * So, we need to add a special mount option to scan for
1133          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
1134          */
1135         bytenr = btrfs_sb_offset(0);
1136         flags |= FMODE_EXCL;
1137         mutex_lock(&uuid_mutex);
1138
1139         bdev = blkdev_get_by_path(path, flags, holder);
1140         if (IS_ERR(bdev)) {
1141                 ret = PTR_ERR(bdev);
1142                 goto error;
1143         }
1144
1145         if (btrfs_read_disk_super(bdev, bytenr, &page, &disk_super))
1146                 goto error_bdev_put;
1147
1148         devid = btrfs_stack_device_id(&disk_super->dev_item);
1149         transid = btrfs_super_generation(disk_super);
1150         total_devices = btrfs_super_num_devices(disk_super);
1151
1152         ret = device_list_add(path, disk_super, devid, fs_devices_ret);
1153         if (ret > 0) {
1154                 if (disk_super->label[0]) {
1155                         pr_info("BTRFS: device label %s ", disk_super->label);
1156                 } else {
1157                         pr_info("BTRFS: device fsid %pU ", disk_super->fsid);
1158                 }
1159
1160                 pr_cont("devid %llu transid %llu %s\n", devid, transid, path);
1161                 ret = 0;
1162         }
1163         if (!ret && fs_devices_ret)
1164                 (*fs_devices_ret)->total_devices = total_devices;
1165
1166         btrfs_release_disk_super(page);
1167
1168 error_bdev_put:
1169         blkdev_put(bdev, flags);
1170 error:
1171         mutex_unlock(&uuid_mutex);
1172         return ret;
1173 }
1174
1175 /* helper to account the used device space in the range */
1176 int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
1177                                    u64 end, u64 *length)
1178 {
1179         struct btrfs_key key;
1180         struct btrfs_root *root = device->fs_info->dev_root;
1181         struct btrfs_dev_extent *dev_extent;
1182         struct btrfs_path *path;
1183         u64 extent_end;
1184         int ret;
1185         int slot;
1186         struct extent_buffer *l;
1187
1188         *length = 0;
1189
1190         if (start >= device->total_bytes || device->is_tgtdev_for_dev_replace)
1191                 return 0;
1192
1193         path = btrfs_alloc_path();
1194         if (!path)
1195                 return -ENOMEM;
1196         path->reada = READA_FORWARD;
1197
1198         key.objectid = device->devid;
1199         key.offset = start;
1200         key.type = BTRFS_DEV_EXTENT_KEY;
1201
1202         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1203         if (ret < 0)
1204                 goto out;
1205         if (ret > 0) {
1206                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1207                 if (ret < 0)
1208                         goto out;
1209         }
1210
1211         while (1) {
1212                 l = path->nodes[0];
1213                 slot = path->slots[0];
1214                 if (slot >= btrfs_header_nritems(l)) {
1215                         ret = btrfs_next_leaf(root, path);
1216                         if (ret == 0)
1217                                 continue;
1218                         if (ret < 0)
1219                                 goto out;
1220
1221                         break;
1222                 }
1223                 btrfs_item_key_to_cpu(l, &key, slot);
1224
1225                 if (key.objectid < device->devid)
1226                         goto next;
1227
1228                 if (key.objectid > device->devid)
1229                         break;
1230
1231                 if (key.type != BTRFS_DEV_EXTENT_KEY)
1232                         goto next;
1233
1234                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1235                 extent_end = key.offset + btrfs_dev_extent_length(l,
1236                                                                   dev_extent);
1237                 if (key.offset <= start && extent_end > end) {
1238                         *length = end - start + 1;
1239                         break;
1240                 } else if (key.offset <= start && extent_end > start)
1241                         *length += extent_end - start;
1242                 else if (key.offset > start && extent_end <= end)
1243                         *length += extent_end - key.offset;
1244                 else if (key.offset > start && key.offset <= end) {
1245                         *length += end - key.offset + 1;
1246                         break;
1247                 } else if (key.offset > end)
1248                         break;
1249
1250 next:
1251                 path->slots[0]++;
1252         }
1253         ret = 0;
1254 out:
1255         btrfs_free_path(path);
1256         return ret;
1257 }
1258
1259 static int contains_pending_extent(struct btrfs_transaction *transaction,
1260                                    struct btrfs_device *device,
1261                                    u64 *start, u64 len)
1262 {
1263         struct btrfs_fs_info *fs_info = device->fs_info;
1264         struct extent_map *em;
1265         struct list_head *search_list = &fs_info->pinned_chunks;
1266         int ret = 0;
1267         u64 physical_start = *start;
1268
1269         if (transaction)
1270                 search_list = &transaction->pending_chunks;
1271 again:
1272         list_for_each_entry(em, search_list, list) {
1273                 struct map_lookup *map;
1274                 int i;
1275
1276                 map = em->map_lookup;
1277                 for (i = 0; i < map->num_stripes; i++) {
1278                         u64 end;
1279
1280                         if (map->stripes[i].dev != device)
1281                                 continue;
1282                         if (map->stripes[i].physical >= physical_start + len ||
1283                             map->stripes[i].physical + em->orig_block_len <=
1284                             physical_start)
1285                                 continue;
1286                         /*
1287                          * Make sure that while processing the pinned list we do
1288                          * not override our *start with a lower value, because
1289                          * we can have pinned chunks that fall within this
1290                          * device hole and that have lower physical addresses
1291                          * than the pending chunks we processed before. If we
1292                          * do not take this special care we can end up getting
1293                          * 2 pending chunks that start at the same physical
1294                          * device offsets because the end offset of a pinned
1295                          * chunk can be equal to the start offset of some
1296                          * pending chunk.
1297                          */
1298                         end = map->stripes[i].physical + em->orig_block_len;
1299                         if (end > *start) {
1300                                 *start = end;
1301                                 ret = 1;
1302                         }
1303                 }
1304         }
1305         if (search_list != &fs_info->pinned_chunks) {
1306                 search_list = &fs_info->pinned_chunks;
1307                 goto again;
1308         }
1309
1310         return ret;
1311 }
1312
1313
1314 /*
1315  * find_free_dev_extent_start - find free space in the specified device
1316  * @device:       the device which we search the free space in
1317  * @num_bytes:    the size of the free space that we need
1318  * @search_start: the position from which to begin the search
1319  * @start:        store the start of the free space.
1320  * @len:          the size of the free space. that we find, or the size
1321  *                of the max free space if we don't find suitable free space
1322  *
1323  * this uses a pretty simple search, the expectation is that it is
1324  * called very infrequently and that a given device has a small number
1325  * of extents
1326  *
1327  * @start is used to store the start of the free space if we find. But if we
1328  * don't find suitable free space, it will be used to store the start position
1329  * of the max free space.
1330  *
1331  * @len is used to store the size of the free space that we find.
1332  * But if we don't find suitable free space, it is used to store the size of
1333  * the max free space.
1334  */
1335 int find_free_dev_extent_start(struct btrfs_transaction *transaction,
1336                                struct btrfs_device *device, u64 num_bytes,
1337                                u64 search_start, u64 *start, u64 *len)
1338 {
1339         struct btrfs_fs_info *fs_info = device->fs_info;
1340         struct btrfs_root *root = fs_info->dev_root;
1341         struct btrfs_key key;
1342         struct btrfs_dev_extent *dev_extent;
1343         struct btrfs_path *path;
1344         u64 hole_size;
1345         u64 max_hole_start;
1346         u64 max_hole_size;
1347         u64 extent_end;
1348         u64 search_end = device->total_bytes;
1349         int ret;
1350         int slot;
1351         struct extent_buffer *l;
1352         u64 min_search_start;
1353
1354         /*
1355          * We don't want to overwrite the superblock on the drive nor any area
1356          * used by the boot loader (grub for example), so we make sure to start
1357          * at an offset of at least 1MB.
1358          */
1359         min_search_start = max(fs_info->alloc_start, 1024ull * 1024);
1360         search_start = max(search_start, min_search_start);
1361
1362         path = btrfs_alloc_path();
1363         if (!path)
1364                 return -ENOMEM;
1365
1366         max_hole_start = search_start;
1367         max_hole_size = 0;
1368
1369 again:
1370         if (search_start >= search_end || device->is_tgtdev_for_dev_replace) {
1371                 ret = -ENOSPC;
1372                 goto out;
1373         }
1374
1375         path->reada = READA_FORWARD;
1376         path->search_commit_root = 1;
1377         path->skip_locking = 1;
1378
1379         key.objectid = device->devid;
1380         key.offset = search_start;
1381         key.type = BTRFS_DEV_EXTENT_KEY;
1382
1383         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1384         if (ret < 0)
1385                 goto out;
1386         if (ret > 0) {
1387                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1388                 if (ret < 0)
1389                         goto out;
1390         }
1391
1392         while (1) {
1393                 l = path->nodes[0];
1394                 slot = path->slots[0];
1395                 if (slot >= btrfs_header_nritems(l)) {
1396                         ret = btrfs_next_leaf(root, path);
1397                         if (ret == 0)
1398                                 continue;
1399                         if (ret < 0)
1400                                 goto out;
1401
1402                         break;
1403                 }
1404                 btrfs_item_key_to_cpu(l, &key, slot);
1405
1406                 if (key.objectid < device->devid)
1407                         goto next;
1408
1409                 if (key.objectid > device->devid)
1410                         break;
1411
1412                 if (key.type != BTRFS_DEV_EXTENT_KEY)
1413                         goto next;
1414
1415                 if (key.offset > search_start) {
1416                         hole_size = key.offset - search_start;
1417
1418                         /*
1419                          * Have to check before we set max_hole_start, otherwise
1420                          * we could end up sending back this offset anyway.
1421                          */
1422                         if (contains_pending_extent(transaction, device,
1423                                                     &search_start,
1424                                                     hole_size)) {
1425                                 if (key.offset >= search_start) {
1426                                         hole_size = key.offset - search_start;
1427                                 } else {
1428                                         WARN_ON_ONCE(1);
1429                                         hole_size = 0;
1430                                 }
1431                         }
1432
1433                         if (hole_size > max_hole_size) {
1434                                 max_hole_start = search_start;
1435                                 max_hole_size = hole_size;
1436                         }
1437
1438                         /*
1439                          * If this free space is greater than which we need,
1440                          * it must be the max free space that we have found
1441                          * until now, so max_hole_start must point to the start
1442                          * of this free space and the length of this free space
1443                          * is stored in max_hole_size. Thus, we return
1444                          * max_hole_start and max_hole_size and go back to the
1445                          * caller.
1446                          */
1447                         if (hole_size >= num_bytes) {
1448                                 ret = 0;
1449                                 goto out;
1450                         }
1451                 }
1452
1453                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1454                 extent_end = key.offset + btrfs_dev_extent_length(l,
1455                                                                   dev_extent);
1456                 if (extent_end > search_start)
1457                         search_start = extent_end;
1458 next:
1459                 path->slots[0]++;
1460                 cond_resched();
1461         }
1462
1463         /*
1464          * At this point, search_start should be the end of
1465          * allocated dev extents, and when shrinking the device,
1466          * search_end may be smaller than search_start.
1467          */
1468         if (search_end > search_start) {
1469                 hole_size = search_end - search_start;
1470
1471                 if (contains_pending_extent(transaction, device, &search_start,
1472                                             hole_size)) {
1473                         btrfs_release_path(path);
1474                         goto again;
1475                 }
1476
1477                 if (hole_size > max_hole_size) {
1478                         max_hole_start = search_start;
1479                         max_hole_size = hole_size;
1480                 }
1481         }
1482
1483         /* See above. */
1484         if (max_hole_size < num_bytes)
1485                 ret = -ENOSPC;
1486         else
1487                 ret = 0;
1488
1489 out:
1490         btrfs_free_path(path);
1491         *start = max_hole_start;
1492         if (len)
1493                 *len = max_hole_size;
1494         return ret;
1495 }
1496
1497 int find_free_dev_extent(struct btrfs_trans_handle *trans,
1498                          struct btrfs_device *device, u64 num_bytes,
1499                          u64 *start, u64 *len)
1500 {
1501         /* FIXME use last free of some kind */
1502         return find_free_dev_extent_start(trans->transaction, device,
1503                                           num_bytes, 0, start, len);
1504 }
1505
1506 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1507                           struct btrfs_device *device,
1508                           u64 start, u64 *dev_extent_len)
1509 {
1510         struct btrfs_fs_info *fs_info = device->fs_info;
1511         struct btrfs_root *root = fs_info->dev_root;
1512         int ret;
1513         struct btrfs_path *path;
1514         struct btrfs_key key;
1515         struct btrfs_key found_key;
1516         struct extent_buffer *leaf = NULL;
1517         struct btrfs_dev_extent *extent = NULL;
1518
1519         path = btrfs_alloc_path();
1520         if (!path)
1521                 return -ENOMEM;
1522
1523         key.objectid = device->devid;
1524         key.offset = start;
1525         key.type = BTRFS_DEV_EXTENT_KEY;
1526 again:
1527         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1528         if (ret > 0) {
1529                 ret = btrfs_previous_item(root, path, key.objectid,
1530                                           BTRFS_DEV_EXTENT_KEY);
1531                 if (ret)
1532                         goto out;
1533                 leaf = path->nodes[0];
1534                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1535                 extent = btrfs_item_ptr(leaf, path->slots[0],
1536                                         struct btrfs_dev_extent);
1537                 BUG_ON(found_key.offset > start || found_key.offset +
1538                        btrfs_dev_extent_length(leaf, extent) < start);
1539                 key = found_key;
1540                 btrfs_release_path(path);
1541                 goto again;
1542         } else if (ret == 0) {
1543                 leaf = path->nodes[0];
1544                 extent = btrfs_item_ptr(leaf, path->slots[0],
1545                                         struct btrfs_dev_extent);
1546         } else {
1547                 btrfs_handle_fs_error(fs_info, ret, "Slot search failed");
1548                 goto out;
1549         }
1550
1551         *dev_extent_len = btrfs_dev_extent_length(leaf, extent);
1552
1553         ret = btrfs_del_item(trans, root, path);
1554         if (ret) {
1555                 btrfs_handle_fs_error(fs_info, ret,
1556                                       "Failed to remove dev extent item");
1557         } else {
1558                 set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags);
1559         }
1560 out:
1561         btrfs_free_path(path);
1562         return ret;
1563 }
1564
1565 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1566                                   struct btrfs_device *device,
1567                                   u64 chunk_tree, u64 chunk_objectid,
1568                                   u64 chunk_offset, u64 start, u64 num_bytes)
1569 {
1570         int ret;
1571         struct btrfs_path *path;
1572         struct btrfs_fs_info *fs_info = device->fs_info;
1573         struct btrfs_root *root = fs_info->dev_root;
1574         struct btrfs_dev_extent *extent;
1575         struct extent_buffer *leaf;
1576         struct btrfs_key key;
1577
1578         WARN_ON(!device->in_fs_metadata);
1579         WARN_ON(device->is_tgtdev_for_dev_replace);
1580         path = btrfs_alloc_path();
1581         if (!path)
1582                 return -ENOMEM;
1583
1584         key.objectid = device->devid;
1585         key.offset = start;
1586         key.type = BTRFS_DEV_EXTENT_KEY;
1587         ret = btrfs_insert_empty_item(trans, root, path, &key,
1588                                       sizeof(*extent));
1589         if (ret)
1590                 goto out;
1591
1592         leaf = path->nodes[0];
1593         extent = btrfs_item_ptr(leaf, path->slots[0],
1594                                 struct btrfs_dev_extent);
1595         btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
1596         btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
1597         btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1598
1599         write_extent_buffer_chunk_tree_uuid(leaf, fs_info->chunk_tree_uuid);
1600
1601         btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1602         btrfs_mark_buffer_dirty(leaf);
1603 out:
1604         btrfs_free_path(path);
1605         return ret;
1606 }
1607
1608 static u64 find_next_chunk(struct btrfs_fs_info *fs_info)
1609 {
1610         struct extent_map_tree *em_tree;
1611         struct extent_map *em;
1612         struct rb_node *n;
1613         u64 ret = 0;
1614
1615         em_tree = &fs_info->mapping_tree.map_tree;
1616         read_lock(&em_tree->lock);
1617         n = rb_last(&em_tree->map);
1618         if (n) {
1619                 em = rb_entry(n, struct extent_map, rb_node);
1620                 ret = em->start + em->len;
1621         }
1622         read_unlock(&em_tree->lock);
1623
1624         return ret;
1625 }
1626
1627 static noinline int find_next_devid(struct btrfs_fs_info *fs_info,
1628                                     u64 *devid_ret)
1629 {
1630         int ret;
1631         struct btrfs_key key;
1632         struct btrfs_key found_key;
1633         struct btrfs_path *path;
1634
1635         path = btrfs_alloc_path();
1636         if (!path)
1637                 return -ENOMEM;
1638
1639         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1640         key.type = BTRFS_DEV_ITEM_KEY;
1641         key.offset = (u64)-1;
1642
1643         ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
1644         if (ret < 0)
1645                 goto error;
1646
1647         BUG_ON(ret == 0); /* Corruption */
1648
1649         ret = btrfs_previous_item(fs_info->chunk_root, path,
1650                                   BTRFS_DEV_ITEMS_OBJECTID,
1651                                   BTRFS_DEV_ITEM_KEY);
1652         if (ret) {
1653                 *devid_ret = 1;
1654         } else {
1655                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1656                                       path->slots[0]);
1657                 *devid_ret = found_key.offset + 1;
1658         }
1659         ret = 0;
1660 error:
1661         btrfs_free_path(path);
1662         return ret;
1663 }
1664
1665 /*
1666  * the device information is stored in the chunk root
1667  * the btrfs_device struct should be fully filled in
1668  */
1669 static int btrfs_add_device(struct btrfs_trans_handle *trans,
1670                             struct btrfs_fs_info *fs_info,
1671                             struct btrfs_device *device)
1672 {
1673         struct btrfs_root *root = fs_info->chunk_root;
1674         int ret;
1675         struct btrfs_path *path;
1676         struct btrfs_dev_item *dev_item;
1677         struct extent_buffer *leaf;
1678         struct btrfs_key key;
1679         unsigned long ptr;
1680
1681         path = btrfs_alloc_path();
1682         if (!path)
1683                 return -ENOMEM;
1684
1685         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1686         key.type = BTRFS_DEV_ITEM_KEY;
1687         key.offset = device->devid;
1688
1689         ret = btrfs_insert_empty_item(trans, root, path, &key,
1690                                       sizeof(*dev_item));
1691         if (ret)
1692                 goto out;
1693
1694         leaf = path->nodes[0];
1695         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1696
1697         btrfs_set_device_id(leaf, dev_item, device->devid);
1698         btrfs_set_device_generation(leaf, dev_item, 0);
1699         btrfs_set_device_type(leaf, dev_item, device->type);
1700         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1701         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1702         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1703         btrfs_set_device_total_bytes(leaf, dev_item,
1704                                      btrfs_device_get_disk_total_bytes(device));
1705         btrfs_set_device_bytes_used(leaf, dev_item,
1706                                     btrfs_device_get_bytes_used(device));
1707         btrfs_set_device_group(leaf, dev_item, 0);
1708         btrfs_set_device_seek_speed(leaf, dev_item, 0);
1709         btrfs_set_device_bandwidth(leaf, dev_item, 0);
1710         btrfs_set_device_start_offset(leaf, dev_item, 0);
1711
1712         ptr = btrfs_device_uuid(dev_item);
1713         write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1714         ptr = btrfs_device_fsid(dev_item);
1715         write_extent_buffer(leaf, fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1716         btrfs_mark_buffer_dirty(leaf);
1717
1718         ret = 0;
1719 out:
1720         btrfs_free_path(path);
1721         return ret;
1722 }
1723
1724 /*
1725  * Function to update ctime/mtime for a given device path.
1726  * Mainly used for ctime/mtime based probe like libblkid.
1727  */
1728 static void update_dev_time(const char *path_name)
1729 {
1730         struct file *filp;
1731
1732         filp = filp_open(path_name, O_RDWR, 0);
1733         if (IS_ERR(filp))
1734                 return;
1735         file_update_time(filp);
1736         filp_close(filp, NULL);
1737 }
1738
1739 static int btrfs_rm_dev_item(struct btrfs_fs_info *fs_info,
1740                              struct btrfs_device *device)
1741 {
1742         struct btrfs_root *root = fs_info->chunk_root;
1743         int ret;
1744         struct btrfs_path *path;
1745         struct btrfs_key key;
1746         struct btrfs_trans_handle *trans;
1747
1748         path = btrfs_alloc_path();
1749         if (!path)
1750                 return -ENOMEM;
1751
1752         trans = btrfs_start_transaction(root, 0);
1753         if (IS_ERR(trans)) {
1754                 btrfs_free_path(path);
1755                 return PTR_ERR(trans);
1756         }
1757         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1758         key.type = BTRFS_DEV_ITEM_KEY;
1759         key.offset = device->devid;
1760
1761         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1762         if (ret < 0)
1763                 goto out;
1764
1765         if (ret > 0) {
1766                 ret = -ENOENT;
1767                 goto out;
1768         }
1769
1770         ret = btrfs_del_item(trans, root, path);
1771         if (ret)
1772                 goto out;
1773 out:
1774         btrfs_free_path(path);
1775         btrfs_commit_transaction(trans);
1776         return ret;
1777 }
1778
1779 /*
1780  * Verify that @num_devices satisfies the RAID profile constraints in the whole
1781  * filesystem. It's up to the caller to adjust that number regarding eg. device
1782  * replace.
1783  */
1784 static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info,
1785                 u64 num_devices)
1786 {
1787         u64 all_avail;
1788         unsigned seq;
1789         int i;
1790
1791         do {
1792                 seq = read_seqbegin(&fs_info->profiles_lock);
1793
1794                 all_avail = fs_info->avail_data_alloc_bits |
1795                             fs_info->avail_system_alloc_bits |
1796                             fs_info->avail_metadata_alloc_bits;
1797         } while (read_seqretry(&fs_info->profiles_lock, seq));
1798
1799         for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
1800                 if (!(all_avail & btrfs_raid_group[i]))
1801                         continue;
1802
1803                 if (num_devices < btrfs_raid_array[i].devs_min) {
1804                         int ret = btrfs_raid_mindev_error[i];
1805
1806                         if (ret)
1807                                 return ret;
1808                 }
1809         }
1810
1811         return 0;
1812 }
1813
1814 struct btrfs_device *btrfs_find_next_active_device(struct btrfs_fs_devices *fs_devs,
1815                                         struct btrfs_device *device)
1816 {
1817         struct btrfs_device *next_device;
1818
1819         list_for_each_entry(next_device, &fs_devs->devices, dev_list) {
1820                 if (next_device != device &&
1821                         !next_device->missing && next_device->bdev)
1822                         return next_device;
1823         }
1824
1825         return NULL;
1826 }
1827
1828 /*
1829  * Helper function to check if the given device is part of s_bdev / latest_bdev
1830  * and replace it with the provided or the next active device, in the context
1831  * where this function called, there should be always be another device (or
1832  * this_dev) which is active.
1833  */
1834 void btrfs_assign_next_active_device(struct btrfs_fs_info *fs_info,
1835                 struct btrfs_device *device, struct btrfs_device *this_dev)
1836 {
1837         struct btrfs_device *next_device;
1838
1839         if (this_dev)
1840                 next_device = this_dev;
1841         else
1842                 next_device = btrfs_find_next_active_device(fs_info->fs_devices,
1843                                                                 device);
1844         ASSERT(next_device);
1845
1846         if (fs_info->sb->s_bdev &&
1847                         (fs_info->sb->s_bdev == device->bdev))
1848                 fs_info->sb->s_bdev = next_device->bdev;
1849
1850         if (fs_info->fs_devices->latest_bdev == device->bdev)
1851                 fs_info->fs_devices->latest_bdev = next_device->bdev;
1852 }
1853
1854 int btrfs_rm_device(struct btrfs_fs_info *fs_info, const char *device_path,
1855                 u64 devid)
1856 {
1857         struct btrfs_device *device;
1858         struct btrfs_fs_devices *cur_devices;
1859         u64 num_devices;
1860         int ret = 0;
1861         bool clear_super = false;
1862
1863         mutex_lock(&uuid_mutex);
1864
1865         num_devices = fs_info->fs_devices->num_devices;
1866         btrfs_dev_replace_lock(&fs_info->dev_replace, 0);
1867         if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
1868                 WARN_ON(num_devices < 1);
1869                 num_devices--;
1870         }
1871         btrfs_dev_replace_unlock(&fs_info->dev_replace, 0);
1872
1873         ret = btrfs_check_raid_min_devices(fs_info, num_devices - 1);
1874         if (ret)
1875                 goto out;
1876
1877         ret = btrfs_find_device_by_devspec(fs_info, devid, device_path,
1878                                            &device);
1879         if (ret)
1880                 goto out;
1881
1882         if (device->is_tgtdev_for_dev_replace) {
1883                 ret = BTRFS_ERROR_DEV_TGT_REPLACE;
1884                 goto out;
1885         }
1886
1887         if (device->writeable && fs_info->fs_devices->rw_devices == 1) {
1888                 ret = BTRFS_ERROR_DEV_ONLY_WRITABLE;
1889                 goto out;
1890         }
1891
1892         if (device->writeable) {
1893                 mutex_lock(&fs_info->chunk_mutex);
1894                 list_del_init(&device->dev_alloc_list);
1895                 device->fs_devices->rw_devices--;
1896                 mutex_unlock(&fs_info->chunk_mutex);
1897                 clear_super = true;
1898         }
1899
1900         mutex_unlock(&uuid_mutex);
1901         ret = btrfs_shrink_device(device, 0);
1902         mutex_lock(&uuid_mutex);
1903         if (ret)
1904                 goto error_undo;
1905
1906         /*
1907          * TODO: the superblock still includes this device in its num_devices
1908          * counter although write_all_supers() is not locked out. This
1909          * could give a filesystem state which requires a degraded mount.
1910          */
1911         ret = btrfs_rm_dev_item(fs_info, device);
1912         if (ret)
1913                 goto error_undo;
1914
1915         device->in_fs_metadata = 0;
1916         btrfs_scrub_cancel_dev(fs_info, device);
1917
1918         /*
1919          * the device list mutex makes sure that we don't change
1920          * the device list while someone else is writing out all
1921          * the device supers. Whoever is writing all supers, should
1922          * lock the device list mutex before getting the number of
1923          * devices in the super block (super_copy). Conversely,
1924          * whoever updates the number of devices in the super block
1925          * (super_copy) should hold the device list mutex.
1926          */
1927
1928         cur_devices = device->fs_devices;
1929         mutex_lock(&fs_info->fs_devices->device_list_mutex);
1930         list_del_rcu(&device->dev_list);
1931
1932         device->fs_devices->num_devices--;
1933         device->fs_devices->total_devices--;
1934
1935         if (device->missing)
1936                 device->fs_devices->missing_devices--;
1937
1938         btrfs_assign_next_active_device(fs_info, device, NULL);
1939
1940         if (device->bdev) {
1941                 device->fs_devices->open_devices--;
1942                 /* remove sysfs entry */
1943                 btrfs_sysfs_rm_device_link(fs_info->fs_devices, device);
1944         }
1945
1946         num_devices = btrfs_super_num_devices(fs_info->super_copy) - 1;
1947         btrfs_set_super_num_devices(fs_info->super_copy, num_devices);
1948         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
1949
1950         /*
1951          * at this point, the device is zero sized and detached from
1952          * the devices list.  All that's left is to zero out the old
1953          * supers and free the device.
1954          */
1955         if (device->writeable)
1956                 btrfs_scratch_superblocks(device->bdev, device->name->str);
1957
1958         btrfs_close_bdev(device);
1959         call_rcu(&device->rcu, free_device);
1960
1961         if (cur_devices->open_devices == 0) {
1962                 struct btrfs_fs_devices *fs_devices;
1963                 fs_devices = fs_info->fs_devices;
1964                 while (fs_devices) {
1965                         if (fs_devices->seed == cur_devices) {
1966                                 fs_devices->seed = cur_devices->seed;
1967                                 break;
1968                         }
1969                         fs_devices = fs_devices->seed;
1970                 }
1971                 cur_devices->seed = NULL;
1972                 __btrfs_close_devices(cur_devices);
1973                 free_fs_devices(cur_devices);
1974         }
1975
1976         fs_info->num_tolerated_disk_barrier_failures =
1977                 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
1978
1979 out:
1980         mutex_unlock(&uuid_mutex);
1981         return ret;
1982
1983 error_undo:
1984         if (device->writeable) {
1985                 mutex_lock(&fs_info->chunk_mutex);
1986                 list_add(&device->dev_alloc_list,
1987                          &fs_info->fs_devices->alloc_list);
1988                 device->fs_devices->rw_devices++;
1989                 mutex_unlock(&fs_info->chunk_mutex);
1990         }
1991         goto out;
1992 }
1993
1994 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_fs_info *fs_info,
1995                                         struct btrfs_device *srcdev)
1996 {
1997         struct btrfs_fs_devices *fs_devices;
1998
1999         WARN_ON(!mutex_is_locked(&fs_info->fs_devices->device_list_mutex));
2000
2001         /*
2002          * in case of fs with no seed, srcdev->fs_devices will point
2003          * to fs_devices of fs_info. However when the dev being replaced is
2004          * a seed dev it will point to the seed's local fs_devices. In short
2005          * srcdev will have its correct fs_devices in both the cases.
2006          */
2007         fs_devices = srcdev->fs_devices;
2008
2009         list_del_rcu(&srcdev->dev_list);
2010         list_del_rcu(&srcdev->dev_alloc_list);
2011         fs_devices->num_devices--;
2012         if (srcdev->missing)
2013                 fs_devices->missing_devices--;
2014
2015         if (srcdev->writeable)
2016                 fs_devices->rw_devices--;
2017
2018         if (srcdev->bdev)
2019                 fs_devices->open_devices--;
2020 }
2021
2022 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_fs_info *fs_info,
2023                                       struct btrfs_device *srcdev)
2024 {
2025         struct btrfs_fs_devices *fs_devices = srcdev->fs_devices;
2026
2027         if (srcdev->writeable) {
2028                 /* zero out the old super if it is writable */
2029                 btrfs_scratch_superblocks(srcdev->bdev, srcdev->name->str);
2030         }
2031
2032         btrfs_close_bdev(srcdev);
2033
2034         call_rcu(&srcdev->rcu, free_device);
2035
2036         /*
2037          * unless fs_devices is seed fs, num_devices shouldn't go
2038          * zero
2039          */
2040         BUG_ON(!fs_devices->num_devices && !fs_devices->seeding);
2041
2042         /* if this is no devs we rather delete the fs_devices */
2043         if (!fs_devices->num_devices) {
2044                 struct btrfs_fs_devices *tmp_fs_devices;
2045
2046                 tmp_fs_devices = fs_info->fs_devices;
2047                 while (tmp_fs_devices) {
2048                         if (tmp_fs_devices->seed == fs_devices) {
2049                                 tmp_fs_devices->seed = fs_devices->seed;
2050                                 break;
2051                         }
2052                         tmp_fs_devices = tmp_fs_devices->seed;
2053                 }
2054                 fs_devices->seed = NULL;
2055                 __btrfs_close_devices(fs_devices);
2056                 free_fs_devices(fs_devices);
2057         }
2058 }
2059
2060 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
2061                                       struct btrfs_device *tgtdev)
2062 {
2063         mutex_lock(&uuid_mutex);
2064         WARN_ON(!tgtdev);
2065         mutex_lock(&fs_info->fs_devices->device_list_mutex);
2066
2067         btrfs_sysfs_rm_device_link(fs_info->fs_devices, tgtdev);
2068
2069         if (tgtdev->bdev)
2070                 fs_info->fs_devices->open_devices--;
2071
2072         fs_info->fs_devices->num_devices--;
2073
2074         btrfs_assign_next_active_device(fs_info, tgtdev, NULL);
2075
2076         list_del_rcu(&tgtdev->dev_list);
2077
2078         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2079         mutex_unlock(&uuid_mutex);
2080
2081         /*
2082          * The update_dev_time() with in btrfs_scratch_superblocks()
2083          * may lead to a call to btrfs_show_devname() which will try
2084          * to hold device_list_mutex. And here this device
2085          * is already out of device list, so we don't have to hold
2086          * the device_list_mutex lock.
2087          */
2088         btrfs_scratch_superblocks(tgtdev->bdev, tgtdev->name->str);
2089
2090         btrfs_close_bdev(tgtdev);
2091         call_rcu(&tgtdev->rcu, free_device);
2092 }
2093
2094 static int btrfs_find_device_by_path(struct btrfs_fs_info *fs_info,
2095                                      const char *device_path,
2096                                      struct btrfs_device **device)
2097 {
2098         int ret = 0;
2099         struct btrfs_super_block *disk_super;
2100         u64 devid;
2101         u8 *dev_uuid;
2102         struct block_device *bdev;
2103         struct buffer_head *bh;
2104
2105         *device = NULL;
2106         ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
2107                                     fs_info->bdev_holder, 0, &bdev, &bh);
2108         if (ret)
2109                 return ret;
2110         disk_super = (struct btrfs_super_block *)bh->b_data;
2111         devid = btrfs_stack_device_id(&disk_super->dev_item);
2112         dev_uuid = disk_super->dev_item.uuid;
2113         *device = btrfs_find_device(fs_info, devid, dev_uuid, disk_super->fsid);
2114         brelse(bh);
2115         if (!*device)
2116                 ret = -ENOENT;
2117         blkdev_put(bdev, FMODE_READ);
2118         return ret;
2119 }
2120
2121 int btrfs_find_device_missing_or_by_path(struct btrfs_fs_info *fs_info,
2122                                          const char *device_path,
2123                                          struct btrfs_device **device)
2124 {
2125         *device = NULL;
2126         if (strcmp(device_path, "missing") == 0) {
2127                 struct list_head *devices;
2128                 struct btrfs_device *tmp;
2129
2130                 devices = &fs_info->fs_devices->devices;
2131                 /*
2132                  * It is safe to read the devices since the volume_mutex
2133                  * is held by the caller.
2134                  */
2135                 list_for_each_entry(tmp, devices, dev_list) {
2136                         if (tmp->in_fs_metadata && !tmp->bdev) {
2137                                 *device = tmp;
2138                                 break;
2139                         }
2140                 }
2141
2142                 if (!*device)
2143                         return BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
2144
2145                 return 0;
2146         } else {
2147                 return btrfs_find_device_by_path(fs_info, device_path, device);
2148         }
2149 }
2150
2151 /*
2152  * Lookup a device given by device id, or the path if the id is 0.
2153  */
2154 int btrfs_find_device_by_devspec(struct btrfs_fs_info *fs_info, u64 devid,
2155                                  const char *devpath,
2156                                  struct btrfs_device **device)
2157 {
2158         int ret;
2159
2160         if (devid) {
2161                 ret = 0;
2162                 *device = btrfs_find_device(fs_info, devid, NULL, NULL);
2163                 if (!*device)
2164                         ret = -ENOENT;
2165         } else {
2166                 if (!devpath || !devpath[0])
2167                         return -EINVAL;
2168
2169                 ret = btrfs_find_device_missing_or_by_path(fs_info, devpath,
2170                                                            device);
2171         }
2172         return ret;
2173 }
2174
2175 /*
2176  * does all the dirty work required for changing file system's UUID.
2177  */
2178 static int btrfs_prepare_sprout(struct btrfs_fs_info *fs_info)
2179 {
2180         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2181         struct btrfs_fs_devices *old_devices;
2182         struct btrfs_fs_devices *seed_devices;
2183         struct btrfs_super_block *disk_super = fs_info->super_copy;
2184         struct btrfs_device *device;
2185         u64 super_flags;
2186
2187         BUG_ON(!mutex_is_locked(&uuid_mutex));
2188         if (!fs_devices->seeding)
2189                 return -EINVAL;
2190
2191         seed_devices = __alloc_fs_devices();
2192         if (IS_ERR(seed_devices))
2193                 return PTR_ERR(seed_devices);
2194
2195         old_devices = clone_fs_devices(fs_devices);
2196         if (IS_ERR(old_devices)) {
2197                 kfree(seed_devices);
2198                 return PTR_ERR(old_devices);
2199         }
2200
2201         list_add(&old_devices->list, &fs_uuids);
2202
2203         memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
2204         seed_devices->opened = 1;
2205         INIT_LIST_HEAD(&seed_devices->devices);
2206         INIT_LIST_HEAD(&seed_devices->alloc_list);
2207         mutex_init(&seed_devices->device_list_mutex);
2208
2209         mutex_lock(&fs_info->fs_devices->device_list_mutex);
2210         list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
2211                               synchronize_rcu);
2212         list_for_each_entry(device, &seed_devices->devices, dev_list)
2213                 device->fs_devices = seed_devices;
2214
2215         mutex_lock(&fs_info->chunk_mutex);
2216         list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
2217         mutex_unlock(&fs_info->chunk_mutex);
2218
2219         fs_devices->seeding = 0;
2220         fs_devices->num_devices = 0;
2221         fs_devices->open_devices = 0;
2222         fs_devices->missing_devices = 0;
2223         fs_devices->rotating = 0;
2224         fs_devices->seed = seed_devices;
2225
2226         generate_random_uuid(fs_devices->fsid);
2227         memcpy(fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2228         memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2229         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2230
2231         super_flags = btrfs_super_flags(disk_super) &
2232                       ~BTRFS_SUPER_FLAG_SEEDING;
2233         btrfs_set_super_flags(disk_super, super_flags);
2234
2235         return 0;
2236 }
2237
2238 /*
2239  * Store the expected generation for seed devices in device items.
2240  */
2241 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
2242                                struct btrfs_fs_info *fs_info)
2243 {
2244         struct btrfs_root *root = fs_info->chunk_root;
2245         struct btrfs_path *path;
2246         struct extent_buffer *leaf;
2247         struct btrfs_dev_item *dev_item;
2248         struct btrfs_device *device;
2249         struct btrfs_key key;
2250         u8 fs_uuid[BTRFS_UUID_SIZE];
2251         u8 dev_uuid[BTRFS_UUID_SIZE];
2252         u64 devid;
2253         int ret;
2254
2255         path = btrfs_alloc_path();
2256         if (!path)
2257                 return -ENOMEM;
2258
2259         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2260         key.offset = 0;
2261         key.type = BTRFS_DEV_ITEM_KEY;
2262
2263         while (1) {
2264                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2265                 if (ret < 0)
2266                         goto error;
2267
2268                 leaf = path->nodes[0];
2269 next_slot:
2270                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2271                         ret = btrfs_next_leaf(root, path);
2272                         if (ret > 0)
2273                                 break;
2274                         if (ret < 0)
2275                                 goto error;
2276                         leaf = path->nodes[0];
2277                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2278                         btrfs_release_path(path);
2279                         continue;
2280                 }
2281
2282                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2283                 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
2284                     key.type != BTRFS_DEV_ITEM_KEY)
2285                         break;
2286
2287                 dev_item = btrfs_item_ptr(leaf, path->slots[0],
2288                                           struct btrfs_dev_item);
2289                 devid = btrfs_device_id(leaf, dev_item);
2290                 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
2291                                    BTRFS_UUID_SIZE);
2292                 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
2293                                    BTRFS_UUID_SIZE);
2294                 device = btrfs_find_device(fs_info, devid, dev_uuid, fs_uuid);
2295                 BUG_ON(!device); /* Logic error */
2296
2297                 if (device->fs_devices->seeding) {
2298                         btrfs_set_device_generation(leaf, dev_item,
2299                                                     device->generation);
2300                         btrfs_mark_buffer_dirty(leaf);
2301                 }
2302
2303                 path->slots[0]++;
2304                 goto next_slot;
2305         }
2306         ret = 0;
2307 error:
2308         btrfs_free_path(path);
2309         return ret;
2310 }
2311
2312 int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path)
2313 {
2314         struct btrfs_root *root = fs_info->dev_root;
2315         struct request_queue *q;
2316         struct btrfs_trans_handle *trans;
2317         struct btrfs_device *device;
2318         struct block_device *bdev;
2319         struct list_head *devices;
2320         struct super_block *sb = fs_info->sb;
2321         struct rcu_string *name;
2322         u64 tmp;
2323         int seeding_dev = 0;
2324         int ret = 0;
2325
2326         if ((sb->s_flags & MS_RDONLY) && !fs_info->fs_devices->seeding)
2327                 return -EROFS;
2328
2329         bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2330                                   fs_info->bdev_holder);
2331         if (IS_ERR(bdev))
2332                 return PTR_ERR(bdev);
2333
2334         if (fs_info->fs_devices->seeding) {
2335                 seeding_dev = 1;
2336                 down_write(&sb->s_umount);
2337                 mutex_lock(&uuid_mutex);
2338         }
2339
2340         filemap_write_and_wait(bdev->bd_inode->i_mapping);
2341
2342         devices = &fs_info->fs_devices->devices;
2343
2344         mutex_lock(&fs_info->fs_devices->device_list_mutex);
2345         list_for_each_entry(device, devices, dev_list) {
2346                 if (device->bdev == bdev) {
2347                         ret = -EEXIST;
2348                         mutex_unlock(
2349                                 &fs_info->fs_devices->device_list_mutex);
2350                         goto error;
2351                 }
2352         }
2353         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2354
2355         device = btrfs_alloc_device(fs_info, NULL, NULL);
2356         if (IS_ERR(device)) {
2357                 /* we can safely leave the fs_devices entry around */
2358                 ret = PTR_ERR(device);
2359                 goto error;
2360         }
2361
2362         name = rcu_string_strdup(device_path, GFP_KERNEL);
2363         if (!name) {
2364                 kfree(device);
2365                 ret = -ENOMEM;
2366                 goto error;
2367         }
2368         rcu_assign_pointer(device->name, name);
2369
2370         trans = btrfs_start_transaction(root, 0);
2371         if (IS_ERR(trans)) {
2372                 rcu_string_free(device->name);
2373                 kfree(device);
2374                 ret = PTR_ERR(trans);
2375                 goto error;
2376         }
2377
2378         q = bdev_get_queue(bdev);
2379         if (blk_queue_discard(q))
2380                 device->can_discard = 1;
2381         device->writeable = 1;
2382         device->generation = trans->transid;
2383         device->io_width = fs_info->sectorsize;
2384         device->io_align = fs_info->sectorsize;
2385         device->sector_size = fs_info->sectorsize;
2386         device->total_bytes = i_size_read(bdev->bd_inode);
2387         device->disk_total_bytes = device->total_bytes;
2388         device->commit_total_bytes = device->total_bytes;
2389         device->fs_info = fs_info;
2390         device->bdev = bdev;
2391         device->in_fs_metadata = 1;
2392         device->is_tgtdev_for_dev_replace = 0;
2393         device->mode = FMODE_EXCL;
2394         device->dev_stats_valid = 1;
2395         set_blocksize(device->bdev, 4096);
2396
2397         if (seeding_dev) {
2398                 sb->s_flags &= ~MS_RDONLY;
2399                 ret = btrfs_prepare_sprout(fs_info);
2400                 BUG_ON(ret); /* -ENOMEM */
2401         }
2402
2403         device->fs_devices = fs_info->fs_devices;
2404
2405         mutex_lock(&fs_info->fs_devices->device_list_mutex);
2406         mutex_lock(&fs_info->chunk_mutex);
2407         list_add_rcu(&device->dev_list, &fs_info->fs_devices->devices);
2408         list_add(&device->dev_alloc_list,
2409                  &fs_info->fs_devices->alloc_list);
2410         fs_info->fs_devices->num_devices++;
2411         fs_info->fs_devices->open_devices++;
2412         fs_info->fs_devices->rw_devices++;
2413         fs_info->fs_devices->total_devices++;
2414         fs_info->fs_devices->total_rw_bytes += device->total_bytes;
2415
2416         spin_lock(&fs_info->free_chunk_lock);
2417         fs_info->free_chunk_space += device->total_bytes;
2418         spin_unlock(&fs_info->free_chunk_lock);
2419
2420         if (!blk_queue_nonrot(bdev_get_queue(bdev)))
2421                 fs_info->fs_devices->rotating = 1;
2422
2423         tmp = btrfs_super_total_bytes(fs_info->super_copy);
2424         btrfs_set_super_total_bytes(fs_info->super_copy,
2425                                     tmp + device->total_bytes);
2426
2427         tmp = btrfs_super_num_devices(fs_info->super_copy);
2428         btrfs_set_super_num_devices(fs_info->super_copy, tmp + 1);
2429
2430         /* add sysfs device entry */
2431         btrfs_sysfs_add_device_link(fs_info->fs_devices, device);
2432
2433         /*
2434          * we've got more storage, clear any full flags on the space
2435          * infos
2436          */
2437         btrfs_clear_space_info_full(fs_info);
2438
2439         mutex_unlock(&fs_info->chunk_mutex);
2440         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2441
2442         if (seeding_dev) {
2443                 mutex_lock(&fs_info->chunk_mutex);
2444                 ret = init_first_rw_device(trans, fs_info);
2445                 mutex_unlock(&fs_info->chunk_mutex);
2446                 if (ret) {
2447                         btrfs_abort_transaction(trans, ret);
2448                         goto error_trans;
2449                 }
2450         }
2451
2452         ret = btrfs_add_device(trans, fs_info, device);
2453         if (ret) {
2454                 btrfs_abort_transaction(trans, ret);
2455                 goto error_trans;
2456         }
2457
2458         if (seeding_dev) {
2459                 char fsid_buf[BTRFS_UUID_UNPARSED_SIZE];
2460
2461                 ret = btrfs_finish_sprout(trans, fs_info);
2462                 if (ret) {
2463                         btrfs_abort_transaction(trans, ret);
2464                         goto error_trans;
2465                 }
2466
2467                 /* Sprouting would change fsid of the mounted root,
2468                  * so rename the fsid on the sysfs
2469                  */
2470                 snprintf(fsid_buf, BTRFS_UUID_UNPARSED_SIZE, "%pU",
2471                                                 fs_info->fsid);
2472                 if (kobject_rename(&fs_info->fs_devices->fsid_kobj, fsid_buf))
2473                         btrfs_warn(fs_info,
2474                                    "sysfs: failed to create fsid for sprout");
2475         }
2476
2477         fs_info->num_tolerated_disk_barrier_failures =
2478                 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
2479         ret = btrfs_commit_transaction(trans);
2480
2481         if (seeding_dev) {
2482                 mutex_unlock(&uuid_mutex);
2483                 up_write(&sb->s_umount);
2484
2485                 if (ret) /* transaction commit */
2486                         return ret;
2487
2488                 ret = btrfs_relocate_sys_chunks(fs_info);
2489                 if (ret < 0)
2490                         btrfs_handle_fs_error(fs_info, ret,
2491                                     "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command.");
2492                 trans = btrfs_attach_transaction(root);
2493                 if (IS_ERR(trans)) {
2494                         if (PTR_ERR(trans) == -ENOENT)
2495                                 return 0;
2496                         return PTR_ERR(trans);
2497                 }
2498                 ret = btrfs_commit_transaction(trans);
2499         }
2500
2501         /* Update ctime/mtime for libblkid */
2502         update_dev_time(device_path);
2503         return ret;
2504
2505 error_trans:
2506         btrfs_end_transaction(trans);
2507         rcu_string_free(device->name);
2508         btrfs_sysfs_rm_device_link(fs_info->fs_devices, device);
2509         kfree(device);
2510 error:
2511         blkdev_put(bdev, FMODE_EXCL);
2512         if (seeding_dev) {
2513                 mutex_unlock(&uuid_mutex);
2514                 up_write(&sb->s_umount);
2515         }
2516         return ret;
2517 }
2518
2519 int btrfs_init_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
2520                                   const char *device_path,
2521                                   struct btrfs_device *srcdev,
2522                                   struct btrfs_device **device_out)
2523 {
2524         struct request_queue *q;
2525         struct btrfs_device *device;
2526         struct block_device *bdev;
2527         struct list_head *devices;
2528         struct rcu_string *name;
2529         u64 devid = BTRFS_DEV_REPLACE_DEVID;
2530         int ret = 0;
2531
2532         *device_out = NULL;
2533         if (fs_info->fs_devices->seeding) {
2534                 btrfs_err(fs_info, "the filesystem is a seed filesystem!");
2535                 return -EINVAL;
2536         }
2537
2538         bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2539                                   fs_info->bdev_holder);
2540         if (IS_ERR(bdev)) {
2541                 btrfs_err(fs_info, "target device %s is invalid!", device_path);
2542                 return PTR_ERR(bdev);
2543         }
2544
2545         filemap_write_and_wait(bdev->bd_inode->i_mapping);
2546
2547         devices = &fs_info->fs_devices->devices;
2548         list_for_each_entry(device, devices, dev_list) {
2549                 if (device->bdev == bdev) {
2550                         btrfs_err(fs_info,
2551                                   "target device is in the filesystem!");
2552                         ret = -EEXIST;
2553                         goto error;
2554                 }
2555         }
2556
2557
2558         if (i_size_read(bdev->bd_inode) <
2559             btrfs_device_get_total_bytes(srcdev)) {
2560                 btrfs_err(fs_info,
2561                           "target device is smaller than source device!");
2562                 ret = -EINVAL;
2563                 goto error;
2564         }
2565
2566
2567         device = btrfs_alloc_device(NULL, &devid, NULL);
2568         if (IS_ERR(device)) {
2569                 ret = PTR_ERR(device);
2570                 goto error;
2571         }
2572
2573         name = rcu_string_strdup(device_path, GFP_NOFS);
2574         if (!name) {
2575                 kfree(device);
2576                 ret = -ENOMEM;
2577                 goto error;
2578         }
2579         rcu_assign_pointer(device->name, name);
2580
2581         q = bdev_get_queue(bdev);
2582         if (blk_queue_discard(q))
2583                 device->can_discard = 1;
2584         mutex_lock(&fs_info->fs_devices->device_list_mutex);
2585         device->writeable = 1;
2586         device->generation = 0;
2587         device->io_width = fs_info->sectorsize;
2588         device->io_align = fs_info->sectorsize;
2589         device->sector_size = fs_info->sectorsize;
2590         device->total_bytes = btrfs_device_get_total_bytes(srcdev);
2591         device->disk_total_bytes = btrfs_device_get_disk_total_bytes(srcdev);
2592         device->bytes_used = btrfs_device_get_bytes_used(srcdev);
2593         ASSERT(list_empty(&srcdev->resized_list));
2594         device->commit_total_bytes = srcdev->commit_total_bytes;
2595         device->commit_bytes_used = device->bytes_used;
2596         device->fs_info = fs_info;
2597         device->bdev = bdev;
2598         device->in_fs_metadata = 1;
2599         device->is_tgtdev_for_dev_replace = 1;
2600         device->mode = FMODE_EXCL;
2601         device->dev_stats_valid = 1;
2602         set_blocksize(device->bdev, 4096);
2603         device->fs_devices = fs_info->fs_devices;
2604         list_add(&device->dev_list, &fs_info->fs_devices->devices);
2605         fs_info->fs_devices->num_devices++;
2606         fs_info->fs_devices->open_devices++;
2607         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2608
2609         *device_out = device;
2610         return ret;
2611
2612 error:
2613         blkdev_put(bdev, FMODE_EXCL);
2614         return ret;
2615 }
2616
2617 void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info *fs_info,
2618                                               struct btrfs_device *tgtdev)
2619 {
2620         u32 sectorsize = fs_info->sectorsize;
2621
2622         WARN_ON(fs_info->fs_devices->rw_devices == 0);
2623         tgtdev->io_width = sectorsize;
2624         tgtdev->io_align = sectorsize;
2625         tgtdev->sector_size = sectorsize;
2626         tgtdev->fs_info = fs_info;
2627         tgtdev->in_fs_metadata = 1;
2628 }
2629
2630 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2631                                         struct btrfs_device *device)
2632 {
2633         int ret;
2634         struct btrfs_path *path;
2635         struct btrfs_root *root = device->fs_info->chunk_root;
2636         struct btrfs_dev_item *dev_item;
2637         struct extent_buffer *leaf;
2638         struct btrfs_key key;
2639
2640         path = btrfs_alloc_path();
2641         if (!path)
2642                 return -ENOMEM;
2643
2644         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2645         key.type = BTRFS_DEV_ITEM_KEY;
2646         key.offset = device->devid;
2647
2648         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2649         if (ret < 0)
2650                 goto out;
2651
2652         if (ret > 0) {
2653                 ret = -ENOENT;
2654                 goto out;
2655         }
2656
2657         leaf = path->nodes[0];
2658         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2659
2660         btrfs_set_device_id(leaf, dev_item, device->devid);
2661         btrfs_set_device_type(leaf, dev_item, device->type);
2662         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2663         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2664         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2665         btrfs_set_device_total_bytes(leaf, dev_item,
2666                                      btrfs_device_get_disk_total_bytes(device));
2667         btrfs_set_device_bytes_used(leaf, dev_item,
2668                                     btrfs_device_get_bytes_used(device));
2669         btrfs_mark_buffer_dirty(leaf);
2670
2671 out:
2672         btrfs_free_path(path);
2673         return ret;
2674 }
2675
2676 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2677                       struct btrfs_device *device, u64 new_size)
2678 {
2679         struct btrfs_fs_info *fs_info = device->fs_info;
2680         struct btrfs_super_block *super_copy = fs_info->super_copy;
2681         struct btrfs_fs_devices *fs_devices;
2682         u64 old_total;
2683         u64 diff;
2684
2685         if (!device->writeable)
2686                 return -EACCES;
2687
2688         mutex_lock(&fs_info->chunk_mutex);
2689         old_total = btrfs_super_total_bytes(super_copy);
2690         diff = new_size - device->total_bytes;
2691
2692         if (new_size <= device->total_bytes ||
2693             device->is_tgtdev_for_dev_replace) {
2694                 mutex_unlock(&fs_info->chunk_mutex);
2695                 return -EINVAL;
2696         }
2697
2698         fs_devices = fs_info->fs_devices;
2699
2700         btrfs_set_super_total_bytes(super_copy, old_total + diff);
2701         device->fs_devices->total_rw_bytes += diff;
2702
2703         btrfs_device_set_total_bytes(device, new_size);
2704         btrfs_device_set_disk_total_bytes(device, new_size);
2705         btrfs_clear_space_info_full(device->fs_info);
2706         if (list_empty(&device->resized_list))
2707                 list_add_tail(&device->resized_list,
2708                               &fs_devices->resized_devices);
2709         mutex_unlock(&fs_info->chunk_mutex);
2710
2711         return btrfs_update_device(trans, device);
2712 }
2713
2714 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
2715                             struct btrfs_fs_info *fs_info, u64 chunk_objectid,
2716                             u64 chunk_offset)
2717 {
2718         struct btrfs_root *root = fs_info->chunk_root;
2719         int ret;
2720         struct btrfs_path *path;
2721         struct btrfs_key key;
2722
2723         path = btrfs_alloc_path();
2724         if (!path)
2725                 return -ENOMEM;
2726
2727         key.objectid = chunk_objectid;
2728         key.offset = chunk_offset;
2729         key.type = BTRFS_CHUNK_ITEM_KEY;
2730
2731         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2732         if (ret < 0)
2733                 goto out;
2734         else if (ret > 0) { /* Logic error or corruption */
2735                 btrfs_handle_fs_error(fs_info, -ENOENT,
2736                                       "Failed lookup while freeing chunk.");
2737                 ret = -ENOENT;
2738                 goto out;
2739         }
2740
2741         ret = btrfs_del_item(trans, root, path);
2742         if (ret < 0)
2743                 btrfs_handle_fs_error(fs_info, ret,
2744                                       "Failed to delete chunk item.");
2745 out:
2746         btrfs_free_path(path);
2747         return ret;
2748 }
2749
2750 static int btrfs_del_sys_chunk(struct btrfs_fs_info *fs_info,
2751                                u64 chunk_objectid, u64 chunk_offset)
2752 {
2753         struct btrfs_super_block *super_copy = fs_info->super_copy;
2754         struct btrfs_disk_key *disk_key;
2755         struct btrfs_chunk *chunk;
2756         u8 *ptr;
2757         int ret = 0;
2758         u32 num_stripes;
2759         u32 array_size;
2760         u32 len = 0;
2761         u32 cur;
2762         struct btrfs_key key;
2763
2764         mutex_lock(&fs_info->chunk_mutex);
2765         array_size = btrfs_super_sys_array_size(super_copy);
2766
2767         ptr = super_copy->sys_chunk_array;
2768         cur = 0;
2769
2770         while (cur < array_size) {
2771                 disk_key = (struct btrfs_disk_key *)ptr;
2772                 btrfs_disk_key_to_cpu(&key, disk_key);
2773
2774                 len = sizeof(*disk_key);
2775
2776                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2777                         chunk = (struct btrfs_chunk *)(ptr + len);
2778                         num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2779                         len += btrfs_chunk_item_size(num_stripes);
2780                 } else {
2781                         ret = -EIO;
2782                         break;
2783                 }
2784                 if (key.objectid == chunk_objectid &&
2785                     key.offset == chunk_offset) {
2786                         memmove(ptr, ptr + len, array_size - (cur + len));
2787                         array_size -= len;
2788                         btrfs_set_super_sys_array_size(super_copy, array_size);
2789                 } else {
2790                         ptr += len;
2791                         cur += len;
2792                 }
2793         }
2794         mutex_unlock(&fs_info->chunk_mutex);
2795         return ret;
2796 }
2797
2798 int btrfs_remove_chunk(struct btrfs_trans_handle *trans,
2799                        struct btrfs_fs_info *fs_info, u64 chunk_offset)
2800 {
2801         struct extent_map_tree *em_tree;
2802         struct extent_map *em;
2803         struct map_lookup *map;
2804         u64 dev_extent_len = 0;
2805         u64 chunk_objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2806         int i, ret = 0;
2807         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2808
2809         em_tree = &fs_info->mapping_tree.map_tree;
2810
2811         read_lock(&em_tree->lock);
2812         em = lookup_extent_mapping(em_tree, chunk_offset, 1);
2813         read_unlock(&em_tree->lock);
2814
2815         if (!em || em->start > chunk_offset ||
2816             em->start + em->len < chunk_offset) {
2817                 /*
2818                  * This is a logic error, but we don't want to just rely on the
2819                  * user having built with ASSERT enabled, so if ASSERT doesn't
2820                  * do anything we still error out.
2821                  */
2822                 ASSERT(0);
2823                 if (em)
2824                         free_extent_map(em);
2825                 return -EINVAL;
2826         }
2827         map = em->map_lookup;
2828         mutex_lock(&fs_info->chunk_mutex);
2829         check_system_chunk(trans, fs_info, map->type);
2830         mutex_unlock(&fs_info->chunk_mutex);
2831
2832         /*
2833          * Take the device list mutex to prevent races with the final phase of
2834          * a device replace operation that replaces the device object associated
2835          * with map stripes (dev-replace.c:btrfs_dev_replace_finishing()).
2836          */
2837         mutex_lock(&fs_devices->device_list_mutex);
2838         for (i = 0; i < map->num_stripes; i++) {
2839                 struct btrfs_device *device = map->stripes[i].dev;
2840                 ret = btrfs_free_dev_extent(trans, device,
2841                                             map->stripes[i].physical,
2842                                             &dev_extent_len);
2843                 if (ret) {
2844                         mutex_unlock(&fs_devices->device_list_mutex);
2845                         btrfs_abort_transaction(trans, ret);
2846                         goto out;
2847                 }
2848
2849                 if (device->bytes_used > 0) {
2850                         mutex_lock(&fs_info->chunk_mutex);
2851                         btrfs_device_set_bytes_used(device,
2852                                         device->bytes_used - dev_extent_len);
2853                         spin_lock(&fs_info->free_chunk_lock);
2854                         fs_info->free_chunk_space += dev_extent_len;
2855                         spin_unlock(&fs_info->free_chunk_lock);
2856                         btrfs_clear_space_info_full(fs_info);
2857                         mutex_unlock(&fs_info->chunk_mutex);
2858                 }
2859
2860                 if (map->stripes[i].dev) {
2861                         ret = btrfs_update_device(trans, map->stripes[i].dev);
2862                         if (ret) {
2863                                 mutex_unlock(&fs_devices->device_list_mutex);
2864                                 btrfs_abort_transaction(trans, ret);
2865                                 goto out;
2866                         }
2867                 }
2868         }
2869         mutex_unlock(&fs_devices->device_list_mutex);
2870
2871         ret = btrfs_free_chunk(trans, fs_info, chunk_objectid, chunk_offset);
2872         if (ret) {
2873                 btrfs_abort_transaction(trans, ret);
2874                 goto out;
2875         }
2876
2877         trace_btrfs_chunk_free(fs_info, map, chunk_offset, em->len);
2878
2879         if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2880                 ret = btrfs_del_sys_chunk(fs_info, chunk_objectid,
2881                                           chunk_offset);
2882                 if (ret) {
2883                         btrfs_abort_transaction(trans, ret);
2884                         goto out;
2885                 }
2886         }
2887
2888         ret = btrfs_remove_block_group(trans, fs_info, chunk_offset, em);
2889         if (ret) {
2890                 btrfs_abort_transaction(trans, ret);
2891                 goto out;
2892         }
2893
2894 out:
2895         /* once for us */
2896         free_extent_map(em);
2897         return ret;
2898 }
2899
2900 static int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
2901 {
2902         struct btrfs_root *root = fs_info->chunk_root;
2903         struct btrfs_trans_handle *trans;
2904         int ret;
2905
2906         /*
2907          * Prevent races with automatic removal of unused block groups.
2908          * After we relocate and before we remove the chunk with offset
2909          * chunk_offset, automatic removal of the block group can kick in,
2910          * resulting in a failure when calling btrfs_remove_chunk() below.
2911          *
2912          * Make sure to acquire this mutex before doing a tree search (dev
2913          * or chunk trees) to find chunks. Otherwise the cleaner kthread might
2914          * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after
2915          * we release the path used to search the chunk/dev tree and before
2916          * the current task acquires this mutex and calls us.
2917          */
2918         ASSERT(mutex_is_locked(&fs_info->delete_unused_bgs_mutex));
2919
2920         ret = btrfs_can_relocate(fs_info, chunk_offset);
2921         if (ret)
2922                 return -ENOSPC;
2923
2924         /* step one, relocate all the extents inside this chunk */
2925         btrfs_scrub_pause(fs_info);
2926         ret = btrfs_relocate_block_group(fs_info, chunk_offset);
2927         btrfs_scrub_continue(fs_info);
2928         if (ret)
2929                 return ret;
2930
2931         trans = btrfs_start_trans_remove_block_group(root->fs_info,
2932                                                      chunk_offset);
2933         if (IS_ERR(trans)) {
2934                 ret = PTR_ERR(trans);
2935                 btrfs_handle_fs_error(root->fs_info, ret, NULL);
2936                 return ret;
2937         }
2938
2939         /*
2940          * step two, delete the device extents and the
2941          * chunk tree entries
2942          */
2943         ret = btrfs_remove_chunk(trans, fs_info, chunk_offset);
2944         btrfs_end_transaction(trans);
2945         return ret;
2946 }
2947
2948 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info)
2949 {
2950         struct btrfs_root *chunk_root = fs_info->chunk_root;
2951         struct btrfs_path *path;
2952         struct extent_buffer *leaf;
2953         struct btrfs_chunk *chunk;
2954         struct btrfs_key key;
2955         struct btrfs_key found_key;
2956         u64 chunk_type;
2957         bool retried = false;
2958         int failed = 0;
2959         int ret;
2960
2961         path = btrfs_alloc_path();
2962         if (!path)
2963                 return -ENOMEM;
2964
2965 again:
2966         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2967         key.offset = (u64)-1;
2968         key.type = BTRFS_CHUNK_ITEM_KEY;
2969
2970         while (1) {
2971                 mutex_lock(&fs_info->delete_unused_bgs_mutex);
2972                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2973                 if (ret < 0) {
2974                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
2975                         goto error;
2976                 }
2977                 BUG_ON(ret == 0); /* Corruption */
2978
2979                 ret = btrfs_previous_item(chunk_root, path, key.objectid,
2980                                           key.type);
2981                 if (ret)
2982                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
2983                 if (ret < 0)
2984                         goto error;
2985                 if (ret > 0)
2986                         break;
2987
2988                 leaf = path->nodes[0];
2989                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2990
2991                 chunk = btrfs_item_ptr(leaf, path->slots[0],
2992                                        struct btrfs_chunk);
2993                 chunk_type = btrfs_chunk_type(leaf, chunk);
2994                 btrfs_release_path(path);
2995
2996                 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
2997                         ret = btrfs_relocate_chunk(fs_info, found_key.offset);
2998                         if (ret == -ENOSPC)
2999                                 failed++;
3000                         else
3001                                 BUG_ON(ret);
3002                 }
3003                 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3004
3005                 if (found_key.offset == 0)
3006                         break;
3007                 key.offset = found_key.offset - 1;
3008         }
3009         ret = 0;
3010         if (failed && !retried) {
3011                 failed = 0;
3012                 retried = true;
3013                 goto again;
3014         } else if (WARN_ON(failed && retried)) {
3015                 ret = -ENOSPC;
3016         }
3017 error:
3018         btrfs_free_path(path);
3019         return ret;
3020 }
3021
3022 static int insert_balance_item(struct btrfs_fs_info *fs_info,
3023                                struct btrfs_balance_control *bctl)
3024 {
3025         struct btrfs_root *root = fs_info->tree_root;
3026         struct btrfs_trans_handle *trans;
3027         struct btrfs_balance_item *item;
3028         struct btrfs_disk_balance_args disk_bargs;
3029         struct btrfs_path *path;
3030         struct extent_buffer *leaf;
3031         struct btrfs_key key;
3032         int ret, err;
3033
3034         path = btrfs_alloc_path();
3035         if (!path)
3036                 return -ENOMEM;
3037
3038         trans = btrfs_start_transaction(root, 0);
3039         if (IS_ERR(trans)) {
3040                 btrfs_free_path(path);
3041                 return PTR_ERR(trans);
3042         }
3043
3044         key.objectid = BTRFS_BALANCE_OBJECTID;
3045         key.type = BTRFS_TEMPORARY_ITEM_KEY;
3046         key.offset = 0;
3047
3048         ret = btrfs_insert_empty_item(trans, root, path, &key,
3049                                       sizeof(*item));
3050         if (ret)
3051                 goto out;
3052
3053         leaf = path->nodes[0];
3054         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3055
3056         memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item));
3057
3058         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
3059         btrfs_set_balance_data(leaf, item, &disk_bargs);
3060         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
3061         btrfs_set_balance_meta(leaf, item, &disk_bargs);
3062         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
3063         btrfs_set_balance_sys(leaf, item, &disk_bargs);
3064
3065         btrfs_set_balance_flags(leaf, item, bctl->flags);
3066
3067         btrfs_mark_buffer_dirty(leaf);
3068 out:
3069         btrfs_free_path(path);
3070         err = btrfs_commit_transaction(trans);
3071         if (err && !ret)
3072                 ret = err;
3073         return ret;
3074 }
3075
3076 static int del_balance_item(struct btrfs_fs_info *fs_info)
3077 {
3078         struct btrfs_root *root = fs_info->tree_root;
3079         struct btrfs_trans_handle *trans;
3080         struct btrfs_path *path;
3081         struct btrfs_key key;
3082         int ret, err;
3083
3084         path = btrfs_alloc_path();
3085         if (!path)
3086                 return -ENOMEM;
3087
3088         trans = btrfs_start_transaction(root, 0);
3089         if (IS_ERR(trans)) {
3090                 btrfs_free_path(path);
3091                 return PTR_ERR(trans);
3092         }
3093
3094         key.objectid = BTRFS_BALANCE_OBJECTID;
3095         key.type = BTRFS_TEMPORARY_ITEM_KEY;
3096         key.offset = 0;
3097
3098         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3099         if (ret < 0)
3100                 goto out;
3101         if (ret > 0) {
3102                 ret = -ENOENT;
3103                 goto out;
3104         }
3105
3106         ret = btrfs_del_item(trans, root, path);
3107 out:
3108         btrfs_free_path(path);
3109         err = btrfs_commit_transaction(trans);
3110         if (err && !ret)
3111                 ret = err;
3112         return ret;
3113 }
3114
3115 /*
3116  * This is a heuristic used to reduce the number of chunks balanced on
3117  * resume after balance was interrupted.
3118  */
3119 static void update_balance_args(struct btrfs_balance_control *bctl)
3120 {
3121         /*
3122          * Turn on soft mode for chunk types that were being converted.
3123          */
3124         if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
3125                 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
3126         if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
3127                 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
3128         if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
3129                 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
3130
3131         /*
3132          * Turn on usage filter if is not already used.  The idea is
3133          * that chunks that we have already balanced should be
3134          * reasonably full.  Don't do it for chunks that are being
3135          * converted - that will keep us from relocating unconverted
3136          * (albeit full) chunks.
3137          */
3138         if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3139             !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3140             !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3141                 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
3142                 bctl->data.usage = 90;
3143         }
3144         if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3145             !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3146             !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3147                 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
3148                 bctl->sys.usage = 90;
3149         }
3150         if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3151             !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3152             !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3153                 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
3154                 bctl->meta.usage = 90;
3155         }
3156 }
3157
3158 /*
3159  * Should be called with both balance and volume mutexes held to
3160  * serialize other volume operations (add_dev/rm_dev/resize) with
3161  * restriper.  Same goes for unset_balance_control.
3162  */
3163 static void set_balance_control(struct btrfs_balance_control *bctl)
3164 {
3165         struct btrfs_fs_info *fs_info = bctl->fs_info;
3166
3167         BUG_ON(fs_info->balance_ctl);
3168
3169         spin_lock(&fs_info->balance_lock);
3170         fs_info->balance_ctl = bctl;
3171         spin_unlock(&fs_info->balance_lock);
3172 }
3173
3174 static void unset_balance_control(struct btrfs_fs_info *fs_info)
3175 {
3176         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3177
3178         BUG_ON(!fs_info->balance_ctl);
3179
3180         spin_lock(&fs_info->balance_lock);
3181         fs_info->balance_ctl = NULL;
3182         spin_unlock(&fs_info->balance_lock);
3183
3184         kfree(bctl);
3185 }
3186
3187 /*
3188  * Balance filters.  Return 1 if chunk should be filtered out
3189  * (should not be balanced).
3190  */
3191 static int chunk_profiles_filter(u64 chunk_type,
3192                                  struct btrfs_balance_args *bargs)
3193 {
3194         chunk_type = chunk_to_extended(chunk_type) &
3195                                 BTRFS_EXTENDED_PROFILE_MASK;
3196
3197         if (bargs->profiles & chunk_type)
3198                 return 0;
3199
3200         return 1;
3201 }
3202
3203 static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
3204                               struct btrfs_balance_args *bargs)
3205 {
3206         struct btrfs_block_group_cache *cache;
3207         u64 chunk_used;
3208         u64 user_thresh_min;
3209         u64 user_thresh_max;
3210         int ret = 1;
3211
3212         cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3213         chunk_used = btrfs_block_group_used(&cache->item);
3214
3215         if (bargs->usage_min == 0)
3216                 user_thresh_min = 0;
3217         else
3218                 user_thresh_min = div_factor_fine(cache->key.offset,
3219                                         bargs->usage_min);
3220
3221         if (bargs->usage_max == 0)
3222                 user_thresh_max = 1;
3223         else if (bargs->usage_max > 100)
3224                 user_thresh_max = cache->key.offset;
3225         else
3226                 user_thresh_max = div_factor_fine(cache->key.offset,
3227                                         bargs->usage_max);
3228
3229         if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max)
3230                 ret = 0;
3231
3232         btrfs_put_block_group(cache);
3233         return ret;
3234 }
3235
3236 static int chunk_usage_filter(struct btrfs_fs_info *fs_info,
3237                 u64 chunk_offset, struct btrfs_balance_args *bargs)
3238 {
3239         struct btrfs_block_group_cache *cache;
3240         u64 chunk_used, user_thresh;
3241         int ret = 1;
3242
3243         cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3244         chunk_used = btrfs_block_group_used(&cache->item);
3245
3246         if (bargs->usage_min == 0)
3247                 user_thresh = 1;
3248         else if (bargs->usage > 100)
3249                 user_thresh = cache->key.offset;
3250         else
3251                 user_thresh = div_factor_fine(cache->key.offset,
3252                                               bargs->usage);
3253
3254         if (chunk_used < user_thresh)
3255                 ret = 0;
3256
3257         btrfs_put_block_group(cache);
3258         return ret;
3259 }
3260
3261 static int chunk_devid_filter(struct extent_buffer *leaf,
3262                               struct btrfs_chunk *chunk,
3263                               struct btrfs_balance_args *bargs)
3264 {
3265         struct btrfs_stripe *stripe;
3266         int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3267         int i;
3268
3269         for (i = 0; i < num_stripes; i++) {
3270                 stripe = btrfs_stripe_nr(chunk, i);
3271                 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
3272                         return 0;
3273         }
3274
3275         return 1;
3276 }
3277
3278 /* [pstart, pend) */
3279 static int chunk_drange_filter(struct extent_buffer *leaf,
3280                                struct btrfs_chunk *chunk,
3281                                u64 chunk_offset,
3282                                struct btrfs_balance_args *bargs)
3283 {
3284         struct btrfs_stripe *stripe;
3285         int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3286         u64 stripe_offset;
3287         u64 stripe_length;
3288         int factor;
3289         int i;
3290
3291         if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
3292                 return 0;
3293
3294         if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP |
3295              BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) {
3296                 factor = num_stripes / 2;
3297         } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) {
3298                 factor = num_stripes - 1;
3299         } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) {
3300                 factor = num_stripes - 2;
3301         } else {
3302                 factor = num_stripes;
3303         }
3304
3305         for (i = 0; i < num_stripes; i++) {
3306                 stripe = btrfs_stripe_nr(chunk, i);
3307                 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
3308                         continue;
3309
3310                 stripe_offset = btrfs_stripe_offset(leaf, stripe);
3311                 stripe_length = btrfs_chunk_length(leaf, chunk);
3312                 stripe_length = div_u64(stripe_length, factor);
3313
3314                 if (stripe_offset < bargs->pend &&
3315                     stripe_offset + stripe_length > bargs->pstart)
3316                         return 0;
3317         }
3318
3319         return 1;
3320 }
3321
3322 /* [vstart, vend) */
3323 static int chunk_vrange_filter(struct extent_buffer *leaf,
3324                                struct btrfs_chunk *chunk,
3325                                u64 chunk_offset,
3326                                struct btrfs_balance_args *bargs)
3327 {
3328         if (chunk_offset < bargs->vend &&
3329             chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
3330                 /* at least part of the chunk is inside this vrange */
3331                 return 0;
3332
3333         return 1;
3334 }
3335
3336 static int chunk_stripes_range_filter(struct extent_buffer *leaf,
3337                                struct btrfs_chunk *chunk,
3338                                struct btrfs_balance_args *bargs)
3339 {
3340         int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3341
3342         if (bargs->stripes_min <= num_stripes
3343                         && num_stripes <= bargs->stripes_max)
3344                 return 0;
3345
3346         return 1;
3347 }
3348
3349 static int chunk_soft_convert_filter(u64 chunk_type,
3350                                      struct btrfs_balance_args *bargs)
3351 {
3352         if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
3353                 return 0;
3354
3355         chunk_type = chunk_to_extended(chunk_type) &
3356                                 BTRFS_EXTENDED_PROFILE_MASK;
3357
3358         if (bargs->target == chunk_type)
3359                 return 1;
3360
3361         return 0;
3362 }
3363
3364 static int should_balance_chunk(struct btrfs_fs_info *fs_info,
3365                                 struct extent_buffer *leaf,
3366                                 struct btrfs_chunk *chunk, u64 chunk_offset)
3367 {
3368         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3369         struct btrfs_balance_args *bargs = NULL;
3370         u64 chunk_type = btrfs_chunk_type(leaf, chunk);
3371
3372         /* type filter */
3373         if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
3374               (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
3375                 return 0;
3376         }
3377
3378         if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3379                 bargs = &bctl->data;
3380         else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3381                 bargs = &bctl->sys;
3382         else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3383                 bargs = &bctl->meta;
3384
3385         /* profiles filter */
3386         if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
3387             chunk_profiles_filter(chunk_type, bargs)) {
3388                 return 0;
3389         }
3390
3391         /* usage filter */
3392         if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
3393             chunk_usage_filter(fs_info, chunk_offset, bargs)) {
3394                 return 0;
3395         } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3396             chunk_usage_range_filter(fs_info, chunk_offset, bargs)) {
3397                 return 0;
3398         }
3399
3400         /* devid filter */
3401         if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
3402             chunk_devid_filter(leaf, chunk, bargs)) {
3403                 return 0;
3404         }
3405
3406         /* drange filter, makes sense only with devid filter */
3407         if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
3408             chunk_drange_filter(leaf, chunk, chunk_offset, bargs)) {
3409                 return 0;
3410         }
3411
3412         /* vrange filter */
3413         if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
3414             chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
3415                 return 0;
3416         }
3417
3418         /* stripes filter */
3419         if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) &&
3420             chunk_stripes_range_filter(leaf, chunk, bargs)) {
3421                 return 0;
3422         }
3423
3424         /* soft profile changing mode */
3425         if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
3426             chunk_soft_convert_filter(chunk_type, bargs)) {
3427                 return 0;
3428         }
3429
3430         /*
3431          * limited by count, must be the last filter
3432          */
3433         if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) {
3434                 if (bargs->limit == 0)
3435                         return 0;
3436                 else
3437                         bargs->limit--;
3438         } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) {
3439                 /*
3440                  * Same logic as the 'limit' filter; the minimum cannot be
3441                  * determined here because we do not have the global information
3442                  * about the count of all chunks that satisfy the filters.
3443                  */
3444                 if (bargs->limit_max == 0)
3445                         return 0;
3446                 else
3447                         bargs->limit_max--;
3448         }
3449
3450         return 1;
3451 }
3452
3453 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
3454 {
3455         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3456         struct btrfs_root *chunk_root = fs_info->chunk_root;
3457         struct btrfs_root *dev_root = fs_info->dev_root;
3458         struct list_head *devices;
3459         struct btrfs_device *device;
3460         u64 old_size;
3461         u64 size_to_free;
3462         u64 chunk_type;
3463         struct btrfs_chunk *chunk;
3464         struct btrfs_path *path = NULL;
3465         struct btrfs_key key;
3466         struct btrfs_key found_key;
3467         struct btrfs_trans_handle *trans;
3468         struct extent_buffer *leaf;
3469         int slot;
3470         int ret;
3471         int enospc_errors = 0;
3472         bool counting = true;
3473         /* The single value limit and min/max limits use the same bytes in the */
3474         u64 limit_data = bctl->data.limit;
3475         u64 limit_meta = bctl->meta.limit;
3476         u64 limit_sys = bctl->sys.limit;
3477         u32 count_data = 0;
3478         u32 count_meta = 0;
3479         u32 count_sys = 0;
3480         int chunk_reserved = 0;
3481         u64 bytes_used = 0;
3482
3483         /* step one make some room on all the devices */
3484         devices = &fs_info->fs_devices->devices;
3485         list_for_each_entry(device, devices, dev_list) {
3486                 old_size = btrfs_device_get_total_bytes(device);
3487                 size_to_free = div_factor(old_size, 1);
3488                 size_to_free = min_t(u64, size_to_free, SZ_1M);
3489                 if (!device->writeable ||
3490                     btrfs_device_get_total_bytes(device) -
3491                     btrfs_device_get_bytes_used(device) > size_to_free ||
3492                     device->is_tgtdev_for_dev_replace)
3493                         continue;
3494
3495                 ret = btrfs_shrink_device(device, old_size - size_to_free);
3496                 if (ret == -ENOSPC)
3497                         break;
3498                 if (ret) {
3499                         /* btrfs_shrink_device never returns ret > 0 */
3500                         WARN_ON(ret > 0);
3501                         goto error;
3502                 }
3503
3504                 trans = btrfs_start_transaction(dev_root, 0);
3505                 if (IS_ERR(trans)) {
3506                         ret = PTR_ERR(trans);
3507                         btrfs_info_in_rcu(fs_info,
3508                  "resize: unable to start transaction after shrinking device %s (error %d), old size %llu, new size %llu",
3509                                           rcu_str_deref(device->name), ret,
3510                                           old_size, old_size - size_to_free);
3511                         goto error;
3512                 }
3513
3514                 ret = btrfs_grow_device(trans, device, old_size);
3515                 if (ret) {
3516                         btrfs_end_transaction(trans);
3517                         /* btrfs_grow_device never returns ret > 0 */
3518                         WARN_ON(ret > 0);
3519                         btrfs_info_in_rcu(fs_info,
3520                  "resize: unable to grow device after shrinking device %s (error %d), old size %llu, new size %llu",
3521                                           rcu_str_deref(device->name), ret,
3522                                           old_size, old_size - size_to_free);
3523                         goto error;
3524                 }
3525
3526                 btrfs_end_transaction(trans);
3527         }
3528
3529         /* step two, relocate all the chunks */
3530         path = btrfs_alloc_path();
3531         if (!path) {
3532                 ret = -ENOMEM;
3533                 goto error;
3534         }
3535
3536         /* zero out stat counters */
3537         spin_lock(&fs_info->balance_lock);
3538         memset(&bctl->stat, 0, sizeof(bctl->stat));
3539         spin_unlock(&fs_info->balance_lock);
3540 again:
3541         if (!counting) {
3542                 /*
3543                  * The single value limit and min/max limits use the same bytes
3544                  * in the
3545                  */
3546                 bctl->data.limit = limit_data;
3547                 bctl->meta.limit = limit_meta;
3548                 bctl->sys.limit = limit_sys;
3549         }
3550         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3551         key.offset = (u64)-1;
3552         key.type = BTRFS_CHUNK_ITEM_KEY;
3553
3554         while (1) {
3555                 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
3556                     atomic_read(&fs_info->balance_cancel_req)) {
3557                         ret = -ECANCELED;
3558                         goto error;
3559                 }
3560
3561                 mutex_lock(&fs_info->delete_unused_bgs_mutex);
3562                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3563                 if (ret < 0) {
3564                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3565                         goto error;
3566                 }
3567
3568                 /*
3569                  * this shouldn't happen, it means the last relocate
3570                  * failed
3571                  */
3572                 if (ret == 0)
3573                         BUG(); /* FIXME break ? */
3574
3575                 ret = btrfs_previous_item(chunk_root, path, 0,
3576                                           BTRFS_CHUNK_ITEM_KEY);
3577                 if (ret) {
3578                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3579                         ret = 0;
3580                         break;
3581                 }
3582
3583                 leaf = path->nodes[0];
3584                 slot = path->slots[0];
3585                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3586
3587                 if (found_key.objectid != key.objectid) {
3588                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3589                         break;
3590                 }
3591
3592                 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3593                 chunk_type = btrfs_chunk_type(leaf, chunk);
3594
3595                 if (!counting) {
3596                         spin_lock(&fs_info->balance_lock);
3597                         bctl->stat.considered++;
3598                         spin_unlock(&fs_info->balance_lock);
3599                 }
3600
3601                 ret = should_balance_chunk(fs_info, leaf, chunk,
3602                                            found_key.offset);
3603
3604                 btrfs_release_path(path);
3605                 if (!ret) {
3606                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3607                         goto loop;
3608                 }
3609
3610                 if (counting) {
3611                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3612                         spin_lock(&fs_info->balance_lock);
3613                         bctl->stat.expected++;
3614                         spin_unlock(&fs_info->balance_lock);
3615
3616                         if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3617                                 count_data++;
3618                         else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3619                                 count_sys++;
3620                         else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3621                                 count_meta++;
3622
3623                         goto loop;
3624                 }
3625
3626                 /*
3627                  * Apply limit_min filter, no need to check if the LIMITS
3628                  * filter is used, limit_min is 0 by default
3629                  */
3630                 if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
3631                                         count_data < bctl->data.limit_min)
3632                                 || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) &&
3633                                         count_meta < bctl->meta.limit_min)
3634                                 || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) &&
3635                                         count_sys < bctl->sys.limit_min)) {
3636                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3637                         goto loop;
3638                 }
3639
3640                 ASSERT(fs_info->data_sinfo);
3641                 spin_lock(&fs_info->data_sinfo->lock);
3642                 bytes_used = fs_info->data_sinfo->bytes_used;
3643                 spin_unlock(&fs_info->data_sinfo->lock);
3644
3645                 if ((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
3646                     !chunk_reserved && !bytes_used) {
3647                         trans = btrfs_start_transaction(chunk_root, 0);
3648                         if (IS_ERR(trans)) {
3649                                 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3650                                 ret = PTR_ERR(trans);
3651                                 goto error;
3652                         }
3653
3654                         ret = btrfs_force_chunk_alloc(trans, fs_info,
3655                                                       BTRFS_BLOCK_GROUP_DATA);
3656                         btrfs_end_transaction(trans);
3657                         if (ret < 0) {
3658                                 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3659                                 goto error;
3660                         }
3661                         chunk_reserved = 1;
3662                 }
3663
3664                 ret = btrfs_relocate_chunk(fs_info, found_key.offset);
3665                 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3666                 if (ret && ret != -ENOSPC)
3667                         goto error;
3668                 if (ret == -ENOSPC) {
3669                         enospc_errors++;
3670                 } else {
3671                         spin_lock(&fs_info->balance_lock);
3672                         bctl->stat.completed++;
3673                         spin_unlock(&fs_info->balance_lock);
3674                 }
3675 loop:
3676                 if (found_key.offset == 0)
3677                         break;
3678                 key.offset = found_key.offset - 1;
3679         }
3680
3681         if (counting) {
3682                 btrfs_release_path(path);
3683                 counting = false;
3684                 goto again;
3685         }
3686 error:
3687         btrfs_free_path(path);
3688         if (enospc_errors) {
3689                 btrfs_info(fs_info, "%d enospc errors during balance",
3690                            enospc_errors);
3691                 if (!ret)
3692                         ret = -ENOSPC;
3693         }
3694
3695         return ret;
3696 }
3697
3698 /**
3699  * alloc_profile_is_valid - see if a given profile is valid and reduced
3700  * @flags: profile to validate
3701  * @extended: if true @flags is treated as an extended profile
3702  */
3703 static int alloc_profile_is_valid(u64 flags, int extended)
3704 {
3705         u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
3706                                BTRFS_BLOCK_GROUP_PROFILE_MASK);
3707
3708         flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
3709
3710         /* 1) check that all other bits are zeroed */
3711         if (flags & ~mask)
3712                 return 0;
3713
3714         /* 2) see if profile is reduced */
3715         if (flags == 0)
3716                 return !extended; /* "0" is valid for usual profiles */
3717
3718         /* true if exactly one bit set */
3719         return (flags & (flags - 1)) == 0;
3720 }
3721
3722 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
3723 {
3724         /* cancel requested || normal exit path */
3725         return atomic_read(&fs_info->balance_cancel_req) ||
3726                 (atomic_read(&fs_info->balance_pause_req) == 0 &&
3727                  atomic_read(&fs_info->balance_cancel_req) == 0);
3728 }
3729
3730 static void __cancel_balance(struct btrfs_fs_info *fs_info)
3731 {
3732         int ret;
3733
3734         unset_balance_control(fs_info);
3735         ret = del_balance_item(fs_info);
3736         if (ret)
3737                 btrfs_handle_fs_error(fs_info, ret, NULL);
3738
3739         atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3740 }
3741
3742 /* Non-zero return value signifies invalidity */
3743 static inline int validate_convert_profile(struct btrfs_balance_args *bctl_arg,
3744                 u64 allowed)
3745 {
3746         return ((bctl_arg->flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3747                 (!alloc_profile_is_valid(bctl_arg->target, 1) ||
3748                  (bctl_arg->target & ~allowed)));
3749 }
3750
3751 /*
3752  * Should be called with both balance and volume mutexes held
3753  */
3754 int btrfs_balance(struct btrfs_balance_control *bctl,
3755                   struct btrfs_ioctl_balance_args *bargs)
3756 {
3757         struct btrfs_fs_info *fs_info = bctl->fs_info;
3758         u64 allowed;
3759         int mixed = 0;
3760         int ret;
3761         u64 num_devices;
3762         unsigned seq;
3763
3764         if (btrfs_fs_closing(fs_info) ||
3765             atomic_read(&fs_info->balance_pause_req) ||
3766             atomic_read(&fs_info->balance_cancel_req)) {
3767                 ret = -EINVAL;
3768                 goto out;
3769         }
3770
3771         allowed = btrfs_super_incompat_flags(fs_info->super_copy);
3772         if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
3773                 mixed = 1;
3774
3775         /*
3776          * In case of mixed groups both data and meta should be picked,
3777          * and identical options should be given for both of them.
3778          */
3779         allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
3780         if (mixed && (bctl->flags & allowed)) {
3781                 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
3782                     !(bctl->flags & BTRFS_BALANCE_METADATA) ||
3783                     memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
3784                         btrfs_err(fs_info,
3785                                   "with mixed groups data and metadata balance options must be the same");
3786                         ret = -EINVAL;
3787                         goto out;
3788                 }
3789         }
3790
3791         num_devices = fs_info->fs_devices->num_devices;
3792         btrfs_dev_replace_lock(&fs_info->dev_replace, 0);
3793         if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
3794                 BUG_ON(num_devices < 1);
3795                 num_devices--;
3796         }
3797         btrfs_dev_replace_unlock(&fs_info->dev_replace, 0);
3798         allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE | BTRFS_BLOCK_GROUP_DUP;
3799         if (num_devices > 1)
3800                 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1);
3801         if (num_devices > 2)
3802                 allowed |= BTRFS_BLOCK_GROUP_RAID5;
3803         if (num_devices > 3)
3804                 allowed |= (BTRFS_BLOCK_GROUP_RAID10 |
3805                             BTRFS_BLOCK_GROUP_RAID6);
3806         if (validate_convert_profile(&bctl->data, allowed)) {
3807                 btrfs_err(fs_info,
3808                           "unable to start balance with target data profile %llu",
3809                           bctl->data.target);
3810                 ret = -EINVAL;
3811                 goto out;
3812         }
3813         if (validate_convert_profile(&bctl->meta, allowed)) {
3814                 btrfs_err(fs_info,
3815                           "unable to start balance with target metadata profile %llu",
3816                           bctl->meta.target);
3817                 ret = -EINVAL;
3818                 goto out;
3819         }
3820         if (validate_convert_profile(&bctl->sys, allowed)) {
3821                 btrfs_err(fs_info,
3822                           "unable to start balance with target system profile %llu",
3823                           bctl->sys.target);
3824                 ret = -EINVAL;
3825                 goto out;
3826         }
3827
3828         /* allow to reduce meta or sys integrity only if force set */
3829         allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3830                         BTRFS_BLOCK_GROUP_RAID10 |
3831                         BTRFS_BLOCK_GROUP_RAID5 |
3832                         BTRFS_BLOCK_GROUP_RAID6;
3833         do {
3834                 seq = read_seqbegin(&fs_info->profiles_lock);
3835
3836                 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3837                      (fs_info->avail_system_alloc_bits & allowed) &&
3838                      !(bctl->sys.target & allowed)) ||
3839                     ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3840                      (fs_info->avail_metadata_alloc_bits & allowed) &&
3841                      !(bctl->meta.target & allowed))) {
3842                         if (bctl->flags & BTRFS_BALANCE_FORCE) {
3843                                 btrfs_info(fs_info,
3844                                            "force reducing metadata integrity");
3845                         } else {
3846                                 btrfs_err(fs_info,
3847                                           "balance will reduce metadata integrity, use force if you want this");
3848                                 ret = -EINVAL;
3849                                 goto out;
3850                         }
3851                 }
3852         } while (read_seqretry(&fs_info->profiles_lock, seq));
3853
3854         if (btrfs_get_num_tolerated_disk_barrier_failures(bctl->meta.target) <
3855                 btrfs_get_num_tolerated_disk_barrier_failures(bctl->data.target)) {
3856                 btrfs_warn(fs_info,
3857                            "metadata profile 0x%llx has lower redundancy than data profile 0x%llx",
3858                            bctl->meta.target, bctl->data.target);
3859         }
3860
3861         if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3862                 fs_info->num_tolerated_disk_barrier_failures = min(
3863                         btrfs_calc_num_tolerated_disk_barrier_failures(fs_info),
3864                         btrfs_get_num_tolerated_disk_barrier_failures(
3865                                 bctl->sys.target));
3866         }
3867
3868         ret = insert_balance_item(fs_info, bctl);
3869         if (ret && ret != -EEXIST)
3870                 goto out;
3871
3872         if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
3873                 BUG_ON(ret == -EEXIST);
3874                 set_balance_control(bctl);
3875         } else {
3876                 BUG_ON(ret != -EEXIST);
3877                 spin_lock(&fs_info->balance_lock);
3878                 update_balance_args(bctl);
3879                 spin_unlock(&fs_info->balance_lock);
3880         }
3881
3882         atomic_inc(&fs_info->balance_running);
3883         mutex_unlock(&fs_info->balance_mutex);
3884
3885         ret = __btrfs_balance(fs_info);
3886
3887         mutex_lock(&fs_info->balance_mutex);
3888         atomic_dec(&fs_info->balance_running);
3889
3890         if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3891                 fs_info->num_tolerated_disk_barrier_failures =
3892                         btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3893         }
3894
3895         if (bargs) {
3896                 memset(bargs, 0, sizeof(*bargs));
3897                 update_ioctl_balance_args(fs_info, 0, bargs);
3898         }
3899
3900         if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
3901             balance_need_close(fs_info)) {
3902                 __cancel_balance(fs_info);
3903         }
3904
3905         wake_up(&fs_info->balance_wait_q);
3906
3907         return ret;
3908 out:
3909         if (bctl->flags & BTRFS_BALANCE_RESUME)
3910                 __cancel_balance(fs_info);
3911         else {
3912                 kfree(bctl);
3913                 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3914         }
3915         return ret;
3916 }
3917
3918 static int balance_kthread(void *data)
3919 {
3920         struct btrfs_fs_info *fs_info = data;
3921         int ret = 0;
3922
3923         mutex_lock(&fs_info->volume_mutex);
3924         mutex_lock(&fs_info->balance_mutex);
3925
3926         if (fs_info->balance_ctl) {
3927                 btrfs_info(fs_info, "continuing balance");
3928                 ret = btrfs_balance(fs_info->balance_ctl, NULL);
3929         }
3930
3931         mutex_unlock(&fs_info->balance_mutex);
3932         mutex_unlock(&fs_info->volume_mutex);
3933
3934         return ret;
3935 }
3936
3937 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
3938 {
3939         struct task_struct *tsk;
3940
3941         spin_lock(&fs_info->balance_lock);
3942         if (!fs_info->balance_ctl) {
3943                 spin_unlock(&fs_info->balance_lock);
3944                 return 0;
3945         }
3946         spin_unlock(&fs_info->balance_lock);
3947
3948         if (btrfs_test_opt(fs_info, SKIP_BALANCE)) {
3949                 btrfs_info(fs_info, "force skipping balance");
3950                 return 0;
3951         }
3952
3953         tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
3954         return PTR_ERR_OR_ZERO(tsk);
3955 }
3956
3957 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
3958 {
3959         struct btrfs_balance_control *bctl;
3960         struct btrfs_balance_item *item;
3961         struct btrfs_disk_balance_args disk_bargs;
3962         struct btrfs_path *path;
3963         struct extent_buffer *leaf;
3964         struct btrfs_key key;
3965         int ret;
3966
3967         path = btrfs_alloc_path();
3968         if (!path)
3969                 return -ENOMEM;
3970
3971         key.objectid = BTRFS_BALANCE_OBJECTID;
3972         key.type = BTRFS_TEMPORARY_ITEM_KEY;
3973         key.offset = 0;
3974
3975         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
3976         if (ret < 0)
3977                 goto out;
3978         if (ret > 0) { /* ret = -ENOENT; */
3979                 ret = 0;
3980                 goto out;
3981         }
3982
3983         bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
3984         if (!bctl) {
3985                 ret = -ENOMEM;
3986                 goto out;
3987         }
3988
3989         leaf = path->nodes[0];
3990         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3991
3992         bctl->fs_info = fs_info;
3993         bctl->flags = btrfs_balance_flags(leaf, item);
3994         bctl->flags |= BTRFS_BALANCE_RESUME;
3995
3996         btrfs_balance_data(leaf, item, &disk_bargs);
3997         btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
3998         btrfs_balance_meta(leaf, item, &disk_bargs);
3999         btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
4000         btrfs_balance_sys(leaf, item, &disk_bargs);
4001         btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
4002
4003         WARN_ON(atomic_xchg(&fs_info->mutually_exclusive_operation_running, 1));
4004
4005         mutex_lock(&fs_info->volume_mutex);
4006         mutex_lock(&fs_info->balance_mutex);
4007
4008         set_balance_control(bctl);
4009
4010         mutex_unlock(&fs_info->balance_mutex);
4011         mutex_unlock(&fs_info->volume_mutex);
4012 out:
4013         btrfs_free_path(path);
4014         return ret;
4015 }
4016
4017 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
4018 {
4019         int ret = 0;
4020
4021         mutex_lock(&fs_info->balance_mutex);
4022         if (!fs_info->balance_ctl) {
4023                 mutex_unlock(&fs_info->balance_mutex);
4024                 return -ENOTCONN;
4025         }
4026
4027         if (atomic_read(&fs_info->balance_running)) {
4028                 atomic_inc(&fs_info->balance_pause_req);
4029                 mutex_unlock(&fs_info->balance_mutex);
4030
4031                 wait_event(fs_info->balance_wait_q,
4032                            atomic_read(&fs_info->balance_running) == 0);
4033
4034                 mutex_lock(&fs_info->balance_mutex);
4035                 /* we are good with balance_ctl ripped off from under us */
4036                 BUG_ON(atomic_read(&fs_info->balance_running));
4037                 atomic_dec(&fs_info->balance_pause_req);
4038         } else {
4039                 ret = -ENOTCONN;
4040         }
4041
4042         mutex_unlock(&fs_info->balance_mutex);
4043         return ret;
4044 }
4045
4046 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
4047 {
4048         if (fs_info->sb->s_flags & MS_RDONLY)
4049                 return -EROFS;
4050
4051         mutex_lock(&fs_info->balance_mutex);
4052         if (!fs_info->balance_ctl) {
4053                 mutex_unlock(&fs_info->balance_mutex);
4054                 return -ENOTCONN;
4055         }
4056
4057         atomic_inc(&fs_info->balance_cancel_req);
4058         /*
4059          * if we are running just wait and return, balance item is
4060          * deleted in btrfs_balance in this case
4061          */
4062         if (atomic_read(&fs_info->balance_running)) {
4063                 mutex_unlock(&fs_info->balance_mutex);
4064                 wait_event(fs_info->balance_wait_q,
4065                            atomic_read(&fs_info->balance_running) == 0);
4066                 mutex_lock(&fs_info->balance_mutex);
4067         } else {
4068                 /* __cancel_balance needs volume_mutex */
4069                 mutex_unlock(&fs_info->balance_mutex);
4070                 mutex_lock(&fs_info->volume_mutex);
4071                 mutex_lock(&fs_info->balance_mutex);
4072
4073                 if (fs_info->balance_ctl)
4074                         __cancel_balance(fs_info);
4075
4076                 mutex_unlock(&fs_info->volume_mutex);
4077         }
4078
4079         BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running));
4080         atomic_dec(&fs_info->balance_cancel_req);
4081         mutex_unlock(&fs_info->balance_mutex);
4082         return 0;
4083 }
4084
4085 static int btrfs_uuid_scan_kthread(void *data)
4086 {
4087         struct btrfs_fs_info *fs_info = data;
4088         struct btrfs_root *root = fs_info->tree_root;
4089         struct btrfs_key key;
4090         struct btrfs_key max_key;
4091         struct btrfs_path *path = NULL;
4092         int ret = 0;
4093         struct extent_buffer *eb;
4094         int slot;
4095         struct btrfs_root_item root_item;
4096         u32 item_size;
4097         struct btrfs_trans_handle *trans = NULL;
4098
4099         path = btrfs_alloc_path();
4100         if (!path) {
4101                 ret = -ENOMEM;
4102                 goto out;
4103         }
4104
4105         key.objectid = 0;
4106         key.type = BTRFS_ROOT_ITEM_KEY;
4107         key.offset = 0;
4108
4109         max_key.objectid = (u64)-1;
4110         max_key.type = BTRFS_ROOT_ITEM_KEY;
4111         max_key.offset = (u64)-1;
4112
4113         while (1) {
4114                 ret = btrfs_search_forward(root, &key, path, 0);
4115                 if (ret) {
4116                         if (ret > 0)
4117                                 ret = 0;
4118                         break;
4119                 }
4120
4121                 if (key.type != BTRFS_ROOT_ITEM_KEY ||
4122                     (key.objectid < BTRFS_FIRST_FREE_OBJECTID &&
4123                      key.objectid != BTRFS_FS_TREE_OBJECTID) ||
4124                     key.objectid > BTRFS_LAST_FREE_OBJECTID)
4125                         goto skip;
4126
4127                 eb = path->nodes[0];
4128                 slot = path->slots[0];
4129                 item_size = btrfs_item_size_nr(eb, slot);
4130                 if (item_size < sizeof(root_item))
4131                         goto skip;
4132
4133                 read_extent_buffer(eb, &root_item,
4134                                    btrfs_item_ptr_offset(eb, slot),
4135                                    (int)sizeof(root_item));
4136                 if (btrfs_root_refs(&root_item) == 0)
4137                         goto skip;
4138
4139                 if (!btrfs_is_empty_uuid(root_item.uuid) ||
4140                     !btrfs_is_empty_uuid(root_item.received_uuid)) {
4141                         if (trans)
4142                                 goto update_tree;
4143
4144                         btrfs_release_path(path);
4145                         /*
4146                          * 1 - subvol uuid item
4147                          * 1 - received_subvol uuid item
4148                          */
4149                         trans = btrfs_start_transaction(fs_info->uuid_root, 2);
4150                         if (IS_ERR(trans)) {
4151                                 ret = PTR_ERR(trans);
4152                                 break;
4153                         }
4154                         continue;
4155                 } else {
4156                         goto skip;
4157                 }
4158 update_tree:
4159                 if (!btrfs_is_empty_uuid(root_item.uuid)) {
4160                         ret = btrfs_uuid_tree_add(trans, fs_info,
4161                                                   root_item.uuid,
4162                                                   BTRFS_UUID_KEY_SUBVOL,
4163                                                   key.objectid);
4164                         if (ret < 0) {
4165                                 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4166                                         ret);
4167                                 break;
4168                         }
4169                 }
4170
4171                 if (!btrfs_is_empty_uuid(root_item.received_uuid)) {
4172                         ret = btrfs_uuid_tree_add(trans, fs_info,
4173                                                   root_item.received_uuid,
4174                                                  BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4175                                                   key.objectid);
4176                         if (ret < 0) {
4177                                 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4178                                         ret);
4179                                 break;
4180                         }
4181                 }
4182
4183 skip:
4184                 if (trans) {
4185                         ret = btrfs_end_transaction(trans);
4186                         trans = NULL;
4187                         if (ret)
4188                                 break;
4189                 }
4190
4191                 btrfs_release_path(path);
4192                 if (key.offset < (u64)-1) {
4193                         key.offset++;
4194                 } else if (key.type < BTRFS_ROOT_ITEM_KEY) {
4195                         key.offset = 0;
4196                         key.type = BTRFS_ROOT_ITEM_KEY;
4197                 } else if (key.objectid < (u64)-1) {
4198                         key.offset = 0;
4199                         key.type = BTRFS_ROOT_ITEM_KEY;
4200                         key.objectid++;
4201                 } else {
4202                         break;
4203                 }
4204                 cond_resched();
4205         }
4206
4207 out:
4208         btrfs_free_path(path);
4209         if (trans && !IS_ERR(trans))
4210                 btrfs_end_transaction(trans);
4211         if (ret)
4212                 btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret);
4213         else
4214                 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
4215         up(&fs_info->uuid_tree_rescan_sem);
4216         return 0;
4217 }
4218
4219 /*
4220  * Callback for btrfs_uuid_tree_iterate().
4221  * returns:
4222  * 0    check succeeded, the entry is not outdated.
4223  * < 0  if an error occurred.
4224  * > 0  if the check failed, which means the caller shall remove the entry.
4225  */
4226 static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info *fs_info,
4227                                        u8 *uuid, u8 type, u64 subid)
4228 {
4229         struct btrfs_key key;
4230         int ret = 0;
4231         struct btrfs_root *subvol_root;
4232
4233         if (type != BTRFS_UUID_KEY_SUBVOL &&
4234             type != BTRFS_UUID_KEY_RECEIVED_SUBVOL)
4235                 goto out;
4236
4237         key.objectid = subid;
4238         key.type = BTRFS_ROOT_ITEM_KEY;
4239         key.offset = (u64)-1;
4240         subvol_root = btrfs_read_fs_root_no_name(fs_info, &key);
4241         if (IS_ERR(subvol_root)) {
4242                 ret = PTR_ERR(subvol_root);
4243                 if (ret == -ENOENT)
4244                         ret = 1;
4245                 goto out;
4246         }
4247
4248         switch (type) {
4249         case BTRFS_UUID_KEY_SUBVOL:
4250                 if (memcmp(uuid, subvol_root->root_item.uuid, BTRFS_UUID_SIZE))
4251                         ret = 1;
4252                 break;
4253         case BTRFS_UUID_KEY_RECEIVED_SUBVOL:
4254                 if (memcmp(uuid, subvol_root->root_item.received_uuid,
4255                            BTRFS_UUID_SIZE))
4256                         ret = 1;
4257                 break;
4258         }
4259
4260 out:
4261         return ret;
4262 }
4263
4264 static int btrfs_uuid_rescan_kthread(void *data)
4265 {
4266         struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data;
4267         int ret;
4268
4269         /*
4270          * 1st step is to iterate through the existing UUID tree and
4271          * to delete all entries that contain outdated data.
4272          * 2nd step is to add all missing entries to the UUID tree.
4273          */
4274         ret = btrfs_uuid_tree_iterate(fs_info, btrfs_check_uuid_tree_entry);
4275         if (ret < 0) {
4276                 btrfs_warn(fs_info, "iterating uuid_tree failed %d", ret);
4277                 up(&fs_info->uuid_tree_rescan_sem);
4278                 return ret;
4279         }
4280         return btrfs_uuid_scan_kthread(data);
4281 }
4282
4283 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info)
4284 {
4285         struct btrfs_trans_handle *trans;
4286         struct btrfs_root *tree_root = fs_info->tree_root;
4287         struct btrfs_root *uuid_root;
4288         struct task_struct *task;
4289         int ret;
4290
4291         /*
4292          * 1 - root node
4293          * 1 - root item
4294          */
4295         trans = btrfs_start_transaction(tree_root, 2);
4296         if (IS_ERR(trans))
4297                 return PTR_ERR(trans);
4298
4299         uuid_root = btrfs_create_tree(trans, fs_info,
4300                                       BTRFS_UUID_TREE_OBJECTID);
4301         if (IS_ERR(uuid_root)) {
4302                 ret = PTR_ERR(uuid_root);
4303                 btrfs_abort_transaction(trans, ret);
4304                 btrfs_end_transaction(trans);
4305                 return ret;
4306         }
4307
4308         fs_info->uuid_root = uuid_root;
4309
4310         ret = btrfs_commit_transaction(trans);
4311         if (ret)
4312                 return ret;
4313
4314         down(&fs_info->uuid_tree_rescan_sem);
4315         task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid");
4316         if (IS_ERR(task)) {
4317                 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4318                 btrfs_warn(fs_info, "failed to start uuid_scan task");
4319                 up(&fs_info->uuid_tree_rescan_sem);
4320                 return PTR_ERR(task);
4321         }
4322
4323         return 0;
4324 }
4325
4326 int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
4327 {
4328         struct task_struct *task;
4329
4330         down(&fs_info->uuid_tree_rescan_sem);
4331         task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
4332         if (IS_ERR(task)) {
4333                 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4334                 btrfs_warn(fs_info, "failed to start uuid_rescan task");
4335                 up(&fs_info->uuid_tree_rescan_sem);
4336                 return PTR_ERR(task);
4337         }
4338
4339         return 0;
4340 }
4341
4342 /*
4343  * shrinking a device means finding all of the device extents past
4344  * the new size, and then following the back refs to the chunks.
4345  * The chunk relocation code actually frees the device extent
4346  */
4347 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
4348 {
4349         struct btrfs_fs_info *fs_info = device->fs_info;
4350         struct btrfs_root *root = fs_info->dev_root;
4351         struct btrfs_trans_handle *trans;
4352         struct btrfs_dev_extent *dev_extent = NULL;
4353         struct btrfs_path *path;
4354         u64 length;
4355         u64 chunk_offset;
4356         int ret;
4357         int slot;
4358         int failed = 0;
4359         bool retried = false;
4360         bool checked_pending_chunks = false;
4361         struct extent_buffer *l;
4362         struct btrfs_key key;
4363         struct btrfs_super_block *super_copy = fs_info->super_copy;
4364         u64 old_total = btrfs_super_total_bytes(super_copy);
4365         u64 old_size = btrfs_device_get_total_bytes(device);
4366         u64 diff = old_size - new_size;
4367
4368         if (device->is_tgtdev_for_dev_replace)
4369                 return -EINVAL;
4370
4371         path = btrfs_alloc_path();
4372         if (!path)
4373                 return -ENOMEM;
4374
4375         path->reada = READA_FORWARD;
4376
4377         mutex_lock(&fs_info->chunk_mutex);
4378
4379         btrfs_device_set_total_bytes(device, new_size);
4380         if (device->writeable) {
4381                 device->fs_devices->total_rw_bytes -= diff;
4382                 spin_lock(&fs_info->free_chunk_lock);
4383                 fs_info->free_chunk_space -= diff;
4384                 spin_unlock(&fs_info->free_chunk_lock);
4385         }
4386         mutex_unlock(&fs_info->chunk_mutex);
4387
4388 again:
4389         key.objectid = device->devid;
4390         key.offset = (u64)-1;
4391         key.type = BTRFS_DEV_EXTENT_KEY;
4392
4393         do {
4394                 mutex_lock(&fs_info->delete_unused_bgs_mutex);
4395                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4396                 if (ret < 0) {
4397                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4398                         goto done;
4399                 }
4400
4401                 ret = btrfs_previous_item(root, path, 0, key.type);
4402                 if (ret)
4403                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4404                 if (ret < 0)
4405                         goto done;
4406                 if (ret) {
4407                         ret = 0;
4408                         btrfs_release_path(path);
4409                         break;
4410                 }
4411
4412                 l = path->nodes[0];
4413                 slot = path->slots[0];
4414                 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
4415
4416                 if (key.objectid != device->devid) {
4417                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4418                         btrfs_release_path(path);
4419                         break;
4420                 }
4421
4422                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
4423                 length = btrfs_dev_extent_length(l, dev_extent);
4424
4425                 if (key.offset + length <= new_size) {
4426                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4427                         btrfs_release_path(path);
4428                         break;
4429                 }
4430
4431                 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
4432                 btrfs_release_path(path);
4433
4434                 ret = btrfs_relocate_chunk(fs_info, chunk_offset);
4435                 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4436                 if (ret && ret != -ENOSPC)
4437                         goto done;
4438                 if (ret == -ENOSPC)
4439                         failed++;
4440         } while (key.offset-- > 0);
4441
4442         if (failed && !retried) {
4443                 failed = 0;
4444                 retried = true;
4445                 goto again;
4446         } else if (failed && retried) {
4447                 ret = -ENOSPC;
4448                 goto done;
4449         }
4450
4451         /* Shrinking succeeded, else we would be at "done". */
4452         trans = btrfs_start_transaction(root, 0);
4453         if (IS_ERR(trans)) {
4454                 ret = PTR_ERR(trans);
4455                 goto done;
4456         }
4457
4458         mutex_lock(&fs_info->chunk_mutex);
4459
4460         /*
4461          * We checked in the above loop all device extents that were already in
4462          * the device tree. However before we have updated the device's
4463          * total_bytes to the new size, we might have had chunk allocations that
4464          * have not complete yet (new block groups attached to transaction
4465          * handles), and therefore their device extents were not yet in the
4466          * device tree and we missed them in the loop above. So if we have any
4467          * pending chunk using a device extent that overlaps the device range
4468          * that we can not use anymore, commit the current transaction and
4469          * repeat the search on the device tree - this way we guarantee we will
4470          * not have chunks using device extents that end beyond 'new_size'.
4471          */
4472         if (!checked_pending_chunks) {
4473                 u64 start = new_size;
4474                 u64 len = old_size - new_size;
4475
4476                 if (contains_pending_extent(trans->transaction, device,
4477                                             &start, len)) {
4478                         mutex_unlock(&fs_info->chunk_mutex);
4479                         checked_pending_chunks = true;
4480                         failed = 0;
4481                         retried = false;
4482                         ret = btrfs_commit_transaction(trans);
4483                         if (ret)
4484                                 goto done;
4485                         goto again;
4486                 }
4487         }
4488
4489         btrfs_device_set_disk_total_bytes(device, new_size);
4490         if (list_empty(&device->resized_list))
4491                 list_add_tail(&device->resized_list,
4492                               &fs_info->fs_devices->resized_devices);
4493
4494         WARN_ON(diff > old_total);
4495         btrfs_set_super_total_bytes(super_copy, old_total - diff);
4496         mutex_unlock(&fs_info->chunk_mutex);
4497
4498         /* Now btrfs_update_device() will change the on-disk size. */
4499         ret = btrfs_update_device(trans, device);
4500         btrfs_end_transaction(trans);
4501 done:
4502         btrfs_free_path(path);
4503         if (ret) {
4504                 mutex_lock(&fs_info->chunk_mutex);
4505                 btrfs_device_set_total_bytes(device, old_size);
4506                 if (device->writeable)
4507                         device->fs_devices->total_rw_bytes += diff;
4508                 spin_lock(&fs_info->free_chunk_lock);
4509                 fs_info->free_chunk_space += diff;
4510                 spin_unlock(&fs_info->free_chunk_lock);
4511                 mutex_unlock(&fs_info->chunk_mutex);
4512         }
4513         return ret;
4514 }
4515
4516 static int btrfs_add_system_chunk(struct btrfs_fs_info *fs_info,
4517                            struct btrfs_key *key,
4518                            struct btrfs_chunk *chunk, int item_size)
4519 {
4520         struct btrfs_super_block *super_copy = fs_info->super_copy;
4521         struct btrfs_disk_key disk_key;
4522         u32 array_size;
4523         u8 *ptr;
4524
4525         mutex_lock(&fs_info->chunk_mutex);
4526         array_size = btrfs_super_sys_array_size(super_copy);
4527         if (array_size + item_size + sizeof(disk_key)
4528                         > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
4529                 mutex_unlock(&fs_info->chunk_mutex);
4530                 return -EFBIG;
4531         }
4532
4533         ptr = super_copy->sys_chunk_array + array_size;
4534         btrfs_cpu_key_to_disk(&disk_key, key);
4535         memcpy(ptr, &disk_key, sizeof(disk_key));
4536         ptr += sizeof(disk_key);
4537         memcpy(ptr, chunk, item_size);
4538         item_size += sizeof(disk_key);
4539         btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
4540         mutex_unlock(&fs_info->chunk_mutex);
4541
4542         return 0;
4543 }
4544
4545 /*
4546  * sort the devices in descending order by max_avail, total_avail
4547  */
4548 static int btrfs_cmp_device_info(const void *a, const void *b)
4549 {
4550         const struct btrfs_device_info *di_a = a;
4551         const struct btrfs_device_info *di_b = b;
4552
4553         if (di_a->max_avail > di_b->max_avail)
4554                 return -1;
4555         if (di_a->max_avail < di_b->max_avail)
4556                 return 1;
4557         if (di_a->total_avail > di_b->total_avail)
4558                 return -1;
4559         if (di_a->total_avail < di_b->total_avail)
4560                 return 1;
4561         return 0;
4562 }
4563
4564 static u32 find_raid56_stripe_len(u32 data_devices, u32 dev_stripe_target)
4565 {
4566         /* TODO allow them to set a preferred stripe size */
4567         return SZ_64K;
4568 }
4569
4570 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
4571 {
4572         if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK))
4573                 return;
4574
4575         btrfs_set_fs_incompat(info, RAID56);
4576 }
4577
4578 #define BTRFS_MAX_DEVS(r) ((BTRFS_MAX_ITEM_SIZE(r->fs_info)             \
4579                         - sizeof(struct btrfs_chunk))           \
4580                         / sizeof(struct btrfs_stripe) + 1)
4581
4582 #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE        \
4583                                 - 2 * sizeof(struct btrfs_disk_key)     \
4584                                 - 2 * sizeof(struct btrfs_chunk))       \
4585                                 / sizeof(struct btrfs_stripe) + 1)
4586
4587 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4588                                u64 start, u64 type)
4589 {
4590         struct btrfs_fs_info *info = trans->fs_info;
4591         struct btrfs_fs_devices *fs_devices = info->fs_devices;
4592         struct list_head *cur;
4593         struct map_lookup *map = NULL;
4594         struct extent_map_tree *em_tree;
4595         struct extent_map *em;
4596         struct btrfs_device_info *devices_info = NULL;
4597         u64 total_avail;
4598         int num_stripes;        /* total number of stripes to allocate */
4599         int data_stripes;       /* number of stripes that count for
4600                                    block group size */
4601         int sub_stripes;        /* sub_stripes info for map */
4602         int dev_stripes;        /* stripes per dev */
4603         int devs_max;           /* max devs to use */
4604         int devs_min;           /* min devs needed */
4605         int devs_increment;     /* ndevs has to be a multiple of this */
4606         int ncopies;            /* how many copies to data has */
4607         int ret;
4608         u64 max_stripe_size;
4609         u64 max_chunk_size;
4610         u64 stripe_size;
4611         u64 num_bytes;
4612         u64 raid_stripe_len = BTRFS_STRIPE_LEN;
4613         int ndevs;
4614         int i;
4615         int j;
4616         int index;
4617
4618         BUG_ON(!alloc_profile_is_valid(type, 0));
4619
4620         if (list_empty(&fs_devices->alloc_list))
4621                 return -ENOSPC;
4622
4623         index = __get_raid_index(type);
4624
4625         sub_stripes = btrfs_raid_array[index].sub_stripes;
4626         dev_stripes = btrfs_raid_array[index].dev_stripes;
4627         devs_max = btrfs_raid_array[index].devs_max;
4628         devs_min = btrfs_raid_array[index].devs_min;
4629         devs_increment = btrfs_raid_array[index].devs_increment;
4630         ncopies = btrfs_raid_array[index].ncopies;
4631
4632         if (type & BTRFS_BLOCK_GROUP_DATA) {
4633                 max_stripe_size = SZ_1G;
4634                 max_chunk_size = 10 * max_stripe_size;
4635                 if (!devs_max)
4636                         devs_max = BTRFS_MAX_DEVS(info->chunk_root);
4637         } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
4638                 /* for larger filesystems, use larger metadata chunks */
4639                 if (fs_devices->total_rw_bytes > 50ULL * SZ_1G)
4640                         max_stripe_size = SZ_1G;
4641                 else
4642                         max_stripe_size = SZ_256M;
4643                 max_chunk_size = max_stripe_size;
4644                 if (!devs_max)
4645                         devs_max = BTRFS_MAX_DEVS(info->chunk_root);
4646         } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
4647                 max_stripe_size = SZ_32M;
4648                 max_chunk_size = 2 * max_stripe_size;
4649                 if (!devs_max)
4650                         devs_max = BTRFS_MAX_DEVS_SYS_CHUNK;
4651         } else {
4652                 btrfs_err(info, "invalid chunk type 0x%llx requested",
4653                        type);
4654                 BUG_ON(1);
4655         }
4656
4657         /* we don't want a chunk larger than 10% of writeable space */
4658         max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
4659                              max_chunk_size);
4660
4661         devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info),
4662                                GFP_NOFS);
4663         if (!devices_info)
4664                 return -ENOMEM;
4665
4666         cur = fs_devices->alloc_list.next;
4667
4668         /*
4669          * in the first pass through the devices list, we gather information
4670          * about the available holes on each device.
4671          */
4672         ndevs = 0;
4673         while (cur != &fs_devices->alloc_list) {
4674                 struct btrfs_device *device;
4675                 u64 max_avail;
4676                 u64 dev_offset;
4677
4678                 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
4679
4680                 cur = cur->next;
4681
4682                 if (!device->writeable) {
4683                         WARN(1, KERN_ERR
4684                                "BTRFS: read-only device in alloc_list\n");
4685                         continue;
4686                 }
4687
4688                 if (!device->in_fs_metadata ||
4689                     device->is_tgtdev_for_dev_replace)
4690                         continue;
4691
4692                 if (device->total_bytes > device->bytes_used)
4693                         total_avail = device->total_bytes - device->bytes_used;
4694                 else
4695                         total_avail = 0;
4696
4697                 /* If there is no space on this device, skip it. */
4698                 if (total_avail == 0)
4699                         continue;
4700
4701                 ret = find_free_dev_extent(trans, device,
4702                                            max_stripe_size * dev_stripes,
4703                                            &dev_offset, &max_avail);
4704                 if (ret && ret != -ENOSPC)
4705                         goto error;
4706
4707                 if (ret == 0)
4708                         max_avail = max_stripe_size * dev_stripes;
4709
4710                 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes)
4711                         continue;
4712
4713                 if (ndevs == fs_devices->rw_devices) {
4714                         WARN(1, "%s: found more than %llu devices\n",
4715                              __func__, fs_devices->rw_devices);
4716                         break;
4717                 }
4718                 devices_info[ndevs].dev_offset = dev_offset;
4719                 devices_info[ndevs].max_avail = max_avail;
4720                 devices_info[ndevs].total_avail = total_avail;
4721                 devices_info[ndevs].dev = device;
4722                 ++ndevs;
4723         }
4724
4725         /*
4726          * now sort the devices by hole size / available space
4727          */
4728         sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
4729              btrfs_cmp_device_info, NULL);
4730
4731         /* round down to number of usable stripes */
4732         ndevs -= ndevs % devs_increment;
4733
4734         if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) {
4735                 ret = -ENOSPC;
4736                 goto error;
4737         }
4738
4739         if (devs_max && ndevs > devs_max)
4740                 ndevs = devs_max;
4741         /*
4742          * the primary goal is to maximize the number of stripes, so use as many
4743          * devices as possible, even if the stripes are not maximum sized.
4744          */
4745         stripe_size = devices_info[ndevs-1].max_avail;
4746         num_stripes = ndevs * dev_stripes;
4747
4748         /*
4749          * this will have to be fixed for RAID1 and RAID10 over
4750          * more drives
4751          */
4752         data_stripes = num_stripes / ncopies;
4753
4754         if (type & BTRFS_BLOCK_GROUP_RAID5) {
4755                 raid_stripe_len = find_raid56_stripe_len(ndevs - 1,
4756                                                          info->stripesize);
4757                 data_stripes = num_stripes - 1;
4758         }
4759         if (type & BTRFS_BLOCK_GROUP_RAID6) {
4760                 raid_stripe_len = find_raid56_stripe_len(ndevs - 2,
4761                                                          info->stripesize);
4762                 data_stripes = num_stripes - 2;
4763         }
4764
4765         /*
4766          * Use the number of data stripes to figure out how big this chunk
4767          * is really going to be in terms of logical address space,
4768          * and compare that answer with the max chunk size
4769          */
4770         if (stripe_size * data_stripes > max_chunk_size) {
4771                 u64 mask = (1ULL << 24) - 1;
4772
4773                 stripe_size = div_u64(max_chunk_size, data_stripes);
4774
4775                 /* bump the answer up to a 16MB boundary */
4776                 stripe_size = (stripe_size + mask) & ~mask;
4777
4778                 /* but don't go higher than the limits we found
4779                  * while searching for free extents
4780                  */
4781                 if (stripe_size > devices_info[ndevs-1].max_avail)
4782                         stripe_size = devices_info[ndevs-1].max_avail;
4783         }
4784
4785         stripe_size = div_u64(stripe_size, dev_stripes);
4786
4787         /* align to BTRFS_STRIPE_LEN */
4788         stripe_size = div_u64(stripe_size, raid_stripe_len);
4789         stripe_size *= raid_stripe_len;
4790
4791         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
4792         if (!map) {
4793                 ret = -ENOMEM;
4794                 goto error;
4795         }
4796         map->num_stripes = num_stripes;
4797
4798         for (i = 0; i < ndevs; ++i) {
4799                 for (j = 0; j < dev_stripes; ++j) {
4800                         int s = i * dev_stripes + j;
4801                         map->stripes[s].dev = devices_info[i].dev;
4802                         map->stripes[s].physical = devices_info[i].dev_offset +
4803                                                    j * stripe_size;
4804                 }
4805         }
4806         map->sector_size = info->sectorsize;
4807         map->stripe_len = raid_stripe_len;
4808         map->io_align = raid_stripe_len;
4809         map->io_width = raid_stripe_len;
4810         map->type = type;
4811         map->sub_stripes = sub_stripes;
4812
4813         num_bytes = stripe_size * data_stripes;
4814
4815         trace_btrfs_chunk_alloc(info, map, start, num_bytes);
4816
4817         em = alloc_extent_map();
4818         if (!em) {
4819                 kfree(map);
4820                 ret = -ENOMEM;
4821                 goto error;
4822         }
4823         set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
4824         em->map_lookup = map;
4825         em->start = start;
4826         em->len = num_bytes;
4827         em->block_start = 0;
4828         em->block_len = em->len;
4829         em->orig_block_len = stripe_size;
4830
4831         em_tree = &info->mapping_tree.map_tree;
4832         write_lock(&em_tree->lock);
4833         ret = add_extent_mapping(em_tree, em, 0);
4834         if (!ret) {
4835                 list_add_tail(&em->list, &trans->transaction->pending_chunks);
4836                 atomic_inc(&em->refs);
4837         }
4838         write_unlock(&em_tree->lock);
4839         if (ret) {
4840                 free_extent_map(em);
4841                 goto error;
4842         }
4843
4844         ret = btrfs_make_block_group(trans, info, 0, type,
4845                                      BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4846                                      start, num_bytes);
4847         if (ret)
4848                 goto error_del_extent;
4849
4850         for (i = 0; i < map->num_stripes; i++) {
4851                 num_bytes = map->stripes[i].dev->bytes_used + stripe_size;
4852                 btrfs_device_set_bytes_used(map->stripes[i].dev, num_bytes);
4853         }
4854
4855         spin_lock(&info->free_chunk_lock);
4856         info->free_chunk_space -= (stripe_size * map->num_stripes);
4857         spin_unlock(&info->free_chunk_lock);
4858
4859         free_extent_map(em);
4860         check_raid56_incompat_flag(info, type);
4861
4862         kfree(devices_info);
4863         return 0;
4864
4865 error_del_extent:
4866         write_lock(&em_tree->lock);
4867         remove_extent_mapping(em_tree, em);
4868         write_unlock(&em_tree->lock);
4869
4870         /* One for our allocation */
4871         free_extent_map(em);
4872         /* One for the tree reference */
4873         free_extent_map(em);
4874         /* One for the pending_chunks list reference */
4875         free_extent_map(em);
4876 error:
4877         kfree(devices_info);
4878         return ret;
4879 }
4880
4881 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans,
4882                                 struct btrfs_fs_info *fs_info,
4883                                 u64 chunk_offset, u64 chunk_size)
4884 {
4885         struct btrfs_root *extent_root = fs_info->extent_root;
4886         struct btrfs_root *chunk_root = fs_info->chunk_root;
4887         struct btrfs_key key;
4888         struct btrfs_device *device;
4889         struct btrfs_chunk *chunk;
4890         struct btrfs_stripe *stripe;
4891         struct extent_map_tree *em_tree;
4892         struct extent_map *em;
4893         struct map_lookup *map;
4894         size_t item_size;
4895         u64 dev_offset;
4896         u64 stripe_size;
4897         int i = 0;
4898         int ret = 0;
4899
4900         em_tree = &fs_info->mapping_tree.map_tree;
4901         read_lock(&em_tree->lock);
4902         em = lookup_extent_mapping(em_tree, chunk_offset, chunk_size);
4903         read_unlock(&em_tree->lock);
4904
4905         if (!em) {
4906                 btrfs_crit(fs_info, "unable to find logical %Lu len %Lu",
4907                            chunk_offset, chunk_size);
4908                 return -EINVAL;
4909         }
4910
4911         if (em->start != chunk_offset || em->len != chunk_size) {
4912                 btrfs_crit(fs_info,
4913                            "found a bad mapping, wanted %Lu-%Lu, found %Lu-%Lu",
4914                             chunk_offset, chunk_size, em->start, em->len);
4915                 free_extent_map(em);
4916                 return -EINVAL;
4917         }
4918
4919         map = em->map_lookup;
4920         item_size = btrfs_chunk_item_size(map->num_stripes);
4921         stripe_size = em->orig_block_len;
4922
4923         chunk = kzalloc(item_size, GFP_NOFS);
4924         if (!chunk) {
4925                 ret = -ENOMEM;
4926                 goto out;
4927         }
4928
4929         /*
4930          * Take the device list mutex to prevent races with the final phase of
4931          * a device replace operation that replaces the device object associated
4932          * with the map's stripes, because the device object's id can change
4933          * at any time during that final phase of the device replace operation
4934          * (dev-replace.c:btrfs_dev_replace_finishing()).
4935          */
4936         mutex_lock(&fs_info->fs_devices->device_list_mutex);
4937         for (i = 0; i < map->num_stripes; i++) {
4938                 device = map->stripes[i].dev;
4939                 dev_offset = map->stripes[i].physical;
4940
4941                 ret = btrfs_update_device(trans, device);
4942                 if (ret)
4943                         break;
4944                 ret = btrfs_alloc_dev_extent(trans, device,
4945                                              chunk_root->root_key.objectid,
4946                                              BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4947                                              chunk_offset, dev_offset,
4948                                              stripe_size);
4949                 if (ret)
4950                         break;
4951         }
4952         if (ret) {
4953                 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4954                 goto out;
4955         }
4956
4957         stripe = &chunk->stripe;
4958         for (i = 0; i < map->num_stripes; i++) {
4959                 device = map->stripes[i].dev;
4960                 dev_offset = map->stripes[i].physical;
4961
4962                 btrfs_set_stack_stripe_devid(stripe, device->devid);
4963                 btrfs_set_stack_stripe_offset(stripe, dev_offset);
4964                 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
4965                 stripe++;
4966         }
4967         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4968
4969         btrfs_set_stack_chunk_length(chunk, chunk_size);
4970         btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
4971         btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
4972         btrfs_set_stack_chunk_type(chunk, map->type);
4973         btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
4974         btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
4975         btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
4976         btrfs_set_stack_chunk_sector_size(chunk, fs_info->sectorsize);
4977         btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
4978
4979         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
4980         key.type = BTRFS_CHUNK_ITEM_KEY;
4981         key.offset = chunk_offset;
4982
4983         ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
4984         if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
4985                 /*
4986                  * TODO: Cleanup of inserted chunk root in case of
4987                  * failure.
4988                  */
4989                 ret = btrfs_add_system_chunk(fs_info, &key, chunk, item_size);
4990         }
4991
4992 out:
4993         kfree(chunk);
4994         free_extent_map(em);
4995         return ret;
4996 }
4997
4998 /*
4999  * Chunk allocation falls into two parts. The first part does works
5000  * that make the new allocated chunk useable, but not do any operation
5001  * that modifies the chunk tree. The second part does the works that
5002  * require modifying the chunk tree. This division is important for the
5003  * bootstrap process of adding storage to a seed btrfs.
5004  */
5005 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
5006                       struct btrfs_fs_info *fs_info, u64 type)
5007 {
5008         u64 chunk_offset;
5009
5010         ASSERT(mutex_is_locked(&fs_info->chunk_mutex));
5011         chunk_offset = find_next_chunk(fs_info);
5012         return __btrfs_alloc_chunk(trans, chunk_offset, type);
5013 }
5014
5015 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
5016                                          struct btrfs_fs_info *fs_info)
5017 {
5018         struct btrfs_root *extent_root = fs_info->extent_root;
5019         u64 chunk_offset;
5020         u64 sys_chunk_offset;
5021         u64 alloc_profile;
5022         int ret;
5023
5024         chunk_offset = find_next_chunk(fs_info);
5025         alloc_profile = btrfs_get_alloc_profile(extent_root, 0);
5026         ret = __btrfs_alloc_chunk(trans, chunk_offset, alloc_profile);
5027         if (ret)
5028                 return ret;
5029
5030         sys_chunk_offset = find_next_chunk(fs_info);
5031         alloc_profile = btrfs_get_alloc_profile(fs_info->chunk_root, 0);
5032         ret = __btrfs_alloc_chunk(trans, sys_chunk_offset, alloc_profile);
5033         return ret;
5034 }
5035
5036 static inline int btrfs_chunk_max_errors(struct map_lookup *map)
5037 {
5038         int max_errors;
5039
5040         if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
5041                          BTRFS_BLOCK_GROUP_RAID10 |
5042                          BTRFS_BLOCK_GROUP_RAID5 |
5043                          BTRFS_BLOCK_GROUP_DUP)) {
5044                 max_errors = 1;
5045         } else if (map->type & BTRFS_BLOCK_GROUP_RAID6) {
5046                 max_errors = 2;
5047         } else {
5048                 max_errors = 0;
5049         }
5050
5051         return max_errors;
5052 }
5053
5054 int btrfs_chunk_readonly(struct btrfs_fs_info *fs_info, u64 chunk_offset)
5055 {
5056         struct extent_map *em;
5057         struct map_lookup *map;
5058         struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
5059         int readonly = 0;
5060         int miss_ndevs = 0;
5061         int i;
5062
5063         read_lock(&map_tree->map_tree.lock);
5064         em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
5065         read_unlock(&map_tree->map_tree.lock);
5066         if (!em)
5067                 return 1;
5068
5069         map = em->map_lookup;
5070         for (i = 0; i < map->num_stripes; i++) {
5071                 if (map->stripes[i].dev->missing) {
5072                         miss_ndevs++;
5073                         continue;
5074                 }
5075
5076                 if (!map->stripes[i].dev->writeable) {
5077                         readonly = 1;
5078                         goto end;
5079                 }
5080         }
5081
5082         /*
5083          * If the number of missing devices is larger than max errors,
5084          * we can not write the data into that chunk successfully, so
5085          * set it readonly.
5086          */
5087         if (miss_ndevs > btrfs_chunk_max_errors(map))
5088                 readonly = 1;
5089 end:
5090         free_extent_map(em);
5091         return readonly;
5092 }
5093
5094 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
5095 {
5096         extent_map_tree_init(&tree->map_tree);
5097 }
5098
5099 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
5100 {
5101         struct extent_map *em;
5102
5103         while (1) {
5104                 write_lock(&tree->map_tree.lock);
5105                 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
5106                 if (em)
5107                         remove_extent_mapping(&tree->map_tree, em);
5108                 write_unlock(&tree->map_tree.lock);
5109                 if (!em)
5110                         break;
5111                 /* once for us */
5112                 free_extent_map(em);
5113                 /* once for the tree */
5114                 free_extent_map(em);
5115         }
5116 }
5117
5118 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5119 {
5120         struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
5121         struct extent_map *em;
5122         struct map_lookup *map;
5123         struct extent_map_tree *em_tree = &map_tree->map_tree;
5124         int ret;
5125
5126         read_lock(&em_tree->lock);
5127         em = lookup_extent_mapping(em_tree, logical, len);
5128         read_unlock(&em_tree->lock);
5129
5130         /*
5131          * We could return errors for these cases, but that could get ugly and
5132          * we'd probably do the same thing which is just not do anything else
5133          * and exit, so return 1 so the callers don't try to use other copies.
5134          */
5135         if (!em) {
5136                 btrfs_crit(fs_info, "No mapping for %Lu-%Lu", logical,
5137                             logical+len);
5138                 return 1;
5139         }
5140
5141         if (em->start > logical || em->start + em->len < logical) {
5142                 btrfs_crit(fs_info, "Invalid mapping for %Lu-%Lu, got %Lu-%Lu",
5143                            logical, logical+len, em->start,
5144                            em->start + em->len);
5145                 free_extent_map(em);
5146                 return 1;
5147         }
5148
5149         map = em->map_lookup;
5150         if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
5151                 ret = map->num_stripes;
5152         else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5153                 ret = map->sub_stripes;
5154         else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
5155                 ret = 2;
5156         else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5157                 ret = 3;
5158         else
5159                 ret = 1;
5160         free_extent_map(em);
5161
5162         btrfs_dev_replace_lock(&fs_info->dev_replace, 0);
5163         if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace))
5164                 ret++;
5165         btrfs_dev_replace_unlock(&fs_info->dev_replace, 0);
5166
5167         return ret;
5168 }
5169
5170 unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info,
5171                                     struct btrfs_mapping_tree *map_tree,
5172                                     u64 logical)
5173 {
5174         struct extent_map *em;
5175         struct map_lookup *map;
5176         struct extent_map_tree *em_tree = &map_tree->map_tree;
5177         unsigned long len = fs_info->sectorsize;
5178
5179         read_lock(&em_tree->lock);
5180         em = lookup_extent_mapping(em_tree, logical, len);
5181         read_unlock(&em_tree->lock);
5182         BUG_ON(!em);
5183
5184         BUG_ON(em->start > logical || em->start + em->len < logical);
5185         map = em->map_lookup;
5186         if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5187                 len = map->stripe_len * nr_data_stripes(map);
5188         free_extent_map(em);
5189         return len;
5190 }
5191
5192 int btrfs_is_parity_mirror(struct btrfs_mapping_tree *map_tree,
5193                            u64 logical, u64 len, int mirror_num)
5194 {
5195         struct extent_map *em;
5196         struct map_lookup *map;
5197         struct extent_map_tree *em_tree = &map_tree->map_tree;
5198         int ret = 0;
5199
5200         read_lock(&em_tree->lock);
5201         em = lookup_extent_mapping(em_tree, logical, len);
5202         read_unlock(&em_tree->lock);
5203         BUG_ON(!em);
5204
5205         BUG_ON(em->start > logical || em->start + em->len < logical);
5206         map = em->map_lookup;
5207         if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5208                 ret = 1;
5209         free_extent_map(em);
5210         return ret;
5211 }
5212
5213 static int find_live_mirror(struct btrfs_fs_info *fs_info,
5214                             struct map_lookup *map, int first, int num,
5215                             int optimal, int dev_replace_is_ongoing)
5216 {
5217         int i;
5218         int tolerance;
5219         struct btrfs_device *srcdev;
5220
5221         if (dev_replace_is_ongoing &&
5222             fs_info->dev_replace.cont_reading_from_srcdev_mode ==
5223              BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
5224                 srcdev = fs_info->dev_replace.srcdev;
5225         else
5226                 srcdev = NULL;
5227
5228         /*
5229          * try to avoid the drive that is the source drive for a
5230          * dev-replace procedure, only choose it if no other non-missing
5231          * mirror is available
5232          */
5233         for (tolerance = 0; tolerance < 2; tolerance++) {
5234                 if (map->stripes[optimal].dev->bdev &&
5235                     (tolerance || map->stripes[optimal].dev != srcdev))
5236                         return optimal;
5237                 for (i = first; i < first + num; i++) {
5238                         if (map->stripes[i].dev->bdev &&
5239                             (tolerance || map->stripes[i].dev != srcdev))
5240                                 return i;
5241                 }
5242         }
5243
5244         /* we couldn't find one that doesn't fail.  Just return something
5245          * and the io error handling code will clean up eventually
5246          */
5247         return optimal;
5248 }
5249
5250 static inline int parity_smaller(u64 a, u64 b)
5251 {
5252         return a > b;
5253 }
5254
5255 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
5256 static void sort_parity_stripes(struct btrfs_bio *bbio, int num_stripes)
5257 {
5258         struct btrfs_bio_stripe s;
5259         int i;
5260         u64 l;
5261         int again = 1;
5262
5263         while (again) {
5264                 again = 0;
5265                 for (i = 0; i < num_stripes - 1; i++) {
5266                         if (parity_smaller(bbio->raid_map[i],
5267                                            bbio->raid_map[i+1])) {
5268                                 s = bbio->stripes[i];
5269                                 l = bbio->raid_map[i];
5270                                 bbio->stripes[i] = bbio->stripes[i+1];
5271                                 bbio->raid_map[i] = bbio->raid_map[i+1];
5272                                 bbio->stripes[i+1] = s;
5273                                 bbio->raid_map[i+1] = l;
5274
5275                                 again = 1;
5276                         }
5277                 }
5278         }
5279 }
5280
5281 static struct btrfs_bio *alloc_btrfs_bio(int total_stripes, int real_stripes)
5282 {
5283         struct btrfs_bio *bbio = kzalloc(
5284                  /* the size of the btrfs_bio */
5285                 sizeof(struct btrfs_bio) +
5286                 /* plus the variable array for the stripes */
5287                 sizeof(struct btrfs_bio_stripe) * (total_stripes) +
5288                 /* plus the variable array for the tgt dev */
5289                 sizeof(int) * (real_stripes) +
5290                 /*
5291                  * plus the raid_map, which includes both the tgt dev
5292                  * and the stripes
5293                  */
5294                 sizeof(u64) * (total_stripes),
5295                 GFP_NOFS|__GFP_NOFAIL);
5296
5297         atomic_set(&bbio->error, 0);
5298         atomic_set(&bbio->refs, 1);
5299
5300         return bbio;
5301 }
5302
5303 void btrfs_get_bbio(struct btrfs_bio *bbio)
5304 {
5305         WARN_ON(!atomic_read(&bbio->refs));
5306         atomic_inc(&bbio->refs);
5307 }
5308
5309 void btrfs_put_bbio(struct btrfs_bio *bbio)
5310 {
5311         if (!bbio)
5312                 return;
5313         if (atomic_dec_and_test(&bbio->refs))
5314                 kfree(bbio);
5315 }
5316
5317 static int __btrfs_map_block(struct btrfs_fs_info *fs_info,
5318                              enum btrfs_map_op op,
5319                              u64 logical, u64 *length,
5320                              struct btrfs_bio **bbio_ret,
5321                              int mirror_num, int need_raid_map)
5322 {
5323         struct extent_map *em;
5324         struct map_lookup *map;
5325         struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
5326         struct extent_map_tree *em_tree = &map_tree->map_tree;
5327         u64 offset;
5328         u64 stripe_offset;
5329         u64 stripe_end_offset;
5330         u64 stripe_nr;
5331         u64 stripe_nr_orig;
5332         u64 stripe_nr_end;
5333         u64 stripe_len;
5334         u32 stripe_index;
5335         int i;
5336         int ret = 0;
5337         int num_stripes;
5338         int max_errors = 0;
5339         int tgtdev_indexes = 0;
5340         struct btrfs_bio *bbio = NULL;
5341         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
5342         int dev_replace_is_ongoing = 0;
5343         int num_alloc_stripes;
5344         int patch_the_first_stripe_for_dev_replace = 0;
5345         u64 physical_to_patch_in_first_stripe = 0;
5346         u64 raid56_full_stripe_start = (u64)-1;
5347
5348         read_lock(&em_tree->lock);
5349         em = lookup_extent_mapping(em_tree, logical, *length);
5350         read_unlock(&em_tree->lock);
5351
5352         if (!em) {
5353                 btrfs_crit(fs_info, "unable to find logical %llu len %llu",
5354                         logical, *length);
5355                 return -EINVAL;
5356         }
5357
5358         if (em->start > logical || em->start + em->len < logical) {
5359                 btrfs_crit(fs_info,
5360                            "found a bad mapping, wanted %Lu, found %Lu-%Lu",
5361                            logical, em->start, em->start + em->len);
5362                 free_extent_map(em);
5363                 return -EINVAL;
5364         }
5365
5366         map = em->map_lookup;
5367         offset = logical - em->start;
5368
5369         stripe_len = map->stripe_len;
5370         stripe_nr = offset;
5371         /*
5372          * stripe_nr counts the total number of stripes we have to stride
5373          * to get to this block
5374          */
5375         stripe_nr = div64_u64(stripe_nr, stripe_len);
5376
5377         stripe_offset = stripe_nr * stripe_len;
5378         if (offset < stripe_offset) {
5379                 btrfs_crit(fs_info,
5380                            "stripe math has gone wrong, stripe_offset=%llu, offset=%llu, start=%llu, logical=%llu, stripe_len=%llu",
5381                            stripe_offset, offset, em->start, logical,
5382                            stripe_len);
5383                 free_extent_map(em);
5384                 return -EINVAL;
5385         }
5386
5387         /* stripe_offset is the offset of this block in its stripe*/
5388         stripe_offset = offset - stripe_offset;
5389
5390         /* if we're here for raid56, we need to know the stripe aligned start */
5391         if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5392                 unsigned long full_stripe_len = stripe_len * nr_data_stripes(map);
5393                 raid56_full_stripe_start = offset;
5394
5395                 /* allow a write of a full stripe, but make sure we don't
5396                  * allow straddling of stripes
5397                  */
5398                 raid56_full_stripe_start = div64_u64(raid56_full_stripe_start,
5399                                 full_stripe_len);
5400                 raid56_full_stripe_start *= full_stripe_len;
5401         }
5402
5403         if (op == BTRFS_MAP_DISCARD) {
5404                 /* we don't discard raid56 yet */
5405                 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5406                         ret = -EOPNOTSUPP;
5407                         goto out;
5408                 }
5409                 *length = min_t(u64, em->len - offset, *length);
5410         } else if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
5411                 u64 max_len;
5412                 /* For writes to RAID[56], allow a full stripeset across all disks.
5413                    For other RAID types and for RAID[56] reads, just allow a single
5414                    stripe (on a single disk). */
5415                 if ((map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
5416                     (op == BTRFS_MAP_WRITE)) {
5417                         max_len = stripe_len * nr_data_stripes(map) -
5418                                 (offset - raid56_full_stripe_start);
5419                 } else {
5420                         /* we limit the length of each bio to what fits in a stripe */
5421                         max_len = stripe_len - stripe_offset;
5422                 }
5423                 *length = min_t(u64, em->len - offset, max_len);
5424         } else {
5425                 *length = em->len - offset;
5426         }
5427
5428         /* This is for when we're called from btrfs_merge_bio_hook() and all
5429            it cares about is the length */
5430         if (!bbio_ret)
5431                 goto out;
5432
5433         btrfs_dev_replace_lock(dev_replace, 0);
5434         dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
5435         if (!dev_replace_is_ongoing)
5436                 btrfs_dev_replace_unlock(dev_replace, 0);
5437         else
5438                 btrfs_dev_replace_set_lock_blocking(dev_replace);
5439
5440         if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
5441             op != BTRFS_MAP_WRITE && op != BTRFS_MAP_DISCARD &&
5442             op != BTRFS_MAP_GET_READ_MIRRORS && dev_replace->tgtdev != NULL) {
5443                 /*
5444                  * in dev-replace case, for repair case (that's the only
5445                  * case where the mirror is selected explicitly when
5446                  * calling btrfs_map_block), blocks left of the left cursor
5447                  * can also be read from the target drive.
5448                  * For REQ_GET_READ_MIRRORS, the target drive is added as
5449                  * the last one to the array of stripes. For READ, it also
5450                  * needs to be supported using the same mirror number.
5451                  * If the requested block is not left of the left cursor,
5452                  * EIO is returned. This can happen because btrfs_num_copies()
5453                  * returns one more in the dev-replace case.
5454                  */
5455                 u64 tmp_length = *length;
5456                 struct btrfs_bio *tmp_bbio = NULL;
5457                 int tmp_num_stripes;
5458                 u64 srcdev_devid = dev_replace->srcdev->devid;
5459                 int index_srcdev = 0;
5460                 int found = 0;
5461                 u64 physical_of_found = 0;
5462
5463                 ret = __btrfs_map_block(fs_info, BTRFS_MAP_GET_READ_MIRRORS,
5464                              logical, &tmp_length, &tmp_bbio, 0, 0);
5465                 if (ret) {
5466                         WARN_ON(tmp_bbio != NULL);
5467                         goto out;
5468                 }
5469
5470                 tmp_num_stripes = tmp_bbio->num_stripes;
5471                 if (mirror_num > tmp_num_stripes) {
5472                         /*
5473                          * BTRFS_MAP_GET_READ_MIRRORS does not contain this
5474                          * mirror, that means that the requested area
5475                          * is not left of the left cursor
5476                          */
5477                         ret = -EIO;
5478                         btrfs_put_bbio(tmp_bbio);
5479                         goto out;
5480                 }
5481
5482                 /*
5483                  * process the rest of the function using the mirror_num
5484                  * of the source drive. Therefore look it up first.
5485                  * At the end, patch the device pointer to the one of the
5486                  * target drive.
5487                  */
5488                 for (i = 0; i < tmp_num_stripes; i++) {
5489                         if (tmp_bbio->stripes[i].dev->devid != srcdev_devid)
5490                                 continue;
5491
5492                         /*
5493                          * In case of DUP, in order to keep it simple, only add
5494                          * the mirror with the lowest physical address
5495                          */
5496                         if (found &&
5497                             physical_of_found <= tmp_bbio->stripes[i].physical)
5498                                 continue;
5499
5500                         index_srcdev = i;
5501                         found = 1;
5502                         physical_of_found = tmp_bbio->stripes[i].physical;
5503                 }
5504
5505                 btrfs_put_bbio(tmp_bbio);
5506
5507                 if (!found) {
5508                         WARN_ON(1);
5509                         ret = -EIO;
5510                         goto out;
5511                 }
5512
5513                 mirror_num = index_srcdev + 1;
5514                 patch_the_first_stripe_for_dev_replace = 1;
5515                 physical_to_patch_in_first_stripe = physical_of_found;
5516         } else if (mirror_num > map->num_stripes) {
5517                 mirror_num = 0;
5518         }
5519
5520         num_stripes = 1;
5521         stripe_index = 0;
5522         stripe_nr_orig = stripe_nr;
5523         stripe_nr_end = ALIGN(offset + *length, map->stripe_len);
5524         stripe_nr_end = div_u64(stripe_nr_end, map->stripe_len);
5525         stripe_end_offset = stripe_nr_end * map->stripe_len -
5526                             (offset + *length);
5527
5528         if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5529                 if (op == BTRFS_MAP_DISCARD)
5530                         num_stripes = min_t(u64, map->num_stripes,
5531                                             stripe_nr_end - stripe_nr_orig);
5532                 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5533                                 &stripe_index);
5534                 if (op != BTRFS_MAP_WRITE && op != BTRFS_MAP_DISCARD &&
5535                     op != BTRFS_MAP_GET_READ_MIRRORS)
5536                         mirror_num = 1;
5537         } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
5538                 if (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_DISCARD ||
5539                     op == BTRFS_MAP_GET_READ_MIRRORS)
5540                         num_stripes = map->num_stripes;
5541                 else if (mirror_num)
5542                         stripe_index = mirror_num - 1;
5543                 else {
5544                         stripe_index = find_live_mirror(fs_info, map, 0,
5545                                             map->num_stripes,
5546                                             current->pid % map->num_stripes,
5547                                             dev_replace_is_ongoing);
5548                         mirror_num = stripe_index + 1;
5549                 }
5550
5551         } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
5552                 if (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_DISCARD ||
5553                     op == BTRFS_MAP_GET_READ_MIRRORS) {
5554                         num_stripes = map->num_stripes;
5555                 } else if (mirror_num) {
5556                         stripe_index = mirror_num - 1;
5557                 } else {
5558                         mirror_num = 1;
5559                 }
5560
5561         } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5562                 u32 factor = map->num_stripes / map->sub_stripes;
5563
5564                 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
5565                 stripe_index *= map->sub_stripes;
5566
5567                 if (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS)
5568                         num_stripes = map->sub_stripes;
5569                 else if (op == BTRFS_MAP_DISCARD)
5570                         num_stripes = min_t(u64, map->sub_stripes *
5571                                             (stripe_nr_end - stripe_nr_orig),
5572                                             map->num_stripes);
5573                 else if (mirror_num)
5574                         stripe_index += mirror_num - 1;
5575                 else {
5576                         int old_stripe_index = stripe_index;
5577                         stripe_index = find_live_mirror(fs_info, map,
5578                                               stripe_index,
5579                                               map->sub_stripes, stripe_index +
5580                                               current->pid % map->sub_stripes,
5581                                               dev_replace_is_ongoing);
5582                         mirror_num = stripe_index - old_stripe_index + 1;
5583                 }
5584
5585         } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5586                 if (need_raid_map &&
5587                     (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS ||
5588                      mirror_num > 1)) {
5589                         /* push stripe_nr back to the start of the full stripe */
5590                         stripe_nr = div_u64(raid56_full_stripe_start,
5591                                         stripe_len * nr_data_stripes(map));
5592
5593                         /* RAID[56] write or recovery. Return all stripes */
5594                         num_stripes = map->num_stripes;
5595                         max_errors = nr_parity_stripes(map);
5596
5597                         *length = map->stripe_len;
5598                         stripe_index = 0;
5599                         stripe_offset = 0;
5600                 } else {
5601                         /*
5602                          * Mirror #0 or #1 means the original data block.
5603                          * Mirror #2 is RAID5 parity block.
5604                          * Mirror #3 is RAID6 Q block.
5605                          */
5606                         stripe_nr = div_u64_rem(stripe_nr,
5607                                         nr_data_stripes(map), &stripe_index);
5608                         if (mirror_num > 1)
5609                                 stripe_index = nr_data_stripes(map) +
5610                                                 mirror_num - 2;
5611
5612                         /* We distribute the parity blocks across stripes */
5613                         div_u64_rem(stripe_nr + stripe_index, map->num_stripes,
5614                                         &stripe_index);
5615                         if ((op != BTRFS_MAP_WRITE && op != BTRFS_MAP_DISCARD &&
5616                             op != BTRFS_MAP_GET_READ_MIRRORS) && mirror_num <= 1)
5617                                 mirror_num = 1;
5618                 }
5619         } else {
5620                 /*
5621                  * after this, stripe_nr is the number of stripes on this
5622                  * device we have to walk to find the data, and stripe_index is
5623                  * the number of our device in the stripe array
5624                  */
5625                 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5626                                 &stripe_index);
5627                 mirror_num = stripe_index + 1;
5628         }
5629         if (stripe_index >= map->num_stripes) {
5630                 btrfs_crit(fs_info,
5631                            "stripe index math went horribly wrong, got stripe_index=%u, num_stripes=%u",
5632                            stripe_index, map->num_stripes);
5633                 ret = -EINVAL;
5634                 goto out;
5635         }
5636
5637         num_alloc_stripes = num_stripes;
5638         if (dev_replace_is_ongoing) {
5639                 if (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_DISCARD)
5640                         num_alloc_stripes <<= 1;
5641                 if (op == BTRFS_MAP_GET_READ_MIRRORS)
5642                         num_alloc_stripes++;
5643                 tgtdev_indexes = num_stripes;
5644         }
5645
5646         bbio = alloc_btrfs_bio(num_alloc_stripes, tgtdev_indexes);
5647         if (!bbio) {
5648                 ret = -ENOMEM;
5649                 goto out;
5650         }
5651         if (dev_replace_is_ongoing)
5652                 bbio->tgtdev_map = (int *)(bbio->stripes + num_alloc_stripes);
5653
5654         /* build raid_map */
5655         if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK &&
5656             need_raid_map &&
5657             ((op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS) ||
5658             mirror_num > 1)) {
5659                 u64 tmp;
5660                 unsigned rot;
5661
5662                 bbio->raid_map = (u64 *)((void *)bbio->stripes +
5663                                  sizeof(struct btrfs_bio_stripe) *
5664                                  num_alloc_stripes +
5665                                  sizeof(int) * tgtdev_indexes);
5666
5667                 /* Work out the disk rotation on this stripe-set */
5668                 div_u64_rem(stripe_nr, num_stripes, &rot);
5669
5670                 /* Fill in the logical address of each stripe */
5671                 tmp = stripe_nr * nr_data_stripes(map);
5672                 for (i = 0; i < nr_data_stripes(map); i++)
5673                         bbio->raid_map[(i+rot) % num_stripes] =
5674                                 em->start + (tmp + i) * map->stripe_len;
5675
5676                 bbio->raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE;
5677                 if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5678                         bbio->raid_map[(i+rot+1) % num_stripes] =
5679                                 RAID6_Q_STRIPE;
5680         }
5681
5682         if (op == BTRFS_MAP_DISCARD) {
5683                 u32 factor = 0;
5684                 u32 sub_stripes = 0;
5685                 u64 stripes_per_dev = 0;
5686                 u32 remaining_stripes = 0;
5687                 u32 last_stripe = 0;
5688
5689                 if (map->type &
5690                     (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID10)) {
5691                         if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5692                                 sub_stripes = 1;
5693                         else
5694                                 sub_stripes = map->sub_stripes;
5695
5696                         factor = map->num_stripes / sub_stripes;
5697                         stripes_per_dev = div_u64_rem(stripe_nr_end -
5698                                                       stripe_nr_orig,
5699                                                       factor,
5700                                                       &remaining_stripes);
5701                         div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
5702                         last_stripe *= sub_stripes;
5703                 }
5704
5705                 for (i = 0; i < num_stripes; i++) {
5706                         bbio->stripes[i].physical =
5707                                 map->stripes[stripe_index].physical +
5708                                 stripe_offset + stripe_nr * map->stripe_len;
5709                         bbio->stripes[i].dev = map->stripes[stripe_index].dev;
5710
5711                         if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5712                                          BTRFS_BLOCK_GROUP_RAID10)) {
5713                                 bbio->stripes[i].length = stripes_per_dev *
5714                                                           map->stripe_len;
5715
5716                                 if (i / sub_stripes < remaining_stripes)
5717                                         bbio->stripes[i].length +=
5718                                                 map->stripe_len;
5719
5720                                 /*
5721                                  * Special for the first stripe and
5722                                  * the last stripe:
5723                                  *
5724                                  * |-------|...|-------|
5725                                  *     |----------|
5726                                  *    off     end_off
5727                                  */
5728                                 if (i < sub_stripes)
5729                                         bbio->stripes[i].length -=
5730                                                 stripe_offset;
5731
5732                                 if (stripe_index >= last_stripe &&
5733                                     stripe_index <= (last_stripe +
5734                                                      sub_stripes - 1))
5735                                         bbio->stripes[i].length -=
5736                                                 stripe_end_offset;
5737
5738                                 if (i == sub_stripes - 1)
5739                                         stripe_offset = 0;
5740                         } else
5741                                 bbio->stripes[i].length = *length;
5742
5743                         stripe_index++;
5744                         if (stripe_index == map->num_stripes) {
5745                                 /* This could only happen for RAID0/10 */
5746                                 stripe_index = 0;
5747                                 stripe_nr++;
5748                         }
5749                 }
5750         } else {
5751                 for (i = 0; i < num_stripes; i++) {
5752                         bbio->stripes[i].physical =
5753                                 map->stripes[stripe_index].physical +
5754                                 stripe_offset +
5755                                 stripe_nr * map->stripe_len;
5756                         bbio->stripes[i].dev =
5757                                 map->stripes[stripe_index].dev;
5758                         stripe_index++;
5759                 }
5760         }
5761
5762         if (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS)
5763                 max_errors = btrfs_chunk_max_errors(map);
5764
5765         if (bbio->raid_map)
5766                 sort_parity_stripes(bbio, num_stripes);
5767
5768         tgtdev_indexes = 0;
5769         if (dev_replace_is_ongoing &&
5770            (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_DISCARD) &&
5771             dev_replace->tgtdev != NULL) {
5772                 int index_where_to_add;
5773                 u64 srcdev_devid = dev_replace->srcdev->devid;
5774
5775                 /*
5776                  * duplicate the write operations while the dev replace
5777                  * procedure is running. Since the copying of the old disk
5778                  * to the new disk takes place at run time while the
5779                  * filesystem is mounted writable, the regular write
5780                  * operations to the old disk have to be duplicated to go
5781                  * to the new disk as well.
5782                  * Note that device->missing is handled by the caller, and
5783                  * that the write to the old disk is already set up in the
5784                  * stripes array.
5785                  */
5786                 index_where_to_add = num_stripes;
5787                 for (i = 0; i < num_stripes; i++) {
5788                         if (bbio->stripes[i].dev->devid == srcdev_devid) {
5789                                 /* write to new disk, too */
5790                                 struct btrfs_bio_stripe *new =
5791                                         bbio->stripes + index_where_to_add;
5792                                 struct btrfs_bio_stripe *old =
5793                                         bbio->stripes + i;
5794
5795                                 new->physical = old->physical;
5796                                 new->length = old->length;
5797                                 new->dev = dev_replace->tgtdev;
5798                                 bbio->tgtdev_map[i] = index_where_to_add;
5799                                 index_where_to_add++;
5800                                 max_errors++;
5801                                 tgtdev_indexes++;
5802                         }
5803                 }
5804                 num_stripes = index_where_to_add;
5805         } else if (dev_replace_is_ongoing &&
5806                    op == BTRFS_MAP_GET_READ_MIRRORS &&
5807                    dev_replace->tgtdev != NULL) {
5808                 u64 srcdev_devid = dev_replace->srcdev->devid;
5809                 int index_srcdev = 0;
5810                 int found = 0;
5811                 u64 physical_of_found = 0;
5812
5813                 /*
5814                  * During the dev-replace procedure, the target drive can
5815                  * also be used to read data in case it is needed to repair
5816                  * a corrupt block elsewhere. This is possible if the
5817                  * requested area is left of the left cursor. In this area,
5818                  * the target drive is a full copy of the source drive.
5819                  */
5820                 for (i = 0; i < num_stripes; i++) {
5821                         if (bbio->stripes[i].dev->devid == srcdev_devid) {
5822                                 /*
5823                                  * In case of DUP, in order to keep it
5824                                  * simple, only add the mirror with the
5825                                  * lowest physical address
5826                                  */
5827                                 if (found &&
5828                                     physical_of_found <=
5829                                      bbio->stripes[i].physical)
5830                                         continue;
5831                                 index_srcdev = i;
5832                                 found = 1;
5833                                 physical_of_found = bbio->stripes[i].physical;
5834                         }
5835                 }
5836                 if (found) {
5837                         struct btrfs_bio_stripe *tgtdev_stripe =
5838                                 bbio->stripes + num_stripes;
5839
5840                         tgtdev_stripe->physical = physical_of_found;
5841                         tgtdev_stripe->length =
5842                                 bbio->stripes[index_srcdev].length;
5843                         tgtdev_stripe->dev = dev_replace->tgtdev;
5844                         bbio->tgtdev_map[index_srcdev] = num_stripes;
5845
5846                         tgtdev_indexes++;
5847                         num_stripes++;
5848                 }
5849         }
5850
5851         *bbio_ret = bbio;
5852         bbio->map_type = map->type;
5853         bbio->num_stripes = num_stripes;
5854         bbio->max_errors = max_errors;
5855         bbio->mirror_num = mirror_num;
5856         bbio->num_tgtdevs = tgtdev_indexes;
5857
5858         /*
5859          * this is the case that REQ_READ && dev_replace_is_ongoing &&
5860          * mirror_num == num_stripes + 1 && dev_replace target drive is
5861          * available as a mirror
5862          */
5863         if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) {
5864                 WARN_ON(num_stripes > 1);
5865                 bbio->stripes[0].dev = dev_replace->tgtdev;
5866                 bbio->stripes[0].physical = physical_to_patch_in_first_stripe;
5867                 bbio->mirror_num = map->num_stripes + 1;
5868         }
5869 out:
5870         if (dev_replace_is_ongoing) {
5871                 btrfs_dev_replace_clear_lock_blocking(dev_replace);
5872                 btrfs_dev_replace_unlock(dev_replace, 0);
5873         }
5874         free_extent_map(em);
5875         return ret;
5876 }
5877
5878 int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
5879                       u64 logical, u64 *length,
5880                       struct btrfs_bio **bbio_ret, int mirror_num)
5881 {
5882         return __btrfs_map_block(fs_info, op, logical, length, bbio_ret,
5883                                  mirror_num, 0);
5884 }
5885
5886 /* For Scrub/replace */
5887 int btrfs_map_sblock(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
5888                      u64 logical, u64 *length,
5889                      struct btrfs_bio **bbio_ret, int mirror_num,
5890                      int need_raid_map)
5891 {
5892         return __btrfs_map_block(fs_info, op, logical, length, bbio_ret,
5893                                  mirror_num, need_raid_map);
5894 }
5895
5896 int btrfs_rmap_block(struct btrfs_fs_info *fs_info,
5897                      u64 chunk_start, u64 physical, u64 devid,
5898                      u64 **logical, int *naddrs, int *stripe_len)
5899 {
5900         struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
5901         struct extent_map_tree *em_tree = &map_tree->map_tree;
5902         struct extent_map *em;
5903         struct map_lookup *map;
5904         u64 *buf;
5905         u64 bytenr;
5906         u64 length;
5907         u64 stripe_nr;
5908         u64 rmap_len;
5909         int i, j, nr = 0;
5910
5911         read_lock(&em_tree->lock);
5912         em = lookup_extent_mapping(em_tree, chunk_start, 1);
5913         read_unlock(&em_tree->lock);
5914
5915         if (!em) {
5916                 btrfs_err(fs_info, "couldn't find em for chunk %Lu",
5917                         chunk_start);
5918                 return -EIO;
5919         }
5920
5921         if (em->start != chunk_start) {
5922                 btrfs_err(fs_info, "bad chunk start, em=%Lu, wanted=%Lu",
5923                        em->start, chunk_start);
5924                 free_extent_map(em);
5925                 return -EIO;
5926         }
5927         map = em->map_lookup;
5928
5929         length = em->len;
5930         rmap_len = map->stripe_len;
5931
5932         if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5933                 length = div_u64(length, map->num_stripes / map->sub_stripes);
5934         else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5935                 length = div_u64(length, map->num_stripes);
5936         else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5937                 length = div_u64(length, nr_data_stripes(map));
5938                 rmap_len = map->stripe_len * nr_data_stripes(map);
5939         }
5940
5941         buf = kcalloc(map->num_stripes, sizeof(u64), GFP_NOFS);
5942         BUG_ON(!buf); /* -ENOMEM */
5943
5944         for (i = 0; i < map->num_stripes; i++) {
5945                 if (devid && map->stripes[i].dev->devid != devid)
5946                         continue;
5947                 if (map->stripes[i].physical > physical ||
5948                     map->stripes[i].physical + length <= physical)
5949                         continue;
5950
5951                 stripe_nr = physical - map->stripes[i].physical;
5952                 stripe_nr = div_u64(stripe_nr, map->stripe_len);
5953
5954                 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5955                         stripe_nr = stripe_nr * map->num_stripes + i;
5956                         stripe_nr = div_u64(stripe_nr, map->sub_stripes);
5957                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5958                         stripe_nr = stripe_nr * map->num_stripes + i;
5959                 } /* else if RAID[56], multiply by nr_data_stripes().
5960                    * Alternatively, just use rmap_len below instead of
5961                    * map->stripe_len */
5962
5963                 bytenr = chunk_start + stripe_nr * rmap_len;
5964                 WARN_ON(nr >= map->num_stripes);
5965                 for (j = 0; j < nr; j++) {
5966                         if (buf[j] == bytenr)
5967                                 break;
5968                 }
5969                 if (j == nr) {
5970                         WARN_ON(nr >= map->num_stripes);
5971                         buf[nr++] = bytenr;
5972                 }
5973         }
5974
5975         *logical = buf;
5976         *naddrs = nr;
5977         *stripe_len = rmap_len;
5978
5979         free_extent_map(em);
5980         return 0;
5981 }
5982
5983 static inline void btrfs_end_bbio(struct btrfs_bio *bbio, struct bio *bio)
5984 {
5985         bio->bi_private = bbio->private;
5986         bio->bi_end_io = bbio->end_io;
5987         bio_endio(bio);
5988
5989         btrfs_put_bbio(bbio);
5990 }
5991
5992 static void btrfs_end_bio(struct bio *bio)
5993 {
5994         struct btrfs_bio *bbio = bio->bi_private;
5995         int is_orig_bio = 0;
5996
5997         if (bio->bi_error) {
5998                 atomic_inc(&bbio->error);
5999                 if (bio->bi_error == -EIO || bio->bi_error == -EREMOTEIO) {
6000                         unsigned int stripe_index =
6001                                 btrfs_io_bio(bio)->stripe_index;
6002                         struct btrfs_device *dev;
6003
6004                         BUG_ON(stripe_index >= bbio->num_stripes);
6005                         dev = bbio->stripes[stripe_index].dev;
6006                         if (dev->bdev) {
6007                                 if (bio_op(bio) == REQ_OP_WRITE)
6008                                         btrfs_dev_stat_inc(dev,
6009                                                 BTRFS_DEV_STAT_WRITE_ERRS);
6010                                 else
6011                                         btrfs_dev_stat_inc(dev,
6012                                                 BTRFS_DEV_STAT_READ_ERRS);
6013                                 if (bio->bi_opf & REQ_PREFLUSH)
6014                                         btrfs_dev_stat_inc(dev,
6015                                                 BTRFS_DEV_STAT_FLUSH_ERRS);
6016                                 btrfs_dev_stat_print_on_error(dev);
6017                         }
6018                 }
6019         }
6020
6021         if (bio == bbio->orig_bio)
6022                 is_orig_bio = 1;
6023
6024         btrfs_bio_counter_dec(bbio->fs_info);
6025
6026         if (atomic_dec_and_test(&bbio->stripes_pending)) {
6027                 if (!is_orig_bio) {
6028                         bio_put(bio);
6029                         bio = bbio->orig_bio;
6030                 }
6031
6032                 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
6033                 /* only send an error to the higher layers if it is
6034                  * beyond the tolerance of the btrfs bio
6035                  */
6036                 if (atomic_read(&bbio->error) > bbio->max_errors) {
6037                         bio->bi_error = -EIO;
6038                 } else {
6039                         /*
6040                          * this bio is actually up to date, we didn't
6041                          * go over the max number of errors
6042                          */
6043                         bio->bi_error = 0;
6044                 }
6045
6046                 btrfs_end_bbio(bbio, bio);
6047         } else if (!is_orig_bio) {
6048                 bio_put(bio);
6049         }
6050 }
6051
6052 /*
6053  * see run_scheduled_bios for a description of why bios are collected for
6054  * async submit.
6055  *
6056  * This will add one bio to the pending list for a device and make sure
6057  * the work struct is scheduled.
6058  */
6059 static noinline void btrfs_schedule_bio(struct btrfs_device *device,
6060                                         struct bio *bio)
6061 {
6062         struct btrfs_fs_info *fs_info = device->fs_info;
6063         int should_queue = 1;
6064         struct btrfs_pending_bios *pending_bios;
6065
6066         if (device->missing || !device->bdev) {
6067                 bio_io_error(bio);
6068                 return;
6069         }
6070
6071         /* don't bother with additional async steps for reads, right now */
6072         if (bio_op(bio) == REQ_OP_READ) {
6073                 bio_get(bio);
6074                 btrfsic_submit_bio(bio);
6075                 bio_put(bio);
6076                 return;
6077         }
6078
6079         /*
6080          * nr_async_bios allows us to reliably return congestion to the
6081          * higher layers.  Otherwise, the async bio makes it appear we have
6082          * made progress against dirty pages when we've really just put it
6083          * on a queue for later
6084          */
6085         atomic_inc(&fs_info->nr_async_bios);
6086         WARN_ON(bio->bi_next);
6087         bio->bi_next = NULL;
6088
6089         spin_lock(&device->io_lock);
6090         if (op_is_sync(bio->bi_opf))
6091                 pending_bios = &device->pending_sync_bios;
6092         else
6093                 pending_bios = &device->pending_bios;
6094
6095         if (pending_bios->tail)
6096                 pending_bios->tail->bi_next = bio;
6097
6098         pending_bios->tail = bio;
6099         if (!pending_bios->head)
6100                 pending_bios->head = bio;
6101         if (device->running_pending)
6102                 should_queue = 0;
6103
6104         spin_unlock(&device->io_lock);
6105
6106         if (should_queue)
6107                 btrfs_queue_work(fs_info->submit_workers, &device->work);
6108 }
6109
6110 static void submit_stripe_bio(struct btrfs_bio *bbio, struct bio *bio,
6111                               u64 physical, int dev_nr, int async)
6112 {
6113         struct btrfs_device *dev = bbio->stripes[dev_nr].dev;
6114         struct btrfs_fs_info *fs_info = bbio->fs_info;
6115
6116         bio->bi_private = bbio;
6117         btrfs_io_bio(bio)->stripe_index = dev_nr;
6118         bio->bi_end_io = btrfs_end_bio;
6119         bio->bi_iter.bi_sector = physical >> 9;
6120 #ifdef DEBUG
6121         {
6122                 struct rcu_string *name;
6123
6124                 rcu_read_lock();
6125                 name = rcu_dereference(dev->name);
6126                 btrfs_debug(fs_info,
6127                         "btrfs_map_bio: rw %d 0x%x, sector=%llu, dev=%lu (%s id %llu), size=%u",
6128                         bio_op(bio), bio->bi_opf,
6129                         (u64)bio->bi_iter.bi_sector,
6130                         (u_long)dev->bdev->bd_dev, name->str, dev->devid,
6131                         bio->bi_iter.bi_size);
6132                 rcu_read_unlock();
6133         }
6134 #endif
6135         bio->bi_bdev = dev->bdev;
6136
6137         btrfs_bio_counter_inc_noblocked(fs_info);
6138
6139         if (async)
6140                 btrfs_schedule_bio(dev, bio);
6141         else
6142                 btrfsic_submit_bio(bio);
6143 }
6144
6145 static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical)
6146 {
6147         atomic_inc(&bbio->error);
6148         if (atomic_dec_and_test(&bbio->stripes_pending)) {
6149                 /* Should be the original bio. */
6150                 WARN_ON(bio != bbio->orig_bio);
6151
6152                 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
6153                 bio->bi_iter.bi_sector = logical >> 9;
6154                 bio->bi_error = -EIO;
6155                 btrfs_end_bbio(bbio, bio);
6156         }
6157 }
6158
6159 int btrfs_map_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
6160                   int mirror_num, int async_submit)
6161 {
6162         struct btrfs_device *dev;
6163         struct bio *first_bio = bio;
6164         u64 logical = (u64)bio->bi_iter.bi_sector << 9;
6165         u64 length = 0;
6166         u64 map_length;
6167         int ret;
6168         int dev_nr;
6169         int total_devs;
6170         struct btrfs_bio *bbio = NULL;
6171
6172         length = bio->bi_iter.bi_size;
6173         map_length = length;
6174
6175         btrfs_bio_counter_inc_blocked(fs_info);
6176         ret = __btrfs_map_block(fs_info, bio_op(bio), logical,
6177                                 &map_length, &bbio, mirror_num, 1);
6178         if (ret) {
6179                 btrfs_bio_counter_dec(fs_info);
6180                 return ret;
6181         }
6182
6183         total_devs = bbio->num_stripes;
6184         bbio->orig_bio = first_bio;
6185         bbio->private = first_bio->bi_private;
6186         bbio->end_io = first_bio->bi_end_io;
6187         bbio->fs_info = fs_info;
6188         atomic_set(&bbio->stripes_pending, bbio->num_stripes);
6189
6190         if ((bbio->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
6191             ((bio_op(bio) == REQ_OP_WRITE) || (mirror_num > 1))) {
6192                 /* In this case, map_length has been set to the length of
6193                    a single stripe; not the whole write */
6194                 if (bio_op(bio) == REQ_OP_WRITE) {
6195                         ret = raid56_parity_write(fs_info, bio, bbio,
6196                                                   map_length);
6197                 } else {
6198                         ret = raid56_parity_recover(fs_info, bio, bbio,
6199                                                     map_length, mirror_num, 1);
6200                 }
6201
6202                 btrfs_bio_counter_dec(fs_info);
6203                 return ret;
6204         }
6205
6206         if (map_length < length) {
6207                 btrfs_crit(fs_info,
6208                            "mapping failed logical %llu bio len %llu len %llu",
6209                            logical, length, map_length);
6210                 BUG();
6211         }
6212
6213         for (dev_nr = 0; dev_nr < total_devs; dev_nr++) {
6214                 dev = bbio->stripes[dev_nr].dev;
6215                 if (!dev || !dev->bdev ||
6216                     (bio_op(first_bio) == REQ_OP_WRITE && !dev->writeable)) {
6217                         bbio_error(bbio, first_bio, logical);
6218                         continue;
6219                 }
6220
6221                 if (dev_nr < total_devs - 1) {
6222                         bio = btrfs_bio_clone(first_bio, GFP_NOFS);
6223                         BUG_ON(!bio); /* -ENOMEM */
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 0;
6232 }
6233
6234 struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid,
6235                                        u8 *uuid, u8 *fsid)
6236 {
6237         struct btrfs_device *device;
6238         struct btrfs_fs_devices *cur_devices;
6239
6240         cur_devices = fs_info->fs_devices;
6241         while (cur_devices) {
6242                 if (!fsid ||
6243                     !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
6244                         device = __find_device(&cur_devices->devices,
6245                                                devid, uuid);
6246                         if (device)
6247                                 return device;
6248                 }
6249                 cur_devices = cur_devices->seed;
6250         }
6251         return NULL;
6252 }
6253
6254 static struct btrfs_device *add_missing_dev(struct btrfs_fs_devices *fs_devices,
6255                                             u64 devid, u8 *dev_uuid)
6256 {
6257         struct btrfs_device *device;
6258
6259         device = btrfs_alloc_device(NULL, &devid, dev_uuid);
6260         if (IS_ERR(device))
6261                 return NULL;
6262
6263         list_add(&device->dev_list, &fs_devices->devices);
6264         device->fs_devices = fs_devices;
6265         fs_devices->num_devices++;
6266
6267         device->missing = 1;
6268         fs_devices->missing_devices++;
6269
6270         return device;
6271 }
6272
6273 /**
6274  * btrfs_alloc_device - allocate struct btrfs_device
6275  * @fs_info:    used only for generating a new devid, can be NULL if
6276  *              devid is provided (i.e. @devid != NULL).
6277  * @devid:      a pointer to devid for this device.  If NULL a new devid
6278  *              is generated.
6279  * @uuid:       a pointer to UUID for this device.  If NULL a new UUID
6280  *              is generated.
6281  *
6282  * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
6283  * on error.  Returned struct is not linked onto any lists and can be
6284  * destroyed with kfree() right away.
6285  */
6286 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
6287                                         const u64 *devid,
6288                                         const u8 *uuid)
6289 {
6290         struct btrfs_device *dev;
6291         u64 tmp;
6292
6293         if (WARN_ON(!devid && !fs_info))
6294                 return ERR_PTR(-EINVAL);
6295
6296         dev = __alloc_device();
6297         if (IS_ERR(dev))
6298                 return dev;
6299
6300         if (devid)
6301                 tmp = *devid;
6302         else {
6303                 int ret;
6304
6305                 ret = find_next_devid(fs_info, &tmp);
6306                 if (ret) {
6307                         kfree(dev);
6308                         return ERR_PTR(ret);
6309                 }
6310         }
6311         dev->devid = tmp;
6312
6313         if (uuid)
6314                 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE);
6315         else
6316                 generate_random_uuid(dev->uuid);
6317
6318         btrfs_init_work(&dev->work, btrfs_submit_helper,
6319                         pending_bios_fn, NULL, NULL);
6320
6321         return dev;
6322 }
6323
6324 /* Return -EIO if any error, otherwise return 0. */
6325 static int btrfs_check_chunk_valid(struct btrfs_fs_info *fs_info,
6326                                    struct extent_buffer *leaf,
6327                                    struct btrfs_chunk *chunk, u64 logical)
6328 {
6329         u64 length;
6330         u64 stripe_len;
6331         u16 num_stripes;
6332         u16 sub_stripes;
6333         u64 type;
6334
6335         length = btrfs_chunk_length(leaf, chunk);
6336         stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6337         num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6338         sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
6339         type = btrfs_chunk_type(leaf, chunk);
6340
6341         if (!num_stripes) {
6342                 btrfs_err(fs_info, "invalid chunk num_stripes: %u",
6343                           num_stripes);
6344                 return -EIO;
6345         }
6346         if (!IS_ALIGNED(logical, fs_info->sectorsize)) {
6347                 btrfs_err(fs_info, "invalid chunk logical %llu", logical);
6348                 return -EIO;
6349         }
6350         if (btrfs_chunk_sector_size(leaf, chunk) != fs_info->sectorsize) {
6351                 btrfs_err(fs_info, "invalid chunk sectorsize %u",
6352                           btrfs_chunk_sector_size(leaf, chunk));
6353                 return -EIO;
6354         }
6355         if (!length || !IS_ALIGNED(length, fs_info->sectorsize)) {
6356                 btrfs_err(fs_info, "invalid chunk length %llu", length);
6357                 return -EIO;
6358         }
6359         if (!is_power_of_2(stripe_len) || stripe_len != BTRFS_STRIPE_LEN) {
6360                 btrfs_err(fs_info, "invalid chunk stripe length: %llu",
6361                           stripe_len);
6362                 return -EIO;
6363         }
6364         if (~(BTRFS_BLOCK_GROUP_TYPE_MASK | BTRFS_BLOCK_GROUP_PROFILE_MASK) &
6365             type) {
6366                 btrfs_err(fs_info, "unrecognized chunk type: %llu",
6367                           ~(BTRFS_BLOCK_GROUP_TYPE_MASK |
6368                             BTRFS_BLOCK_GROUP_PROFILE_MASK) &
6369                           btrfs_chunk_type(leaf, chunk));
6370                 return -EIO;
6371         }
6372         if ((type & BTRFS_BLOCK_GROUP_RAID10 && sub_stripes != 2) ||
6373             (type & BTRFS_BLOCK_GROUP_RAID1 && num_stripes < 1) ||
6374             (type & BTRFS_BLOCK_GROUP_RAID5 && num_stripes < 2) ||
6375             (type & BTRFS_BLOCK_GROUP_RAID6 && num_stripes < 3) ||
6376             (type & BTRFS_BLOCK_GROUP_DUP && num_stripes > 2) ||
6377             ((type & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 &&
6378              num_stripes != 1)) {
6379                 btrfs_err(fs_info,
6380                         "invalid num_stripes:sub_stripes %u:%u for profile %llu",
6381                         num_stripes, sub_stripes,
6382                         type & BTRFS_BLOCK_GROUP_PROFILE_MASK);
6383                 return -EIO;
6384         }
6385
6386         return 0;
6387 }
6388
6389 static int read_one_chunk(struct btrfs_fs_info *fs_info, struct btrfs_key *key,
6390                           struct extent_buffer *leaf,
6391                           struct btrfs_chunk *chunk)
6392 {
6393         struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
6394         struct map_lookup *map;
6395         struct extent_map *em;
6396         u64 logical;
6397         u64 length;
6398         u64 stripe_len;
6399         u64 devid;
6400         u8 uuid[BTRFS_UUID_SIZE];
6401         int num_stripes;
6402         int ret;
6403         int i;
6404
6405         logical = key->offset;
6406         length = btrfs_chunk_length(leaf, chunk);
6407         stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6408         num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6409
6410         ret = btrfs_check_chunk_valid(fs_info, leaf, chunk, logical);
6411         if (ret)
6412                 return ret;
6413
6414         read_lock(&map_tree->map_tree.lock);
6415         em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
6416         read_unlock(&map_tree->map_tree.lock);
6417
6418         /* already mapped? */
6419         if (em && em->start <= logical && em->start + em->len > logical) {
6420                 free_extent_map(em);
6421                 return 0;
6422         } else if (em) {
6423                 free_extent_map(em);
6424         }
6425
6426         em = alloc_extent_map();
6427         if (!em)
6428                 return -ENOMEM;
6429         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
6430         if (!map) {
6431                 free_extent_map(em);
6432                 return -ENOMEM;
6433         }
6434
6435         set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
6436         em->map_lookup = map;
6437         em->start = logical;
6438         em->len = length;
6439         em->orig_start = 0;
6440         em->block_start = 0;
6441         em->block_len = em->len;
6442
6443         map->num_stripes = num_stripes;
6444         map->io_width = btrfs_chunk_io_width(leaf, chunk);
6445         map->io_align = btrfs_chunk_io_align(leaf, chunk);
6446         map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
6447         map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6448         map->type = btrfs_chunk_type(leaf, chunk);
6449         map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
6450         for (i = 0; i < num_stripes; i++) {
6451                 map->stripes[i].physical =
6452                         btrfs_stripe_offset_nr(leaf, chunk, i);
6453                 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
6454                 read_extent_buffer(leaf, uuid, (unsigned long)
6455                                    btrfs_stripe_dev_uuid_nr(chunk, i),
6456                                    BTRFS_UUID_SIZE);
6457                 map->stripes[i].dev = btrfs_find_device(fs_info, devid,
6458                                                         uuid, NULL);
6459                 if (!map->stripes[i].dev &&
6460                     !btrfs_test_opt(fs_info, DEGRADED)) {
6461                         free_extent_map(em);
6462                         return -EIO;
6463                 }
6464                 if (!map->stripes[i].dev) {
6465                         map->stripes[i].dev =
6466                                 add_missing_dev(fs_info->fs_devices, devid,
6467                                                 uuid);
6468                         if (!map->stripes[i].dev) {
6469                                 free_extent_map(em);
6470                                 return -EIO;
6471                         }
6472                         btrfs_warn(fs_info, "devid %llu uuid %pU is missing",
6473                                    devid, uuid);
6474                 }
6475                 map->stripes[i].dev->in_fs_metadata = 1;
6476         }
6477
6478         write_lock(&map_tree->map_tree.lock);
6479         ret = add_extent_mapping(&map_tree->map_tree, em, 0);
6480         write_unlock(&map_tree->map_tree.lock);
6481         BUG_ON(ret); /* Tree corruption */
6482         free_extent_map(em);
6483
6484         return 0;
6485 }
6486
6487 static void fill_device_from_item(struct extent_buffer *leaf,
6488                                  struct btrfs_dev_item *dev_item,
6489                                  struct btrfs_device *device)
6490 {
6491         unsigned long ptr;
6492
6493         device->devid = btrfs_device_id(leaf, dev_item);
6494         device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
6495         device->total_bytes = device->disk_total_bytes;
6496         device->commit_total_bytes = device->disk_total_bytes;
6497         device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
6498         device->commit_bytes_used = device->bytes_used;
6499         device->type = btrfs_device_type(leaf, dev_item);
6500         device->io_align = btrfs_device_io_align(leaf, dev_item);
6501         device->io_width = btrfs_device_io_width(leaf, dev_item);
6502         device->sector_size = btrfs_device_sector_size(leaf, dev_item);
6503         WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
6504         device->is_tgtdev_for_dev_replace = 0;
6505
6506         ptr = btrfs_device_uuid(dev_item);
6507         read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
6508 }
6509
6510 static struct btrfs_fs_devices *open_seed_devices(struct btrfs_fs_info *fs_info,
6511                                                   u8 *fsid)
6512 {
6513         struct btrfs_fs_devices *fs_devices;
6514         int ret;
6515
6516         BUG_ON(!mutex_is_locked(&uuid_mutex));
6517
6518         fs_devices = fs_info->fs_devices->seed;
6519         while (fs_devices) {
6520                 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE))
6521                         return fs_devices;
6522
6523                 fs_devices = fs_devices->seed;
6524         }
6525
6526         fs_devices = find_fsid(fsid);
6527         if (!fs_devices) {
6528                 if (!btrfs_test_opt(fs_info, DEGRADED))
6529                         return ERR_PTR(-ENOENT);
6530
6531                 fs_devices = alloc_fs_devices(fsid);
6532                 if (IS_ERR(fs_devices))
6533                         return fs_devices;
6534
6535                 fs_devices->seeding = 1;
6536                 fs_devices->opened = 1;
6537                 return fs_devices;
6538         }
6539
6540         fs_devices = clone_fs_devices(fs_devices);
6541         if (IS_ERR(fs_devices))
6542                 return fs_devices;
6543
6544         ret = __btrfs_open_devices(fs_devices, FMODE_READ,
6545                                    fs_info->bdev_holder);
6546         if (ret) {
6547                 free_fs_devices(fs_devices);
6548                 fs_devices = ERR_PTR(ret);
6549                 goto out;
6550         }
6551
6552         if (!fs_devices->seeding) {
6553                 __btrfs_close_devices(fs_devices);
6554                 free_fs_devices(fs_devices);
6555                 fs_devices = ERR_PTR(-EINVAL);
6556                 goto out;
6557         }
6558
6559         fs_devices->seed = fs_info->fs_devices->seed;
6560         fs_info->fs_devices->seed = fs_devices;
6561 out:
6562         return fs_devices;
6563 }
6564
6565 static int read_one_dev(struct btrfs_fs_info *fs_info,
6566                         struct extent_buffer *leaf,
6567                         struct btrfs_dev_item *dev_item)
6568 {
6569         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6570         struct btrfs_device *device;
6571         u64 devid;
6572         int ret;
6573         u8 fs_uuid[BTRFS_UUID_SIZE];
6574         u8 dev_uuid[BTRFS_UUID_SIZE];
6575
6576         devid = btrfs_device_id(leaf, dev_item);
6577         read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
6578                            BTRFS_UUID_SIZE);
6579         read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
6580                            BTRFS_UUID_SIZE);
6581
6582         if (memcmp(fs_uuid, fs_info->fsid, BTRFS_UUID_SIZE)) {
6583                 fs_devices = open_seed_devices(fs_info, fs_uuid);
6584                 if (IS_ERR(fs_devices))
6585                         return PTR_ERR(fs_devices);
6586         }
6587
6588         device = btrfs_find_device(fs_info, devid, dev_uuid, fs_uuid);
6589         if (!device) {
6590                 if (!btrfs_test_opt(fs_info, DEGRADED))
6591                         return -EIO;
6592
6593                 device = add_missing_dev(fs_devices, devid, dev_uuid);
6594                 if (!device)
6595                         return -ENOMEM;
6596                 btrfs_warn(fs_info, "devid %llu uuid %pU missing",
6597                                 devid, dev_uuid);
6598         } else {
6599                 if (!device->bdev && !btrfs_test_opt(fs_info, DEGRADED))
6600                         return -EIO;
6601
6602                 if(!device->bdev && !device->missing) {
6603                         /*
6604                          * this happens when a device that was properly setup
6605                          * in the device info lists suddenly goes bad.
6606                          * device->bdev is NULL, and so we have to set
6607                          * device->missing to one here
6608                          */
6609                         device->fs_devices->missing_devices++;
6610                         device->missing = 1;
6611                 }
6612
6613                 /* Move the device to its own fs_devices */
6614                 if (device->fs_devices != fs_devices) {
6615                         ASSERT(device->missing);
6616
6617                         list_move(&device->dev_list, &fs_devices->devices);
6618                         device->fs_devices->num_devices--;
6619                         fs_devices->num_devices++;
6620
6621                         device->fs_devices->missing_devices--;
6622                         fs_devices->missing_devices++;
6623
6624                         device->fs_devices = fs_devices;
6625                 }
6626         }
6627
6628         if (device->fs_devices != fs_info->fs_devices) {
6629                 BUG_ON(device->writeable);
6630                 if (device->generation !=
6631                     btrfs_device_generation(leaf, dev_item))
6632                         return -EINVAL;
6633         }
6634
6635         fill_device_from_item(leaf, dev_item, device);
6636         device->in_fs_metadata = 1;
6637         if (device->writeable && !device->is_tgtdev_for_dev_replace) {
6638                 device->fs_devices->total_rw_bytes += device->total_bytes;
6639                 spin_lock(&fs_info->free_chunk_lock);
6640                 fs_info->free_chunk_space += device->total_bytes -
6641                         device->bytes_used;
6642                 spin_unlock(&fs_info->free_chunk_lock);
6643         }
6644         ret = 0;
6645         return ret;
6646 }
6647
6648 int btrfs_read_sys_array(struct btrfs_fs_info *fs_info)
6649 {
6650         struct btrfs_root *root = fs_info->tree_root;
6651         struct btrfs_super_block *super_copy = fs_info->super_copy;
6652         struct extent_buffer *sb;
6653         struct btrfs_disk_key *disk_key;
6654         struct btrfs_chunk *chunk;
6655         u8 *array_ptr;
6656         unsigned long sb_array_offset;
6657         int ret = 0;
6658         u32 num_stripes;
6659         u32 array_size;
6660         u32 len = 0;
6661         u32 cur_offset;
6662         u64 type;
6663         struct btrfs_key key;
6664
6665         ASSERT(BTRFS_SUPER_INFO_SIZE <= fs_info->nodesize);
6666         /*
6667          * This will create extent buffer of nodesize, superblock size is
6668          * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will
6669          * overallocate but we can keep it as-is, only the first page is used.
6670          */
6671         sb = btrfs_find_create_tree_block(fs_info, BTRFS_SUPER_INFO_OFFSET);
6672         if (IS_ERR(sb))
6673                 return PTR_ERR(sb);
6674         set_extent_buffer_uptodate(sb);
6675         btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
6676         /*
6677          * The sb extent buffer is artificial and just used to read the system array.
6678          * set_extent_buffer_uptodate() call does not properly mark all it's
6679          * pages up-to-date when the page is larger: extent does not cover the
6680          * whole page and consequently check_page_uptodate does not find all
6681          * the page's extents up-to-date (the hole beyond sb),
6682          * write_extent_buffer then triggers a WARN_ON.
6683          *
6684          * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
6685          * but sb spans only this function. Add an explicit SetPageUptodate call
6686          * to silence the warning eg. on PowerPC 64.
6687          */
6688         if (PAGE_SIZE > BTRFS_SUPER_INFO_SIZE)
6689                 SetPageUptodate(sb->pages[0]);
6690
6691         write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
6692         array_size = btrfs_super_sys_array_size(super_copy);
6693
6694         array_ptr = super_copy->sys_chunk_array;
6695         sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array);
6696         cur_offset = 0;
6697
6698         while (cur_offset < array_size) {
6699                 disk_key = (struct btrfs_disk_key *)array_ptr;
6700                 len = sizeof(*disk_key);
6701                 if (cur_offset + len > array_size)
6702                         goto out_short_read;
6703
6704                 btrfs_disk_key_to_cpu(&key, disk_key);
6705
6706                 array_ptr += len;
6707                 sb_array_offset += len;
6708                 cur_offset += len;
6709
6710                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
6711                         chunk = (struct btrfs_chunk *)sb_array_offset;
6712                         /*
6713                          * At least one btrfs_chunk with one stripe must be
6714                          * present, exact stripe count check comes afterwards
6715                          */
6716                         len = btrfs_chunk_item_size(1);
6717                         if (cur_offset + len > array_size)
6718                                 goto out_short_read;
6719
6720                         num_stripes = btrfs_chunk_num_stripes(sb, chunk);
6721                         if (!num_stripes) {
6722                                 btrfs_err(fs_info,
6723                                         "invalid number of stripes %u in sys_array at offset %u",
6724                                         num_stripes, cur_offset);
6725                                 ret = -EIO;
6726                                 break;
6727                         }
6728
6729                         type = btrfs_chunk_type(sb, chunk);
6730                         if ((type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) {
6731                                 btrfs_err(fs_info,
6732                             "invalid chunk type %llu in sys_array at offset %u",
6733                                         type, cur_offset);
6734                                 ret = -EIO;
6735                                 break;
6736                         }
6737
6738                         len = btrfs_chunk_item_size(num_stripes);
6739                         if (cur_offset + len > array_size)
6740                                 goto out_short_read;
6741
6742                         ret = read_one_chunk(fs_info, &key, sb, chunk);
6743                         if (ret)
6744                                 break;
6745                 } else {
6746                         btrfs_err(fs_info,
6747                             "unexpected item type %u in sys_array at offset %u",
6748                                   (u32)key.type, cur_offset);
6749                         ret = -EIO;
6750                         break;
6751                 }
6752                 array_ptr += len;
6753                 sb_array_offset += len;
6754                 cur_offset += len;
6755         }
6756         clear_extent_buffer_uptodate(sb);
6757         free_extent_buffer_stale(sb);
6758         return ret;
6759
6760 out_short_read:
6761         btrfs_err(fs_info, "sys_array too short to read %u bytes at offset %u",
6762                         len, cur_offset);
6763         clear_extent_buffer_uptodate(sb);
6764         free_extent_buffer_stale(sb);
6765         return -EIO;
6766 }
6767
6768 int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info)
6769 {
6770         struct btrfs_root *root = fs_info->chunk_root;
6771         struct btrfs_path *path;
6772         struct extent_buffer *leaf;
6773         struct btrfs_key key;
6774         struct btrfs_key found_key;
6775         int ret;
6776         int slot;
6777         u64 total_dev = 0;
6778
6779         path = btrfs_alloc_path();
6780         if (!path)
6781                 return -ENOMEM;
6782
6783         mutex_lock(&uuid_mutex);
6784         mutex_lock(&fs_info->chunk_mutex);
6785
6786         /*
6787          * Read all device items, and then all the chunk items. All
6788          * device items are found before any chunk item (their object id
6789          * is smaller than the lowest possible object id for a chunk
6790          * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
6791          */
6792         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
6793         key.offset = 0;
6794         key.type = 0;
6795         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6796         if (ret < 0)
6797                 goto error;
6798         while (1) {
6799                 leaf = path->nodes[0];
6800                 slot = path->slots[0];
6801                 if (slot >= btrfs_header_nritems(leaf)) {
6802                         ret = btrfs_next_leaf(root, path);
6803                         if (ret == 0)
6804                                 continue;
6805                         if (ret < 0)
6806                                 goto error;
6807                         break;
6808                 }
6809                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6810                 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
6811                         struct btrfs_dev_item *dev_item;
6812                         dev_item = btrfs_item_ptr(leaf, slot,
6813                                                   struct btrfs_dev_item);
6814                         ret = read_one_dev(fs_info, leaf, dev_item);
6815                         if (ret)
6816                                 goto error;
6817                         total_dev++;
6818                 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
6819                         struct btrfs_chunk *chunk;
6820                         chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
6821                         ret = read_one_chunk(fs_info, &found_key, leaf, chunk);
6822                         if (ret)
6823                                 goto error;
6824                 }
6825                 path->slots[0]++;
6826         }
6827
6828         /*
6829          * After loading chunk tree, we've got all device information,
6830          * do another round of validation checks.
6831          */
6832         if (total_dev != fs_info->fs_devices->total_devices) {
6833                 btrfs_err(fs_info,
6834            "super_num_devices %llu mismatch with num_devices %llu found here",
6835                           btrfs_super_num_devices(fs_info->super_copy),
6836                           total_dev);
6837                 ret = -EINVAL;
6838                 goto error;
6839         }
6840         if (btrfs_super_total_bytes(fs_info->super_copy) <
6841             fs_info->fs_devices->total_rw_bytes) {
6842                 btrfs_err(fs_info,
6843         "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu",
6844                           btrfs_super_total_bytes(fs_info->super_copy),
6845                           fs_info->fs_devices->total_rw_bytes);
6846                 ret = -EINVAL;
6847                 goto error;
6848         }
6849         ret = 0;
6850 error:
6851         mutex_unlock(&fs_info->chunk_mutex);
6852         mutex_unlock(&uuid_mutex);
6853
6854         btrfs_free_path(path);
6855         return ret;
6856 }
6857
6858 void btrfs_init_devices_late(struct btrfs_fs_info *fs_info)
6859 {
6860         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6861         struct btrfs_device *device;
6862
6863         while (fs_devices) {
6864                 mutex_lock(&fs_devices->device_list_mutex);
6865                 list_for_each_entry(device, &fs_devices->devices, dev_list)
6866                         device->fs_info = fs_info;
6867                 mutex_unlock(&fs_devices->device_list_mutex);
6868
6869                 fs_devices = fs_devices->seed;
6870         }
6871 }
6872
6873 static void __btrfs_reset_dev_stats(struct btrfs_device *dev)
6874 {
6875         int i;
6876
6877         for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6878                 btrfs_dev_stat_reset(dev, i);
6879 }
6880
6881 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
6882 {
6883         struct btrfs_key key;
6884         struct btrfs_key found_key;
6885         struct btrfs_root *dev_root = fs_info->dev_root;
6886         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6887         struct extent_buffer *eb;
6888         int slot;
6889         int ret = 0;
6890         struct btrfs_device *device;
6891         struct btrfs_path *path = NULL;
6892         int i;
6893
6894         path = btrfs_alloc_path();
6895         if (!path) {
6896                 ret = -ENOMEM;
6897                 goto out;
6898         }
6899
6900         mutex_lock(&fs_devices->device_list_mutex);
6901         list_for_each_entry(device, &fs_devices->devices, dev_list) {
6902                 int item_size;
6903                 struct btrfs_dev_stats_item *ptr;
6904
6905                 key.objectid = BTRFS_DEV_STATS_OBJECTID;
6906                 key.type = BTRFS_PERSISTENT_ITEM_KEY;
6907                 key.offset = device->devid;
6908                 ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
6909                 if (ret) {
6910                         __btrfs_reset_dev_stats(device);
6911                         device->dev_stats_valid = 1;
6912                         btrfs_release_path(path);
6913                         continue;
6914                 }
6915                 slot = path->slots[0];
6916                 eb = path->nodes[0];
6917                 btrfs_item_key_to_cpu(eb, &found_key, slot);
6918                 item_size = btrfs_item_size_nr(eb, slot);
6919
6920                 ptr = btrfs_item_ptr(eb, slot,
6921                                      struct btrfs_dev_stats_item);
6922
6923                 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
6924                         if (item_size >= (1 + i) * sizeof(__le64))
6925                                 btrfs_dev_stat_set(device, i,
6926                                         btrfs_dev_stats_value(eb, ptr, i));
6927                         else
6928                                 btrfs_dev_stat_reset(device, i);
6929                 }
6930
6931                 device->dev_stats_valid = 1;
6932                 btrfs_dev_stat_print_on_load(device);
6933                 btrfs_release_path(path);
6934         }
6935         mutex_unlock(&fs_devices->device_list_mutex);
6936
6937 out:
6938         btrfs_free_path(path);
6939         return ret < 0 ? ret : 0;
6940 }
6941
6942 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
6943                                 struct btrfs_fs_info *fs_info,
6944                                 struct btrfs_device *device)
6945 {
6946         struct btrfs_root *dev_root = fs_info->dev_root;
6947         struct btrfs_path *path;
6948         struct btrfs_key key;
6949         struct extent_buffer *eb;
6950         struct btrfs_dev_stats_item *ptr;
6951         int ret;
6952         int i;
6953
6954         key.objectid = BTRFS_DEV_STATS_OBJECTID;
6955         key.type = BTRFS_PERSISTENT_ITEM_KEY;
6956         key.offset = device->devid;
6957
6958         path = btrfs_alloc_path();
6959         if (!path)
6960                 return -ENOMEM;
6961         ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
6962         if (ret < 0) {
6963                 btrfs_warn_in_rcu(fs_info,
6964                         "error %d while searching for dev_stats item for device %s",
6965                               ret, rcu_str_deref(device->name));
6966                 goto out;
6967         }
6968
6969         if (ret == 0 &&
6970             btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
6971                 /* need to delete old one and insert a new one */
6972                 ret = btrfs_del_item(trans, dev_root, path);
6973                 if (ret != 0) {
6974                         btrfs_warn_in_rcu(fs_info,
6975                                 "delete too small dev_stats item for device %s failed %d",
6976                                       rcu_str_deref(device->name), ret);
6977                         goto out;
6978                 }
6979                 ret = 1;
6980         }
6981
6982         if (ret == 1) {
6983                 /* need to insert a new item */
6984                 btrfs_release_path(path);
6985                 ret = btrfs_insert_empty_item(trans, dev_root, path,
6986                                               &key, sizeof(*ptr));
6987                 if (ret < 0) {
6988                         btrfs_warn_in_rcu(fs_info,
6989                                 "insert dev_stats item for device %s failed %d",
6990                                 rcu_str_deref(device->name), ret);
6991                         goto out;
6992                 }
6993         }
6994
6995         eb = path->nodes[0];
6996         ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
6997         for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6998                 btrfs_set_dev_stats_value(eb, ptr, i,
6999                                           btrfs_dev_stat_read(device, i));
7000         btrfs_mark_buffer_dirty(eb);
7001
7002 out:
7003         btrfs_free_path(path);
7004         return ret;
7005 }
7006
7007 /*
7008  * called from commit_transaction. Writes all changed device stats to disk.
7009  */
7010 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans,
7011                         struct btrfs_fs_info *fs_info)
7012 {
7013         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7014         struct btrfs_device *device;
7015         int stats_cnt;
7016         int ret = 0;
7017
7018         mutex_lock(&fs_devices->device_list_mutex);
7019         list_for_each_entry(device, &fs_devices->devices, dev_list) {
7020                 if (!device->dev_stats_valid || !btrfs_dev_stats_dirty(device))
7021                         continue;
7022
7023                 stats_cnt = atomic_read(&device->dev_stats_ccnt);
7024                 ret = update_dev_stat_item(trans, fs_info, device);
7025                 if (!ret)
7026                         atomic_sub(stats_cnt, &device->dev_stats_ccnt);
7027         }
7028         mutex_unlock(&fs_devices->device_list_mutex);
7029
7030         return ret;
7031 }
7032
7033 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
7034 {
7035         btrfs_dev_stat_inc(dev, index);
7036         btrfs_dev_stat_print_on_error(dev);
7037 }
7038
7039 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
7040 {
7041         if (!dev->dev_stats_valid)
7042                 return;
7043         btrfs_err_rl_in_rcu(dev->fs_info,
7044                 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7045                            rcu_str_deref(dev->name),
7046                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7047                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7048                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7049                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7050                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7051 }
7052
7053 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
7054 {
7055         int i;
7056
7057         for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7058                 if (btrfs_dev_stat_read(dev, i) != 0)
7059                         break;
7060         if (i == BTRFS_DEV_STAT_VALUES_MAX)
7061                 return; /* all values == 0, suppress message */
7062
7063         btrfs_info_in_rcu(dev->fs_info,
7064                 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7065                rcu_str_deref(dev->name),
7066                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7067                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7068                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7069                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7070                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7071 }
7072
7073 int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info,
7074                         struct btrfs_ioctl_get_dev_stats *stats)
7075 {
7076         struct btrfs_device *dev;
7077         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7078         int i;
7079
7080         mutex_lock(&fs_devices->device_list_mutex);
7081         dev = btrfs_find_device(fs_info, stats->devid, NULL, NULL);
7082         mutex_unlock(&fs_devices->device_list_mutex);
7083
7084         if (!dev) {
7085                 btrfs_warn(fs_info, "get dev_stats failed, device not found");
7086                 return -ENODEV;
7087         } else if (!dev->dev_stats_valid) {
7088                 btrfs_warn(fs_info, "get dev_stats failed, not yet valid");
7089                 return -ENODEV;
7090         } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
7091                 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7092                         if (stats->nr_items > i)
7093                                 stats->values[i] =
7094                                         btrfs_dev_stat_read_and_reset(dev, i);
7095                         else
7096                                 btrfs_dev_stat_reset(dev, i);
7097                 }
7098         } else {
7099                 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7100                         if (stats->nr_items > i)
7101                                 stats->values[i] = btrfs_dev_stat_read(dev, i);
7102         }
7103         if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
7104                 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
7105         return 0;
7106 }
7107
7108 void btrfs_scratch_superblocks(struct block_device *bdev, const char *device_path)
7109 {
7110         struct buffer_head *bh;
7111         struct btrfs_super_block *disk_super;
7112         int copy_num;
7113
7114         if (!bdev)
7115                 return;
7116
7117         for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX;
7118                 copy_num++) {
7119
7120                 if (btrfs_read_dev_one_super(bdev, copy_num, &bh))
7121                         continue;
7122
7123                 disk_super = (struct btrfs_super_block *)bh->b_data;
7124
7125                 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
7126                 set_buffer_dirty(bh);
7127                 sync_dirty_buffer(bh);
7128                 brelse(bh);
7129         }
7130
7131         /* Notify udev that device has changed */
7132         btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
7133
7134         /* Update ctime/mtime for device path for libblkid */
7135         update_dev_time(device_path);
7136 }
7137
7138 /*
7139  * Update the size of all devices, which is used for writing out the
7140  * super blocks.
7141  */
7142 void btrfs_update_commit_device_size(struct btrfs_fs_info *fs_info)
7143 {
7144         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7145         struct btrfs_device *curr, *next;
7146
7147         if (list_empty(&fs_devices->resized_devices))
7148                 return;
7149
7150         mutex_lock(&fs_devices->device_list_mutex);
7151         mutex_lock(&fs_info->chunk_mutex);
7152         list_for_each_entry_safe(curr, next, &fs_devices->resized_devices,
7153                                  resized_list) {
7154                 list_del_init(&curr->resized_list);
7155                 curr->commit_total_bytes = curr->disk_total_bytes;
7156         }
7157         mutex_unlock(&fs_info->chunk_mutex);
7158         mutex_unlock(&fs_devices->device_list_mutex);
7159 }
7160
7161 /* Must be invoked during the transaction commit */
7162 void btrfs_update_commit_device_bytes_used(struct btrfs_fs_info *fs_info,
7163                                         struct btrfs_transaction *transaction)
7164 {
7165         struct extent_map *em;
7166         struct map_lookup *map;
7167         struct btrfs_device *dev;
7168         int i;
7169
7170         if (list_empty(&transaction->pending_chunks))
7171                 return;
7172
7173         /* In order to kick the device replace finish process */
7174         mutex_lock(&fs_info->chunk_mutex);
7175         list_for_each_entry(em, &transaction->pending_chunks, list) {
7176                 map = em->map_lookup;
7177
7178                 for (i = 0; i < map->num_stripes; i++) {
7179                         dev = map->stripes[i].dev;
7180                         dev->commit_bytes_used = dev->bytes_used;
7181                 }
7182         }
7183         mutex_unlock(&fs_info->chunk_mutex);
7184 }
7185
7186 void btrfs_set_fs_info_ptr(struct btrfs_fs_info *fs_info)
7187 {
7188         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7189         while (fs_devices) {
7190                 fs_devices->fs_info = fs_info;
7191                 fs_devices = fs_devices->seed;
7192         }
7193 }
7194
7195 void btrfs_reset_fs_info_ptr(struct btrfs_fs_info *fs_info)
7196 {
7197         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7198         while (fs_devices) {
7199                 fs_devices->fs_info = NULL;
7200                 fs_devices = fs_devices->seed;
7201         }
7202 }